High_Voltage_Engineering_2E.pdf

High Voltage Engineering
Fundamentals

High Voltage Engineering
Fundamentals
Second edition
E. Kuffel
Dean Emeritus,
University of Manitoba,
Winnipeg, Canada
W.S. Zaengl
Professor Emeritus,
Electrical Engineering Dept.,
Swiss Federal Institute of Technology,
Zurich, Switzerland
J. Kuffel
Manager of High Voltage and Current Laboratories,
Ontario Hydro Technologies,
Toronto, Canada
Newnes
OXFORD AUCKLAND BOSTON JOHANNESBURG MELBOURNE NEW DELHI

Newnes
An imprint of Butterworth-Heinemann
Linacre House, Jordan Hill, Oxford OX2 8DP
225 Wildwood Avenue, Woburn, MA 01801-2041
A division of Reed Educational and Professional Publishing Ltd
First published 1984 by Pergamon Press
Reprinted 1986
Second edition 2000, published by Butterworth-Heinemann
 E. Kuffel and W.S. Zaengl 1984
 E. Kuffel, W.S. Zaengl and J. Kuffel 2000
All rights reserved. No part of this publication
may be reproduced in any material form (including
photocopying or storing in any medium by electronic
means and whether or not transiently or incidentally
to some other use of this publication) without the
written permission of the copyright holder except
in accordance with the provisions of the Copyright,
Designs and Patents Act 1988 or under the terms of a
licence issued by the Copyright Licensing Agency Ltd,
90 Tottenham Court Road, London, England W1P 9HE.
Applications for the copyright holder’s written permission
to reproduce any part of this publication should be addressed
to the publishers
British Library Cataloguing in Publication Data
A catalogue record for this book is available from the British Library
Library of Congress Cataloguing in Publication Data
A catalogue record for this book is available from the Library of Congress
ISBN 0 7506 3634 3
Typeset by Laser Words, Madras, India
Printed in Great Britain

Contents
Preface to second edition xi
Preface to first edition xv
Chapter 1 Introduction 1
1.1 Generation and transmission of electric energy 1
1.2 Voltage stresses 3
1.3 Testing voltages 5
1.3.1 Testing with power frequency voltages 5
1.3.2 Testing with lightning impulse voltages 5
1.3.3 Testing with switching impulses 6
1.3.4 D.C. voltages 6
1.3.5 Testing with very low frequency voltage 7
References 7
Chapter 2 Generation of high voltages 8
2.1 Direct voltages 9
2.1.1 A.C. to D.C. conversion 10
2.1.2 Electrostatic generators 24
2.2 Alternating voltages 29
2.2.1 Testing transformers 32
2.2.2 Series resonant circuits 40
2.3 Impulse voltages 48
2.3.1 Impulse voltage generator circuits 52
2.3.2 Operation, design and construction of impulse generators 66
2.4 Control systems 74
References 75
Chapter 3 Measurement of high voltages 77
3.1 Peak voltage measurements by spark gaps 78
3.1.1 Sphere gaps 79
3.1.2 Reference measuring systems 91

vi Contents
3.1.3 Uniform field gaps 92
3.1.4 Rod gaps 93
3.2 Electrostatic voltmeters 94
3.3 Ammeter in series with high ohmic resistors and high ohmic resistor voltage
dividers 96
3.4 Generating voltmeters and field sensors 107
3.5 The measurement of peak voltages 109
3.5.1 The Chubb–Fortescue method 110
3.5.2 Voltage dividers and passive rectifier circuits 113
3.5.3 Active peak-reading circuits 117
3.5.4 High-voltage capacitors for measuring circuits 118
3.6 Voltage dividing systems and impulse voltage measurements 129
3.6.1 Generalized voltage generation and measuring circuit 129
3.6.2 Demands upon transfer characteristics of the measuring system 132
3.6.3 Fundamentals for the computation of the measuring system 139
3.6.4 Voltage dividers 147
3.6.5 Interaction between voltage divider and its lead 163
3.6.6 The divider’s low-voltage arm 171
3.7 Fast digital transient recorders for impulse measurements 175
3.7.1 Principles and historical development of transient digital recorders
176
3.7.2 Errors inherent in digital recorders 179
3.7.3 Specification of ideal A/D recorder and parameters required for h.v.
impulse testing 183
3.7.4 Future trends 195
References 196
Chapter 4 Electrostatic fields and field stress control 201
4.1 Electrical field distribution and breakdown strength of insulating materials
201
4.2 Fields in homogeneous, isotropic materials 205
4.2.1 The uniform field electrode arrangement 206
4.2.2 Coaxial cylindrical and spherical fields 209
4.2.3 Sphere-to-sphere or sphere-to-plane 214
4.2.4 Two cylindrical conductors in parallel 218
4.2.5 Field distortions by conducting particles 221
4.3 Fields in multidielectric, isotropic materials 225
4.3.1 Simple configurations 227
4.3.2 Dielectric refraction 232
4.3.3 Stress control by floating screens 235
4.4 Numerical methods 241
4.4.1 Finite difference method (FDM) 242

Contents vii
4.4.2 Finite element method (FEM) 246
4.4.3 Charge simulation method (CSM) 254
4.4.4 Boundary element method 270
References 278
Chapter 5 Electrical breakdown in gases 281
5.1 Classical gas laws 281
5.1.1 Velocity distribution of a swarm of molecules 284
5.1.2 The free path of molecules and electrons 287
5.1.3 Distribution of free paths 290
5.1.4 Collision-energy transfer 291
5.2 Ionization and decay processes 294
5.2.1 Townsend first ionization coefficient 295
5.2.2 Photoionization 301
5.2.3 Ionization by interaction of metastables with atoms 301
5.2.4 Thermal ionization 302
5.2.5 Deionization by recombination 302
5.2.6 Deionization by attachment–negative ion formation 304
5.2.7 Mobility of gaseous ions and deionization by diffusion 308
5.2.8 Relation between diffusion and mobility 314
5.3 Cathode processes – secondary effects 316
5.3.1 Photoelectric emission 317
5.3.2 Electron emission by positive ion and excited atom impact 317
5.3.3 Thermionic emission 318
5.3.4 Field emission 319
5.3.5 Townsend second ionization coefficient 321
5.3.6 Secondary electron emission by photon impact 323
5.4 Transition from non-self-sustained discharges to breakdown 324
5.4.1 The Townsend mechanism 324
5.5 The streamer or ‘Kanal’ mechanism of spark 326
5.6 The sparking voltage–Paschen’s law 333
5.7 Penning effect 339
5.8 The breakdown field strength (Eb) 340
5.9 Breakdown in non-uniform fields 342
5.10 Effect of electron attachment on the breakdown criteria 345
5.11 Partial breakdown, corona discharges 348
5.11.1 Positive or anode coronas 349
5.11.2 Negative or cathode corona 352
5.12 Polarity effect – influence of space charge 354
5.13 Surge breakdown voltage–time lag 359

viii Contents
5.13.1 Breakdown under impulse voltages 360
5.13.2 Volt–time characteristics 361
5.13.3 Experimental studies of time lags 362
References 365
Chapter 6 Breakdown in solid and liquid dielectrics 367
6.1 Breakdown in solids 367
6.1.1 Intrinsic breakdown 368
6.1.2 Streamer breakdown 373
6.1.3 Electromechanical breakdown 373
6.1.4 Edge breakdown and treeing 374
6.1.5 Thermal breakdown 375
6.1.6 Erosion breakdown 381
6.1.7 Tracking 385
6.2 Breakdown in liquids 385
6.2.1 Electronic breakdown 386
6.2.2 Suspended solid particle mechanism 387
6.2.3 Cavity breakdown 390
6.2.4 Electroconvection and electrohydrodynamic model of dielectric
breakdown 391
6.3 Static electrification in power transformers 393
References 394
Chapter 7 Non-destructive insulation test techniques 395
7.1 Dynamic properties of dielectrics 395
7.1.1 Dynamic properties in the time domain 398
7.1.2 Dynamic properties in the frequency domain 404
7.1.3 Modelling of dielectric properties 407
7.1.4 Applications to insulation ageing 409
7.2 Dielectric loss and capacitance measurements 411
7.2.1 The Schering bridge 412
7.2.2 Current comparator bridges 417
7.2.3 Loss measurement on complete equipment 420
7.2.4 Null detectors 421
7.3 Partial-discharge measurements 421
7.3.1 The basic PD test circuit 423
7.3.2 PD currents 427
7.3.3 PD measuring systems within the PD test circuit 429
7.3.4 Measuring systems for apparent charge 433
7.3.5 Sources and reduction of disturbances 448
7.3.6 Other PD quantities 450
7.3.7 Calibration of PD detectors in a complete test circuit 452

Contents ix
7.3.8 Digital PD instruments and measurements 453
References 456
Chapter 8 Overvoltages, testing procedures and insulation coordination 460
8.1 The lightning mechanism 460
8.1.1 Energy in lightning 464
8.1.2 Nature of danger 465
8.2 Simulated lightning surges for testing 466
8.3 Switching surge test voltage characteristics 468
8.4 Laboratory high-voltage testing procedures and statistical treatment of results
472
8.4.1 Dielectric stress–voltage stress 472
8.4.2 Insulation characteristics 473
8.4.3 Randomness of the appearance of discharge 473
8.4.4 Types of insulation 473
8.4.5 Types of stress used in high-voltage testing 473
8.4.6 Errors and confidence in results 479
8.4.7 Laboratory test procedures 479
8.4.8 Standard test procedures 484
8.4.9 Testing with power frequency voltage 484
8.4.10 Distribution of measured breakdown probabilities (confidence in
measured PV) 485
8.4.11 Confidence intervals in breakdown probability (in measured values)
487
8.5 Weighting of the measured breakdown probabilities 489
8.5.1 Fitting of the best fit normal distribution 489
8.6 Insulation coordination 492
8.6.1 Insulation level 492
8.6.2 Statistical approach to insulation coordination 495
8.6.3 Correlation between insulation and protection levels 498
8.7 Modern power systems protection devices 500
8.7.1 MOA – metal oxide arresters 500
References 507
Chapter 9 Design and testing of external insulation 509
9.1 Operation in a contaminated environment 509
9.2 Flashover mechanism of polluted insulators under a.c. and d.c. 510
9.2.1 Model for flashover of polluted insulators 511
9.3 Measurements and tests 512
9.3.1 Measurement of insulator dimensions 513

x Contents
9.3.2 Measurement of pollution severity 514
9.3.3 Contamination testing 517
9.3.4 Contamination procedure for clean fog testing 518
9.3.5 Clean fog test procedure 519
9.3.6 Fog characteristics 520
9.4 Mitigation of contamination flashover 520
9.4.1 Use of insulators with optimized shapes 520
9.4.2 Periodic cleaning 520
9.4.3 Grease coating 521
9.4.4 RTV coating 521
9.4.5 Resistive glaze insulators 521
9.4.6 Use of non-ceramic insulators 522
9.5 Design of insulators 522
9.5.1 Ceramic insulators 523
9.5.2 Polymeric insulators (NCI) 526
9.6 Testing and specifications 530
9.6.1 In-service inspection and failure modes 531
References 531
Index 533

Preface to Second Edition
The first edition as well as its forerunner of Kuffel and Abdullah published in
1970 and their translations into Japanese and Chinese languages have enjoyed
wide international acceptance as basic textbooks in teaching senior under-
graduate and postgraduate courses in High-Voltage Engineering. Both texts
have also been extensively used by practising engineers engaged in the design
and operation of high-voltage equipment. Over the years the authors have
received numerous comments from the text’s users with helpful suggestions
for improvements. These have been incorporated in the present edition. Major
revisions and expansion of several chapters have been made to update the
continued progress and developments in high-voltage engineering over the
past two decades.
As in the previous edition, the principal objective of the current text is to
cover the fundamentals of high-voltage laboratory techniques, to provide an
understanding of high-voltage phenomena, and to present the basics of high-
voltage insulation design together with the analytical and modern numerical
tools available to high-voltage equipment designers.
Chapter 1 presents an introduction to high-voltage engineering including
the concepts of power transmission, voltage stress, and testing with various
types of voltage. Chapter 2 provides a description of the apparatus used in the
generation of a.c., d.c., and impulse voltages. These first two introductory
chapters have been reincorporated into the current revision with minor
changes.
Chapter 3 deals with the topic of high-voltage measurements. It has under-
gone major revisions in content to reflect the replacement of analogue instru-
mentation with digitally based instruments. Fundamental operating principles
of digital recorders used in high-voltage measurements are described, and the
characteristics of digital instrumentation appropriate for use in impulse testing
are explained.
Chapter 4 covers the application of numerical methods in electrical stress
calculations. It incorporates much of the contents of the previous text, but the
section on analogue methods has been replaced by a description of the more
current boundary element method.
Chapter 5 of the previous edition dealt with the breakdown of gaseous,
liquid, and solid insulation. In the new edition these topics are described in

xii Preface to Second Edition
two chapters. The new Chapter 5 covers the electrical breakdown of gases.
The breakdown of liquid and solid dielectrics is presented in Chapter 6 of the
current edition.
Chapter 7 of the new text represents an expansion of Chapter 6 of the
previous book. The additional areas covered comprise a short but fundamental
introduction to dielectric properties of materials, diagnostic test methods, and
non-destructive tests applicable also to on-site monitoring of power equipment.
The expanded scope is a reflection of the growing interest in and development
of on-site diagnostic testing techniques within the electrical power industry.
This area represents what is perhaps the most quickly evolving aspect of high-
voltage testing. The current drive towards deregulation of the power industry,
combined with the fact that much of the apparatus making up the world’s
electrical generation and delivery systems is ageing, has resulted in a pressing
need for the development of in-service or at least on-site test methods which
can be applied to define the state of various types of system assets. Assessment
of the remaining life of major assets and development of maintenance practices
optimized both from the technical and economic viewpoints have become
critical factors in the operation of today’s electric power systems. Chapter 7
gives an introduction and overview of the fundamental aspects of on-site test
methods with some practical examples illustrating current practices.
Chapter 8 is an expansion of Chapter 7 from the previous edition. However,
in addition to the topics of lightning phenomena, switching overvoltages and
insulation coordination, it covers statistically based laboratory impulse test
methods and gives an overview of metal oxide surge arresters. The statistical
impulse test methods described are basic tools used in the application of
insulation coordination concepts. As such, an understanding of these methods
leads to clearer understanding of the basis of insulation coordination. Similarly,
an understanding of the operation and application of metal oxide arresters is
an integral part of today’s insulation coordination techniques.
Chapter 9 describes the design, performance, application and testing of
outdoor insulators. Both ceramic and composite insulators are included.
Outdoor insulators represent one of the most critical components of
transmission and distribution systems. While there is significant experience
in the use of ceramic insulators, composite insulators represent a relatively
new and quickly evolving technology that offers a number of performance
advantages over the conventional ceramic alternative. Their use and
importance will continue to increase and therefore merits particular attention.
The authors are aware of the fact that many topics also relevant to the
fundamentals of high-voltage engineering have again not been treated. But
every textbook about this field will be a compromise between the limited
space available for the book and the depth of treatment for the selected topics.
The inclusion of more topics would reduce its depth of treatment, which should

Preface to Second Edition xiii
be good enough for fundamental understanding and should stimulate further
reading.
The authors would like to express their thanks to Professors Yuchang Qiu of
X’ian Jaotong University, Stan. Grzybowski of Mississippi State University,
Stephen Sebo of Ohio State University for their helpful suggestions in the
selection of new material, Ontario Power Technologies for providing help
in the preparation of the text and a number of illustrations and Mrs Shelly
Gerardin for her skilful efforts in scanning and editing the text of the first
edition. Our special thanks go to Professor Yuchang Qiu for his laborious
proof reading of the manuscript.
Finally we would like to express our personal gratitude to Mr Peter Kuffel
and Dr Waldemar Ziomek for their invaluable help in the process of continued
review and preparation of the final manuscript and illustrations.

Preface to First Edition
The need for an up-to-date textbook in High Voltage Engineering fundamentals
has been apparent for some time. The earlier text of Kuffel and Abdullah
published in 1970, although it had a wide circulation, was of somewhat limited
scope and has now become partly outdated.
In this book an attempt is made to cover the basics of high voltage laboratory
techniques and high voltage phenomena together with the principles governing
design of high voltage insulation.
Following the historical introduction the chapters 2 and 3 present a compre-
hensive and rigorous treatment of laboratory, high voltage generation and
measurement techniques and make extensive references to the various inter-
national standards.
Chapter 4 reviews methods used in controlling electric stresses and intro-
duces the reader to modern numerical methods and their applications in the
calculation of electric stresses in simple practical insulations.
Chapter 5 includes an extensive treatment of the subject of gas discharges
and the basic mechanisms of electrical breakdown of gaseous, liquid and solid
insulations.
Chapter 6 deals with modern techniques for discharge detection and
measurement. The final chapter gives an overview treatment of systems
overvoltages and insulation coordination.
It is hoped the text will fill the needs of senior undergraduate and grad-
uate students enrolled in high voltage engineering courses as well as junior
researchers engaged in the field of gas discharges. The in-depth treatment of
high voltage techniques should make the book particularly useful to designers
and operators of high voltage equipment and utility engineers.
The authors gratefully acknowledge Dr. M. M. Abdullah’s permission to
reproduce some material from the book High Voltage Engineering, Pergamon
Press, 1970.
E. KUFFEL, W.S. ZAENGAL
March 1984

Chapter 1
Introduction
1.1 Generation and transmission of electric energy
The potential benefits of electrical energy supplied to a number of consumers
from a common generating system were recognized shortly after the develop-
ment of the ‘dynamo’, commonly known as the generator.
The first public power station was put into service in 1882 in London
(Holborn). Soon a number of other public supplies for electricity followed
in other developed countries. The early systems produced direct ccurrent at
low-voltage, but their service was limited to highly localized areas and were
used mainly for electric lighting. The limitations of d.c. transmission at low-
voltage became readily apparent. By 1890 the art in the development of an a.c.
generator and transformer had been perfected to the point when a.c. supply
was becoming common, displacing the earlier d.c. system. The first major
a.c. power station was commissioned in 1890 at Deptford, supplying power
to central London over a distance of 28 miles at 10 000 V. From the earliest
‘electricity’ days it was realized that to make full use of economic genera-
tion the transmission network must be tailored to production with increased
interconnection for pooling of generation in an integrated system. In addition,
the potential development of hydroelectric power and the need to carry that
power over long distances to the centres of consumption were recognized.
Power transfer for large systems, whether in the context of interconnection
of large systems or bulk transfers, led engineers invariably to think in terms
of high system voltages. Figure 1.1 lists some of the major a.c. transmission
systems in chronological order of their installations, with tentative projections
to the end of this century.
The electric power (P) transmitted on an overhead a.c. line increases approx-
imately with the surge impedance loading or the square of the system’s oper-
ating voltage. Thus for a transmission line of surge impedance ZL (¾
D250 )
at an operating voltage V, the power transfer capability is approximately
P D V2
/ZL, which for an overhead a.c. system leads to the following results:
V kV 400 700 1000 1200 1500
P MW 640 2000 4000 5800 9000

2 High Voltage Engineering: Fundamentals
0
100
200
300
400
500
600
700
800
1885 1905 1925 1945 1965 1985 2005
Year of installation
A.C.
voltage
(kV)
1 1890 10 kV Deptford
2 1907 50 kV Stadtwerke München
3 1912 110 kV Lauchhammer − Riesa
4 1926 220 kV N. Pennsylvania
5 1936 287 kV Boulder Dam
6 1952 380 kV Harspränget − Hallsberg
7 1959 525 kV USSR
8 1965 735 kV Manicouagan − Montreal
9 2003 (Est) 500 kV Three Gorges (China)
1 2
3
4
5
6
7
9
8
Figure 1.1 Major a.c. systems in chronological order of their installations
The rapidly increasing transmission voltage level in recent decades is a
result of the growing demand for electrical energy, coupled with the devel-
opment of large hydroelectric power stations at sites far remote from centres
of industrial activity and the need to transmit the energy over long distances
to the centres. However, environmental concerns have imposed limitations
on system expansion resulting in the need to better utilize existing transmis-
sion systems. This has led to the development of Flexible A.C. Transmission
Systems (FACTS) which are based on newly developing high-power elec-
tronic devices such as GTOs and IGBTs. Examples of FACTS systems include
Thyristor Controlled Series Capacitors and STATCOMS. The FACTS devices
improve the utilization of a transmission system by increasing power transfer
capability.
Although the majority of the world’s electric transmission is carried on
a.c. systems, high-voltage direct current (HVDC) transmission by overhead
lines, submarine cables, and back-to-back installations provides an attractive
alternative for bulk power transfer. HVDC permits a higher power density
on a given right-of-way as compared to a.c. transmission and thus helps the
electric utilities in meeting the environmental requirements imposed on the
transmission of electric power. HVDC also provides an attractive technical
and economic solution for interconnecting asynchronous a.c. systems and for
bulk power transfer requiring long cables.

Introduction 3
Table 1.1 summarizes a number of major HVDC schemes in order of their
in-service dates. Figure 1.2 provides a graphic illustration of how HVDC trans-
mission voltages have developed. As seen in Figure 1.2 the prevailing d.c.
voltage for overhead line installations is 500 kV. This ‘settling’ of d.c. voltage
has come about based on technical performance, power transfer requirements,
environmental and economic considerations. Current trends indicate that d.c.
voltage levels will not increase dramatically in the near future.
0
100
200
300
400
500
600
700
1950 1960 1970 1980 1990 2000 2010
Year of installation
D.C. voltage
(kV)
Figure 1.2 Major d.c. systems in chronological order of their installations
1.2 Voltage stresses
Normal operating voltage does not severely stress the power system’s insula-
tion and only in special circumstances, for example under pollution conditions,
may operating voltages cause problems to external insulation. Nevertheless,
the operating voltage determines the dimensions of the insulation which forms
part of the generation, transmission and distribution equipment. The voltage
stresses on power systems arise from various overvoltages. These may be of
external or internal origin. External overvoltages are associated with lightning
discharges and are not dependent on the voltage of the system. As a result,
the importance of stresses produced by lightning decreases as the operating
voltage increases. Internal overvoltages are generated by changes in the oper-
ating conditions of the system such as switching operations, a fault on the
system or fluctuations in the load or generations.
Their magnitude depends on the rated voltage, the instance at which a
change in operating conditions occurs, the complexity of the system and so
on. Since the change in the system’s conditions is usually associated with
switching operations, these overvoltages are generally referred to as switching
overvoltages.

Table 1.1 Major HVDC schemes
Scheme Year Power D.C. Line or cable Location
(MW) voltage length (km)
(kv)
Gottland 1 1954 20 š100 96 Sweden
English Channel 1961 160 š100 64 England–
France
Pacific Intertie 1970 1440 š400 1362 USA
Nelson River 1 1972 1620 š450 892 Canada
Eel River 1972 320 2 ð 80 Back to Canada
back
Cabora Bassa 1978 1920 š533 1414 Mozambique–
South Africa
Nelson River 2 1978 900 š250 930 Canada
1985 1800 š500
Chateauguay 1984 1000 2 ð 140 Back to Canada
back
Itaipu 1 1984 200 š300 785 Brazil
1985 1575
1986 2383 š600
Intermountain 1986 1920 š500 784 USA
Cross Channel 1986 2000 2 ð š270 72 England–
France
Itaipu 2 1987 3150 š600 805 Brazil
Gezhouba–
Shanghai 1989 600 500 1000 China
1990 1200 š500
Fenno-Skan 1989 500 400 200 Finland–
Sweden
Rihand-Delhi 1991 1500 š500 910 India
Hydro Quebec–
New England 1990 2000 š450 1500 Canada–USA
Baltic Cable 1994 600 450 250 Sweden–
Germany
Tian Guang 2000 1800 š500 960 China
(est)
Three Gorges 2002 3000 š500 – China
(est)
Source: HVDC Projects Listing, D.C. Flexible A.C. Transmission Subcommittee of the IEEE Transmission and Distribution
Committee, Working Group on HVDC, and Bibliography and Records, January 1998 Issue.

Introduction 5
In designing the system’s insulation the two areas of specific importance
are:
(i) determination of the voltage stresses which the insulation must withstand,
and
(ii) determination of the response of the insulation when subjected to these
voltage stresses.
The balance between the electric stresses on the insulation and the dielectric
strength of this insulation falls within the framework of insulation coordination
and will be discussed in Chapter 8.
1.3 Testing voltages
Power systems equipment must withstand not only the rated voltage (Vm),
which corresponds to the highest voltage of a particular system, but also
overvoltages. Accordingly, it is necessary to test h.v. equipment during its
development stage and prior to commissioning. The magnitude and type of
test voltage varies with the rated voltage of a particular apparatus. The stan-
dard methods of measurement of high-voltage and the basic techniques for
application to all types of apparatus for alternating voltages, direct voltages,
switching impulse voltages and lightning impulse voltages are laid down in
the relevant national and international standards.
1.3.1 Testing with power frequency voltages
To assess the ability of the apparatus’s insulation withstand under the system’s
power frequency voltage the apparatus is subjected to the 1-minute test under
50 Hz or 60 Hz depending upon the country. The test voltage is set at a level
higher than the expected working voltage in order to be able to simulate
the stresses likely to be encountered over the years of service. For indoor
installations the equipment tests are carried out under dry conditions only. For
outdoor equipment tests may be required under conditions of standard rain as
prescribed in the appropriate standards.
1.3.2 Testing with lightning impulse voltages
Lightning strokes terminating on transmission lines will induce steep rising
voltages in the line and set up travelling waves along the line and may
damage the system’s insulation. The magnitude of these overvoltages may
reach several thousand kilovolts, depending upon the insulation. Exhaustive
measurements and long experience have shown that lightning overvoltages are
characterized by short front duration, ranging from a fraction of a microsecond

to several tens of microseconds and then slowly decreasing to zero. The stan-
dard impulse voltage has been accepted as an aperiodic impulse that reaches
its peak value in 1.2 µsec and then decreases slowly (in about 50 µsec) to half
its peak value. Full details of the waveshape of the standard impulse voltage
together with the permitted tolerances are presented in Chapter 2, and the
prescribed test procedures are discussed in Chapter 8.
In addition to testing equipment, impulse voltages are extensively used in
research laboratories in the fundamental studies of electrical discharge mech-
anisms, notably when the time to breakdown is of interest.
1.3.3 Testing with switching impulses
Transient overvoltages accompanying sudden changes in the state of power
systems, e.g. switching operations or faults, are known as switching impulse
voltages. It has become generally recognized that switching impulse volt-
ages are usually the dominant factor affecting the design of insulation in h.v.
power systems for rated voltages of about 300 kV and above. Accordingly,
the various international standards recommend that equipment designed for
voltages above 300 kV be tested for switching impulses. Although the wave-
shape of switching overvoltages occurring in the system may vary widely,
experience has shown that for flashover distances in atmospheric air of prac-
tical interest the lowest withstand values are obtained with surges with front
times between 100 and 300 µsec. Hence, the recommended switching surge
voltage has been designated to have a front time of about 250 µsec and half-
value time of 2500 µsec. For GIS (gas-insulated switchgear) on-site testing,
oscillating switching impulse voltages are recommended for obtaining higher
efficiency of the impulse voltage generator Full details relating to generation,
measurements and test procedures in testing with switching surge voltages
will be found in Chapters 2, 3 and 8.
1.3.4 D.C. voltages
In the past d.c. voltages have been chiefly used for purely scientific research
work. Industrial applications were mainly limited to testing cables with rela-
tively large capacitance, which take a very large current when tested with a.c.
voltages, and in testing insulations in which internal discharges may lead to
degradation of the insulation under testing conditions. In recent years, with
the rapidly growing interest in HVDC transmission, an increasing number of
industrial laboratories are being equipped with sources for producing d.c. high
voltages. Because of the diversity in the application of d.c. high voltages,
ranging from basic physics experiments to industrial applications, the require-
ments on the output voltage will vary accordingly. Detailed description of the
various main types of HVDC generators is given in Chapter 2.

Introduction 7
1.3.5 Testing with very low-frequency voltage
In the earlier years when electric power distribution systems used mainly
paper-insulated lead covered cables (PILC) on-site testing specifications called
for tests under d.c. voltages. Typically the tests were carried out at 4–4.5V0.
The tests helped to isolate defective cables without further damaging good
cable insulation. With the widespread use of extruded insulation cables of
higher dielectric strength, the test voltage levels were increased to 5–8V0. In
the 1970s premature failures of extruded dielectric cables factory tested under
d.c. voltage at specified levels were noted1
. Hence on-site testing of cables
under very low frequency (VLF) of ¾0.1 Hz has been adopted. The subject
has been recently reviewed1,2
.
References
1. Working Group 21.09. After-laying tests on high voltage extruded insulation cable systems,
Electra, No. 173 (1997), pp. 31–41.
2. G.S. Eager et al. High voltage VLF testing of power cables, IEEE Trans Power Delivery, 12,
No. 2 (1997), pp. 565–570.

Chapter 2
Generation of high voltages
A fundamental knowledge about generators and circuits which are in use for
the generation of high voltages belongs to the background of work on h.v.
technology.
Generally commercially available h.v. generators are applied in routine
testing laboratories; they are used for testing equipment such as transformers,
bushings, cables, capacitors, switchgear, etc. The tests should confirm the effi-
ciency and reliability of the products and therefore the h.v. testing equipment
is required to study the insulation behaviour under all conditions which the
apparatus is likely to encounter. The amplitudes and types of the test voltages,
which are always higher than the normal or rated voltages of the apparatus
under test, are in general prescribed by national or international standards or
recommendations, and therefore there is not much freedom in the selection of
the h.v. testing equipment. Quite often, however, routine testing laboratories
are also used for the development of new products. Then even higher volt-
ages might be necessary to determine the factor of safety over the prospective
working conditions and to ensure that the working margin is neither too high
nor too low. Most of the h.v. generator circuits can be changed to increase
the output voltage levels, if the original circuit was properly designed. There-
fore, even the selection of routine testing equipment should always consider
a future extension of the testing capabilities.
The work carried out in research laboratories varies considerably from one
establishment to another, and the type of equipment needed varies accordingly.
As there are always some interactions between the h.v. generating circuits used
and the test results, the layout of these circuits has to be done very carefully.
The classes of tests may differ from the routine tests, and therefore specially
designed circuits are often necessary for such laboratories. The knowledge
about some fundamental circuits treated in this chapter will also support the
development of new test circuits.
Finally, high voltages are used in many branches of natural sciences or other
technical applications. The generating circuits are often the same or similar
to those treated in the following sections. It is not the aim, however, of this
introductory text to treat the broad variations of possible circuits, due to space
limitation. Not taken into account are also the differing problems of electrical
power generation and transmission with high voltages of a.c. or d.c., or the

Generation of high voltages 9
pure testing technique of h.v. equipment, the procedures of which may be
found in relevant standards of the individual equipment. Power generation
and transmission problems are treated in many modern books, some of which
are listed within the bibliography of an earlier report.1Ł
This chapter discusses the generation of the following main classes of volt-
ages: direct voltages, alternating voltages, and transient voltages.
2.1 Direct voltages
In h.v. technology direct voltages are mainly used for pure scientific research
work and for testing equipment related to HVDC transmission systems. There
is still a main application in tests on HVAC power cables of long length, as
the large capacitance of those cables would take too large a current if tested
with a.c. voltages (see, however, 2.2.2: Series resonant circuits). Although
such d.c. tests on a.c. cables are more economical and convenient, the validity
of this test suffers from the experimentally obtained stress distribution within
the insulating material, which may considerably be different from the normal
working conditions where the cable is transmitting power at low-frequency
alternating voltages. For the testing of polyethylene h.v. cables, in use now
for some time, d.c. tests are no longer used, as such tests may not confirm the
quality of the insulation.50
High d.c. voltages are even more extensively used in applied physics
(accelerators, electron microscopy, etc.), electromedical equipment (X-rays),
industrial applications (precipitation and filtering of exhaust gases in thermal
power stations and the cement industry; electrostatic painting and powder
coating, etc.), or communications electronics (TV, broadcasting stations).
Therefore, the requirements on voltage shape, voltage level, and current rating,
short- or long-term stability for every HVDC generating system may differ
strongly from each other. With the knowledge of the fundamental generating
principles it will be possible, however, to select proper circuits for a special
application.
In the International Standard IEC 60-12
or IEEE Standard. 4-19953
the
value of a direct test voltage is defined by its arithmetic mean value, which
will be designated as V . Therefore, this value may be derived from
V D
1
T
T
0
Vt dt. 2.1
where T equals a certain period of time if the voltage Vt is not constant, but
periodically oscillating with a frequency of f D 1/T. Test voltages as applied
to test objects then deviate periodically from the mean value. This means that
Ł Superscript numbers are to References at the end of the chapter.

a ripple is present. The amplitude of the ripple, υV, is defined as half the
difference between the maximum and minimum values, or
υV D 0.5Vmax Vmin. 2.2
The ripple factor is the ratio of the ripple amplitude to the arithmetic mean
value, or υV/V. For test voltages this ripple factor should not exceed 3 per
cent unless otherwise specified by the appropriate apparatus standard or be
necessary for fundamental investigations.
The d.c. voltages are generally obtained by means of rectifying circuits
applied to a.c. voltages or by electrostatic generation. A treatment of the
generation principles according to this subdivision is appropriate.
2.1.1 A.C. to D.C. conversion
The rectification of alternating currents is the most efficient means of obtaining
HVDC supplies. Although all circuits in use have been known for a long time,
the cheap production and availability of manifold solid state rectifiers has
facilitated the production and application of these circuits fundamentally. Since
some decades, there is no longer a need to employ valves, hot cathode gas-
filled valves, mercury pool or corona rectifiers, or even mechanical rectifiers
within the circuits, for which the auxiliary systems for cathode heating, etc.,
have always aggravated their application. The state of the art of such earlier
circuits may be found in the work of Craggs and Meek,4
which was written
in 1954. All rectifier diodes used now adopt the Si type, and although the
peak reverse voltage is limited to less than about 2500 V, rectifying diode
units up to tens and hundreds of kVs can be made by series connections if
appropriate means are applied to provide equal voltage distribution during the
non-conducting period. One may treat and simulate, therefore, a rectifier within
the circuits – independently of the voltage levels – simply by the common
symbol for a diode.
The theory of rectifier circuits for low voltages and high power output is
discussed in many standard handbooks. Having the generation of high d.c.
voltages in mind, we will thus restrict the treatment mainly to single-phase
a.c. systems providing a high ratio of d.c. output to a.c. input voltage. As,
however, the power or d.c. output is always limited by this ratio, and because
very simple rectifier circuits are in use, we will treat only selected examples
of the many available circuits.
Simple rectifier circuits
For a clear understanding of all a.c. to d.c. conversion circuits the single-phase
half-wave rectifier with voltage smoothing is of basic interest (Fig. 2.1(a)).
If we neglect the leakage reactance of the transformer and the small internal

(a)
(b)
V~(t)
V~(t)
V (t)
t
a.T
V max
V min
D
C
h.t.
transformer
V
c
2.d V
a
iL (t)
RL (load)
i (t)
i (t)
T = 1/f
Figure 2.1 Single-phase half-wave rectifier with reservoir capacitance C.
(a) Circuit. (b) Voltages and currents with load RL
impedance of the diodes during conduction – and this will be done throughout
unless otherwise stated – the reservoir or smoothing capacitor C is charged to
the maximum voltage CVmax of the a.c. voltage V¾t of the h.t. transformer,
when D conducts. This is the case as long as V V¾t for the polarity of D
assumed. If I D 0, i.e. the output load being zero RL D 1, the d.c. voltage
across C remains constant CVmax, whereas V¾t oscillates between šVmax.
The diode D must be dimensioned, therefore, to withstand a peak reverse
voltage of 2Vmax.
The output voltage V does not remain any more constant if the circuit
is loaded. During one period, T D 1/f of the a.c. voltage a charge Q is
transferred to the load RL, which is represented as
Q D

T
iLt dt D
1
RL

T
Vt dt D IT D
I
f
. 2.3

I is therefore the mean value of the d.c. output iLt, and Vt the d.c. voltage
which includes a ripple as shown in Fig. 2.1(b). If we introduce the ripple
factor υV from eqn (2.2), we may easily see that Vt now varies between
Vmax ½ Vt ½ Vmin; Vmin D Vmax 2υV. 2.4
The charge Q is also supplied from the transformer within the short conduction
time tc D ˛T of the diode D during each cycle. Therefore, Q equals also to
Q D

˛T
it dt D

T
iLt dt. 2.5
As ˛T − T, the transformer and diode current it is pulsed as shown idealized
in Fig. 2.l(b) and is of much bigger amplitudes than the direct current iL
¾
D I.
The ripple υV could be calculated exactly for this circuit based upon the expo-
nential decay of Vt during the discharge period T1 ˛. As, however, for
practical circuits the neglected voltage drops within transformer and rectifiers
must be taken into account, and such calculations are found elsewhere,3
we
may assume that ˛ D 0. Then υV is easily found from the charge Q transferred
to the load, and therefore
Q D 2υVC D IT; υV D
IT
2C
D
I
2fC
. 2.6
This relation shows the interaction between the ripple, the load current and
circuit parameter design values f and C. As, according to eqn (2.4), the mean
output voltage will also be influenced by υV, even with a constant a.c. voltage
V¾t and a lossless rectifier D, no load-independent output voltage can be
reached. The product fC is therefore an important design factor.
For h.v. test circuits, a sudden voltage breakdown at the load RL ! 0
must always be taken into account. Whenever possible, the rectifiers should
be able to carry either the excessive currents, which can be limited by fast,
electronically controlled switching devices at the transformer input, or they
can be protected by an additional resistance inserted in the h.t. circuit. The
last method, however, increases the internal voltage drop.
Half-wave rectifier circuits have been built up to voltages in the megavolt
range, in general by extending an existing h.v. testing transformer to a d.c.
current supply. The largest unit has been presented by Prinz,5
who used a 1.2-
MV cascaded transformer and 60-mA selenium-type solid state rectifiers with
an overall reverse voltage of 3.4 MV for the circuit. The voltage distribution
of this rectifier, which is about 12 m in length, is controlled by sectionalized
parallel capacitor units, which are small in capacitance value in comparison
with the smoothing capacitor C (see Fig. 2.14). The size of such circuits,
however, would be unnecessarily large for pure d.c. supplies.
The other disadvantage of the single-phase half-wave rectifier concerns the
possible saturation of the h.v. transformer, if the amplitude of the direct current

is comparable with the nominal alternating current of the transformer. The
biphase half-wave (or single-phase full-wave) rectifier as shown in Fig. 2.2
overcomes this disadvantage, but it does not change the fundamental effi-
ciency, considering that two h.v. windings of the transformer are now avail-
able. With reference to the frequency f during one cycle, now each of the
diodes D1 and D2 is conducting for one half-cycle with a time delay of
T/2. The ripple factor according to eqn (2.6) is therefore halved. It should
be mentioned that the real ripple will also be increased if both voltages V1¾
and V2¾ are not exactly equal. If V2 max would be smaller than V1 max 2υV
or Vmin, this h.v. winding would not charge the capacitance C. The same effect
holds true for multiphase rectifiers, which are not treated here.
V1∼(t)
V2∼(t)
D1
D2
h.t.
transformer
C V RL
Figure 2.2 Biphase half-wave rectifier circuit with smoothing capacitor C
Thus single-phase full-wave circuits can only be used for h.v. applications
if the h.t. winding of the transformer can be earthed at its midpoint and if the
d.c. output is single-ended grounded. More commonly used are single-phase
voltage doublers, a circuit of which is contained in the voltage multiplier or
d.c. cascade of Fig. 2.6, see stage 1. Although in such a circuit grounding
of the h.v. winding is also not possible, if asymmetrical d.c. voltages are
produced, the potential of this winding is fixed. Therefore, there is no danger
due to transients followed by voltage breakdowns.
Cascade circuits
The demands from physicists for very high d.c. voltages forced the improve-
ment of rectifying circuits quite early. It is obvious that every multiplier circuit
in which transformers, rectifiers and capacitor units have only to withstand a
fraction of the total output voltage will have great advantages. Today there are
many standard cascade circuits available for the conversion of modest a.c. to
high d.c. voltages. However, only few basic circuits will be treated.

In 1920 Greinacher, a young physicist, published a circuit6
which was
improved in 1932 by Cockcroft and Walton to produce high-energy positive
ions.7
The interesting and even exciting development stages of those circuits
have been discussed by Craggs and Meek.4
To demonstrate the principle
only, an n-stage single-phase cascade circuit of the ‘Cockcroft–Walton type’,
shown in Fig. 2.3, will be presented.
HV output open-circuited: I D 0. The portion 0 n0
Vt is a half-wave
rectifier circuit in which C0
n charges up to a voltage of CVmax if Vt has
reached the lowest potential, Vmax. If Cn is still uncharged, the rectifier Dn
conducts as soon as Vt increases. As the potential of point n0
swings up
to CV2 max during the period T D 1/f, point n attains further on a steady
potential of C2Vmax if Vt has reached the highest potential of CVmax. The
part n0
n 0 is therefore a half-wave rectifier, in which the voltage across
D0
n can be assumed to be the a.c. voltage source. The current through Dn that
∼
V(t); Vmax
C′n Cn
D′n
Dn
C′n−1 Cn−1
D′n−1
Dn−1
n′ n
4′
3′
D3
D2
D1
C1
C2
C3
2′
1′
1
(n−1)′ (n−1)
C3
′
D3
′
C2
′
D2
′
C1
′
D1
′
2
3
4
0
I
H.V. output
(a)
Figure 2.3 (a) Cascade circuit according to Cockroft–Walton or
Greinacher. (b) Waveform of potentials at the nodes, no load

1 = H.V. output
1′
2′
V
0
=
n
.
2
V
max
(n −1)
Stages(n−2) to 3
(n −1)′
2
n
n′
2Vmax
V(t )
t
0
t1
t2
D1...Dn
Conducting
D1
′...Dn
′
Conducting
Vmax
(b)
Figure 2.3 (continued)
charged the capacitor Cn was not provided by D0
n, but from Vt and C0
n. We
assumed, therefore, that C0
n was not discharged, which is not correct. As we
will take this into consideration for the loaded circuit, we can also assume
that the voltage across Cn is not reduced if the potential n0
oscillates between
zero and C2Vmax. If the potential of n0
, however, is zero, the capacitor C0
n 1
is also charged to the potential of n, i.e. to a voltage of C2Vmax. The next
voltage oscillation of Vt from Vmax to CVmax will force the diode Dn 1
to conduct, so that also Cn 1 will be charged to a voltage of C2Vmax.
In Fig. 2.3(b) the steady state potentials at all nodes of the circuit are
sketched for the circuit for zero load conditions. From this it can be seen, that:
ž the potentials at the nodes 10
, 20
. . . n0
are oscillating due to the voltage
oscillation of Vt;
ž the potentials at the nodes 1, 2 . . . n remain constant with reference to
ground potential;
ž the voltages across all capacitors are of d.c. type, the magnitude of which is
2Vmax across each capacitor stage, except the capacitor C0
n which is stressed
with Vmax only;

ž every rectifier D1, D0
1 . . . Dn, D0
n is stressed with 2Vmax or twice a.c. peak
voltage; and
ž the h.v. output will reach a maximum voltage of 2nVmax.
Therefore, the use of several stages arranged in this manner enables very
high voltages to be obtained. The equal stress of the elements used is very
convenient and promotes a modular design of such generators. The number of
stages, however, is strongly limited by the current due to any load. This can
only be demonstrated by calculations, even if ideal rectifiers, capacitors and
an ideal a.c. voltage source are assumed.
Finally it should be mentioned that the lowest stage n of the cascade circuit
(Fig. 2.3(a)) is the Cockcroft–Walton voltage doubler. The a.c. voltage source
Vt is usually provided by an h.t. transformer, if every stage is built for
high voltages, typically up to about 300 kV. This source is always symmet-
rically loaded, as current is withdrawn during each half-cycle (t1 and t2 in
Fig. 2.3(b)). The voltage waveform does not have to be sinusoidal: every
symmetrical waveform with equal positive and negative peak values will give
good performance. As often high-frequency input voltages are used, this hint
is worth remembering.
H.V. output loaded: I 0. If the generator supplies any load current I, the
output voltage will never reach the value 2nVmax as shown in Fig. 2.3(b).
There will also be a ripple on the voltage, and therefore we have to deal with
two quantities: the voltage drop V0 and the peak-to-peak ripple 2υV. The
sketch in Fig. 2.4 shows the shape of the output voltage and the definitions of
2n Vmax
(no load)
V0 max
V0 (t) with load
2 δ V
∆V0
+Vmax
t1
0
t2
V(t)
T = 1/f
t
Figure 2.4 Loaded cascade circuit, definitions of voltage drop V0 and
ripple υV

V0 and 2υV. The time instants t1 and t2 are in agreement with Fig. 2.3(b).
Therefore, the peak value of Vo is reached at t1, if Vt was at CVmax and the
rectifiers D1 . . . Dn just stopped to transfer charge to the ‘smoothing column’
C1 . . . Cn. After that the current I continuously discharges the column, inter-
rupted by a sudden voltage drop shortly before t2: this sudden voltage drop
is due to the conduction period of the diodes D0
1 . . . D0
n, during which the
‘oscillating column’ C0
1 . . . C0
n is charged.
Now let a charge q be transferred to the load per cycle, which is obviously
q D I/f D IT. This charge comes from the smoothing column, the series
connection of C1 . . . Cn. If no charge would be transferred during T from this
stack via D0
1 . . . D0
n to the oscillating column, the peak-to-peak ripple would
merely be
2υV D IT
n

iD1
1/Ci.
As, however, just before the time instant t2 every diode D0
1 . . . D0
n transfers
the same charge q, and each of these charges discharges all capacitors on the
smoothing column between the relevant node and ground potential, the total
ripple will be
υV D
1
2f

1
C1
C
2
C2
C
3
C3
C . . .
n
Cn

. 2.7
Thus in a cascade multiplier the lowest capacitors are responsible for most
ripple and it would be desirable to increase the capacitance in the lower
stages. This is, however, very inconvenient for h.v. cascades, as a voltage
breakdown at the load would completely overstress the smaller capacitors
within the column. Therefore, equal capacitance values are usually provided,
and with C D C1 D C2 . . . Cn, eqn (2.7) is
υV D
I
fC
ð
nn C 1
4
. 2.7a
To calculate the total voltage drop V0, we will first consider the stage n.
Although the capacitor C0
n at time t1 will be charged up to the full voltage
Vmax, if ideal rectifiers and no voltage drop within the a.c.-source are assumed,
the capacitor Cn will only be charged to a voltage
Vcn
max D 2Vmax
nq
C0
n
D 2Vmax Vn
as Cn has lost a total charge of nq during a full cycle before and C0
n has to
replace this lost charge. At time instant t2, Cn transfers the charge q to C0
n 1

equal amounts q to C0
n2, . . . C0
2, C0
1 and q to the load during T. Therefore,
C0
n 1 can only be charged up to a maximum voltage of
VC0
n 1
max D

2Vmax
nq
C0
n

nq
Cn
D VCn
max
nq
Cn
.
As the capacitor C0
n 1 will be charged up to this voltage minus n 1q/c0
n 1,
etc., one can easily form the general rules for the total voltage drop at the
smoothing stack C1 . . . Cn
If all the capacitors within the cascade circuit are equal or
C1 D C0
1 D C2 D C0
2 D . . . Cn D C0
n D C,
then the voltage drops across the individual stages are
Vn D q/cn;
Vn 1 D q/c[2n C n 1];
.
.
.
V1 D q/c[2n C 2n 1 C 2n 2 C . . . C 2 ð 2 C 1]. 2.8
By summation, and with q D I/f, we find
V0 D
1
fC

2n3
3
C
n2
2

n
6

. 2.9
Thus the lowest capacitors are most responsible for the total V0 as is the case
of the ripple, eqn (2.7). However, only a doubling of C0
n is convenient, since
this capacitor has to withstand only half the voltage of the other capacitors;
namely Vmax. Therefore, Vn decreases by an amount of 0.5 nq/c, which
reduces V of every stage by the same amount, thus n times. Hence,
V0 D
1
fC

2n3
3

n
6

. 2.10
For this case and n ½ 4 we may neglect the linear term and therefore approx-
imate the maximum output voltage by
V0 max
¾
D 2nVmax
I
fC
ð
2n3
3
. 2.11
For a given number of stages, this maximum voltage or also the mean value
V0 D V0 max υV will decrease linearly with the load current I at constant

frequency, which is obvious. For a given load, however, V0 may rise initially
with the number of stages n, but reaches an optimum value and even decreases
if n is too large. Thus – with respect to constant values of I Vmax, f and
C – the highest value can be reached with the ‘optimum’ number of stages,
obtained by differentiating eqn (2.11) with respect to n. Then
nopt D

VmaxfC
I
2.12
For a generator with Vmax D 100 kV, f D 500 Hz, C D 7 µF and I D 500 mA,
nopt D 10. It is, however, not desirable to use the optimum number of stages,
as then V0 max is reduced to 2/3 of its maximum value 2nVmax. Also the
voltage variations for varying loads will increase too much.
The application of this circuit to high power output, which means high prod-
ucts of IV0 is also limited by eqns (2.9) and (2.11), in which again the large
influence of the product fC can be seen. An increase of supply frequency
is in general more economical than an increase of the capacitance values;
small values of C also provide a d.c. supply with limited stored energy, which
might be an essential design factor, i.e. for breakdown investigations on insu-
lating materials. A further advantage is related to regulation systems, which
are always necessary if a stable and constant output voltage V0 is required.
Regulation can be achieved by a measurement of V0 with suitable voltage
dividers (see Chapter 3, section 3.6.4) within a closed-loop regulation system,
which controls the a.c. supply voltage Vt. For fast response, high supply
frequencies and small stored energy are prerequisites.
For tall constructions in the MV range, the circuit of Fig. 2.3(a) does not
comprise all circuit elements which are influencing the real working condi-
tions. There are not only the impedances of the diodes and the supply trans-
former which have to be taken into consideration; stray capacitances between
the two capacitor columns and capacitor elements to ground form a much
more complex network. There are also improved circuits available by adding
one or two additional ‘oscillating’ columns which charge the same smoothing
stack. This additional column can be fed by phase-shifted a.c. voltages, by
which the ripple and voltage drop can further be reduced. For more details
see reference 8.
Cascade generators of Cockcroft–Walton type are used and manufactured
today worldwide. More information about possible constructions can be found
in the literature9,10
or in company brochures. The d.c. voltages produced with
this circuit may range from some 10 kV up to more than 2 MV, with current
ratings from some 10 µA up to some 100 mA. Supply frequencies of 50/60 Hz
are heavily limiting the efficiency, and therefore higher frequencies up to about
1000 Hz (produced by single-phase alternators) or some 10 kHz (produced by
electronic circuits) are dominating.

Also for this kind of generators, voltage reversal can be performed by
a reversal of all diodes. For some special tests on components as used for
HVDC transmission, a fast reversal of the d.c. voltages is necessary. This can
be done with special mechanical arrangements of the diodes, as published by
W. Hauschild et al.50,51
Figure 2.5 shows such a unit for a d.c. voltage up to
Figure 2.5 A Cockroft–Walton d.c. generator for voltages up to
900 kV/10 mA with fast polarity reversal at ETH Zurich (courtesy HIGH
VOLT, Dresden, Germany)

900 kV. Here, also the general structure of the Cockroft–Walton circuit can
be identified.
Voltage multiplier with cascaded transformers
The multiple charge transfer within the cascade circuit of the Cock-
croft–Walton type demonstrated the limitations in d.c. power output. This
disadvantage can be reduced if single- or full-wave rectifier systems, each
having its own a.c. power source, are connected in series at the d.c. output
only. Then the a.c. potentials remain more or less at d.c. potentials. Although
there are many modifications possible, the principle that will be demonstrated
here is based upon a very common circuit, which is shown in Fig. 2.6. Every
transformer per stage consists of an l.v. primary (1), h.v. secondary (2), and l.v.
tertiary winding (3), the last of which excites the primary winding of the next
upper stage. As none of the h.v. secondary windings is on ground potential,
a d.c. voltage insulation within each transformer (T1, T2, etc.) is necessary,
which can be subdivided within the transformers. Every h.v. winding feeds
two half-wave rectifiers, which have been explained before. Although there
Further states
(up to n)
Stage 2
Stage 1
3
2
1 n.2V
1
∼
3
2 T1
T2
V
Figure 2.6 D.C. cascade circuit with cascaded transformers

are limitations as far as the number of stages is concerned, as the lower trans-
formers have to supply the energy for the upper ones, this circuit, excited with
power frequency, provides an economical d.c. power supply for h.v. testing
purposes with moderate ripple factors and high power capabilities.
The ‘Engetron’ circuit (Deltatron)
A very sophisticated cascade transformer HVDC generator circuit was
described by Enge in a US Patent.11
Although such generators might be limited
in the power output up to about 1 MV and some milliamperes, the very small
ripple factors, high stability, fast regulation and small stored energies are
essential capabilities of this circuit.
The circuit is shown in Fig. 2.7. It consists primarily of a series connection
of transformers, which do not have any iron core. These transformers are
coupled by series capacitors Cs which compensate most of the stray inductance
Termination
HV d.c. − output
Further stages
Module
Stage 2
Stage 1
Cockcroft−Walton
multipliers
Cs
Cp
∼
Oscillator
(50 . . . 100 kc/s)
Figure 2.7 The ‘Engetron’ or Deltatron principle

of the transformers. In addition to this, to every primary and secondary winding
a capacitor Cp is connected in parallel, which provides an overcompensation
of the magnetizing currents. The whole chain of cascaded transformers is
loaded by a terminating resistor; thus the network acts similarly to a terminated
transmission line along which the a.c. voltage remains nearly constant and
has a phase shift between input (high-frequency power supply) and output
(termination). The transformers, therefore, are not used to increase the a.c.
voltage.
It is now possible to connect to every stage indicated as usual
Cockcroft–Walton cascade circuit, with only a small input voltage (some
kV), producing, however, output voltages of some 10 kV per stage. The
storage columns of these Cockcroft–Walton cascades are then directly series
connected, providing the high d.c. output voltage for the whole cascade
transformer HVDC generator unit. Typically up to about 25 stages can be used,
every stage being modular constructed. As these modules are quite small, they
can be stacked in a cylindrical unit which is then insulated by SF6. Not shown
in Fig. 2.7 is the voltage regulation system, which is controlled by a parallel
mixed R-C voltage divider and a high-frequency oscillator, whose frequency
ranges from 50 to 100 kHz. As for these high frequencies the capacitors within
the Cockcroft–Walton circuits can be very small, and the energy stored is
accordingly low; regulation due to load variations or power voltage supply
variations is very fast (response time typically about 1 msec). The small ripple
factor is not only provided by the storage capacitor, but also by the phase-
shifted input voltages of the cascade circuits. Amongst the disadvantages is
the procedure to change polarity, as all modules have to be reversed.
Summary and concluding remarks to 2.1.1
It has been shown that all a.c. to d.c. voltage conversion systems could be
classed between the circuits of Figs 2.1 and 2.3, if single-phase a.c. voltages
are converted into d.c. voltages. A high d.c. to a.c. voltage ratio can only
be gained with a high product of a.c. frequency and energy stored in the
smoothing capacitors, as they have to store electrical energy within each cycle,
during which the a.c. power is oscillating. If, therefore, the d.c. output should
be very stable and continuous, a high product (fC) is necessary. A reduction
of stored energy is possible if the a.c. power is not only provided at ground
potential, this means if a.c. power is injected into the circuits at different
potential levels. The savings, therefore, can be made either on the a.c. or d.c.
side. The large variety of possible circuits and technical expenditure is always
strongly related to the ‘quality’ of the d.c. power needed, this means to the
stability and the ripple of the output voltage.

2.1.2 Electrostatic generators
Electrostatic generators convert mechanical energy directly into electrical
energy. In contrast to electromagnetic energy conversion, however,
electrical charges are moved in this generator against the force of electrical
fields, thus gaining higher potential energies and consuming mechanical
energy. All historical electrostatic machines, such as the Kelvin water dropper
or the Wimshurst machine, are therefore forerunners of modern generators of
this type. A review of earlier machines may be found in reference 12.
Besides successful developments of ‘dust generators’ presented by Pauthe-
nier et al.13
the real breakthrough in the generation of high and ultra-high
d.c. voltages is linked with Van de Graaff, who in 1931 succeeded with the
development of electrostatic belt-driven generators.14
These generators are
in common use today in nuclear physics research laboratories. Figure 2.8
demonstrates the principle of operation, which is described in more detail
in reference 4. Charge is sprayed onto an insulating moving belt by means
of corona discharge points (or direct contact) which are at some 10 kV from
earth potential. The belt, the width of which may vary widely (some cm up to
metres), is driven at about 15–30 m/sec by means of a motor and the charge
is conveyed to the upper end where it is removed from the belt by discharging
points connected to the inside of an insulated metal electrode through which
Upper spray
points
Collector
Motor driven
pulley
Controllable spray
voltage
Lower spray
points
Insulating belt
H.V. terminal
Upper pulley
(insulated from earth)
Figure 2.8 Outline of electrostatic belt-driven generator

the belt passes. The entire equipment is usually enclosed in an earthed metal
tank filled with insulating compressed gases of good performance such as air,
mixtures of N2 –CO2, Freon 12 (CCl2, F2) or SF6. For simple applications the
metal tank can be omitted, so that the insulation is provided by atmospheric
air only.
The potential of the h.v. terminal at any instant is V D Q/C above earth,
where Q is the charge stored and C is the capacitance of the h.v. electrode to
ground. The potential of the terminal rises at a rate given by dV/dt D I/C,
where
I D O
Sb 2.13
is the net charging current to the terminal. In this equation, O
S is the charge
density at the belt in coulombs/m2
, b its width in m, and v the belt speed
in m/sec. In practice, dV/dt may reach a value of 1 MV/sec and it appears
that the final potential of the h.v. electrode would be infinite in the absence
of any mechanism of charge loss. Equilibrium is in practice established at a
potential such that the charging current equals the discharge current which
includes load currents – also due to voltage dividers, leakage currents and
corona losses, if present – and by voltage regulating systems which are based
on voltage measurement and the controllable spray unit.
While the h.v. terminal electrode can easily be shaped in such a way that
local discharges are eliminated from its surface, the field distribution between
this electrode and earth along the fast moving belt is of greatest importance.
The belt, therefore, is placed within properly shaped field grading rings, the
grading of which is provided by resistors and sometimes additional corona
discharge elements.
The lower spray unit, shown in Fig. 2.8, may consist of a number of needles
connected to the controllable d.c. source so that the discharge between the
points and the belt is maintained. The collector needle system is placed near
the point where the belt enters the h.v. terminal.
A self-inducing arrangement is commonly used for spraying on the down-
going belt charges of polarity opposite to that of the h.v. terminal. The rate of
charging of the terminal, for a given speed of the belt, is therefore doubled. To
obtain a self-charging system, the upper pulley is connected to the collector
needle and is therefore maintained at a potential higher than that of the h.v.
terminal. The device includes another system of points (shown as upper spray
points in Fig. 2.8) which is connected to the inside of the h.v. terminal and is
directed towards the pulley at the position shown. As the pulley is at a higher
positive potential, the negative charges of the corona at the upper spray points
are collected by the belt. This neutralizes any remaining positive charges on
the belt and leaves any excess negative charges which travel down with it and
are neutralized at the lower spray points.

For a rough estimation of the current I which can be provided by such
generators, we may assume a homogeneous electrical field E normal to the
belt running between the lower spray points and the grounded lower pulley.
As E D D/ε0 D O
S/ε0, D being the flux density, ε0 the permittivity and O
S
the charge density according to eqn (3.13) deposited at the belt, with ε0 D
8.85 ð 1012
As/Vm, the charge density cannot be larger than about 2.7 ð
105
As/m2
if E D 30 kV/cm. For a typical case the belt speed might be v D
20 m/sec and its width b D 1 m. The charging current according to eqn (2.13)
is then I ¾
D 540 µA. Although with sandwiched belts the output current might
be increased as well as with self-inducing arrangements mentioned above, the
actual short-circuit currents are limited to not more than a few mA with the
biggest generators.
The main advantages of belt-driven electrostatic generators are the high d.c.
voltages which can easily be reached, the lack of any fundamental ripple, and
the precision and flexibility, though any stability of the voltage can only be
achieved by suitable stabilizing devices. Then voltage fluctuations and voltage
stability may be in the order down to 105
.
The shortcomings of these generators are the limited current output, as
mentioned above, the limitations in belt velocity and its tendency for vibra-
tions, which aggravates an accurate grading of the electrical fields, and the
maintenance necessary due to the mechanically stressed parts.
The largest generator of this type was set into operation at Oak Ridge
National Laboratory.15
A view of this tandem-type heavy ion accelerator is
shown in Fig. 2.9. This generator operates with 25 MV, and was tested up to
internal flashovers with about 31 MV.
For h.v. testing purposes only a limited amount of generators are in use
due to the limited current output. A very interesting construction, however,
comprising the Van de Graaff generator as well as a coaxial test arrangement
for testing of gases, is used at MIT16
by Cooke. This generator, with an
output of about 4 MV, may be controlled to provide even very low frequency
a.c. voltages.
The disadvantages of the belt-driven generators led Felici to develop elec-
trostatic machines with insulating cylindrical rotors which can sustain perfectly
stable movement even at high speeds. The schematic diagram of such a
machine17
is shown in Fig. 2.10. To ensure a constant narrow air gap, the
stator is also made in the form of a cylinder. If the stator is a perfect insulator,
ions are deposited on its surface which tend to weaken the field. In order to
avoid such ion screening, a slight conductivity has to be provided for the stator
and resistivities in the range 1011
–1013
/cm have been found satisfactory.
The overall efficiency of the machine is higher than 90 per cent and the life
expectancies are only limited by mechanical wearing of the bearings, provided
the charge density on the rotor surface is kept within limits which depend upon
the insulating material employed. Epoxy cylinders have a practically unlimited

Figure 2.9 25-MV electrostatic tandem accelerator (Oak Ridge National
Laboratory)
life if the density remains sufficiently low. Unlike the rectifier circuit, the cylin-
drical generator delivers a smooth and continuous current without any ripple.
Sames of France have built two-pole generators of the Felici type. They
give an output of 600 kV at 4 mA and are suitable for use with particle
accelerator, electrostatic paint spray equipment, electrostatic precipitator, X-
ray purposes and testing h.v. cables. A cross-sectional view of the generator

4
+V
−V
1
2
3
7
5
6
+
+
+
+
+
+
−
−
−
−
−
−
Figure 2.10 Diagrammatic cross-section of the Felici generator.
(1) Cylindrical stator. (2) Insulating rotor. (3) Ionizer. (4) Contact metallic
segments. (5) Auxiliary generator. (6) Load. (7) Stationary insulating core
(V D 30 kV; CV D 200 kV)
is shown in Fig. 2.11. The distinctive features include a cylindrical pressure
vessel enclosing the generator, the rotor of which is driven at 3000 rpm by
means of an induction motor. Ions from an exciting source are sprayed onto
the rotor at the charging poles and are transported to the output poles with a
consequent rise of potential. The transfer of charge takes place by means of
thin blades placed a short distance from the rotor, and in the absence of any
rubbing contact the efficiency of the machine is about 90 per cent. The char-
acteristics of the 600-kV generator are such that the fluctuation in the voltage
is less than 104
per cent and the voltage drop at full load current of 4 mA
is only 500 V. For a 5 per cent variation in the main voltage, the generator
voltage remains within 105
per cent.
The main applications of these ‘rotating barrel’ generators are in physics as
well as in different areas of industrial applications, but rarely in h.v. testing.
The maximum voltages are limited to less than 750 kV.
Finally, another type of electrostatic generator is the vacuum-insulated
‘varying capacitance machine’, first discussed in detail by Trump18
and
recently again investigated by Philp.19
This machine provides a high voltage
in the range up to about 1 MV and/or high power in the range of megawatts.
The high efficiency, however, could only be reached by high field gradients
within the generator, which up to now can only be obtained theoretically
by assuming the possible high E values in vacuum. It is, however, doubtful
whether the stresses necessary can be reached within the large electrode areas

Series
resistor chain
Output cable
Voltage divider
resistor chain
Inductor
Frame
Glass
cylinder
Collecting electrodes
Spraying
electrode
Inductor
Rotor
Stator
Figure 2.11 Sames electrostatic generator
present in such generators, and therefore only a reference to this type of
generator might be useful.
2.2 Alternating voltages
As electric power transmission with high a.c. voltages predominates in our
transmission and distribution systems, the most common form of testing h.v.
apparatus is related to high a.c. voltages. It is obvious then that most research
work in electrical insulation systems has to be carried out with this type of
voltage.
In every laboratory HVAC supplies are therefore in common use. As far
as the voltage levels are concerned, these may range from about 10 kV
r.m.s. only up to more than 1.5 MV r.m.s. today, as the development of

transmission voltages up to about 1200 kV has proceeded for many years.
For routine testing, the voltage levels for power-frequency testing are always
related to the highest r.m.s. phase-to-phase voltage Vm of power transmission
systems. This ‘rated power-frequency short duration withstand voltage’ Vt is
different for different apparatus used within the transmission systems and also
dependent upon the type of insulation coordination applied (see Chapter 8).
For Vm 300 kV, the ratio Vt/Vm is up to about 1.9 and may decrease with
higher values of Vm. If, nevertheless, higher nominal voltages for the a.c.
testing supplies are foreseen, the necessity for the determination of safety
factors are most responsible for this fact.
In general, all a.c. voltage tests are made at the nominal power frequency
of the test objects. Typical exceptions are related to the testing of iron-cored
windings, i.e. potential or instrument transformers, or to fundamental studies
on insulating materials or systems. For iron-cored windings, the frequency
has to be raised to avoid saturation of the core. Depending upon the type of
testing equipment used, the methods for the generation of variable-frequency
voltages might be expensive.
A fundamental design factor for all a.c. testing supplies is an adequate
control system for a continuous regulation of the high output voltages. In
general, this will be performed by a control of the primary or l.v. input of
the voltage step-up systems. It is not the aim of this chapter to deal with the
details of these systems. Some hints related to the different methods will be
given in section 2.4.
Although power transmission systems are mostly of three-phase type, the
testing voltages are usually single-phase voltages to ground. The waveshapes
must be nearly pure sinusoidal with both half-cycles closely alike, and
according to the recommendations2,3
the results of a high-voltage test are
thought to be unaffected by small deviations from a sinusoid if the ratio of
peak-to-r.m.s. values equals
p
2 within š5 per cent, a requirement which can
be assumed to be met if the r.m.s. value of the harmonics does not exceed
5 per cent of the r.m.s. value of the fundamental. The r.m.s. value is for a
cycle of T
Vr.m.s. D

1
T
T
0
V2t dt.
The nominal value of the test voltage, however, is defined by its peak value
divided by
p
2, i.e. Vpeak/
p
2. The reason for this definition can be found in
the physics of breakdown phenomena in most of the insulating materials, with
the breakdown mainly following the peak voltages or the highest values of
field strength.
Testing of h.v. apparatus or h.v. insulation always involves an application
of high voltages to capacitive loads with low or very low power dissipation

only. In general, power dissipation can be completely neglected if the nominal
power output of the supply is determined. If Ct is the capacitance of the equip-
ment or sample under test, and Vn the nominal r.m.s. voltage of the h.v. testing
supply, the nominal KVA rating Pn may be calculated from the design formula
Pn D kV2
nωCt 2.14
in which the factor k ½ 1 accounts for additional capacitances within the whole
test circuit and some safety factor. Examples for additional capacitances are
h.v. electrodes and connections between test object and voltage source, which
might have large diameters and dimensions to avoid heavy discharges or even
partial discharges, or measurement devices as, e.g., capacitor voltage dividers
or sphere gaps frequently incorporated within the test circuit. This safety factor
k might range from only about 2 for very high voltages of ½1 MV, and may
increase to higher values for lower nominal voltages, as over-dimensioning
is economically possible. The capacitance of test equipment Ct may change
considerably, depending upon the type of equipment. Typical values are:
Simple post or suspension insulators some 10 pF
Bushings, simple and graded ¾100–1000 pF
Potential transformers ¾200–500 pF
Power transformers
1000 kVA ¾1000 pF
1000 kVA ¾1000–10 000 pF
H.V. power cables:
Oil-paper impregnated ¾250–300 pF/m
Gaseous insulated ¾60 pF/m
Metal clad substation, SF6 insulated ¾1000–10 000 pF
One may calculate the nominal currents In D Pn/Vn from eqn (2.14) for
different test voltages, different Ct values as shown above, and proper safety
factors k. From such estimations it may be seen that these currents may range
from some 10 mA for testing voltages of 100 kV only, up to amperes in the
megavolt range. Although these currents are not high and the nominal power
is moderate, many efforts are necessary to keep the test equipment as small
as possible, as the space is limited and expensive within any h.v. laboratory.
Frequently the equipment will be used also for field testing. Then the porta-
bility and transportation calls for lightweight equipment. Some facilities are
possible by the fact that most of the test voltages are only of short dura-
tion. The nominal ratings are, therefore, often related to short time periods
of 15 min. Due to the relatively large time constants for the thermal tempera-
ture rise, no sophisticated cooling systems are in general necessary within the
voltage testing supplies.
A final introductory remark is related to the necessity that all supplies can
withstand sudden voltage breakdowns of the output voltage. The stress to the

windings and coils accompanied by the breakdown events is usually not related
to the short-circuit currents and thus the magnetic forces within the windings,
as those currents are not large either; more frequently it is the stray potential
distribution between the windings which will cause insulation failures. One
may also provide proper damping resistors between h.v. testing supply and the
test equipment to reduce the rate of the sudden voltage drop and to avoid any
overvoltages within the test circuit caused by interruptions of the breakdown
phenomena. Nominal values of such damping resistors between 10 and 100 k
will usually not influence the test conditions. These resistors, however, are
expensive for very high voltages and it should be checked whether the a.c.
voltage supply can withstand the stresses without the damping resistors.
Most of the above remarks are common to the two main methods for the
generation of high a.c. testing voltages: transformers and resonant circuits.
2.2.1 Testing transformers
The power frequency single-phase transformer is the most common form of
HVAC testing apparatus. Designed for operation at the same frequency as the
normal working frequency of the test objects (i.e., 60 or 50 Hz), they may also
be used for higher frequencies with rated voltage, or for lower frequencies, if
the voltages are reduced in accordance to the frequency, to avoid saturation
of the core.
From the considerations of thermal rating, the kVA output and the funda-
mental design of the iron core and windings there is not a very big difference
between a testing and a single-phase power transformer. The differences are
related mainly to a smaller flux density within the core to avoid unneces-
sary high magnetizing currents which would produce higher harmonics in the
voltage regulator supplying the transformer, and to a very compact and well-
insulated h.v. winding for the rated voltage. Therefore, a single-phase testing
unit may be compared with the construction of a potential transformer used
for the measurement of voltage and power in power transmission systems.
For a better understanding of advanced circuits, the fundamental design of
such ‘single unit testing transformers’ will be illustrated. Figure 2.12(a) shows
the well-known circuit diagram. The primary winding ‘2’ is usually rated for
low voltages of 1 kV, but might often be split up in two or more windings
which can be switched in series or parallel (not shown here) to increase the
regulation capabilities. The iron core ‘l’ is fixed at earth potential as well as
one terminal of each of the two windings. Simplified cross-sections of two
possible constructions for the unit itself are given in Figs 2.12(b) and (c).
In both cases the layout arrangement of core and windings is basically the
same. Figure 2.12(b), however, shows a grounded metal tank unit, for which
an h.v. bushing ‘6’ is necessary to bring the high voltage out of the tank
‘5’. Instead of a bushing, a coaxial cable could also be used if this improves

1
3
2
Low
voltage
High
voltage
(a)
8
6
4
3
2
1
5
(b) (c)
8
7
5
Figure 2.12 Single unit testing transformers. (a) Diagram. (b c) different
construction units. (1) Iron core. (2) Primary l.v. or exciting winding.
(3) Secondary h.v. winding. (4) Field grading shield. (5) Grounded metal tank
and base. (6) H.V. bushing. (7) Insulating shell or tank. (8) H.V. electrode
the connection between testing transformer and test object. In Fig. 2.12(c)
the active part of the transformer is housed within an isolating cylinder ‘7’
avoiding the use of the bushing. This construction reduces the height, although
the heat transfer from inside to outside is aggravated. In both cases the vessels

would be filled with high-quality transformer oil, as most of the windings are
oil-paper insulated.
The sectional view of the windings shows the primary winding close to
the iron core and surrounded by the h.v. winding ‘3’. This coaxial arrange-
ment reduces the magnetic stray flux and increases, therefore, the coupling
of both windings. The shape of the cross-sectional view of winding no. 3 is
a hint to the usual layout of this coil: the beginning (grounded end) of the
h.v. winding is located at the side close to the core, and the end close to a
sliced metal shield, which prevents too high field intensities at h.v. potential.
Between both ends the single turns are arranged in layers, which are carefully
insulated from each other by solid materials (kraft paper sheets for instance).
Adjacent layers, therefore, form coaxial capacitors of high values, and if those
capacitances are equal – produced by the reduced width of the single layers
with increasing diameters – the potential distribution for transient voltages can
be kept constant. By this procedure, the trapezoidal shape of the cross-section
is originated.
It may well be understood that the design of the h.v. winding becomes
difficult if voltages of more than some 100 kV must be produced within one
coil. Better constructions are available by specialized techniques, mainly by
‘cascading’ transformers.
The first step in this technique is to place two h.v. windings on one iron
core, to join both windings in series and to connect this junction with the
core.20
For illustration, the circuit diagram is shown in Fig. 2.13 in combi-
nation with a simplified cross-section of the active part. The arrangement
could still be treated as a single unit transformer, as only one core exists.
The mid-point of the h.v. winding is connected to the core and to a metal
tank, if such a tank is used as a vessel. The cross-section shows that the
primary winding ‘2’ is, however, placed now around the first part ‘3a’ of the
whole h.t. winding, whose inner layer, which is at half-potential of the full
output voltage, is connected to the core. There are two additional windings,
‘4a’ and ‘4b’, rated for low voltages, which act as compensating windings.
These are placed close to the core and reduce the high leakage reactance
between ‘3b’ and the primary ‘2’. Often an exciting winding ‘5’, again a
winding rated for low voltages as the primary winding, is also available. This
exciting winding is introduced here as it will be needed for the cascading
of transformers. Note that this winding is at the full output potential of the
transformer.
Although no vessel is shown in which such a unit would be immersed, it
can easily be understood that for metal tank construction (see Fig. 2.12(b))
two h.v. bushings are now necessary. The tank itself must be insulated from
earth for half-output voltage. This typical view for testing transformers can
be seen in Fig. 2.14. If, however, insulating tanks are employed, this internal
layout may not necessarily be recognized from outside.

5
1
2
(a)
3a
3b
5
High
voltage
1
4a
4b
3b
3a
4a 4b
2
High voltage
(b)
Figure 2.13 Single unit testing transformer with mid-point potential at
core: Diagram (a) and cross-section (b). (1) Iron core. (2) Primary winding.
(3a b) High-voltage windings. (4a b) compensating windings.
(5) Exciting winding
Cascaded transformers
For voltages higher than about 300 to 500 kV, the cascading of transformers
is a big advantage, as the weight of a whole testing set can be subdivided into
single units and therefore transport and erection becomes easier. A review of
earlier constructions is given in reference 4.
A prerequisite to apply this technique is an exciting winding within each
transformer unit as already shown in Fig. 2.13. The cascading principle will
be illustrated with the basic scheme shown in Fig. 2.15. The l.v. supply
is connected to the primary winding ‘l’ of transformer I, designed for an
h.v. output of V as are the other two transformers. The exciting winding

Figure 2.14 Testing transformer for 1200 kV r.m.s. comprising three single
unit transformers according to Fig. 2.13, with metallic tanks and bushings
(High Voltage Laboratory, Technical University of Munich, Germany). (Note.
Suspended at ceiling and connected with transformer is a selenium-type
rectifier with a reverse voltage of 3.4 MV, see ref. 5.)
‘3’ supplies the primary of the second transformer unit II; both windings
are dimensioned for the same low voltage, and the potential is fixed to the
high potential V. The h.v. or secondary windings ‘2’ of both units are series
connected, so that a voltage of 2 V is produced hereby. The addition of the
stage III needs no further explanation. The tanks or vessels containing the

Transf. III
Transf. II
Transf. I
3
3
3
P
P
P P
2P 2P
V
2V
3V
3P
1
1
1
2
2
2
I = P/V
Figure 2.15 Basic circuit of cascaded transformers. (1) Primary windings.
(2) Secondary h.t. windings. (3) Tertiary exciting windings
active parts (core and windings) are indicated by dashed lines only. For a
metal tank construction and the non-subdivided h.v. winding assumed in this
basic scheme, the core and tank of each unit would be tapped to the l.v.
terminal of each secondary winding as indicated. Then the tank of transformer
I can be earthed; the tanks of transformers II and III are at high potentials,
namely V and 2 V above earth, and must be suitably insulated. Through h.t.
bushings the leads from the exciting coils ‘3’ as well as the tappings of the
h.v. windings are brought up to the next transformer. If the h.v. windings
of each transformer are of mid-point potential type (see Fig. 2.13), the tanks
are at potentials of 0.5 V, 1.5 V and 2.5 V respectively, as shown in Fig. 2.14.
Again, an insulating shell according to Fig. 2.12 could avoid the h.t. bushings,
rendering possible the stacking of the transformer units as shown in Fig. 2.16.
The disadvantage of transformer cascading is the heavy loading of primary
windings for the lower stages. In Fig. 2.15 this is indicated by the letter P, the
product of current and voltage for each of the coils. For this three-stage cascade
the output kVA rating would be 3P, and therefore each of the h.t. windings ‘2’
would carry a current of I D P/V. Also, only the primary winding of trans-
former III is loaded with P, but this power is drawn from the exciting winding

of transformer II. Therefore, the primary of this second stage is loaded with
2P. Finally, the full power 3P must be provided by the primary of trans-
former I. Thus an adequate dimensioning of the primary and exciting coils is
necessary. As for testing of insulation, the load is primarily a capacitive one,
a compensation of this capacitive load by l.v. reactors, which are in parallel
Figure 2.16 (a) Cascaded testing transformers with insulating shell
construction (courtesy IREQ, Canada)

Figure 2.16 (b) Cascaded testing transformers with metal tanks and
coolers. Total voltage 3000 kV, 4 A (courtesy HIGH-VOLT Dresden,
Germany)
to the primary windings, is possible. As these reactors must be switched in
accordance to the variable load, however, one usually tries to avoid this addi-
tional expense. It might also be necessary to add tuned filters to improve the
waveshape of the output voltage, that is to reduce higher harmonics.21
Without any compensation, the overloading of the lower stage transformers
introduces a relatively high internal impedance of the whole cascade circuit.
In a simplified equivalent circuit of each transformer unit, which consists of
a three-windings-type, we may define leakage or stray reactances X for each
winding, the primary Xp, the h.t. winding Xh and the exciting winding Xe.
Neglecting losses within the windings and magnetizing currents, the somewhat
simplified calculation of the resultant reactance Xres of a cascade unit with n
transformers having the individual reactances Xpv, Xhv and Xev shows
Xres D
n

vD1
[Xhv C n 2
Xev C n C 1 2
Xpv]. 2.15
(All reactances related to same voltage.)
Assuming three equal transformer units, the equation leads to a resultant
reactance of
Xres D 3Xh C 5Xe C 14Xp
instead of only 3Xh C Xe C Xp which might be expected.

Cascaded transformers are the dominating HVAC testing units in all large
testing laboratories. In Fig. 2.16(a) the 2.4-MV cascade of the Quebec Hydro
Research and Testing Laboratory can be seen. Here, each of the six 600 kV
single units are of insulating case type and the two lower stages consist of
four units to avoid overloading for full rated current. The world’s largest a.c.
testing station at WEI Istra near Moscow, Russia, is equipped with a cascaded
testing transformer rated for 3 MV, 12 MVA, which is shown in Fig. 2.16(b).
This very large unit was designed and built by the former firm TuR, Dresden,
Germany, and commissioned by Siemens.
A final remark relates to the effect that for all transformers the output voltage
will increase with load, as this is formed by capacitors. The equivalent circuit
of a transformer loaded by capacitors forms a series resonant circuit, which
is shown in Fig. 2.17 and will be used to introduce the resonant circuits for
testing purposes. With nominal load, the exciting frequency is well below
resonance frequency, so that the voltage increase is only about proportional to
the load. If the testing transformer, however, is switched to a primary voltage
higher than about half the rated voltage, the output voltage will oscillate with
resonance frequency, and the amplitude may easily become higher than the
rated voltage. The impedance of the voltage regulators used must also be taken
into account for quantitative calculations.
L1 r1 L2
L
r2
∼ C
Figure 2.17 Equivalent circuit of straight test set consisting of a
transformer and test capacitance
2.2.2 Series resonant circuits
The tuned series resonant h.v. testing circuit arose as a means of overcoming
the accidental and unwanted resonance to which the more conventional test
sets are more prone. If we consider a conventional ‘straight’ test set such as
the first unit in Fig. 2.15 used in, say, testing a capacitor C, then its equivalent
circuit will be that shown in Fig. 2.17. In this circuit (r1 C jωL1) represents the

impedances of the voltage regulator and the transformer primary. ωL represents
the transformer shunt impedance which is usually large compared with L1 and
L2 and can normally be neglected. (r2 C jωL2) represents the impedance of
the transformer secondary. 1/ωC represents the impedance of the load.
If by chance ωL1 C L2 D l/ωC, accidental resonance occurs. At supply
frequency the effect can be extremely dangerous, as the instantaneous voltage
application can be of the order of 20 times the intended high voltage. This
has given rise to some vicious explosions during cable testing. The greatest
possibility of this occurring is when testing at the maximum limit of current
and relatively low voltages, i.e. high capacitive load. Unfortunately the induc-
tance of most of the supply regulators varies somewhat over its range, so
that resonance does not necessarily occur when the voltages are at their low
switch-on value, but rather suddenly at the higher voltage range.
Resonance of a harmonic can similarly occur, as harmonic currents are
present due to the transformer iron core. In recent years, also the power supply
voltages contain a large amount of harmonics due to the still increasing appli-
cation of power electronics. These resonances are not quite so disastrous, but
third harmonics have been observed of greater amplitude than the fundamental,
and even the thirteenth harmonic can give a 5 per cent ripple on the voltage
waveform. This form of harmonic resonance causes greater voltage distor-
tion than other effects and occurs insidiously at particular capacitance loads,
usually unnoticed by conventional instrumentation.
With the series resonant set, however, the resonance is controlled at funda-
mental frequency and no unwanted resonance can therefore occur.
Historically, in the period 1935–45, power engineers were increasingly
aware of the potentialities of tuned circuits. It was not, however, until the late
1940s that engineers at Ferranti, England, and Standard Telefon of Kabelfabrik,
Norway, combined to make this a practical proposition culminating in a 600-
kV, 2400-kVA a.c. testing equipment completed in 1950, although an earlier
version of a resonance transformer for supplying X-ray equipment has been
described by Charlton.22
The development of this technique will be demonstrated based upon some
circuits shown in Fig. 2.18. In each circuit, the capacitance Ct represents
the almost pure capacitive load of the test objects and a constant power
supply frequency is assumed. In Fig. 2.18(a) a continuously variable induc-
tance (reactor) is connected to the l.v. winding of a step-up transformer,
whose secondary winding is rated for the full test voltage. By this means,
the impedance of the reactor is converted to the h.v. side. If the inductance
of the reactor is tuned to match the impedance of the capacitive load, the
idle power of the load is completely compensated. The step-up transformer,
however, has to carry the full load current, which is a disadvantage of this
circuit. The same disadvantage applies to the circuit of Fig. 2.18(b), although
no special means are necessary to cascade two or more units. The inductors are

Ct
Feed transf.
Reactor
Main transf.
Regulator
a.c. supply
Ct
Ct
No. 2 transf./reactor
(mounted on
insulators)
Feed transf.
No. 1 transf./reactor Regulator
Feed transf.
Regulator
r L
C
∼
f
(a) (b)
(c)
(d)
Mechanical
drive
No. 3 reactor
No. 2 reactor
No. 1 reactor
Figure 2.18 Series resonant circuit for transformer/reactor. (a) Single
transformer/reactor. (b) Two or more units in series. (c) Simplified diagram
of s.r. circuit for h.t. reactor units in series. (d) Equivalent diagram of s.r.
circuits
designed for high-quality factors Q D ωL/R within the limits of the inductance
variation. The feed transformer therefore injects the losses of the circuits only.
These types of s.r. circuits have been produced mainly up to about the late
1960s, since it was not possible to design continuously variable reactors for
high voltages. Then, as described in reference 23, a new technique was devel-
oped with split iron cores, the gap of which is variable. This technique now
provides h.t. continuously variable reactors up to about 300 kV per unit. Thus
the testing step-up transformers can be omitted, as indicated in Fig. 2.18(c).
The inductance of these h.t. reactors may be changed by up to 10 to 20 times,
offering the opportunity to tune the circuits with capacitances Ct, which vary
within the same order.
The equivalent circuit diagram for all these circuits is simply a low damped
s.r. circuit sketched in Fig. 2.18(d). Because the equations of such circuits
are well known, detailed designs will not be discussed here. It should be

emphasized that the high output voltage may best be controlled by a contin-
uously variable a.c. supply voltage, i.e. by a voltage regulator transformer, if
the circuit was tuned before. The feed transformers are rated for the nominal
currents of the inductors; the voltage rating can be as low as V/Q, if V is the
full output voltage and Q the worst quality factor of the whole circuit. The
additional advantages of the s.r. circuits may be summarized as follows.
Additional advantages of the series resonant circuit
1. The voltage waveshape is improved not only by the elimination of
unwanted resonances, but also by attenuation of harmonics already in the
power supply. A realistic figure for the amplification of the fundamental
voltage amplitude on resonance is between 20 and 50 times for power
frequencies of 50/60 Hz. Higher harmonic voltages are divided in the series
circuit with a decreasing proportion across the capacitive load. It is easily seen
that harmonics in the supply become insignificant.
2. The power required from the supply is lower than the kVA in the main
test circuit. It represents only about 5 per cent of the main kVA with a unity
power factor.
3. If a failure of the test specimen occurs, no heavy power arc will develop,
as only the load capacitance will be discharged. This is of great value to the
cable industry where a power arc can sometimes lead to the dangerous explo-
sion of the cable termination. It has also proved invaluable for development
work as the weak part of the test object is not completely destroyed. Addition-
ally, as the arc is self-extinguishing due to this voltage collapse, it is possible
to delay the tripping of the supply circuit.
4. The series or parallel operation of h.t. reactor or l.t. reactor/h.t. trans-
former units is simple and very efficient. Any number of units may be put
in series without the high impedance problems associated with a cascaded
testing transformer group (see eqn (2.15)). Equal voltage distributions for
series connections are easily provided by a proper control of the individual
reactor impedances. For heavy current testing it is possible to parallel the
reactor or reactor/transformer units, even if the impedances are different,
merely by controlling each associated reactance.
5. Various degrees of sophistication are possible concerning auto-tuning
devices keeping the set in tune, if supply frequency or load capacitance varies
during a long-term test, or concerning auto-voltage control.
Figure 2.19 shows cascaded h.t. reactors for an s.r. circuit according to
Fig. 2.18(c). From this figure a further advantage may be seen not mentioned
before, namely the reduction in the size and weight of such units in comparison
to testing transformers. For testing transformers typically, a specific weight of
about 10 to 20 kg/kVA (not including the necessary regulating and control
equipment) can be assumed. According to reference 24 and Fig. 2.18(c), this

Figure 2.19 2.2 MV series resonant circuit (Hitachi Research Laboratory,
supplied by Hipotronix, Brewster, USA)
weight for oil-insulated continuously variable h.v. reactors can be reduced to
about 3 to 6 kg/kVA for a power frequency of 60 Hz.
For field testing of cables, large rotating machines or metal-clad gas-
insulated substations (GIS), a still further reduction of weight and size
of the testing equipment is very desirable. This goal was reached by the

development25
of a different kind of series resonant circuit, for which chokes
with constant inductances are used. As the load capacitance cannot be kept
within narrow limits, the supply frequency must thus be continuously variable
to achieve resonance. This disadvantage, however, may be eliminated by the
novel features, which may be briefly explained by the schematic diagram
sketched in Fig. 2.20. An exciter supply, connected to the l.v. mains, excites
the s.r. circuit with a variable frequency; this supply is therefore designed as
a controlled frequency converter, which are standard devices nowadays. This
converter supplies again only the losses of the testing circuit, which are usually
less than about 2 per cent of the reactive power of Ct, if frequencies equal
or higher than 50 Hz are used. The chokes can easily be designed for such
high-quality factors, which increase with frequency up to some 100 Hz. In
Fig. 2.20 only one nominal inductance Ln of the h.v. reactor is indicated; this
inductance, however, might be provided by any number of chokes in series
and/or in parallel. Ct represents the test object and other shunt capacitances,
e.g. capacitor voltage dividers or some frequency-adjusting capacitor units,
if a specified testing frequency f must be achieved. Due to the resonance
condition, this frequency is always
f D
1
2- LnCt
. 2.16
Main
supply
f′
f′′
∼
∼
Frequency
converter
Ve
I
Ln (reactor or reactor units)
V
Ct
Feed
transformer
Figure 2.20 Schematic diagram of s.r. test circuit with variable test
frequency
The nominal inductance Ln will predominantly be designed according to
a nominal capacitance Cn D Ct which is the highest capacitance that can
be tested with the full rated voltage V D Vn of the circuit, and a nominal
frequency fn, which is the lowest frequency within this rated voltage. With

the above equation we thus obtain
Ln D
1
2-2
f2
nCn
. 2.17
A further criterion for the design of the choke is the maximum or nominal
current I D In, which either overheats the coil or saturates the iron core, if
any. As the losses are very small, we may neglect R − ωLn within the whole
frequency range; In may thus directly be derived from the voltage drop across
Ln, which is nearly the full rated voltage Vn, or from the fact that for all
frequencies or every cycle the magnetic energy of the choke is equivalent to
the electric energy stored within the test specimen. Thus
In D
Vn
2-fnLn
D Vn

Cn
Ln
. 2.18
These three equations are used to demonstrate the normalized operating char-
acteristics of the circuit. For test objects with capacitance values Ct different
from Cn, the resulting testing frequency f will also be different from fn. The
variation of the frequency then becomes, according to eqn (2.16),
f
fn
D

Cn
Ct
D
1
Ct/Cn
. 2.19
For Ct Cn, the reactor Ln can be used up to the full rated voltage Vn.
Although the frequency increases according to eqn (2.19), the load current
will always be lower than In. Ohm’s law or eqn (2.18) can be used to derive
the relationship of the normalized current for Ct Cn,
I
In
D
fn
f
D

Ct
Cn
. 2.20
For Ct Cn, this circuit may still and conveniently be applied, if the testing
voltage V D Vt is decreased to keep the current at its nominal value In. As
the current I is always proportional to the testing voltage, we may extend
eqn (2.20) to
I
In
D
Vt
Vn
D

Ct
Cn
2.20a
and apply this equation to show the necessary reduction of the testing voltage
for Ct Cn, if we limit I to In:
Vt
Vn
D
1
Ct/Cn
. 2.21

The normalized operating conditions given by eqns (2.19), (2.20) and (2.21)
are illustrated in Fig. 2.21. Whereas for quite small test specimens the test
frequency f may conveniently be limited by the addition of a permanent h.t.
capacitor as, e.g., a capacitor voltage divider; the relatively modest variation
of this frequency for large capacitors under test will improve the flexibility of
applications, i.e. for the testing of power cables with a.c. voltages. The actual
limitations in testing of very large test specimens with lower voltages than Vn
are given by the reduction of Q for too low frequencies, and the frequency
for which the exciter transformer saturates.
V/Vn
Vt/Vn
Ct/Cn
I/In
I/In
f/fn
f/fn
1 1 2
0 0 0
0 1 2 3 4
1
Figure 2.21 Operating characteristics of circuit according to Fig. 2.20
The prototype reactor described in reference 25 was designed for Vn D
200 kV, In D 6 A, fn D 100 Hz. The total weight is about 300 kg only,
yielding a specific weight of 0.25 kg/kVA for the nominal frequency. The
reactor has a cylindrical bar iron core, which is at half-potential of the
subdivided h.t. winding; this winding is coaxially placed across the core.
The construction provides excellent and high Q values between 50 and 150
within a frequency range of 50 to more than 1000 Hz. Thus a very small
a.c. testing supply is available which can easily be handled and conveyed. A
further advantage of this circuit is obviously related to the cheap generation of
frequencies higher than power frequencies, which may be used for ageing tests.
Since about 1980, powerful series resonance circuits with variable frequencies
were being used more often,26,52
as it was recognized that at least for on-site
tests the influence of frequencies, which are not very far from the nominal
working frequency, is of minor importance. Apart from tests on GIS, the
main application is now related to on-site tests of polymeric cables, for which
d.c. voltage tests are no longer applied due to well-known reasons.50,51
Figure 2.22 shows the test set-up comprising 12 modular reactor units of

Figure 2.22 Modular reactors of a series resonant circuit with variable
frequency during an on-site test of very long polymeric cables (courtesy FKH,
Zurich, Switzerland)
200 kV/6 A each as used during on-site a.c. voltage tests of 110 kV XLPE
cables with a length of about 7 km.
Whereas s.r. circuits are still used less in h.v. laboratories than testing
transformers, especially designed resonant circuits have often been applied
in conjunction with X-ray sets even for voltages in the MV range.4
2.3 Impulse voltages
As explained in detail in Chapter 8, disturbances of electric power transmis-
sion and distribution systems are frequently caused by two kinds of transient
voltages whose amplitudes may greatly exceed the peak values of the normal
a.c. operating voltage.

The first kind are lightning overvoltages, originated by lightning strokes
hitting the phase wires of overhead lines or the busbars of outdoor substa-
tions. The amplitudes are very high, usually in the order of 1000 kV or more,
as every stroke may inject lightning currents up to about 100 kA and even
more into the transmission line;27
each stroke is then followed by travelling
waves, whose amplitude is often limited by the maximum insulation strength
of the overhead line. The rate of voltage rise of such a travelling wave is
at its origin directly proportional to the steepness of the lightning current,
which may exceed 100 kA/µsec, and the voltage levels may simply be calcu-
lated by the current multiplied by the effective surge impedance of the line.
Too high voltage levels are immediately chopped by the breakdown of the
insulation and therefore travelling waves with steep wave fronts and even
steeper wave tails may stress the insulation of power transformers or other
h.v. equipment severely. Lightning protection systems, surge arresters and the
different kinds of losses will damp and distort the travelling waves, and there-
fore lightning overvoltages with very different waveshapes are present within
the transmission system.
The second kind is caused by switching phenomena. Their amplitudes are
always related to the operating voltage and the shape is influenced by the
impedances of the system as well as by the switching conditions. The rate of
voltage rise is usually slower, but it is well known that the waveshape can also
be very dangerous to different insulation systems, especially to atmospheric air
insulation in transmission systems with voltage levels higher than 245 kV.28
Both types of overvoltages are also effective in the l.v. distribution systems,
where they are either produced by the usual, sometimes current-limiting,
switches or where they have been transmitted from the h.v. distribution
systems. Here they may often cause a breakdown of electronic equipment,
as they can reach amplitudes of several kilovolts, and it should be mentioned
that the testing of certain l.v. apparatus with transient voltages or currents is a
need today.29
Such tests also involve ‘electromagnetic compatibility (EMC)
tests’, which will not be discussed here.
Although the actual shape of both kinds of overvoltages varies strongly,
it became necessary to simulate these transient voltages by relatively simple
means for testing purposes. The various national and international standards
define the impulse voltages as a unidirectional voltage which rises more or less
rapidly to a peak value and then decays relatively slowly to zero. In the relevant
IEC Standard 60,2
widely accepted today through national committees,3
a
distinction is made between lightning and switching impulses, i.e. according
to the origin of the transients. Impulse voltages with front durations varying
from less than one up to a few tens of microseconds are, in general, consid-
ered as lightning impulses. Figure 2.23(a) shows the shape for such a ‘full’
lightning impulse voltage as well as sketches for the same voltage chopped
at the tail (Fig. 2.23(b)) or on the front (Fig. 2.23(c)), i.e. interrupted by a

V
1.0
0.9
0.5
0.3
V
1.0
0.9
0.3
0
A
B
C
D
0.7α
0.1α
α
Tail
Front
O1
A
B
C
0
O1
T1
T2
Tc
T
T ′
Tail
Front
T1 = 1.67T
T ′ = 0.3T1 = 0.5T
t
t
V
1.0
0.9
0.7
0.3
0
0.1
A
B
C
D
O1 Tc
t
t
(b)
(a)
(c)
Figure 2.23 General shape and definitions of lightning impulse (LI)
voltages. (a) Full LI. (b) LI chopped on the tail. (c) LI chopped on the front.
T1 : front time. T2 : time to half-value. Tc: time to chopping. O1 : virtual origin
disruptive discharge. Although the definitions are clearly indicated, it should
be emphasized that the ‘virtual origin’ O1 is defined where the line AB cuts
the time axis. The ‘front time’ T1, again a virtual parameter, is defined as 1.67
times the interval T between the instants when the impulse is 30 per cent and
90 per cent of the peak value for full or chopped lightning impulses.

For front-chopped impulses the ‘time to chopping’ Tc is about equal to
T1. The reason for defining the point A at 30 per cent voltage level can be
found in most records of measured impulse voltages. It is quite difficult to
obtain a smooth slope within the first voltage rise, as the measuring systems as
well as stray capacitances and inductances may cause oscillations. For most
applications, the (virtual) front time T1 is 1.2 µs, and the (virtual) time to
half-value T2 is 50 µs. In general the specifications2
permit a tolerance of
up to š30 per cent for T1 and š20 per cent for T2. Such impulse voltages
are referred to as a T1/T2 impulse, and therefore the 1.2/50 impulse is the
accepted standard lightning impulse voltage today. Lightning impulses are
therefore of very short duration, mainly if they are chopped on front. Due
to inherent measurement errors (see Chapter 3, section 3.6) and uncertainties
in the evaluation the ‘time parameters’ T1, T2 and Tc or especially the time
difference between the points C and D (Figs 2.23(b) and (c)) can hardly be
quantified with high accuracy.
V
1.0
0.9
0.5
0
Tp
Td
T2
t
Figure 2.24 General shape of switching impulse voltages. Tp: time to peak.
T2 : time to half-value. Td: time above 90 per cent
Figure 2.24 illustrates the slope of a switching impulse. Whereas the time
to half-value T2 is defined similarly as before, the time to peak Tp is the time
interval between the actual origin and the instant when the voltage has reached
its maximum value. This definition could be criticized, as it is difficult to
establish the actual crest value with high accuracy. An additional parameter is
therefore the time Td, the time at 90 per cent of crest value. The different defi-
nitions in comparison to lightning impulses can be understood if the time scale

is emphasized: the standard switching impulse has time parameters (including
tolerances) of
Tp D 250 µs š 20%
T2 D 2500 µs š 60%
and is therefore described as a 250/2500 impulse. For fundamental investiga-
tions concerning the insulation strength of long air gaps or other apparatus, the
time to peak has to be varied between about 100 and 1000 µs, as the break-
down strength of the insulation systems may be sensitive upon the voltage
waveshape.28
2.3.1 Impulse voltage generator circuits
The introduction to the full impulse voltages as defined in the previous section
leads to simple circuits for the generation of the necessary waveshapes. The
rapid increase and slow decay can obviously be generated by discharging
circuits with two energy storages, as the waveshape may well be composed
by the superposition of two exponential functions. Again the load of the gener-
ators will be primarily capacitive, as insulation systems are tested. This load
will therefore contribute to the stored energy. A second source of energy could
be provided by an inductance or additional capacitor. For lightning impulses
mainly, a fast discharge of pure inductor is usually impossible, as h.v. chokes
with high energy content can never be built without appreciable stray capaci-
tances. Thus a suitable fast discharge circuit will always consist essentially of
two capacitors.
Single-stage generator circuits
Two basic circuits for single-stage impulse generators are shown in Fig. 2.25.
The capacitor C1 is slowly charged from a d.c. source until the spark gap G
breaks down. This spark gap acts as a voltage-limiting and voltage-sensitive
switch, whose ignition time (time to voltage breakdown) is very short in
comparison to T1. As such single-stage generators may be used for charging
voltages from some kV up to about 1 MV, the sphere gaps (see Chapter 3,
section 3.1) will offer proper operating conditions. An economic limit of the
charging voltage V0 is, however, a value of about 200 to 250 kV, as too large
diameters of the spheres would otherwise be required to avoid excessive inho-
mogeneous field distributions between the spheres. The resistors R1, R2 and
the capacitance C2 form the waveshaping network. R1 will primarily damp the
circuit and control the front time T1. R2 will discharge the capacitors and there-
fore essentially control the wavetail. The capacitance C2 represents the full
load, i.e. the object under test as well as all other capacitive elements which
are in parallel to the test object (measuring devices; additional load capacitor

to avoid large variations of T1/T2, if the test objects are changed). No induc-
tances are assumed so far, and are neglected in the first fundamental analysis,
which is also necessary to understand multistage generators. In general this
approximation is permissible, as the inductance of all circuit elements has to
be kept as low as possible.
R1
R2 C2 V(t)
V0 C1
G
R1
R2 C2 V(t)
R1
R2
R2 V(s)
V0 C1
G
1
sC2
V0
s
1/sC1
(c)
(b)
(a)
Figure 2.25 Single-stage impulse generator circuits (a) and (b). C1 :
discharge capacitance. C2 : load capacitance. R1 : front or damping
resistance. R2 : discharge resistance. (c) Transform circuit
Before starting the analysis, we should mention the most significant param-
eter of impulse generators. This is the maximum stored energy
W D 1
2 C1V0max
2
2.22
within the ‘discharge’ capacitance C1. As C1 is always much larger than C2,
this figure determines mainly the cost of a generator.

For the analysis we may use the Laplace transform circuit sketched in
Fig. 2.25(c), which simulates the boundary condition, that for t 0 C1 is
charged to V0 and for t 0 this capacitor is directly connected to the wave-
shaping network. For the circuit Fig. 2.25(a) the output voltage is thus given
by the expression
Vs D
V0
s
Z2
Z1 C Z2
,
where
Z1 D
1
C1s
C R1;
Z2 D
R2/C2s
R2 C 1/C2s
.
By substitution we find
Vs D
V0
k
1
s2
C as C b
2.23
where
a D

1
R1C1
C
1
R1C2
C
1
R2C2

;
b D

1
R1R2C1C2

;
k D R1C2. 2.24
For circuit Fig. 2.25(b) one finds the same general expression (eqn (2.23)),
with the following constants; however,
a D

1
R1C1
C
1
R1C2
C
1
R2C1

;
b D

1
R1R2C1C2

;
k D R1C2.



as above 2.25
For both circuits, therefore, we obtain from the transform tables the same
expression in the time domain:
Vt D
V0
k
1
˛2 ˛1
[exp˛1t exp˛2t] 2.26

where ˛1 and ˛2 are the roots of the equation s2
C as C b D 0, or
˛1, ˛2 D
a
2
Ý

a
2
2
b. 2.27
The output voltage Vt is therefore the superposition of two exponential func-
tions of different signs. According to eqn (2.27), the negative root leads to a
larger time constant 1/˛1 than the positive one, which is 1/˛2. A graph of the
expression (eqn (2.26)) is shown in Fig. 2.26, and a comparison with Figs 2.23
and 2.24 demonstrates the possibility to generate both types of impulse volt-
ages with these circuits.
0
v(t)
V0
k(α2−α1)
exp (−α1t)
V0
k(α2−α1)
exp (−α2t)
V0
k(α1−α2)
Vp
1/α2
1/α1 Time (t)
t max
Figure 2.26 The impulse voltage wave and its components according to
circuits in Fig. 2.25
Although one might assume that both circuits are equivalent, a larger differ-
ence may occur if the voltage efficiency, 5, is calculated. This efficiency is
defined as
5 D
Vp
V0
; 2.28
Vp being the peak value of the output voltage as indicated in Fig. 2.26. Obvi-
ously this value is always smaller than 1 or 100 per cent. It can be calculated

by finding tmax from dVt/dt D 0; this time for the voltage Vt to rise to its
peak value is given by
tmax D
ln˛2/˛1
˛2 ˛1
. 2.29
Substituting this equation into eqn (2.26), one may find
5 D
˛2/˛1[˛2/˛1˛1]
˛2/˛1[˛2/˛2˛1]
k˛2 ˛1
. 2.30
For a given impulse shape T1/T2 or Tp/T2 of the impulse voltages the values
of ˛1 and ˛2 must be equal. The differences in efficiency 5 can only be due,
therefore, to differences in the value of k D R1C2 for both circuits. We may
first calculate this term for the circuit Fig. 2.25(b), which has always a higher
efficiency for a given ratio of C2/C1, as during the discharge the resistors R1
and R2 do not form a voltage-dividing system. The product R1C2 is found by
eqn (2.27) by forming
˛1 Ð ˛2 D b
˛1 C ˛2 D a 2.31
and by the substitution of a and b from eqn (2.25). Then we obtain
k D R1C2 D
1
2

˛2 C ˛1
˛2 Ð ˛1

1

1 4
˛2 Ð ˛1
˛2 C ˛12

1 C
C2
C1

. 2.32
For C2 C1, which is fulfilled in all practical circuits, and with ˛2 × ˛1 for
all normalized waveshapes, one may simplify this equation to
k ¾
D
1 C C2/C1
˛2 C ˛1
. 2.33
The substitution of this expression in eqn (2.30) finally results in
5 D
C1
C1 C C2
D
1
1 C C2/C1
2.34
if again the inequality ˛2 × ˛1 is taken into account. The voltage efficiency
for this circuit will therefore rise continuously, if (C2/C1) decreases to zero.
Equation (2.34) indicates the reason why the discharge capacitance C1 should
be much larger than the load C2.
Less favourable is the circuit Fig. 2.25(a). The calculation of 5 may be
based upon the substitution of ˛1 and ˛2 in eqn (2.30) from eqn (2.27), and
a treatment of the ratio R1/R2 D fC2/C1, which increases heavily with

decreasing values of C2/C1. With minor approximations and the inequality
˛2 × ˛1 one may find the result
5 ¾
D
C1
C1 C C2
R2
R1 C R2
D
1
1 C C2/C1
1
1 C R1/R2
. 2.35
The comparison with eqn (2.34) shows the decrease in 5 due to an additional
factor. As the ratio R1/R2 is dependent upon the waveshape, the simple depen-
dency from (C2/C1) only is lost. For a 1.2/50 µs impulse and similar impulse
voltages the rapid increase of R1/R2 leads to a decrease of 5 for C2/C1 0.1;
therefore, the efficiency moves through an optimum value and decreases for
high C2/C1 values as well as for small ones. One could even show that for
very small C2/C1 ratios this circuit will fail to work.
In practice, both circuits are in use, often, however, in mixed and modified
form. If resistive h.v. dividers are placed in parallel to the test object, their
resistor value may contribute to discharge the circuits. The front resistor R1
is often subdivided, mainly in multistage generators treated later on. Never-
theless, the dependency of the voltage efficiency factors 5 is displayed in
Fig. 2.27 for the standard lightning impulse voltage 1.2/50 µsec as well as for
some other waveshapes. More information is available in the literature.30
100
80
60
40
20
0 0.1 0.2 0.3 0.4 0.5
C2/C1
h (%)
Circuit Fig. 2.26(a)
1.2/50 ms and approx.
1.2/200 ms
Circuit Fig. 2.25(b)
1.2/50 ms
1.2/5 ms
Figure 2.27 Voltage efficiency factors 5 in dependency of the capacitance
ratio C2 /C1 for lightning impulses T1 /T2
Dimensioning of circuit elements. The common task is to find the resistor
values for R1 and R2, as C2 and C1 are known in general. For larger

generators, the discharge capacitors are always given and dimensioned for
a good efficiency (see eqns (2.34) and (2.35)) within a certain range of
C2. This total load capacitance can easily be measured if it is not known
in advance. The unknown resistance values can then be calculated using
eqn (2.31) and the circuit-dependent values for a and b due to eqns (2.24)
and (2.25). The result will be for:
Circuit Fig. 2.25(a):
R1 D
1
2C1



1
˛1
C
1
˛2

1
˛1
C
1
˛2
2

4C1 C C2
˛1˛2 Ð C2

 . 2.36
R2 D
1
2C1 C C2



1
˛1
C
1
˛2

C

1
˛1
C
1
˛2
2

4C1 C C2
˛1˛2C2

 . 2.37
Circuit Fig. 2.25(b):
R1 D
1
2C2



1
˛1
C
1
˛2

1
˛1
C
1
˛2
2

4C1 C C2
˛1˛2C1

 . 2.38
R2 D
1
2C1 C C2



1
˛1
C
1
˛2

C

1
˛1
C
1
˛2
2

4C1 C C2
˛1˛2C1

 . 2.39
All these equations contain the time constants 1/˛1 and 1/˛2, which depend
upon the waveshape. There is, however, no simple relationship between these
time constants and the times T1, T2 and Tp as defined in the national or
international recommendations, i.e. in Figs 2.23 and 2.24. This relationship
can be found by applying the definitions to the analytical expression for Vt,
this means to eqn (2.26). The relationship is irrational and must be computed
numerically. The following table shows the result for some selected wave-
shapes:
T1/T2 Tp/T2 1/˛1 1/˛2
(µs) (µs) (µs) (µs)
1.2/5 – 3.48 0.80
1.2/50 – 68.2 0.405
1.2/200 – 284 0.381
250/2500 – 2877 104
– 250/2500 3155 62.5

The standardized nominal values of T1 and T2 are difficult to achieve in
practice, as even for fixed values of C1 the load C2 will vary and the exact
values for R1 and R2 according to eqns (2.38) and (2.39) are in general not
available. These resistors have to be dimensioned for the rated high voltage of
the generator and are accordingly expensive. The permissible tolerances for
T1 and T2 are therefore necessary and used to graduate the resistor values. A
recording of the real output voltage Vt will in addition be necessary if the
admissible impulse shape has to be testified.
Another reason for such a measurement is related to the value of the test
voltage as defined in the recommendations.2,3
This magnitude corresponds
to the crest value, if the shape of the lightning impulse is smooth. However,
oscillations or an overshoot may occur at the crest of the impulse. If the
frequency of such oscillations is not less than 0.5 MHz or the duration of
overshoot not over 1 µsec, a ‘mean curve’ (see Note below) should be drawn
through the curve. The maximum amplitude of this ‘mean curve’ defines the
value of the test voltage. Such a correction is only tolerated, provided their
single peak amplitude is not larger than 5 per cent of the crest value. Oscil-
lations on the front of the impulse (below 50 per cent of the crest value) are
tolerated, provided their single peak amplitude does not exceed 25 per cent
of the crest value. It should be emphasized that these tolerances constitute the
permitted differences between specified values and those actually recorded by
measurements. Due to measuring errors the true values and the recorded ones
may be somewhat different.
Note. With the increasing application of transient or digital recorders in
recording of impulse voltages it became very obvious that the definition of
a ‘mean curve’ for the evaluation of lightning impulse parameters of wave-
forms with oscillations and/or overshoot, as provided by the standards,2,3
is insufficient. Any software, written to evaluate the parameters, needs clear
instructions which are not yet available. As this matter is still under considera-
tion (by CIGRE Working Group 33.03) and a revision of the current standards
may provide solutions, no further comments to this problem are given.
The origin of such oscillations or the overshoot can be found in measuring
errors (heavily oscillating ‘step response’, see Chapter 3, section 3.6) as well
as by the inductances within every branch of the circuit or the stray capaci-
tances, which will increase with the physical dimensions of the circuit. As far
as inductances are concerned, a general rule for the necessary critical damping
of single-stage or – with less accuracy – of multistage generators can easily
be demonstrated by Fig. 2.28. If individual inductances L1, L2 are considered
within the discharge circuit as indicated in Fig. 2.28(a), a second order differ-
ential equation determines the output voltage across the load capacitance C2.
However, such an equivalent circuit cannot be exact, as additional circuits
related to stray capacitances are not taken into account. Thus we may only

combine the total inductance within the C1 –C2 circuit to single inductance
L, as shown in Fig. 2.28(b), and neglect the positions of the tail resistors,
which have no big influence. This reduces the circuit to a simple damped
series resonant circuit, and the critical resistance R D R1 for the circuit to be
non-oscillatory is given by the well-known equation
R1
¾
D R D 2

L
C
2.40
where
1
C
D
1
C1
C
1
C2
L1
C1
C1 C2
R2 R2 C2
L2
G
G L R
R1
Alternative
positions of R1
(a)
(b)
Figure 2.28 Simplified circuit of impulse generator and load. Circuit
showing alternative positions of the wave tail control resistance. (b) Circuit
for calculation of wave front oscillations
This equation is in general suitable for predicting the limiting values
for the front resistor R1. The extremely tedious analytical analysis of
circuits containing individual inductances is shown elsewhere.31–34
Computer
programs for transients may also be used to find the origin of oscillations,
although it is difficult to identify good equivalent circuits.
Multistage impulse generator circuits
The difficulties encountered with spark gaps for the switching of very high
voltages, the increase of the physical size of the circuit elements, the efforts

necessary in obtaining high d.c. voltages to charge C1 and, last but not least,
the difficulties of suppressing corona discharges from the structure and leads
during the charging period make the one-stage circuit inconvenient for higher
voltages.
In order to overcome these difficulties, in 1923 Marx35
suggested an
arrangement where a number of condensers are charged in parallel through
high ohmic resistances and then discharged in series through spark gaps.
There are many different, although always similar, multistage circuits in use.
To demonstrate the principle of operation, a typical circuit is presented in
Fig. 2.29 which shows the connections of a six-stage generator. The d.c.
voltage charges the equal stage capacitors C0
1 in parallel through the high
value charging resistors R0
as well as through the discharge (and also charging)
Multistage generator
R1
N
F
L
D
I
B
G
A
M
E
K
C
H
C ′
1
C ′
1
C ′
1
C ′
1
C ′
1 C ′
C ′
1
C ′
C ′′
C ′′′
C ′
1
R′
2
R′
2
R′
2
R′
2
R′
2
R′
2
G6
G5
G4
G3
G2
C2
G1
R ′
R ′
R ′
R ′
R ′
R ′
d.c.
(−V)
Figure 2.29 Basic circuit of a six-stage impulse generator (Marx generator)

resistances R0
2, which are much smaller than the resistors R0
and are compa-
rable with R2 in Fig. 2.25. At the end of the relatively long charging period
(typically several seconds up to 1 minute), the points A, B, . . . , F will be at
the potential of the d.c. source, e.g. V with respect to earth, and the points
G, H, . . . , N will remain at the earth potential, as the voltage drop during
charging across the resistors R0
2 is negligible. The discharge or firing of the
generator is initiated by the breakdown of the lowest gap G1 which is followed
by a nearly simultaneous breakdown of all the remaining gaps. According to
the traditional theory, which does not take into account the stray capacitances
indicated by the dotted lines, this rapid breakdown would be caused by high
overvoltages across the second and further gaps: when the first gap fires, the
potential at point A changes rapidly from V to zero, and thus the point H
increases its potential to CV. As the point B still would remain at the charging
potential, V, thus a voltage of 2V would appear across G2. This high over-
voltage would therefore cause this gap to break down and the potential at
point I would rise to C2V, creating a potential difference of 3V across gap
G3, if again the potential at point C would remain at the charging potential.
This traditional interpretation, however, is wrong, since the potentials B and
C can – neglecting stray capacitances – also follow the adjacent potentials of
the points A and B, as the resistors R0
are between. We may only see up to
now that this circuit will give an output voltage with a polarity opposite to
that of the charging voltage.
In practice, it has been noted that the gap G2 must be set to a gap distance
only slightly greater than that at which G1 breaks down; otherwise it does
not operate. According to Edwards, Husbands and Perry31
for an adequate
explanation one may assume the stray capacitances C0
, C00
and C000
within the
circuit. The capacitances C0
are formed by the electrical field between adjacent
stages; C000
has a similar meaning across two stages. C00
is the capacitance of
the spark gaps. If we assume now the resistors as open circuits, we may
easily see that the potential at point B is more or less fixed by the relative
magnitudes of the stray capacitances. Neglecting C0
between the points H
and C and taking into account that the discharge capacitors C0
1 are large in
comparison to the stray capacitances, point B can be assumed as mid-point of
a capacitor voltage divider formed by C00
and C0
/C000
. Thus the voltage rise
of point A from V to zero will cause the potential B to rise from V to a
voltage of
VB D V C V

C00
C0
C C00
C C000

D V

C0
C C000
C0
C C00
C C000

.
Hence the potential difference across G2 becomes
VG2 D CV VB D V

1 C
C0
C C000
C0
C C00
C C000

.

If C00
equals zero, the voltage across G2 will reach its maximum value 2 V.
This gap capacitance, however, cannot be avoided. If the stage capacitances
C0
and C000
are both zero, VG2 will equal V, and a sparking of G2 would
not be possible. It is apparent, therefore, that these stray capacitances enhance
favourable conditions for the operation of the generator. In reality, the condi-
tions set by the above equations are approximate only and are, of course,
transient, as the stray capacitances start to discharge via the resistors. As the
values of C0
to C000
are normally in the order of some 10 pF only, the time
constants for this discharge may be as low as 107
to 108
sec.
Thus the voltage across G2 appears for a short time and leads to breakdown
within several tens of nanoseconds. Transient overvoltages appear across the
further gaps, enhanced also by the fact that the output terminal N remains at
zero potential mainly, and therefore additional voltages are built up across the
resistor R0
2. So the breakdown continues and finally the terminal N attains a
voltage of C6V, or nV, if n stages are present.
The processes associated with the firing of such generators are even more
sophisticated. They have been thoroughly analysed and investigated experi-
mentally.31,36,37
In practice for a consistent operation it is necessary to set the distance for
the first gap G1 only slightly below the second and further gaps for earliest
breakdown. It is also necessary to have the axes of the gaps in one vertical
plane so that the ultraviolet illumination from the spark in the first gap
irradiates the other gaps. This ensures a supply of electrons released from
the gap to initiate breakdown during the short period when the gaps are
subjected to the overvoltage. If the first gap is not electronically triggered,
the consistency of its firing and stability of breakdown and therefore output
voltage is improved by providing ultraviolet illumination for the first gap.
These remarks indicate only a small part of the problems involved with the
construction of spark gaps and the layout of the generator. Before some of
these additional problems are treated, we shall treat more realistic Marx
circuits as used for the explanations so far.
In Fig. 2.29, the wavefront control resistor R1 is placed between the gener-
ator and the load only. Such a single ‘external’ front resistor, however, has to
withstand for a short time the full rated voltage and therefore is inconveniently
long or may occupy much space. This disadvantage can be avoided if either a
part of this resistance is distributed or if it is completely distributed within the
generator. Such an arrangement is illustrated in Fig. 2.30, in which in addition
the series connection of the capacitors C0
1 and gaps (as proposed originally by
Goodlet38
) is changed to an equivalent arrangement for which the polarity of
the output voltage is the same as the charging voltage. The charging resistors
R0
are always large compared with the distributed resistors R0
1 and R0
2, and
R0
2 is made as small as is necessary to give the required time to halve-value
T2. Adding the external front resistor R00
1 helps to damp oscillations otherwise

Multistage generator
R′2 R′1 R′2 R′1 R′2 R′1 R′2
C ′1
Gn
G1 C ′1
G2 C ′1 C ′1
G3
R′
R′ R′ R′
R′1
C2
R′′1
Further
stages
Impulse
voltage
d.c.
Figure 2.30 Multistage impulse generator with distributed discharge and
front resistors. R0
2 : discharge resistors. R0
1 : internal front resistors. R00
1 :
external front resistor
excited by the inductance and capacitance of the external leads between the
generator and the load, if these leads are long. It may be readily seen that this
circuit can be reduced to the single-stage impulse generator circuit shown in
Fig. 2.25(b). If the generator has fired, the total discharge capacitance C1 may
be calculated as
1
C1
D
n
1
C0
1
;
the effective front resistance R1 as
R1 D R00
1 C
n

R0
1;
and the effective discharge resistance R2 – neglecting the charging resistances
R0
– as
R2 D nR0
2 D
n

R0
2;
where n is the number of stages.
The consistent firing of such circuits could be explained as for the generator
of Fig. 2.29. For both generator circuits, the firing is aggravated if the resis-
tances R0
2 have relatively low values. According to eqns (2.22) and (2.39) such
low values appear with generators of high energy content and/or short times to
half-value, T2. Then the time constant for discharging the stray capacitances
to ground C000
(Fig. 2.29) will be too low and accordingly the overvoltages
for triggering the upper stages too short. By additional means providing high
resistance values within the firing period, this disadvantage can be avoided.39
Special circuits for generating switching impulse voltages
The common impulse generator circuits discussed so far are well capable of
producing standard switching impulses with adequate voltage efficiency 5, if
the circuit is well designed and the ratio C2/C1 is kept adequately small.

Other methods, however, have taken advantage of utilizing testing trans-
formers to step up the amplitudes from impulse voltages also. One such circuit
is shown in Fig. 2.31.2
An initially charged capacitor C1 is discharged into
the waveshaping circuit R1, C2 as well as into the l.v. winding of the trans-
former. The elements R1 and C2 or other suitable components, in the dotted
rectangle, may be used to control the waveshape. The wave tail is not only
controlled by the resistive voltage divider included, but also from the main
inductance of the transformer equivalent circuit. The time to crest Tp is even
without R1, C2 limited by the series inductance of the transformer, Ls, which
forms a series resonant circuit in combination with C1 and the load capacitance
C0
2. Neglecting any losses within the circuit, the voltage across the test object
would therefore start with a 1 cos ωt function, and as Tp ³ T/2 D -/fr,
fr being the resonance frequency, the time to crest is approximately
Tcr
¾
D - LsC
Transformer
Test object
Voltage
divider
C′2
C2
R1
C1
d.c.
Figure 2.31 Circuit for generation of switching impulses using a
transformer
where, neglecting transformer ratio,
C D
C1C0
2
C1 C C0
2
.
In general, low values of Tp are difficult to achieve, as Ls is quite large and also
the capacitance of the h.v. winding of the transformer contributes to the load
C0
2. Further problems arise with transient oscillations within the transformer
windings, mainly if cascaded testing transformers are used.40
The physical phenomena of disruptive discharges in long air gaps as well
as in other insulating systems are often related to the front of switching
impulses only. Therefore, with switching surges it is not always necessary
to produce double-exponential waveshapes as recommended in Fig. 2.24. In

fact, many investigations are sometimes made with unusual shapes of high
voltages, and one can establish many circuits for mainly oscillating volt-
ages, whose variety cannot be treated here. Only the common impulse voltage
circuit with a strongly increased inductance will demonstrate this variety, as
such circuits came into use for field testing of GIS.41,42
The principle of
such circuits is demonstrated in Figs 2.32(a) and (b). If the front resistor R1
in Fig. 2.25(a) is replaced by a series inductance, the circuit of Fig. 2.32(a)
results, which was first described for the generation of high switching impulse
voltages up to 500 kV by Bellaschi and Rademacher.43
A typical waveshape
of the output voltage is also included; it may easily be calculated, since the
circuit is a damped series resonant circuit only. The advantage of such a circuit
is the nearly doubling of the output voltage, if C2 − C1, in comparison to the
charging voltage V0. Also a proper damping does not decrease the amplitude
of the oscillation very much, and therefore the first increase of the voltage
may be used as the front of a switching impulse. This fundamental circuit can
be applied to multistage impulse generators, in which the front resistors are
replaced by h.t. chokes.42
V0 C1 R2
L1
C2
C1 R2
L1 R1
C2
t
t
(b)
(a)
Figure 2.32 Circuits for the generation of oscillating switching impulses
A modular design of such generators offers the opportunity for easy trans-
portation and erection on field site.
The second circuit of Fig. 2.32 uses an additional resistor R1 in series to L1,
and is therefore simply a pure impulse voltage circuit with a high inductance
within the discharge circuit. Therefore, unidirectional damped oscillations are
produced. Also this circuit can be used for high voltages, as originally shown
by Kojema and Tanaka.44
2.3.2 Operation, design and construction of impulse generators
Although the main aim of this book is not concerned with detailed information
about the design and construction of h.v. equipment, some additional remarks

are necessary for a better understanding of the problems involved with the
operation and use of impulse generators. The advice given within this chapter
is mainly related to section 2.3.1 and concerns multistage generators.
Every generator needs a d.c. power supply to charge the discharge capaci-
tance C1. This supply may simply consist of an h.t. transformer and rectifiers
providing unidirectional currents, as the voltage smoothing is made by C1. The
d.c. supply should primarily act as a current source, so that the charging time
can be controlled. The charging times should not be shorter than 5 to 10 sec,
as every voltage application to an object under test may lead to prestressing
effects within the insulation, influencing the withstand or breakdown strength.
Much longer charging times or time intervals between successive voltage
applications may be necessary, depending upon the material tested. For the
rough-controlled charging, the d.c. supply is usually only voltage controlled
by a voltage regulator at the primary of the h.t. transformer. Manufacturers
nowadays provide thyristor-controlled charging supplies with current-limiting
output. By this method, a programmed charging of generators is possible to
reach equal charging times for all levels of impulse voltages.
The layout of the construction of multistage impulse generators is largely
governed by the type of capacitors involved. Oil-paper-insulated capacitors of
low inductance and high capability for fast discharging are in common use;
mineral oil is often replaced by special fluids, providing higher permittivity to
increase the capacitance per volume.
For earlier constructions predominantly capacitor units have been used,
having the dielectric assembled in an insulating cylinder of porcelain or
varnished paper with plane metal end-plates. This construction provided
the obvious advantage that the stages of capacitors could be built up in
the vertical columns, each stage being separated from the adjacent one by
supports of the same or similar form as the capacitor units without dielectric.
Such a construction is illustrated by the generator shown in Fig. 2.33. The
disadvantage relates to the difficult replacement of failing capacitor units,
and therefore this originally preferred construction is not much used today.
New designs prefer complete modular constructions with capacitor units within
insulating cylinders or vessels, or within metal tanks and bushings. This design
originated from improvements in the capacitor dielectrics, which could reduce
the size of the capacitors significantly even for voltages up to 100 kV per unit
or more. To improve the consistency in firing the spark gaps, especially in
cases of large generators with many stages, the gaps are mounted inside a
cylindrical housing in which the pressure can be controlled, and hence the
gaps do not require adjusting. Such a construction is shown in Fig. 2.34 for a
20-stage, 4-MV, 200-kJ indoor generator. Besides such indoor constructions,
many of the generators for very high impulse voltages are used under outdoor
conditions for the direct testing of outdoor material and to avoid the use of too
large laboratories. To eliminate the detrimental influence of weather conditions

Figure 2.33 2.4-MV impulse generator UMIST
on electrical insulation as well as the mechanical influences (corrosion, etc.),
most of these generators are housed in huge insulating cylinders providing
for the full insulation of the output voltage and providing the opportunity to
have the generator itself under air conditioning.
Besides the charging resistances R0
, all of the waveshaping resistors should
be placed in such a way that they can easily be exchanged and replaced, as
they must be changed often to ensure the waveshapes necessary.

Figure 2.34 Indoor impulse generator, 20-stage, 4-MV, 200-kJ, with
encapsulated sphere gaps (courtesy of CEPEL, Rio de Janero, manufactured
by Haefely)
The resistors may be composed of wire, liquids or composite resistive
materials. Although the high heat capacity involved with liquid and composite
resistors would give preference to these resistor types, the instability of the
resistance values is of big disadvantage. Therefore highly non-inductively
wound wire resistors are best for the front and discharge resistors R1 and

R2 of the circuits. Wavefront resistors are quite satisfactory if their L/R value
is less than about 0.1 µsec.
The spark gaps are usually mounted on horizontal arms and the setting
of the gaps is adjusted by a remotely controlled motor in conjunction with
an indicator. This remark is, of course, related to gaps working in open air
only. Sometimes encapsulated and pressurized gaps are used, for which the
breakdown voltage is controlled by the gas pressure. The use of proper gas
mixtures gives good switching performance.45
Each generator should have a device to earth the capacitors when it is not
in operation. Due to relaxation phenomena, d.c.-operated capacitors can build
up high voltages after a short-time short-circuit.
All leads and electrodes within the generator should be dimensioned
properly to avoid too heavy corona discharges during the charging period.
During the short time of discharge and therefore impulse generation, partial
discharges cannot be prevented. A complete immersion of the generators in
improved insulation materials, as mineral-oil or high-pressure gases, could
reduce the size effectively; such solutions, however, are only used for special
purposes.46,47
Finally, some explanations refer to the tripping and synchronization of the
operation of impulse generators. The simple method of tripping the genera-
tors by non-triggered sphere gaps suffers from the disadvantage that the exact
instant of firing is not predictable. Furthermore, the presence of unavoidable
dust can cause irregular operation of multistage generators due to the following
main reason: dust particles are likely to be attracted to the spheres stressed
with d.c. voltages during charging, and the breakdown voltage can strongly
be reduced by these particles.48
If dust is randomly deposited on the lowest
gap (G1 in Fig. 2.29 or 2.30), the dispersion of the d.c. breakdown voltage
of this gap increases and thus the output voltage will not be stable. To over-
come this irregular firing of generators, which even with triggered gaps may
occur, a protection against dust may be provided. As separate enclosures of
the gaps with any insulating material prevents any ultraviolet illumination
from reaching the other gaps, only a common enclosure of all gaps is satis-
factory. Early investigations demonstrating those effects have been described
by Edwards et al.31
The common enclosure of all gaps provides, today, a
reliable method to ensure stable tripping (see Fig. 2.34).
A stable self-tripping by a fast mechanical closure of the first gap is a
simple means to avoid self-firing due to dust particles; however, this method
cannot be used to synchronize the impulse voltage with other events within a
very short time interval of a microsecond or less. The early need for synchro-
nization arose from the necessity to initiate the time sweep of oscilloscopes
used for the voltage recording. Later on, analogue impulse oscilloscopes have
been manufactured with built-in time delays, so that a pre-trigger may not be
necessary. Today, triggered firing of impulse generators would not be even

required as transient recorders are available (see Chapter 3) which replace the
analogue impulse oscilloscopes and which provide a continuous pre-trigger
system. However, as such recorders may still not yet be in common use and
since controlled or triggered switching of high voltages has so many applica-
tions in different fields, a brief review of controlled switching devices will be
presented.
There are many factors and properties which have to be considered if
controlled switching of voltages has to be achieved. The most essential factors
include:
ž the magnitude of the switched voltage (some kV up to MV),
ž the magnitude of the control voltage or signal,
ž the time delay between control signal and final stage of the switching,
ž the jitter of the time delay,
ž the conductivity of the switch in open and closed position,
ž the inductance of the switch,
ž the magnitude of the current switched,
ž the repetition or recurrence frequency, and finally,
ž the number of switching operations admissible.
For voltages higher than about 10 kV no solid state electronic element is able
to operate. Up to some 10 kV, different types of thyratrons may be used,
especially those with heated cathode and hydrogen content. Special types of
thyratrons may switch voltages up to 100 kV and currents up to 20 kA with
time delays of about 10 µsec and very low jitter down to a few nanoseconds.
These elements are expensive, however, and the application is aggravated by
the energy supply for the heated cathode, if the cathode is at high potential.
Thus, for voltages higher than 100 kV only spark gaps are nearly unlimited in
application, if they are properly controlled.
The physical mechanism responsible for the very fast transition of the resis-
tance value of a spark gap begins with the streamer breakdown of the insulating
gas and finishes with the arc, which has unlimited current-carrying capabilities.
The time-dependent resistance of a gap may be calculated from the well-known
‘spark laws’ due to Toepler, Rompe-Weizel or Braginskii.49
Investigations
show that the oldest law associated with Toepler may conveniently be used for
the computation of this time-dependent resistance, Rt. If it is the current
flowing in the gap, this resistance is given as
Rt D
kTd
t
0
it dt
where d is the gap distance and kT the ‘Toepler spark constant’. The integration
may be started t D 0 by a finite, not too small, current, due to the early
beginning of the spark formation. The values of kT are not real constants; they

are slightly dependent upon the gas involved and the field strength within the
gap before breakdown. For air, many measurements have yielded a value of
kT
¾
D 0.5 ð 104
š 20 per cent Vs/cm. The above relationship may be applied
to a discharge circuit consisting of a discharge capacitor C in series with a
resistance R and a homogeneous spark gap to calculate the current it and
the time-dependent voltage drop across the gap. If then a time to breakdown
Tb is defined as the time from the 90 to the 10 per cent instant values of the
decreasing voltage, for C
¾
D2 nF, R
¾
D100 one may derive the dependency
Tb
¾
D 13 ð 106 kT
E
,
where Tb is in ns, kT in Vs/cm and E in kV/cm. Thus for short switching times
high field strength E before breakdown is necessary. Such high values can be
achieved by pressurizing the gap, as the breakdown strength will increase
about proportionally with the gas pressure (see Chapter 5, section 5.5). Also
in air at atmospheric pressure switching times of about 20 ns will be reached
for voltages up to some 100 kV.
Insulating bushing Annular or pilot gap
Trigger rod
Earthed main
electrode
High voltage
main electrode
Voltage V ′
Main
gap
d′
Figure 2.35 The Trigatron spark gap
The development of triggered and therefore controlled spark gaps cannot be
discussed in detail. Only the principle will be considered using the arrangement
displayed in Fig. 2.35, which provides good operating characteristics. This
arrangement, known as ‘Trigatron’, consists essentially of a three-electrode
gap. The main electrodes – indicated as h.v. and earthed electrodes – may
consist of spheres, hemispheres or other nearly homogeneous electrode config-
urations. A small hole is drilled into the earthed electrode into which a metal
rod projects. The annular gap between the rod and the surrounding sphere is
typically about 1 mm. The metal rod or trigger electrode forms the third elec-
trode, being essentially at the same potential as the drilled electrode, as it is
connected to it through a high resistance, so that the control or tripping pulse

can be applied between these two electrodes. For this special arrangement, a
glass tube is fitted across the rod and is surrounded by a metal foil connected
to the potential of the main electrode. The function of this tube is to promote
corona discharges around the rod as this causes photoionization in the pilot
gap, if a tripping impulse is applied to the rod. Due to this photoionization
primary electrons are available in the annular gap which start the breakdown
without appreciable time delay. The glass tube (or a tube of different solid
insulation material, such as epoxy resin) may also fill the annular gap, so that
the rod as well as the tube with its face is flush with the outside surface of
the sphere. Thus a surface discharge is caused by the tripping pulse.
If a voltage V stresses the main gap, which is lower than the peak voltage
at which self-firing occurs, this main gap will break down at a voltage even
appreciably lower than the self-firing voltage Vs, if a tripping pulse is applied.
The Trigatron requires a pulse of some kilovolts, typically 10 kV, and the
tripping pulse should have a steep front with steepness
¾
D0.5 kV/nsec to keep
the jitter of the breakdown as small as possible. The first essential operating
characteristic refers to the voltage operating limits, at which a steady operation
or switching is possible. Such a characteristic is sketched in Fig. 2.36, where
the operating voltage V, the voltage across the main gap, is shown in depen-
dency of the main gap distance. The upper operation limit is identical with the
self-firing voltage as defined earlier; the lower operating limit is that at which
still a steady operation or breakdown is obtained with a predetermined jitter,
for instance 100 ns, or time delay. Such a characteristic is clearly dependent
Main gap distance, d
Starting
Lower limit
Upper limit
Operating
voltage,
V
Figure 2.36 Sketch of an operating characteristic of a Trigatron

upon the detailed construction of the Trigatron. These characteristics are also
polarity sensitive, mainly if the field distribution within the gap is not very
homogeneous; the polarity sensitivity refers also to the polarity of the tripping
pulse, which should always be of opposite polarity of the main voltage which
had to be switched.
The physical mechanism which causes the main gap to break down is funda-
mentally understood, although it might be quite complex in detail. Indeed, it is
recognized today that two types of mechanism are active. For small spacings d
and a given tripping voltage V, the breakdown may directly be initiated by the
distortion and enhancement of the electrical field between trigger electrode and
the opposite main electrode, leading to a direct breakdown between these two
electrodes. The arc then commutates from the larger electrode to the drilled
electrode for the main current. The second type of breakdown takes place for
larger gap distances. The trigger pulse causes a breakdown of the annular or
pilot gap, and the large amount of charge carriers of all types available after
sparking will initiate the breakdown of the main gap.
2.4 Control systems
In sections 2.1 to 2.3 only the basic principles for the generation of high d.c, a.c.
and impulse voltages have been described. In general, the specific circuit and
its components are only a part of a whole test system which is at least supple-
mented by sophisticated measuring (see Chapter 3) and control devices. In the
simplest case of ‘manual operation’, measurement and control are performed
by an operator and all results are recorded and further evaluated by hand. With
the availability of quite cheap and extremely efficient PCs it is now easy to
fulfil the demands for partly or fully automatic test sequences as well as to
perform automatic recording and evaluation of test results. Even older avail-
able voltage generators can, in general, be updated with the necessary hard-
and software. New types of h.v. generating systems are always prepared and
equipped with the necessary interfaces for full automatic operation.
The main problem in designing such computer-aided control, measuring
and evaluation systems is related to the hard electromagnetic environment
present during high voltage tests, for which breakdown phenomena emitting
very fast and strong electromagnetic transients are common. Industrial PCs of
high ‘electromagnetic compatibility’ must thus be used as control computers,
if a control room is not additionally shielded. As electronic measuring devices
are also quite sensitive to the electromagnetic environment, larger and well-
shielded control racks containing all items related to automatic testing, e.g.
computer, monitor, keyboard, printer, transient recorders, etc., can be used.
No limitations are set by applying special software for running the tests
as prescribed by special test procedures, by recording individual results and
evaluating test sequences by, e.g., statistical procedures, or for presenting test

records. Of big advantage is the application of interface systems which are
also commonly used in other industrial control systems.
Additional information is usually provided in the pamphlets or websites of
the manufacturers of high-voltage generators.
References
1. Report of the Task Force on Power System Textbooks of PEEC/IEEE; Electric Power Systems
Textbooks. Trans. IEEE. PAS 100 (1981), pp. 4255–4262.
2. IEC Publication 60: High-voltage test techniques. Part 1: General definitions and test require-
ments, Second Edition, 1989-11. Part 2: Measuring Systems, Second Edition, 1994-11.
3. IEEE Std 4-1995. IEEE Standard Techniques for High-voltage Testing.
4. J.D. Craggs and J.M. Meek. High Voltage Laboratory Technique. Butterworth, London, 1954.
5. H. Prinz. Feuer, Blitz und Funke. F. Bruckmann-Verlag, Munich, 1965.
6. H. Greinacher. Erzeugung einer Gleichspannung vom vielfachen Betrag einer Wechsels-
pannung ohne Transformator. Bull. SEV 11 (1920), p. 66.
7. J.D. Cockcroft and E.T.S. Walton. Experiments with high velocity ions. Proc. Roy. Soc.
London, Series A, 136 (1932), pp. 619–630.
8. K. Hammer and K. Kluge. Besonderheiten bei der Entwicklung von Gleich- spannungsprue-
fanlagen mit grossen Abgabestroemen. Elektrie 35 (1981), pp. 127–131.
9. H.P.J. Brekoo and A. Verhoeff. Phil. Tech. Rev. 23 (1962), p. 338.
10. M. Wagstaff. Direct Current 7 (1962), p. 304.
11. H.A. Enge. Cascade transformer high voltage generator. US Patent No. 3,596,167 (July
1971).
12. N.J. Felici. Elektrostatische Hochspannungs-Generatoren. Verlag G. Braun, Karlsruhe, 1957.
13. M. Pauthenier and M. Moreau-Hanot. J. de Phys. et le Radium 8 (1937), p. 193.
14. Van Atta et al. The design, operation and performance of the Round Hill electrostatic gener-
ator. Phys. Rev. 49 (1936), p. 761.
15. Holifield Heavy Ion Research Facility. Pamphlet of Oak Ridge National Laboratory, Oak
Ridge/Tenn., USA, 1981.
16. MIT. Study of gas dielectrics for cable insulation. EPRI Report No. EL 220 (October 1977).
17. N.J. Felici. Direct Current 1 (1953), p. 122.
18. J.G. Trump. Elect. Eng. 66 (1947), p. 525.
19. S.F. Philps. The vacuum-insulated, varying capacitance machine. Trans. IEEE. EI 12 (1977),
p. 130.
20. E.T. Norris and F.W. Taylor. J. IEE. 69 (1931), p. 673.
21. W. Müller. Untersuchung der Spannungskurvenform von Prüftranformatoren an einem
Modell. Siemens-Zeitschrift 35 (1961), pp. 50–57.
22. E.E. Charlton, W.F. Westendorp, L.E. Dempster and G. Hotaling. J. Appl. Phys. 10 (1939),
p. 374.
23. R. Reid. High voltage resonant testing. IEEE PES Winter Meeting 1974, Conf. Paper C74
038-6.
24. R. Reid. New method for power frequency testing of metal clad gas insulated substations and
larger rotary machines in the field. World Electrotechn. Congress, Moscow 1977, Section 1,
Report 29.
25. F. Bernasconi, W.S. Zaengl and K. Vonwiller. A new HV-series resonant circuit for dielectric
tests. 3rd Int. Symp. on HV Engg., Milan, Report 43.02,1979.
26. W.S. Zaengl et al. Experience of a.c. voltage tests with variable frequency using a lightweight
on-site s.r. device. CIGRE-Session 1982, Report 23.07.

27. R.H. Golde. Lightning, Vols I and II. Academic Press, London/New York/San Francisco,
1977.
28. Les Renardieres Group. Positive discharges in long air gaps at Les Renardieres. Electra
No. 53, July 1977.
29. F.A. Fisher and F.D. Martzloff. Transient control levels, a proposal for insulation coordina-
tion in low-voltage systems. Trans. IEEE PAS 95 (1976), pp. 120–129.
30. O. Etzel and G. Helmchen. Berechnung der Elemente des Stossspannungs-Kreises für die
Stossspannungen 1,2/50, 1,2/5 und 1,2/200. ETZ-A 85 (1964), pp. 578–582.
31. F.S. Edwards, A.S. Husbands and F.R. Perry. Proc. IEE 981 (1951), p. 155.
32. A. Vondenbusch. Ein allgemeines Berechnungsverfahren fuer Stossschaltungen mit
voneinander unabhaengigen Energiespeichern. Ph.D. Thesis, TH Aachen, 1968.
33. J.R. Eaton and J.P. Gebelein. GE Rev. 43 (1940), p. 322.
34. J.L. Thomason. Trans. AIEE 53 (1934), p. 169.
35. Marx, E. Deutsches Reichspatent (German Patent) No. 455933.
36. A. Rodewald. Ausgleichsvorgaenge in der Marxschen Vervielfachungsschaltung nach der
Zuendung der ersten Schaltfunkenstrecke. Bull. SEV 60 (1969), pp. 37–44.
37. F. Heilbronner. Das Durchzünden mehrstufiger Stossgeneratoren. ETZ-A 92 (1971),
pp. 372–376.
38. B.L. Goodlet. J. IEE 67 (1929), p. 1177, and British Patent No. 344 862.
39. A. Rodewald and K. Feser. The generation of lightning and switching impulse voltages in
the UHV region with an improved Marx circuit. Trans. IEEE PAS 93 (1974), pp. 414–420.
40. D. Kind and H. Wehinger. Transients in testing transformers due to the generation of
switching voltages. Trans. IEEE PAS 97 (1978), pp. 563–568.
41. W. Boeck, D. Kind and K.H. Schneider. Hochspannungspruefungen von SF6-Anlagen vor
Ort. ETZ-A 94 (1973), pp. 406–412.
42. K. Feser. High voltage testing of metal-enclosed, gas-insulated substations on-site with
oscillating switching impulse voltages. Gaseous Dielectrics II, Pergamon Press (1980), pp.
332–340 (Proc. of the 2nd Int. Symp. on Gas. Diel., Knoxville/Tenn., USA).
43. P.L. Bellaschi and L.B. Rademacher. Trans. AIEE 65 (1946), p. 1047.
44. K. Kojema and S. Tanaka. JIEE Japan 83 (1963), p. 42.
45. J.M. Christensen et al. A versatile automatic 1.2 MV impulse generator. 2nd Int. Symp. on
High Voltage Engg., Zurich, 1975, paper 2.1-03.
46. F. Brändlin, K. Feser and H. Sutter. Eine fahrbare Stossanlage für die Prüfung von
gekapselten SF6-isolierten Schaltanlagen. Bull. SEV 64 (1973), pp. 113–119.
47. F. Jamet and G. Thomer. Flash Radiography, Elsevier, Amsterdam, 1976.
48. T.E. Allibone and J.C. Saunderson. The influence of particulate matter on the breakdown of
large sphere-gaps. Third Int. Symp. on Gaseous Diel., Knoxville/Tenn., USA, 1982. (Gaseous
Dielectrics III, Pergamon Press, 1982, pp. 293–299.)
49. K. Moeller. Ein Beitrag zur experimentellen Ueberprüfung der Funkengesetze von Toepler,
Rompe-Weizel und Braginskii. ETZ-A 92 (1971), pp. 37–42.
50. H. Frank, W. Hauschild, I. Kantelberg and H. Schwab. HV DC Testing Generator for Short-
Time Reversal on Load. 4th Int. Symposium on High Voltage Engineering (ISH), Athens
1983, paper 5105.
51. W. Schufft and Y. Gotanda. A new DC Voltage Test System with Fast Polarity Reversal. 10th
Int. Symposium on High Voltage Engineering (ISH), Montreal 1997, Vol. 4, pp. 37–40.
52. W. Hauschild et al. Alternating voltage on-site testing of XLPE cables: The parameter selec-
tion of frequency-tuned resonant test systems. 10th Int. Symposium on High Voltage Engi-
neering (ISH), Montreal 1977, Vol. 4, pp. 75–78.

Chapter 3
Measurement of high voltages
Measurement of high voltages – d.c., a.c. or impulse voltages – involves
unusual problems that may not be familiar to specialists in the common
electrical measurement techniques. These problems increase with the
magnitude of the voltage, but are still easy to solve for voltages of some
10 kV only, and become difficult if hundreds of kilovolts or even megavolts
have to be measured. The difficulties are mainly related to the large structures
necessary to control the electrical fields, to avoid flashover and sometimes to
control the heat dissipation within the circuits.
This chapter is devoted to the measurement of voltages applied for the
testing of h.v. equipment or in research. Voltage-measuring methods used
within the electric power transmission systems, e.g. instrument transformers,
conventional or non-conventional ones, are not discussed. Such methods are
summarized in specialized books as, for instance, reference 2, distributed
publications,3,4Ł
or a summary given in reference 1. An introduction into
some measuring methods related to non-destructive insulation testing is pre-
sented separately (Chapter 7), and a brief reference related to the measurement
of electrical fields is included in Chapter 4, section 4.4.
The classification of the measuring methods by sections according to the
type of voltages to be measured would be difficult and confusing. A basic
principle of quantifying a voltage may cover all kinds of voltage shapes and
thus it controls the classification. The essential part of a measuring system
relates also to the elements or apparatus representing the individual circuit
elements. These could be treated separately, but a preferred treatment is within
the chapter, in which special problems first arise. Due to space limitation no
constructional details are given, but the comments referring to such prob-
lems should carefully be noted. The classification used here could introduce
difficulties in selecting proper methods for the measurement of given volt-
ages. Therefore, at this point a table is included (Table 3.1) which correlates
the methods treated within the corresponding sections to the type of voltages
to be measured.
Ł Superscript numbers are to references at the end of the chapter.

Table 3.1 (Note ‘C’ means ‘in combination with’)
Type of voltage d.c. voltages a.c. voltages Impulse
Quantity voltages
Mean value 3.3 Not Not
3.4 Applicable Applicable
3.6.4
r.m.s. value 3.2 3.2 Not
3.6.4 C 3.2 3.6.4 C 3.2 Applicable
Crest values 3.1 3.1 3.1
3.4 (special des.) 3.4 (special des.)
3.6.4 (special) C
3.3 (divider) C 3.5.1 3.5.3 or 3.7
ripple by CRO or
3.7 3.5.2 C 3.5.3
(see also 3.6.4) (see also 3.6.4)
Voltage shape 3.4 (special des.) 3.4 (special des.) 3.6.4 (special) C
or or CRO or 3.7
3.6.4 C CRO or 3.6.4 C CRO or
3.7 3.7
3.1 Peak voltage measurements by spark gaps
Simple spark gaps insulated by atmospheric air can be used to measure the
amplitude of a voltage above about 10 kV. The complex mechanism of this
physical effect, often employed in protecting equipment from overvoltages
(protection gaps), is treated in Chapter 5. Although spark gaps for measure-
ment purposes might be applied following given rules and recommendations
only, a misuse can be avoided through an adequate study of the physical
phenomena. As the fast transition from an either completely insulating or still
highly insulating state of a gap to the high conducting arc state is used to deter-
mine a voltage level, the disruptive discharge does not offer a direct reading
of the voltage across the gap. A complete short-circuit is the result of a spark,
and therefore the voltage source must be capable to allow such a short-circuit,
although the currents may and sometimes must be limited by resistors in series
with the gap. Strictly speaking, spark gaps according to sections 3.1.1 and 3.1.3
can be considered as approved calibration devices with a limited accuracy, i.e
known measuring uncertainty, but with a high reliability. Because of their high
reliability and simplicity, spark gaps will probably never completely disappear

Measurement of high voltages 79
from h.v. laboratories. More accurate and easier-to-use devices incorporating
electronic circuits are generally applied for routine measurements. But these
circuits are often sensitive to the electromagnetic effects and may sometimes
fail to work. A regular calibration of such devices against approved spark
gaps thus eliminates the possibility of large measuring errors and awkward
consequences.
The geometry of a spark gap is a decisive factor for its application. For
some decades the international and also national standards recommend the
sphere gap (section 3.1.1) and now also the rod/rod gap for approved voltage
measurements, as their reliability are best confirmed. The uniform field gaps
(section 3.1.3) are merely included here to demonstrate their disadvantages
and to save the beginner troublesome experiments.
3.1.1 Sphere gaps
Two adjacent metal spheres of equal diameters whose separation distance is
limited, as discussed later, form a sphere gap for the measurement of the peak
value of either d.c., a.c. or both kinds of impulse voltages. The ability to
respond to peak values of voltages, if the duration of the peak region is not
too short in time (1–3 µsec), is governed by a short statistical time lag, i.e.
the waiting time for an electron to appear to initiate an electron avalanche
and breakdown streamer, and an equally short formative time lag required for
the voltage breakdown or fast current increase within the breakdown channel
(see Fig. 5.42). The limitation in gap distance provides a fairly homogeneous
field distribution so that no predischarge or corona appears before breakdown;
the formative time lags are, therefore, also short. The permanent presence of
primary or initiatory electrons within the regions of maximum field gradients
to start critical avalanches within a short time lag is of great importance. The
electrical field distribution within the high field regions must sufficiently be
controlled by the geometry of the electrode and the air density as well as its
composition must be known. Air is composed of various types of molecules
which will influence the breakdown voltage. All these influences can be
accounted for by the well-known breakdown criteria of gases (see Chapter 5)
besides the primary electron impact, whose presence is a prerequisite.
All instructions as given in the still relevant IEC Recommendation5
or
National Standards6
in detail can be related to these effects. The two stan-
dardized arrangements for the construction of the sphere gaps are shown in
Figs 3.1(a) and 3.1(b). It should be noted also that in the horizontal arrange-
ment one sphere must be earthed.
These figures contain most of the instructions necessary to define the geo-
metry, except for values A and B which require some explanation. These
two parameters define clearances such as to maintain the field distribution
between the points on the two spheres that are closest to each other (sparking

1
4
≤0.5D
≤0.20
≤0.2D
≤0.5D
≤1.5D
X 2
5
P
B
≥2
D
≥15
D
D
S
3
A
Figure 3.1(a) Vertical sphere gap. 1. Insulating support. 2. Sphere shank.
3. Operating gear, showing maximum dimensions. 4. High-voltage connection
with series resistor. 5. Stress distributor, showing maximum dimensions.
P. Sparking point of h.v. sphere. A. Height of P above ground plane.
B. Radius of space free from external structures. X. Item 4 not to pass
through this plane within a distance B from P. Note: The figure is drawn to
scale for a 100-cm sphere gap at radius spacing. (Reproduced from ref. 5)
points) within narrow limits. The height of the sparking point P above the
horizontal ground plane, which can be a conducting network in or on the
floor of the laboratory, or a conducting surface on the support in which the
sphere gap is placed, must be within given limits related to the sphere diameter
D. To be accepted as a standard measuring device, a minimum clearance B
around the sphere must also be available, within which no extraneous objects
(such as walls, ceilings, transformer tanks, impulse generators) or supporting
framework for the spheres are allowed. Table 3.2 gives the required clearances.
Related to the accuracy of the field distribution are also requirements for the
construction of the spheres and their shanks. The most important rules are
reproduced partly:

Tolerances on size, shape and surface of spheres and their shanks
The spheres shall be carefully made so that their surfaces are smooth and their
curvature is as uniform as possible. The diameter shall nowhere differ by more
than 2 per cent from the nominal value. They should be reasonably free from
surface irregularities in the region of the sparking points. This region is defined
by a circle such as would be drawn on the spheres by a pair of dividers set
to an opening of 0.3D and centred on the sparking point. The freedom from
surface irregularities shall be checked by adequate measuring devices (for
more details see reference 5 or 6).
The surfaces of the spheres in the neighbourhood of the sparking points
shall be free from any trace of varnish, grease or other protective coating.
They shall be clean and dry, but need not to be polished. If the spheres become
4
3
3
2
X
2D
2D
≤0.5D ≤0.5D
≤0.2D
≤0.2D
≤1.5
D
≥
A
min
≥15
D
B
≥2D
≥2D
P
S
A
1
Figure 3.1(b) Horizontal sphere gap. 1. Insulating support. 2. Sphere
shank. 3. Operating gear, showing maximum dimensions. 4. High-voltage
connection with series resistor. P. Sparking point of h.v. sphere. A. Height of
P above ground plane. B. Radius of space free from external structures.
X. Item 4 not to pass through this plane within a distance B from P. Note:
The figure is drawn to scale for a 25-cm sphere gap at a radius spacing.
(Reproduced from ref. 5).

Table 3.2 Clearance around the spheres
Sphere Minimum Maximum Minimum
diameter value of value of Value of
D (mm) A A B
62.5 7D 9D 14S
125 6 8 12
250 5 7 10
500 4 6 8
750 4 6 8
1000 3.5 5 7
1500 3 4 6
2000 3 4 6
excessively roughened or pitted in use, they shall be refinished or replaced.
For relative air humidity exceeding 90 per cent, moisture may condense on
the surface and the measurement will then cease to be accurate.
The sphere shanks shall be reasonably in line and the shanks of the h.v.
sphere shall be free from sharp edges or corners, but the diameter of the shank
shall not exceed 0.2D over a length D. If a stress distributor is used at the end
of the shank, its greatest dimension shall be in accordance with Fig. 3.1.
Disruptive discharge voltages
If these and all otherwise recommended conditions are fulfilled, a sphere gap of
diameter D and spacing S will spark at a peak voltage whose value will be close
to the nominal values shown in Tables 3.3 and 3.4. These ‘calibration data’,
related to the atmospheric reference conditions (temperature 20°C; air pressure
101.3 kPa or 760 mmHg) and the kind and polarity of voltage applied, are a
result of joint international measurements within the period 1920 to about 1955;
a summary of this research work is found in the bibliography of reference 6.
Note. For every sphere diameter the sparking voltage is a non-linear function
of the gap distance, which is mainly due to the increasing field inhomogeneity
and only less to the physics of breakdown. All table values could well be
simulated by polynominals of order 6 or even less. Note also, that many table
values are the result of only linear interpolation between points which have
been the result of actual measurements.
For d.c. voltages the measurement is generally subject to larger errors,
caused by dust or fibres in the air.23,24
In this case the results are considered
to have an estimated uncertainty of š5 per cent provided that the spacing is
less than 0.4D and excessive dust is not present.

Table 3.3
(PART 1) Sphere gap with one sphere grounded
Peak values of disruptive discharge voltages (50% for
impulse tests) are valid for:
alternating voltages,
negative lightning impulse voltages,
negative switching impulse voltages,
direct voltages of either polarity.
Atmospheric reference conditions: 20°C and 101.3 kPa
Sphere gap spacing (mm) Voltage, kV peak
Sphere diameter (cm)
6.25 12.5 25
5 17.2 16.8
10 31.9 31.7
15 45.5 45.5
20 58.5 59.0
25 69.5 72.5 72.5
30 79.5 85.0 86
35 (87.5) 97.0 99
40 (95.0) 108 112
45 (101) 119 125
50 (107) 129 137
55 (112) 138 149
60 (116) 146 161
65 154 173
70 (161) 184
80 (174) 206
90 (185) 226
100 (195) 244
110 (203) 261
120 (212) 275
125 (214) 282
150 (314)
175 (342)
200 (366)
225 (385)
250 (400)
(continued overleaf )

Table 3.3 (continued)
Voltage, kV peak
Sphere gap
spacing (mm) 50 75 100 150 200
50 138 138 138 138
75 202 203 203 203 203
100 263 265 266 266 266
125 320 327 330 330 330
150 373 387 390 390 390
175 420 443 443 450 450
200 460 492 510 510 510
250 530 585 615 630 630
300 (585) 665 710 745 750
350 (630) 735 800 850 855
400 (670) (800) 875 955 975
450 (700) (850) 945 1050 1080
500 (730) (895) 1010 1130 1180
600 (970) (1110) 1280 1340
700 (1025) (1200) 1390 1480
750 (1040) (1230) 1440 1540
800 (1260) (1490) 1600
900 (1320) (1580) 1720
1000 (1360) (1660) 1840
1100 (1730) (1940)
1200 (1800) (2020)
1300 (1870) (2100)
1400 (1920) (2180)
1500 (1960) (2250)
1600 (2320)
(2320)
1700 (2370)
1800 (2410)
1900 (2460)
2000 (2490)
Note. The figures in parentheses, which are for spacing of more than 0.5D, will be within š5 per cent if the maximum clearances
in Table 3.2 are met. On errors for direct voltages, see text.

Table 3.4
Peak values of disruptive discharge voltages (50%
values) are valid for:
positive lightning impulses,
positive switching impulses,
direct voltages of either polarity.
Atmospheric reference conditions: 20°C and 101.3 kPa
Sphere gap spacing (mm) Voltage, kV peak
6.25 12.5 25
5 17.2 16.8 –
10 31.9 31.7 31.7
15 45.9 45.5 45.5
20 59 59 59
25 71.0 72.5 72.7
30 82.0 85.5 86
35 (91.5) 98.0 99
40 (101) 110 112
45 (108) 122 125
50 (115) 134 138
55 (122) 145 151
60 (127) 155 163
65 (164) 175
70 (173) 187
80 (189) 211
90 (203) 233
100 (215) 254
110 (229) 273
120 (234) 291
125 (239) 299
150 (337)
175 (368)
200 (395)
225 (416)
250 (433)
(continued overleaf )

Sphere gap Voltage, kV peak
spacing (mm)
50 75 100 150 200
50 138 138 138 138 138
75 203 202 203 203 203
100 263 265 266 266 266
125 323 327 330 330 330
150 380 387 390 390 390
175 432 447 450 450 450
200 480 505 510 510 510
250 555 605 620 630 630
300 (620) 695 725 745 750
350 (670) 770 815 858 860
400 (715) (835) 900 965 980
450 (745) (890) 980 1060 1090
500 (775) (940) 1040 1150 1190
600 (1020) (1150) 1310 1380
700 (1070) (1240) (1430) 1550
750 (1090) (1280) (1480) 1620
800 (1310) (1530) 1690
900 (1370) (1630) 1820
1000 (1410) (1720) 1930
1100 (1790) (2030)
1200 (1860) (2120)
1300 (1930) (2200)
1400 (1980) (2280)
1500 (2020) (2350)
1600 (2410)
1700 (2470)
1800 (2510)
1900 (2550)
2000 (2590)
Note. The figures in parentheses, which are for spacing of more than 0.5D, will be within š5 per cent if the maximum clearances
in Table 3.2 are met.

For a.c. and impulse voltages, the tables are considered to be ‘accurate’ (to
have an estimated uncertainty) within š3 per cent for gap lengths up to 0.5D.
The tables are not valid for impulses below 10 kV and gaps less than 0.05D
due to the difficulties to adjust the gap with sufficient accuracy. Values for
spacing larger than 0.5D are regared with less accuracy and, for that reason,
are shown in parentheses.
Remarks on the use of the sphere gap
The sphere gap represents a capacitance, which may form a series resonant
circuit with its leads. Heavy predischarges across a test object will excite
superimposed oscillations that may cause erratic breakdown. To avoid exces-
sive pitting of the spheres, protective series resistances may be placed between
test object and sphere gap, whose value may range from 0.1 to 1 M for d.c.
and a.c. power frequency voltages. For higher frequencies, the voltage drop
would increase and it is necessary to reduce the resistance. For impulse volt-
ages such protective resistors should not be used or should not exceed a value
of 500 (inductance less than 30 µH).
The disruptive discharge values of Tables 3.3 and 3.4 apply to measure-
ments made without irradiation other than random ionization already present,
except in
ž the measurement of voltages below 50 kV peak, irrespective of the sphere
diameters,
ž the measurement of voltages with spheres of 125 mm diameter and less,
whatever the voltage.
Therefore, for measurements under these conditions, additional irradiation is
recommended and is essential if accurate and consistent results are to be
obtained, especially in the case of impulse voltages and small spacing (see also
below). For irradiation a quartz tube mercury vapour lamp having a minimum
rating of 35 W and a current of at least 1 A is best applicable. Irradiation
by capsules containing radioactive materials having activities not less than
0.2 mCi (7,4 106
Bq) and preferably of about 0.6 mCi (22,2 106
Bq), inserted
in the h.v. sphere near the sparking points, needs precautions in handling the
radioactive materials.
The application of spark gaps is time consuming. The procedure usually
consists of establishing a relation between a high voltage, as measured by the
sphere gap, and the indication of a voltmeter, an oscilloscope, or other device
connected in the control circuit of the equipment. Unless the contrary can be
shown, this relation ceases to be valid if the circuit is altered in any respect
other than a slight change of the spacing of the spheres. The voltage measured
by the sphere gap is derived from the spacing. The procedure in establishing
the relationship varies with the type of voltage to be measured, as follows:
for the measurement of direct and alternating voltages, the voltage shall be

applied with an amplitude low enough not to cause disruptive discharge during
the switching transient and it is then raised sufficiently slowly for the l.v.
indicator to be read accurately at the instant of disruptive discharge of the
gap. Alternatively, a constant voltage may be applied across the gap and the
spacing between the spheres slowly reduced until disruptive discharge occurs.
If there is dust or fibrous material in the air, numerous low and erratic
disruptive discharges may occur, especially when direct voltages are being
measured, and it may be necessary to carry out a large number of tests before
consistent results can be obtained.
The procedure for the measurement of impulse voltages is different: in order
to obtain the 50 per cent disruptive discharge voltage, the spacing of the sphere
gap or the charging voltage of the impulse generator shall be adjusted in steps
corresponding to not more than 2 per cent of the expected disruptive discharge
value. Six applications of the impulse should be made at each step. The interval
between applications shall not be less than 5 sec. The value giving 50 per
cent probability of disruptive discharge is preferably obtained by interpolation
between at least two gap or voltage settings, one resulting in two disruptive
discharges or less, and the other in four disruptive discharges or more. Another,
less accurate, method is to adjust the settings until four to six disruptive
discharges are obtained in a series of ten successive applications.
Since in general the actual air density during a measurement differs from
the reference conditions, the disruptive voltage of the gap will be given as
Vd D kdVd0 3.1
where Vd0 corresponds to the table values and kd is a correction factor related
to air density. The actual relative air density (RAD) is given in general
terms by
υ D
p
p0
273 C t0
273 C t
D
p
p0
T0
T
3.2
where p0 D air pressure of standard condition, p D air pressure at test condi-
tions, t0 D 20°C, t D temperature in degrees Centigrade at test conditions.
The correction factor kd, given in Table 3.5, is a slightly non-linear function
of RAD, a result explained by Paschen’s law (see Chapter 5).
The influence of humidity is neglected in the recommendations, as its influ-
ence (an increase in breakdown voltage with increasing humidity) is unlikely
to exceed 2 or 3 per cent over the range of humidity normally encountered in
laboratories.
Some factors influencing the gap breakdown such as effects of nearby
earthed objects, of humidity, of dust particles, of irradiation and voltage
polarity are discussed fully in the previous book131
and will not be dealt
with here. The details can be found in references (7 to 24).

Table 3.5 Air-density
correction factor
Relative air Correction
density factor
RAD kd
0.70 0.72
0.75 0.77
0.80 0.82
0.85 0.86
0.90 0.91
0.95 0.95
1.00 1.00
1.05 1.05
1.10 1.09
1.15 1.13
Final remarks
It shall be emphasized that all relevant standards related to the sphere gap
are quite old and are essentially based on reference 5, which was submitted
to the National Committees for approval in 1958. The publication of IEC
52 in 1960 was then a compromise, accepted from most of the National
Committees, as Tables 3.3 and 3.4 are based on calibrations made under condi-
tions which were not always recorded in detail. Also, results from individual
researchers have not been in full agreement, especially for impulse voltages.
As, however, sphere gaps have been used since then world wide and – apart
from the following remarks – no significant errors could be detected during
application of this measuring method, the sparking voltages as provided by
the tables are obviously within the estimated uncertainties.
IEC Publication 52, since about 1993, has been under revision, which may
be finished in about 2000. The main aim of this revision is the inclusion of
switching surges and additional hints to the application of irradiation. Although
no final decisions have been made up to now, the following information may
be valuable:
ž Switching surges. Some later investigations demonstrated the applicability
of the table values for full standard switching impulse voltages, which are
identical to those of lightning impulses. This is already considered in refer-
ence 6 and in Tables 3.3 and 3.4.
ž Irradiation. Apart from the requirements as already given in the stan-
dards, the special importance of irradiation for the measurement of impulse

voltages will be mentioned. As shown in reference 22, additional irradiation
is required if the sphere gap is used in laboratories in which impulse
generators with encapsulated gaps are used. Current investigations are also
concerned with the influence of irradiation from different kinds of u.v. lamps
on breakdown. Only lamps having emission in the far ultraviolet (u.v.-C)
are efficient.
ž Influence of humidity. The systematic influence of humidity to the disruptive
voltages, which is about 0.2 per cent per g/m3
, will be mentioned, which
is the main source of the uncertainty.19
In this context, a calculation of
all disruptive voltages as provided by Table 3.3 shall be mentioned, see
reference 134. These calculations, completely based on the application of the
‘streamer breakdown criterion’, on the very well-known ‘effective ionization
coefficients’ of dry air, on a very accurate field distribution calculation
within the sphere gaps, and on the systematic (see Chapter 5.5) influence
of humidity on breakdown, essentially confirmed the validity of the table
values with only some exceptions.
3.1.2 Reference measuring systems
Up until the late 1980s the main method for calibration of high-voltage
measuring systems for impulse voltages was through the use of sphere gaps
in conjunction with step response measurements.3
The most recent revi-
sion of IEC 60-2:199453
contains significant differences from the previous
version.3
One of the fundamental changes has been to introduce the appli-
cation of Reference Measuring Systems in the area of impulse testing. The
concept of Reference Measuring Systems in high-voltage impulse testing was
introduced to address questions of quality assurance in measurements, an area
which has seen a significant increase in attention over the past decade.
The need for better quality assurance in high-voltage impulse measurements
was convincingly demonstrated in the 1980s and 1990s through the perfor-
mance of several round-robin tests designed to quantify the repeatability of
measurements between different laboratories. These tests comprised circu-
lating reference divider systems amongst different laboratories and comparing
the voltage and time parameters of impulses measured with the reference
systems to those derived from the measurement of the same impulses using the
regular laboratory dividers. Analysis of the results of these tests showed that
while some laboratories were able to make repeatable simultaneous measure-
ments of the voltage and time parameters of impulses using two Measuring
Systems with good agreement, others were not.135,136
For example, refer-
ence 135 gives the results of a round-robin test series performed under the
sponsorship of the IEEE High Voltage Test Techniques subcommittee. The
paper describes the results found when two reference dividers were circulated
to a number of laboratories, each having a Measuring System thought to be
adequately calibrated in accordance with the previous version of IEC 60-2.

The study revealed significant discrepancies in some laboratories between the
results obtained with the Measuring Systems currently in everyday use and
the Measuring System using the reference divider which was being circu-
lated. Based on these findings, the concept of Reference Measuring Systems
was introduced with the aim of improving the quality of high-voltage impulse
measurements.
A Reference Measuring System is defined in IEC Publication 60-2:1994
as a Measuring System having sufficient accuracy and stability for use in the
approval of other systems by making simultaneous comparative measurements
with specific types of waveforms and ranges of voltage or current. The require-
ments on a Reference Measuring System for use in high-voltage impulse
testing are clearly laid out in IEC Publication 60-2:1994. Reference dividers
meeting these requirements are available from several manufacturers or can
be constructed by the user.135
Figure 3.2 shows a photograph of a refer-
ence divider which is designed for use in calibrating a.c., d.c., lightning and
switching impulse voltages and is referred to as a Universal Reference Divider.
3.1.3 Uniform field gaps
It is often believed that some disadvantages of sphere gaps for peak voltage
measurements could be avoided by using properly designed plate electrodes
providing a uniform field distribution within a specified volume of air. The
procedure to control the electrical field within such an arrangement by appro-
priately shaped electrodes is discussed in Chapter 4, section 4.2 (Rogowski or
Bruce profile). It will also be shown in Chapter 5, section 5.6 that the break-
down voltage of a uniform field gap can be calculated based upon fundamental
physical processes and their dependency upon the field strength. According to
eqn (5.103) the breakdown voltage Vb can be expressed also by
Vb D EcυS C B
p
υS 3.3
if the gas pressure p in eqn (5.102) is replaced by the air density υ (see
eqn (3.2)) and if the gap distance is designated by S. The values Ec and
B in eqn (3.3) are also constants as the values E/pc and
p
K/C within
eqn (5.102). They are, however, dependent upon reference conditions. An
equivalent calculation as performed in Chapter 5, section 5.6 shows that
Ec D

p0
T
T0

ð

E
p

c
3.4
B D

Kp0T
CT0
3.5
where all values are defined by eqns (5.102) and (3.2). Equation (3.3) would
thus simply replace Tables 3.3 and 3.4 which are necessary for sphere gaps.

Figure 3.2 Universal Reference Voltage Divider for 500 kV lightning and
switching impulse, 200 kV a.c. (r.m.s.) and 250 kV d.c. voltage (courtesy
Presco AG, Switzerland)

Apart from this advantage of a uniform field gap, no polarity effect and
no influence of nearby earthed objects could be expected if the dimensions
are properly designed. All these advantages, however, are compensated by the
need for a very accurate mechanical finish of the electrodes, the extremely
careful parallel alignment, and – last but not least – the problem arising by
unavoidable dust, which cannot be solved for usual air conditions within a
laboratory. As the highly stressed electrode areas become much larger than
for sphere gaps, erratic disruptive discharges will tend to occur. Therefore, a
uniform field gap insulated in atmospheric air is not applicable for voltage
measurements.
3.1.4 Rod gaps
Rod gaps have earlier been used for the measurement of impulse voltages,
but because of the large scatter of the disruptive discharge voltage and the
uncertainties of the strong influence of the humidity, they are no longer allowed
to be used as measuring devices. A summary of these difficulties may be found
in reference 4 of Chapter 2.
Later investigations of Peschke,14
however, have demonstrated how the
simple electrode configuration rod/rod gap may be used for the measurement
of d.c. voltages, if the air density and the humidity is taken into account,
and if some rules relating to the electrode arrangement are followed. This
arrangement must comprise two hemispherically capped rods of about 20 mm
diameter as sketched in Fig. 3.3. The earthed rod must be long enough to
Earth plate
∼20 mm dia.
∼20 mm dia.
≈

≈

S (variable)
Rods capped by
half spheres
h.v.
1000 cm
35 cm
Figure 3.3 Electrode arrangement for a rod/rod gap to measure high d.c.
voltages

initiate positive breakdown streamers if the h.v. rod is the cathode. Then for
both polarities the breakdown will always be initiated by positive streamers
giving a very small scatter and being humidity dependent. Apart from too low
voltages 130 kV, for which the proposed rod/rod gap is not sufficiently
inhomogeneous, the breakdown voltage Vb then follows the relationship
Vb D υA C BS 4

5.1 ð 102h C 8.65 in kV 3.6
where S D gap distance in cm, υ D relative air density according to eqn (3.2),
and h D absolute humidity in g/m3
.
This empirical equation is limited to 4 h 20 g/m3
and has been shown
to apply in the voltage range up to 1300 kV. Vb shows a linear increase with
the gap length S, and the steepness B for the gap configuration shown in
Fig. 3.3 is not very dependent on polarity. Also the constant A displays a
small polarity effect, and numerical values are
A D 20 kV; B D 5.1 kV/cm; for positive polarity
A D 15 kV; B D 5.45 kV/cm; for negative polarity
of the h.v. electrode. The estimated uncertainty of eqn (3.6) is lower than š2
per cent and therefore smaller than the ‘accuracy’ provided by sphere gaps.
These investigations of Peschke14
triggered additional work, the results of
which are provisionally included within Appendix C of IEC Standard 60-1,
1989, see reference 2 of Chapter 2. The rod/rod gap thus became an approved
measuring device for d.c. voltages. The additional investigations showed, that
with somewhat different electrode configurations, which are not displayed
here, the disruptive voltage U0 even becomes equal for both voltage polarities,
namely
U0 D 2 C 0.534d 3.6a
where U0 is in kV and d is the gap spacing in millimetres. This equation is
valid for gap spacing between 250 and 2500 mm, an air humidity between 1
and 13 g/m3
, and its measurement uncertainty is estimated to be less than š3
per cent for these boundary conditions. A disadvantage of the electrode config-
urations as shown in Figs 19a/b of IEC 60-1 are the much larger dimensions
as those displayed in Fig. 3.3.
3.2 Electrostatic voltmeters
Coulomb’s law defines the electrical field as a field of forces, and since elec-
trical fields may be produced by voltages, the measurement of voltages can
be related to a force measurement. In 1884 Lord Kelvin suggested a design

for an electrostatic voltmeter based upon this measuring principle. If the field
is produced by the voltage V between a pair of parallel plane disc electrodes,
the force F on an area A of the electrode, for which the field gradient E is
the same across the area and perpendicular to the surface, can be calculated
from the derivative of the stored electrical energy Wel taken in the field direc-
tion (x). Since each volume element A dx contains the same stored energy
dWel D εE2
A dx/2, the attracting force F D dWel/dx becomes
jFj D
εAE2
2
D
εA
2S2
V2
, 3.7
where ε D permittivity of the insulating medium and S D gap length between
the parallel plane electrodes.
The attracting force is always positive independent of the polarity of the
voltage. If the voltage is not constant, the force is also time dependent. Then
the mean value of the force is used to measure the voltage, thus
1
T
T
0
Ft dt D
εA
2S2
1
T
T
0
v2
t dt D
εA
2S2
Vr.m.s2
, 3.8
where T is a proper integration time. Thus, electrostatic voltmeters are r.m.s.-
indicating instruments!
The design of most of the realized instruments is arranged such that one
of the electrodes or a part of it is allowed to move. By this movement, the
electrical field will slightly change which in general can be neglected. Besides
differences in the construction of the electrode arrangements, the various volt-
meters differ in the use of different methods of restoring forces required to
balance the electrostatic attraction; these can be a suspension of the moving
electrode on one arm of a balance or its suspension on a spring or the use of
a pendulous or torsional suspension. The small movement is generally trans-
mitted and amplified by a spotlight and mirror system, but many other systems
have also been used. If the movement of the electrode is prevented or mini-
mized and the field distribution can exactly be calculated, the electrostatic
measuring device can be used for absolute voltage measurements, since the
calibration can be made in terms of the fundamental quantities of length and
forces.
The paramount advantage is the extremely low loading effect, as only elec-
trical fields have to be built up. The atmospheric air, high-pressure gas or even
high vacuum between the electrodes provide very high resistivity, and thus
the active power losses are mainly due to the resistance of insulating mate-
rials used elsewhere. The measurement of voltages lower than about 50 V is,
however, not possible, as the forces become too small.
The measuring principle displays no upper frequency limit. The load induc-
tance and the electrode system capacitance, however, form a series resonant

circuit, thus limiting the frequency range. For small voltmeters the upper
frequency is generally in the order of some MHz.
Many designs and examples of electrostatic voltmeters have been
summarized or described in the books of Schwab,1
Paasche,30
Kuffel and
Abdullah,26
Naidu and Kamaraju,29
and Bowdler.127
High-precision-type
electrostatic voltmeters have been built for very high voltages up to 1000 kV.
The construction of such an absolute voltmeter was described by House
et al.31
In spite of the inherent advantages of this kind of instrument, their
application for h.v. testing purposes is very limited nowadays. For d.c. voltage
measurements, the electrostatic voltmeters compete with resistor voltage
dividers or measuring resistors (see next chapter), as the very high input
impedance is in general not necessary. For a.c. voltage measurements, the
r.m.s. value is either of minor importance for dielectric testing or capacitor
voltage dividers (see section 3.6) can be used together with low-voltage
electronic r.m.s. instruments, which provide acceptable low uncertainties.
Thus the actual use of these instruments is restricted and the number of
manufacturers is therefore extremely limited.
3.3 Ammeter in series with high ohmic resistors and
high ohmic resistor voltage dividers
The two basic principles
Ohm’s law provides a method to reduce high voltages to measurable quanti-
ties, i.e. adequate currents or low voltages. The simplest method employs a
microammeter in series with a resistor R of sufficiently high value to keep
the loading of an h.v. source as small as possible (Fig. 3.4(a)). Thus for a
pure resistance R, the measured quantities are related to the unknown high
voltage by
vt D Rit 3.9
or
V D RI 3.10
if the voltage drop across the ammeter is neglected, which is usually allowable
due to the small terminal impedance of such instruments. For d.c. voltage
measurements, average current-indicating instruments such as moving coil or
equivalent electronic meters are used giving the arithmetic mean value of V
according to eqn (3. 10). Less recommendable is the measurement of r.m.s.
values as the polarity of the high voltage would not be shown. Fundamentally

(a) (b)
l;i(t)
V2;v2 (t)
V;v(t) V;v(t)
OP
R R1
R2
i1
i2
i 0
Figure 3.4 Measurement of high d.c. and a.c. voltages by means of:
(a) ammeter in series with resistor R; (b) voltage divider R1 , R2 and
voltmeter of negligible current input. OP, over voltage protection
the time-dependency vt according to eqn (3.9) could also be measured by, for
instance, an oscilloscope. The difficulties, however, in treating the resistance
R as a pure resistance are limiting this application. This problem will be
discussed later on. It is recommended that the instrument be protected against
overvoltage with a glow tube or zener diode for safety reasons.
The main difficulties encountered in this method are related to the stability of
the resistance R. All types of resistors are more or less temperature dependent
and often may show some voltage dependency. Such variations are directly
proportional to the voltage to be measured and increase the uncertainty of the
measurement result.
Before discussing some details concerning resistor technology, the alter-
native method shown in Fig. 3.4(b) will be described. If the output voltage
of this voltage divider is measured with instruments of negligible current
consumption i ! 0 or i/i2 − 1, the high voltage is now given by
vt D v2t

1 C
R1
R2

3.11
V D V2

1 C
R1
R2

3.12
Apart from the uncertainty of the output voltage measurement (V2 or v2t,
the magnitude of the high voltage is now only influenced by the ratio R1/R2.

As both resistors pass the same current i1 D i2, the influence of voltage and
temperature coefficients of the resistors can be eliminated to a large extent, if
both resistors employ equal resistor technology, are subjected to equal voltage
stresses, and if provisions are made to prevent accumulation of heat within any
section of the resistor column. Thus the uncertainty of the measurement can
be greatly reduced. Accurate measurement of V2 was difficult in earlier times
as only electrostatic voltmeters of limited accuracy had been available. Today
electronic voltmeters with terminal impedances high enough to keep i − i2
and giving high accuracy for d.c. voltage measurements are easy to use.
So far it appears that either method could easily be used for measurement of
even very high voltages. The design of the methods starts with dimensioning
the h.v. resistor R or R1 respectively. The current through these resistors is
limited by two factors. The first one is set by the heat dissipation and heat
transfer to the outside and defines the upper limit of the current. A calculation
assuming heat transfer by natural convection only would demonstrate upper
limits of 1 to 2 mA. The second factor is due to the loading of the h.v. source;
in general, very low currents are desirable. As the resistors predominantly
at the input end of the h.v. column are at high potential and thus high field
gradients have to be controlled, even with the best insulating materials the
leakage along the resistor column or the supporting structure controls the
lower limit of the current, which in general shall not be smaller than about
100 µA. This magnitude results in a resistance of 1010
for a voltage of
1000 kV, and thus the problem of the resistor technology arises.
Comment regarding the resistor technology and design of the h.v. arm
In practice this high ohmic resistor R, R1 is composed of a large number of
individual elements connected in series, as no commercial types of single unit
resistors for very high voltages are available.
Wire-wound metal resistors made from Cu–Mn, Cu–Ni and Ni–Cr alloys
or similar compositions have very low temperature coefficients down to
about 105
/K and provide adequate accuracy for the method prescribed in
Fig. 3.4(a). As, however, the specific resistivity of these materials is not very
large, the length of the wire required becomes very considerable even for
currents of 1 mA and even for the finest gauge which can be made. Individual
units of about 1 M each then must be small in size as only a voltage
drop of 1 kV arises, and thus the manner of winding will enhance self-
inductive and self-capacitive components. In addition, the distributed stray
capacitance to ground, discussed in more detail in section 3.6 and briefly
below, causes a strongly non-linear voltage distribution along a resistor column
and overstresses the individual elements during a sudden load drop originated
by voltage breakdown of a test object. Wire-wound resistors are thus not only
very expensive to produce, but also quite sensitive to sudden voltage drops.

Many constructions have been described in the literature and summaries can
be found in references 1, 26, 30 and 127.
Especially for the voltage-dividing system, Fig. 3.4(b), common carbon,
carbon composition or metal oxide film resistors are preferably used. They
should be carefully selected due to the usually larger temperature coefficients
(TC) which may even be different for the same type of such resistors. Nowa-
days, however, metal oxide products with TC values of about 20 to 30 ppm/K
only can be produced. The resistor value of all these resistors may change also
with voltage magnitude, and the – in general – negative voltage coefficients
may be found in the manufacturer’s catalogue. The self-inductance of such
resistors is always negligible, as the high values of the individual film resis-
tors are often reached by a bifilar arrangement of the film. Too thin films are
generally destroyed by fast voltage breakdown across the resistor column. This
effect may well be understood if the stray capacitances to earth are considered,
or if high field gradients at the film surfaces are encountered. If the voltage
suddenly disappears, high capacitive or displacement currents are injected into
the thin film material, which cannot dissipate the heat within a very short time.
Thus the temperature rise within the material may be so high that some of
the material even explodes. The result is an increase of the original resistance
value. Carbon composition resistors have large energy absorption capabilities.
Their resistor value may, however, decrease due to short-time overloads, as
the individual particles may be additionally sintered. A conditioning performed
by prestressing of such resistors with short overloading voltages may decrease
the effect. Thus the selection of resistors is not a simple task.
Other problems involved in a skilful design of the h.v. resistor concern
the prevention of too high field gradients within the whole arrangement and,
related to this, is the effect of stray capacitances upon the frequency-dependent
transfer characteristics. To demonstrate these problems the design of a 100-kV
standard resistor described by Park32
will be discussed here. This resistor,
shown in Fig. 3.5, is made up of a hundred 1-M wirewound resistors
connected in series and arranged to form a vertical helix. Some of these
individual resistors are forming resistor elements, as they are placed within
small cylindrical housings predominantly made from metal. Figure 3.6 shows a
cross-section of such a resistor element; the metal cylinders or ‘shields’ enclose
the individual resistors of small size and thus increase the diameter of the
resistors. The metal shield is separated by a gap whose insulation can withstand
and insulate the voltage drop V1 V2 across the element. As the absolute
values of the potentials V1, V2 can be high, the field gradients at the surface of
small wires or small individual resistor units would be too high to withstand the
insulation strength of the atmospheric air used for the construction. Therefore,
the larger diameter of the shields lowers the field gradients to an acceptable
magnitude. A further reduction of these gradients is achieved by the helical

Ch
′
Cp
′
Cp
′
Cc
′
Ce
′
Figure 3.5 100-M , 100-kV standard resistor according to Park32
Metal shield
Insulation
V2
V1
Figure 3.6 Sketch of cross-section of an h.v. resistor element
arrangement, as now the helix might be assumed to form a cylinder of much
larger diameter, across which the potential continuously decreases from the
top to the bottom. These statements could be confirmed by a computation
of the very complex field distribution of the three-dimensional structure. The
h.v. end of the resistor is fitted with a large ‘stress ring’ which again prevents
concentration of electrical field and thus corona or partial discharge formation.
A corona-free design is absolutely necessary to avoid leakage currents, which
would decrease the overall resistance value.
For voltages higher than about 100 kV such an air-insulated design becomes
difficult. The resistor elements then need improved insulation commonly
achieved by mineral oil or highly insulating gases. They have to be placed,
therefore, in insulating vessels. Additional oil or gas flow provided by pumps
will improve the temperature equalization.

Frequency-dependent transfer characteristics
This problem is closely related to the field distribution phenomena. As charges
are the origin and the end of electrostatic field lines, and such field lines will
exist between points of differing potentials, the electrostatic field distribution
may well be represented by ‘stray capacitances’. Such stray capacitances have
been included in Fig. 3.5 showing the 100-kV resistor, and three different
kinds of capacitances are distinguished: the parallel capacitances C0
p between
neighbouring resistor elements within the helix, the stray capacitances to the
h.v. electrode C0
h and the stray capacitances C0
e to earth potential. Thus a very
complex equivalent network is formed which is shown in Fig. 3.7 by assuming
five resistor elements R0
only and neglecting any residual inductances of the
resistors. For equal values of R0
, the real values of the different stray capaci-
tances would not be equal as is assumed. Depending upon the magnitude of the
individual capacitances the ratio I1/V will therefore change with frequency.
As the number of elements used in Fig. 3.7 is too small in reality, a very large
number of results would appear by assuming any combinations of capacitive
elements. Thus an ingenious reduction of the circuit parameters is necessary,
which can be done by assuming homogeneous ladder networks.
Ch
′
Ch
′
Ch
′
Ch
′
Cp
′
Cp
′
Cp
′
Cp
′
Cp
′
R′
R′
R′
R′
R′
I1
Ce
′
Ce
′
Ce
′
Ce
′
V
I
Figure 3.7 Equivalent network of an h.v. resistor
Although such ladder networks are treated in more detail in section 3.6, a
short calculation is included at this point, originally published by Davis.33

C′
C′
C′
C′
C′
C′
C′
R x
X i
x v
I0
I1
V = 0
R
V1
P
C = ΣC′
Figure 3.8 Shielded resistor
This calculation is based upon a ‘shielded resistor’ network, shown in Fig. 3.8.
Here it is assumed that a resistor R of equally distributed resistance values
per unit length dx is enclosed by a metal shield, whose potential is P. In
comparison with Fig. 3.7, the interturn capacitances C0
p are neglected. This
metal shield also suppresses the stray capacitances to h.v. electrode C0
h, and
this structure leads to only one type of stray capacitance C D

C0
which is
uniformly distributed from the resistance to the shield. Taking a point X at a
distance x from the earthed end, the resistance between X and the earthed end
is Rx.
Let the ratio Rx/R D K, so that Rx D RK and an element of resistance dRx D
R dK. The amount of capacitance associated with dRx is then C0
D C dK. If
# is the potential at X and i the current in the resistor at this point,
di D jωC# P dK; d# D iR dK.
and
d2
#
dK2
D R
di
dK
D jωCR# P.
The general solution of this equation is
# D A eaK
C B eaK
C P,

where A and B are constants and a D
p
jωCR. The constants are obtained by
putting
# D V1, where K D 1,
# D 0, where K D 0.
The equation then becomes
# D
eaK
[V1 P1 ea
] eaK
[V1 P1 ea
]
ea
ea C P 3.13
The current i at any point is then
i D
1
R
d#
dK
D
1
R
a
ea
ea [eaK
fV1 P1 ea
g C eaK
fV1 P1 ea
g]. 3.14
Here, the equations for the currents at the earthed end and the h.v. end can be
derived by inserting the appropriate values of K.
The current at the earthed end is obtained by putting K D 0, and is
I0 D
1
R
a
ea
ea [V1 P1 ea
C V1 P1 ea
]
D
a
R sinh a
[V1 P C P cosh a].
By expanding the hyperbolic functions, the result will be:
I0 D
a[V1 P C Pf1 C a2
/2 C a4
/24 C . . .g]
Rfa C a3
/6 C a5
/120 C . . .g
D
V1 C Pa2
/2 C Pa4
/24
Rf1 C a2
/6 C a4
/120 C . . .g
. 3.15
The current I1 at the h.v. end is obtained by putting K D 1 and by similar
treatment
I1 D
V1 C fV1 Pa2
/2g C fV1 Pa4
/24g
Rf1 C a2
/6 C a4
/120g
3.16
The above analysis shows that the current is a function of the shield potential
P and it will be of interest to express the currents for the following two special
cases:

Case I. When P D 0, the uniformly distributed capacitance C is a stray
capacitance to earth, Ce (compare with Fig. 3.7), and the current to ground
becomes
I0 D
V1
Rf1 C a2
/6 C a4
/120 C . . .g
.
The terms containing higher powers of a than a2
may be neglected, as a2
D
jωRC and the following alternating signs as well as decreasing values of the
terms do scarcely contribute. Thus
I0 ³
V1
R

1 C j
ωRCe
6
D
V1
R

1 C

j
ωRCe
6
2

1 j
ωRCe
6

. 3.17
The phase angle between the input voltage V1 and the current to earth is then
ωRCe/6. Similarly, the current at the h.v. end is
I1 ³
V11 C a2
/2
R1 C a2
/6
D
V1
R
ð

1 C
ωRCe
12
C j
ωRCe
3

1 C

ωRCe
6
2
.
For not too high frequencies, we may neglect the real frequency terms, and
thus
I1 ³
V1
R

1 C j
ωRCe
3

3.18
The phase angle becomes CωRCe/3.
For a.c. voltage measurements only eqn (3.17) is important. Apart from
the phase shift the relative change of the current amplitudes with increasing
frequency contains the amplitude errors. We thus may define the normalized
transfer characteristic
H0jω D
I0ω
I0ω D 0
D
1

1 C j
ωRCe
6
.
The amplitude frequency response becomes
H0ω D jH0jωj D 1

1 C

ωRCe
6
2
3.19

This equation shows the continuous decrease of the current with frequency.
The 3 dB bandwidth fB, defined by H0ω D 1/
p
2, is thus
fB D
3
)RCe
D
0.95
RCe
. 3.20
For a decrease of the current amplitude by only 2 per cent, the corresponding
frequency is much lower (³0.095/RC, or one-tenth of fB). An h.v. resistor
for 100 kV is assumed, with a resistance of 200 M and a stray capacitance
Ce of 10 pF. Then eqn (3.20) gives a bandwidth of 475 Hz, demonstrating the
limited accuracy for a.c. measurements. As the resistance values cannot be
reduced very much due to the heat dissipation, only a decrease of Ce can
improve the frequency range.
Case II. One possible way of shielding and thus reducing the stray capaci-
tances to ground would be to raise the potential of the metal shield indicated
in Fig. 3.8. When P D V1/2, the expressions for I0 and I1 can be obtained
in a similar manner as in Case I. Neglecting again in eqns (3.15) and (3.16)
powers higher than 2, we obtain for both currents
I0 ³
V1
R

1 C j
ωRC
12

3.21
I1 ³
V1
R

1 C j
ωRC
12

3.22
Thus the expressions for the two currents are the same. In comparison to
eqn (3.17) the change in the sign of the phase angle should be emphasized.
The output current I0 thus increases in amplitude also with frequency. Such
phenomena are always associated with stray capacitances to h.v. potential
C0
h as shown in Fig. 3.7. However, for h.v. resistors or resistor dividers as
treated in this chapter, cylindrical metal shields of the type assumed cannot be
applied as the external voltage withstand strength would be lowered. But the
calculations demonstrated a strategy to enlarge the bandwidth of such systems.
In Fig. 3.9 two suitable methods are therefore sketched, the efficiency of
which may well be understood from the results of the above calculation.
Figure 3.9(a) shows stress control or grading rings surrounding the resistor.
Apart from the toroid fixed to h.v. potential, the other ring potentials would
float as long as their potentials are not bound to any voltage-dividing system
which is independent of the resistor, i.e. an additional resistor or capacitor
voltage divider (see section 3.6). Apart from the additional cost, such voltage
dividers are again influenced by stray capacitances and thus it is difficult to
control the shield potentials with high accuracy. If the ring potentials are
equivalent to the potentials provided by the current of the resistor at the
corresponding plane of the toroids, the electrostatic field distribution along

V
P = V
P = V
R′= R
R
1
6
6
R
6
R
6
R
X
6
R
6
5
6
4
6
P = V
3
6
P = V
Equipotential
lines
2
6
P = V
1
6
P = V
(a) (b)
Figure 3.9 Suitable methods for the shielding of h.v. resistors or resistor
dividers. (a) Grading rings. (b) Grading top electrode
the resistance would have nearly no field component perpendicular to the
x-direction. Thus all stray capacitances to ground C0
e or h.v. potential C0
h
(Fig. 3.7) are converted to parallel capacitances C0
p, the voltage distribution
of which for a.c. voltages equals exactly the voltage distribution along the
resistor. With a small number of shielding electrodes equal field distribution
can only be approximated.
The top stress ring of the standard resistor in Fig. 3.5 indicates an alternative
method of shielding. The comparison of eqns (3.17) and (3.21) shows opposite
influences of stray capacitances to ground and to h.v. potentials. Therefore a
properly shaped stress control electrode fixed to h.v. potential may also grade
the potentials along the resistor, as sketched in Fig. 3.9(b). For a linearly
distributed resistor in the x-direction, however, an ideal grading is difficult
to achieve. A non-linear resistor distribution originally proposed by Goosens
and Provoost34
for impulse resistor voltage dividers gives an elegant solution
to solve the disadvantage. The numerical calculation of the field distribution
between h.v. electrode and earthed plane would demonstrate, however, the
sensitivity of the distribution to surrounding objects at any potential. Thus
the stray capacitance distribution will change with the surroundings, and will
influence the frequency-dependent transfer characteristics.
Summarizing the above discussions, the high ohmic resistor in series with
an ammeter or the improved method of a voltage dividing system are excellent
means for the measurement of high d.c. voltages and, for resistors of smaller
size and thus lower amplitudes (about 100–200 kV), also a.c. voltages. A
very recent development of a 300 kV d.c. measuring device of very high

Figure 3.10 300-kV divider for d.c. height 210 cm (PTB, Germany)35
accuracy, described by Peier and Graetsch,35
takes advantage of all principles
discussed before (see Fig. 3.10). Here, 300 equal wire-wound resistors each
of about 2 M are series connected, and one of these resistors is used to
form the l.v. arm of a divider (ratio ¾300:1). The resistors are aged by a
temperature treatment. They form a helix of 50 windings and are installed in
a PMMA housing containing insulating oil. The pitch of the helix varies so
that the potential distribution of the resistor column equals approximately the
electrostatic field potential distribution, although the divider is not provided
for the precise measurement of a.c. voltages. Freedom of leakage currents due
to corona was confirmed by partial discharge measurements. A very careful
investigation of all sources of errors and uncertainties for this device shows a
relative uncertainty of š28 ð 106
. The final limit of the uncertainty for d.c.
voltage measurement up to 300 kV is now obviously better than 1 ð 105
, see
reference 132.

3.4 Generating voltmeters and field sensors
Similar to electrostatic voltmeters the generating voltmeter, also known as
the rotary voltmeter or field mill, provides a lossless measurement of d.c.
and, depending upon the construction, a.c. voltages by simple but mainly
mechanical means. The physical principle refers to a field strength measure-
ment, and preliminary construction was described by Wilson,36
who used the
principle for the detection of atmospheric fields which are of small magnitude.
The principle of operation is explained by Fig. 3.11. An adequately shaped,
corona-free h.v. electrode excites the electrostatic field within a highly insu-
lating medium (gas, vacuum) and ground potential. The earthed electrodes
are subdivided into a sensing or pick-up electrode A, a guard electrode G
and a movable electrode M, all of which are at same potential. Every field
line ending at these electrodes binds free charges, whose density is locally
dependent upon the field gradient E acting at every elementary surface area.
For measurement purposes, only the elementary surface areas dA D a of the
electrode A are of interest. The local charge density is then ,a D εEa, with
ε the permittivity of the dielectric.
If the electrode M is fixed and the voltage V (or field-distribution Ea)
is changed, a current it would flow between electrode A and earth. This
V
E
G
G
A i (t)
q (t)
q = ∫A σ da
M
Figure 3.11 Principle of generating voltmeters and field sensors

current results then from the time-dependent charge density ,t, a, which is
sketched as a one-dimensional distribution only. The amount of charge can be
integrated by
qt D

A
,t, a da D ε

A
Et, a da,
where A is the area of the sensing electrode exposed to the field. This time-
varying charge is used by all kinds of field sensors, which use pick-up elec-
trodes (rods, plates, etc.) only.
If the voltage V is constant, again a current it will flow but only if M is
moved, thus steadily altering the surface field strength from full to zero values
within the covered areas. Thus the current is
it D
dq
dt
D
d
dt

At
,a da D ε
d
dt

At
Ea da. 3.23
The integral boundary denotes the time-varying exposed area At and ,a
as well as Ea are also time dependent if the voltage is not constant.
The field lines between h.v. and sensing electrode comprise a capacitive
system. Thus the charge q can be computed by an electrostatic field compu-
tation or by calibration of the system. The integration across the time-varying
area At, however, provides a time-varying capacitance Ct, and also if the
voltage changes with time, qt D CtVt and
it D
d
dt
[CtVt]. 3.24
Various kinds of generating voltmeters use these basic equations and the
manifold designs differ in the constructional means for providing Ct and
interpreting the current it. Such designs and examples can be found in the
literature, see, for example, references 1, 29, 131 and 133.
Generating voltmeters are very linear instruments and applicable over a wide
range of voltages. The sensitivity may be changed by the area of the sensing
electrodes (or iris) as well as by the current instrument or amplification. Their
early application for the output voltage measurement of a Van de Graaff’s thus
may well be understood. Excessive space charge accumulation within the gap
between h.v. electrode and generating voltmeter, however, must be avoided.
The presence of space charges will be observed if the voltage is switched off.
Vibrating electrometers are also generating voltmeters, but will only be
mentioned here as they are not widely used. The principle can well be under-
stood with reference to Fig. 3.11 neglecting the movable disc. If the sensing
electrode would oscillate in the direction of the h.v. electrode, again a current
it D dq/dt is excited with constant voltage V due to a variation of the
capacitance C D Ct. This principle was developed by Gahlke and Neubert
(see reference 30, p. 77). The sensing electrode may also pick up charges
when placed just behind a small aperture drilled in a metal plate. Commercial

types of such an instrument are able to measure d.c. voltages down to 10 µV,
or currents down to 1017
A, or charges down to 1015
pC, and its terminal
resistance is as high as 1016
.
3.5 The measurement of peak voltages
Disruptive discharge phenomena within electrical insulation systems or high-
quality insulation materials are in general caused by the instantaneous
maximum field gradients stressing the materials. Alternating voltages or
impulse voltages may produce these high gradients, and even for d.c. voltages
with ripple, the maximum amplitude of the instantaneous voltage may initiate
the breakdown. The standards for the measurement and application of test
voltages therefore limit the ripple factors for d.c. testing voltages, as the peak
value of d.c. voltages is usually not measured, and claim for a measurement
of the peak values of a.c. and impulse voltages whenever this is adequate.
Up to this point the spark gaps (section 3.1) have been treated to be an
adequate means for the measurement of the peak values of all types of voltages.
The necessary calibration procedure, however, and the limited accuracy are
hindering its daily application and call for more convenient methods. We
could already adequately show the disadvantages encountered with high-ohmic
resistor voltage dividers (see section 3.3) applied to a.c. voltage measurements,
which resulted in limitations within the voltage range of 100–200 kV.
The simplest way to obtain the output peak voltage of a testing transformer
is by measuring and recording the primary voltage and then multiplying the
value by the transformer ratio. However, the load-dependent magnitude of the
ratio as well as unavoidable waveshape variations caused by the transformer
impedances which magnify or reduce the higher harmonics render such a
method unacceptable. Even simpler would be to calculate the peak value of
an impulse voltage from the charging voltage of the impulse voltage generator
multiplied by the voltage efficiency factor . (see eqn (2.28), Chapter 2). Here,
the unknown voltage drops within the generator and the loading effects by the
object under test do not allow, in general, the use of such methods.
The direct measurement of the high voltages across test objects and of their
peak values is therefore of great importance. Many of the methods treated
in this chapter require voltage dividing systems providing adequate voltage
levels for the circuits used to process the peak or crest values. A detailed
study and generalized theory of voltage dividing systems will be presented
in section 3.6. Therefore, within this chapter the voltage divider’s equivalent
circuits are simplified and assumed ideal. A treatment of the construction and
performance of h.v. capacitors for measuring purposes is, however, added to
this chapter, as their application is closely related to the circuits described here.
The measurement of peak voltages by means of oscilloscopes is not
treated in detail. Apart from the measurement of impulse crest values their

application to a.c. voltages is not convenient and thus unusual. For accurate
measurements a very careful adjustment and calibration of the oscilloscope
would be necessary. This, however, is beyond the scope of this book.
3.5.1 The Chubb–Fortescue method
This simple but accurate method for the measurement of peak values of a.c.
voltages was proposed by Chubb and Fortescue,37
who as early as 1913
became interested in the use of a sphere gap as a measuring device. The basic
V(t )
C C
R
OP
I
(a) (b)
ic (t)
+ ic (t) −ic (t)
Figure 3.12 A.C. peak voltage measurement by Chubb and Fortescue.
(a) Fundamental circuit. (b) Recommended, actual circuit
diagram (Fig. 3.12(a)) comprises a standard capacitor, two diodes and a current
integrating ammeter (i.e. moving coil or equivalent instrument) only. The
displacement current ict is subdivided into positive and negative components
by the back-to-back connected diodes. The voltage drop across these diodes
(less than 1 V for Si diodes) may completely be neglected when high voltages
are to be measured. The measuring instrument may be included in one of the
two branches. In either case it reads a magnitude of charge per cycle, or the
mean value of the current ict D C dV/dt, and thus
I D
1
T
t2
t1
ict dt D
C
T
t2
t1
dV D
C
T
VC max C jV maxj
according to Fig. 3.13 which illustrates the integral boundaries and the magni-
tudes related to Fig. 3.12(a). The difference between the positive and negative
peak values may be designated as Vpp, and if both peak values are equal, a

V−max T = 1
f
t2
t1 t
V(t )
V+max
ic(t)
Figure 3.13 Diagram of voltage Vt and current ict from circuit
Fig 3.12(a)
condition which usually applies, we may write
I D CfVpp D 2CfVmax. 3.25
An increased current would be measured if the current reaches zero more
than once during one half-cycle. This means the waveshape of the voltage
would contain more than one maximum per half-cycle. A.C. testing voltages
with such high harmonics contents are, however, not within the limits of
standards and therefore only very short and rapid voltage drops caused by
heavy predischarges within the test circuit could introduce errors. A filtering
of the a.c. voltage by a damping resistor placed between the capacitor C and
the object tested will eliminate this problem.
The relationship in eqn (3.25) shows the principal sources of errors. First,
the frequency f must be accurately known. In many countries the power
frequency often used for testing voltages is very stable and accurately known.
The independent measurement of the frequency with extremely high precision
(i.e. counters) is possible. The current measurement causes no problem, as
these currents are in the mA range. The effective value of the capacitance
should also be accurately known, and because of the different constructions
available, which will be discussed in section 3.5.4, a very high precision is
possible. The main source of error is often introduced by imperfect diodes.
These have to subdivide the a.c. current ict with high precision, this
means the charge transferred in the forward direction, which is limited by
the capacitance C, must be much higher (104
–105
times) than the charge
in the reversed voltage direction. But due to the back-to-back connection
of the diodes, the reverse voltages are low. However, the diodes as well
as the instrument become highly stressed by short impulse currents during
voltage breakdowns. A suitable protection of the rectifying circuit is thus
recommended as shown in Fig. 3.12(b). The resistor R introduces a required

voltage drop during breakdown to ignite the overvoltage protector OP (e.g. a
gas discharge tube).
The influence of the frequency on the reading can be eliminated by
electronically controlled gates and by sensing the rectified current by analogue-
to-digital converters. By this means (see Boeck38
) and using pressurized
standard capacitors, the measurement uncertainty may reach values as low as
0.05 per cent.
3.5.2 Voltage dividers and passive rectifier circuits
Passive circuits are nowadays rarely used in the measurement of peak values
of high a.c. or impulse voltages. The rapid development of very cheap inte-
grated operational amplifiers and circuits during the last decades has offered
many possibilities to ‘sample and hold’ such voltages and thus displace passive
circuits. Nevertheless, a short treatment of basic passive crest voltmeters will
be included because the fundamental problems of such circuits can be shown.
The availability of excellent semiconductor diodes has eliminated the earlier
difficulties encountered in the application of the circuits to a large extent.
Simple, passive circuits can be built cheaply and they are reliable. And, last
but not least, they are not sensitive to electromagnetic impact, i.e. their electro-
magnetic compatibility (EMC) is excellent. In contrast, sophisticated electronic
instruments are more expensive and may suffer from EMC problems. Passive
as well as active electronic circuits and instruments as used for peak voltage
measurements are unable to process high voltages directly and they are always
used in conjunction with voltage dividers which are preferably of the capaci-
tive type.
A.C. voltages
The first adequately usable crest voltmeter circuit was described in 1930
by Davis, Bowdler and Standring.39
This circuit is shown in Fig. 3.14. A
V
V2
C2
C1
R2 Cs Rd Vm
D
Figure 3.14 Simple crest voltmeter for a.c. measurements, according to
Davis, Bowdler and Standring

capacitor divider reduces the high voltage V to a low magnitude. If R2 and Rd
are neglected and the voltage V increases, the storage capacitor Cs is charged
to the crest value of V2 neglecting the voltage drop across the diode. Thus
the d.c. voltage Vm ³ CV2 max could be measured by a suitable instrument of
very high input resistance. The capacitor Cs will not significantly discharge
during a period, if the reverse current through the diode is very small and
the discharge time constant of the storage capacitor very large. If V2 is now
decreased, C2 will hold the charge and the voltage across it and thus Vm
no longer follows the crest value of V2. Hence, a discharge resistor Rd must
be introduced into the circuit. The general rules for the measuring technique
require that a measured quantity be indicated within a few seconds. Thus
the time constant RdCs should be within about 0.5–1 sec. Three new errors,
however, are now introduced: an experiment would readily show that the
output voltage Vm decreases steadily if a constant high voltage V is switched
to the circuit. This effect is caused by a continuous discharge of Cs as well as
of C2. Thus the mean potential of V2t will gain a negative d.c. component,
which finally equals to about CV2 max. Hence a leakage resistor R2 must be
inserted in parallel with C2 to equalize these unipolar discharge currents. The
second error refers to the voltage shape across the storage capacitor. This
voltage contains a ripple discussed in Chapter 2, section 2.1. Thus the error,
almost independent of the type of instrument used (i.e. mean or r.m.s. value
measurement), is due to the ripple and recorded as the difference between peak
and mean value of Vm. The error is approximately proportional to the ripple
factor (see eqn (2.2)) and thus frequency dependent as the discharge time
constant is a fixed value. For RdCs D 1 sec, this ‘discharge error’ amounts to
¾1 per cent for 50 Hz, ¾0.33 per cent for 150 Hz and ¾0.17 per cent for
300 Hz. The third source of systematic error is related to this discharge error:
during the conduction time of the diode the storage capacitor is recharged to
the crest value and thus Cs is in parallel to C2. If the discharge error is ed,
this ‘recharge error’ er is approximately given by
er ³ 2ed
Cs
C1 C C2 C Cs
3.26
Hence Cs should be small compared to C2, which for h.v. dividers is the
largest capacitance in the circuit. There still remains a negative d.c. component
of the mean potential of the voltage V2, as the equalizing effect of R2 is not
perfect. This ‘potential error’ ep is again a negative term, and amounts to
ep D R2/Rd. Hence R2 should be much smaller than Rd.
This leakage resistor R2 introduces another error directly related to the now
frequency-dependent ratio or attenuation factor of the voltage divider. Apart
from a phase shift between V2 and V, which is not of interest, the rela-
tive amplitudes of V2 decrease with decreasing frequency and the calculation

shows the relative error term
efd D
1
2fωR2C1 C C2g2
³
1
2ωR2C22
3.27
Apart from a negligible influence caused by the diode’s inherent junction
capacitance, we see that many systematic error terms aggravate the exact
crest voltage measured.
A numerical example will demonstrate the relative magnitudes of the
different errors. Let C1 D 100 pF, C2 D 100 nF, a realistic measure for a
HVAC divider with attenuation or scale factor of 1000. For RdCs D 1 sec,
the inherent error term ed D 1 per cent for 50 Hz. Allowing an error of
one-half of this value for the recharge error er requires a Cs value C2/3
approximately, and thus Cs D 33 nF. From RdCs D 1 s the discharge resistor
is calculated to be about 30 M . This value is a measure for the high input
resistance of the voltmeter and the diode’s reverse resistance necessary. Let
the potential error ep again be 0.5 per cent. Hence R2 D Rd/200 or 150 k .
For a frequency of 50 Hz this leakage resistor gives efd ³ 2.25 percent. Thus
the sum of errors becomes about 4.25 per cent, still neglecting the voltage
drop across the diode.
Hence, for passive rectifying circuits comprising capacitor voltage dividers
acting as voltage source, at least too small ‘leakage resistors’ (R2) must be
avoided. The possible solution to bleed also the h.v. capacitor is too expensive,
as it requires an additional h.v. resistor. The addition of an equalizing branch
to the l.v. arm of the voltage divider provides an attractive solution. This
can be accomplished again using a peak rectifier circuit as already shown in
Fig. 3.14 by the addition of a second network comprising D, Cs and R, but
for negative polarities. Thus the d.c. currents in both branches are opposite in
polarity and compensate each other. All errors related to R2 are then cancelled.
The most advanced passive circuit to monitor crest values of power
frequency voltages was developed in 1950 by Rabus. This ‘two-way booster
circuit’ reduces the sum of systematic error terms to less than 1 per cent
even for frequencies down to 20 Hz. More information about this principle is
provided in references 1 and 131.
Impulse voltages
The measurement of peak values of impulse voltages with short times to crest
(lightning impulses) with passive elements only was impossible up to about
1950. Then the availability of vacuum diodes with relatively low internal
resistance and of vacuum tubes to build active d.c. amplifiers offered the
opportunity to design circuits for peak impulse voltage measurement but of
relatively low accuracy. Now, active highly integrated electronic devices can
solve all problems involved with passive circuits, see 3.5.3. The problems,
however, shall shortly be indicated by the following explanations.

Impulse voltages are single events and the crest value of an impulse is
theoretically available only during an infinitely short time. The actual crest
value may less stringently be defined as a crest region in which the voltage
amplitude is higher than 99.5 per cent. For a standard 1.2/50 µsec wave the
available time is then about 1.1 µsec. Consider now the simple crest voltmeter
circuit of Fig. 3.14 discussed earlier, omitting the discharge resistor Rd as well
as R2. The diode D will then conduct for a positive voltage impulse applied
to the voltage divider, and the storage capacitor must be charged during the
rising front only. But instantaneous charging is only possible if the diode has
no forward (dynamic) resistance. The actual forward resistance RD gives rise
to a changing time constant RDCs and it will be shown in section 3.6 that a
‘response time’, which is equal to the time constant RDCs for such an RC
circuit, of about 0.2 µsec would be necessary to record the crest value with
adequate accuracy. For a low Cs value of 1000 pF the required RD D 200 .
As also the diode’s junction capacitance must be very small in comparison to
Cs, diodes with adequate values must be properly selected. The more difficult
problem, however, is the time required to read the voltage across Cs. The
voltage should not decrease significantly, i.e. 1 per cent for at least about
10 sec. Hence the discharge time constant of Cs must be longer than 103
sec,
and thus the interaction between the diode’s reverse resistance and the input
resistance of the instrument necessary to measure the voltage across Cs should
provide a resultant leakage resistance of 1012
. A measurement of this voltage
with electrostatic or electronic electrometers is essential, but the condition for
the diode’s reverse resistance can hardly be met. To avoid this problem, a
charge exchange circuit shown in Fig. 3.15 was proposed.
Impulse voltage dividing
system (see section 3.6)
Rt
Cs1 Cs2 V m
Figure 3.15 Peak voltmeter within dashed line with continuous charge
exchange
If the capacitor Cs1 originally charged to crest value transfers most of its
charge to a much larger second storage capacitor Cs2 within a short time,
i.e. much shorter than 1 sec, Cs1 cannot lose much of the charge through the
finite reverse resistance of the diode and the discharge time constant after the
charge transfer is greatly increased because Cs1 and Cs2 are paralleled. As

Cs2 × Cs1, the output voltage Vm becomes quite low and therefore sensitive
(electronic) d.c. voltmeters must be applied. Also, other peak reading devices
must always be combined with active electronic circuits and earlier solutions
are described elsewhere.1
3.5.3 Active peak-reading circuits
Due to the demand within other technical fields, analogue or digital circuits
and instruments are now widely commercially available. The main problem
encountered with these instruments when applied in h.v. laboratories is in
general only related to their electromagnetic compatibility resulting from the
transient disturbances following breakdown phenomena. It is not the aim of
this section to discuss all possible solutions and instruments as available today.
Again, only some hints to basic principles are provided.
The main properties of amplifying circuits may be summarized as follows:
a high and linear input impedance (1–2 M ) is necessary to avoid excessive
loading of the h.v. dividers of any kind. Thus the error terms er, ep and efd
discussed in section 3.5.2 can be minimized. In circuits used for continuous
measurement of a.c. peak voltages, the reduction of the discharge error ed
is much more difficult. In active analogue circuits this may be achieved by
a continuous compensation of the ripple area. To demonstrate the principle
only, a simplified circuit is sketched in Fig. 3.16 related to an actual circuit
of a specialized manufacturer. A voltage attenuator for low voltages (1 kV)
reduces and adapts the input voltages to be processed by the first operational
amplifier OP1, which forms together with the diodes D, the storage capacitor
Cs and the discharge resistor Rd an active peak rectifier. OP1 charges the
storage capacitor Cs and reduces the forward voltage of D to a large extent.
1 MΩ R1 OP 1
D
OP 2
R2
(0.3 µF)
Rd
Cs
(3 MΩ)
t
1
1
1
Digital
voltmeter
OP 4
OP 3
D1
+
−
C
Figure 3.16 Simplified circuit of a crest voltmeter for a.c. voltages with
discharge error compensation (courtesy Haefely Co.)

Cs is discharged through Rd with a time constant of 1 sec. The second voltage
follower OP2 still amplifies the ripple produced by the active peak rectifier.
Its feedback to the connection point of the diodes D, however, avoids leakage
of Cs by the reversed biased diodes. The ripple is detected by the capacitively
coupled operational amplifier OP3, which rectifies the voltage by the diode D1
whose forward conduction voltage is again strongly linearized by the amplifier.
By this means, the ripple voltage appears across an adjustable voltage divider.
The inversion of the output voltage of OP2 and summation with the ripple
voltage performed by OP4 thus eliminate the ripple and thus also the discharge
error to a large extent. The output voltage may then be monitored by an
analogue or digital voltmeter.
The principle of the active peak rectifier can also be used for impulse
voltage measurements. The discharge of the storage capacitor through Rd is
then avoided and replaced by a reset switch. Very high-quality operational
amplifiers with high slew rates are necessary, and the long storage time is
usually achieved by two or three additional active rectifying circuits. More
details may be found in the literature.39–41,128
The increasing availability of specialized, fully integrated analogue and
digital circuits contributed to the development of a large variety of peak
holding circuits. An earlier publication by Schulz42
describes a mixed circuit
for a very precise a.c. peak measurement with a statistical uncertainty of
2.8 ð 104
which includes a capacitor voltage divider for 200 kV, composed
of a pressurized gas capacitor and mica capacitors for the l.v. arm. The
peak detecting circuit consists essentially of a special sample-and-hold ampli-
fier (a.c. to d.c converter) and a very precise digital voltmeter, both being
controlled by a microprocessor. The main aim of this control is to avoid any
discharge error as mentioned earlier. Such precise measuring units are built
for calibration purposes only and not for laboratory applications.
For impulse voltage measurements, the transient recorders comprising fast
parallel ADCs will increasingly be used. The description of the principle
of transient recorders may be found elsewhere43
and a recent publication
by Malewski and Dechamplain44
demonstrates the necessity of additional
shielding of such commercial equipment. Transient recorders are rapidly
replacing the CRO technique for impulse voltage measurements.
3.5.4 High-voltage capacitors for measuring circuits
The important influence of the effective capacitance of any h.v. capacitors as
used, e.g., in the Chubb–Fortescue circuit of section 3.5.1 or in most of the
peak reading circuits for a.c. voltages, makes it necessary to present a short
treatment about the technology of h.v. capacitor units widely used in testing
and research laboratories.
In comparison to h.v. capacitors used within h.v. transmission and distribu-
tion systems for load or series compensation, the requirements for ‘measuring

capacitors’ are different. First, the effective capacitor values are quite low and
range between some 10 and 100 pF only. These low values are sufficient to
provide the energy or power needed for the measurement and to provide low
load for the voltage source. The second requirement is related to the stability
of the C values relative to atmospheric conditions (temperature, humidity),
external fields and voltage range, including all effects associated with this
magnitude, i.e. partial discharges or non-linearity.
An h.v. capacitor may consist of a single capacitance unit, defined basi-
cally as a two-electrode arrangement, or of a chain of capacitor units rated
for relative ‘low’ voltages (kV range) electrically connected in series. The
technology as well as the electrical behaviour is quite different for the two
cases and therefore a separate discussion is appropriate.
Single capacitor units
Ultra high vacuum would provide the ideal dielectric between metal plates
forming an arrangement with known and fixed field distribution. Ultra high
vacuum has excellent electrical strength although it is limited by well-known,
electrode effects. The difficulties and associated costs, however, to place such
electrodes in large vessels or tanks providing ultra high vacuum conditions
without maintenance are the reasons why vacuum is not used for very high
voltages.
According to Paschen’s law (Chapter 5) high electric strength can also be
achieved with gases at high pressure. Atmospheric pressure may be treated as
the lower limit of a high pressure and, dependent upon the type of gas used,
the upper limit is set again by predominantly electrode surface effects which
place an economic limit given by the decreasing relative dielectric strength of
the gas and the increasing cost of pressure vessels. Gases are dielectrics with
predominantly electronic polarization only (see section 7.1), providing a very
low relative permittivity which is not influenced up to very high frequencies
and only by the particle density. Hence a gaseous dielectric is adequate for
the construction of h.v. capacitors.
Thus the problem reduces to finding electrode arrangements which provide
unchangeable and proper field distributions between two electrodes forming
the capacitance. As a certain maximum field strength will limit the insulation
strengths of any gas, a uniform field electrode arrangement (see section 4.2)
would obviously seem to be most convenient. If the centre part of such an
arrangement only would be used to form the effective capacitance, which is
easily possible by subdividing the low potential electrode into a ‘guard ring’
and measuring section, the best field distribution is achieved. The disadvantage
of this solution is, however, the very low value of the capacitance for the
gap distances necessary for the high voltages. It is also difficult to control
exactly the gap distance, if temperature differences and the consequent material
movements are considered.

(b)
(a)
Figure 3.17 Cross-section (a) and (b) typical view of a ‘standard capacitor’
for a voltage of 2 kV (r.m.s.) (courtesy Presco AG, Weiningen, Switzerland)

These disadvantages can be avoided if multiple plate arrangements are used
as sketched in Fig. 3.17(a). A larger number of circular metal plates which
are insulated by, e.g., dry and very clean air or SF6 from each other make it
possible to realize capacitance values up to some nanofarads for voltages up to
some kilovolts. A very careful surface finish of the metal plates is necessary.
The plate arrangement is deposited in a grounded metal vessel and forms the
guard for field control. Very low temperature coefficients of a few ppm/°K
can be reached by a proper selection of the materials used for construction.
Such capacitor units are used as ‘etalons’ or ‘capacitance standards’ as also
the losses are extremely low.
The coaxial cylindrical electrode configuration provides the second oppor-
tunity to achieve a fairly good field distribution, if the difference between
the two radii of the electrodes is not too large. In Chapter 4, section 4.2, the
two-dimensional coaxial field is treated and it is shown that the radii can be
optimized to keep the diameter of the outer electrode as small as possible
for a given voltage and a limited field strength at the inner electrode. Thus,
the radial dimensions do not become very large if the system is pressurized.
As the capacitance C per unit axial length l is C/l D 2)ε/ lnr2/rl, where
r1 and r2 are the radii of the inner and outer cylinders respectively, even
with the optimum ratio r2/r1 D e this capacitance is about 56 pF/m and thus
large enough to achieve adequate capacitance values (30–100 pF) with limited
length of the electrodes. A further advantage relates to the possible variation
of the capacitance if the inner electrode is not completely centralized. The
central position is a position of minimum value of capacitance as shown by
a computation of the capacitance varying with eccentricity according to the
relevant formula,45
C
l
D
2)ε
cosh1 r2
1 C r2
2 D2
2r1r2
, 3.28
where D is the distance between the axes of both cylinders. The expression
shows that a small eccentricity does not contribute much to a change in capaci-
tance. This is the main reason why most of the ‘standard capacitors’ used today
comprise this coaxial cylinder system. Originally suggested by Schering and
Vieweg in 1928,1
a cross-section of such a compressed gas capacitor is shown
in Fig. 3.18. The main h.v. electrode 1 encloses the l.v. electrode with guard
ring 2 completely and thus shields the electrode from the influence of all
external fields. The pressure vessel 5 is of dielectric material and contributes
to minimize the height. The supporting tube 3 is at earth potential. One main
insulation problem involved in this construction relates to the tangential field
distribution outside the dielectric vessel, as the limited electrical strength of
atmospheric air must withstand the strong field concentration in the vicinity of

4
3
5
1
1′
2′
2
Figure 3.18 Cross-section of a compressed gas capacitor (standard
capacitor). 1. Internal h.v. electrode. 10
. External h.v. electrode.
2. Low-voltage electrode with guarding, 20
. 3. Supporting tube. 4. Coaxial
connection to l.v. sensing electrode. 5. Insulating cylinder
the lower end of the h.v. electrode. Even a rough plot of the equipotential lines
surrounding the electrode system may show this field concentration, which is
sketched in Fig. 3.19, a result obtained by Keller.46
The maximum stress
which occurs at the end of the h.t. electrode remains approximately the same,
and is independent of the length of the dielectric cylinder. The reduction of
this external field by simple means is not possible; even the simplest solution
to increase the diameter of the vessel and to distribute the equipotential lines
within the cylinder is difficult due to the necessary increase in mechanical
strength of the vessel construction.
Compressed gas capacitors provide, if well designed and constructed, a
h.v. capacitance of highest possible stability; they are, however, expensive if
designed for voltages of 100 kV and more. Due to their outstanding perfor-
mance with regard to the precision of the capacitance value and very low

100
80
60
a
b
c
V
(%)
40
20
0 20
(%)
40 60
c a b
80 100
Figure 3.19 Potential distribution along a compressed capacitor for
various lengths of h.t. electrode
tan υ values they are predominantly used as the standard capacitor within h.v.
bridges for C tan υ measurements (see Chapter 7). Variations in the construc-
tion are, of course, possible.47–49
These have been built for a.c. voltages up
to 1500 kV. CO2, N2 or SF6 are convenient gases for insulation. SF6 provides
the highest electric strength and thus only pressures up to about 0.4 MPa (in
comparison to 1 to 1.5 MPa with other gases) are necessary. The relative influ-
ence of the pressure-dependent permittivity upon capacitance value may be
calculated taking into account the increase of the relative permittivity εr with
gas density, given by
εr D 1 C ˛
273
100
p
T
;
p in kPa
T in °K
3.29
where
˛ ³ 0.00232 for SF6,
˛ ³ 0.00055 for N2,
˛ ³ 0.00076 for CO2.
As the actual gas density in a vessel may also be influenced by the construction,
the actual variation with p and T will be specified by the manufacturer. Dissi-
pation factors tan υ are in general about 105
for power frequency.

Figure 3.20 shows a physical picture of a standard capacitor for a rated
voltage of 1000 kV. The increased diameter and thickness of the upper part of
the insulating cylinder is made to reduce the electric field stress in the vicinity
of the h.v. electrode outside of this cylinder.
Figure 3.20 Standard (compressed gas) capacitor for 1000 kV r.m.s.
(courtesy Micafil, Switzerland)
‘Stacked’ capacitor units
This second type of basic capacitor construction consists of a large number
of single capacitor units in series. Single units of conventional capacitors
with, e.g., oil-kraft paper or pure solid dielectric insulation cannot be built

for voltages higher than about 10 kV, and hence this series connection is
necessary. These capacitor units are piled up and thus a stretched stack of
large height/diameter ratio is formed.
Whatever the construction of an individual unit, there are always charges
located at some parts of the electrodes which do not contribute to the actual
series capacitance. These (intermediate) electrodes are at a potential which
is essentially given by the terminals of the capacitor elements which form a
voltage-dividing system. The ‘foreign’ charges are thus related to stray capac-
itances in the same way as discussed in section 3.3.
A realistic equivalent circuit of a stacked capacitor unit established from a
certain, usually large number of single capacitors C0
is shown in Fig. 3.21.
For capacitors within laboratories the stray capacitances to earth may also
be influenced by the walls, which are often electromagnetically shielded; this
situation is assumed in this figure. The h.v. lead including the top electrode
of the whole capacitor may contribute to smaller stray capacitances to h.v.
potential V. A calculation of the current I0 flowing to ground from the last
earthed capacitor could be carried out essentially in the same way as the
computation performed for the equivalent circuit of resistors in section 3.3,
eqns (3.13)–(3.18). Assuming similar approximations within the expressions
for the current I0 we achieve by expansion of the hyperbolic functions the
following result:
I0 D VωC
1 C
Ch
4C
1 C
Ce
6C
C
Ch
12C
D VωCeq 3.30
Ground
I0
Wall
C′
C′ Ce
′
Ch
′
V
Figure 3.21 Stray capacitances influencing the equivalent or effective
capacitance of h.v. capacitors

where C D C0
/n, Ce D nC0
e, Ch D nC0
h, and n a large (infinite) number of
capacitor elements C0
. The capacitance C D C0
/n is obviously the resultant
capacitance of the chain computed from the usual series circuit, i.e. 1/C D
1/C1 C 1/C2 C . . . 1/Cn. Equation (3.30) demonstrates that the ‘equivalent’
or ‘effective’ capacitance Ceq of a stacked capacitor cannot be calculated
from individual elements, as the stray capacitances Ch and Ce are not exactly
known. Therefore, the l.v. capacitor C2 within the voltage divider of Fig. 3.14
or the diodes within the Chubb–Fortescue circuit (Fig. 3.12) will only ‘see’
the current I0.
Many measurements performed with huge stacked capacitor units50
confirmed the fundamental applicability of eqn (3.30). It was also shown that
the influence of the stray capacitance to the h.v. side, C0
h, in general can be
neglected. Nevertheless, it is necessary to rate the series capacitors so that
the term Ce/6C does not exceed 1 to 2 per cent. Thus we may simplify
eqn (3.30) to
Ceq ³ C

1
Ce
6C

. 3.31
This effect of decreasing capacitance can experimentally be checked by
a correct measurement of Ceq with an h.v. bridge (Schering or current
comparator bridge, see Chapter 7). In such bridge circuits, the unknown
capacitor is placed during measurements at its working condition. If the high
voltage is applied, the l.v. end of this unknown capacitor remains essentially
at earth potential, as the bridge potential is very low; hence, the potential
distribution across the test object remains unchanged. A measurement of Ceq
with a usual two-terminal capacitance bridge should never be made and would
indicate wrong results.
The dimensioning of stacked capacitor units for the measurement of high
voltages must take this effect into consideration. Ce can approximately be
calculated by the assumption that the stacked capacitors are of cylindrical
shape, thus forming a metalized vertical cylinder placed upon a horizontal
plane, as sketched in Fig. 3.22.
The well-known formula for this arrangement51
is
Ce D
2)εl
ln

2l
d

4s C l
4s C 3l


3.32
and for s − 1:
Ce ³
2)εl
ln
1.15l
d
. 3.33

d
s
Plane
Figure 3.22 Cylindrical conductor above plane (see eqn (3.32))
Since even more accurate approximations will not contribute much to the result
as shown by Zaengl51
and Luehrmann,52
one may evaluate eqn (3.33) only,
and one can easily see that the absolute values Ce/l range within about 10 to
15 pF/m.
The effect, that the effective capacitance, eqn (3.31), will change with the
dimensions of the capacitor or its surroundings is known as ‘proximity effect’.
The accurate influence of the surroundings on Ce could only be computed by
numerical field calculation (see Chapter 4). As the variations of Ce due to
changed surrounding conditions are usually less than 20 per cent, the error
term in this equation may be as large as 5 per cent to get variations of Ceq
smaller than 1 per cent. This condition leads to
Ce
6C
D
Ce/l
6C/l
5%.
With Ce/l D 10–15 pF/m we obtain
C/l ³ Ceq/l ¾
D 30–50 pF/m. 3.34
If this condition is not fulfilled, capacitors within measuring circuits should be
fixed and placed within a laboratory and no moveable structures or equipment
disturbing the potential distribution of the capacitor should be in the vicinity.
Technology of H.V. capacitors
The technology of compressed gas standard capacitors was treated earlier.
The following explanations refer thus to the technology of ‘stacked’, discrete
capacitors with special emphasis on those used for measurement purposes.
The main requirements for this kind of application are:

ž the capacitance C shall be independent of magnitude of voltage level and
shall not change with time of application (no ageing effects);
ž the temperature coefficient (TC) (in ppm/°K) shall be small or very small,
dependent on the kind and temperature range of application, and shall at
least be known;
ž the effective inductivity of C shall be as small as possible, if used for
high-frequency applications, i.e. voltage dividers for impulse voltages.
Foil capacitors. Most of the discrete capacitor units for stacked capacitors
are made of long strips of at least two layers of, e.g., ‘paper’ and one thin layer
of Al foil. The condenser is then made by winding up two such strips to form
a roll. The rolls are then in general pressed flat to reduce the size, but may
also be used in cylindrical form. A suitable number of units are assembled
and pressed into an insulating case. After assembly the condenser is vacuum
dried and impregnated with a dielectric fluid. Instead of a high-quality paper
(‘condenser paper’, 10 to 30 µm thick) plain plastic films (e.g., polystyrene) or
mixtures with plastic films and paper are now in use. Rolled condensers will
have high inductance, if the metal electrodes are only joined at their ends or
at some intermediate positions. Very low inductivity can only be gained if the
technique of ‘extended foil construction’ is applied. In this method each of
the two metal foil electrodes are on one side extended beyond the dielectric
foils and after rolling and pressing continuously joined (welded) on each side
to form the end caps.
Ceramic capacitors. Only a few manufacturers in the world are able
to produce excellent and special types of ceramic capacitors applicable
for moderate high voltage. Most types of these ‘ceramic radio frequency
power capacitors’ are made from ‘class/type 2’ ceramic. The dielectric
material used for this kind of ceramic contains mostly alkaline earth titanates
giving quite high values of permittivity. This material is prone to ageing
caused by a gradual realignment of the crystalline structure, has very high
temperature coefficients (TC), is voltage dependent and is not recommended
for measurement purposes. There are, however, some kinds of ‘class/type 1’
ceramic without these disadvantages. Within this class 1 ceramic, mainly
manufactured from titanium dioxide or magnesium silicate, the materials with
relative permittivities below 1000 show TC values down to some 10 ppm/°K
within a temperature range of 25 to C95°C and very low dissipation factors
up to the Megahertz range. The main advantage of all types of ceramic
capacitors is the extremely low inductance due to their design, which is either
of tubular or plate/pot type with conductive noble metal electrodes. The only
disadvantage is due to the limitations in rated continuous a.c. voltage (up to
about 40 kV peak) and capacitance (up to about 6 nF) per unit.
Both types of capacitors are in general use for voltage dividing systems.

3.6 Voltage dividing systems and impulse voltage
measurements
The measurement of impulse voltages even of short duration presents no
difficulties, if the amplitudes are low or are in the kilovolt range only. The
tremendous developments during the last three decades related to the technique
of common CROs, digital scopes or transient recorders provide instruments
with very high bandwidth and the possibility to capture nearly every kind of
short-duration single phenomena. Although the usual input voltage range of
these instruments is low, h.v. probes or attenuators for voltages up to some
10 kV are commercially available.
The problems arise with much higher voltages and it is well known that
impulse voltages with magnitudes up to some megavolts are used for testing
and research. The voltage dividers necessary to accommodate these voltages
are specialized apparatus, and there are only a few manufacturers throughout
the world who are ready to produce such dividers with adequate accuracy. Self-
provided constructions are often adequate if the problems are known. But also
the application of such voltage dividers needs a fundamental understanding of
the interactions present in voltage dividing systems. Hence an attempt is made
to introduce the reader to the theory as well as to some hints on constructional
details on this quite difficult field of h.v. measuring techniques.
We will start with a generalized voltage generation and dividing system and
briefly discuss the layout (section 3.6.1). Depending upon the voltage shape to
be measured, the requirements related to the whole measuring system must be
well defined (section 3.6.2). A generalized analytical treatment of the transfer
characteristics of this system involves the complex interactions between the
different parts of the circuit (section 3.6.3). The theory of the ‘isolated’ voltage
dividers as the most essential part of the circuit demonstrates the different types
of these devices and their possible applications (section 3.6.4). For fast tran-
sient voltages the interactions between the dividers and their adherent circuits
are briefly discussed and methods for the evaluation of the transfer properties
are presented (section 3.6.5). Some advice on a proper design of the 1.v. arm
of the voltage dividers is given (section 3.6.6). As the transient digital recorder
has recently become the most powerful tool for the evaluation of impulse volt-
ages, an up-to-date introduction in this kind of instrument is provided in a sepa-
rate section (see 3.7), which is partly still related to voltage dividing systems.
3.6.1 Generalized voltage generation and measuring circuit
Figure 3.23 illustrates the common and most adequate layout of any voltage
testing circuit within an h.v. testing area. The voltage generator 1 is connected
to a test object 3 by a lead 2. These three elements form a voltage generating

Voltage generating system Voltage measuring system
1
5
2
8
3 4
8
6
7
Figure 3.23 Basic voltage testing system. 1. Voltage supply. 2. Lead to test
object. 3. Test object. 4. Lead to voltage divider. 5. Voltage divider. 6. Signal
or measuring cable. 7. Recording instrument. 8. Ground return
system. The lead 2 to the test object may comprise any impedance or resistance
to damp oscillations, if necessary, or to limit the short-circuit currents if the
test object fails. The measuring system starts at the terminals of the test object
and comprises a connecting lead 4 to the voltage divider 5, and a recording
instrument 7, whose signal or measuring cable 6 is placed between its input
terminals and the bottom or l.v. part of the divider. The appropriate ground
return 8 should assure no significant voltage drops for even highly transient
phenomena and keep the ground potential to earth as close as possible.
This layout is sometimes altered and there can be acceptable reasons for
such a change. For d.c. voltages and small currents drawn by the test object,
the voltage divider can be incorporated within the voltage supply, if the lead 2
has no or only a protecting resistance, the voltage drop across which can be
neglected. Essentially the same statements are applicable to low-frequency a.c.
voltages, but a possible influence of the lead inductance should be checked. In
practice, also for impulse voltage testing circuits the voltage divider may form
part of the impulse generator. The simple reasons can easily be understood
from the impulse generator circuits (see Chapter 2, Fig. 2.25). There, the wave
shaping load capacitance C2 is often combined with an l.v. capacitor connected
in series, thus forming an adequate voltage divider. An undamped connection
to the object under test then leads to the erroneous assumption that negligible
voltage drop can occur across the lead. This assumption may be correct for
slowly rising impulse voltages and quite short leads. Connecting leads with
lengths of many metres, however, are often used and thus this assumption may
become unacceptable. It must be remembered that the test object is a capacitor
and thus the circuit formed by the lead and test object is a series resonant
circuit. These oscillations are likely to be excited by firing the generator,
but will only partly be detected by the voltage divider. Completely wrong
is the assumption that such a voltage divider being a part of the generator

is measuring the correct voltage across the test object following a voltage
collapse or disruptive discharge. The whole generator including voltage divider
will be discharged by this short-circuit at the test object and thus the voltage
divider is loaded by the voltage drop across lead 2. This lead forms to first
approximation an inductance, and hence the oscillatory discharge currents
produce heavy (induced) voltage oscillations which are then measured by
the capacitor divider. These voltages are often referred to as overvoltages
across the test object, but this statement is incorrect. For the measurement
of predominantly lightning impulses, therefore, only the layout of the circuit
according to Fig. 3.23 shall be used if an accurate measurement of full and
chopped voltages is desired.
There is a further reason for placing the voltage dividers away from any
energized objects. High-voltage dividers consist of ‘open’ networks and cannot
be shielded against external fields. All objects in the vicinity of the divider
which may acquire transient potentials during a test will disturb the field distri-
bution and thus the divider performance. The lead from the voltage divider
to the test object 4 is therefore an integral part of the measuring system. The
influence of this lead will theoretically be treated in section 3.6.3. There it will
be established that a damping resistor at the input end of this lead contributes
to improved transfer characteristics of the system.
In order to avoid heavy electromagnetic interactions between the recording
instrument and the h.v. test area as well as safety hazards, the length of
the signal cable 6 must be adequately chosen. For any type of voltage to
be measured, the signal cable should be of a coaxial and shielded type. The
shield or outer conductor picks up the transient electrostatic fields and thus
prevents the penetration of this field to the inner conductor. Although even
transient magnetic fields can penetrate into the cable, no appreciable voltage
(noise) is induced due to the symmetrical arrangement. Ordinary coaxial cables
with braided shields may well be used for d.c. and a.c. voltages. For impulse
voltage measurements, these cables must provide very low losses to reduce
distortion of the voltage pulses to be transmitted. As it is impossible to avoid
induced currents within the cable shields which are not related to the trans-
mitted signal, these currents can heavily distort these signals if the so-called
coupling impedance of the cable is not very low. In the frequency domain,
this impedance Zcω is defined by
Zcω D
Vn/l
Id
, 3.35
where Id is the disturbing current flowing in the shield, and Vn/l the voltage
drop at the inner surface of the shield per unit length l of the cable. More
information about the origin of disturbing cable shield currents may be found
in references 1 and 54 and in other publications.55,56
For a pure d.c. current
within the shield, the coupling impedance is given by the voltage drop due to

the d.c. resistance of the shield. If the frequency of these currents increases, the
coupling impedance will continuously decrease if the shield is of rigid cross-
section; then the eddy currents will attenuate the current density at the inner
surface of the cylindrical shield. Hence rigid or corrugated shields, i.e. flexwell
cables, are best suited for noise reduction. For braided shields, the coupling
impedance is in general not a stable quantity, as the current distribution within
the shield is likely to be influenced by resistive contacts within the braid.
Double-shielded cables with predominantly two insulated braided shields will
improve the behaviour. Best conditions are gained by placing the coaxial
cable into an additional, non-braided metal tube, which is connected to ground
potential at least at the input end of the measuring cable and also at its end.
In Fig. 3.23 there is finally the ground return 8. For h.v. test circuits disrup-
tive discharge must always be taken into account. Large and heavily oscillating
short-circuit currents are developed and hence every ground return with simple
leads only cannot keep the voltage drops small. The impedance, therefore, must
be reduced. Large metal sheets of highly conducting material such as copper
or aluminium are best. Many h.v. laboratories provide such ground returns in
combination with a Faraday cage for a complete shielding of the laboratory.
Expanded metal sheets give similar performance. At least metal tapes of large
width should be used to reduce the impedance. A parallel connection of tapes
within flat areas will further decrease the inductance and thus approximate the
efficiency of huge metal sheets.
Information concerning the layout of testing and measuring circuits is also
provided in reference 57. The measuring system thus comprises four main
components with quite different electrical behaviour. The simulation of these
components will depend upon the necessary frequency range to measure the
voltage across the test object. An evaluation of this frequency range shall thus
precede this simulation.
3.6.2 Demands upon transfer characteristics of the
measuring system
The voltage measuring system defined in Fig. 3.23 is a four-terminal network
and can thus be represented as shown in Fig. 3.24. Vi indicates the voltage
across the test object (3 in Fig. 3.23), and the output voltage V0 appears at
the recording instrument, i.e. at the screen of a CRO or transient recorder.
The input voltages Vi are either continuous steady state voltages for d.c. and
a.c. generating systems, or single events for impulse voltages. In both cases,
the instantaneous amplitudes will change with time, even for d.c. voltages
with a periodic ripple.
For a sinusoidal input voltage vit D Vmi sinωt C #i the magnitude Vm0
and phase angle #0 of the output voltage v0t D Vm0 sinωt C #0 can be
determined either by calculation with known network parameters or by

Ii I0
Vi [A] Vo
Figure 3.24 Representation of the measuring system as a four-terminal
network
measurements, although such measurements are difficult to perform for very
high ratios of Vmi/Vm0. The frequency response of the system can then
be subdivided into an amplitude (frequency) response Hω D Vm0/Vmi and
a phase (frequency) response #ω D #0ω #iω. It is well known that
both quantities are also displayed by assuming complex amplitudes Vi D
Vmi expj#i and V0 D Vm0 expj#0, and the system transfer or network
response function
Hjω D
V0
Vi
D jHjωj expfj[#0ω #iω]g 3.36
where jHjωj D Hω as defined above.
Neither d.c. voltages with ripple nor a.c. testing voltages are pure sinu-
soidal, but periodic in nature. The input voltages may then be described by
a – in general – limited number of complex amplitudes Vik obtained by the
application of Fourier series,
Vik D
1
T
T/2
T/2
vit expjkωt dt 3.37
D jVikj expj#ik,
where ω D 2)/T, T is the time period and k are discrete numbers. The periodic
input quantity is thus analysed into sinusoidal frequency components, and the
complex amplitudes are displayed by the amplitude line spectrum jVikj and
the angular frequency line spectrum. For every component with the frequency
ωk D kω, the network response may easily be found with eqn (3.36), and the
responses can be summed up using the principle of superposition. Applying
again the complex form of the Fourier series, this summation gives:
v0t D
1

kD1
VikHjωk expjkωt. 3.38
A direct comparison between v0t and vit can thus be made and the errors
evaluated.

For the single events of impulse voltages, only an infinite number of sinu-
soidal voltages are able to represent the input voltage vit. This continuous
frequency spectrum is defined by the Fourier integral or Fourier transform
of vit
Vijω D
1
tD1
vit expjω8 d8 3.39
and contains amplitude and phase spectra. The linearity and homogeneity
of the time invariant systems assumed enable us again to calculate the time
response of the system by a convolution of the continuous frequency spectrum
with the network response function and the transition from frequency to time
domain by means of the inverse Fourier transform:
v0t D
1
2)
1
ωD1
VijωHjω expjωt dω. 3.40
In practice, the real input quantity vit is not known, as only v0t can be
measured. This output voltage, however, has suffered from the loss of infor-
mation contained in Hjω. No appreciable transmission errors could occur, if
at least the amplitude frequency response Hω D jHjωj would be constant
within a frequency range, in which the line or continuous frequency spectra,
Vik or Vijω, cannot be neglected. Thus the computation of the spectra of
an estimated input quantity is a very efficient tool to judge the necessary
frequency range or bandwidth of our measuring system and its individual
components.
The highest demands upon the measuring system transfer functions are
clearly imposed by impulse voltages. The analysis of the impulse voltage
generating circuits (see Chapter 2, section 2.3.1) displayed a waveshape of the
generator output voltage, which is a double exponential function. Neglecting
the possible interactions between the voltage measuring and generating
systems, we thus may assume an input voltage for the measuring system,
given by vit D A[expt/81 expt/82], where A is a constant value
and 81, 82 the time constants according to eqn (2.27). This voltage can be
chopped at any instantaneous time Tc as defined in Fig. 2.24 caused by a
disruptive discharge of the test object, but the voltage collapse is extremely
rapid. The input voltage is then given by
vit D

0 for 0; t Tc
A[expt/81 expt/82] for 0 t Tc.
3.41
Applying this voltage and its boundary conditions to eqn (3.39) gives Vijω.
The calculation implies no fundamental difficulties; the result, however, is
lengthy and is obtained as
Vijω D ARe C jIm 3.42

where
Re D
81
1 C ω812
f1 C [ω81 sinωTc cosωTc] expTc/81g
. . .
82
1 C ω822
f1 C [ω82 sinωTc cosωTc] expTc/82g;
Im D
81
1 C ω812
fω81 [ω81 cosωTc C sinωTc] expTc/81g
. . .
82
1 C ω822
fω82 [ω82 cosωTc C sinωTc] expTc/82g.
For the special case of a non-chopped voltage Tc ! 1, the Fourier
transform of the input voltage is merely
Vijω D A

81
1 C ω812

82
1 C ω822

j
ω82
1
1 C ω812

ω82
2
1 C ω822

. 3.43
The numerical evaluation of eqns (3.42) and (3.43) is shown in Fig. 3.25 for
a full lightning impulse of 1.2/50 µsec (81 D 68.2 µsec: 82 D 0.405 µsec) and
different instants of chopping, Tc. A normalization was made by vijω D
Vijω/Viω D 0 and only the relative amplitudes jvijωj D viω are
displayed on a linear scale. From the result the following conclusions can
be made.
The relative amplitudes for a full lightning impulse (Tc ! 1) become
already very small in a frequency range of about 0.5–1 MHz; hence an
Vi (t) 1.2/50 ms
Tc t
Tc = 0.1 ms
Tc = 0.5 ms
Tc = 4 ms
Tc = ∞
0.02 0.1 1 10
f (MHz)
0
0.5
1
|
v
i
(
i
w)|
Figure 3.25 Normalized amplitude frequency spectra (Fourier transform) of
a lightning impulse voltage of 1.2/50 µsec, wave full and chopped

amplitude frequency response of our measuring circuit, which is really flat
up to this frequency range, would obviously not provide significant errors.
Depending upon the decay of the amplitude frequency response, the bandwidth
(3 dB point) has to be much higher, i.e. about 5–10 MHz.
The chopping of the voltage introduces a heavy increase of the harmonics
content. For Tc D 4 µsec, i.e. a chopping at the impulse tail, an accurate
measurement of the crest voltage may still be provided by the above-mentioned
amplitude response, although appreciable errors might appear during the instant
of chopping. The voltages chopped within the front Tc D 0.5–0.1 µsec,
however, will require a very wide bandwidth which must obviously increase
with decreasing chopping time. Desirable values of fB for Tc D 0.5 µsec only
shall obviously reach magnitudes of about 100 MHz, but such large values
cannot be achieved with measuring systems for very high voltages.
This frequency domain method described so far for determining a transfer
characteristic quantity to estimate measuring errors is difficult to use, as the
two quantities, Hω and #ω, are difficult to measure due to the large ‘scale
factors’ of the measuring systems. For h.v. measuring systems, the transfer
characteristic is therefore evaluated by means of a measured (experimental
unit) ‘step response’.57,53
This time-domain method is based upon the fact
that the Fourier transform (eqn (3.39)) of a single-step function is proportional
to 1/jω and thus all frequencies are contained. Let us, therefore, represent the
input voltage of our measuring system by such a step function:
vit D

0 for t 0
Vmi for t 0.
3.44
The output voltage of the measuring system, v0t, is then much smaller in
amplitude, it may also be time delayed with reference to the voltage input,
and it will be distorted mainly at its front. This ‘unit step response’ (USR)
is denoted by the term Gt and is sketched in Fig. 3.26(a). The time t D 0
is defined by eqn (3.44), the time delay described by 8de, and for a good
measuring system the final value Vmi1/N will be reached in a short time.
The magnitude N indicates the steady state voltage ratio between input and
output voltage, i.e. the scale factor of the system.
In section 3.6.3, Gt will be calculated based upon equivalent circuits. This
quantity is also easy to measure by means of sensitive recorders or CROs. With
a known value of the USR, Gt, the output voltage response to any arbitrary
input voltage can be calculated from the superposition theorem or Duhamel’s
integral:
v0t D vitGC0 C
t
0
vi8G0
t 8 d8
D GtviC0 C
t
0
v
0
it 8G8 d8. 3.45

G(t)
Vmi (1/N)
0 tde ts t
g (t)
1
0 tde ts t
T1
T2
T3
T4
T 0 = T1 − T2 + T3 − T4 ⋅⋅⋅
(a)
(b)
Figure 3.26 Unit step response and definition of the response time T0
.
(a) Unit step response as output voltage. (b) Normalized unit step response
where G0
t 8 or v0
it 8 is the derivative of Gt or vit with respect
to 8. This integral can always be solved numerically by digital computers, if
analytic expressions are not available.129
The chopping of a lightning impulse voltage at the front (Tc 1 µsec
in Fig. 3.25) is sometimes used for h.v. testing and the demands upon the
measuring circuits become severe. The chopping on front provides a nearly
linearly rising voltage up to Tc. Let us assume an ideally linearly rising voltage,
vit D St, 3.46
where S is the steepness. With eqn (3.45), the output voltage becomes
v0t D S
t
0
G8 d8 D
S
N
t
0
g8 Ð d8 3.47
where gt is the normalized quantity of the USR voltage Gt, whose final
value becomes thus 1 or 100 per cent. Then the term Nv0t represents the
high voltage comparable to vit of eqn (3.46), and we may introduce this
term into eqn (3.47) and expand this equation to
Nv0t D S

t
t
0
[1 g8] d8

. 3.48

This expression relates the output to the input voltage as long as (St) increases.
The integral term will settle to a final value after a time 8s indicated in
Fig. 3.26. This final value is an interesting quantity, it shows that differences
in amplitudes between input (St) and magnified output voltage Nv0t remain
constant. Hence we may write
vit Nv0t D S
t8s
0
[1 g8] d8 D S
1
0
[1 g8] d8 D ST0
3.49
where
T0
D
1
0
[1 g8] d8 3.50
is the ‘response time’ of the measuring system. This quantity gives the time
which can be found by the integration and summation of time areas as shown
in Fig. 3.26(b). T0
includes a real time delay 8de of the output voltage, which
is in general not measured, if the time instant of the application of the unit step
input is not recorded. The former IEC Recommendations57
and the newest
IEC Standard 60-253
therefore neglect this time delay. The justifications for
neglecting this delay are shown in Fig. 3.27. There, the linearly rising input
voltage is suddenly chopped, and the output voltage multiplied by N is approx-
imately sketched for the USR of Fig. 3.26. Equation (3.48) can be applied up
to the instant of chopping, Tc; for later times, eqn (3.45) must be rearranged,
and it can easily be seen that a superposition of three terms (response to St,
negative USR with amplitude STc, and negative response to St for t Tc)
will govern this output voltage.
V(t)
0 tde
tde
t1 Tc t
Vi (t) T 0
T
∆V
V0 (t)⋅N
(recorded voltage)
Figure 3.27 Measuring error V for linearly rising voltages chopped at
Tc. Definition of response time T0
and T

As the sudden change in the output voltage is also delayed, the amplitude
error V is obviously given by
V D viTc Nv0Tc C 8de D ST0
8de D ST
if Tc 8s. Thus the simple relationship
T D T0
8de 3.51
exists, where T is equal to a response time similar to T0, but integrated from
Fig. 3.26 by
T D
1
8de
[1 g8] dt. 3.52
The relative amplitude error υ for a chopped linearly rising voltage thus
becomes
υ D
V
STc
D
T
Tc
. 3.53
For T D 50 ns, and Tc D 0.5 µs, this error is 10 per cent.
Clearly, this simple qualification criterion for a measuring system has some
drawbacks. First, eqn (3.53) can only be used if the assumptions (linearly
rising voltage, time to final value or settling time 8s Tc, ideal chopping) are
fulfilled. Ideal, linearly rising high voltages, however, are difficult to generate
and it is even more difficult to confirm this linearity by measurements, as the
measured values are not accurate.50
Due to its definition, the response time T
or T0
can even be negative, see section 3.6.5. Such problems could easily be
demonstrated by a systematic evaluation of eqn (3.45) but only one example
is shown in Fig. 3.28. This example does not need further explanations and
additional information is given in section 3.6.7.
In spite of these disadvantages, the concept of response times cannot be
disregarded and it is again used in the latest IEC Standard related to measuring
systems.53
The biggest advantage is related to its computation: it will be
demonstrated in the next section that T0
or T can be calculated analytically
even for very sophisticated networks without the knowledge of gt. Hence,
the value of this quantity giving a measure to quantify transfer properties of a
measuring system must be acknowledged, although additional response para-
meters can be used to correlate step responses with other types of measuring
errors.53,129,130
3.6.3 Fundamentals for the computation of the measuring system
Any analysis of a complex and sophisticated network as sketched in Fig. 3.23
either in the frequency or time domain is based on an equivalent circuit,

−80
−60
−40
−20
0
20
40
60
80
100%
a: T = 100 ns; fr = ∞
b: T = 40 ns; fr = 2.52 MHz
c: T = 1 ns; fr = 10 MHz
c
b
a
Vi (t)
0.1 0.5
t (ms)
R L
C v0 (t)
vi (t)
C = 100 pF
a: R = 1 k Ω; L = 0;
b: R = 400 Ω; L = 40 mH;
c: R = 10 Ω; L = 2.5 mH;
V
i
(
t
)
v
0
(
t
)
Figure 3.28 Computed response V0 t of an R–L–C circuit with given
parameters to a linearly rising input voltage Vit chopped at Tc D 0.5 µsec
which may represent, as closely as possible, the physical reality. Although
being aware that the execution of this task is most difficult for h.v. measuring
systems due to their dimensions and complex electromagnetic fields involved,
we have already simulated the system by a simple ‘four-terminal’ or ‘two-port’
network as shown in Fig. 3.24. The analysis or computation of any numer-
ical evaluation of results can be performed nowadays by adequate computer
programs in time or frequency domain. The disadvantage of this method is,
however, that the influence of individual network parameters is quite diffi-
cult to identify. Thus we use as far as possible an analytical treatment of our
systems based on the general network theory.
The representation of the actual measuring system (Fig. 3.23) by a four-
terminal network imposes, however, certain restrictions. As demonstrated later,
the theory of travelling waves or distributed parameters is also used to evaluate
the behaviour of the system during transients, and thus it is assumed that the
electromagnetic phenomena are quasi-stationary in the two directions perpen-
dicular to the direction of wave propagation. These conditions are somewhat
limiting the validity of the calculations when the dimensions of the measuring
systems are too large. The limitations are obviously directly correlated with the

definition of a voltage as an independent quantity from space within an elec-
tromagnetic field, and as the dimensions of our h.v. systems are in the range of
metres, the quasi-stationary nature of the electromagnetic phenomena is more
or less restricted. For example, the travelling time of a wave at the velocity
of light is 20 nsec between two points 6 metres apart. If impulses chopped on
the front at Tc D 200 nsec are considered, the time is only ten times longer
than the field propagation time.
With these restrictions in mind, we nevertheless may start a generalized
computation of our four-terminal network, Fig. 3.24, and apply the results
to equivalent circuits later on. The Laplace transform will now be used
throughout the treatment, with the complex frequency s D , C jω being the
Laplace operator. Input and output parameters can be described by the matrix
equation

Vis
Iis

D

A11s; A12s
A21s; A22s

Ð

V0s
I0s

D [A]

V0s
I0s

3.54
where [A] is the network matrix of the system defined by this equation.
The measuring system will load the generating system and thus the input
impedance of the measuring system is sometimes necessary. As the output
current I0 for a voltage dividing system with large ratios or scale factors N
cannot influence the input, the condition I0 D 0 can always be assumed. From
eqn (3.54) the input impedance is
Zis D
Vis
Iis
D
A11s
A21s
. 3.55
The most important quantity is the voltage transfer function. For I0 D 0, this
function becomes
Hs D
V0s
Vis
D
1
A11s
. 3.56
Embedded in this function is the scale factor N of the voltage dividing system.
This factor or ratio is a constant quantity for moderate frequencies only and
hence we may derive this ratio by
N D lim
s!0

Vis
V0s

D lim
s!0
[A11s] D A110. 3.57
The voltage transfer function, eqn (3.56), is conveniently normalized by N.
Denoting the normalization by hs, we obtain
hs D NHs D
A110
A11s
. 3.58

The unit step voltage Gt, as described and defined in section 3.6.2, can
be found by applying the Laplace inverse integral to the transfer function
multiplied by 1/s, the Laplace transform of a unit step. Thus
Gt D L1

1
s
Hs

D L1

1
sA11s

. 3.59
From eqn (3.58), the normalized unit step response is
gt D NGt. 3.60
For very complex transfer functions often involved in mixed distributed
parameter circuits, the applicability of eqn (3.59) is restricted, as it is too
difficult to find solutions in the time domain. Then the response time T0
cannot
be computed by eqn (3.50). Based upon a well-known final value theorem of
the Laplace transform, which is
lim
t!0
ft D lim
s!0
[sFs],
we may compute the response time from the following equation, which can
be derived by applying this final value theorem to eqn (3.49):
T0
D lim
s!0

1 hs
s

.
As lim
s!0
hs O
D 1 by definition, the rule of Bernoulli–l’Hôpital leads to
T0
D lim
s!0

dhs
ds

D lim
s!0
[h0
s]. 3.61
The final value theorem contains some restraints, i.e. ft and df/dt must be
Laplace transformable and the product sFs must have all its singularities in
the left half of the s-plane. Equation (3.61) thus may fail sometimes.
The response time T can be computed from eqn (3.51), if 8de is known. It
may be difficult, however, to predict an actual time delay based upon hs only.
The comparison of experimental and thus actual time delays with computed
results may suffer from this disadvantage; for more information about this
very specialized question, the reader is referred to the literature.50
These general results can now be applied to more detailed measuring
circuits. Numerous equivalent circuits could be presented. We will, however,
follow a representation, developed by Asner,58
Creed et al.59
and Zaengl.60
In principle it deals with an adequate simulation of the lead to the voltage
divider, i.e. to the simple metal wire or tube used for the connection of test
object and divider. It was impossible for a long time to detect the influence
of this lead, as no CROs have been available to measure the actual unit
step response of the systems. Thus neither the performance of the voltage

dividers used nor the performance of the whole measuring circuit could really
be checked. Many details within the construction of a voltage divider, however,
can completely destroy the fundamentally good layout based upon theoretical
investigations.
With this lead, a more detailed representation of our simple four-
terminal network, Fig. 3.24, is shown in Fig. 3.29. Three two-port sections
are combined, forming a ‘three-component system’. System 1 represents a
damping impedance Zd at the input end of the lead 2, connecting this
impedance with the voltage dividing system 3, which terminates the lead.
Due to their complex structure and frequency-dependent input impedance,
the voltage dividers cannot properly match the leads’ surge impedance ZL.
The damping impedance Zd is therefore placed at the input end of the lead,
as the travelling wave theory may easily show that only at this place is an
efficient damping of oscillations possible. The lead 2 is thus best treated as a
lossless transmission line, simulated by its surge impedance ZL and its travel
time 8L, which implies that the capacitance per unit length is constant. Leads
to the voltage dividers consist of metal tubes or metal wires, the diameter
of which should be such as to avoid any heavy predischarges. That waves
are really travelling with the velocity of light was readily shown in many
investigations50,59,60
. The simple representation of the lead by ZL and 8L
only was also confirmed by these investigations.
1 2 3
Io= 0
V0
[A]t
[A]L
[A]d
ViL ZL , tL
Zd
Ii
Vi
l
Figure 3.29 The ‘three-component system’ comprised of a (1) damping,
(2) transmission and (3) terminating system
Not represented in the circuit of Fig. 3.29 is the signal cable and the
recording instrument. It will be shown in section 3.6.6 that a lossless signal
cable (see item 6 in Fig. 3.23) can be connected to the different kinds of
dividers without appreciably influencing the USR. In this arrangement they
form a part of the divider’s l.v. arm. As also the recording instruments (item 7,
Fig. 3.23) have high input impedances and wide bandwidth, their possible
influence on the response is small and can thus be neglected, if the instruments
are properly connected to the l.v. arms, see section 3.6.6.

Up to now the terminating or voltage dividing system 3 had not been spec-
ified in detail, as its network depends upon the type of divider used. For
the computation of the transfer properties, the relevant matrix representation
according to eqn (3.54) is used. Thus the matrix [A] of the whole measuring
system is
[A] D [A]d
[A]L
[A]t
.
The matrix [A] can partly be solved by inserting the specific matrix elements
for [A]d
and [A]L
defined with the circuit elements of Fig. 3.29. The details
of the computation are lengthy and are omitted here. The following results,
however, are of general interest.
The normalized transfer function, hs D NVos/Vis, is best described by
introducing reflection coefficients K for travelling waves, which are reflected
either from the terminating system (Kt) or from the damping system Kd.
They are defined by
Kts D
At
11s
At
21s
ZL
At
11s
At
21s
C ZL
D
Zts ZL
Zts C ZL
3.62
Kds D
Ad
12s
Ad
11s
ZL
Ad
12s
Ad
11s
C ZL
D
Zds ZL
Zds C ZL
. 3.63
With these coefficients, the transfer function is:
hs D e8Ls Zd0 C ZL
Zds C ZL
1 C Kts
1 C Kt0
ð
1 Kt0Kd0
1 KtsKds exp28Ls
At
110
At
11s
. 3.64
The inherent time delay caused by the travel time of the lead, 8L, can well
be seen from the first factor; the last factor represents the normalized transfer
function of the voltage dividing system. The normalized step response could
be calculated using eqn (3.59). A glance at the transfer function indicates
the difficulties encountered with its transformation into the time domain. A
very simple example, however, will demonstrate the reflection phenomena
introduced by the lead.
Let the damping system be a pure resistor, i.e. Zds D Rd, and the
terminating system be simulated by a pure resistor divider without any
frequency-dependent impedances, i.e. the divider may merely be represented

by its input resistance Rt. Thus Kts D Rt ZL/Rt C ZL D Kt, and
KdS D Rd ZL/RD C ZL D Kd, and both are real numbers only.
According to eqn (3.58), the normalized transfer function of a pure resistor
divider will be equivalent to 1. Then, eqns (3.64) and (3.60) provide the
normalized USR:
gt D L1

exp8Ls
s
1 KtKd
1 KtKd exp28Ls

.
A well-known evaluation of this expression is based upon the expansion of
the last factor by a geometric row:
gt D L1

exp8Ls
s
1 KtKd[1 C KtKde28Ls
C . . .
. . . C KtKd2
e48Ls
C KtKd3
e68Ls
C . . .]

3.65
The infinite number of factors thus represents the possible number of
reflections. Within the time intervals 1 C 2n t/8L 3 C 2n, where
n D 0, 1, 2 . . ., the amplitudes of gt, however, are constant. In Fig. 3.30,
eqn (3.65) is evaluated for the most probable case, that when Rt × ZL, i.e.
Kt D C1, and Rd is smaller or larger than ZL. For Rd D ZL, Kd D 0 and any
reflection phenomena disappear. This is in fact the reason why the damping
resistor is placed at the input end of the lead. This very simplified example
shows also that the response time T or T0
will strongly be influenced by the
damping resistor. The magnitude of this influence is related to the length of
the lead by 8L.
Rd = ZL /2
Rd = 2.ZL
1
0.5
0 1 3 5 7 9
g
(
t
)
t
τL
Figure 3.30 Unit step response according to eqn (3.65)

An exact evaluation of the response time is based upon eqn (3.61),
substituted into eqn (3.64). The result is:
T0
D Tt C 8L




Zt0
Zd0
ZL
C ZL
Zt0 C Zd0



 C
Zd0
ZL
ð

Tkt
Zt0 C ZL
Zt0 C Zd0
Tkd
Zd0 C ZL
Zt0 C Zd0

3.66
In this equation,
Tt D
At0
110
At
110
3.67
is the response time of the voltage divider or terminator,
Tkt D
K0
t0
1 C Kt0
3.68
is the reflection response time of the terminator,
Tkd D
K0
d0
1 C Kd0
3.69
is the reflection response time of the damping system,
Zt0 D
At
110
At
210
3.70
is the d.c. input resistance of the terminator, and finally
Zd0 D
Ad
120
Ad
110
3.71
is the d.c. resistance of the damping system.
The influence of the dividers’ lead is again illustrated by eqn (3.66). The
complexity of this result is further discussed in section 3.6.5. In general, the
voltage dividing system, mainly represented by its response time Tt, will
essentially control the transfer characteristics of the whole system. Thus it
is justified to treat the terminating system in advance and isolated from the
lead to achieve a general understanding of all h.v. dividing systems and their
adequate application.

3.6.4 Voltage dividers
Voltage dividers for d.c., a.c. or impulse voltages consist of resistors or
capacitors or convenient combinations of these elements. Inductors are in
general not used for voltage dividers for testing purposes, although ‘inductance
voltage dividers’ do exist and are used for the measurement of power frequency
voltages,139
independent from inductive voltage transformers as used in
power transmission. Inductance voltage transformers consist in the simplest
case of a high-quality toroidal core with a tapped winding and some of these
elements can be cascaded to form a ‘cascade inductance divider’. Measuring
uncertainties down to a few ppm can be reached if built for quite low voltages
(1 kV or less), but lots of problems arise if they are built for magnitudes of
100 kV or more. Therefore, no further treatment follows here.
The elements of the aforementioned h.v. voltage dividers are usually
installed within insulating vessels of cylindrical shape with the ground and h.v.
terminals at both ends. The height of a voltage divider depends finally upon the
external flashover voltage and this follows from the rated maximum voltage
applied; this flashover voltage is also influenced by the potential distribution
and is thus influenced by the design of the h.v. electrode, i.e. the top electrode.
For voltages in the megavolt region, the height of the dividers becomes large,
as one may assume the following relative clearances between top electrode
and ground:
2.5 to 3 m/MV for d.c. voltages;
2 to 2.5 m/MV for lightning impulse voltages;
up to or more than 5 m/MV (r.m.s.) for a.c. voltages;
up to and more than 4 m/MV for switching impulse voltages.
Because the breakdown voltages in atmospheric air become strongly non-
linear with voltage magnitude for a.c. and lightning impulse voltages, the
above suggested clearances may be considered only as guidelines.
The most difficult problems in a simulation of the actual network of voltage
dividers is in the inadequate representation of the stray capacitances (see
Figs 3.7, 3.8, 3.21). Whereas the location and dimensions of the active parts,
i.e. resistor or capacitor units, within a voltage divider are exactly known, the
same statements are impossible to achieve for stray capacitances. It would
also be too difficult to present equivalent circuits with distributed parameters,
which individually account for the physical size of the units, by assuming a
too high number of elements of unequal values. Apart from the fundamental
difficulties in performing analytical computations of such circuits, the results
are then individually related to the high number of parameters.
It has been acknowledged by many investigators that a recurrent or
distributed parameter network with equally distributed parameters is probably

V
Zq
′
Zl′
Zl′
Zl′
Zl′ V2
Zq
′
Zq
′
Zq
′
1 element
n
elements
Figure 3.31 Distributed parameter equivalent circuit of voltage dividers
the best compromise to simulate transfer characteristics by equivalent circuits.
Such a distributed parameter network for a generalized voltage divider is
shown in Fig. 3.31. Our ‘terminating system’ [A]t
of Fig. 3.29 is now
simulated by a large number (n) of elements or sections, and the n impedances
Z0
l in series are providing the voltage reduction. An equal number of
impedances Z0
q to earth are distributed along this column. The input voltage
V is thus greatly reduced to the low output voltage V2. The total impedances
are then defined by
Zl D

Z0
l D nZ0
l; and Zq D
1
Z0
q
1
D
Z0
q
n
3.72
The number n is by this definition equivalent to the voltage ratio or scale factor
V/V2 of the divider; it may differ from N as defined before, as the impedance
Zd of the lead (Fig. 3.29) may change the ratio of the whole voltage measuring
system.
The matrix representation of such a network, which is equivalent to a trans-
mission line network, is well known. Applying eqn (3.56) and eqn (3.58) to
this network, one may easily find the normalized transfer function (index

t D terminator), which is
hts D
nV2
V
D
n sinh
1
n

Zls/Zqs
sinh

Zls/Zqs
. 3.73
The normalized unit step response is
gtt D L1

1
s
hts

.
Both quantities can now be computed and analysed for different equivalent
circuits, for which the impedances Z0
l and Z0
q are specified. Z0
q, however, will
always be represented by stray capacitances C0
e to earth, as no voltage dividing
system is known which would comprise any other passive elements at these
locations. This stray capacitance is thus assumed to be equally distributed.
Resistor voltage dividers
The most general representation of such dividers has to assume inductive
components L0
of the actual resistor R0
as well as capacitive elements C0
p in
parallel to the resistors (see Fig. 3.32). Inductances are inherent with every
flow of current due to the magnetic field, and the parallel capacitors C0
p may be
formed by the construction and arrangement of the resistors. The neglecting of
any inductance in series to these stray capacitances indicates possible coupling
effects and the simulation of electrical fields within insulation media of low
permittivity only; the individual values are thus of any small magnitude by
the distributed parameter representation.
V
V2
R2
L2
R′ L′ R′ L′
C ′
p
C ′
e
C ′
p
C ′
p
C ′
e C ′
e
Figure 3.32 Equivalent circuit for resistor voltage dividers. R D nR0
;
L D nL0
; Ce D nC0
e; Cp D C0
p/n; R2 D R0
; L2 D L0
; R1 D n 1R0

The normalized transfer function is easily found from eqn (3.73) and is
hts D n
sinh
1
n

R C sLsCe
1 C R C sLsCp
sinh

R C sLsCe
1 C R C sLsCp
. 3.74
The computation of gtt for this and all the other circuits presented thereafter
can be made with minor approximations justifiable for n × 1. The details can
be found in the literature,61,62
only the result is presented:
gtt D 1 C 2eat
1

kD1
1k
coshbkt C
a
bk
sinhbkt
1 C
Cp
Ce
k2
)2
; 3.75
where
a D R/2L;
bk D

a2
k2
)2
LCe[1 C Cp/Cek2
)2
]
;
k D 1, 2, 3, . . . , 1.
Both quantities can be used to demonstrate the limits of applications if repre-
sentative values for the circuit constants are taken into consideration.
First, it is clear that resistor dividers are ideal for d.c. voltage measurements.
The transfer function hts for high R values and accordingly small values of
L/R increase steadily with a decrease of the frequency. For s ! 0, hts O
D1
and therefore
V2 D
V
n
D V
R2
R1 C R2
(see Fig. 3.32 for the definition of R1 and R2). The advantage of this rela-
tionship and its effect upon the accuracy and stability of the divider ratio was
already discussed in section 3.3.
The ability to measure a.c. voltages as well as ripple inherent in d.c. voltages
depends upon the decrease of hts with frequency. Since for all constructions
of high ohmic resistor dividers the L/R values are lower than about 0.1 µsec,
and also Cp − Ce, the controlling factor of the transfer function is given by
the product RCe. We can thus neglect L and Cp in eqn (3.74) as well as in

eqn (3.75) and therefore:
hts ³ n
sinh
1
n

sRCe
sinh

sRCe
3.76
gtt D 1 C 2
1

kD1
1k
exp
k2
)2
RCe
t 3.77
where again
k D 1, 2, 3, . . . , 1.
Equation (3.76) can be used to calculate the bandwidth fB from the amplitude
frequency response jgtsj, if jgtsj D 1/
p
2. The evaluation shows the simple
relationship
fB D
1.46
RCe
. 3.78
Similarly, the response time T0
can be computed applying eqn (3.51) to
eqn (3.77). The result gives
T0
D
RCe
6
³ T. 3.79
V
R/2
R/2
V2
R2
R
CE
Figure 3.33 Common equivalent circuit representing approximately the
distributed parameter circuit, Fig. 3.32, with L D Cp D 0. CE D
2
3

Ce for
equal response times (eqn (3.80)). CE D 0.44Ce for equal bandwidth
(eqn (3.81))

Although the USR starts continuously, since for t D 0; dgt/dt D 0, a very
pronounced time delay 8de cannot be defined. Thus T0
³ T. fB and T0
could
be used to define much simpler equivalent circuits for the distributed parameter
network. Figure 3.33 shows this very common equivalent circuit. For R2 − R1
the USR is obviously
gtt D 1 expt/8;
where 8 D RCE/4. Since for this truly exponential rise the response time
equals to 8, the not distributed capacitance to ground CE in this equivalent
circuit is
T0
D
RCe
6
D
RCE
4
; ! CE D
2
3
Ce, 3.80
if equal response times are used for comparison. Comparing, however, the
bandwidth of both systems, which is equivalent to fB D 1/2)8 for the simpli-
fied circuit, we obtain
4
2)RCE
D
1.46
RCe
; ! CE D 0.44Ce. 3.81
The reasons for these discrepancies can easily be detected if the real unit
step response according to eqn (3.77) is compared with a true exponential rise
provided by the simplified equivalent circuit (Fig. 3.33). This comparison is
shown in Fig. 3.34 for equal response times. The delayed, but faster, increase
of gtt for the distributed circuit is the main reason for the discrepancies.
1.0
0.8
0.6
0.4
0.2
0 1
1
2
2
3 4 5 6 7
g
t
(t)
t
RCe /p2
g(t) = 1 − exp
gt
(t) = 1 + 2 Σ (−1)k exp
t
RCe /6
∞
k = 1
k 2
p2
RCe
t
Figure 3.34 Comparison of the unit step responses with equal response
time. (1) For equivalent circuit Fig. 3.32 with L D Cp D 0 (eqn (3.77)).
(2) For equivalent circuit Fig. 3.33 with CE D
2
3

Ce

In section 3.5.4 it was shown that the stray capacitances C0
e may approxi-
mately be calculated from the dimensions of any structure. In section 3.3 we
have also given some guidance for the dimensioning of the resistor values for
d.c. or a.c. dividers. Combining both these rules, we may summarize that
Ce
[pF]
³ 10 15
H
[m]
;
R
[G ]
³ 1 2
V
[MV]
;
where H equals the height of a divider, and V is the rated high voltage. We
may introduce these magnitudes into eqn (3.78) and find the following simple
relationship:
fB ³
50 . . . 150
HV
with

fB in Hz
H in m.
V in MV
3.82
Assuming a d.c. voltage divider for V D 1 MV, which will be about 3 m in
height, eqn (3.81) shows a bandwidth of not more than 50 Hz. It is, therefore,
impossible to measure d.c. ripple voltages with high-value resistor dividers for
voltages higher than some 100 kV. Equation (3.82) also shows the limitations
for the application of such dividers without preventive measures: an accurate
measurement of power frequency voltages needs fB values
¾
D1 kHz, resulting
in a product HV of about 100 kV m. This product limits the application of the
above to voltages not exceeding 100–200 kV.
The measurement of lightning or even switching impulse voltages demands
a much higher bandwidth as already discussed in section 3.6.2. The decrease
of Ce by very carefully adapted ‘shielding’ or potential grading methods is
limited, although a reduction by a factor of about 5–10 seems possible. But this
is not enough. There is only one practical solution, i.e. to reduce the value of
R by some orders of magnitude. Let us assume that we have to build a resistor
divider with T ³ T0
D 50 nsec, still introducing an amplitude error υ of 10
per cent for linearly rising voltages chopped at Tc D 0.5 µsec (see eqn (3.53)).
Thus the product RCe becomes 300 nsec according to eqn (3.79). Let the
resistance be about 2 m in height, providing a lightning impulse withstand
strength of about 1000 kV. Without excessively large top electrodes for forced
shielding, C0
e is about 10 pF/m and thus R ³ 300 ð 109
/20 ð 1012
D 15 k .
This is indeed the order of magnitude which can be used for voltage dividers
applicable for the measurement of lightning impulse voltages. This low value
of a resistance will load the impulse generators, but this resistive load is
tolerable if the discharge resistors within the generator are adapted. A large
increase of the rated voltage is, however, not possible. The reduction of Ce
by huge shielding electrodes becomes impractical as the dimensions must
increase with the divider’s height. Thus the response time with the resistance
value unchanged increases proportional to Ce or the product HC0
e. Response

times larger than 200 µsec for the measurement of full standard 1.2/50 light-
ning impulses, also chopped on the crest or the tail, have, however, not been
accepted by the former standards6
and the newest IEC Standard53
sets even
more stringent requirements, which shall not be discussed here. A further
problem is created by the heat dissipation within the resistors. For constant R
values and increasing voltage, the energy dissipated in the resistive materials
increases proportionally with V2
, and during the short time of voltage appli-
cation not much heat can be transferred to the surrounding insulation material,
the energy must be stored within the resistor. A calculation of the temperature
increase within the wire-wound metal resistors would indicate the difficulties
of achieving low-inductive resistor units applicable to this h.v. stress. These
are the main reasons why resistor voltage dividers for voltages higher than
1.5–2 MV and resistance values of 10–20 k cannot be built.
There are, however, some possibilities to improve the unit step response of
such dividers, which will only be treated briefly.
Reduction of resistance value. If only front-chopped impulse voltages of
short duration ( 1 µsec) have to be measured, a further reduction of R is
possible if the impulse generator has high stored energy and the waveshaping
front resistors (R1 in Fig. 2.26) are of low value. The heat dissipation problem
is then solved only by the chopping. It is essential, however, to reduce the
inductive time constant L/R of the resistors as far as possible. For assess-
ment, we have to refer to the equivalent circuit, shown in Fig. 3.32, and
the relevant transfer properties. The numerical evaluation of eqn (3.75), an
example of which is given in Fig. 3.35, shows the appearance of oscillations
in the USR with too low resistance values, although L/R was kept constant
as well as Ce and Cp. The reasons for this instability can easily be explained
using eqn (3.75). Although the damping factor expat) of the infinite series
remains constant, the hyperbolic functions will change to trigonometric ones,
depending upon the series number k. The most efficient term within the series
is the first one (k D 1). For this term, the transition takes place if bk becomes
complex. Hence,
Rcrit ³ R 2)

L
Ce
1
1 C )2
Cp/Ce
3.83
This ‘critical’ resistance Rcrit is included in the table of Fig. 3.35, and the
comparison with the computed responses confirms the validity of the above
equation.
Typical examples for such low-resistor voltage dividers are shown by Rohlfs
et al.63
or Pellinen et al.64
Reduction of Ce. The possibility of reducing the stray capacitance to earth
by metal electrodes connected to h.v. potential was theoretically treated in
section 3.3. The practical application of field-controlling top electrodes was

1.5
1
4 3
5
2
1
0.5
g
t
(
t
)
0 50 100
t (ns)
150 200
Figure 3.35 Calculated unit step response for resistor dividers. Equivalent
circuit according to Fig. 3.32
L/R D 10 nsec; Ce D 40 pF; Cp D 1 pF; Rcrit
(1) R D 30 k 15.5 k
(2) R D 10 k 8.9 k
(3) R D 3 k 4.85 k
(4) R D 1 k 2.8 k
L/R D 10 nsec; Ce D 12 pF; Cp D 1 pF;
(5) R D 10 k 13.4 k
introduced by Bellaschi,65
it is a widely used and effective method. The
combination of a field-controlling h.v. electrode with a non-linear distribu-
tion of the resistance values per unit length was also explained earlier.34
The inherent disadvantages of all field-controlling methods are twofold. First,
the unit step response becomes very sensitive to surrounding objects, as a
strong relative change of Ce is likely to be produced by small changes of
the external potential distribution. The second disadvantage is related to the
interaction between the lead and the divider. Large shielding electrodes intro-
duce a relatively large external parallel capacitance across the divider, which
is not equal to Cp in our equivalent circuit. This capacitance loads the lead
and enhances travelling wave oscillations, which can only be damped by the
impedance Zd of the lead. Additional explanations are given in section 3.6.5.
Ce can also be reduced by a decrease of the dimensions of the resistor.
Harada et al.66
proposed a 1-MV divider with R D 9.3 k , the resistor of
which was only 46 cm in axial length, but placed in a much longer insulating

vessel. In this design difficulties arise with the heat dissipation within this
small resistor and with the field gradient control in the neighbourhood of the
resistor. For further details the reader should refer to the original paper.
Compensation methods. Our equivalent circuits assume an equal distribution
of the voltage dividing elements in the resistor column. Also the l.v. arm is
assumed to be equal to a resistor unit of the h.v. arm. This is, of course, not
true, as the connection of the signal cable with the l.v. arm needs a special
construction (see section 3.6.7, Fig. 3.62). For resistor dividers, the voltage
USR is about equal to the step response of the current through the l.v. arm. In
this way the current also increases in a manner that is similar in shape as is
given by the voltage unit step response. As long as R2 (Fig. 3.32) is not larger
than the surge impedance of the signal cable, one may simply increase the
inductance L2 to increase the resistance of the output voltage. The low value of
the surge impedance, which is in parallel with R2, limits the efficiency of this
method. In practice, the actual value of L2 is predominantly determined by the
construction of the l.v. arm. The actual USR may, therefore, be quite different
from the computed one. Other compensating networks at the input end of the
signal cable have been proposed67
which can be evaluated using the well-
known methods of network synthesis.1
The efficiency of such networks is,
however, quite limited.
Parallel-mixed resistor-capacitor dividers
If in the equivalent circuit for resistor dividers of Fig. 3.32 the stray capac-
itances C0
p are increased, i.e. if real capacitor units are placed in parallel to
the resistor R0
, a ‘parallel-mixed resistor–capacitor divider’ is formed. This
parallel arrangement of resistors and capacitors is a well-known technique
used for attenuators within measuring instruments, i.e. CROs, and is often
referred to as a compensated resistor voltage divider. The idea to use this
circuit for h.v. dividers was introduced by Elsner in 1939,68
with the goal
of reducing the effect of the stray capacitances to earth, C0
e. The efficiency of
the C0
p capacitors can actually be seen by comparing unit step responses of
Fig. 3.34, curve 1, with those in Fig. 3.35. Neglecting any Cp values within
the simplified R-C latter network causes the USR to start continuously with
time. Even the small Cp value of 1 pF in Fig. 3.35 excites a small step in
the USR, and the value of this step gtC0 is obviously dependent upon the
capacitance ratio Cp/Ce (compare the curves 1 to 4 with 5). The increase in
the ratio of Cp/Ce increases this step and thus the question arises whether it
is possible to increase this first step to the final value.
This can be accomplished theoretically only if we assume that the repre-
sentation of actual capacitor units placed in parallel to the resistors in the
equivalent circuit of Fig. 3.32 is correct. It is, however, not correct if this
circuit is used to compute high-frequency phenomena or unit step responses in
the nanosecond or even microsecond range. The reason for this is simple. The

inherent inductance L of every capacitor C causes a series resonance frequency
fr D 1/2)
p
LC, which is quite low for capacitance values capable to compen-
sate h.v. dividers (for instance: (fr D 10 MHz for L D 1 µH; C D 200 pF).
The actual USR of parallel-mixed resistor–capacitor dividers is therefore
similar to pure capacitor voltage dividers, which will be treated later. Apart
from the fact that this type of divider is still in use for the measurement of
impulse voltages, with R values in the 10–100 k range and Cp values in the
order of some 100 pF, we shall simulate the transfer properties by a simplified
equivalent circuit only, which will not cover the high-frequency range. This
equivalent circuit is shown in Fig. 3.36.
R′
V2
R2
C2
Cp
′
Ce
′
V
Figure 3.36 Simplified equivalent circuit for parallel-mixed
resistor–capacitor dividers. R D nR0
; Cp D C0
p/n; Ce D nC0
e; R2 D R0
;
C2 D C0
p
The computation of the normalized transfer function and unit step response
yields for equal elements in the h.v. and l.v. arms, i.e. R2C2 D R0
C0
p:
hts D n
sinh
1
n

sRCe
1 C sRCp
sinh

sRCe
1 C sRCp
3.84
gtt D 1 C 2
1

kD1
1k expakt
1 C k2
)2
Cp/Ce
3.85
where
ak D
k2
)2
RCe1 C k2
)2
Cp/Ce
;
k D 1, 2, 3, . . .
The peculiar effect of this circuit is detected by the calculation of the limiting
values for very high and very low frequencies, or very short and very

long times:
lim
s!0
[hts] D 1; lim
t!1
[gtt] D 1.
But
lim
s!1
[hts] ¾
D 1
Ce
6Cp
; lim
t!0
[gtt] ¾
D 1
Ce
6Cp
.
A sketch of the normalized amplitude frequency response and USR in Fig. 3.37
demonstrates the response of this dividing system to different voltage ratios.
The difference of these ratios is formed by the relation Ce/6Cp, and very
high values of Cp would be necessary to reduce this difference to very small
values. It is obvious that these differences in scale factors can be reduced by
a reduction of C2 within the l.v. arm to increase the voltage drop across C2
for high frequencies. A calculation, published by Harada et al.,69
shows the
condition
R2C2 D R1Cp1












Ce
Cp1
1
sinh

Ce
Cp1











³ R1Cp1

1
Ce
6Cp

3.86
ht (f ) gt (t)
1
0
(a) (b)
1
0
Frequency f Time t
Ce
6 Cp
Ce
6 Cp
Figure 3.37 Schematic diagrams for the normalized amplitude frequency
response: (a) and unit step response, (b) for voltage dividers according to
Fig. 3.36
where
Cp1 D
C0
p
n 1
³
C0
p
n
D Cp;
R1 D n 1R0
³ nR0
D R.

In summary then, it is not recommended to compensate resistor dividers for
high impulse voltages with parallel capacitor units, as the equivalent circuit of
Fig. 3.36 is inadequate to treat short-time phenomena. A compensation of high
ohmic dividers commonly used for the measurement of d.c. or a.c. voltages,
however, is very attractive to increase the performance in the intermediate
frequency range (100 Hz up to some 100 kHz, depending upon the size of the
divider).
Capacitor voltage dividers
It was shown in section 3.5.4 that pure capacitor voltage dividers could be
made either by using single h.v. capacitance units, i.e. a compressed gas
capacitor, in series with a l.v. capacitor, or by applying many stacked and
series connected capacitor units to form an h.v. capacitor. The absence of
any stray capacitance to earth with compressed gas capacitors provides a very
well-defined h.v. capacitance, small in value and small in dimensions, and
by this even a pure capacitor voltage divider with quite good high-frequency
performance can be built if the l.v. arm or capacitor is constructively integrated
in the layout of such a capacitor. This means that this capacitor must be very
close to the h.v. capacitance, and this can be provided for instance by inserting
a symmetrical arrangement of l.v. capacitors between the l.v. sensory electrode
2 and the guard ring 20
or supporting tube 3 (see Fig. 3.18). Although such a
construction was proposed by Schwab and Pagel,70
similar systems may well
be formed by other coaxial arrangements.71
The applicability to very high
voltages, however, is mainly restricted by the high cost of such constructions
and the difficulties involved with the replacement and exchange of l.v. arms
to change the scale factors.
A treatment of capacitor voltage dividers with stacked capacitor units
is thus justified. The distributed parameter network is able to simulate the
transfer properties. Figure 3.38 shows such a network, which may encounter
V2
V
R2 L2 C2
C ′
p
C ′
p
R ′ C ′ C ′
R′
L′ L′
C ′
p
C ′
e
C ′
e
C ′
e
Figure 3.38 Equivalent circuit for capacitor voltage dividers. R D nR0
;
L D nL0
; Ce D nC0
e; C D C0
/n; Cp D C0
p/n; R2 D R0
; L2 D L0
; C2 D C0

all possible passive circuit elements. The actual stacked capacitors are now
simulated by the capacitance units C0
, and L takes into account the inherent
inductance. The series resistance R0
may be used to simulate either only small
losses within the capacitor units C0
, or even real resistors in series with these
units. The small values of stray capacitances in parallel to the stacked columns
C0
p and to ground C0
e complete the equivalent circuit.
A glance at the unit step response, which is represented as
gtt D 1
Ce
6C C Cp
C 2 expat
1

kD1
1k
coshbkt C
a
bk
sinhbkt
AB
,
where
A D

1 C
Cp
C
C
Ce
Ck2
)2

, a D
R
2L
,
B D 1 C
Cpk2
)2
Ce
, bk D

k2
)2
Ð A
LCeB
, 3.87
shows a close similarity to the USR of resistor dividers, eqn (3.75). Both
equations are actually the same, if the value C in eqn (3.87) approaches infinite
values. With finite values of C, representing capacitor voltage dividers, the
main difference is at first related to the negative term Ce/6C C Cp ¾
D Ce/6C,
which is independent of the time and thus also the frequency. This term was
also found in the treatment of the ‘equivalent capacitance’, see eqn (3.31). It
appears again as a result of our procedure of the normalization of the USR.
All explanations referring to the proper dimensioning of stacked capacitors,
therefore, also apply to this result, which demonstrates the possible variations
of the ratio n with Ce.
The time dependency of the USR for ‘pure’ capacitor dividers, i.e. with
R D 0 in the equivalent circuit, is obviously very complex. In eqn (3.87),
with R D 0, the damping term expat will be equal to 1, and all hyper-
bolic functions are converted to trigonometric ones. The numerical evaluation
of this equation for this case is impossible due to the infinite number of
sinusoidal terms. It is also not realistic to assume no resistance at all, as
at least the (frequency-dependent) dissipation factor of the dielectric will
cause some damping. For a simple series equivalent of a lossy capacitor, this
dissipation factor is tan υ D ωR0
C0
D ωRC. The relaxation phenomena within
the dielectric materials, however, control in reality this dissipation factor for
high frequencies. It has been confirmed by measurements72
that an adequate
low-resistance value can be assumed to evaluate eqn (3.87) as was done in
Fig. 3.39. Here, the oscillations can be related to the travel time 8 D
p
LCe,
as a step voltage applied to the input of such a ladder network can travel along

4
3
2
1
0
10 50 100
t(ns)
−1
C = 150pF; L = 2.5mH; R = 20Ω
Ce = 40 pF; Cp = 1pF
g
t
(
t
)
Figure 3.39 Calculated unit step response for a capacitor voltage divider;
the equivalent circuit is Fig. 3.38. R D 20 ; L D 2.5 µH; C D 150 pF;
Ce D 40 pF; Cp D 1 pF
the column. If the voltage amplitude is not reduced to a zero value when the
wave reaches the earthed l.v. part, it will be reflected and excites oscillations.
Pure capacitor voltage dividers are therefore sensitive to input voltages with
short rise times and the output voltage may oscillate with non-oscillating input
voltages. In addition, such a capacitance divider within the whole measuring
circuit, i.e. with leads connected to its input, will form a series resonant circuit.
Thus it is obvious that pure capacitor dividers are not adequate to measure
impulse voltages with a steep front (front-chopped lightning impulse voltages)
or any highly transient phenomena (voltage during chopping). Crest values of
switching impulse or even full lightning impulse voltages, however, can be
properly recorded, if the transient phenomena during the front of the impulses
have disappeared.
The similarity of the step response equations for resistor voltage dividers
to those treated in this part stimulated Zaengl to propose and to realize the
possible improvement of pure capacitor dividers by inserting real resistor units
in series with the capacitors.72
If the value of these resistors is not too
high, but just sufficient to damp the oscillations, it is likely to achieve an
excellent transient performance. A very similar equation to that of eqn (3.83)
could be derived by calculating the transition from hyperbolic to trigonometric
functions for the argument bk and k D 1 in eqn (3.87), providing again a
critical resistance. Such a critical value can also be identified in Fig. 3.40,
in which some calculated step responses according to eqn (3.87) are shown.
Adequate values for a capacitor voltage divider for a voltage of about 1 MV
(height ³ 3 m) are used for this simulation.

1.5
1
4
3
2
1
0.5
0 20
t (ns)
g
t
(
t
)
40 60 80 100
Figure 3.40 Computed unit step response Gtt for damped capacitor
dividers according to equivalent circuit, Fig. 3.38
C D 150 pF; L D 2.5 µH; Ce D 40 pF ; Cp D 1 pF
(1) R D 250
(2) R D 750 4
#
L
Ce
D 1000
(3) R D 1000
(4) R D 2000
The influence of the magnitude of the resistance R is obvious. A very well-
damped response is reached by a resistance value of about
R ³ 4

L
Ce
3.88
although the larger overshoot observed with lower values can still be accepted.
The short response time is in accordance with the theory. T0
or T can be
calculated by the transfer function as well as from eqn (3.87). It is equivalent
to eqn (3.79), and thus T0
D RCe/6. The small resistor values as necessary to
damp the oscillations are responsible for these low response times, and thus
a ‘series-damped capacitor divider’ is formed. The input impedance of these
dividers increases with decreasing frequencies, and hence the loading effect of
the voltage generating system is limited. Their application for a.c., switching
or lightning impulse voltages without any restrictions is, therefore, possible.

If a parallel branch of high ohmic resistors is added, d.c. voltages can also be
measured as shown before and an ‘universal voltage divider’ is formed.
These ‘series-damped capacitor dividers’ are not limited in the voltage
range, as a stacking of capacitor units is not limited as well as the insertion
of distributed resistors. These resistors are not heavily loaded, as only tran-
sient input voltages cause displacement currents. A ‘general-purpose’ voltage
divider is therefore available, and have been in general use since about 1970
up to the highest voltage levels.54,122
Figure 3.41(a) shows such a voltage divider for a lightning impulse voltage
of 6 MV. The electrodes are not provided to shield the divider, i.e. to reduce
Ce, but only to prevent discharges and thus to increase the flashover voltage
for switching impulses.
3.6.5 Interaction between voltage divider and its lead
The analytical treatment of our measuring system presented so far is not yet
complete. Whereas the USR of the voltage dividers could readily be calculated,
similar results are missing for the entire circuit. Now it can be shown that the
generalized expression for the response time T and its interaction with the
circuit elements, eqn (3.66), can effectively be applied in practice.
As already mentioned in section 3.6.3, it is too difficult to apply an analyt-
ical solution to the USR of the whole measuring system, which was repre-
sented by the ‘three-component system’ of Fig. 3.29. Numerical solutions by
advanced programming, however, are possible, and many computer programs
are available. The results presented here are calculated with the ‘transient
network program’ published by Dommel.73
Within this program, the loss-
less transmission line (see 2, Fig. 3.29) is simulated by the exact solution of
the partial differential equations of a line and thus does not introduce any
errors. The simulation of the terminating system, i.e. the voltage dividers,
needs, however, a subdivision of the distributed parameter networks into a
finite number of sections. If the number of elements n (for n see Fig. 3.31) is
larger than about 5, the results are close to the infinite number solution.
Numerical computations need numerical values for the surge impedance of
the lead ZL to the divider. For the common set-up of a voltage testing system
(Fig. 3.22), this lead is more or less horizontal above the ground return, which
is assumed to be an extended plane. Many experiments50
demonstrated that
the travel time 8L is controlled by the velocity of light c0. As ZL D
p
LL/CL
and 8L D
p
LLCL D l/c0, with LL being the total inductance and CL the total
capacitance of this lead, ZL D l/c0CL, with l being the length of the lead. The
capacitance of the lead can be computed assuming that a cylindrical lead of
diameter d is at height H above a plane, which is earthed. The well-known

Figure 3.41 Series-damped capacitor voltage divider for 6-MV impulse
voltage (courtesy EdF, Les Renardieres, France)

capacitance formula
CL D
2)ε0l
A
;
where
A D ln



2l
d
$
%
%
%

f1 C 2H/l2g 1

f1 C 2H/l2g C 1



D ln

4H
d

ln
1
2
1 C

1 C 2H/l2
may well be used, although this lead is placed between the test object and
the voltage divider. As c0 D ε0?00.5
, where ε0 D permittivity and ?0 D
permeability of free space, the surge impedance becomes
ZLhor D A
1
2)
#
?0
ε0
D 60 ð Al, d, H [ ] 3.89
for this horizontal lead. Sometimes, the horizontal lead is lengthened by a
vertical lead to measure the experimental USR of the system. Thus we need ZL
for a vertical lead also. According to Fig. 3.22 and eqn (3.33), this capacitance
is known. With the same assumptions as made above, we obtain
ZLvert D
1
2)
#
?0
ε0
ln

2l
d

4s C l
4s C 3l


³ 60 ln

1.15l
d

[ ] for s − l. 3.90
The differences in the surge impedances are not large if the usual dimensions
are taken into account.
In Fig. 3.42(a), a very simplified equivalent circuit represents a 20-k
resistor divider with a lead length of 3 m (8L D 10 ns). The divider is ideal-
ized by the omission of any stray capacitances or inductances, but a parallel
capacitance of Ct D 50 pF across the whole divider represents a top electrode
which may shield the divider. A pure resistor Rd provides ideal damping
conditions for travelling waves. Figure 3.42(b) shows some computed results
of the USR. For Rd D 0, no noticeable damping effect is observed within the
exposed time scale. Although the oscillations are non-sinusoidal, the funda-
mental frequency can clearly be seen. This frequency is obviously close to the
resonance frequency fr, generated by the lead inductance LL and the divider’s
capacitance Ct. As LL D ZL8L, this inductance is 3 µH, giving fr D 13 MHz.

Rd
Rd variable
ZL = 300 Ω
τL = 10 ns
Ct = 50 pF
Rt = 20 kΩ
R2 C2 = Rt Ct
R2 Rt
ZL ; τL
Ct
C2
Rt
R2 gt (t )
2
1.5
1
0.5
450 Ω
300 Ω
150 Ω
Rd = 0
g
t
(
t
)
0 10 100
(a)
(b)
200
t (ns)
300

Figure 3.42 Computed unit step response for idealized resistor or
parallel-mixed resistor–capacitor divider with lead. (a) Equivalent circuit.
(b) Computed USR
Thus, this example also implies the typical USR for pure capacitor or parallel-
mixed resistor–capacitor voltage dividers, as Ct can well be assumed to repre-
sent these types of dividers. Higher values of Ct will decrease the frequency
of the oscillations. Acceptable responses are only provided by a damped lead.
To prevent any overshoot, Rd must be close to ZL. The exponential increase
in the front and increase of the response time T0
or T D T0
8L is obviously
produced by the time constant RdCt, which equals 15 ns for Rd D ZL and the
specific values assumed. Large capacitor dividers with stacked capacitor units
comprise in general much higher capacitance values, and in such cases the
large response time of such measuring systems is produced by the necessary
damping of the lead.
We may easily compute T0
or the actual response time T D T0
8L from
eqns (3.66) to (3.71). It is clear that for this ideal divider Tt D 0 and Tkd D 0
(no frequency dependency of Zds D Rd). With the only frequency-dependent

term for the input impedance of the divider Zts D Rt/l C sRtCt, we may
easily find that Tkt D RtCt/1 C ZL/Rt). The final result may be best repre-
sented in the form
T D T0
8L D
1
1 C Rd/Rt

RdCt 8L

1
ZL
Rt

1
Rd
ZL

. 3.91
Some remarkable findings can be observed.
For Rd D ZL, the length of the lead has no influence upon the response
time. This case corresponds to the ‘infinite line response’, as the same result
would be achieved if a step voltage supplied from an extremely long lead
would be applied to the dividing system.
With no damping resistance, or Rd ZL, the response time taken from the
actual beginning of the USR will always decrease proportionally with the lead
length l D 8Lc0. This decrease of T is clearly produced by an overshoot of
the USR. As is seen from the computed USR, the determining factor is RdCt
providing a positive contribution to T. For capacitor dividers, Rt ! 1 and
the same equation can be applied.
A second example (Fig. 3.43) simulates a pure resistor divider of low resis-
tance value (2.32 k ), which was in reality built from carbon composition
resistors to achieve extremely low values of inductances. The stray induc-
tances are therefore neglected in the equivalent circuit (Fig. 3.43(a)), but it
comprises distributed stray capacitances to earth, which have been calculated
with eqn (3.33). The small input capacitance (5 pF) was estimated as only
a very small top electrode was provided. The voltage divider was used for
steep-front voltage measurements up to 800 kV. The lead length of 6 m was
used for USR measurements only, and the equivalent circuit simulates this
lead length. The computed USR (Fig. 3.43(b)) shows again larger oscillations
with no damping resistance in the lead. The traces of the oscillations deviate
strongly from the USR of a pure resistor network (see Fig. 3.30), due to the
stray capacitances involved. Only the ‘infinite line response’ is smooth. For
Rd D 100 , the computed input voltage of the divider is also plotted to show
the distortion introduced by the divider. The small capacitive reflection is
mostly suppressed by the divider.
For this equivalent circuit, again the general dependency of the response
time from the circuit parameters can be computed by eqn (3.66). The result is
T D T0
8L D
1
1 C Rd/Rt
ð

RtCe
6
C Rd

Cp C
Ce
2

8L

1
ZL
Rt

1
Rd
ZL

. 3.92
Some interesting findings are observed.

Rd
Rd = 0
(gt) input
(gt) input
ZL ; τL
R′
t = 232 Ω; R′
t = 2.32 kΩ ; R′
t; ZL = 272 Ω
τL = 20 ns
Rt = ∑ Re
C ′
e = 5 pF; C ′
e = 25 pF;
Ce = ∑
Cp
R2 gt (t)
(5 pF)
5 elements
100 Ω
1.5
0.5
0 20 100 300
t (ns)
g
t
(
t
)
500
1
272 Ω
500 Ω
R′t
C′
e
C ′
e
R′t
R′t R′t −R2
(a)
(b)
Figure 3.43 Computed USR for low-value resistor voltage divider.
(a) Equivalent circuit. (b) Computed USR (for divider input)
The influence of the lead on the divider (8L) is the same as before. Now
the the divider’s response time (RtCe/6 D Tt appears, as expected. Only a
part of the stray capacitance Ce, but the full value of the input capacitance
Cp, provides positive response times if the lead is damped.
Figure 3.44 shows oscillograms of measured responses. The lead was placed
parallel to the ground and the unit step voltage generator was mounted at the
wall of the laboratory, which was shielded by a Faraday cage. There is a very
good agreement between the computed and measured values, the USR and the
response time.
Finally, the third example (Fig. 3.45) explains the existence of a real time
delay between the output and input voltage of a resistor voltage divider.
This example is similar to the first one, but the resistors are distributed and
comprise a small inductive time constant of Lt/Rt D 5 ns. Stray capacitances
in parallel to each section, however, are neglected. The USR of the output

Rd = 0 Rd = 100 Ω
Rd = 500 Ω
50 ns/div.
Rd = 297 Ω 50 ns/div.
50 ns/div.
50 ns/div.
Figure 3.44 Measured unit step response for the resistor voltage divider.
R D 2320 , with 6-m lead, according to Fig. 3.43
Rd
(gt) input
ZL = 270 Ω
tL = 20 ns
65pF
L′
t L′
t L′
t − Lu
R′
t R′
t − Ru
C′
e
Ru
Lu
gt (t)
4 sections
R′
t = 2k Ω; Rt = 20kΩ; Ru = 20Ω
L′
t = 10mH; Lt = 100mH; Lu = 0.1mH
(Lt /Rt = Lu/Ru = 5 ns)
C ′
e = pF; Ce = 5pF;
Rd = 100 Ω
0 20 100 300 500 700
t(ns)
0.5
1
1.5
g
t
(
t
)
270 Ω
500 Ω
(gt
) input
Figure 3.45 Computed USR of resistor divider with inductance
voltage now starts with a time delay of about 15–20 ns related to the input
voltage. This delay is caused by the travel time of the divider, 8t D
p
LtCe,
which is about 22 ns. A stray capacitance C0
p would only theoretically suppress
this travel time, as was shown in reference 50. The very small Ce values
assumed in this example should simulate a very good shielded divider. This
rise time of the response is accordingly quite short. If the response time is

calculated as before, an additional small negative term within the brackets
(eqn (3.92)) would appear, being RdLt/Rt2
. Whereas this term is negli-
gable, the additional time delay decreases the computed response time of the
dividers, RtCe/6. This effect may thus be taken into account by a reduction
of this value by a factor of 3/4, and for this some standards6
recommend
a theoretical contribution of the response time for resistor dividers according
to RtCe/8.
Although many other investigations54
have confirmed the validity of such
analytical treatments, a theoretical treatment alone will not satisfy the actual
needs. In practice, an experimentally performed step response measurement
has to prove the performance of the systems used, as insufficient constructions
may show worse results. There are still some unsolved problems inherent to
response measurements, but they shall not be treated here in detail. The most
difficult problem is related to the fact that the actual ‘input terminals’ of our
voltage measuring system (Fig. 3.22) are separated by a long distance due to
the dimensions of the test object. But the existence of a step voltage with a rise
time of one nanosecond or less cannot be defined by an electrostatic potential
difference between points the distance of which is in the metre-range, due
to the basic theory of electromagnetic phenomena. But we still may assume
that actual currents charging the test objects within a short but finite time will
produce a quasi-stationary field and through it a potential difference for which
the expression ‘voltage’ is justified.
With such restrictions in mind, which are usually neglected, the measure-
ment of the step response belongs now to the routine procedure of an ‘accep-
tance test’ for an ‘approved measuring system’ as used for the measurement
of lightning and even switching impulses, see reference 53. The step volt-
ages are usually produced by small generators for some hundred volts, and
some different circuit arrangements can nowadays be used to measure the
step response. Although some further details can be found in the relevant IEC
Standard,53
some additional information is presented in Fig. 3.46. In this
a
b
c
f
g
d
e
Rd
Figure 3.46 The unit step method. (a) To d.c. supply. (b) Charging resistor.
(c) Fast switch. (d) (Added) vertical lead. (e) High-voltage lead. (f) Voltage
divider. (g) To recording instrument

figure, a so-called ‘square loop arrangement’ is shown as recommended by
the earlier standards.6,57
The step generator a to c must have approximately
zero impedance while generating the voltage step and during the subsequent
response. Any fast switching device, c, which short-circuits a constant d.c.
voltage as used to charge the measuring system before the short-circuit occurs
is applicable. Very suitable switches are mercury-wetted relays but also a
uniform field gap of about 1 mm spacing at atmospheric air or a uniform
gap with a spacing up to some millimetres under increased gas pressure. The
gas-insulated gaps can short-circuit after breakdown voltages up to some kilo-
volts, but only single events can be produced. Whereas these switches fulfil the
requirement for ‘zero impedance’, commercial types of electronic pulse gener-
ators are inherent with some 10 ohms of internal impedance which contribute
to the damping resistor, see Rd. At least a metallic strip conductor 1 m wide
shall serve as the earth return between divider and the step generator. The
length of the lead, represented by d and e, shall be equal to the length as
used during actual impulse voltage measurement, if the response time is an
essential parameter during the measurements. If only part e is used during
measurements, it would be possible to calculate the relevant response time
due to the theory as shown before and elsewhere.131
To avoid such effects,
the step generator shall be placed at a metallic wall. Then the usual lead length
as applied during voltage tests can be used. This ‘vertical lead arrangement’
was introduced by one of the authors60
and is the preferred circuit today.
One of the uncertainties of the unit step method is related to the starting
point, i.e. the value of 8de in Fig. 3.26, of the response. This starting point
on the ‘toe’ region is influenced by electromagnetic waves radiated from the
leads between the step generator and the divider. These phenomena have
been thoroughly investigated by an International Research Group;50
the
methods for the computation of these phenomena are based upon Maxwell’s
equations, which can either be solved in the time domain76
or in the frequency
domain.77
The solutions are very sophisticated and cannot be treated within
this chapter. The new IEC Standard,53
however, accounts for these effects
by defining a ‘virtual origin O1’ of the USR together with an ‘initial distortion
time T0’, the definitions of which may be found in this standard.
3.6.6 The divider’s low-voltage arm
We assumed for the theory of dividers, section 3.6.4, that the low-voltage arm
is an integral part of the divider and provides an impedance structure which is
equivalent to the high-voltage arm. In reality, the structure, i.e. the composition
of the circuit elements, is quite different. Therefore, some additional problems
may appear concerning adequate construction and layout of the l.v. arm of
our measuring system. Many distortions in the response can be related to this
part of the system.

For d.c. and a.c. voltage dividers, the design of the l.v. arm is not critical, if
only steady state voltages have to be recorded. However, if any fast transients
have to be transmitted from the voltage divider to the recording instrument
(see Fig. 3.23, items 5 to 7), the l.v. arm of the voltage divider itself may
introduce large disturbances to the response. Let us first discuss the adequate
impedance matching necessary to transmit impulse voltages from the divider
to the recording instrument.
In Fig. 3.47 the somewhat simplified equivalent circuits for the matching
procedures for the different types of dividers are sketched. The signal cable is
mainly treated as lossless, so that the surge impedance Zk D
p
Lk/Ck becomes
independent of frequency, and the travel time 8k D
p
LkCk is a plain value.
For resistor voltage dividers, Fig. 3.47(a), the cable matching is simply done
by a pure ohmic resistance R D Zk at the end of the signal cable. The trans-
mission line theory provides the well-known background for this procedure,
the reflection coefficient becomes zero and any unit step voltage appearing
across R2 is undistorted transmitted by the cable. As the input impedance of
the signal cable is R D Zk, this resistance is in parallel to R2 and forms an
integral part of the divider’s l.v. arm. The low-value of this resistance R, i.e.
typically 50 to 75 , should in fact suggest that we consider the losses of the
signal cable. These losses are in reality dependent upon frequency due to the
skin effect, and the response of such a cable becomes very complex. Whereas
the theory of this problem may be found elsewhere,124,125
the result of this
theory shows clearly that the best matching can be achieved with R equal to
the surge impedance Zk defined for high frequencies. For all kinds of signal
cables the d.c. resistance for the conductors (inner conductor and shield) will
form a voltage dividing system between R2 and R, which may decrease the
voltage across R by an amount of 1 per cent in order of magnitude. As this
amount can easily be taken into account by d.c. resistance measurements only,
this value should be taken into account. The unit step response from a lossy
cable is characterized by a steep increase within a few nanoseconds to values
of more than 90 per cent and a slow tripling up to the final value. These
effects will introduce larger errors if the impulses to be transmitted are shorter
than 0.5 to 1 µsec. At least an experimental test is recommended to check the
signal cable with regard to this additional error.
For parallel-mixed resistor-voltage dividers the same procedure for cable
matching, Fig. 3.47(a), applies. A matching resistor R, coaxially designed
to meet the high-frequency requirements, will not reflect energy. The input
impedance of the recording instrument, however, should not comprise appre-
ciable input capacitance, as otherwise too heavy reflections will appear. The
l.v. arm for this type of divider reflects heavily due to the parallel capacitance
to R2.
For capacitor voltage dividers, Fig. 3.47(b) or (c), the signal cable cannot
be matched at its end. A low ohmic resistor in parallel with C2 would load the

R1
R = Zk R = Zk
C1
R2
Ck
(a) (b)
(c) (d)
Zk ; τk
C2
R3
C4
R4
C1
C2
C1
Zk ; Ck
Zk; Ck Zk; Ck
(Zk− R2
)
C2
R1
C1
R2
C2
CRO CRO
CRO CRO
Figure 3.47 Circuits for signal cable matching. (a) Resistor or
parallel-mixed capacitor-dividers. (b) Capacitor dividers, simple matching.
(c) Capacitor dividers, compensated matching. (d) Damped capacitor divider,
simple matching
l.v. arm of the divider too heavily and decrease the output voltage with time.
To avoid travelling wave oscillations, the cable must then be terminated at its
input end. Then, a voltage step of constant amplitude at C2, i.e. C2 ! 1, will
be halved by R D Zk at the cable input end, as R and Zk form a voltage divider.
This halved voltage travels to the open end and is doubled by reflection. Thus
the original amplitude of the voltage across C2 appears at the input of the
recording instrument. The reflected wave charges the cable to its final voltage
amplitude, and is absorbed by R, as the capacitor C2 forms a short-circuit. In
reality, C2 is of finite value and is therefore discharged during these transient
events. The computation shows that the discharge period is very close to
twice the travel time. After this time, the cable capacitance is charged to
the final voltage, and from this we obtain two ratios of the voltage divider,
namely:
no D
C1 C C2
C1
for t D 0;
ne D
C1 C C2 C Ck
C1
for t ½ 28k.

The signal cable, therefore, introduces an initial ‘overshoot’ of the voltage of
V D ne/no 1 D Ck/C1 C C2, which may well be neglected for short
or medium cable length and high values of C2, i.e. high ratios of the voltage
dividers.
But capacitor dividers are often used for field testing of transient voltages
and longer cables thus are often necessary. The response can be improved by
transferring a part of the l.v. capacitor C2 to the cable end and connecting it
in series with a resistor, Fig. 3.47(c). This system, first treated by Burch,120
offers some opportunities to decrease the overshoot effect. Burch proposed to
make both matching resistances equal and R3 D R4 D Zk. If then the condition
C1 C C2 D C3 C Ck is satisfied, the initial and infinite time values of the
voltage become the same, and the original overshoot of about Ck/Cl C C2
is reduced to about 1/6. There are, however, further opportunities to improve
the response as shown by Zaengl.121
From high voltage arm
From high voltage arm
Circuit elements
(R2, C2)
Circuit elements
(R2, C2)
Metal housing
Metal
housing
Matching impedance,
if necessary
Matching impedance,
if necessary
Signal cable
connection
Signal cable
connection
Cu-disc, silver plated
Figure 3.48 Sketched cross-sections of possible layouts for the l.v. arm of
voltage dividers

For damped capacitor dividers, the resistors R1 and R2 necessary within the
l.v. arm are for the reflected wave in series to the matching impedance at the
l.v. arm, see Fig. 3.47(d). As R2 is very small in comparison to R1, the value
of this matching resistor must only be reduced by the small value of R2. The
methods of Fig. 3.47(c) can also be applied.
Whereas matching resistors for coaxial cables, i.e. resistors between inner
and outer conductors, are commercially available, the series resistors for all
capacitor dividers are an integral part of the divider’s l.v. arm. It may well be
recognized that the path to earth for the reflected wave should not be hindered
by too high inductances. This condition dictates the need for every l.v. capac-
itor to have a very low inductance. The theoretical value of this inductance is
given by the ratio of the divider, which divides also the overall inductance of
the stacked capacitor column. The physical size of the capacitance C2 values
necessary to reduce the high-voltage to a measurable quantity is, however,
relatively large. The coaxial arrangement of any circuit elements used within
the l.v. arm is a condition which should be strictly followed. In Fig. 3.48
simple cross-sections of possible layouts of the l.v. arm are sketched. Radi-
ally arranged elements tend to give even lower inductance values. The metal
housing avoids the impact of electromagnetic fields. By the coaxial connection
of the original cable, the input loop of this cable will not ‘see’ any effective
magnetic field, which contributes to reduced voltages by the transient currents.
Therefore, the current paths must be evenly distributed within the coaxially
arranged elements. A too low inductance may easily be increased by the addi-
tion of small wire loops in series with the elements. For more information
reference should be made to the literature.123,126
3.7 Fast digital transient recorders for impulse
measurements
With the advent of high-speed digital recorders the field of high-voltage
impulse testing has gained a powerful tool. Recent technological developments
have made the use of digital recorders possible in the field of high-voltage
impulse measurements.78–84
Their use has important advantages over the use
of traditional analogue oscilloscopes and recorders. Obtaining impulse test
records in digital form allows for the introduction of the wide range of digital
signal processing techniques into the analysis of high-voltage test data. These
techniques enable high-voltage test engineers to correct errors due to non-
ideal voltage dividers, to eliminate the effects of slight variations in the form
of applied impulses in successive test records, to analyse test object transfer
functions rather than merely visually examine the forms of applied stimuli and
recorded responses, and to perform statistical analyses on the results of long
series of impulse tests.85–89

Although digital techniques have been available for many years, it is only
over the last decade that their use for measurements in high-voltage impulse
tests has become widespread: all the standards covering digital recorders have
been published in the last twelve years. Digital recorders for general use are
covered by IEEE Std 1057-1994:135
there are no IEC Standards which cover
digital recorders for general use. There are presently two standards that deal
with digital recorders for measurements in high-voltage impulse tests: IEEE
Std 1122-1998136
(revision of IEEE Std 1122-1987) and IEC document 1083-
1: 1991137
(a revision is being prepared based on IEEE Std 1122-1998).
These standards define terms, identify quantities that affect accuracy, describe
tests and set minimum standards to be met by each tested parameter. In many
cases, it is not possible to isolate one parameter and test it alone and hence
the limits have to allow for other contributions, e.g. noise on the record. Many
test engineers make use of these standards to make a preliminary selection of
digitizers based on the manufacturer’s specifications but the selected digitizer
has then to be tested according to the standard. The standards set limits on the
sampling rate and the rated resolution. In addition limits are set on measured
values of integral non-linearity of the amplitude and the time base, the differ-
ential non-linearity of the amplitude, the impulse scale factor, the rise time,
the internal noise level, and the effects of interference and ripple.
The following sections of this chapter review the development and
fundamental operating principles of digital recorders, outline how these
devices differ from their analogue predecessors, cover the sources of static
and dynamic errors inherent in digital recorders, and finally provide insight
into the test procedures and minimum performance requirements mandated in
current standards136,137
related to the use of digital recorders in h.v. impulse
testing.
3.7.1 Principles and historical development of transient digital
recorders
The first attempts at digital recording of non-repetitive pulses were under-
taken in the late 1950s in order to enable on-line processing of recorded
transients. Research in nuclear physics and radar signature analysis prompted
the design of a hybrid oscilloscope–TV camera system, which employed a
conventional high writing-speed CRT coupled to a television vidicon tube
scanning the CRT screen. Although very useful in this pioneering period,
this hybrid recorder combined the drawbacks of both analogue and digital
systems. However, the concept of fast writing on a temporary storage medium
and scanning this medium later at a slower rate was a valuable innovation,
and the hybrid recorder paved the path to more refined designs. This idea
spawned the development of recorders which formed the functional basis of
present day scan converters. In the late 1970s electronic circuits utilizing solid

state components for high-speed analogue-to-digital conversion came into use.
Since then several other mechanisms for realizing the A/D conversion process
have been implemented.
The recording errors which characterize a recorder’s dynamic accuracy
depend to a certain extent on the design and operating principles of the
instrument.90–93
To gain an understanding of the physical principles respon-
sible for these errors it is useful to review briefly the various available designs
of recorders. There are four basic A/D conversion schemes utilized in present
day high-speed digital recorders. These are: scan conversion, charge coupled
device storage, flash conversion, and ribbon beam conversion.
Scan converters
The highest speed A/D conversion technique presently available utilizes scan
converters. These consist essentially of an analogue cathode ray tube with
the electron beam writing on a matrix of precharged semiconductor elements.
The moving electron beam leaves a trace of discharged elements on the target
matrix. The target is then read by another beam which scans the matrix at a
slower pace. The slowed down replica of the recorded transient is digitized by
a conventional ADC and stored for further display and processing. The main
errors in such an instrument come from the difference between the electron
beam writing speed at the steep and flat portions of the recorded transient.
The variation in writing speed results in blooming or thickening of the trace
in its slower portions similar to that often encountered in storage oscilloscopes
when recording transients with slow and fast portions. However, with a scan
converter, data processing can be used to reduce errors caused by blooming
and the instruments can achieve a vertical resolution of approximately 1 per
cent at a sweep of 5 ns. Further errors are generated by variations in the time
base. In addition to these drawbacks, short record length is a limitation inherent
in this recording technique. Despite these shortcomings, scan converters are
used in many laboratories where high bandwidth and very high equivalent
sampling rates are needed.94
Charge coupled device converters
Development of the charge coupled device (CCD) formed the basis of another
fast digitizer. The CCD input of such a digitizer (often referred to as the
bucket-brigade circuit) consists of a number of capacitors and electronic
switches integrated on a chip and driven by a clock which can operate at
a fast and a slow rate. The first sample of the transient to be recorded is
taken by connecting the first capacitor to the instrument’s input terminal for a
short period. At the next stage, the charge accumulated in the first capacitor is
transferred to the second one and the first capacitor is discharged in preparation
for the next sample. This procedure is repeated at the high clock rate until the

first sample is transferred to the last capacitor. This completes the recording
cycle and the recorded transient is stored in the form of charges accumulated
in each of the capacitors contained in the CCD. In order to read the signal, the
bucket-brigade charge transfer process is repeated in the reverse direction at the
slow clock rate. The subsequent samples are then digitized by a conventional
ADC at the output terminal of the CCD and stored in the memory for further
processing.
By using an advanced control system several MOS integrated circuit regis-
ters can be charged sequentially at a high rate thereby increasing the available
record length. Although the output signal from the registers can be digitized
(at a slower rate) with high resolution, the actual gain accuracy of CCD-based
recorders is limited to approximately 1 per cent which is often much less than
the high resolution of the actual slow sampling rate A/D converter.95
This
limit is set by the analogue techniques of sampling the recorded transients
with CCD elements.
Flash converters
The third type of fast digitizer is based on the conventional flash conver-
sion technique implemented in advanced semiconductor technology.96,97
In
higher resolution instruments two low-resolution monolithic flash converters
operating at very high sampling rates are used in a subranging mode. The tech-
nique is known as dual rank flash conversion. The input signal is digitized by
the first ADC and the digital output is fed into a digital-to-analogue converter
(DAC). The DAC output is subtracted from the delayed input signal and the
residue is fed into the second ADC to give an overall resolution of double
that of each ADC used. The principal limitation of this digitization scheme
is the time required for the DAC to settle so that the subtraction yielding the
second ADC’s input can be performed with sufficient accuracy. This sets the
limit on the maximum sampling rate of dual rank flash converters.
A variation of the flash conversion technique can be implemented by using
two high-resolution but slower sampling rate ADCs to sample the input alter-
nately, thus giving an effective sampling rate of twice that of each ADC. In
this type of digitizer additional errors can arise from differences in the quanti-
zation characteristics of the two ADCs and/or from asymmetry in the sampling
intervals.
Ribbon beam converters
The last design principle used in digital recorders to achieve high resolution
combined with high sampling rate incorporates an electron bombarded
semiconductor (EBS) tube.98
The EBS tube is similar to a conventional
cathode ray tube (such as those used in analogue oscilloscopes) except that the
luminescent screen is replaced by a target made up of a set of N adjacently

positioned strip diodes. The beam is flat in the horizontal plane like a ribbon
rather than being focused to a point, and there is only one set of plates used
to deflect the electron beam. A metal mask, installed between the electron
beam and the diode target, has a pattern of windows cut in it so that as the
horizontal beam moves up and down with the input signal it illuminates various
combinations of exposed diodes. A maximum of 2N
unique combinations of
the N diode states are available. The output of each diode strip is checked
at an interval corresponding to the recorder’s sampling rate to generate an
N-bit binary encoded word which corresponds to the level of the input signal
present at the instant that the diode outputs are checked.
Although the different high-speed digital recorders described above may
appear similar from the input/output perspective the different conversion
schemes can introduce different errors. In fact, the errors obtained using a high-
speed recorder are in many instances a direct consequence of the conversion
techniques implemented in the particular instrument. In order to assess the
actual measuring properties of a digitizer, it is therefore important to examine
not only the sampling rate and resolution, but also the type of A/D conversion
technique utilized.
3.7.2 Errors inherent in digital recorders
In contrast to an analogue oscilloscope which writes a continuous trace of the
measured signal, a digital recorder is able to record and store only instan-
taneous values of the signal rounded to integer numbers and sampled at a
certain rate over a finite period of time. The input signal is reconstructed by
positioning in time (according to the sampling rate) and the vector of consec-
utive values contained in the recorder’s memory. This leads to the presence of
two types of recording errors which are generally referred to as quantization
and discrete time sampling errors. These errors are the only ones present in
an ideal recorder. Since they occur even when recording slowly changing or
even d.c. input signals they are often referred to as static errors.
Static errors
The quantization error is present because the analogue value of each sample is
transformed into a digital word. This A-to-D conversion entails a quantization
of the recorder’s measuring range into a number of bands or code bins, each
represented by its central value which corresponds to a particular digital code
or level. The number of bands is given by 2N
, where N is the resolution
of the A-to-D converter. The digital output to analogue input relationship of
an ideal digitizer is shown diagrammatically in Fig. 3.49. For any input in
the range (iVav 0.5 Ł Vav to iVav C 0.5 Ł Vav), where Vav is the
voltage corresponding to the width of each code bin, or one least significant
bit (LSB), and iVav is the centre voltage corresponding to the ith code, an

ln
nv
(n-1)∆vav
(n-2)∆vav
(n-3)∆vav
(n-4)∆vav
3 ∆vav
2 ∆vav
0
∆vav
ln -1
ln -2
ln -3
ln -4
l3
l2
l1
l0
D
I
G
I
T
A
L
O
U
T
P
U
T
A
N
A
L
O
G
U
E
I
N
P
U
T
Figure 3.49 Analogue input to digital output relation of an ideal A/D
converter
Digital
output
code
Time
Figure 3.50 Response of an ideal A/D converter to a slowly rising ramp
ideal digitizer will return a value of Ii. Therefore, the response of an ideal
digitizer to a slowly increasing linear ramp would be a stairway such as that
shown in Fig. 3.50. A quick study of these figures reveals the character of
the quantization error associated with the ideal A-to-D conversion process.
The maximum error possible is equivalent to a voltage corresponding to š1
2

of an LSB. For an ideal digital recorder, this quantization would be the only
source of error in the recorded samples. For a real digital recorder, this error
sets the absolute upper limit on the accuracy of the readings. In the case of an
8-bit machine, this upper limit would be 0.39 per cent of the recorder’s full-
scale deflection. The corresponding maximum accuracy (lowest uncertainty)
of a 10-bit recorder is 0.10 per cent of its full-scale deflection.
The error caused by discrete time sampling is most easily demonstrated with
reference to the recording of sinusoidal signals. As an example we can look
at the discrete time sampling error introduced in the measurement of a single
cycle of a pure sine wave of frequency f, which is sampled at a rate of four
times its frequency. When the sinusoid and the sampling clock are in phase,
as shown in Fig. 3.51, a sample will fall on the peak value of both positive
and negative half-cycles. The next closest samples will lie at )/2 radians from
the peaks. As the phase of the clock is advanced relative to the input sinusoid
the sample points which used to lie at the peak values will move to lower
amplitude values giving an error () in the measurement of the amplitude
(A) of
 D A1 cos #
Figure 3.51 Sample points with sinusoid and sampling clock in phase.
(Error in peak amplitude D 0)
where # is the phase shift in the sample points. This error will increase until
# )/4 (Fig. 3.52). For # )/4 the point behind the peak value will now
be closer to the peak and the error will decrease for a # in the range of )/4
to )/2. The maximum per unit value of the discrete time sampling error is
given by eqn 3.93,
max D 1 cos)fts 3.93
where ts is the recorder’s sampling interval and f the sinewave frequency.

∆
Figure 3.52 Sample points with sampling clock phase advanced to )/4
with respect to the sinusoid. Error in peak amplitude () is at a maximum
The maximum errors obtained through quantization and sampling when
recording a sinusoidal waveform are shown in Fig. 3.53. The plotted quantities
were calculated for an 8-bit 200-MHz digitizer.
Error
∆MAX OUANT.
∆MAX SAMPL
∆MAX SAMPL
Frequency (MHz)
1
2
1
.4
0
2 3 4 5 6 7 8 10 20
ts
s/t
Figure 3.53 Sampling and quantization errors of an ideal recorder
In a real digital recorder, an additional two categories of errors are
introduced. The first includes the instrument’s systematic errors. These are
generally due to the digitizer’s analogue input circuitry, and are present to
some degree in all recording instruments. They include such errors as gain

drift, linearity errors, offset errors, etc. They can be compensated by regular
calibration without any net loss in accuracy. The second category contains
the digitizer’s dynamic errors. These become important when recording high-
frequency or fast transient signals. The dynamic errors are often random in
nature, and cannot be dealt with as simply as their systematic counterparts and
are discussed below.
Dynamic errors
In an ideal digitizer each sample of the recorded transient is taken in an
infinitely short time window. This precludes any variation in the width of
the time window which is often termed aperture uncertainty or jitter. Simi-
larly, there is no uncertainty in the time at which the sample is taken. The
widths of all the code bins are equal, and symmetrical about the level repre-
senting each bin’s centre. When a real digitizer is constructed, the non-ideal
nature of all these parameters comes into play. While their effects can be
reduced to a negligible level in recorders characterized by slower sampling
times, in those digitizers which push today’s technology to its limits so as to
provide the highest resolution coupled with the fastest sampling rates, these
parameters may have a significant effect on the dynamic accuracy of the
instruments.91,93,99–113
The nature, magnitude and consequences of the random errors
encountered in real digital recorders vary depending on the instrument’s
design.100,114,115,95
As a result the determination of the suitability of a
particular type of recorder to h.v. impulse testing must consist of two phases.
The initial phase consists of specifying the sampling rate and resolution if
ideal digitizer performance can be assumed. Following this, the dynamic
performance of the recorder under consideration must be determined. The
former requires that the digitizer’s nominal resolution in bits and sampling
rate be related to the accuracies required by the standard for instruments used
in recording h.v. impulses. The latter entails determination of the nature and
limits of the recorder’s dynamic errors. This approach was followed in the
development of IEEE Std 1122136
and IEC Pub. 1083,137
the standards
dealing with qualification of digital recorders for h.v. impulse measurements.
3.7.3 Specification of ideal A/D recorder and parameters required
for H.V. impulse testing
Standard impulses used in H.V. testing of power apparatus
Requirements on the accuracy of recording instruments used in high-voltage
impulse testing vary according to the type of tests, and depend upon the nature
of the test object itself, e.g. testing and research into the dielectric strength
of gas-insulated substations (GIS) involve generation and measurement of

very steep-front high-voltage impulses which may be chopped after a time as
short as 0.1–0.2 µs. This type of test does not require a very high measuring
accuracy, i.e. an uncertainty of a few per cent on the impulse crest is
acceptable. Details regarding impulse testing procedures of testing various
types of insulations are described in Chapter 8, section 8.3.1, with appropriate
references to national and international standards. Definitions of the prescribed
standard voltage waveshape to be used in testing can be found in Chapter 2,
Figs 2.23(a) to (c).
Peak measurements
As previously discussed, the maximum ideal quantization error is equal to
1 LSB. This can be restated in terms of per cent of full scale as 100/2N
,
where N is the recorder’s resolution in bits. For the error to be less than or
equal to 2 per cent of full-scale deflection a simple calculation shows that
the recorder’s resolution should be not less than 5.7 bits. This must of course
be rounded up to the closest integer, 6. Therefore, for a full-scale signal, an
ideal 6-bit recorder will meet the amplitude accuracy requirement. In practice
it is not possible, or at least not practical, to ensure that all the signals to be
recorded will span the instrument’s full measuring range. It is more realistic
to assume that the input signals will fall within a range of between 50 per
cent and 90 per cent of full scale. For a signal which covers 50 per cent of
the recorder’s full scale, an accuracy limit of 2 per cent would be maintained
by using an ideal recorder of 7-bit resolution. The magnitude of the maximum
possible discrete time sampling error depends upon the shape of the impulse
rather than on its magnitude. Therefore, the sampling-rate requirement must
be examined for each different impulse shape. Using the fastest allowable
standard lightning impulse as a basis for calculation, the maximum errors in
peak voltage measurement as a function of the recorder’s sampling rate can
be calculated using the same approach as previously illustrated for the case of
a sinusoidal input. The maximum possible errors resulting from sampling are
given in Table 3.6.
As can be seen from the tabulated results the discrete time sampling error
is negligible in comparison to the quantization error when recording the peak
value of a full lightning impulse.
Tests often require the use of chopping the standard lightning impulse wave
on the front or the tail as shown in Figs 2.23(c) and (b) respectively. With a
standard lightning impulse chopped after 2 µs to 5 µs (tail), since the chopping
occurs after the peak, the analysis used to derive nominal recorder character-
istics required for acceptable recording of the peak value is identical to that
outlined above for the case of the standard lightning impulse.
For testing with front-chopped impulse voltages two cases must be consid-
ered. The first of these is termed a front-chopped impulse and the second a

Table 3.6 Maximum error due to
sampling when recording the peak
of a full standard lightning impulse
Sampling rate Error in % FSD
(MHz)
2 0.195
2.8 0.097
4.0 0.048
5.7 0.024
8.6 0.012
12.0 0.006
linearly rising front-chopped impulse. As far as the measurement of their peak
values is concerned these two impulses can be treated under the same analysis.
For impulses chopped on the front, the most demanding situation covered
by today’s standards specifies that the peak value of an impulse with a time-
to-chop of 500 ns be recorded with the recording device contributing an error
of less than 2 per cent. In this situation, the discrete time sampling error must
be considered as well as the quantization error. For an ideal sawtooth input
(the limiting case of a front-chopped impulse) the maximum possible discrete
time sampling error is given by the product of the sampling interval and the
signal’s rate of rise. This is illustrated in Fig. 3.54.
Values of maximum discrete time sampling error (in per cent of the peak
value) vs recorder sampling rate are shown in Table 3.7 for the shortest stan-
dard time-to-chop of 500 ns.
Table 3.7 Maximum error due to sampling
when recording the peak of an ideal
sawtooth waveform with a 500 ns rise time
Sampling rate Error in % of peak value
(MHz)
50 4
100 2
200 1

u
1.0
0.9
0.5
0
E
0.05 Tr
0.05 Tr
Tr /2.5 t
Figure 3.54 Sampling error for an ideal sawtooth signal
From a comparison of Tables 3.6 and 3.7 it can be seen that the requirements
placed on sampling rate are far more severe when recording front-chopped
lightning impulses than when recording standard full lightning impulses or
impulses chopped on the tail. As shown in Table 3.7, a minimum sampling
rate of 100 MHz is required to ensure a discrete time sampling error of 2 per
cent or less in the measurement of the peak.
In reality high-voltage impulses are rounded at the chop. Figure 3.55 shows
an example of a linearly rising impulse applied to a 250 mm sphere gap with
a gap length of 60 mm. The slope of such a linearly rising front-chopped
impulse is, according to the standards, taken as the slope of the best fitting
straight line drawn through the portion of the impulse front between 50 per
cent and 90 per cent of its peak value. The rise time is defined as being the
time interval between the 50 per cent and 90 per cent points multiplied by 2.5.
The impulse shown in Fig. 3.55 was measured using a small divider insulated
with compressed gas and characterized by an extremely low response time.
The measured rate of rise of 10.8 kV/ns and time-to-chop of 36 ns are much
more severe than those typifying standard test impulses. However, even in
this very severe case, the waveshape shows that the slope close to the chop is
very much less than the impulse’s rate of rise as calculated between the 50 per
cent to 90 per cent points on the impulse front. For higher peak voltages, that
is larger breakdown distances and larger (hence slower) impulse measuring
systems, the rounding effect just prior to the chop will be more pronounced.
This means that in practice the slope at the instant of chopping is significantly
less than (Vpk/500 kV/ns. Since this was the figure used in calculating the

0
−365
−370
−375
−380
−385
−100
−200
−300
−400
0 10 20 30 40 50
Time (ns)
Voltage
(kV)
40 42 44 46 48
Figure 3.55 Actual record of a linearly rising front-chopped h.v. impulse.
(Inset shows every fifth sample in the vicinity of the chop)
minimum sampling rates necessary to limit the discrete time sampling errors
to acceptable values, the requirements on the sampling rate can be relaxed
from those given for the case of the sawtooth wave.
There are two factors responsible for the rounding of the impulse wave near
the time of chopping. The first is the mechanism of the breakdown which is
responsible for the voltage collapse. Although not in all cases, this mech-
anism often contributes a significant amount to the rounding. The second,
often more important, factor in rounding the impulse peak is the limited band-
width of high-voltage measuring systems. Because of their physical size, these
systems cannot respond quickly to fast changes in their input. The rounding-off
introduced in the measurement of an ideal sawtooth waveform with a 500 ns
rise time is illustrated in Fig. 3.56. The response to such an input is shown
for high-voltage measuring systems of varying response times. The standards
require that a measuring system have a minimum response time of 0.05 times
the rise time of any linearly rising front-chopped impulse which it is used to
record.6,53
In the worst case of a 500 ns rise time this requirement translates
into a response time of 25 ns or less. Looking at the case of the measuring
system with 25 ns response time shown in Fig. 3.54 a significant rounding in
the area of the peak is evident.
A simple computer analysis can be used to calculate the values of sampling
frequency which are necessary to meet the 2 per cent accuracy requirement.

1.0
0.8
0.6
0.4
0 .2
0
0 1 2 3 4 5
f (MHz)
6 7 8 9 10
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
100
50
75
25
Tresponse(ns)
Front-chopped
Impulse
T(ms)
Figure 3.56 Illustrative example of a front-chopped impulse
(tchop D 0.5 µs) distorted by an analogue measuring system of a limited
bandwidth. Attenuation of higher spectral frequencies of the impulse by the
measuring system and corresponding distortion of the impulse form are
shown for a few values of the response time of the system
However, for the practical case of bandwidth limited measuring systems, the
minimum requirements on sampling rate necessary to ensure recording of the
peak are exceeded by those necessary to meet standard requirements on the
measurement of time parameters and front oscillations discussed below.
Measurement of time parameters
Since lightning impulses are defined by time as well as amplitude parameters,
the sampling rate required for the evaluation of the time parameters must
also be determined. As previously stated, the standards’ present requirement
on oscilloscopes used to record h.v. impulses is that they must allow for the
evaluation of time parameters with an accuracy of better than 4 per cent.
As with the determination of minimum sampling rates required to ensure
the specified accuracy in the measurement of the impulse peaks, each type of
impulse must be examined in turn to derive minimum sampling rates necessary
for required accuracy in the measurement of time parameters. A standard
full lightning impulse is defined by its front time and its time-to-half value
(Fig. 2.23(a)). The front time is defined as 1.67 times the time interval between
the instants at which the impulse is between 30 per cent and 90 per cent of its
peak value. To determine the minimum sampling rates necessary to evaluate
the front time to within 4 per cent the fastest allowable standard lightning
impulse must be examined. It is characterized by a front time of 0.84 µs.
This means that the shortest time interval which must be measured for such
an impulse is 503 ns. The time-to-half value is defined as the time interval

between the virtual origin (O1 in Fig. 2.23(a)) and the instant on the tail at
which the voltage has dropped to half of the peak value. The evaluation of
the time-to-half value entails recording a much greater time interval than that
required to determine the front time. Therefore, the requirement on minimum
sampling rate is set by the latter. For a standard impulse chopped on the tail
the shortest time interval to be recorded is also the front time and its minimum
value is the same as that of a standard full lightning impulse. The second time
parameter used to define a chopped impulse is the time-to-chop. However, by
definition this time interval is between 2 µs and 5 µs, so the minimum required
sampling rate must be based on the front time measurement. For a front-
chopped or linearly rising front-chopped lightning impulse the time interval
between the virtual origin (O1 in Fig. 2.23(c)) and the instant of chopping can
be much shorter. The minimum time-to-chop allowed for in the standards is
500 ns. This means that the minimum sampling rates required to record the
defining time parameters of full, front-chopped, linearly rising front-chopped,
or standard chopped impulses are approximately equal. To evaluate a time
interval of 500 ns with an accuracy of better than 4 per cent it is necessary to
sample at a rate of 500 MHz or more.
Recording of front oscillations
The standard test impulses (Fig. 2.23(a)) are based on pure bi-exponential
waves such as would be produced by the resistive and capacitive components
present in impulse generating and measuring systems. In practical realizations
of large high-voltage systems, there is always some inductance present in
the circuit. This inductance results in oscillations being superimposed on the
test impulses. A typical record of an impulse with superimposed oscillations is
shown in Fig. 3.57. As is expected, the maximum frequency of the oscillations
in the circuit decreases as the physical size of the circuit increases.
There are standard requirements placed on impulse test waveforms with
regards to the maximum amplitude of superimposed oscillations. In order to
0 2 4 6 8 t(ms)
Figure 3.57 Standard full lightning impulse with superimposed oscillations

ensure that these criteria are met, it is necessary to be able to evaluate the
peaks of the oscillations with a certain precision. The standard requirement
intended to ensure sufficiently accurate recording of the oscillations states
that the upper 3 dB cut-off frequency for surge oscilloscopes be at least two
times fmax. The oscillations present on impulses are damped sine waves.
Therefore, the formula which gives the maximum discrete sampling time error
as a function of signal frequency and sampling rate (eqn 3.93) detailing the
errors of ideal recorders can be applied in this instance. Examination of this
formula reveals that a maximum error of 30 per cent or 3 dB is obtained when
the sampling frequency is four times larger than the frequency of the sine
wave being sampled. Therefore, to correspond to the requirement placed on
surge oscilloscopes (f3 dB 2fmax) the sampling rate of the recorder must be
at least 8 times fmax.
Impulse tests on non-self-restoring insulation
H.V. impulse testing of apparatus such as power transformers, which contain
non-self-restoring insulation, often requires using non-destructive test tech-
niques. Such tests usually consist of checking the linearity of the test object
insulation impedance within a range of test voltages up to the basic insula-
tion level (BIL).85,116,117,118
The quantities monitored during these tests are
the applied voltage and the neutral current which is taken as the response
to the voltage application. The analysis of test records comprises a detailed
comparison of records taken at different voltage levels. Meaningful devia-
tions between the records indicate that the characteristics of the test object
are non-linear, and the test results in a failure. Since even small deviations
between records are meaningful and can result in the disqualification of a
very expensive piece of equipment, it is imperative that the recorder used
provides enough accuracy to resolve such differences. Fortunately, the mean-
ingful frequency content of the records to be examined is band limited to a
few MHz (typically less than 2.5 MHz). The requirements on oscilloscopes
used for such tests have never been accurately specified as the same impulse
oscilloscope has been used for monitoring tests on self-restoring and non-
self-restoring insulation. The parameters of an ideal digitizer which meets the
requirements are, therefore, not as easily determined as those of one appro-
priate for testing of self-restoring insulation. The derivation of the necessary
parameters can be approached in two ways. The first of these entails matching
the accuracy of the analogue h.v. impulse oscilloscopes which have become
the de facto standard instruments for use in these tests.119
The second is to
individually examine all the test procedures and methods of analysis asso-
ciated with the testing of objects containing non-self-restoring insulation. A
major problem encountered in realizing this second alternative stems from
the fact that different apparatus standards have different requirements which

must be used in the derivation of the required resolution and accuracy of the
digital recorder. This problem causes the latter approach to be less favourable
than the former. The high-quality surge oscilloscopes generally used in h.v.
impulse testing are characterized by a bandwidth of approximately 35 MHz.
The vertical or amplitude resolution of such an oscilloscope cannot be matched
by an 8-bit digitizer, but it is most certainly exceeded by the resolution of a 10-
bit recorder. The minimum sampling rate required for such tests is not clearly
specified, but can be deduced on the basis of test requirements. When moni-
toring tests on power transformers and reactors, the neutral current is usually
recorded by means of a shunt whose bandwidth is generally below 3 MHz. A
second aspect of the test which can shed light on the required sampling rate
is that in certain design tests involving the use of chopped waveforms, the
time taken for the voltage to collapse is specified as being between 200 ns and
500 ns. The accuracy with which this parameter is recorded is a direct func-
tion of the sampling rate of the recorder used. However, at present there is
no limit specified on the error permissible in the establishment of the duration
of voltage collapse. In order to ensure accuracies superior to those attainable
with presently used oscilloscopes, it appears that an ideal digital recorder with
an amplitude resolution of 10 bits and a sampling rate of 30 MHz is suitable
for recording tests on apparatus containing non-self-restoring insulation.
Digitizer dynamic performance
As mentioned earlier, the performance of a real digitizer does not usually
match that of its ideal counterpart. This deterioration is principally due to
the imperfect performance of the recorder components. The errors caused
by these imperfections are usually referred to as dynamic errors, and their
magnitude is generally proportional to the slew rate or the rate of change
of the input signal. This proportionality results in a deterioration of recorder
performance with increasing input signal steepness. The causes and effect of
the limited dynamic performance of digital recorders are briefly discussed
here. The dynamic errors can be quantified using four parameters: differential
non-linearity (DNL), integral non-linearity (INL), aperture uncertainty, and
internal noise. Three of these four parameters are assessed directly during the
qualification of digital recorders as laid out in references 136 and 137. The
effects of the fourth (aperture uncertainty) are included in the assessment of
the other three.
Differential non-linearity (DNL)
Differential non-linearity is defined as the variation in the widths of the
recorder’s code bins. For an ideal recorder all of the code bin widths are
equal and given by the recorder’s full-scale deflection divided by the number

Code bin centre
Code bin edge
k + 3
k + 2
k + 1
k − 1
k
Input (volts)
Digital
output
code
Figure 3.58 A portion of the quantization characteristic of an ideal digital
recorder
of quantization levels. If the recorder’s full-scale deflection is defined in bits,
then the width of each ideal quantization band or code bin is equal to 1
least significant bit (LSB). The recorder’s measuring range can, therefore, be
thought of as being divided into 2N
code bins or quantization bands (where
N is the recorder resolution) as shown in Fig. 3.58. In this figure the edges of
the quantization bands (or code bins) are basically assigned a zero probability
of occurrence. In a real recorder this sort of performance cannot be expected.
Since the rate of occurrence of the specific codes can only be determined
in statistical terms, the edges of the code bins are defined by a distribution
which may resemble the normal bell-shaped curve. These distributions reflect
the gradual transition from one state to the next. The areas in which this occurs
are referred to as code transition zones. The static profiles of these transition
zones can be obtained by repeated testing with incremented d.c. input volt-
ages. The results of this can be used in the establishment of a probability based
transition between adjacent code bins. An example of a portion of static quan-
tization characteristic obtained using such a method is shown superimposed on
its ideal counterpart in Fig. 3.59. As can be seen from the differences between
Figs. 3.58 and 3.59, even the static quantization characteristic obtained using
a d.c. input signal deviates from the ideal rectangular characteristic. Under
dynamic conditions the limited slew rates and settling times of the recorder’s
electronic components result in further aberrations of the quantization charac-
teristic. These can take the form of a further sloping out of the code transition
zones as well as a non-uniform widening or narrowing of the code bins. These
types of deviations are quantified in terms of the recorder’s differential non-
linearity (DNL), and their magnitudes vary with the signal slope. According
to the definition of DNL it can be seen that an ideal digitizer will have a DNL

10∆V
Wo
Input (i∆v)
A
ctt(k)
B
k − 1
k + 1
k + 2
k
k + 1
2
Output
Figure 3.59 Four codes of a static quantization characteristic. Dashed
curve is for an ideal digitizer whose average code bin width is w0 . The code
transition threshold from code k to k C 1 is marked (ctt(k)). Measured points
are shown as C
0
Avg
1.5 Avg
Probability
of
occurrence
Digital code
Code dnl = .5
Code dnl = −1
Code dnl = 0
Figure 3.60 Illustration of DNL values
of 0 as the width of every code bin is equal (as shown in Fig. 3.58). If a
digitizer is ideal except for one code which is missing while the two adjacent
codes are 50 per cent wider than the average, then the DNL of the missing
code is 1 and the DNL of each of the two adjacent codes is 0.5. An example
illustrating this is shown in Fig. 3.60. The effect of this DNL is to reduce the

local resolution of the digitizer by 50 per cent at the missing code. The figure
of merit used in literature to describe recorder differential non-linearity can
be expressed as a vector with each entry representing the DNL as measured
for each of the code bins, or alternately as a single figure. When the latter
representation is used, the number is taken to represent the maximum DNL
of all of those measured across the recorder’s range.
The differential non-linearity can also be measured in a dynamic test. This
is usually done by repeatedly recording a sinusoidal waveform and relating the
observed distribution of code occurrences to the probability density function
as expected for a sinusoidal input to an ideal recorder.
Integral non-linearity (INL)
A second parameter used to quantify digitizer performance is the integral non-
linearity (INL). Conceptually the INL can be thought of as an assessment of
how much the real quantization characteristic of the recorder deviates away
from the ideal. This is illustrated in Fig. 3.61.
Digital
output
code
Input signal
Ideal
Non-ideal
Figure 3.61 Integral non-linearity as shown by a deviation between the real
and ideal quantization characteristics
The difference between the differential and integral non-linearity lies in the
fact that, while the former evaluates the variations in the individual code bin
widths, the latter integrates across the DNL of all the code bins and gives a
feeling for the deviation of the recorder’s quantization characteristic from its
ideal counterpart. As with the DNL, the INL is assessed both statistically and
dynamically.

Aperture uncertainty
In addition to differential and integral non-linearity related errors a third signif-
icant error source associated with real digitizers is the aperture uncertainty or
sampling time dispersion. This is described as the variation in timing between
successive samples. The recorder will always introduce a delay between the
time that it is instructed to sample and the time at which the sample is actually
taken. If this delay were constant, then it would not project on the digitizer’s
measuring properties as it could be corrected for by realigning the samples in
time. Unfortunately, as is typical of physical processes, the delay follows a
statistical distribution thus precluding any simple corrections. As can be intu-
itively deduced from the definition of the aperture uncertainty, the magnitude
of errors stemming from this cause are dependent on the form of the signal
being recorded. For slowly varying inputs the magnitude of the errors is not
as prominent as for steeply rising signals. The numerical value of the aperture
uncertainty which is often quoted by manufacturers is defined as the standard
deviation of the sample instant in time.
Internal noise
Error due to internally generated noise is a fourth type of error associated
with real digital recorders. Noise in the digital recorder is equivalent to a
smearing of the quantization characteristic. The extent or severity of this effect
is of course dependent on the relative magnitude of the noise and the code
bin widths thus making high-resolution recorders more vulnerable. When the
noise is large, the quantization characteristic degenerates into an approxi-
mately straight line. Figure 3.62 illustrates the effect of various relative noise
amplitudes on the quantization characteristic. The figure shows a quantization
characteristic obtained on a 10-bit recorder. The noise, whose magnitude is
in the same range as the quantization step size or code bin width, causes the
quantization characteristic to appear as a nearly straight line. The effect of
varying degrees of noise magnitude can be illustrated by artificially increasing
the quantization step size. This is equivalent to reducing the resolution of the
recorder. For an effective resolution of 9 bits, adjacent levels in the recorder
were paired together. Similarly 8-bit resolution was simulated by summing
every four adjacent levels into one. Looking at the three quantization char-
acteristics shown in Fig. 3.62, the effect of higher and lower noise becomes
apparent.
3.7.4 Future trends
The preceding sections have described the development and current state of
digital recorders utilized for h.v. impulse testing. Rapid advancements in elec-
tronics technology will undoubtedly result in significant developments in this

494
496
498
500
502
504
506
Level
+4.200 +4.300
Input (volts d.c.)
10 bits
≡ 9 bits
≡ 8 bits
Figure 3.62 Quantization characteristics obtained by treating a 10-bit
digitizer as having 10-, 9-, and 8-bit resolution
area. Depending on the technologies used in the realization of future genera-
tions of these devices, it likely will be necessary to establish new or further
refined tests for establishing the ability of digital recorders to faithfully repro-
duce records of h.v. impulse tests. Evidence of this is reflected in the many
improvements and enhancements which have been made to commercially
available digitizers over the last decade. Today, various digitizers with rated
resolutions corresponding to 8 bits have sampling rates from 1.109
samples/s
to 1.1010
samples/s. These rates are higher than is needed for measurements of
lightning impulses and hence it is possible to trade sampling rate for improved
amplitude performance. Currently, two modes are of particular interest in
h.v. impulse measurements: the ‘peak detect’ mode and the ‘enhanced resolu-
tion’ mode. In both these modes the digitizer samples the input signal at the
maximum sample rate but the sample rate of the record is set at the maximum
sampling rate divided by an integer. Examples of the use of these features are
given in reference 138.
References
1. A.J. Schwab. High Voltage Measurement Techniques. MIT Press, Cambridge, Ma. and
London, England, 1972.
2. B.D. Jenkins. Introduction to Instrument Transformers. Newnes, London, 1967.
3. Comparative Calibration of Reference Measuring Systems for Lightning Impulse Voltages
According to IEC 60-2:1994 Proc. Int. Symp. on H.V. Engineering, Graz 1995, p. 4469.
4. A. Draper. Electrical Machines. Longmans, London, 1976.
5. IEC Publication 52, 2nd Edition (1960). Recommendations for Voltage Measurements by
Means of Sphere-gaps (one sphere earthed).

6. IEEE Standard Techniques for High Voltage Testing. IEEE Std. 4-1995.
7. F.S. Edwards and J.F. Smee. The Calibration of the Sphere Spark-Gap for Voltage Measure-
ment up to One Million Volts (Effective) at 50 Hz. JIEE 82 (1938), pp. 655–657.
8. H.E. Fiegel and W.A. Keen, Jr. Factors influencing the sparkover voltage of asymmetrically
connected sphere gaps. Trans. AIEE, Commun. Electron. 76 (1957), pp. 307–316.
9. E. Kuffel. The Influence of Nearby Earthed Objects and of the Polarity of the Voltage on
the Breakdown of Horizontal Sphere Gaps. Proc. IEE 108A (1961), pp. 302–307.
10. W.G. Standring, D.H. Browning, R.C. Hughes and W.J. Roberts. Effect of humidity on
flashover of air gaps and insulators under alternating (50 Hz) and impulse (1/50 µs) voltages.
Proc. IEE 110 (1963), pp. 1077–1081.
11. S. Guindehi. Influence of humidity on the sparking voltage in air for different kinds of
voltages and electrode configurations (in German). Bull. SEV. 61 (1970), pp. 97–104.
12. T.E. Allibone and D. Dring. Influence of humidity on the breakdown of sphere and
rod gaps under impulse voltages of short and long wavefronts. Proc. IEE. 119 (1972),
pp. 1417–1422.
13. W. Link. PhD Thesis (in German). TU Stuttgart, No. 203 (1975).
14. E. Peschke. Influence of humidity on the breakdown and flashover of air gaps for high d.c.
voltages (in German). ETZ-A vol. 90 (1969), pp. 7–13.
15. E. Kuffel. The Effect of Irradiation on the Breakdown Voltage of Sphere Gaps in Air under
Direct and Alternating Voltages. Proc. IEE 106C (1956), pp. 133–139.
16. F.O. MacMillan and E.G. Starr. Trans. AIEE 49 (1930), p. 859.
17. P.L. Bellaschi and P.H. McAuley. Impulse Calibration of Sphere Gaps. Electric Journal 31
(1934), pp. 228–232.
18. J.R. Meador. Calibration of the Sphere Gap. Trans. AIEE 53 (1934), pp. 942–948.
19. E. Kuffel. Influence of humidity on the breakdown voltage of sphere-gaps and uniform field
gaps. Proc. IEE, Monograph No. 3322 M, 108C (1961), 295.
20. R. Davis and G.W. Bowdler. The Calibration of Sphere Gaps with Impulse Voltages.
Journal of the IEE (London) 82 (1938), pp. 645–654.
21. J.M. Meek. The Influence of Irradiation on the Measurement of Impulse Voltages with
Sphere Gaps. Journal of the IEE (London) 93 (1946), pp. 97–115.
22. A.J. Kachler. Contribution to the problem of impulse voltage measurements by means of
sphere gaps. 2nd Int. Symp. on High Voltage Engg., Zurich, 1975, pp. 217–221.
23. W. Schultz. Erratic breakdown in air due to impurities in the presence of direct and alter-
nating voltages. 3rd Int. Symp. on High Voltage Engg., Milan, 1979, Report 52.05.
24. D. Peier and H. Groschopp. PTB-Mitteilungen 87 (1977), pp. 396–398.
25. W.O. Schumann. Elektrische Durchbruchfeldstaerke von Gasen. Springer, Berlin, 1923.
26. E. Kuffel and M. Abdullah. High-Voltage Engineering. Pergamon Press, 1970.
27. G.A. Schroeder. Zeitschr.f. Angew. Physik 13 (1967), pp. 296–303.
28. H.A. Boyd, F.M. Bruce and D.J. Tedford. Nature 210 (1966), pp. 719–720.
29. M.S. Naidu and V. Kamaraju. High Voltage Engineering. McGraw-Hill, 1995.
30. P. Paasche. Hochspannungsmessungen. VEB Verlag Technik, Berlin, 1957.
31. H. House, F.W. Waterton and J. Chew. 1000 kV standard voltmeter. 3rd Int. Symp. on High
Voltage Engg., Milan, 1979, Report 43.05.
32. J.H. Park. J. Res. Nat. Bur. Stand. 66C, 1 (1962), p. 19.
33. R. Davis. J. Sci. Inst. 5 (1928), pp. 305–354.
34. R.F. Goosens and P.G. Provoost. Bull. SEV. 37 (1946), pp. 175–184.
35. D. Peier and V. Graetsch. 3rd Int. Symp. on High Voltage Engg., Milan, 1979, Report 43.08.
36. C.T.R. Wilson. Phil. Trans. (A)221 (1920), p. 73.
37. L.W. Chubb and C. Fortescue. Trans. AIEE 32 (1913), pp. 739–748.
38. W. Boeck. ETZ-A 84 (1963), pp. 883–885.
39. R. Davies, G.W. Bowdler and W.G. Standring. The measurement of high voltages with
special reference to the measurement of peak voltages. J. IEE, London, 68 (1930), 1222.

40. R. Peiser and W. Strauss. Impulse peak voltmeter with extended measuring possibilities.
3rd Int. Symp. on High Voltage Engg. (ISH), Milan, 1979, Report 72.07.
41. J.G. Graeme. Designing with Operational Amplifiers. McGraw-Hill, New York, 1977.
42. W. Schulz. High-voltage ac peak measurement with high accuracy. 3rd Int. Symp. on High
Voltage Techn., Milan, 1979, Report 43.12.
43. J.C. Whitaker (ed.). The Electronics Handbook. IEEE Press, 1996.
44. R. Malewski and A. Dechamplain. Digital impulse recorder for high-voltage laboratories.
Trans. IEEE PAS 99 (1980), pp. 636–649.
45. W. Clausnitzer. Trans. IEEE IM 17 (1968), p. 252.
46. A. Keller. Symposium on Precision Electrical Measurements, NPL, London, 1955.
47. H.R. Lucas and D.D. McCarthy. Trans. IEEE PAD 89 (1970), pp. 1513–1521.
48. U. Brand and M. Marckmann. Outdoor high-voltage compressed-gas capacitors using SF6.
2nd Int. Symp. on High Voltage Engg., Zurich, 1975.
49. D.L. Hillhouse and A.E. Peterson. Trans. IEEE IM 22 (1973), No. 4.
50. IRR-IMS Group. Facing uhv measuring problems. Electra No. 35 (1974), pp. 157–254.
51. W. Zaengl. Arch. Techn. Messen (ATM), Blatt Z 130-3 (1969).
52. H. Luehrmann. ETZ-A 91 (1971), pp. 332–335.
53. IEC Publication 60: High-voltage test techniques. Part 2: Measuring Systems, 2nd Edition,
1994-11.
54. High voltage measurements, present state and future developments. Rev. Gen. Electr.,
Special Issue, June 1978.
55. A. Rodewald. Fast transient currents on the shields of auxiliary cables after switching
operations in hv substations and hv laboratories. IEEE PES Winter Meeting, New York,
1979, Paper No. A79 086-0.
56. A. Rodewald. Bull. SEV. 69 (1978), pp. 171–176.
57. IEC Publication 60-4 (1977). High-voltage test techniques, Part 4: Application Guide for
Measuring Devices.
58. A. Asner. Bull. SEV. 52 (1961), pp. 192–203.
59. F. Creed, R. Kawamura and G. Newi. Trans. IEEE PAS 86 (1967), pp. 1408–1420.
60. W. Zaengl. Bull. SEV. 61 (1970), pp. 1003–1017.
61. W. Zaengl and K. Feser. Bull. SEV. 55 (1964), pp. 1250–1256.
62. P.R. Howard. Errors in recording surge voltages. Proc. IEE II/99 (1952), pp. 371–383.
63. A.F. Rohlfs, J.F. Kresge and F.A. Fisher. Trans. AIEE 76 (1957), Part 1, pp. 634–646.
64. D. Pellinen and M.S. Di Capua. Rev. Sci. Inst. 51 (1980), pp. 70–73.
65. P.L. Bellaschi. Trans. AIEE. 52 (1933), pp. 544–567.
66. T. Harada, T. Kawamura, Y. Akatsu, K. Kimura and T. Aizawa. Trans. IEEE PAS 90
(1971), pp. 2247–2250.
67. R. Krawczynski. Correction of the high voltage measuring system by means of l.v. trans-
mission line. 2nd Int. Symp. on High Voltage Engg., Zurich, 1975, Report 3.1-07.
68. R. Elsner. Arch. Elektrot. 33 (1939), 23–40.
69. T. Harada et al. Trans. IEEE PAS 95 (1976), pp. 595–602.
70. A. Schwab and J. Pagel. Trans. IEEE PAS 91 (1972), pp. 2376–2382.
71. W. Breilmann. Effects of the leads on the transient behavior of coaxial divider for the
measurement of high ac and impulse voltage. 3rd Int. Symp. on High Voltage Engg., Milan,
1979, Report 42.12.
72. W. Zaengl. Bull. SEV. 56 (1965), pp. 232–240.
73. H. Dommel. Trans. IEEE PAS 88 (1969), pp. 388–399.
74. A. Schwab, H. Bellm and D. Sautter. 3rd Int. Symp. on High Voltage Engg., Milan, 1979,
Report 42.13.
75. H. Luehrmann. Archiv. fur Elektrot. 57 (1975), pp. 253–264.
76. N. Ari. Electromagnetic phenomena in impulse voltage measuring systems. Trans. IEEE
PAS 96 (1977), pp. 1162–1172.

77. K.H. Gonschorek. 3rd Int. Symp. on High Voltage Engg., Milan, 1979, Report 42.02.
78. R. Malewski and A. De Champlain. IEEE PAS 99 (1980), pp. 636–649.
79. H.G. Tempelaar and C.G.A. Koreman. 4th Int. Symp. on High Voltage Engg., Paper 65.07,
1983.
80. R. Malewski. IEEE PAS 101 (1982), pp. 4508–4517.
81. R. Malewski and B. Poulin. IEEE PAS 104 (1982), pp. 3108–3114.
82. B. Beaumont et al. IEEE PAS 96 (1987), pp. 376–382.
83. P.P. Schumacher and M.E. Potter. IEEE Power Engineering Summer Meeting, paper A.78,
1978.
84. T. Miyamoto. et al. IEEE IM 24 (1975), pp. 379–384.
85. K. Schon, W. Gitt. IEEE PAS 101 (1982), pp. 4147–4152.
86. R. Malewski et al. Canadian IEEE Communications and Power Conference, 1980.
87. Guo-Xiong, De-Xiang. 4th Int. Symp. on High Voltage Engg., paper 65.09, 1983.
88. S.J. Kiersztyn. IEEE PAS 99 (1980), pp. 1984–1998.
89. R. Malewski and B. Poulin. IEEE PD. 3 (1988), pp. 476–489.
90. K. Schon et al. 4th Int. Symp. on High Voltage Engg., paper 65.05, 1983.
91. T.M. Souders and D.R. Flach. National Bureau of Standard Special Publication 634,
Proceedings of the Waveform Recorder Seminar, 1982.
92. H.K. Schoenwetter. IEEE IM 33 (1984), pp. 196–200.
93. R.K. Elsley. National Bureau of Standards, Special Publication 596, Ultrasonic Materials
Characterization, pp. 311–317, 1980.
94. W.B. Boyer. National Bureau of Standards, Special Publication 634, Proceedings of the
Waveform Recorder Seminar, pp. 88–96, 1982.
95. T.R. McComb. IEEE PWRD 2(3) (1987), pp. 661–670.
96. W. Bucklen. Electronic Design, pp. 93–97, Sept. 1980.
97. W. Buchele. Electrical Design News, 1983.
98. D.E. Hartman. Electronic Packaging Production, Mar. 1980.
99. R. Maleski et al. IEEE IM 32 (1983), pp. 17–22.
100. R. Maleski et al. 4th Int. Symp. on High Voltage Engg., paper 65.06, 1983.
101. National Bureau of Standards, Special Publication 634, 1982.
102. R.A. Lawton. IEEE Instrumentation and Measurements Technology Conference, 1985.
103. M. Neil and A. Muto. Electronics 55(4) (1982), pp. 127–132.
104. B. Peetz. IEEE IM 32(1) (1983), pp. 12–17m.
105. L. De Witt. Handshake (Tektronix Newsletter) 5(1) (1980), pp. 9–12.
106. Gould. Biomation, Application Note 307.
107. LeCroy. Application Note, 1980.
108. LeCroy. Application Note, #-2004.
109. L. Ocho and P. McClellan. Handshake Application Library, 1977.
110. W.J. Hopkins, M.Sc. Thesis, Arizona State University, 1984.
111. T.R. McComb et al. CIGRE WG 33.03, Internal Working Document, No. 33-85, 1985.
112. H. Korff and K. Schon. IEEE IM 36(2) (1987), pp. 423–427.
113. Monsanto Research Company. Technical Report Commissioned by the US Department of
Energy, 1985.
114. R. Maleski and J. Kuffel. CIGRE WG 33.03 Internal Working Document No. 38-84, 1984.
115. J. Kuffel, et al. IEEE IM 35(4) (1986), pp. 591–596.
116. IEEE Guide for Transformer Impulse Tests. IEEE Std 93, 1968.
117. E.C. Rippon and G.H. Hickling. AIEE 96(2) (1945), pp. 640–644.
118. G.W. Lengnick and S.L. Foster. AIEE 76(3) (1957), pp. 977–980.
119. Products Specifications #1432 and #F143.4, High Voltage Test Systems, Haefely Ltd. Basel,
Switzerland.
120. F.G. Burch. On potential dividers for cathode-ray oscillographs. Phil. Magazine, Series 7,
13 (1932), pp. 760–774.

121. W. Zaengl. ETZ-A 98 (1977), pp. 792–795.
122. K. Feser. Trans. IEEE PAS 93 (1974), pp. 116–127.
123. R. Malewski and N. Hylten-Cavallius. Trans. IEEE PAS 93 (1974), pp. 1797–1804.
124. Ramo, Whinnery and van Duzer. Fields and Waves in Communication Electronics. J. Wiley,
1965.
125. H.G. Unger. Theorie der Leitungen. Vieweg, 1966.
126. T. Harada et al. Development of high-performance low voltage arms for capacitive voltage
dividers. 3rd Int. Symp. on High Voltage Engg., Milan, 1979, Report 42.14.
127. G.W. Bowdler. Measurements in High-voltage Test Circuits. Pergamon Press, 1973.
128. E. Rinaldi, F. Poletti and A. Zingales. Constructive improvements in impulse peak volt-
meters. 4th Int. Symp. on High Voltage Engg., Athens, 1983, Report 61.02.
129. Q.-C. Qi and W. Zaengl. Investigations of errors related to the measured virtual front time
T of lightning impulses. Trans. IEEE PAS 102 (1983), pp. 2379–2390.
130. N. Hylten-Cavallius et al. A new approach to minimize response errors in the measurement
of high voltages. Trans. IEEE PAS 102 (1983), pp. 2077–2091.
131. E. Kuffel and W.S. Zaengl. High Voltage Engineering Fundamentals. Pergamon Press,
1984.
132. K. Schon. Zur Rückführbarkeit von Hochspannungsmessungen (To the traceability of high-
voltage measurements). PTB-Report No. E-49, Braunschweig/Germany (1994), pp. 41–60.
133. M. Beyer, W. Boeck, K. Moeller and W. Zaengl. Hochspannungstechnik (High Voltage
Techniques). Springer-Verlag, 1986.
134. Komson Petcharaks. Applicability of the Streamer Breakdown Criterion to Inhomogeneous
Gas Gaps. Dissertation ETH Zurich/Switzerland No. 11’192, 1995.
135. IEEE Std 1057:1994 (revision of Trial-use IEEE Std 1057-1989), IEEE Standard for digi-
tizing waveform recorders.
136. IEEE Std 1122:1998 (revision of IEEE Std 1122-1987), IEEE Standard for digital recorders
for measurements in high-voltage impulse tests.
137. IEC Document 61083:1991 Instruments and software used for measurements in high voltage
impulse tests – Part 1: Requirements for instruments and its revision, Committee Draft
42/144/CD.
138. T.R. McComb, J. Dunn and J. Kuffel. Digital Impulse Measurements Meeting Standards
While Pushing the Limits. ISH-99, August 1999, London, England.
139. L. Schnell (Editor). Technology of Electrical Measurements. John Wiley Sons Ltd. Chich-
ester, 1993.

Chapter 4
Electrostatic fields and field
stress control
In response to an increasing demand for electrical energy, operating trans-
mission level voltages have increased considerably over the last decades.
Designers are therefore forced to reduce the size and weight of electrical
equipment in order to remain competitive. This, in turn, is possible only
through a thorough understanding of the properties of insulating materials
and knowledge of electric fields and methods of controlling electric stress.
This chapter is therefore devoted to a discussion of some of the problems
encountered when analysing even relatively simple but practical insulating
systems. Teaching experience has shown that this is a necessary prerequi-
site in order to gain a clearer understanding of the behaviour of insulating
materials. However, no attempt will be made here to introduce the basic field
equations, or to treat systematically the numerous methods available for calcu-
lating electrostatic fields as this may be found in many books.1–4Ł
Rather,
this chapter is intended to provide some fundamental understanding of the
importance of the interaction between fields and materials involved within an
electrical insulation system by discussing some selected examples.
In h.v. engineering most of the problems concerned with the electrical
insulation of high direct, alternating and impulse voltages are related to electro-
static and sometimes electrical conduction fields only. It should be emphasized
however, that the permissible field strengths in the materials are interlinked
with the electrostatic field distributions and thus the problems may become
extremely difficult to solve.
4.1 Electrical field distribution and breakdown strength
of insulating materials
It is often assumed that a voltage V between two electrodes may be adequately
insulated by placing a homogeneous insulating material of breakdown strength
Eb which is considered as a characteristic constant of the material, between
these electrodes. The necessary separation d may then simply be calculated as
d D V/Eb. Although the electrodes are usually well defined and are limited

in size, the experienced designer will be able to take care of the entire field
distribution between the electrodes and will realize that in many cases only a
small portion of the material is stressed to a particular maximum value Emax.
One may conclude that the condition Emax D Eb would provide the optimal
solution for the insulation problem, which thus could be solved merely by field
analysis. This is true only when Eb has a very specific value directly related
to the actual field distribution and can be calculated for very well-known
insulating materials, such as gases (see Chapter 5, section 5.8). However, for
most solid and liquid dielectrics such values are only approximately known.
Hence a special approach is necessary to solve the insulation problem with
fair accuracy.
Plane
d
x
D
E (x)
Emax
Diameter
Voltage V
Figure 4.1 Rod-to-plane electrode configuration (with different field
efficiency factor D V/dEmax
These statements will be elucidated and confirmed by considering the
simple example of an insulation system shown in Fig. 4.1, which represents a
rod–plane electrode configuration insulated by atmospheric air at atmospheric
pressure. Whereas the gap length and the air density are assumed to remain
constant, the diameter D of the hemispherical-shaped rod will change over a
very wide range as indicated by the dashed lines. Two field quantities may
be defined for rods of any diameter D. These are the maximum field strength
Emax at the rod tip and the mean value of the field strength Emean D V/d. With

Electrostatic fields and field stress control 203
these two quantities a ‘field efficiency factor’ is defined as
D
Emean
Emax
D
V
dEmax
4.1
originally proposed by Schwaiger.7
This factor is clearly a pure quantity
related to electrostatic field analysis only. In a more complex electrode arrange-
ment Emax may appear at any point on an electrode, not necessarily coinciding
with the points providing the shortest gap distance, d. equals unity or 100
per cent for a uniform field, and it approaches zero for an electrode with an
edge of zero radius.
If the breakdown of the gap is caused by Emax only, then the breakdown
voltage Vb is obtained from eqn (4.1) as
Vb D Emaxd D Ebd (with Emax D Eb. 4.2
This equation illustrates the concept of the field efficiency factor. As 1 ½ ½ 0
for any field distribution, it is obvious that field non-uniformities reduce the
breakdown voltage.
Let us now check the validity of eqn (4.2) with experimental results. In
Fig. 4.2 the d.c. breakdown voltage Vb is shown for the electrode arrangement
of Fig. 4.1 for d D 10 cm as function of . The dashed straight line corre-
sponds to eqn (4.2) with Eb D 26.6 kV/cm, a value which agrees well with
measured breakdown field intensities in atmospheric air under normal condi-
tions (temperature 20°C; pressure 101.3 kPa; humidity 11 g/m3
) for a uniform
field D 1. The highest breakdown voltage of the gap Vb D 26.6 ð 10 D
266 kV can also be found in Chapter 5, eqn (5.103), or in the calibration
tables for measuring sphere gaps discussed in Chapter 3, Table 3.3, for spheres
of large diameters, i.e. D ½ 100 cm. With small gaps the field distribution
is uniform in the highly stressed regions. The measured breakdown volt-
ages, obtained with positive and negative d.c. voltages, are also shown over
wide ranges of or D, the correlation of which can be computed approx-
imately using eqn (4.20), or more accurately by a numerical computation
for this special rod-plane system using the charge simulation method.7
The
differences are remarkable. The lowest measured Vb values are polarity depen-
dent; the reason for the dependence of breakdown voltage upon polarities is
explained in Chapter 5, section 5.12. Except when D 100 per cent, the break-
down voltages are always higher than those predicted by eqn (4.2). For 0.3
for negative and about 0.1 for positive polarity, the breakdown is not preceded
by any noticeable predischarge phenomenon (corona, partial discharge; see
Chapter 5); thus it is obvious that Eb in eqn (4.2) is not a constant value for a
given gap length. A calculation of breakdown field strength in atmospheric air
using the streamer breakdown criterion (see eqn (5.90)) and the relevant field
distribution within the gap would confirm the dependence of the breakdown

0 10 25 50
0
0
100
200
300
20 40 60 80 100
100 250 500 ∞
Breakdown, neg. pol.
Breakdown, pos. pol.
Corona - inception,
pos. polarity
Pos. d.c.
Neg. d.c.
Neg. + pos. pol.
Eqn (4.2)
266 kV
η (%)
D (mm)
Voltage
(kV)
Figure 4.2 Breakdown and corona inception voltage for the electrode
arrangement of Fig. 4.1 in atmospheric air (normal conditions) with
d D 10 cm, for positive and negative d.c. voltage ( see eqn (4.1))
strength Eb upon rod or sphere diameter D or – more accurately – upon the
actual field distribution. In reality, the lowest breakdown voltage is not reached
with the smallest values of . Below the minimum breakdown voltages, the
sparkover of the gap is influenced by predischarges, which, for lower voltages,
partially bridge the gap and thus produce charged particles, completely altering
the field distribution due to space charges. Computation of the breakdown volt-
ages in this region based upon physical parameters only is inaccurate due to a
lack of precise knowledge of the physical data and complications introduced
due to the moving space charge.
This example, which is typical for most insulation media, demonstrates the
complexity of the problems, i.e. the interaction between the static field distri-
bution, field changes due to discharge development, and parameters related
to the insulation materials. Further complications arise from differences in

behaviour with direct, alternating and impulse voltages. For any other mate-
rial, the results would be different, even for the same electrode configuration.
The proper design of insulation systems is therefore very difficult. Neverthe-
less, the maximum field intensity Emax within any insulation system may be
considered as a significant quantity, even though it only serves as a guide.
In practice, data on the dielectric stresses in the insulation materials used in
h.v. equipment obtained by field analysis must be supplemented by extensive
tests in which the breakdown stresses are experimentally determined for similar
insulation arrangements. Computations of the stresses are most advanced in
gaseous dielectrics. Tests necessary for most of the other materials need not,
however, involve complete experimental models which precisely simulate the
actual equipment. In general, breakdown stresses are dependent upon the
field distribution within high field regions, as will be shown in Chapter 5
for gaseous dielectrics. Thus, models representing only those regions in which
high stresses occur are, in general, sufficient; this offers definite advantages.
Apart from saving time and costs by simplifying the experimental insulation
assemblies, the required voltage levels may often also be reduced significantly,
as the models can be reduced in size using electrode configurations in which
the low field regions are absent.
4.2 Fields in homogeneous, isotropic materials
Many electrical insulation systems contain only one type of dielectric mate-
rial. Most materials may be considered to be isotropic, i.e. the electric field
vector E and the displacement D are parallel. At least on the macroscopic scale
many materials at uniform temperature may also be assumed to be homoge-
neous. The homogeneity is well confirmed in insulating gases and purified
liquids. Solid dielectrics are often composed of large molecular structures
forming crystalline and amorphous regions so that the homogeneity of the
electrical material properties may not be assured within microscopic struc-
tures. The materials will also be assumed to be linear; that means the electric
susceptibility – see Chapter 7 – is not a function of electric field strength. On
a macroscopic basis, the permittivity ε will then simply be a scalar quantity
correlating D and E, with D D εE.
At this stage it is assumed here that the influence of electrical conductivity
on the field distribution may be ignored; this is justified for most insulating
materials when they are stressed by alternating voltages at frequencies above
about 1 Hz. Thus, simple electrostatic field theory may be applied to most of
the practical applications concerned with power frequency or impulse voltages.
With direct or slowly alternating voltages the use of simple electrostatic field
theory is greatly impeded by conduction phenomena. In the limiting case, the
field is purely given by conduction and the correlation between field strength

E and current density j is j D E, where (or the complex permittivity, see
Chapter 7) may be highly dependent upon time due to relaxation phenomena,
upon temperature and often also upon field intensity. This problem is only
mentioned here to emphasize the difficulties encountered with d.c. voltage
applications.
The following examples for electrostatic field distributions are typical for
h.v. insulation systems.
4.2.1 The uniform field electrode arrangement
The realization of homogeneous fields within a finite volume of insulating
material is very difficult. Using parallel metal plates of limited dimensions
creates the problem of a proper stress control at the edges of the plates. The
field problem becomes thus three dimensional, although a rotational symmetry
exists if the parallel plates are circular discs.
Depending upon the material to be tested, the breakdown strength may be
very sensitive to local high fields within the whole electrode arrangement.
Therefore, the highest stress should only be present in the homogeneous field
region, where the plates are in parallel. A certain profile of electrodes is
necessary outside the plane region to limit the dimensions, but the field strength
at the curved edges should never exceed the value E D V/d, if V is the applied
voltage and d the distance between the parallel plates. Rogowski6
proposed
electrodes for uniform fields for axially symmetrical systems whose profile
follows the analytical function first introduced by Maxwell,
z D
a

w C 1 C ew
4.3
where z and w represent the complex coordinates in the z- and w-planes.
Substitution of the coordinates for the complex values z D x C iy and w D
u C iv and separation of the real and imaginary parts gives
x D
a

u C 1 C eu
cos ;
y D
a

v C eu
sin . 4.4
Assuming two infinite, parallel ‘plates’ in the w-plane, the coordinates of
which are given by v D š D const, it can be recognized from eqn (4.4) that
these plates are transformed into the z-plane to the left half-plane only. All
other lines v D const with v C can be assumed to be other equipo-
tential lines, and all lines u D const with 1 u C1 can be assumed to
be field lines in the w-plane, representing a uniform field distribution. These
lines appear in the z-plane as shown in Fig. 4.3, providing the electrical field
distribution of parallel plates terminating at x D 0. The concentration of the

a
a
υ=π
u
=0
u
=−2π
u
=−π
0.9π
0.8π
0.7π
0.6π
u
=0.6π
0.5π
0.5π
0.4π
0.4π
0.3π
0.2π
0.1π
0
.
3
π
0.2π
υ=
0.1π
0.1π
y
υ=0
x
n = −π
n = −0.5π
Figure 4.3 Transformation of a square grid from a w-plane in the displayed
z-plane by eqn (4.3): Rogowski’s profile (š2)
equipotential lines, v D const, within the z-plane may well be recognized at,
or in the vicinity of, the edges of the plates.
The parallel plates, v D š, are thus inadequate to fulfil the demand for field
distribution whose intensity is limited to the field strength within the homo-
geneous part of the arrangement, i.e. for u . It is obvious that the field
strength along equipotential lines for which v C provides better
conditions. For quantitative assessment the field strength within the z-plane
may be computed in several ways, as follows.
From the conjugate complex field strength in the z-plane
EŁ
z D Ex jEy D j
dw
dz
D j
1
dz/dw
4.5
the absolute values could be computed by

EŁ
z

D

E2
x C E2
y.
A second possibility is given by
Ez D Ex C jEy D grad D

∂
∂x

C j

∂
υy

4.6
which needs a partial differentiation only.

Finally, the absolute value of Ez may be computed by
jEzj D
1

∂x
∂
2
C

∂y
∂
2
4.7
a method which is easiest to apply to our separated analytical function,
eqn (4.4). Combining eqns (4.4) and (4.7), we may easily find the field strength
as
jEzj D

a

1 C e2u C 2eu cos v
D fu; . 4.8
To quantify this expression with any applied voltage it is necessary to perform
a calibration with the field intensity within the original w-plane. If the line
v D is at potential D V and the line v D at potential D V, the
magnitude of the field strength in the w-plane is jEwj D 2V/2 D V/. Hence,
the absolute magnitude in the z-plane becomes jEwj jEzj or
jEzj D
V
a

1 C e2u C 2eu cos
4.9
For u
' 3 5, jEzj is practically constant and equals V/a, but for u D 0 and
v D š i.e. at the edges of the plates, jEzj increases to infinite values. There
are, however, many equipotential lines in the z-plane for which jEzj is always
limited to values
'V/a. The general condition for this behaviour is given by
cos
' 0 or within š/2. As the strongest curvature of an equipotential
line will provide the smallest possible electrode arrangement, Rogowski has
chosen the profile cos D 0 or D š/2, the so-called 90° Rogowski profile,
which is marked by a heavier solid line in Fig. 4.3. Along this line the field
strength has its maximum values between the plates in the ‘homogeneous
field region’ u
' 3 5 and decreases gradually within the curvature with
increasing values u. As for all field lines starting at the curved part, the field
strength decreases to a minimum value for v D 0, a breakdown should not
occur between the curved regions of the electrodes. The actual distance of
two metal electrodes shaped in this way would be d D a, and eqns (4.4) and
(4.9) indicate the necessity of dimensioning the electrodes in accordance to
the maximum gap length d D a, necessary for breakdown tests. For smaller
gap lengths and the same profile, the field strength at the curved profile will
decrease relative to the homogeneous field region. Disc-shaped electrodes
would have the rotational centre at a field line for u ¾
D 5 or less providing
any size or volume of a homogeneous field region desired. The rotation of the
profile about the rotational centre converts the field to the third dimension.
The additional increase of the field strength components in the x-direction by
this additional curvature is, however, in general negligible. Machining of such
profiles has to be carried out very carefully. A very efficient test can be made to

demonstrate the performance of the electrodes: breakdown tests in pressurized
sulphur hexafluoride (SF6), a gas very sensitive to local field enhancements,
must display all sparking events in the plane centre of the electrodes.
The decrease of field intensity at the outer curvature of the Rogowski profile
could be prevented by a decrease of the radius of curvature, providing smaller
dimensions or diameters of the disc electrodes. Profiles approaching constant
field intensities at the electrode surface with magnitudes V/d also outside
of the uniform field regions are, for instance, Bruce’s profile5
and Borda’s
profile.6
Borda’s profiles give a completely constant field intensity along the
electrode surface, but as they are also based on a two-dimensional calculation,
the uniformity will disappear if this profile is applied to an axisymmetric
electrode. Improvements can be made by very accurate numerical, computer-
aided field calculations, taking the actual surroundings as additional boundary
conditions into account. For Borda’s profile, such optimization was already
performed by Okubo et al.13
4.2.2 Coaxial cylindrical and spherical fields
Electrode configurations providing two-dimensional cylindrical or three-
dimensional spherical fields are used in h.v. equipment as well as in
laboratories for fundamental research or field stress control. In a short
treatment of the well-known coaxial arrangements, we shall demonstrate the
fundamental differences only; some special cases give useful comparison.
Cross-sections of coaxial cylinders and concentric spheres are sketched in
Figs 4.4(a) and (b), and different notations are used to distinguish between the
radii of cylinders r1, r2 and spheres R1, R2. The electrical field distribution
is symmetrical with reference to the centre of the cylinder axis or the centre
point of the sphere. In both cases the lines of force are radial and the field
strength E is only a function of the distance x from the centres. The cylinders
are then uniformly charged over their surface with a charge per unit length
Q/l, and the spheres with a charge Q, if a voltage V is applied to the two
electrodes. Using Gauss’s law, the field strength Ex at x is derived from the
following:
ž Coaxial cylinder:
Ex D
Q/l
2ε
1
x
D
V
lnr2/r1
1
x
, 4.10
ž Coaxial spheres:
Ex D
Q
4ε
1
x2
D
V
R2 R1/R1R2
1
x2
, 4.11

r2
R2
r1
R1
Emax
Emax
x
x
E(x)
E(x)
(a) (b)
Figure 4.4 Coaxial cylinders (a) and concentric spheres (b)
where the subscripts 1 and 2 refer to inner and outer radii respectively. The
main difference between the two field distributions is the much faster decrease
of the field strength with distance x in the three-dimensional case.
Therefore, for equal geometries (r1 D R1; r2 D R2) Emax will always be
higher in the sphere configuration. As Emax is reached for x D r1 or x D R1
respectively, we obtain for:
ž Coaxial cylinders:
Emax D
V
r1 lnr2/r1
. 4.12
ž Coaxial spheres:
Emax D
V
R11 R1/R2
. 4.13
Note that the denominator in eqn (4.12) will always be larger than that in
eqn (4.13), confirming the statement made above.
Let us consider a few simple examples. Spheres or sphere-like electrodes
are often used as terminating electrodes of h.v. equipment, placed at the top
of a bushing or a voltage divider, etc. Neglecting the influence of the structure
connected to the sphere, we may roughly estimate its necessary diameter 2R1
assuming the ground potential is far away, i.e. R2/R1 × 1 in eqn (4.13). There-
fore, Emax ³ V/R1. Theoretically, atmospheric air insulation would provide a
breakdown strength for large sphere diameters of about 25 kV/cm under normal
conditions. Irregularities involved in the production of large electrodes and
unavoidable dust particles in the air (see Chapter 3, section 3.1.1) will reduce

the permissible breakdown field strength to about Eb D 12–15 kV/cm. There-
fore, the diameters necessary to avoid discharge inception, or even breakdown,
will be
2R1 ³ 2Vpeak/Eb. 4.14
For an a.c. voltage of 1 MV (r.m.s. value) diameters of about 1.9 to 2.4 m are
acceptable. In this case, the greatest uncertainty is related to the breakdown
strength Eb at the electrode surface used, i.e. the surface irregularities of the
electrodes.15
A cylindrical conductor used for partial discharge-free connections in h.v.
test circuits in laboratories is always limited in length, and no discharges
should occur at the end of the cylinder. Obviously, a sphere of larger diameter
than that for the cylindrical conductor must be located at the end, as shown in
Fig. 4.5(a). The earthed walls of the laboratory will form the outer diameters
of the sphere and the cylinder, and we may approximately assume that the
field distributions at both electrodes are independent upon each other. Equal
maximum values Emax are then achieved by setting eqns (4.12) and (4.13)
equal. Thus the
R1
r1
D
lnr2/r1
1 R1/R2
¾
D lnr2/r1 4.15
condition displays the necessary ratio of the diameters. As the ‘radius’ r2 or
R2 of the laboratory may well be assumed to be twenty times the radii of the
electrodes, this ratio becomes at least 3. For small diameters, the breakdown
field strength of gases is not equal for even the same radii, as the increase of
Eb is larger with decreasing radii for spherical fields. Exact values, therefore,
can only be obtained by exact field computations and taking the properties of
the insulation medium into account.
Busbars for SF6-insulated, metal-enclosed equipment (GIS) are typical
coaxial cylindrical arrangements. If the busbar must change the direction,
a ‘knee’ or elbow will be necessary, as shown in Fig. 4.5(b). This problem
can approximately be solved by an interconnection of a coaxial sphere with
coaxial cylinder configurations, if the edges at the earthed conductors arising at
the intersections are adequately rounded. All dimensions are now interlinked,
and as a starting point it will first be necessary to demonstrate optimum
dimensioning. For every coaxial or concentric system there is an optimum ratio
of the radii, as the field stresses Emax reach high values for a given voltage V
for small ratios of the radii as well as with too small dimensions. For coaxial
cylinders we may rewrite eqn (4.12) as
V D Emaxr2

r1
r2

ln

r2
r1

4.16

E max (cylinder)
E max (sphere)
2r1
r1 r2
2 R1
R1
R2
V
V
(a)
(b)
r2, R2 r1;R2
Figure 4.5 Typical ‘coaxial’ arrangements. (a) Cylinder ended by a sphere
within a laboratory. (b) Busbar arrangement in GIS
and search for an optimum ratio r2/r1, for which the highest voltage can
be insulated with a given breakdown strength Eb D Emax of the insulation
material. For not too small diameters we may well neglect the fact that Eb
depends upon r1 for all gases or other insulation materials (see Chapter 5,
section 5.9, eqn (5.111)). Thus Emax D Eb as well as r2 can be treated to
be constant and the differentiation of eqn (4.16) with respect to r1 gives the
condition dV/dr D 0 for
lnr2/r1 D 1; r2/r1opt D e ³ 2.72; 4.17
Vbopt D Ebr1.
This ratio is obviously a very important one in dimensioning h.v. cables or
coaxial conductors insulated by homogeneous materials of any permittivity.
The field efficiency factor defined by eqn (4.1) in coaxial cylindrical system

is, according to eqn (4.12),
cy1 D
1

r2
r1
1
ln

r2
r1

. 4.18
For r2/r1 D e, this efficiency factor becomes 58 per cent, and is therefore
quite high. Highest breakdown voltages can actually be reached with ratios
of r2/r1 very close to the optimum value, which is demonstrated in Fig. 4.6
for SF6-insulated cylindrical conductors within the most interesting range of
r1/r2. For small ratios, i.e. for small diameters of the inner conductor, no direct
0
1 2 4 6 8 2
1
6
1.5
2
2.5
p = 0.5 bar
4 6 8 100
10−1
50
100
150
200
250
V
i
,
V
b
(kV
peak)
r1/r2
Vi
Vb
3
4
r2 = 31
2r1
AC
1/e
Figure 4.6 Breakdown (Vb) and discharge inception (Vi) voltages in a
coaxial cylindrical system with SF6 insulation, in dependence of the ratio
r1 /r2 . Parameter: gas pressure p. Temperature: 20°C (see reference 16)

breakdown will occur for Emax D Eb; similar to Fig. 4.2, the actual breakdown
voltage is increased by corona discharges.
For the concentric sphere arrangement, the same statements will be appli-
cable. The optimum values for breakdown can be derived from eqn (4.13),
resulting in:
R2/R1 D 2; Vbopt D EbR1/2. 4.19
The field efficiency factor becomes in general terms
sphere D R1/R2 4.20
and thus is only slightly smaller for R2/R1opt than that found for the coaxial
cylinders with optimum conditions.
Now we may solve the example in Fig. 4.5(b). If the busbar is optimally
designed, i.e. r2 D r1e, and r1 was calculated by eqn (4.12) for given values
of breakdown voltage V D Vb and breakdown field strength Emax D Eb, one
may apply equal breakdown conditions for the concentric sphere arrangement.
Equating the values Vb/Eb for the two systems, we obtain
R11 R1/R2 D r1 lnr2/r1 D r1,
a condition which obviously has many solutions depending upon the magni-
tude of R2. We may, however, select the optimum ratio R2/R1 for spheres,
and thus we obtain the conditions
R1 D 2r1, R2 D r24/e
and accordingly the different gap distances related to r1:
R2 R1 D

4
e
r2
r1
2

r1 D 2r1,
r2 r1 D

r2
r1
1

r1
¾
D 1.72r1.
These conditions are quite favourable in practice, as the outer sphere diam-
eter is not much bigger than that of the cylindrical system. The gap distance
R2 R1, however, is larger than r2 r1, which could be expected by the
more inhomogeneous field distribution within the three-dimensional field of
the sphere arrangement.
4.2.3 Sphere-to-sphere or sphere-to-plane
In practice, the sphere-to-sphere arrangement is used for measuring high volt-
ages with sphere gaps (Chapter 3, section 3.1.1); sphere-to-plane gaps are
widely used for fundamental breakdown studies. The field distribution can be

computed analytically if the spheres are assumed to become charged to their
potential without any connecting leads. The influence of connecting leads
upon the field distribution was recently investigated with a charge simulation
program by Steinbigler.7
The analytical results are presented here based upon
the method of image charges.1
Another possible solution based upon bipolar
coordinates can be found in the literature.17
+V −V
V = 0
R
R
S
P
b
b/2
M M′
Q0
x1
x2
x3
x1
x2
Emax
Q1 Q2 −Q2 −Q1 −Q0
1 2
Figure 4.7 Sphere-to-sphere or sphere-to-plane arrangement
In Fig. 4.7 two spheres of equal diameter 2R separated by distance b
between centres are assumed to have the potential CV and V respectively.
Then – and only then – the field distribution is completely symmetrical with
reference to an imaginary plane P placed between the two spheres, if the plane
has zero potential. Zero potential also exists at distances far away from the
spheres. With a point charge Q0 D 4ε0RV at the centre of the left sphere (1)
the surface of this sphere would exactly represent an equipotential surface and
could be replaced by a metal conductor, if the right sphere (2) and the plane
were not present. A charge Q0 placed at the centre of sphere (2) will produce
a symmetrical field distribution with reference to the plane P, but this charge
makes the potentials at the surface of the imaginary spheres non-equipotential.
An improvement of these potentials is possible by placing additional image
charges CQ1 and Q1 in both spheres at a certain distance from their centres.
This statement is confirmed by a well-known solution of the electrostatic field
problem concerning a point charge in the vicinity of a conducting sphere by
the image charge technique. A point charge Q and a smaller point charge Q0
distant from Q and opposite in polarity are producing a field distribution, for
which the zero equipotential surface is formed by a sphere. This sphere does

include Q0
, but not at its centre. The amount of the charge Q0
with reference
to Q and the distance from the centre of the imaginary sphere may easily be
computed by consideration of boundary conditions. Applying this principle
to our problem, one may treat the disturbing charge Q0 of sphere (2) such
as the above-mentioned point charge outside of the system (1) and find the
necessary image charge CQ, within this sphere by
jQ1j D Q0
R
b
D 4ε0RV
R
b
placed at a distance
x1 D
R2
b
from the centre. The charges CQ0 and CQ1 inside of sphere (1) and the
charge Q0 outside would make the surface of sphere (1) precisely equipo-
tential to CV; but there is also a charge Q1 within sphere (2) necessary to
gain symmetry with reference to P, and this charge again disturbs the equipo-
tential character of both sphere surfaces. To compensate for these charges,
further image charges CQ2 inside sphere (1) and Q2 inside sphere (2) with
magnitudes
jQ2j D Q1
R
b x1
D 4ε0RV
R
b x1
Ð
R
b
at distances x2 D R2
/b x1 from their centres must be added, and this
process must be continued indefinitely to reach precisely equipotential sphere
surfaces. The potentials or field intensities between the two spheres could
now be computed with the knowledge of the charge intensities and their posi-
tion with reference to the sphere centres. The most interesting quantity is the
field strength along a field line of highest field intensity, which is obviously
within the shortest distance M–M0
of both spheres. As the potentials at any
distance r from a point charge are proportional to 1/r and the field strength
Er D grad thus proportional to 1/r2
, the total field intensity is equal
to the sum of the single intensities of all image point charges inside of both
spheres. The maximum field strength at the points M and M0
is, therefore,
given by
ER D Emax D
1
4ε0
1
nD0
Qn
R xn2
C
1
nD0
Qn
b R xn2
D RV
1
nD0





n

kD1



R
b xk1
k D 1









1
R xn2
C
1
b R xn2

, 4.21

where
Qn D Qn1
R
b xn1
D 4ε0RV
n

kD1

R
b xk1

;
Xn D
R2
b xn1
; with n D 1, 2, 3 . . .
x0 D 0
The same expression can be used to compute the field intensity at any point on
the line M–M0
, if the R values in the expressions R xn2
and b R xn2
are replaced by a distance x measured from the centre of the sphere (1) and
the point considered between M and the plane P, i.e. R − x − b/2.
The capacitance between the two spheres can be calculated according to
Gauss’s law, as the real total charge on metal spheres replacing the imaginary
spheres is equal to the sum of all charges Qn:
C D
1
nD0
Qn
2V
D 2ε0R
1
nD0
n

kD1

R
b xk1

,
where again x0 D 0.
Numerical evaluation of eqn (4.21) for different b/R ratios displays the
following approximation for the maximum field strength Emax, if S R:
Emax
¾
D 0.9
V
S/2
R C S/2
R
4.22
where S D b 2R is equal to the distance M–M0
, and V equals the potentials
as defined in Fig. 4.4, i.e. half the voltage across the two spheres.
For a sphere-to-plane arrangement, the same equation can be used, if S/2
is then equal to the gap distance and V identical to the voltage applied.
As mentioned before, eqn (4.21) may be applied to compute the field inten-
sities between oppositely charged metal spheres along a field line of highest
field strength, i.e. between the shortest distance M–M0
. Numerical examples
for the evaluation of this equation are shown in Fig. 4.8 for different values of
S/R to demonstrate the increasing non-uniformity of the electrostatic field with
increasing S/R ratios. The field strength values are normalized with reference
to the mean values Emean according to eqn (4.1); by this the field efficiency
factor may directly be computed from the maximum values of the field
intensity. This ‘isolated’ sphere-to-sphere arrangement is only an approxima-
tion of actual electrode arrangements, i.e. sphere gaps for the measurement of
the peak values of high voltages (see Chapter 3, section 3.1.1).

5.82
4.67
3.63
2.66
1.77
1.36
1.0
E
/
E
mean
S
/
R
10
8
6
4
2
1
0 20.00 40.00
Distance M − M′ (%)
60.00 80.00 100.00
Figure 4.8 Electric stress distribution along the axis M–M0
of the
sphere-to-sphere arrangement (Fig. 4.7) for various S/R ratios
4.2.4 Two cylindrical conductors in parallel
We choose this electrode configuration for comparison with the field distri-
bution between two oppositely charged spheres as treated above. If two or
more cylindrical conductors would be at the same potential with reference
to predominantly earth potential far away from the parallel conductors, the
configuration of so-called ‘bundle conductors’ is formed, a system extensively
applied in h.v. transmission lines. Due to the interaction of the single conduc-
tors the maximum field intensity at the conductors is reduced in comparison
to a single cylindrical conductor, so that the corona inception voltage can
significantly be increased. Solutions of the field distributions for such bundle
conductors are possible by the complex variable technique, i.e. conformal
mapping.6
For our comparison, we have to charge the two cylindrical conductors with
opposite polarity to each other. Thus the field distribution can be calculated by
assuming only two line charges šQ/l D š.l running in parallel and eccen-
trically placed within the conductors. This statement is confirmed by a short
calculation based upon Fig. 4.9, in which the two line charges š.l are spaced
by a distance b. At any point P within the plane the potential p may be
found by the principle of superposition. As the field intensity of an individual
line charge is Er D ./2εr with r being the distance from the charge, the
potentials may be found by integration. Superposition leads to
p D
.l
2ε
ln
r00
r0 C K 4.23

b
P
−rl
+rl
r ′′
r′
Plane of symmetry
Figure 4.9 Two line charges š.l in parallel
when r0
and r00
are defined in Fig. 4.9 and K is a constant found from boundary
conditions. For equal line charges of opposite polarity and the potential zero
at infinite distances, there is also zero potential, i.e. p D 0 at the plane of
symmetry, r0
D r00
. Thus K D 0 for this special case of equal charges.
For all other ratios r00
/r0
D const, also p is constant and may lead to any
positive or negative potentials. However, all constant ratios of r00
/r0
generate
cylindrical surfaces. These surfaces T may be assumed to be cylindrical
conductors of different diameters.
Interested in two conductors of equal diameters, the two line charges will be
eccentrically but symmetrically placed within these two conductors as shown
in Fig. 4.10. The eccentric position, indicated by the distance c between the
Plane of symmetry
f = V = 0
+V/2 −V/2
P1
A
B C
D
M1 M2
r1 =r r2 =r
r′
1
r ′
2
r2
r1
′′
y
+rl −rl
S
b c
c
a
Figure 4.10 Two equal cylindrical conductors in parallel, symmetrically
charged, Fig. 4.7, and the parallel cylindrical conductors, Fig. 4.10, for
equal voltages applied

line charges and the centres M of the conductors, can easily be found for
constant ratios r00
1 /r0
1 and r00
2 /r0
2 for the points P1 or P2 positioned at A,B or
C,D. Omitting this simple calculation, we find for equal radii r1 D r2 D r
c D

b
2
2
C r2
b
2
D
a
2

a
2
2
r2;
a D

b2 C 2r2 4.24
The distance c D a b/2 becomes for r − a/2 very small which demon-
strates that for larger gaps the fields in the vicinity of the conductor surface
will not be much disturbed in comparison to single conductors. For thinner
conductors, we may calculate the field distribution along the flux line for the
highest density, i.e. between B and C, where the field strength is highest. The
potential y along this line starting at By D 0 is provided by eqn (4.23), as
y D A ln

r00
r0

D A ln





b C S
2

y

b S
2
C y





where A is a constant given by boundary conditions and S is the gap distance.
Assuming a total potential difference or voltage of V between the two conduc-
tors, A is given by y D CV/2 for y D 0 and thus
A D
V/2
ln

b C S
b S

The field strength Ey becomes therefore
Ey D
dy
dy
D A




1

b C S
2

y
C
1

b S
2

C y




D
V
2
b
%
b
2
2

y
S
2
2

ln

b C S
b S
4.25

The field distribution is symmetrical to y D S/2. For convenience, the distance
b D fa, r might be expressed by the gap distance also. Then
Ey D
V
S

S
2r
2
C

S
r

%
1 C
y
r

y2
rS

ln

1 C
S
2r
C

S
2r
2
C
S
r


. 4.26
The field distribution between two conductors can easiest be discussed by
relating eqn (4.26) with the maximum field intensity Emax for y D 0. This
ratio becomes
Ey
Emax
D
1
1 C
y
r

y2
rS
D
r
r C y

1
y
S

In comparison to a single charged cylindrical conductor, for which this
field strength ratio would be given by r/r C y only – see eqn (4.10) – it
is obvious that for all values y/S − 1 the parallel conductor is of diminishing
influence. As the minimum value of E is reached for y D S/2, the ratio
Emin/Emax becomes
Emin
Emax
D
1
1 C S/4r
A comparison of the field distributions between the sphere-to-sphere gap and
the parallel cylindrical conductors is plotted in Fig. 4.11. Again we can recog-
nize that the cylindrical fields are more uniform for the same ratios of gap
distance and radii.
4.2.5 Field distortions by conducting particles
Up to now we have treated ‘macroscopic’ fields acting between conducting
electrodes with dimensions suitable to insulate high voltages by controlling
the maximum electrical field strength by large curvatures of the electrodes.
In actual insulation systems the real surface of any conductor may not be
really plane or shaped as assumed by macroscopic dimensions, or the real
homogeneous insulation material may be contaminated by particles of a more
or less conducting nature. Although a real surface roughness of an electrode,
or the real shape of particles within the insulating material, may be very
complex, the local distortion of the electrical field which can be assumed to
be ‘microscopic’ in dimensions can easily lead to partial discharges or even
to a breakdown of the whole insulation system.

0 20.00 40.00 60.00 80.00 100.00
y/S (%)
1.00
1.16
1.32
1.60
1.88
2.14
2.39
E/E
mean
1
2
4
6
8
10
S
/
r
Figure 4.11 Field strength distribution between two cylindrical conductors
of equal radii r, for different ratios S/r, with S D gap distance (see
Fig. 4.10). (Compare with Fig. 4.8)
To account for such phenomena, two results of field distributions produced
by spheroidal conducting particles are shown. The results are based upon one
of the most powerful methods for solving Laplace’s equation, the method of
separation of variables, extensively treated in the book of Moon and Spencer.3
The first example is related to prolate spheroids formed within a prolate
spheroidal coordinate system (, , ) shown in Fig. 4.12, which is related
to rectangular coordinates by the equations
x D a sinh sin  cos ;
y D a sinh sin  sin ;
z D a cosh cos . 4.27
The prolate spheroids are surfaces of constant values, for which
x
b
2
C
y
b
2
C
z
c
2
D 1, 4.28
where b D a sinh ; c D a cosh . The variable may be changed from 0 to
C1. For ! 1, sinh ¾
D cosh , and thus b ¾
D c, i.e. the spheroid becomes
a sphere. For ! 0, the spheroid approaches a straight line segment of length
2a on the z-axis, as z D a for  D 0. Due to the rotational symmetry with
reference to the z-axis, the cross-sections of the spheroid for constant z value
planes are circles. The surfaces of constant  values are hyperboloids, and for
the special case of  D /2 the hyperboloid becomes the x–y-plane. is the

h = p y = const.
h = const.
q = const.
q = 0
F
F
O
y
y = π/2
y = 0
a
x
y
z
a
Figure 4.12 Prolate spheroidal coordinates. The coordinate surfaces are
prolate spheroids, D const, hyperboloids, 5 D const, and meridian planes,
D const (see reference 3, p. 237)
angle measured about the z-axis, and the range of is taken as 0 2.
Surfaces of constant are half-planes containing the z-axis.
The solution of Laplace’s equation for this coordinate system is treated in
reference 3. The results depend upon the boundary conditions, i.e. assuming
scalar potentials for constant values D 0, as well as for distances far away
from the centre, ! 1, i.e. for a sphere of infinite large diameter. The lengthy
calculations are not shown here, but it may well be recognized that two special
cases are of interest. A charged spheroid of potential D V for D 0 with
a reference potential D 0 far away from the spheroid, and a spheroid within
an otherwise uniform field E D E0 D constant. In both cases the field strength
Ez along a flux line in the z-direction (x D y D 0) is of main interest; no
simple analytical expressions, however, can be achieved as Legendre functions
are involved in the solutions. Therefore, only some field distributions and
maximum potential gradients are reproduced from computations.3

E max
E max
f = V
2b
O
a
a cosh h0
Figure 4.13 Electrostatic field about the spheroid of Fig. 4.12 with D 0
D V (see reference 3, p. 245)
In Fig. 4.13 the electrostatic field about a charged spheroid with potential
D V within free space ( D 0 for ! 1) displays the field enhancement
along the z-axis for a ratio of b/a ¾
D 0.436. The maximum field strength
Emax will heavily increase with decreasing ratios b/a, as the curvature at
this point increases. The numerical evaluation of Emax is shown in Fig. 4.14.
Slim spheroids may be assumed to simulate capped wires whose length is
large in comparison to its diameter.
Of more importance is the second case, for which a spheroid of either high
permittivity equivalent to a conducting spheroid or a real metal particle placed
within a dielectric material in which an originally constant uniform field E0
was present. A field map is shown in Fig. 4.15. The potential D 0 being
present not only at the surface of the spheroid, but also for all values z D 0, i.e.
a plane in the xy-direction simulates also a macroscopic plate-to-plate electrode
arrangement, which would produce a uniform field. If a protrusion is present
at the plates, whose shape is identical with half of the spheroid, the field is
distorted heavily in the vicinity of this protrusion only. The map indicates that
the large distortion is limited to dimensions about equivalent to the dimensions
of the protrusion only, a region of the field which can be named ‘microscopic’.

0 0.5 1.0
1
2
3
4
5
6
7
8
9
10
20
30
40
50
60
70
80
90
100
aE
max
/
V
Sphere
b/a
Figure 4.14 Maximum potential gradient near a charged metal spheroid as
affected by the shape of the spheroid. Major semi-axes a, minor semi-axes b,
potential of spheroid V. The ordinate (aEmax/V) approaches infinity as
b/a ! 0 and falls to a value of unity for a sphere (see reference 3, p. 246)
Again, for different shapes of the spheroid the maximum values Emax can be
calculated with reference to the uniform field strength E0, the result of which
is shown in Fig. 4.16. For b D a, i.e. a sphere, Emax/E0 equals to 3, the well-
known field enhancement factor for a half-sphere placed upon a plate electrode
within a parallel plane-to-plane arrangement. Again, for slender spheroids the
Emax values will increase to very high values, independently of the absolute
size of the spheroids. Such high Emax values are responsible for electron
emission at metal surfaces. Critical electron avalanches in gases, however, are
produced not only by this high value, but also from the field distribution in
the vicinity of Emax, so that the absolute values of the dimensions a and b
become significant also.
4.3 Fields in multidielectric, isotropic materials
Many actual h.v. insulation systems, e.g. a transformer insulation, are
composed of various insulation materials, whose permittivities ε are different

−100
−80
−60
−40
−20
0
20
40
60
80
f =
100
volts
E0
z
O
f = 0
Figure 4.15 Field distribution produced by a spheroid of high permittivity
(ε2 /ε1 ! 1) within a uniform electrostatic field, E0 (see reference 3, p. 257)
0 0.5 1.0
b/a
1
2
3
4
5
6
7
E
max
/
E
0
Sphere
Figure 4.16 Maximum potential gradient at a metal spheroid introduced
into a uniform electric field. Here a and b are semi-axes of the ellipse,
Fig. 4.13 (see reference 3, p. 258)

from each other. The main reasons for the application of such a multidielectric
system are often mechanical ones, as gaseous or fluid materials are not able to
support or separate the conductors. Layer arrangements may also be applied
to control electric stresses. The aim of this section is, therefore, to treat
fundamental phenomena for such systems.
Only a few examples have been chosen to demonstrate principally the
dangerous effects. Analytical methods for field computations in multidielec-
tric systems containing predetermined shapes of the electrodes as well as the
interfaces of the dielectrics are severely restricted. Adequate solutions are in
general only possible by numerical computations or experimental field plotting
techniques.
4.3.1 Simple configurations
Due to the effect of reduced electrical breakdown at the interface of two
different insulation materials, the interfaces in highly stressed field regions
should be normal to the field lines. The ‘parallel-plate capacitor’ containing
two layers of different materials represented by their permittivities ε1 and
ε2 is therefore typical for many applications. Figure 4.17 shows the arrange-
ment and the dimensions assumed. For usual dielectric materials and power-
frequency a.c. voltages, the conductivity of the materials can be neglected and
hence no free charges are built up at the interface between the two layers. The
displacement vectors D1 and D2 are then equal, starting from and ending at
the equal free charges on the plates only. As D D εE, and identical in both
materials, the ratio of the field strength becomes
E1
E2
D
ε2
ε1
4.29
V d
1 D1,E1 d1;V1
d2;V2
D2,E2
2
Figure 4.17 Parallel plate capacitors comprising two layers of different
materials
and as the field remains uniform in each layer, the voltage V or potential
difference between the two plates is
V D E1d1 C E2d2

where d1, d2 are the individual values of the thickness of the two dielectrics.
Introducing eqn (4.29) into this equation, we obtain the following absolute
values of E1 and E2 with reference to the voltage applied:
E1 D
V
ε1

d1
ε1
C
d2
ε2
D
V
d
ε2/ε1
d1
d

ε2
ε1
1

C 1
D
V
d1 C d2

ε1
ε2
4.30
E2 D
V
ε2

d1
ε1
C
d2
ε2
D
V
d
1
d1
d

ε2
ε1
1

C 1
4.31
This relationship demonstrates some essential effects:
(a) The partial replacement of a given dielectric material of ε1, for instance
a gas within a gap of uniform field, by a material of higher permittivity
ε2 decreases according to eqn (4.30) the ‘effective gap distance’ d0
D d1 C
d2ε1/ε2 defined by the unaltered field strength E1 in the original gap, as the
equivalent thickness of the layer ‘2’ becomes d2ε1/ε2 only. Alternatively, for
V1, d and ε1/ε2 remaining constant, the field stress E1 will always increase
if the thickness of the layer ‘2’ with higher permittivity is increased.
Although no distinct relationships exist between the permittivity of an insu-
lation material and its permissible breakdown field strength, gases with the
lowest values of ε very close to ε0, the permittivity of the free space, are in
general most sensitive to high field stresses, primarily if the gas pressure is
only equal to atmospheric pressure or even lower. Any partial replacement of
the gas with solid materials thus does not improve the dielectric strength of
an air or gas-insulated system, as the gas will now be even more stressed than
in the original system.
(b) The continuous increase of both field intensities E1 and E2 in the parallel
plate system with increasing thickness d2 for ε1 ε2 given by eqns (4.30)
and (4.31) can numerically be demonstrated in Fig. 4.18. The worst case is
displayed for conditions when d1 ! 0, i.e. for very thin layers of the low
permittivity material, as the field strength increases to a value ε2/ε1 times
the field in a system filled with one type of a material of any permittivity.
‘Sandwiched’ or multi-dielectric insulation systems can therefore be
dangerous if the layers are of very different permittivities. However, it is
also very difficult in h.v. insulation technology to avoid such or similar
arrangements due to production problems. Examples are the continuous
tight contact between metal electrodes and solid insulation materials, or
between insulation material interfaces. The remaining voids may then become
filled with gases, the breakdown strength of which may be calculated by
applying Paschen’s law, treated in Chapter 5, if the dimensions and discharge
parameters of the gases are known. Only for very thin gaseous layers may the
breakdown strength of the gas be high enough to fulfil the requirements. Thus

0 0.2 0.4 0.6 0.8 1
1/4
1
2
3
4
E1
V/d
E
V/d
E2
V/d
d1
d
Figure 4.18 Relative values of field strength E, and E2 in the system of
Fig. 4.17 for constant voltage V and gap distance d by varying d1 and
d2 D d d1 for ε2 /ε1 D 4
it is essential to avoid any voids or bubbles within a solid or fluid insulation
system, although this was demonstrated by a uniform field configuration only.
Actual voids can be more complex in shape, and then the field strength will
be more or less reduced (see section 4.3.2).
(c) Either eqn (4.30) or (4.31) may in general be used to calculate the resul-
tant or an ‘effective’ mean value of the permittivity of any homogeneous
mixture of dielectric materials, such as in the case of resin- or oil-impregnated
kraft papers which are extensively used in h.v. apparatus. As such layers are
usually oriented in parallel to the electrodes, the two-dielectric system can
be subdivided into an infinite number of layers with materials designated by
their intrinsic properties ε1 and ε2 and the resultant permittivity εres can be
defined as
D D εresE 4.32
where D and E are macroscopic mean values. As the microscopic values E1
or E2 will remain unchanged by multiple layers, we can write
D D εresE D ε1E1 D ε2E2
or after replacement of E1 or E2 from eqn (4.30) or (4.31) and rearranging
the numbers
εresE D

V
d

1
d1/d
ε1
C
d2/d
ε2

As before, V/d represents the mean value of the field strength within the
mixture, and the distances can be replaced by relative volumes v1 and v2
as the relationships d1/d and d2/d represent also the volumes of the two
materials. Therefore
εres D
1
v1/ε1 C v2/ε2
4.33
or for a mixture of n materials
εres D
1
v1/ε1 C v2/ε2 C . . . C v3/ε3
4.34
with
n
iD1
vi D 1
or 100 per cent.
A kraft paper in which 75 per cent of the volume is filled with cellulose
(ε2
¾
D 6ε0) should be impregnated with mineral oil (ε1
¾
D 2.2ε0). Then v1 D 25
per cent and εres
¾
D 4.19ε0, which is less than one could expect by a merely
linear interpolation.
Please note the assumption which has been made for deriving eqn (4.34):
if the materials are not put in layers parallel to the electrodes, but in normal
directions, different results will apply which can also easily be calculated.
(d) Multi-dielectric insulation systems provide distinct advantages if made
of thin layers making up flexible slabs and which are well impregnated by
fluids or even gases of high breakdown strength such as SF6. Single layers may
have weak points of low breakdown strength; overlapping of many layers will
provide a statistical distribution of the weak points not spread throughout the
insulation. Oil-impregnated h.v. power cables are typical multilayer insulation
systems.
(e) The consistency of the electric flux density at interfaces without free
charges can in non-uniform electrode arrangements be used to make the field
stress more uniform. A typical example is the coaxial cable or coaxial capac-
itor with sandwiched dielectric materials sketched in Fig. 4.19(a). Applying
Gauss’s law to each of the individual interfaces forming equipotential areas
within the field being symmetrical with reference to the centre of the cylinder
axis, one may easily derive the field strength Ex as
EX D
V
εxX
m
nD1
1
εn
ln

rn C 1
rn
4.35

r3
r4
rm
r1
r2
V V = f = 0
X
1
(a)
(b) 0 r1 r2 r3 r4 x
E(x)
E max
2
Figure 4.19 Coaxial cable with layers of different permittivity. (a)
Geometry. (b) Field distribution for ε1 r1 D ε2 r2 D . . . εnrn D const
with V the voltage applied across all of the m layers, and εx the inherent value
of permittivity within the layer of distance x from the centre.
For cylindrical conductors within each layer, Ex is proportional to 1/x,
as this is the case for cylindrical conductors; the discontinuities within the
field distributions caused by the interfaces are recognized in eqn (4.35) as for
x rn and x ½ rn two different values of Ex will appear in the equation. As
the maximum values of Ex are always at the locations x ½ rn it is possible
to maintain the same values Emax within every layer of the dielectric, if εxrn
remains constant. With εx D ε1, ε2, . . . , εn for the individual layers the condi-
tions can be written as
ε1r1 D ε2r2 D . . . D εnrn D const.

The field distribution for this condition is sketched in Fig. 4.19(b). The actual
applicability is, however, restricted by the limited availability of dielectric
materials capable of taking full advantage of this effect. However, in h.v. oil-
filled power cables high-density cellulose papers may be used for the layers
close to the inner conductors, whose resultant permittivity εres is somewhat
higher after impregnation than that for a lower density paper used for larger
diameters.
4.3.2 Dielectric refraction
In the case when the electrical displacement vector D meets the interface
between two media of different permittivities at an angle other than 90°, the
direction of this vector will change in the second dielectric. In general, it
can be assumed that no free charges are present at the interface and only
(dipolar) polarization charges define the boundary conditions. Then the angles
of incidence and refraction are related as follows:
tan ˛1
tan ˛2
D
Et1/En1
Et2/En2
D
En2
En1
D
Dn2/ε2
Dn1/ε1
D
ε1
ε2
. 4.36
These quantities are illustrated in Fig. 4.20 for the conditions ε1 ε2. In
practical systems stressed with d.c. voltages the accumulation of free surface
charges at the interface will take place, caused by the differing conductivities
En2
En1
Et2
Et1
E1
E2
α1
α2
2
1
Figure 4.20 The law of refraction applied to field intensities E for ε1 ε2
of the materials (‘interfacial polarization’, see section 7.1). For a.c. voltage
applications eqn (4.36) may be applied.
Figure 4.21 shows the case when two different dielectrics are placed
between parallel plane electrodes, the interface of which is not perpendicular
to the electrode surface. We observe a compression of equipotential lines at
the corner P increasing the field strength at that point.

100
80
60
40
20
/ 0 = 1
%
P
Interface
/ 0 = 4
Figure 4.21 Two different dielectric materials between plane electrodes
If the angle between interface and electrode in this corner is 90°, the field
intensity at point P becomes theoretically infinite.8,9
This may correspond to
the case when a solid dielectric is only partly attached to the electrode, leaving
a void filled with dielectric materials of inadequate breakdown strength. A
typical example occurs during testing of breakdown strength of solid dielectrics
in the form of plates only shown in Fig. 4.22. The metal disc electrodes may
be of Rogowski’s profile, for which the breakdown could always be achieved
within the uniform field region if only one insulation material is present.
If plates of solid material with permittivity ε2 are tested in atmospheric air
only, for which the breakdown strength as well as the permittivity ε1
¾
D ε0
is much lower than the corresponding values for the solid material, even
for voltages much lower than the breakdown voltage, many partial discharges
will appear starting from the edges as indicated in the figure. These discharges
will spread over the surface of the solid dielectric and will cause breakdown
1
Locations of high electric field
stresses for 1 2
1
2
Figure 4.22 Breakdown tests on solid dielectric plate materials (ε2 ). :
locations of high electrical field stresses for ε1 ε2

outside the uniform field region. To avoid this phenomenon, either compressed
gases of very high dielectric strength must be used or insulation fluids, whose
permittivity ε1 should be higher than ε2 to avoid field enhancement, if the
breakdown strength of the fluid is not as high as that of the solid dielectric.
Therefore, the testing of the insulation strength of solid materials in which no
electrodes can be embedded becomes a troublesome and very difficult task!
However, the law of refraction given by eqn (4.36) can be used to control
the electric field, i.e. to improve the dielectric strength of an insulation system.
Typical examples include spacers of solid materials used in metal-enclosed
gas-insulated substations discussed briefly in section 4.2. The coaxial cylin-
drical conductors are not only insulated by compressed sulphur hexafluoride
(SF6) but also partly by spacers necessary for mechanical support of the
inner conductor. If only a disc of solid material would be used as shown
in Fig. 4.23(a), the flux lines would not be refracted or distorted and the field
strength Ex along the interface between gas and solid material would follow
eqn (4.10). This means that only tangential components of the electric field,
Et, are stressing the interface and Et is not constant along the surface. As the
permissible Et values at boundaries are always lower than field magnitudes
within the adjacent materials, the spacers can be formed in such a way, that all
V = 0
20%
40%
60%
80%
100%
Disc insulator
(a) (b)
Figure 4.23 Epoxy disc insulator supporting the inner conductor of a
coaxial cylinder system. (a) Simple disc spacer: no refraction of equipotential
lines. (b) Shaped spacer for approximate constant tangential field stress at the
interface gas insulator

Et components along the interface remain nearly constant. One possible solu-
tion is shown in Fig. 4.23(b), and the same equipotential lines as in Fig. 4.23(a)
are used to demonstrate the change of the field distribution. The field map for
this example was computed by numerical methods (see section 4.4).
4.3.3 Stress control by floating screens
The necessity for applying electrostatic stress control in h.v. apparatus was
demonstrated up to now for fields in homogeneous materials as well as for
multi-dielectric insulation systems. But in all examples only two metal elec-
trodes have been used whose potential was fixed by the applied voltage. For
homogeneous or single dielectric materials the field stress control was thus
merely possible by providing an adequate shape or contour for these elec-
trodes, and Rogowski’s profile (Fig. 4.3) may be considered an example. The
insertion of multi-dielectric systems between the main electrodes also provided
a means for stress control, as shown for the case of coaxial cables and its two-
dimensional field configuration (Fig. 4.19). For this special case, the interface
between layers of differing permittivity was equipotential. Dielectric interfaces
for general three-dimensional insulation systems, however, are often difficult
to shape such as to provide equipotential surfaces, which would avoid any
tangential field intensities with its limited breakdown strength.
Equation (4.36) indicates that flux lines penetrating from a dielectric of
high permittivity into one of much lower permittivity are forced to leave the
material nearly perpendicular to its surface. This means that the equipotential
lines or surfaces in the dielectric of smaller permittivity are forced to be nearly
parallel to the interface as is found for metal electrodes. A dielectric of very
high εr values thus behaves similarly to an electrically conducting material,
and for ε ! 1 the boundary conditions for metal surfaces are reached. For
this reason insulation systems, including floating screens, whose potential is
solely controlled by the field distribution of the dielectric materials attached
to the screens, can be treated as a multi-dielectric system.
Field stresses are controlled by means of such screens in many h.v. apparatus
such as capacitor-type cable terminations,14
bushings, potential transformers,
etc. The ‘capacitor bushing’ or ‘field stress-controlled bushing’ will be treated
as a typical example and will demonstrate the complexity of the problems
involved. Bushings are used to run a high potential cylindrical conductor H
through a grounded wall or barrier W (see Fig. 4.24). The wall may consist of
a partially conducting concrete or brick, a grounded metal tank of a transformer
or any other metal-enclosed h.v. apparatus. The insulation materials used on
both sides of the wall can, therefore, be different. For transformers mineral oil
insulation inside the tank is typical and atmospheric air is commonly used
outside. For this case, the bushing also provides sealing. The main task,
however, is provided by the electrical insulation of the conductor H from the

I I
I I
W
H
f = 0
f = V
q = b
q = a
W
H
h = h0
O
z
(a)
(b)
Figure 4.24 Simple bushing arrangements for introduction into the
problems solved by capacitor bushings, Fig. 4.25. (a) The problem. (b) A
theoretical solution3

wall and its mechanical support by an insulation system, which is as compact
as possible.
To demonstrate the actual complex problems involved in the design of field
stress controlled bushings, reference is first made to Fig. 4.24(a). The barrier
W perpendicular to the plane of drawing contains a circular opening in which
the cylindrical conductor H is centred. Even without taking into account the
mechanical support and assuming a homogeneity of the insulation material
used, the numerical calculation or graphical field mapping would show that
the high field intensity regions are at the conductor surface within the plane
of the wall and at the edged contours I of the wall opening. To support the
cylindrical conductor and to avoid a breakdown between the wall and the
conductor caused by the high field regions, we may add a solid insulation
material as shown by dashed lines. The solid dielectric would withstand the
high field stresses in the vicinity of the cylindrical conductor; at I at the wall
opening, however, the high tangential components of the field intensities at
the interface between solid and gaseous (or liquid) dielectrics used on both
sides of the wall would cause surface discharges and lead to relatively low
flashover voltages.
One solution to the problem is to use special contours conducting electrodes
and the solid-type insulator supporting the h.v. conductor. An adequate
solution proposed by Moon and Spencer3
is shown in Fig. 4.24(b), displaying
a field map for a three-dimensional arrangement computed with an oblate
spheroidal coordinate system. The electrodes W and H are shaped to give
equipotential lines. The solid dielectric is shaped such that it prevents
refraction of flux lines; the equipotential lines, calculated analytically, remain
unchanged. Although the field intensities are still highest at the shaped
conductors, the improvement of field distribution in comparison to the simple
configurations of Fig. 4.24(a) is clear. As far as we know, bushings of this
type have never been used, as it is much too difficult to produce such a device.
In practice, the solution is in the introduction of ‘floating’ electrodes, as will
be shortly demonstrated. Let the cylindrical h.v. conductor H be surrounded by
many layers of thin dielectric sheets of permittivity ε where ε is considerably
higher than ε0, the permittivity of vacuum or air used for the ‘external’ insu-
lation of the bushing. Figure 4.25(a) shows a simplified cross-sectional view
of such an arrangement, in which the dielectric sheets of different lengths are
interleaved with thin conducting foils providing the floating electrodes; these
are shown by the thicker lines. Neglecting now the influence of the dielectric
conductivity, i.e. the permittivity ε0, of the external insulation, which is accept-
able for a large number of conducting foils, we may treat this system as an
arrangement of coaxial cylindrical capacitor units which are series connected.
Thus a ‘capacitor bushing’ is formed. As indicated, the length l0, l1, . . . , ln
of the sheets is increasing from the wall W to the centre conductor H, and the
conditions for the different lengths can be provided by boundary conditions.

l2
l1
l0
r0
2ri r
r1
r2
1
2
3
d
d
f = V
f = V
f = 0
f = 0
H
W
(a)
(b)
(c)
z
Ez
Et
Er
r
r0
ri
l0 /2 li /2
0 z
Figure 4.25 Capacitor bushing. (a) Coaxial capacitor arrangement.
(b) Profile of foils for constant radial field intensity Er (mean value).
(c) Definition of field intensity components
Let us assume the simplest boundary condition, for which the mean value of
the field intensity Er acting within the sheets remains constant. If every sheet
is of equal thickness υ, each of the coaxial capacitor units is stressed by equal
voltages V D Erυ, if all capacitances are equal. Then C1 D C2 D . . . Cn
with
C1 D
2εl0
lnr0/r1
,
C2 D
2εl1
lnr1/r2
,
or
l0
lnr0/r1
D
l1
lnr1/r2
D . . . D
ln
lnrn/rnC1
4.37

Apart from this exact solution, an approximation is possible for thin sheets.
Then rnC1 D rn υ and υ/rn − 1 even for the smallest radius ri of the inner
conductor, yielding
lnrn/rnC1 D ln

1
1 υ/rn

¾
D υ/rn
With this approximation, eqn (4.37) becomes
l0r0
¾
D l1r1
¾
D . . . ¾
D lnrn, 4.38
where 0 n N, with N equal to the total number of sheets. As N is quite
high, we may replace the discrete numbers ln and rn by the variables z D 1/2
(Fig. 4.25(a)) and r. Equation (4.38) then defines a two-dimensional profile
or contour of the conducting foil edges as sketched in Fig. 4.25(b). The given
boundary condition provides a hyperbolic profile, along which the potential
increases steadily between r0 and ri. Neglecting the very local increase of the
field intensities produced by the edges of the conducting foils, we can now
assume quite constant values between two foils, i.e. a mean radial field strength
Er as indicated in Fig. 4.25(c). Whereas Er stresses the insulation material of
the sheets only, an even more significant axial component of a field intensity
Ez is introduced between the conducting foil edges, as sketched in Fig. 4.25(c).
The solid material from the active part of the capacitor bushing also shares a
boundary with the surrounding dielectric material, in general atmospheric air,
or mineral oil.
Therefore, this interface is in summary stressed by a tangential field intensity
Et, which has the components of Er and Ez, the latter defined as a mean value
of the potential difference  between each adjacent foil and the increase
l D 2z in sheet length, i.e. Ez D /l. For the small values of ,
l and r provided by the large amount of sections we may neglect all
discontinuities and write in differential terms
d D Er dr D Ez dz
where in addition dz D dl/2 with the dimensions assumed in Fig. 4.25.
Apart from the sign, the gradient d can be assumed to be a voltage drop
dV across the capacitor elements as formed by adjacent electrodes. As each
capacitor element C0
D 2εlr/dr, in which the differential term dr is used
to quantify the quite small distances between the electrodes or thickness of
the dielectric sheets, the voltage drop becomes
dV D
i
ωC0 D
i
ω2εlr
dr D K0
dr
lr
In this equation, the product lr can be taken from any locus of an electrode
as indicated in eqn (4.38). As all capacitor elements are series connected, the

displacement current i is always the same and thus K0 is a constant. Now we
are able to indicate the field stresses in general terms as
Er D K0
1
lr
4.39a
Ez D 2K0
1
lr
ž
dr
dl
4.39b
From both equations, different kinds of boundary conditions may be intro-
duced to find criteria for the field stresses for the tangential components
which in general control the external flashover voltages. We use one example
only, defined by the assumption that the radial stress Er shall be constant.
Then, according to eqn (4.39a), the product lr D K0/Er0 where Er0 will be
a convenient design criteria. Applying this term to eqn (4.39b) provides the
axial component as
Ez D 2K0
1
l2
D 2Er0
r
l
D 2
E2
r0
K0
r2
4.40
This dependency shows the strong increase of the axial field strength
with increasing diameter of the dielectric sheets. It contributes to a non-
homogeneous potential distribution at the surface of the laminated unit and a
highest stress at the grounded flange promoting surface flashover, as the mean
value of the tangential field intensity according to Fig. 4.25(c) is
Et D

Er2 C Ez2, 4.41
if the surface is very close to the foil edges.
In practice, such a dimensioning of a capacitor bushing due to constant
mean values of the radial field intensity is not at all ideal and the calculations
performed only indicate the problems. But one could readily see that the
conducting foils can be used to control the internal fields Er as well as the field
strength distribution along the boundaries Et, and that it will not be possible to
keep both these values constant. The dimensioning of bushings thus becomes
a difficult task, as also other important factors have to be taken into account.
First, the surrounding insulation materials cannot be neglected. Secondly, h.v.
bushings, in general, are not made from a single dielectric material, which
is often provided by oil- or resin-impregnated kraft paper or plastic films.
Protection of the active part is provided by porcelain or other solid insulation
material housings, having different permittivities and introducing additional
field refraction at the interfaces of the differing materials. Due to the heat
generated within the h.v. conductor H the permissible radial field intensity
may be lower than within the outer regions. Finally, careful attention must be
paid to the edges of the conducting foils which form regions of locally high

fields, as the equipotential lines will not necessarily leave the foils at the edges,
but in its vicinity only. Therefore, foils made from semiconducting materials
of still adequate conductivity are sometimes used to adapt the potentials at the
foil edges to the field distribution forced by the dielectric materials outside
the field-controlled regions. Analytical computations of bushing designs are,
therefore, supplemented by numerical computations, which take into account
the very different boundary conditions.18,19
(a)
(b)
Figure 4.26 Photographs of bushing (courtesy Micafil, Switzerland).
(a) Wall bushing, outdoor–indoor, rated 123 kV/1250 A. (b) Transformer
bushing with ‘dry’ insulation, rated 170 kV a.c./630 A, BIL 750 kV
Figure 4.26 shows two typical types of bushings, a transformer bushing
with its asymmetry due to different external insulation, and a common wall
bushing for air-to-air insulation, but one side indoor, the other side (porcelain)
outdoor.
4.4 Numerical methods
In recent years several numerical methods for solving partial differential
equations and thus also Laplace’s and Poisson’s equations have become
available. There are inherent difficulties in solving partial differential

equations and thus in Laplace’s or Poisson’s equations for general two-
or three-dimensional fields with sophisticated boundary conditions, or for
insulating materials with different permittivities and/or conductivities. Each
of the different numerical methods, however, has inherent advantages or
disadvantages, depending upon the actual problem to be solved, and thus
the methods are to some extent complementary.20
The aim of this chapter is to introduce the most widely used methods in
such a way that a fundamental knowledge is provided and to give the user
of a computer program an understanding of the limitations of the results and
computations.
4.4.1 Finite difference method (FDM)
Apart from other numerical methods for solving partial differential equations,
the finite difference method (FDM) is quite universally applicable to linear and
even non-linear problems. Although this method can be traced back to C.F.
Gauss (1777–1855), and Boltzmann had already demonstrated in 1892 in his
lectures in Munich the applicability of difference equations to solve Laplace’s
equation, it was not until the 1940s that FDMs had been used widely.
The applicability of FDMs to solutions of general partial differential
equations is well documented in specialized books.21,22
More specific
references concerning the treatment of electric and magnetic field problems
with the FDM can be found in reference 23.
This introduction is illustrated by two-dimensional problems for which
Laplace’s equation, or Poisson’s equation,
r2
D
∂2

∂x2
C
∂2

∂y2
D fx, y
applies. The field problem is then given within an x–y-plane, the area of
which has to be limited by given boundary conditions, i.e. by contours on
which some field quantities are known. It is also known that every potential
and its distribution within the area under consideration will be continuous in
nature. Therefore, an unlimited number of x, y values would be necessary to
ascribe the potential distribution. As every numerical computation can provide
a limited amount of information, only a discretization of the area will be
necessary to exhibit nodes for which the solution may be found. Such nodes
are produced by any net or grid laid down upon the area.
As any irregular net, however, would lead to inadequate difference
equations replacing the original partial differential equation, and would thus
be prohibitive for numerical computations, the FDM is in general applied
to regular nets or polygons only. These restrictions will be understood more
clearly by the derivation of the differential equations. Regular polygons which
can fill a plane are squares, triangles or hexagons, but squares or equilateral

triangles are the only regular nets in common use. As also such square or
triangular nets will in general not fit into the boundaries, we will derive the
difference equations for rectangles, which can at least at given boundaries be
formed in such a way that nodes can also be laid down upon the boundary. As
squares are a particular case of rectangles, the result applies also for squares,
and the inherent difficulties in using irregular nets are better understood.
0 XW XP XE x
XN
XS
(SW )
(NW )
N
(NE)
(SE )
S
W P E
hs
hW
hN hE
y
Figure 4.27 Irregular rectangular net composed of horizontal and vertical
lines, with node abbreviations
In Fig. 4.27 such an irregular net of rectangles is sketched within the x–y-
plane, with the sides of all rectangles parallel to the x- or y-axis. All points
of intersection between the vertical and horizontal lines create nodes, but only
five nodes will be of immediate special interest. These are the four neigh-
bouring nodes, N, W, S, and E around a point P, which are given compass
notations (N D north, etc.). Let us assume now that the potentials at these
nodes, i.e. S, E, N and W are known either from given boundary
conditions or other computational results. As the potential within the field
region is continuous, it is obviously possible to expand the potential at any
point x, y by the use of Taylor’s series. If this point is identical with node
P, the series for the two variables x and y is given by
x, y D P C
1
1!
[x xPxP C y ypyP]
C
1
2!
[x xP2
xxP C 2x xPy yPxyP
C y yP2
yyP]

C
1
3!
[x xP3
xxxP C 3x xP2
y yPxxyP
C 3x xPy yP2
xyyP C y yP3
yyyP]
C R0
xyP 4.42
where the rest of the series R0
xyP is of still higher order. In eqn (4.42) the
derivatives
xP D

∂x, y
∂x

P
, yP D

∂x, y
∂y

P
, xyP D

∂x, y
∂x∂y

P
, etc.
are used for abbreviation.
Every potential x, y in the close vicinity of node P may be expressed
by eqn (4.42) with adequate accuracy, if the Taylor’s series is interrupted
by ignoring terms containing third derivatives of the potential, as they will be
multiplied by small distances h to the power of 3 or more. Thus, the potentials
of the nodes E, N, W and S can be expressed by the following equations, in
which the small distances x xP and y yP are substituted by the proper
values hE, hN, hW and hS:
E D P C hExP C 1
2 h2
ExxP 4.43a
N D P C hNyP C 1
2 h2
NyyP 4.43b
W D P C hWxP C 1
2
h2
WxxP 4.43c
S D P C hSyP C 1
2
h2
SyyP 4.43d
The sums of eqns (4.43a) and (4.43c), and eqns (4.43b) and (4.43d) respec-
tively yield the following two equations:
E C W 2P D hE hWxP C 1
2
h2
E C h2
WxxP, 4.44a
N C S 2P D hN hSyP C 1
2 h2
N C h2
SyyP, 4.44b
The derivatives xP and yP may be expressed by the well-known first
order approximations
xP ¾
D
hW
hE
[E P] C
hE
hw
[P W]
hE C hW
D
hW
hEhE C hW
E C
hE hW
hEhW
P
hE
hWhE C hW
W 4.45a
yP ¾
D
hS
hNhN C hS
N C
hN hS
hNhS
P
hN
hShN C hS
S 4.45b

Introducing eqn (4.45a) into (4.44a) and eqn (4.45b) into (4.44b) will result
in
xxP D
2E
hEhE C hW
C
2W
hWhE C hW

2P
hEhW
, 4.46a
yyP D
2N
hNhN C hS
C
2S
hShN C hS

2P
hNhS
, 4.46b
With these approximations for the second derivatives of the potential functions
in the x- and y-direction at node P it is now possible to solve Laplace’s or
Poisson’s equation,
r2
D xx C yy D

0 (Laplacian region)
Fx, y (Poissonian region)
4.47
where Fx, y D ./ε for electrostatic fields within a medium of permittivity ε
and containing distributed charges of density .x, y. The solution may then
be written as
DEPE C DNPN C DWPW C DSPS C DPPP C 1
2 FP D 0
4.48
with
DEP D
1
hEhE C hW
, DNP D
1
hNhN C hS
,
DWP D
1
hWhE C hW
, DSP D
1
hShN C hS
,
DPP D

1
hEhW
C
1
hNhS

.
This difference equation is a valid approximation of the original differential
equation (4.47), but it should be recalled that the validity is restricted to the
individual point P under consideration. The same form is, however, valid for
every node within a net.
Before further considerations we shall discuss briefly the common simpli-
fications. For every two-dimensional problem most of the field regions can
be subdivided by a regular square net. Then hE D hN D hW D hS D h, and
eqn (4.48) is reduced to
E C N C W C S 4P C h2
FP D 0 4.49
It may well be understood now that difference equations similar to eqn (4.48)
can be derived for other nets or other neighbouring nodes to P within our
rectangular net shown in Fig. 4.27 if the proper derivations are performed. In
this figure, for instance, one could involve the nodes NE, NW, SW and SE

either neglecting the nodes N, W, S and E or including the nodes. In all cases,
the unknown potential P can be expressed by the surrounding potentials
which are assumed to be known for the single difference equation.
All difference equations, however, are approximations to the field equation
due to the omission of higher order terms in eqns (4.42) and (4.45). The error
due to these approximations is known as truncation error, and it is important
to investigate this error carefully if the values h are not chosen properly. The
treatment of the truncation error is beyond the scope of this chapter and may
be found elsewhere.21,23
The numerical evaluation of the difference equation (4.49) is obviously
simple, but time consuming and therefore FDM is now seldom used.
4.4.2 Finite element method (FEM)
By reviewing the theory of the FDM it was readily demonstrated that the partial
derivatives of the basic field equations (4.47) have been replaced by their
algebraic difference form, eqns (4.46a, b), resulting in a system of algebraic
equations which have to be solved. Due to the approximations made during
this derivation the algorithm was linear of the first order (‘first order FDM
algorithm’).
Although there are different approaches to arrive mathematically at finite
element approximations24
and the most general approach is traced back to the
variational problem of extremization of a specific functional, the most common
basis is related to a very well-known physical property of fields. The FEM
concerns itself with minimizing the energy in the whole field region of interest,
when the field may be electric or magnetic, of Laplacian or Poissonian type.
In this section a specific rather than general treatment of the method will be
presented. To reduce the size of equations, we will restrict ourselves to two-
dimensional electric fields of Laplacian type. Convenient applications even for
complicated Poissonian electric fields as, for instance, present around coro-
nating h.v. lines are documented in references 28 and 29, as well as the
practical application to magnetic fields.25–27
Let us consider a steady state electrostatic field within a dielectric material
whose conductivity may be neglected and whose permittivity may be depen-
dent upon the direction of the field strength E (anisotropic material) or not
(isotropic dielectric). Then as no space charge should be present or accu-
mulated, the potentials would be excited from boundaries (metal electrodes)
between which the dielectric material is placed. Assuming a Cartesian coordi-
nate system, for such a Laplacian field, the electrical energy W stored within
the whole volume R of the region under consideration is
W D
' ' '
V
%
1
2
εx

∂
∂x
2
C εy

∂
∂y
2
C εz

∂
∂z
2

dx dy dz. 4.50

εx, εy and εz would be anisotropic permittivity coefficients, and it should be
noted that even in an isotropic material with εx D εy D εz D ε, the absolute
values of ε may change at boundaries between different dielectric materials.
The reader may easily verify from any small volume element dV D dx dy dz
that the expressions εr2
/2 within eqn (4.50) are energy densities per unit
volumes dV.
y
x
−z
A
f = fb
f = fa
I
I
II
Il
Figure 4.28 Limited field area A within x–y-plane representing a
two-dimensional field within space (x, y, z-coordinates). Dielectric material
subdivided by dashed line into regions I and II
Furthermore, it is assumed that the potential distribution does not change in
the z-direction, i.e. a two-dimensional case. Figure 4.28 displays the situation
for which the field space is reduced from the volume R to the area A limited
by boundaries with given potentials a and b (Dirichlet boundaries). The
dielectric may be subdivided into two parts, I and II, indicated by the dashed
interface, for which the boundary condition is well known (see section 4.3),
if no free charges are built up at the interface. The total stored energy within
this area-limited system is now given according to eqn (4.50) by
W D z
''
A
%
1
2
εx

∂
∂x
2
C εy

∂
∂y
2

dx dy 4.51
where z is a constant. W/z is thus an energy density per elementary area dA.
Before any minimization criteria based upon eqn (4.51) can be applied,
appropriate assumptions about the potential distribution x, y must be made.
It should be emphasized that this function is continuous and a finite number
of derivatives may exist. As it will be impossible to find a continuous function
for the whole area A, an adequate discretization must be made.

For our two-dimensional problem it is possible to use rectangular or square
elements, as was done for the FDM (see Fig. 4.27), or multiple node composite
elements for three-dimensional regions. There are, however, definite advan-
tages in using simple, irregularly distributed elements with an arbitrary trian-
gular shape (or tetrahedrons for three-dimensional problems). Such triangles
can easily be fitted to coincide with boundary shapes, i.e. the nodes of a trian-
gular element system can be placed upon curved boundaries, a situation often
met in h.v. insulation systems.
y
j
i
m (e)
yj
ym
yi
xm xj xi x
Figure 4.29 A section of area A (Fig. 4.28) subdivided into irregular
triangular elements. Notation of nodes i, j, m for element e
Figure 4.29 shows such a subdivision of a part of a two-dimensional region
A (in Fig. 4.28) into triangular elements. Let us consider one of these elements
(indicated by e) and the nodes i, j and m, and formulate basic functions for
the potential distributions x, y within this element. In the development of
the FEM no a priori restrictions are placed on this basic function. However,
for triangular elements, or a ‘triangular element family’,29
polynomials can
be of higher order, such as
x, y D ˛1 C ˛2x C ˛3y C ˛4x2
C ˛5xy C ˛6y2
C . . . , 4.52
for which the inter-element compatibility can be improved. The increase in
accuracy by applying higher order functions is compensated, however, by an
increase in computation time and computation complexity, and thus most of the
algorithms used are based upon a first order approximation, i.e. a linear depen-
dency of on x and y in eqn (4.52). Following this simple basic function,
this equation is reduced to
x, y D D ˛1 C ˛2x C ˛3y. 4.53

This means that the potentials within each element are linearly distributed
and the field intensity, whose components in the x- and y-directions can be
computed for eqn (4.53) by simple derivation, is constant. In this respect, the
FEM and the FDM coincide.
For such a first order approximation, the three coefficients ˛1, ˛2 and ˛3
for element e are easily computed by means of the three a priori unknown
potentials at the respective nodes i, j and m, given by the equations
i D ˛1 C ˛2xi C ˛3yi
j D ˛1 C ˛2xj C ˛3yj
m D ˛1 C ˛2xm C ˛3ym
4.54
The coefficients may be computed applying Cramer’s rule, the result being
˛1 D
1
2e
aii C ajj C amm; 4.55a
˛2 D
1
2e
bii C bjj C bmm; 4.55b
˛3 D
1
2e
cii C cjj C cmm; 4.55c
where
ai D xjym xmyj
aj D xmyi xiym
am D xiyj xjyi
4.55d
bi D yj ym
bj D ym yi
bm D yi yj



4.55e
ci D xm xj
cj D xi xm
cm D xj xi
4.55f
and
2e D ai C aj C am
D bicj bjci

4.56
From eqn (4.56) and Fig. 4.29, one may easily see that the symbol e is used
to describe the area of the triangular element i, j, m.
With eqns (4.53), (4.54) and (4.55), the potential distribution of the element
can thus be related to the potentials of the adjoining nodes, and simple numbers
ai, bi, etc. for each element can be computed once division of the two-
dimensional region into triangular elements has been performed. Introducing

these values into eqn (4.53) the result is (index e used for ‘element’):
ex, y D
1
2e
[ai C bi C cii C . . .
C aj C bj C cjj C am C bm C cmm] 4.57
This equation may also be written as
e D
(
Ni, Nj, Nm
) i
j
m
4.58
in which the functions N are the ‘shape functions’, as they will depend upon
the shape of the finite elements used. Such shape functions can be derived
for many kinds and shapes of elements including the rectangles used for the
FDM.24
With eqn (4.57) or eqn (4.53), the energy noted within the element is
easily computed. According to eqn (4.51), the partial derivatives for each
element are:
∂
∂x
D ˛2 D fi, j, m
∂
∂y
D ˛3 D fi, j, m 4.59
However, as we are not interested in the absolute values of these energies,
the components of the electric field intensities should not be introduced into
eqn (4.51) at this stage. The FEM is based upon the minimization of the energy
within the whole system, and thus only derivatives of the energies with respect
to the potential distribution are of interest. According to eqn (4.51), the energy
functional, i.e. the energy per unit length in the z-direction for our specific
case, is for the element under consideration
?e
D
We
z
D
1
2
e εx

∂
∂x
2
C εy

∂
∂y
2
e
4.60
as
**
dx dy provides the area of the element, e. For further consideration
only, isotropic dielectric material is assumed within each individual element,
i.e εx D εy D εe.
Whereas the functional ?e
in eqn (4.60) is only dependent upon the node
potentials of the individual element (eqn (4.59)), an equivalent functional ? for
the whole system (area A, Fig. 4.28) will exist. The formulation regarding the
minimization of the energy within the complete system may thus be written as
∂?
∂fg
D 0 4.61

where fg is the potential vector for all nodes within this system. For our
specific element, the minimizing equations can easily be derived by differenti-
ating eqn (4.60) partially with respect to i, j and m. Taking also eqns (4.55)
and (4.59) into account, the differentiation with respect to i yields
∂?e
∂i
D
1
2
εee

2˛2
∂˛2
∂i
C 2˛3
∂˛3
∂i

D
1
2
εe˛2bi C ˛3ci
D
εe
4e
[b2
i C c2
i i C bibj C cicjj C bibm C cicmm] 4.62
The set of all three equations may best be expressed in matrix form as
∂?e
∂fge D
εe
4e


b2
i C c2
i bibj C cicj bibm C cicm
b2
j C c2
j bjbm C cjcm
sym b2
m C c2
m


i
j
m
D [h]e
fe
g 4.63
The matrix [h]e
is well known as the ‘stiffness matrix’ for the individual
element, as it contains the sensitivity of the functional with respect to the
potentials. (Within a mechanical, elastic system, this matrix relates mechanical
nodal forces to displacements.) It contains well-known geometric quantities
(eqns (4.55), (4.56)) and the material’s permittivity εe.
It is now possible to establish a set of algebraic equations with which
the still unknown potentials can be computed. No assumptions have been
made so far concerning the a priori known potentials at the boundaries, and
Fig. 4.28 displayed only the finite field regions with a Dirichlet boundary.
The triangular element e within Fig. 4.29 is surrounded by other triangular
elements and it is seen that any node potential within such a system will
depend upon the potentials of the surrounding nodes. The number of these
nodes is dependent upon the triangular network, but that number is always
small. Thus it is sufficient to demonstrate the last step with a set of only four
triangular elements as shown in Fig. 4.30. The elements are numbered from
1 to 4, and the nodes by 1 to 5.
Application of eqn (4.61) to this set of elements yields
∂?
∂5
D 0, 4.64
where ? is the energy functional of the system with the four elements.
Before this equation is evaluated, it is convenient to write the stiffness matrix,

x
y
3
3
4
4
2
2
5
1
1
(m)
(i )
(j)
Figure 4.30 Node 5 connected to four triangular elements (i, j, m identify
element l)
eqn (4.63), as
[h]e
D
%
hiie hije hime
hjje hjme
sym hmme

4.65
where
hiie D
εe
4e
b2
i C c2
i ;
hije D
εe
4e
bibj C cicj;
.
.
.
etc.
Replacing the index e by the individual numbers of the elements of Fig. 4.30
results in
∂?
∂5
D 0 D 4.66
(from element 1) D [him12 C hjm11 C hmm15 C . . .
(from element 4) D [him41 C hjm44 C hmm45].

This equation may be written as
H151 C H252 C H353 C H454 C H555 D 0 4.67
where
H15 D [him4 C hjm1]
H25 D [him1 C hjm2]
.
.
.
H55 D [hmm1 C hmm2 C hmm3 C hmm4] D
4
rD1
hmmr
If the potentials 1 to 4 were known, 5 could immediately be calculated
from this equation. As, however, the potentials of the nodes 1 to 4 might
still be embedded in a larger triangular network, for every unknown poten-
tial a corresponding equation has to be set up. For our system with Laplace
conditions the FEM solution may thus be written as
∂?
∂fg
D 0 D [H]fg 4.68
indicating the assembly of the whole set of minimizing equations, which can
be solved following the usual rules (see section 4.4.1).
This short, detailed introduction of the FEM cannot demonstrate all the
advantages and disadvantages of the method. In the application to electric
field problems within insulation systems, the advantages may be summarized
as follows:
(a) It is readily applicable to non-homogeneous systems (i.e. with materials
of different permittivities) as well as to anisotropic systems (refer to
eqn (4.51)).
(b) The shapes and sizes of the elements may be chosen to fit arbitrary
boundaries and the grid size may easily be adapted to the gradient of
the potentials, i.e. small elements can be placed into regions with high
gradients and vice versa.
(c) Accuracy may also be improved using higher order elements (compare
with eqn (4.52)), without complicating boundary conditions.
(d) Dielectric materials may also be treated as the case where conduction
currents contribute to the potential distribution. This can be done by
assuming complex permittivity with real and imaginary parts (i.e. ε D
ε0
–jε00
, where tan υ D ε00
/ε0
).30
For the calculation of electric field intensities within electric insulation
systems, the only disadvantage of the FEM is still related to the limited and a

priori unknown accuracy which can be achieved. Even for two-dimensional
problems and highly divergent fields, a very large number of triangular
elements or nodes would be necessary to obtain an adequate accuracy within
the highly divergent field regions, which are responsible for the breakdown
of the whole system. It should be remembered that the often used first order
algorithm (see eqn (4.53)) does result in a constant field strength within each
element (see eqn (4.59)), which is only approximately correct for the case of
continuous field distribution within homogeneous materials. Although the size
of the elements can well be adapted to the divergence of the field distribution,
too large a number of elements or nodes would be required for high accuracy.
Efficient computation algorithms are necessary to solve eqn (4.68), as the
stiffness matrix [H], although highly sparse and symmetric, will become
very large.
Finally, Fig. 4.31 shows an example of a field computation using the FEM.
Figure 4.31(a) displays the original triangular grid used for computation of
a coaxial section of a GIS comprising a conical space rεr D 6.5 within
the gaseous insulation system εr ³ 1. The result of this grid displayed by
5 per cent equipotential lines (Fig. 4.31(b)) still shows some discontinuities,
although much smaller triangular elements have been used in regions with
high field non-uniformity. Figures 4.31(c) and (d) are sections of the figures
shown before. The same sections computed by a much higher grid density
(see Figs 4.31(e), (f)) confirm the large improvement displayed by the new
equipotential lines, the discontinuity of which disappeared.
4.4.3 Charge simulation method (CSM)
A third numerical method widely and successfully used today to calculate
electric fields is known as the charge simulation method (CSM). Though the
fundamentals of this method may be familiar to most electrical engineers, as
it is based upon frequently used analytical field computation methods, it may
be useful to review some fundamentals of Maxwell’s equations.
The Poisson’s equation may be written as
divε grad D div D D . 4.69
(which is a differential form of Maxwell’s equation) where D is the electric
flux density and p is the volume charge density. Equation (4.69) is independent
of any particular coordinate system. This equation may be integrated by means
of a volume integral, resulting in
'
V
divD dV D
+
A
D dA D
'
V
. dV 4.70
Here the volume integral of divergence is transferred to a surface integral
with the closed surface A, and the volume integral applied to the charge

density can easily be identified with the total charge enclosed by the surface
A. Equation (4.70) represents the well-known Gauss’s law. This law gives
an exact solution of Poisson’s equation, and many direct methods for field
computations are based upon this law.
Figure 4.31 Field computation by FEM. Coaxial section of GIS.
(a) Triangular elements for the computation of a limited axial section
comprising the spacer. (b) Result of the computation (5 per cent equipotential
lines) based upon the grid displayed in (a). (c), (d) A section of (a) and (b)
enlarged. Same axial and radial notations. (e) A section of the improved grid
with a much higher number of elements in comparison to (a) or (c)
respectively. (f) Equipotential lines due to (e) (courtesy of BBC, Baden,
Switzerland)

We have seen that both the FDM (section 4.4.1) and the FEM (section 4.4.2)
are directly based upon the differential form of a Maxwell equation (see
eqn (4.69)). We noted also, in previous sections, that solving the differential
equations, either analytically or numerically, involves difficulties inherent in
the formulation of boundary conditions as well as due to inaccuracies arising
within numerical procedures. On the other hand, Gauss’s law is much easier
to apply, at least for the cases where some symmetry boundary conditions
are apparent. This advantage was used by applying eqn (4.70) to calculate
analytically some simple field configurations, e.g. coaxial cylindrical or spher-
ical fields (see section 4.2), for which the integrals of the left-hand side of
eqn (4.70) could easily be solved due to symmetry conditions, arising from
a concentration of the charge distribution (right-hand side of this equation)
within line or point charges respectively.
Directly related to the application of Gauss’s law is the method of images (or
image charges), which could be used to compute analytically some important
problems by means of ready-made solutions, thus eliminating the need for
formal solutions of Laplace’s or Poisson’s equations in differential form. This
method, which can be traced to Lord Kelvin31
and Maxwell,32
was also
used for field computation of a sphere-to-sphere arrangement.
Steinbigler7
introduced this technique as an efficient method for digital
computation of electric fields. Since its publication in English34
this method
(CSM) has been recognized to be very competitive and often superior to
FEM or FDM, at least for treating two- or three-dimensional fields within h.v.
insulation systems, particularly where high accuracies within highly divergent
field areas are demanded. Although the efficiency and applicability of the CSM

may not have been fully developed up to now, many recent publications have
shown interest in this technique.20,35,42
The basic principle of CSM is very easy to formulate. Using the superposi-
tion principle, the potential functions of the fields of individual charges of any
type (point, line or ring charges, for instance) can be found by a summation
of the potentials (scalars) resulting from the individual charges. Let Qj be a
number n of individual charges, and Qj be the potential at any point within the
space (independent of the coordinate system used). The superposition principle
results in
i D
n
jD1
pijQj 4.71
where pij are the potential coefficients, which are known for many types of
individual charges by particular solutions of Laplace’s or Poisson’s equations
mentioned earlier. Figure 4.32 displays a point charge QP and a line charge
Ql placed at the x- and y-axis respectively and an arbitrary point Pi at which
the potential @ would apply.
y
z
x
Qp = Q1
Q2
= Ql Pij = Pi1
Pij = Pi 2
Pi
Figure 4.32 A finite line charge Ql and a point charge QP related to a field
point Pi
Whereas the potential coefficients, pij, . . ., are known, only additional
boundary conditions enable us to relate i with Qj quantitatively. If the
individual charges are placed outside the space in which the field is to be
computed (or inside a closed metal electrode, whose surface is an quipotential
area), the magnitudes of these charges are related to the distributed surface
charges which are physically bonded by the electric flux leaving or entering

the surface of any electrode or conductor surrounding these charges. If n
charges Qj are assumed, we require also at least n known potentials to solve
eqn (4.71) for the a priori unknown charge magnitudes. This can easily be
done by identifying the potentials i with n potentials on the surface of the
conductors (‘contour points’), which are adequately placed at a given electrode
configuration. If this potential is @ D ?, we may rewrite eqn (4.71) as
n
jD1
pijQj D c. 4.72
This equation leads to a system of n linear equations for the n unknown
charges




p11 p12 Ð Ð Ð p1n
p21 p22 Ð Ð Ð p2n
.
.
.
pn1 pn2 Ð Ð Ð pnn











Q1
Q2
.
.
.
Qn







D







1
2
.
.
.
n







4.72a
or
[p]fQg D fg.
After this system has been solved, it is necessary to check whether the set of
calculated charges fits the actual boundary conditions. It must be emphasized
that only n discrete contour points of the real electrode system have been
used to solve eqn (4.72), and thus the potentials at any other contour points
considered in this calculation might still be different from X. Therefore,
eqn (4.71) must be additionally used to compute the potentials at a number of
‘check points’ located on the electrode boundary (with known potential). The
difference between these potentials and the given boundary potential is then a
measure of the accuracy and applicability of the simulation. The development
and introduction of special objective functions is thus an important procedure
within the optimization of the CSM.34–36
As soon as an adequate charge system has been adopted, the potentials and
the field strength within the space can be computed. Whereas the potentials are
found by superposition, i.e. by eqn (4.71) or the corresponding set of linear
equations (compare with eqn (4.72)), the field stresses are calculated by super-
position of magnitudes and directional components. For a Cartesian coordinate
system, for instance, the x-coordinate Ex would then be for a number of n
charges.
Ex D
n
jD1
∂pij
∂x
Qj D
n
jD1
fijxQj 4.73

where fij are ‘field intensity coefficients’ in the x-direction. Before further
considerations, the computation algorithm may be applied to a simple example.
In Fig. 4.33(a), a symmetrical sphere-to-sphere electrode system is sketched
symmetrically charged to šV. This condition implies zero potential for the
plane z D 0 as well as for the dielectric space at a distance from the spheres
(unlimited dielectric space). Thus the field configuration is axisymmetric with
the rotation centre being the z-axis. This simple example would be difficult to
compute by FDM or FEM, as the space is unlimited.
Let us consider the case of two point charges šQ1 and šQ2 symmetrically
placed along the axis at r D 0; z D š0.75/1.25D and only two contour points
P1, P2 at r D 0 as shown in Fig. 4.33(a). The symmetric arrangement of the
charges (imaging) gives V D 0 at z D 0. Thus also a sphere-to-plane geometry
is computed.
To solve eqn (4.72), the potential coefficients for a point charge are neces-
sary. The potential related to a point charge Q at distant d is given by:
D
Q
4ε
1
d
D pQ 4.74
or
p D
1
4εd
Thus the potential coefficients pij are dependent upon the distance d between
the charges Qj and the contour points Pi. For our r–z-coordinate system, the
coefficients may be expressed by
pij D
1
4ε

r2
i C
,
zi zj
-2
4.74a
from simple geometric considerations. Let the computer now solve the four
simple equations using eqn (4.72), in order to obtain the magnitudes of Q1
and Q2, and to compute a sufficient set of other potentials within the r–z-
coordinate system. These potentials can be used to draw equipotential lines;
such lines are shown in Fig. 4.33(b) for part of the positive z-axis. The result
may appear disappointing, since the equipotential line CV deviates grossly
from the circle, representing the cross-section of the sphere. An agreement of
the computed and given potential is only found for the contour points P1, P2,
but for other contour points a disagreement extending up to about 39 per cent
can be observed. This suggests a very poor simulation and bad assumption of
point charges. Therefore, we may add a third point charge Q3 and contour point
P3, as also indicated in Fig. 4.33(a), and repeat the calculations. The result is
now shown in Fig. 4.33(c). The disagreement between the real contour of the

z
P2
P1
D
+V
−V
−Q 3
−Q 2
−Q 1
V = 0
P3
D2
D
Q3
Q2
Q1
0.75 D
1.25 D
D
0 r
r
P′
1
P′
2
P′
3
D2
(a)
(b)
z
P2
+V
P1
Q 1
Q 2
Figure 4.33 Example for CSM. (a) Sphere-to-sphere electrode arrangement.
(b) to (d) Computed results. (For more information see text)

z
r
(c)
(d)
P2
Q1
Q3 P3
Q2
P1
z
r
P2
Q3
Q2
P1

electrode and the computed equipotential line D CV is now very small, not
exceeding 1.98 per cent, an error difficult to establish within the figure. This
means that the simulation was greatly improved and it is easy to recognize that
more charges improve the computation. We can, however, also find excellent
solutions using only two simulation charges placed at proper positions. This
was done in Fig. 4.33(d), where again only two charges and two contour
points were used to solve the problem. The largest deviation for the computed
potential D V from the sphere is now less than 0.2 per cent.
This simple example demonstrates two essential features concerning an
effective application of the CSM. The first relates to the proper selection
of the types of simulation charges, and the second to a suitable arrangement
of the charges and contour points.
Various other charge types are available for which the potential coeffi-
cients are known from analytical solutions. For our example, the application
of toroidal line charges (ring charges) of constant charge density and centred
on the axis of symmetry would have been an effective method of discretiza-
tion. One could also use infinite or finite line charges, or even plane or curved
surface charges. The complexity of computation, however, in general increases
with the complexity of the simulation charges used, as the potential coeffi-
cients become more difficult to compute numerically. As an exercise, only
the coefficients for finite line charges and toroidal line charges are reproduced
here. With the notations of Fig. 4.34, the potential coefficients are for:
Finite straight line charges (Fig. 4.34(a)):
pij D
1
4εzj2 zj1
ln
zj2 zi C A1zj1 C zi C A2
zj1 zi C υ1zj2 C zi C υ2
4.75
z z
zj 2
zj
Qj
Qj
zj 1
ri r −rj rj ri r
zi
Pi
zi Pi
(a) (b)
Figure 4.34 Finite straight line charges (a) and toroidal line (ring)
charges (b), with constant charge densities. Total charge: Qj

where
A1 D

r2
i C zj2 zi2; A2 D

r2
i C zj1 C zi2,
υ1 D

r2
i C zj1 zi2; υ2 D

r2
i C zj2 C zi2.
A suitable application of eqn (4.73) leads also to an analytical expression for
the field strength components in the r- and z-directions (Qj individual line
charges):
Er D
n
jD1
Qj
4εzj2 zj1
ð

zj2 zi
riA1

zj1 zi
riυ1
C
zj1 C zi
riA2

zj2 C zi
riυ2

, 4.76a
Ez D
n
jD1
Qj
4εzj2 zj1

1
A1

1
υ1

1
A2
C
1
υ2

. 4.76b
Ring charges (Fig. 4.34(b)):
pij D
1
4ε
2

Kk1
˛1

Kk2
˛2

, 4.77
where
˛1 D

ri C rj2 C zi zj2, ˛2 D

ri C rj2 C zi C zj2,
ˇ1 D

ri rj2 C zi zj2, ˇ2 D

ri rj2 C zi C zj2,
and
k1 D
2
p
rjri
˛1
, k2 D
2
p
rjri
˛2
with the complete elliptic integrals of the first kind Kk and the second
kind Ek.
The field stress components become
Er D
n
jD1
Qj
4ε
1
ri



.
r2
j r2
i C zi C zj2
/
Ek1 ˇ2
1Kk1
˛1ˇ2
1

.
r2
j r2
i C zi C zj2
/
Ek2 ˇ2
2KK2
˛2ˇ2
2



4.78a
Ez D
n
jD1
Qj
4ε
2
ri
,
zi zj
-
Ek1
˛1ˇ2
1
C
,
zi C zj
-
Ek2
˛2ˇ2
2
. 4.78b

As far as the most suitable arrangement of discrete charges within an
electrode is concerned, these may either be found by optimization techniques
based upon objective functions35
or a more practical approach is by the defi-
nition of an assignment factor,34
which relates the successive distances of the
contour points with the distances between a contour point and the adjoining
corresponding charge. Details of this method may be found in the literature.
For a field space containing only one type of dielectric material ε D
constant, the application of the CSM to real three-dimensional problems does
not present fundamental difficulties. Even sophisticated electrode configura-
tions can be treated by means of discrete charges and images, at least if
types of charges with variable charge densities are used (ring charges with
periodically variable charge distribution,34
multipoles,39
elliptic cylinder
charges,37
axispheroidal charges.38
Even electric fields with even moving
space charges can be treated.34
In contrast to the simple solutions within the FDM or FEM for treating
multi-dielectric materials, the CSM when used for field calculations in systems
composed of two or more materials increases the expenditure. This may be
understood by considering the fundamental mathematical solutions and the
physical mechanisms involved. The CSM is directly based upon physical
charges and in every dielectric material polarization processes take place.
Whereas in a homogeneous material placed between electrodes the absolute
value of its permittivity does not contribute to the field strength (or poten-
tials), but only the flux density D, the field distribution at the boundaries
of different materials is heavily distorted due to the dipole charges at the
boundaries which do not have counterparts at the adjacent medium. The law
of dielectric refraction (section 4.3.2, eqn (4.36)) results from this physical
effect and is associated with an infinitely thin layer of bonded charges located
in the two media. The free surface charges physically present due to electrical
conduction of the interface surface also contribute to field distortions, but the
common dielectric refraction is not related to such additional charges.
This realignment of dipoles within different dielectric materials must thus
be considered within the CSM. An exact solution with CSM must be based
upon the physical dipole surface charge density as has been shown recently.47
But continuous surface charges can also be simulated by discrete charges by
replacing the surface charge density at metal electrodes, whose potential is a
fixed value, by discrete charges within this electrode. This method, originally
presented by P. Weiss,34
will be presented briefly through a simple example.
Figure 4.35 displays a cross-section of a part of an insulation system, in
which a metal electrode with fixed potential, D c, meets two adjoining
dielectric materials I and II. The actual shapes of the two-dimensional surfaces
of the three different boundaries (electrode–dielectric I, electrode–dielectric
II, dielectric I–dielectric II) determine the optimal types of discrete charges
simulating the problem. Thus, the localized charges 1 to 7 will represent point

Electrode
3
2
4
6
7
5
1
Charges
Diel. II
Diel.I
f = variable
f = fc
Contour points
Figure 4.35 Simulation of a dielectric boundary by discrete charges
charges as well as intersections with line or ring charges. From earlier consid-
erations it is obvious that a part of the charges (nos 1–3, denoted as nE) have
been placed inside the electrode, i.e. behind the metal surface. However, the
same is correct for the charges placed on both sides of the dielectric interface
(nos 4–7), as the influence of the dipolar charges within dielectric I upon the
field in dielectric II can be simulated by the discrete charges nos 4 and 5 within
dielectric I and vice versa. It was also shown earlier that a limited number of
contour points placed at a ‘ D constant’ boundary is necessary, which is equal
to the number of simulated charges within an electrode, and thus a number
of nE D 3 contour points (nos 1–3) is adequate. For the dielectric interface,
however, it will be sufficient for our example to place only two contour points
corresponding to the two pairs of simulation charges (nos 4 and 6, nos 5 and
7), as each contour point belongs to dielectric I as well as to dielectric II. Equal
numbers of charges, designated by nB, on both sides of the dielectric interface
are thus convenient and they should be placed at positions equally distributed
between the mutual contour points and adjacent charges respectively. For our
example, nB is only 2. Now it is possible to set up a system of equations
for our unknown charges based upon well-known boundary conditions. These
boundary conditions can be subdivided into three parts:
(1) The electrode–dielectric interface is a boundary with known potential,
D ?. The absolute magnitude of the surface charge density at this elec-
trode is, due to the polarization mechanisms in both dielectric materials,

dependent upon the relative permittivity εr of the dielectric materials, as
D D εE D εrε0E, where ε0 is the permittivity of vacuum. Also the absolute
magnitudes of our simulation charges would depend upon these materials’
constants. However, it is not necessary to take these physical effects into
account, which are indeed included within our potential coefficients (see
eqns (4.74), (4.74a), etc.). For any homogeneous dielectric material, the elec-
tric field may be computed independent of any relative permittivity εr, and
the potential coefficients, for which eqn (4.74) is one example only, are in
general always computed by assuming ε D ε0, or any other number as long
as the simulated charges are not used to derive capacitance values from the
results, which is also possible. It is easy to understand, however, that the abso-
lute magnitudes of the discrete charges used within our system are based upon
a superposition of potentials. And thus we can use the known potential at the
electrode interface to derive two sets of equations due to the two dielectrics
used. The first set of three equations based upon the three or nE contour points
takes only dielectric I into account, for which the charges within dielectric II
can be neglected:
nE
jD1
Qjpij C
nECnB
jDnEC1
Qjpij D c 4.79a
1–3 4–5
Using eqns (4.80) and (4.81) subject to two new boundary conditions, the
electric field within dielectric II could be computed, as all Qj charges within
eqn (4.87a) which are not yet known define the potentials within this material.
For the computation of the field distribution within dielectric I, the same
considerations apply. But now we neglect the charges within dielectric I, which
results in an equal set of three or nE equations, as
nE
jD1
Qjpij C
nEC2nB
jDnEC1
Qjpij D c 4.79b
1–3 6–7
(2) The potential at the dielectric interface is unknown. We know, however,
that due to the continuity of the potential at either side of the interface, the
potentials must be equal at each contour point. As the charges within the
electrode (nos 1–3) will not disturb the continuity condition, the potentials
due to the charges within the dielectric materials must satisfy the condition
nECnB
jDnEC1
Qjpij D
nEC2nB
jDnECnBC1
Qjpij 4.80
4–5 6–7

This equation refers to a number nb contour points giving an equal number of
new equations, in which those charges Qj are involved, which have not yet
been used within eqn (4.79a) or eqn (4.79b) respectively. It should be noticed
that this potential continuity condition implies that the field stress components
tangential to the interface are equal.
(3) Finally, the third boundary condition refers to the continuity of the
normal component of the electric flux density crossing the dielectric inter-
face, or the discontinuity of the normal components of the field intensity (see
eqn (4.29)). To include this condition, the ‘field intensity coefficient’ fij must
be considered (see eqn (4.73)), being the contribution of the charge j to that
component of the field vector, which is normal to the dielectric boundary at
a contour point i. These factors are in general also known from analytical
computations, as this applies to the potential coefficients pij, and specific
fij values can be taken directly from the earlier equations ((4.76) or (4.78))
for line or ring charges. Then for any normal component Eni D Qjfij, this
condition may be written as
εI


nE
jD1
Qjfij C
nEC2nB
jDnECnBC1
Qjfij

D εII


nE
jD1
Qjfij C
nECnB
jDnEC1
Qjfij

4.81
1–3 6–7 1–3 4–5
where ε, εII are the permittivities of the two dielectrics.
This equation refers again to a number of nB contour points, and thus a
total number of nE C 2nB linear equations are given for the calculations
of the same number of unknown charges. This procedure demonstrates the
difficulties involved with the implementation of dielectric boundaries, as a
significant number of additional charges increase the computational efforts.
This section will be concluded with an example of a numerical field compu-
tation based upon the CSM with surface simulation charges.45
Figure 4.36
shows the computed arrangement as well as the essential sections of the
fields computed. A cylindrical epoxy spacer εr D 3.75 with recessments
is placed between parallel electrodes shaped at the outer parts similar to
Bruce’s profile, but also recessed to reduce the field intensity at the triple point
(gas–solid–electrode interface). Outside the spacer, gas insulation (εr D 1) is
provided, and the whole system is placed within a cylindrical metal pressure
vessel with zero potential, the vessel not being shown. As the diameter of the
vessel is large compared with the diameter of the electrode system, for the
field calculation zero potential is assumed also in infinite space.
The result of the computation is displayed by a number of field stress arrows
starting at the points at which the field intensity is computed. These sites
are located at the electrode contours as well as at the interface between gas
and dielectric, for which the normal and tangential components of the field

r = 1
r = 3.75
Metal Metal
Axis
Figure 4.36 Field computation by CSM with surface charges.45
Epoxy
spacer between parallel plate electrodes
intensities are of utmost interest. The length of each arrow is proportional
to the absolute value of the field strength, and the direction of an arrow
displays the field direction at each site. No tangential components act at the
electrode–dielectric interfaces, and normal components can barely be noted
in the upper part of the spacer. This example is taken from an investigation
concerning surface charge accumulation at the gas–dielectric interface under
a d.c. voltage application. The experimental investigations showed a high
accumulation of positive or negative surface charges after applying high d.c.
voltages to the electrode for a long time, i.e. up to hours, but the polarity and
magnitude of these charges are directly related to this original electrostatic
field, i.e. the field before charge is deposited and fixed to the interface.

4.4.4 Boundary element method
The boundary element method (BEM) is a relatively new technique for
solving Laplace’s and Poisson’s equations (and other partial differential
equations).33,47,48,49,50
The unknown function u is first solved on the
boundary of the domain, the value of u and its partial derivatives are then
calculated by integration of the number of elements on the boundary. In this
way, the number of elements and thus the number of unknowns of the resulting
linear equations is greatly reduced compared with domain approaches such as
finite element and final difference methods. The boundary approach also makes
it possible to handle problems with infinite domain.
The formulation of boundary integral equations48
TheproblemistosolvePoisson’sequation(orLaplace’sequationif.x, y D 0):
r2
ux, y D .x, y x, y in D 4.82
ux, y D fx, y x, y in 1 4.83
qx, y D ∂u/∂n D gx, y x, y in 2 4.84
 D 1 C 2 4.85
where  is the boundary of the domain D, and q is the directional derivative
of ux, y with respect to the outwards normal n of the boundary.
If the solution is approximated, the residual Resx, y is not zero,
Resx, y D 2
ux, y C .x, y 6D 0 4.86
Therefore we force it to zero by a weighting function wx, y,
'
D
r2
ux, y C .x, ywx, y dx dy D 0 4.87
or
'
D
r2
ux, ywx, y dx dy D
'
D
.x, ywx, y dx dy 4.88
Integrating the left-hand side by parts twice yields
+

∂u
∂n
w ds
'
D

∂u
∂x
∂w
∂x
C
∂u
∂y
∂w
∂y

dx dy D
'
D
+


∂u
∂n
wu
∂w
∂n

ds C
'
D
ur2
w dx dy D
'
D

or
'
D
ur2
w dx dy D
+


u
∂w
∂n

∂u
∂n
w

ds
'
D
Choosing
wx, y D ln r
r D

x x02 C y y02 4.92
where r is the distance between points Px, y and P0x0,yy0, and P0 are
either in the domain D or on the boundary .
It is easy to verify that ln r is a solution to Laplace’s equation in the domain
with the singular point P0 being excluded:
r2
ln r D 0 for P 6D P0 4.93
The function ln r is called the fundamental solution of Laplace’s equation in
an infinite region.
C
C
q1
q2
P0
P0
σ
s
e
n
n
Figure 4.37 Boundary integral equation (courtesy of Ming Yu)
If P0 is located in D (as shown in Fig. 4.37), to exclude it from the domain,
P0 is encircled with a circle of very small radius ε centred at P0. The domain
and the boundary of the circle are denoted by and c respectively.
+
Cc

u
∂ ln r
∂n
q ln r

ds
'
D
Qx, y ln r ds dy
D
'
D
ur2
ln r dx dy D 0 4.94

Note that the normal vector of c is inward,
+
Cc

u
∂ ln r
∂n
q ln r

ds D
+


u
∂ ln r
∂n
q ln r

ds
C
+
c

u
∂ ln r
∂n
q ln r

ds 4.95
∂ ln r
∂n
D
∂ ln r
∂r
∂r
∂n
D
1
r
1 D
1
r
4.96
Since it is a first type line integration, it does not matter whether it is integrated
from 0 to 2 or from 2 to 0, so long as we keep ds 0. Thus if integrating
from 0 to 2 (note that if it is integrated from 2 to 0, then d5 0, ds D
rd5 D rd5, which will give the same result),
ds D rd5 4.97
+
c

u
∂ ln r
∂n
q ln r

ds D
' 2
0

u
1
ε

∂u
∂n
ln ε

ε d5 4.98
D
' 2
0

u C
∂u
∂n
ε ln ε

d5 4.99
D

uε, I C
∂uε, I
∂n
ε ln ε

2 4.100
Note that
uε, I D uε cos I C x0, ε sin I C y0 4.101
0 I 2 4.102
Let ε ! 0,  C c becomes  while D approaches D, and
lim
ε!0
ε ln ε D 0 4.103
lim
ε!0
uε, I D ux0, y0 4.104
which leads to
ux0, y0 D
1
2
+


u
∂ ln r
∂n
q ln r

ds
'
D
Qx, y ln r dx dy

4.105
This is the boundary integral equation which links the values of the unknown
function in the domain with the line integral along the boundary. If u and
q are known on the boundary, ux, y in the domain can be calculated by

eqn (4.105). Because on the boundary either u or q is known but not both, it
is necessary to find out another half of the u and q on the boundary.
Let us now locate the P0 on the boundary. It can be done similarly as in the
case when P0 was located in D (as in Fig. 4.37), except that the integral is now
taken from 51 to 52, and when ε ! 0, 52 51 D (if the boundary is smooth
at P0), therefore the boundary integral equation for P0 on the boundary is
ux0, y0 D
1

+


u
∂ ln r
∂nn
q ln r

ds
'
D
QX, Y ln r dx dy

4.106
Another approach to obtain (4.91) is to use Green’s identity:
'
D
ur2
w wr2
u dx dy D
+


u
∂w
∂n
w
∂u
∂n

ds 4.107
Since r2
u D Q, the above equation yields (4.91) naturally. From this
approach it is clear that if u and q are known exactly on the boundary,
theoretically u in the domain can be calculated exactly, not approximately
(the fundamental solution ln r is actually the Green’s function for an infinite
region).
In formulating the boundary element equations consider the case shown in
Fig. 4.37 with the boundary  discretized into elements which can be modelled
by curves or straight lines as shown in Fig. 4.38. On each element, u and q
are approximated by constant, linear, quadratic or other basis functions. For
simplicity, we will use constant elements. The boundary  is discretized into



element
non-element Si
n
2
3
1 i
i + l
Ω
node
S
Figure 4.38 Discretization of the boundary constant element

n straight line elements Si i D 1, . . . , n,
S D
n
iD1
Si 4.108
where S is the discretized boundary (refer to Fig. 4.38). That is, S is an
approximation of boundary .
On each element a node is located at the centre. The values of u and q on
the boundary are approximated as follows:
ux, y D
n
iD1
Uiix, y 4.110
qx, y D
n
iD1
Qiix, y 4.111
ix, y D
0
1 x, y in Si
0 otherwise
4.112
where Ui and Qi are nodal values of u and q. For node i, a weighting function
is defined,
wix, y D ln ri 4.113
ri D

x xi2 C y yi2 i D 1, 2, . . . , n 4.114
where n is the number of boundary elements. For every element we have the
following equation,
usi, yi D
+


u
∂ ln r
∂n
q ln r

ds 4.115
³
+
s

u
∂ ln r
∂n
q ln r

ds 4.116
D
n
jD1
'
sj

Uj
∂ ln ri
∂n
Qj ln ri

ds 4.117
uxi, yi³
n
jD1
Uj
'
sj
∂ ln ri
∂n
ds
n
jD1
Qj
'
sj
ln ri ds iD1, 2, . . . , n 4.118
or
n
jD1
HijUj D
n
jD1
GijQj i D 1, 2, . . . , n 4.119

where
Hij D
'
sj
∂ ln ri
∂n
ds i D 1, 2, . . . , n 4.120
Hij D
'
sj
∂ ln ri
∂n
ds i D 1, 2, . . . , n 4.121
Gij D
'
sj
ln ri ds i D 1, 2, . . . , n 4.122
Now there are n equations. Because half of the values of Ui and Qi are known,
there are only n unknowns. Therefore the above equations can be rearranged
to obtain a set of algebraic equations in matrix form,
Anz D b 4.123
After solving this equation, all the values of Ui and Qi on the boundary are
known. Thus the values of ux0, y0 in the domain D can be calculated by
(refer to eqn 4.105)
ux0, y0 D
1
2


n
jD1
Uj
'
sj
∂ ln r
∂n
ds
n
jD1
Qj
'
sj
ln r ds



'
D
Qx, y ln r dx dy 4.124
r D

x x02 C y y02 4.125
The partial derivatives of u can also be calculated,
∂ux, y
∂x0
D
1
2



n
jD1
'
Sj
∂
∂x0

Uj
∂ ln r
∂n
Qj ln r

ds

'
D
Qx, y
∂ ln r
∂x0
ds

4.126
∂ux, y
∂y0
D
1
2



n
jD1
'
Sj
∂
∂y0

Uj
∂ ln r
∂n
Qj ln r

ds

'
D
Qx, y
∂ ln r
∂y0
ds

4.127
Note that in the above equations x and y are integral variables, and the partial
derivatives are taken with respect to x0 and y0.

The third type boundary condition au C bq D h can be treated as follows,
aiUi C biQi D hi i D 1, 2, . . . , k 4.128
where k is the number of nodes with third type boundary condition.
Combining eqn (4.128) with eqn (4.119), there are n C k unknowns and
n C k equations which are sufficient for finding the unique solutions.
Numerical examples
In the following illustration, the symbols below are used:
N : number of points being checked on an equipotential line
Ne : number of boundary elements on a closed boundary
u : exact solution
O
u : approximates solution
O
uavg : average O
u on a equipotential line
O
umax : maximum O
O
umin : minimum O
E : exact solution of field strengh
O
Eavg : average O
E on a boundary
O
Emax : maximum O
E on a boundary
O
Emin : minimum O
E on a boundary
Note that the reference direction of E is opposite to n on the conductor
surface, thus on the conductor surface E D q. The numbers in parentheses
are percentage errors defined by,
error D
approximation exact
exact
100% 4.129
Example 1 A conductor of infinite length above the ground
A conductor with R D 1 is centred at (0.5), i.e. its height h D 5. Boundary condition:
U1 D 100 4.130
The exact solution can be obtained by the method of image (conductor 2 is the image),
ux, y D u1
lnr1/r2
ln[2h d C a/a]
4.131

where
d D

h2 R2, a D R h d.
The equipotential line with potential U0 is given by,
x2
C y y02
D b2
4.132
y0 D d
1 C K2
0
1 K2
0
4.133
b D
2dk0
1 K2
0
4.134
K0 D exp

U0
u1
ln
2h d C a
a

4.135
The boundary elements used is are 12. Comparison of results is given in Table 4.1.
Table 4.1 Comparison for Example 1 N D 1048
u(exact) 0 30 50
O
uavg (error %) 0.0010 (N/A) 29.9128 0.29% 49.8845 0.23%
O
umax (error %) 0.0017 (N/A) 29.9179 0.27% 49.9003 0.20%
O
umin (error %) 0.0007 (N/A) 29.9016 0.33% 49.8838 0.23%
N/A: not applicable.
Example 2 Coaxial cylinders
Boundary condition (subscripts 1 and 2 denote the inner and outer cylinders respec-
tively),
E0 D 144.2695041 u2 D 10 4.136
Exact solution,
ux, y D E1R1 ln
R2
r
C u2 4.137
Ex, y D E1
R1
r
4.138
r D

x2 C y2 4.139
The boundary element used is also 12. Comparison of results is given in Table 4.2

ground
image
h
x
h
conductor
R
y
R2
R1
y
Q (xi,yi)
P(x,y)
Q ′ (xi,−yi )
r1
r2
(a) Example 1 (b) Example 2
O
Figure 4.39 Conductor models for the numerical examples. (a) Conductor
above ground. (b) Two coaxial cylinders48
Table 4.2 Comparison for Example 248
10 N 32
u (exact) 51.5037 E (exact) 72.1348
O
uavg (error %) 50.8005(1.37%) O
Eavg (error %) 71.9526(0.25%
O
umax (error %) 50.8016(1.36%) O
Emax (error %) 72.0011(0.19%
O
umin (error %) 50.7995(1.37%) O
Emin (error %) 71.8905(0.34%
It can be observed that the numerical solution is very accurate even with constant
elements. The errors of potentials in the second example are greater than that of the
first example. This is due to the use of the Neumann boundary condition.
With a higher order interpolation function on the boundary, more accurate numerical
results can be achieved.48
References
1. B.D. Popovic. Introductory Engineering Electromagnetics. Addison-Wesley, 1971.
2. D. Vitkovitch. Field Analysis: Experimental and Computational Methods. D. Van Nostrand,
1966.

3. P. Moon and D.E. Spencer. Field Theory for Engineers. D. Van Nostrand, 1961; Field Theory
Handbook. Springer, 1961.
4. J.D. Kraus and K.R. Carver. Electromagnetics (2nd edn). McGraw-Hill, New York, 1973.
5. F.M. Bruce. Calibration of uniform field spark gaps for high voltage measurements at power
frequencies. Proc. IEE 94, Part II (1947), p. 138.
6. H. Prinz. Hochspannungsfelder. Oldenburg, Munich, 1969.
7. H. Steinbigler. Anfangsfeldstaerken und Ausnutzungsfaktoren rotations-symmetrischer Elek-
trodenanordnungen in Luft. Dr.-Thesis, TH Munich, 1969.
8. T. Takuma, T. Kouno and H. Matsuba. Field behaviour near singular points in composite
dielectric arrangements. Trans. IEEE EI 13 (1978), pp. 426–435.
9. P. Weiss. Fictious peaks and edges in electric fields. 3rd Int. Symp. on High Voltage Engg.,
Milan, 1979, Report 11.21.
10. K. Wechsler and M. Riccitiello. Electric breakdown of a parallel solid and liquid dielectric
system. Trans. AIEE 80, Part III (1961), pp. 365–369.
11. E.A. Cherney. High voltage flashover along a solid–liquid interface. Conf. on El. Ins. and
Diel. Phenomena, Annual Report 1970, pp. 187–190.
12. H. Kaerner and H.-J. Voss. The particle influenced breakdown of insulating surfaces in SF6
under oscillating switching impulse voltage. 3rd Int. Symp. on High Voltage Engg., Milan,
1979, Report 32.04.
13. H. Okubo, T. Amemiya and M. Honda. Borda’s profile and electrical field optimization by
using charge simulation method. 3rd Int. Symp. on High Voltage Engg., Milan, 1979, Report
11.16.
14. B.M. Weedy. Underground Transmission of Electric Power. J. Wiley, 1980.
15. K. Feser. Bemessung von Elektroden im UHV-Bereich am Beispiel von Toroidelektroden
für Spannungsteiler. ETZ-A 96 (1975), pp. 207–210.
16. S. Sangkasaad. Dielectric strength of compressed SF6 in nonuniform fields. Dr.-Thesis, ETH
Zurich, No. 5738 (1976).
17. G.W. Carter and S.C. Loh. The calculation of the electric field in a sphere-gap by means of
bipolar coordinates. Proc. IEE 106C (1959), pp. 108–111.
18. A. Roth. Hochspannungstechnik (5th edn). Springer-Verlag, Vienna/New York, 1965.
19. H. Singer. Ein Rechenverfahren von Steuerbelaegen in Durchfuehrungen und Kabelendver-
schluessen. 2nd Int. Symp. on High Voltage Engg., Zurich, 1975, Report 1.1–12.
20. M.D.R. Beasley et al. Comparative study of three methods for computing electric fields.
Proc. IEE 126 (1979), pp. 126–134.
21. G.E. Forsythe and W.R. Wasow. Finite-Difference Methods for Partial Differential
Equations. John Wiley, 1960.
22. H. Rutishauser. Vorlesungen über numerische Mathematik. Bd. 2: Differentialgleichungen
und Eigenwertprobleme. Birkhäuser Verlag, 1976.
23. K.J. Binns and P.J. Lawrenson. Analysis and Computation of Electric and Magnetic Field
Problems, (2nd edn). Pergamon Press, 1973.
24. O.C. Zienkiewicz. The Finite Element Method in Engineering Science. McGraw-Hill, London,
1971 (3rd edn 1977) (in German: Methode der finiten Elemente, Carl Hanser,1975).
25. N.A. Demerdash and T.W. Nehl. An evaluation of the methods of finite elements and finite
differences in the solution of nonlinear electromagnetic fields in electrical machines. Trans.
IEEE PAS 98 (1979), pp. 74–87.
26. P. Unterweger. Computation of magnetic fields in electrical apparatus. Trans. IEEE PAS 93
(1974), pp. 991–1002.
27. E.F. Fuchs and G.A. McNaughton. Comparison of first-order finite difference and finite
element algorithms for the analysis of magnetic fields. Part 1: Theoretical analysis, Part
II: Numerical results. Trans. IEEE PAS 101 (1982), pp. 1170–1201.
28. W. Janischewskyj and G. Gela. Finite element solution for electric fields of coronating dc
transmission lines. Trans. IEEE PAS 98 (1979), pp. 1000–1012.

29. W. Janischewskyj, P. Sarma Maruvada and G. Gela. Corona losses and ionized fields of
HVDC transmission lines. CIGRE-Session 1982, Report 3609.
30. O.W. Andersen. Finite element solution of complex potential electric fields. Trans. IEEE
PAS 96 (1977), pp. 1156–1160.
31. W. Thomson (Lord Kelvin). Reprint of Papers on Electrostatics and Magnetism. Macmillan,
London, 1872.
32. J.C. Maxwell. A Treatise on Electricity and Magnetism. (3rd edn). Clarendon Press, Oxford,
1891.
33. L. Lapidus and G.F. Pinder. Numerical Solutions of Differential Equations in Science and
Engineering. J. Wiley, New York, 1982.
34. H. Singer, H. Steinbigler and P. Weiss. A charge simulation method for the calculation of
high voltage fields. Trans. IEEE PAS 93 (1974), pp. 1660–1668.
35. A. Yializis, E. Kuffel and P.H. Alexander. An optimized charge simulation method for the
calculation of high voltage fields. Trans. IEEE PAS 97 (1978), pp. 2434–2440.
36. F. Yousef. Ein Verfahren zur genauen Nachbildung und Feldoptimierung von Elektrodensys-
temen auf der Basis von Ersatzladungen. Dr.-Thesis, Techn. Univ. Aachen, 1982.
37. S. Sato et al. Electric field calculation in two-dimensional multiple dielectric by the use of
elliptic cylinder charge. 3rd Int. Symp. on High Voltage Engg., Milan, 1979, Report 11.03.
38. S. Sato et al. Electric field calculation by charge simulation method using axi-spheroidal
charge. 3rd Int. Symp. on High Voltage Engg., Milan, 1979, Report 11.07.
39. H. Singer. Numerische Feldberechnung mit Hilfe von Multipolen. Arch. für Elektrotechnik
59 (1977), pp. 191–195.
40. H. Singer. Feldberechnung mit Oberflaechenleitschichten und Volumenleitfaehigkeit. ETZ
Archiv 3 (1981), pp. 265–267.
41. H. Singer and P. Grafoner. Optimization of electrode and insulator contours. 2nd Int. Symp.
on High Voltage Engg., Zurich, 1975, Report 1.3–03, pp. 111–116.
42. D. Metz. Optimization of high voltage fields. 3rd Int. Symp. on High Voltage Engg., Milan,
1979, Report 11.12.
43. H. Singer. Flaechenladungen zur Feldberechnung von Hochspannungs-systemen. Bull. SEV.
65 (1974), pp. 739–746.
44. J.H. McWhirter and J.J. Oravec. Three-dimensional electrostatic field solutions in a rod gap
by a Fredholm integral equation. 3rd Int. Symp. on High Voltage Engg., Milan, 1979, Report
11.14.
45. S. Sato et al. High speed charge simulation method. Trans. IEE Japan, 101 A (1981), pp. 1–8
(in Japanese).
46. A. Knecht. Development of surface charges on epoxy resin spacers stressed with direct
applied voltages. Proc. 3rd Int. Symp. on Gaseous Diel., Knoxville, 1982. (Gaseous
Dielectrics III, 1982, pp. 356–364.)
47. C.A. Brebbia, J.F.C. Telles and L.C. Wrobel. Boundary Element Techniques; Applications
Theory and Applications in Engineering. Berlin, New York, Springer, Verlag, 1984.
48. Ming Yu. The Studies of Corona and Ion Flow Fields Associated with HVDC Power Trans-
mission Lines. Ph.D. Thesis, Elec. Eng. Dpt. University of Manitoba, 1993.
49. Ming Yu and E. Kuffel. The Integration Functions with Irregularities in Boundary Element
Method for Poisson’s Equations. Pro. 6th Int. Symp. on Electromagnetic Fields, Sept. 1993.
Warsaw, Poland.
50. Ming Yu and E. Kuffel. Spline Element for Boundary Element Method. COMPUMAG 93,
Miami Fl. USA, Nov. 1993.

Chapter 5
Electrical breakdown in gases
Before proceeding to discuss breakdown in gases a brief review of the funda-
mental principles of kinetic theory of gases, which are pertinent to the study of
gaseous ionization and breakdown, will be presented. The review will include
the classical gas laws, followed by the ionization and decay processes which
lead to conduction of current through a gas and ultimately to a complete
breakdown or spark formation.
5.1 Classical gas laws
In the absence of electric or magnetic fields charged particles in weakly ionized
gases participate in molecular collisions. Their motions follow closely the
classical kinetic gas theory.
The oldest gas law established experimentally by Boyle and Mariotte states
that for a given amount of enclosed gas at a constant temperature the product
of pressure p and volume V is constant or
pV D C D const. 5.1
In the same system, if the pressure is kept constant, then the volumes
V and V0 are related to their absolute temperatures T and T0 (in K) by
Gay–Lussac’s law:
V
V0
D
T
T0
. 5.2
When temperatures are expressed in degrees Celsius, eqn (5.2) becomes
V
V0
D
273 C
273
5.3
Equation (5.3) suggests that as we approach D 273°C the volume of gas
shrinks to zero. In reality, all gases liquefy before reaching this value.
According to eqn (5.2) the constant C in eqn (5.1) is related to a given
temperature T0 for the volume V0:
pV0 D C0. 5.4

Substituting V0 from eqn (5.2) gives
pV D

C0
T0

T. 5.5
The ratio C0/T0 is called the universal gas constant and is denoted by R.
Equation (5.5) then becomes
pV D RT D C. 5.6
Numerically R is equal to 8.314 joules/°K mol. If we take n as the number of
moles, i.e. the mass m of the gas divided by its mol-mass, then for the general
case eqn (5.1) takes the form
pV D nC D nRT, 5.7
Equation (5.7) then describes the state of an ideal gas, since we assumed
that R is a constant independent of the nature of the gas. Equation (5.7) may
be written in terms of gas density N in volume V containing N1 molecules.
Putting N D NA where NA D 6.02 ð 1023
molecules/mole, NA is known as
the Avogadro’s number. Then eqn (5.7) becomes
N1
V
D N D
NA
R
p
T
or
pV D
N1
NA
RT D N1kT or p D NkT 5.8
The constant k D R/NA is the universal Boltzmann’s constant (D1.3804 ð
1023
joules/°K) and N is the number of molecules in the gas.
If two gases with initial volumes V1 and V2 are combined at the same
temperature and pressure, then the new volume will be given by
V D V1 C V2
or in general
V D V1 C V2 C V2 C . . . C Vn. 5.9
Combining eqns (5.7) and (5.9) gives
V D
n1RT
p
C
n2RT
p
C . . . C
nnRT
p
rearranging
p D
n1RT
V
C
n2RT
V
C . . . C
nnRT
V
or
p D p1 C p2 C . . . pn. 5.10

Electrical breakdown in gases 283
where p1, p2, . . . , pn denote the partial pressures of gases 1, 2, . . . , n.
Equation (5.10) is generally referred to as the law of partial pressures.
Equations (5.1) to (5.10) can be derived directly from the kinetic theory of
gases developed by Maxwell in the middle of the nineteenth century. A brief
derivation will be presented.
The fundamental equation for the kinetic theory of gas is derived with the
following assumed conditions:
1. Gas consists of molecules of the same mass which are assumed spheres.
2. Molecules are in continuous random motion.
3. Collisions are elastic – simple mechanical.
4. Mean distance between molecules is much greater than their diameter.
5. Forces between molecules and the walls of the container are negligible.
Consider a cubical container of side l D 1 m as shown in Fig. 5.1 with N1
molecules, each of mass m and r.m.s. velocity u. Let us resolve the velocity
into components, ux, uy, uz where u2
D u2
x C u2
y C u2
z . Suppose a molecule
of mass m is moving in the x-direction with velocity ux. As it strikes the
wall of container plane YZ it rebounds with the velocity ux. The change in
momentum, therefore, is
m D mux mux D 2mux.
z
l
l
x
uz
uy
ux
u
0
y
l
Figure 5.1 Resolution of molecular forces

For the cube of side l the number of collisions per second with the right-hand
wall is ux/2l, therefore
m/sec/molecule D
2muxux
2l
D
mu2
x
l
,
but the same molecule will experience the same change in momentum at
the opposite wall. Hence m/sec/molecule in the x-direction D 2mu2
x/l. For
the three-dimensional cube with total change in momentum per second per
molecule (which is the force) we obtain the force per particle as
F D
2m
l
u2
x C u2
y C u2
z D
2mu2
l
. 5.11
As kinetic energy for a particle W D 1
2
mu2
, therefore,
F D 4
W
l
.
For N1 particles the energy due to different velocities u of particles will
become the mean energy, and therefore
F D 4
N1W
l
.
Force leads to pressure p, taking into account the total area of the cube
A D 6l2

p D
F
A
D
4N1W
6l Ð l2
D
2
3
N1W
l3
5.12
with l3
D V D volume. Comparing eqns (5.8) and (5.12) leads to:
pV D 2
3 N1W.
Comparing eqn (5.12) with eqn (5.1) we note that these equations are identical
for constant temperature. Using eqn (5.8) gives
p D
2
3
N1
V
W D
2
3
NW D NkT
which leads to the expression for mean energy per molecule:
W D 3
2
kT. 5.13
5.1.1 Velocity distribution of a swarm of molecules
It has been shown using probability consideration that the distribution of
molecular velocities depends on both the temperature and the molecular weight

of the gas. The mathematical analysis shows the most probable velocity is
neither the average nor the r.m.s. velocity of all the molecules.
The velocity u of gas molecules or particles has a statistical distribution and
follows the Boltzmann–Maxwell distribution given by the expression1Ł
fu du D
dNu
N
D
4
p

u
uP
2
eu/uP2
du
uP
5.14
where up is the most probable velocity and dNu/N the relative number of
particles whose instantaneous velocities lie in the range u/up and u C du/up.
Let
f

u
up

D
dNu
N

du
up
and
ur D
u
up
(relative velocity).
Introducing this dimensionless variable into eqn (5.14) gives the function
representing velocity distribution
fur D
4
p

u2
r eu2
r 5.14a
with
dNu
N
D fur dur.
The distribution function corresponding to eqn (5.14a) is shown in Fig. 5.2.
It should be noted that the function is asymmetrical about the most probable
velocity up. A greater number of particles has a velocity higher than up. The
average velocity u is obtained from integrating ur from 0 to 1.
ur D
1
ur D0
urfur dur D
4
p

1
0
u3
r eu2
r dur

1/2
D
2
p

;
or
u D urup D 1.128up 5.15

1
0.5
0 1 2
ur =
up
ueff
up
u
u
3 4
4
π
1
e
= 0.83
Figure 5.2 Distribution of velocities (up most probable; u average; ueff
effective or r.m.s.)
The r.m.s. or effective value of velocity is obtained by squaring ur and
obtaining the average square value
ur2
eff D
1
ur D0
u2
r fur dur D
4
p

1
0
u4
r eu2
r

3/8
p

dur D
3
2
ueff D ureffup D
3
2
up D 1.224up. 5.16
The mean kinetic energy of the particle given by eqn (5.13) relates its effective
velocity to the temperature (1
2 mu2
eff D 3
2 kT) and we obtain
ueff D
3kT
m
; u D
8kT
m
; up D
2KT
m
. 5.17
Hence the respective velocities remain in the ratio up : u : ueff D 1 : 1.128 :
1.224.
It should be noted that the foregoing considerations apply only when the
molecules or particles remain in thermal equilibrium, and in the absence of
particle acceleration by external fields, diffusion, etc. If the gas contains elec-
trons or ions or other atoms that are at the same temperature, the average
particle energy of such mixture is
1
2 mu2
eff D 1
2 meu2
eeff D 1
2 miu2
ieff D . . . D 3
2 kT 5.18
where m, mi, me are the respective masses of gas molecules, ions, electron,
and ueff, uieff, ueeff are their corresponding velocities.

Table 5.1 Mean molecular velocities at 20°C and 760 Torr2
Gas Electron H2 O2 N2 Air CO2 H2O SF6
(vapour)
u (m/sec) 100 ð 103
1760 441 470 465 375 556 199
The values of the mean molecular velocities calculated for 20°C and
760 Torr for several of the common gases are included in Table 5.1.
5.1.2 The free path l of molecules and electrons
Knowledge of dependency and distribution of free paths # may explain (with
restrictions) the dependency of ˛ D fE, N discussed later, even assuming a
simple ‘ballistic’ model. For this reason a short treatment of free paths will be
presented. The free path # is defined as the distance molecules or particles
travel between collisions. The free path is a random quantity and as we shall
see its mean value depends upon the concentration of particles or the density
of the gas.
To derive the mean free path # assume an assembly of stationary molecules
of radius r1, and a moving layer of smaller particles of radius r2 as particles
move, their density will decrease as shown in Fig. 5.3. As the smaller particles
move, their density will decrease due to scattering caused by collisions with
gas molecules. If we assume that the moving particles and molecules behave
as solid spheres, then a collision will occur every time the centres of two
particles come within a distance r1 C r2. The area for collision presented by a
x dx
2r1
2r2
r1
r2
Figure 5.3 Model for determining free paths

molecule is then r1 C r22
and in a unit volume it is Nr1 C r22
. This is
often called the effective area for interception where N D number of particles
per unit volume of gas.
If we consider a layer of thickness dx, distant x from the origin (Fig. 5.3)
and nx the number of particles that survived the distance x, then the decrease
in the moving particles due to scattering in layer dx is
dn D nxNr1 C r22
dx.
Assuming the number of particles entering (at x D 0) is n0, integration gives
nx D N0 eNr1Cr22x
. 5.19
The probability of free path of length x is equal to the probability of collisions
between x and x C dx. The mean free path # D x is obtained as follows.
Differentiating eqn (5.19) we obtain
fx D
dn
n0
D Nr1 C r22
eNr1Cr22
x
dx.
For the mean free path
x D # D
1
xD0
xfx dx
D Nr1 C r22
1
xD0
x eNr1Cr22x
dx
D
1
Nr1 C r22
5.20
The denominator in eqn (5.20) has the dimensions of area and the value
r1 C r22
is usually called the cross-section for interception or simply colli-
sion cross-section and is denoted by :
D
1
N#
5.21
We shall see later that the collisions between the incoming particles and the
stationary molecules may lead to processes such as ionization, excitation,
attachment, etc.
If we put in eqn (5.21) Q D N, then Q will represent the effective cross-
section presented by molecules or particles in unit volume of gas for all
collisions for density of N molecules/volume. If, for example, only a fraction
Pi of collisions between the incoming particles and the gas particles leads to
ionization then Pi is the probability of ionization. Thus if only ionizing colli-
sions are counted, the molecules present an effective area of only PiQ D Qi;

Qi is the effective cross-section for ionization. Similarly for other processes,
excitation Qe, photoionization Qph attachment Qa, etc., including elastic
collisions can be taken into account
Q D Qelastic C Qi C Qe C Qa C . . . 5.22
Atomic cross-sections for different processes vary over a wide range. For
ionization they can rise to some 2 ð 1016
cm2
, but for collisions resulting in
nuclear reactions they may be 1024
cm2
or less.
In deriving the expression (5.20) it was assumed that the struck molecules
were stationary, i.e. the molecules of gas 2 had no thermal velocity. In reality
this is not true. It can be shown that the expression giving the collisional
cross-section must be still multiplied by a factor
* D 1 C
m1
m2
with m1 and m2 the mass of each gas component. In a gas mixture the colli-
sional cross-section of particles of type 1 of gas m1, r1, N1 becomes equal
to the sum of all collisional cross-sections of the other particles of types of
gas m2, m3, . . . , r2, r3, . . . , N2, N3, . . .. Thus the mean free path of particles
of type 1 is
#1 D
1

n
iD1
Nir1 C ri2
1 C
m1
mi
5.23
For an atom in its own gas r1 D r2 D r; u1 D u2. Then
#a D
1
4
p
2r2
N
. 5.24
For an electron in a gas r1 − r2 and m1 − m2 eqn (5.23) gives
#e D
1
r2
2N
or
#e D 4
p
2#a D 5.66#a. 5.25
Table 5.2 shows examples of mean free path (gas) for gases of different molec-
ular weight.
From eqn (5.8), N D p/kT, it follows that the mean free path is directly
proportional to temperature and inversely as the gas pressure
#p, T D #0
p0
p
T
T0
. 5.26

Table 5.2 Mean free paths measured at 15°C and 760 Torr2
Type of gas H2 O2 N2 CO2 H2O Dimensions
# 11.77 6.79 6.28 4.19 4.18 108
m
Molecular
weight 2.016 32.00 28.020 44.00 18.00
Considering a typical practical case with values for average velocity of gas
u ³ 500 m/sec and the mean free path # ³ 107
m we obtain the number of
collisions per second:
+ D
u
#
D 5 ð 109 1
sec
³ 5 collisions/nsec.
The average time between two collisions
t D
1
+
D
1
5 ð 109
D 0.2 nsec.
5.1.3 Distribution of free paths
In the earlier sections it was shown that molecular collisions are random events
and these determine free paths. Hence, free path is a random quantity and will
have a distribution about a mean value. For the system in Fig. 5.3 the mean
free path is given by eqn (5.20)
# D
1
Nr1 C r22
,
N being the gas density and r1 and r2 the radii of the two types of particles.
The distribution function of free paths is obtained from eqn (5.19)
n
n0
dn D
x
xD0
ln
dx
#
or
nx D n0 ex/#
5.27
where nx D number of molecules reaching a distance x without collision,
dn D number of molecules colliding thereafter within a distance dx, n0 D
total number of molecules at x D 0. Equation (5.27) is plotted in Fig. 5.4.
It is seen that the percentage of molecules that survive collisions is only
37 per cent. The exponent in eqn (5.27) may also be written in terms of

n(x)
n0
0.37
0
0 1
x
λ =
λ
λ
n0
n
1
N . π (r1 + r2)2
Figure 5.4 Distribution of free paths
collision cross-sections defined by eqn (5.21), to represent absorption or decay
of particles along the path x or
n D n0eNx
5.28
where may include photoabsorption, attachment, etc.
5.1.4 Collision-energy transfer
The collisions between gas particles are of two types: (i) elastic or simple
mechanical collisions in which the energy exchange is always kinetic, and
(ii) inelastic in which some of the kinetic energy of the colliding particles is
transferred into potential energy of the struck particle or vice versa. Examples
of the second type of collisions include excitation, ionization, attachment, etc.,
which will be discussed later.
To derive an expression for energy transfer between two colliding particles,
let us consider first the case of an elastic collision between two particles2
of masses m and M. Assume that before collision the particle of large mass
M was at rest and the velocity of the smaller particle was u0 in the direction
shown in Fig. 5.5. After collision let the corresponding velocities be u1 and
V, the latter along line of centres as shown. is the incidence angle and is
the scattering angle.
The fractional energy loss by the incoming particle during a collision at an
angle is then given by
 D
u2
0 u2
1
u2
0

. 5.29

mu0
mu1
MV
q
dq
y
Figure 5.5 Energy transfer during elastic collision
Since the collision is assumed to be kinetic, the equations for conservation of
momentum and energy are
mu0 mu1 cos D MV cos 5.30
mu1 sin D MV sin 5.31
1
2
mu2
0 1
2
mu2
1 D 1
2
MV2
. 5.32
Squaring eqns (5.30) and (5.31) and adding and combining with eqn (5.32)
we obtain
V D
2mu0 cos
M C m
.
Rearranging eqn (5.32) and combining with eqn (5.29) gives
 D
MV2
mu2
0
D
4mM cos2

m C M2
. 5.33
To obtain the mean fractional energy loss per collision, let P be the proba-
bility of a collision at an angle of incidence between and C d. The total
area presented for collision is r1 C r22
. The probability of a collision taking
place between and C d is the ratio of the projected area (Fig. 5.5) to the
whole area or
P d D

2r1 C r2 sin cos d
r1 C r22
D sin 2 d

for 0
for /2
.
The mean fractional loss of energy per collision allowing for collisions at all
angles is
 D
/2
0
P d
/2
0
P d. 5.34

Using eqns (5.33) and (5.34), we obtain
 D
2mM
m C M2
. 5.35
If we consider the case when the incoming particle is an ion of the same
mass as the struck particle, then m D M and eqn (5.35) gives  D 1
2
which
indicates a high rate of energy loss in each elastic collision. On the other
hand, if the incoming particle is an electron, then m − M and eqn (5.35)
gives  D 2m/M. The average fraction of energy lost by an electron in an
elastic collision is therefore very small. For example, if we consider the case
of electrons colliding with He gas atoms, the average fractional energy loss
per collision  is 2.7 ð 104
and in argon it is 2.7 ð 105
. Thus electrons
will not readily lose energy in elastic collisions whereas ions will.
Let us now consider the case when part of the kinetic energy of the incoming
particle is converted into potential energy of the struck particle. Then applying
the laws of energy and momentum conservation we obtain
1
2
mu2
0 D 1
2
mu2
1 C 1
2
MV2
C Wp 5.36
mu0 D mu1 C MV 5.37
where Wp is the increase in potential energy of the particle of mass M initially
at rest. Substituting eqn (5.37) into eqn (5.36) and rearranging we obtain
Wp D
1
2

mu2
0 u2
1
m2
M
u0 u12

. 5.38
For the conditions of constant kinetic energy of the incoming particles,
differentiation of eqn (5.38) with respect to u1 gives the maximum energy
transfer when
dWpmax
du
D 0
or
u1
u0
D
m
m C M
. 5.39
Equation (5.39) shows that the potential energy gained from the incident
particle reaches a maximum value when the ratio of its final to initial velocity
equals the ratio of its mass to the sum of masses of the individual particles.
When the colliding particles are identical, the maximum kinetic to potential
energy transfer occurs when u1 D u0/2. On the other hand, if the colliding
particle is an electron of mass m − M the maximum energy transfer corre-
sponds to u1 m/Mu0 which means that the new velocity u1 becomes only
a small fraction of the original velocity.

For the case when the target particle was initially at rest, the maximum
amount of potential energy gained will be given by the expression obtained
by inserting the value of velocity u1 from eqn (5.39) into eqn (5.38) or
Wpmax D
M
m C M
mu2
0
2
. 5.40
For an electron m − M, eqn (5.40) becomes
Wpmax
¾
D
1
2
mu2
0
2
5.41
or almost all its kinetic energy is converted into potential energy. Thus we
shall see later that electrons are good ionizers of gas, while ions are not. To
cause ionization the incoming electron must have a kinetic energy of at least
1
2
mu2
0 ½ eVi, where Vi is the ionization potential of the atom or molecule.
5.2 Ionization and decay processes
At normal temperature and pressure gases are excellent insulators. The conduc-
tion in air at low field is in the region 1016
1017
A/cm2
.* This current
results from cosmic radiations and radioactive substances present in earth and
the atmosphere. At higher fields charged particles may gain sufficient energy
between collisions to cause ionization on impact with neutral molecules. It
was shown in the previous section that electrons on average lose little energy
in elastic collisions and readily build up their kinetic energy which may be
supplied by an external source, e.g. an applied field. On the other hand, during
inelastic collisions a large fraction of their kinetic energy is transferred into
potential energy, causing, for example, ionization of the struck molecule.
Ionization by electron impact is for higher field strength the most impor-
tant process leading to breakdown of gases. The effectiveness of ionization by
electron impact depends upon the energy that an electron can gain along the
mean free path in the direction of the field.
If #e is the mean free path in the field direction of strength E then the average
energy gained over a distance # is W D eE#e. This quantity is proportional
to E/p since #e / 1/p (eqn (5.26)). To cause ionization on impact the energy
W must be at least equal to the ionization energy of the molecule eVi. Elec-
trons with lower energy than eVi may excite particles and the excited particles
* The figure 1016 1017 A/cm2
correlates with the current flowing to the whole surface of
earth (due to natural electric field). This current is 1000–1200 A. With earth surface of about
5 ð 1019 cm2, we get
j D
I
surface
D
1000
5 ð 1019
D 0.2 ð 1016
A/cm2
.

on collision with electrons of low energy may become ionized. Furthermore,
not all electrons having gained energy W ½ eVi upon collision will cause
ionization. This simple model is not applicable for quantitative calculations,
because ionization by collision, as are all other processes in gas discharges, is a
probability phenomenon, and is generally expressed in terms of cross-section
for ionization defined as the product Pi D i where Pi is the probability
of ionization on impact and is the molecular or atomic cross-sectional
area for interception defined earlier. The cross-section i is measured using
monoenergetic electron beams of different energy. The variation of ionization
cross-sections for H2, O2, and N2 with electron energy is shown in Fig. 5.6.3
It is seen that the cross-section is strongly dependent upon the electron energy.
At energies below ionization potential the collision may lead to excitation of
the struck atom or molecule which on collision with another slow moving
electron may become ionized. This process becomes significant only when
densities of electrons are high. Very fast moving electrons may pass near
an atom without ejecting an electron from it. For every gas there exists an
optimum electron energy range which gives a maximum ionization probability.
102
101
102
101
O2
N2
H2
Electron energy (eV)
s
i
-ionization
cross
section
(×
10
−20
m
2
)
Figure 5.6 Variation of ionization cross-sections for O2 , N2 , H2 with
electron energy
5.2.1 Townsend first ionization coefficient
In the absence of electric field the rate of electron and positive ion genera-
tion in an ordinary gas is counterbalanced by decay processes and a state of
equilibrium exists. This state of equilibrium will be upset upon the application

of a sufficiently high field. The variation of the gas current measured between
two parallel plate electrodes was first studied as a function of the applied
voltage by Townsend.4
Townsend found that the current at first increased proportionately with the
applied voltage and then remained nearly constant at a value i0 which corre-
sponded to the background current (saturation current), or if the cathode was
irradiated with a u.v. light, i0 gave the emitted photocurrent. At still higher
voltage the current increased above the value i0 at an exponential rate. The
general pattern of the current–voltage relationship is shown schematically in
Fig. 5.7.
The increase in current beyond V2 Townsend ascribed to ionization of the
gas by electron collision. As the field increases, electrons leaving the cathode
are accelerated more and more between collisions until they gain enough
energy to cause ionization on collision with gas molecules or atoms.
V1
i0
V2
V3
V4
Voltage
Current
Figure 5.7 Current–voltage relationship in prespark region
To explain this current increase Townsend introduced a quantity ˛, known
as Townsend’s first ionization coefficient, defined as the number of electrons
produced by an electron per unit length of path in the direction of the field.
Thus if we assume that n is the number of electrons at a distance x from the
cathode in field direction (Fig. 5.8) the increase in electrons dn in additional
distance dx is given by
dn D ˛ndx.
Integration over the distance d from cathode to anode gives
n D n0e˛d
5.42

where n0 is the number of primary electrons generated at the cathode. In terms
of current, with I0 the current leaving the cathode, eqn (5.42) becomes
I D I0e˛d
. 5.43
The term e˛d
in eqn (5.42) is called the electron avalanche and it represents
the number of electrons produced by one electron in travelling from cathode
to anode. The electron multiplication within the avalanche is shown diagram-
matically in Fig. 5.8.
Anode
Cathode
dx
E
d
(a) (b)
n0
nx
λi
x
Figure 5.8 Schematic representation of electron multiplication (a) gap
arrangement, (b) electron avalanche
The increase of current (avalanche growth) shown in the diagram
(Fig. 5.8(b)) would be I D I0ek
, with k D number of ionizing steps k D x/#i.
The transition for infinitely small values of
dx

lim #i
#i ! dx

leads to the expression e˛x
.
The quantity ˛, although a basic quantity describing the rate of ionization
by electron collision, cannot be readily calculated from the measured cross-
section for ionization i. The latter is determined for monoenergetic electrons
and calculation of ˛ from value of i is only possible when the electron energy
distribution in the gas is known. For ‘swarm’ conditions Raether5
derived a
relationship between ˛ and i, which is of the form
˛
N
D
1
ue
1
0
vivfv dv
with N the concentration, molecules/atoms, fvth the distribution of veloci-
ties of electrons, and ue the drift velocity of electrons in the field direction.
A simple derivation is possible for simple gases (non-attaching) using the
Clausius distribution of free paths (Fig. 5.4) and applying it to electrons.
We have seen that at a constant temperature for a given gas the energy
distribution W depends only on the value E/p. Also for a given energy

distribution the probability of an ionization occurring will depend on the gas
density or pressure.
Therefore, we can write
˛ D pf

E
p

or
˛
p
D f

E
p

. 5.44
Equation (5.44) describes a general dependence of ˛/p upon E/p which has
been confirmed experimentally.
A derivation of expression for this dependence is possible for simple gases,
using the Clausius distribution (eqn (5.27)) for free paths applied to electrons.
This means that we assume that this distribution will not be altered by the
additional velocity of electrons in field direction. Then all electrons which
acquire energy W ½ eVi, where Vi is the ionization potential, will ionize
the gas. These electrons have travelled a distance x, and using eqn (5.27) the
fraction of electrons with paths exceeding a given value x is
f0
x D ex/#
.
Therefore, only with a very small probability electrons can gain high energies
if they reach long distances.
The number of successful collisions – the ionization coefficient ˛ – is
clearly related to this distribution, and is certainly directly proportional to
the decay of collisions in the intervals between x and x C dx, or
˛ D
df0
x
dx
D
1
#
e#i/#
5.44a
where #i D x is the ionizing free path. The above treatment assumes #E D #,
i.e. the velocity distribution is not altered by the additional velocity of electrons
in the field direction. In reality there is a difference between # and #E as shown
below.
Hence
v
#
D v
#E
and #E D 1
nE
λ1
λ2
λ3
λ4
λE3
λE2
λE1
v being the electron drift velocity.
Then eqn (5.44) when corrected for field drift velocity becomes
˛ D nE e#i/#E
D
1
#E
e#i/#E
. . . . 5.45

Using eqn (5.21), with i as true cross-section for ionization and N the gas
density, we obtain
# D
1
Ni
.
Introducing from eqn (5.8) N D p/kT, for a gas pressure p the mean free path
becomes
# D
kT
pi
.
If in addition we put #i D Vi/E, then
˛ D
df0
x
dx
D
pi
kT
eVi/Epi/kT
, 5.46
or
˛
p
D
i
kT
ei/kT[Vi/E/p]
D AT e[BT/E/p]
5.47
where
AT D
i
kT
; BT D
Vii
kT
. 5.47a
It cannot be expected that the real dependence of ˛/p upon E/p agrees
with measured values within the whole range of E/p, because phenomena
which have not been taken into account are influencing the ionization rate.
However, even with constant values of A and B, eqn (5.47) determines the
ionization process within certain ranges of E/p. Therefore, for various gases
the ‘constants’ A and B have been determined experimentally and can be found
in the literature.6
Some of these experimental values for several of the more common gases
are listed in Table 5.3.
Table 5.3 Ionization constants A and B T D 20°C
Gas A B E/p range Vi
ion pairs V cm1
V cm1
Torr1
volts
cm1
Torr1
Torr1
H2 5 130 150–600 15.4
N2 12 342 100–600 15.5
air 15 365 100–800 –
CO2 20 466 500–1000 12.6
He 3 34 20–150 24.5
Hg 20 370 200–600 –

The constants A and B in eqn (5.47a), as derived from kinetic theory, rarely
agree with the experimentally determined values. The reasons for this disagree-
ment lies in the assumptions made in our derivations. We assumed that every
electron whose energy exceeds eVi will automatically lead to ionization. In
reality the probability of ionization for electrons with energy just above the
ionization threshold is small and it rises slowly to a maximum value of
about 0.5 at 4 to 6 times the ionization energy. Beyond that it decreases.
We have also assumed that the mean free path is independent of electron
energy which is not necessarily true. A rigorous treatment would require taking
account of the dependence of the ionization cross-section upon the electron
energy.
Using the experimental values for the constants A and B for N2 and H2 in
eqn (5.47), the graphical relationship between the parameters ˛/p and E/p
has been plotted in Fig. 5.9. The values have been corrected to T D 0°C.
4
3
2
1
0 100 300 500
N2 H2
a/
p
0
(cm
torr)
−1
E/p0 (volts/cm torr)
Figure 5.9 Dependence of ˛/p on E/p in N2 and H2 , reduced to 0°C
It should be noted that theoretically ˛/p begins at zero value of E/p, which
follows from the distribution of free paths which have values from 0 to 1. In
practice in many gases attachment q will also be present, and at low values of
E/p it is difficult to obtain the values for ‘real’ ˛ and for ‘real’ *. Experimental
measurements yield the ‘effective’ ionization coefficient ˛ D ˛ *. In this
case ˛p begins at a finite value of E/p corresponding to the lowest breakdown
strength.
Numerous measurements of ˛ in various gases have been made by
Townsend4
and subsequent workers and the data can be found in the
literature.7–9
The Geballe and Harrison’s data are included in Table 5.5.

5.2.2 Photoionization
Electrons of lower energy than the ionization energy eVi may on collision
excite the gas atoms to higher states. The reaction may be symbolically
represented as A C e C K energy ! AŁ
C e; AŁ
! A C hv; AŁ
represents the
atom in an excited state. On recovering from the excited state in some
107
–1010
sec, the atom radiates a quantum of energy of photon hv which
in turn may ionize another atom whose ionization potential energy is equal
to or less than the photon energy. The process is known as photoionization
and may be represented as A C hv ! AC
C e, where A represents a neutral
atom or molecule in the gas and hv the photon energy. For ionization to
occur hv ½ eVi or the photon wavelength # c0h/eVi, c0 being the velocity
of light and h Planck’s constant. Therefore, only very short wavelength light
quanta can cause photoionization of gas. For example, the shortest wavelength
radiated from a u.v. light with quartz envelope is 145 nm, which corresponds
to eVi D 8.5 eV, lower than the ionization potential of most gases.
The probability of photon ionizing a gas or molecule is maximum when
hv eVi is small (0.1–1 eV). Photoionization is a secondary ionization
process and may be acting in the Townsend breakdown mechanism and
is essential in the streamer breakdown mechanism and in some corona
discharges. If the photon energy is less than eVi it may still be absorbed by
the atom and raise the atom to a higher energy level. This process is known
as photoexcitation.
5.2.3 Ionization by interaction of metastables with atoms
In certain elements the lifetime in some of the excited electronic states extends
to seconds. These states are known as metastable states and the atoms in these
states are simply referred to as metastables represented by Am
. Metastables
have a relatively high potential energy and are therefore able to ionize neutral
particles. If Vm
, the energy of a metastable Am
, exceeds Vi, the ionization
of another atom B, then on collision ionization may result according to the
reaction
Am
C B ! AC
C B C e.
For Vm
of an atom Am
Vi of an atom B the reaction may lead to the exciting
of the atom B which may be represented by Am
C B ! A C BŁ
.
Another possibility for ionization by metastables is when 2Vm
for Am
is
greater than Vi for A. Then the reaction may proceed as
Am
C Am
! AC
C A C e C K.E.
This last reaction is important only when the density of metastables is high.

Another reaction may follow as
Am
C 2A ! AŁ
2 C A;
AŁ
2 A C A C hv.
The photon released in the last reaction is of too low energy to cause ionization
in pure gas, but it may release electrons from the cathode.
Ionization by metastable interactions comes into operation long after exci-
tation, and it has been shown that these reactions are responsible for long time
lags observed in some gases.10
It is effective in gas mixtures.
5.2.4 Thermal ionization
The term thermal ionization, in general, applies to the ionizing actions of
molecular collisions, radiation and electron collisions occurring in gases at
high temperature. If a gas is heated to sufficiently high temperature many of the
gas atoms or molecules acquire sufficiently high velocity to cause ionization
on collision with other atoms or molecules. Thermal ionization is the principal
source of ionization in flames and high-pressure arcs.
In analysing the process of thermal ionization, the recombination between
positive ions and electrons must be taken into account. Under thermodynamic
equilibrium conditions the rate of new ion formation must be equal to the rate
of recombination. Using this assumption Saha11
derived an expression for the
degree of ionization in terms of the gas pressure and absolute temperature
as follows:
2
1 2
D
1
p
2me3/2
h
kT5/2
ewi/kT
or
2
1 2
D
2.4 ð 104
p
T5/2
ewi/kT
5.48
where p is the pressure in torr, Wi the ionization energy of the gas, k Boltz-
mann’s constant, the ratio of ni/n, and ni the number of ionized particles
of total n particles. The strong dependence of on temperature in eqn (5.48)
shows that the degree of ionization is negligible at room temperature. On
substitution of values Wi, kT, p and T in eqn (5.48) we find that thermal
ionization becomes significant for temperatures above 1000 K.
5.2.5 Deionization by recombination
Whenever there are positively and negatively charged particles present, recom-
bination takes place. The potential energy and the relative kinetic energy of the

recombining electron–ion is released as quantum of radiation. Symbolically
the reaction may be represented as
or
AC
C e ! A C hv
AC
C e ! Am
C hv

radiation
recombination
Alternatively a third body C may be involved and may absorb the excess
energy released in the recombination. The third body C may be another heavy
particle or electron. Symbolically
AC
C C C e ! AŁ
C C ! A C C C hv
or
AC
C e C e ! AŁ
C e ! A C e C hv.
At high pressures, ion–ion recombination takes place. The rate of recom-
bination in either case is directly proportional to the concentration of both
positive ions and negative ions. For equal concentrations of positive ions nC
and negative ions n the rate of recombination
dnC
dt
D
dn
dt
D ˇnCn 5.49
where ˇ is a constant known as the recombination rate coefficient.
Since nC ³ n D ni and if we assume at time t D 0: ni D ni0 and at time
t: ni D nit, then eqn (5.49) becomes
dni
dt
D ˇ2
i .
Integration gives
ni
ni0
dni
ni
2
D ˇ
t
0
dt
or
nit D
ni0
1 C ni0ˇt
. 5.50
The half-time duration, during which time the concentration of ions has
decreased to half its original value, is given by
tn D
1
ni0ˇ
. 5.51

3
2
1
0
102
103
p (torr)
b
(10
−6
cm
3
/s)
104
Figure 5.10 Recombination coefficient (ion–ion) in air at 20°C
The variation of the recombination rate coefficient ˇ with pressure in air is
shown in Fig. 5.10. The recombination process is particularly important at
high pressures for which diffusion is relatively unimportant.
5.2.6 Deionization by attachment – negative ion formation
Electron affinity
Certain atoms or molecules in their gaseous state can readily acquire a free
electron to form a stable negative ion. Gases, whether atomic or molecular,
that have this tendency are those that are lacking one or two electrons in
their outer shell and are known as electronegative gases. Examples include
the halogens (F, Cl, Br, I and At) with one electron missing in their outer
shell, O, S, Se with two electrons deficient in the outer shell.
For a negative ion to remain stable for some time, the total energy must
be lower than that of an atom in the ground state. The change in energy that
occurs when an electron is added to a gaseous atom or molecule is called the
electron affinity of the atom and is designated by Wa. This energy is released
as a quantum or kinetic energy upon attachment. Table 5.4 shows electron
affinities of some elements.
There are several processes of negative ion formation:
(1) The simplest mechanism is one in which the excess energy upon attach-
ment is released as quantum known as radiative attachment. This process
is reversible, that is the captured electron can be released by absorption of a
photon known as photodetachment. Symbolically the process is represented as:
A C e , A
C hv Wa D hv.

Table 5.4 Electron affinities of some
elements
Element Ion formed Wa (kJ/mole)
H H
72
O O
135
F F
330
Cl Cl
350
Br Br
325
I I
295
(2) The excess energy upon attachment can be acquired as kinetic energy of
a third body upon collision and is known as a third body collision attachment,
represented symbolically as:
e C A C B ! A
C B C Wk Wa D Wk.
(3) A third process is known as dissociative attachment which is predom-
inant in molecular gases. Here the excess energy is used to separate the
molecule into a neutral particle and an atomic negative ion, symbolically
expressed as:
e C AB , AB
Ł , A
C B.
(4) In process (3) in the intermediate stage the molecular ion is at a higher
potential level and upon collision with a different particle this excitation energy
may be lost to the colliding particle as potential and/or kinetic energy. The
two stages of the process here are:
e C AB , AB
Ł
AB
Ł CA , AB
C A C Wk C Wp.
Other processes of negative ion formation include splitting of a molecule into
positive and negative ions upon impact of an electron without attaching the
electron:
e C AB , AC
C B
C e
and a charge transfer following heavy particle collision, yielding an ion pair
according to:
A C B ! AC
C B
.

All the above electron attachment processes are reversible, leading to electron
detachment.
The process of electron attachment may be expressed by cross-section for
negative ion formation A in an analogous way to ionization by electron
impact. Typical examples of the variation of attachment cross-section with
electron energy for processes (2) and (3) measured in SF6 and CO2 are shown
in Figs 5.11 and 5.12 respectively.
1.0
0.8
0.6
s
A
(
×
10
−19
m
2
)
0.4
0.2
0
1 2
0.2 0.4 0.6
Electron energy E (eV)
0.8 1.0 1.2
SF5 (×100 approx.)
−
Figure 5.11 Variation of attachment cross-section with electron energy in
SF6 . 1. Radiative attachment. 2. Dissociative attachment
5
3
1
0 2 4 6 8 10
Electron energy E(eV)
4.4 eV
8.2 eV
s
A
×
10
−23
m
2
Figure 5.12 Variation of electron attachment cross-section with electron
energy in CO2 (both peaks O
)
Cumulatively the process of electron attachment describing the removal of
electrons by attachment from ionized gas by any of the above processes may
be expressed by a relation analogous to the expression (5.43) which defines
electron multiplication in a gas. If * is the attachment coefficient defined by
analogy with the first Townsend ionization coefficient ˛, as the number of

attachments produced in a path of a single electron travelling a distance of
1 cm in the direction of field, then the loss of electron current in a distance
dx due to this cause is
dI D *I dx
or for a gap of length d with electron current I0 starting at cathode
I D I0e*d
5.52
Several methods for the measurements of the attachment coefficient have been
described in the literature.12
Methods for the determination of attachment
coefficient utilizing eqn (5.52) rely on the measurement of the surviving elec-
tronic current13
at grids distance d apart inserted at two points along the
path of the current between the electrodes. Such methods are applicable only
at relatively low values of E/p when ionization by electron collision can be
neglected. At higher values of E/p it becomes necessary to measure both the
ionization coefficient ˛ and the attachment coefficient * simultaneously.
If the processes of electron multiplication by electron, collision and electron
loss by attachment are considered to operate simultaneously, then neglecting
other processes the number of electrons produced by collision in distance dx is
dni D n˛ dx
where x is the distance from the cathode. At the same time the number of
electrons lost in dx by attachment is
dnA D n* dx
so that the number of electrons still free is
dn D dni C dnA D n˛ * dx.
Integration from x D 0 to x with n0 electrons starting from the cathode gives
the number of electrons at any point in the gap as
n D n0e˛*X
5.53
The steady state current under such conditions will have two components,
one resulting from the flow of electrons and the other from negative ions. To
determine the total current we must find the negative ion current component.
We note that the increase in negative ions in distance dx is
dn D n* dx D n0*e˛*
dx.
Integration from 0 to x gives
n D
*0*
˛ *
[e˛*x
1]

The total current equals the sum of the two components or
n C n
n0
D
˛
˛ *
e˛*d

*
˛ *
5.54
and the expression for current becomes
I D I0

˛
˛ *
e˛*d

*
˛ *

. 5.55
In the absence of attachment when * is zero the expression (5.55) reduces to
the form i D i0e˛d
and the log i d plot of eqn (5.55) gives a straight line,
with ˛ representing the slope. When the value of * is appreciable, there may be
a decrease in currents, especially at large values of d, such that the log i against
d curve drops below the straight line relation. The departure from linearity in
plotting log i against d gives a measure of the attachment coefficient. Several
workers9
have used this method for determining the ˛ and * coefficients. The
results obtained by this method by Geballe and Harrison for ionization ˛ and
attachment * in oxygen and in air are included in Table 5.5. It is convenient to
represent the observed ionization coefficient by a single coefficient ˛ D ˛ *
defined as the effective ionization coefficient.
As electron attachment reduces electron amplification in a gas, gases with a
high attachment coefficient such as sulphur hexafluoride or freon have much
higher dielectric strength than air or nitrogen. The measured data for ionization
and attachment coefficients for SF6 are included in Table 5.6. These gases are
technically important and are widely used as insulating medium in compact
h.v. apparatus including totally enclosed substations and h.v. cables as will be
discussed later in this chapter.
5.2.7 Mobility of gaseous ions and deionization by diffusion
Mobility
In the presence of an electric field charged particles in a gas will experience
a force causing them to drift with a velocity that varies directly with the field
and inversely with the density of the gas through which it moves. The drift
velocity component in the field direction of unit strength is defined as the
mobility K or symbolically
K D
u
E
m2
/V sec,
where u is the average drift velocity in field direction and E is the elec-
tric field strength. The mobility K is mainly a characteristic of the gas
through which the ion moves and is independent of E/p over a wide range

Table 5.5 Geballe and Harrison’s values for ˛/p and */p in oxygen and
air
E/p Oxygen Air
V/cm. torr
˛/p */p ˛/p */p ˛/p */p ˛/p */p
25.0 0.0215 0.0945 0.0730 0.00120 0.00495 0.00375
27.5 0.0293 0.0900 0.0607 0.00205 0.00473 0.00268
30.0 0.0400 0.0851 0.0451 0.00340 0.00460 0.00120
32.5 0.0532 0.0795 0.0263 0.00560 0.00460 C0.00100
35.0 0.0697 0.0735 0.0038 0.00880 0.00475 C0.00405
37.5 0.0862 0.0685 C0.0177 0.0130 0.00497 C0.0080
40.0 0.107 0.0645 C0.043 0.0190 0.00530 C0.0137
42.5 0.128 0.0605 C0.068 0.0260 0.00575 C0.0203
45.0 0.152 0.0570 C0.095 0.0340 0.00635 C0.0227
47.5 0.179 0.0535 C0.126 0.0460 0.00700 C0.0390
50.0 0.206 0.052 C0.154 0.057 0.00780 C0.049
52.5 0.234 0.049 C0.185 0.070 0.00870 C0.061
55.0 0.263 0.047 C0.216 0.087 0.00967 C0.077
57.5 0.292 0.045 C0.247 0.102 0.0108 C0.091
60.0 0.323 0.043 C0.280 0.120 0.0119 C0.108
62.5 0.355 0.0415 C0.314 0.140 – –
65.0 0.383 0.040 C0.343 0.170 – –
70.0 0.450 – – – – –
72.5 0.480 – – – – –
75.0 0.518 – – – – –
of E/p so long as the velocity gained by the ion from the field is consider-
ably less than the average thermal velocity of the gas through which the ion
moves.
To derive an expression for mobility of ions in a gas under an influence
of electric field in the region of low values of E/p we assume that the ions
are in thermal equilibrium with the gas molecules. Their drift velocity is
small compared to the thermal velocity. If ;, the time interval between two
successive collisions, is independent of E, then
; D
#i
c
where #i is the ionic mean free path and c is the mean thermal velocity of the
ion. During time ; the ion is accelerated by the field E with an acceleration

Table 5.6 Experimental values of the ionization and attachment coefficients
in SF6 (temp. D 20°C
E/p20 p ˛/p ð 103
˛/p */p ð 107
V/cm. torr torr cm1
cm1
cm1
torr1
torr1
torr1
115.0 5.2 90 1.05 1.14
125.0 5.2 200 1.32 1.12
135.0 5.2 480 1.52 1.04
145.0 5.2 760 1.73 0.97
154.0 5.2 1000
155.0 5.2 1050
165.0 5.2 1300
175.0 5.2 1550
185.0 5.2 1850
200.0 5.2 2250
115.0 19.5 90 1.04 1.13
120.0 19.5 50 1.18 1.13
125.0 19.5 200 1.30 1.10
135.0 19.5 505 1
116.0 50.2 25
118.0 50.2 8 1.15 1.14
120.0 50.2 60 1.18 1.12
122.0 50.2 115
125.0 50.2 225
126.0 50.2 240 3
116.0 99.1 38
118.0 99.1 10
119.0 99.1 33
120.0 99.1 56
122.0 99.1 120
115.0 202.0 75
117.0 202.0 16
118.0 202.0 4
119.0 202.0 29
119.25 202.0 36 60
122.0 202.0 110 6
117.0 402.4 30
118.0 402.4 5
118.5 402.4 16 75

a D eE/m, where m is the ionic mass and e is its charge. Therefore, in time
; it moves a distance
s D
eE
2m
;2
and the drift velocity becomes
u D
eE
2m
; D
e;
2m

E D
e#i
2mc

E
and
K D
u
E
D
e#i
2mc
. 5.56
In deriving eqn (5.56) we assumed that #i is unaffected by the drift motion,
that is all ions are moving with the same random velocity and all ions have
the same mean free path #i. To take the statistical distribution of mean free
paths #i into account, let us assume that the ions are moving with an average
velocity c in zig-zag projections of lengths which are distributed about the
mean free path #i. Then if a is the acceleration caused by the field E, the
distance between two collisions is
s D
1
2
at2
D
1
2
eE
m
x2
c2

.
x denotes the total distance travelled between these collisions. The average
value of s is obtained by averaging x2
over the distribution of free paths
s D
eE
2m
1
c2
1
0
x2
ex/#i
dx
#i
1
0
ex/#i
dx
#i
D
eE
mc2
#
2
i .
If the mean free time ; D #i/c
s D
eE
m
;2
,
the drift velocity
u D
s
;
D
eE
mc
#i,
and
K D
u
E
D
e#i
mc
. 5.57

Thus when the distribution of free paths is taken into account the expression
for mobility is increased by a factor of 2. The expression (5.57) ignores the
fact that after collision the ions may have initial velocities in the direction of
field. Langevin14
deduced a more exact expression which takes into account
this effect of ‘persistence of motion’ and for an ion of mass m moving through
a gas consisting of molecules of mass M the expression becomes
K D
0.815e#
Mc
m C M
m
5.58
where c is the r.m.s. velocity of agitation of the gas molecules and is an
approximation to the ionic mean free path # ³ #i. For condition of thermal
equilibrium
mc2
1
2
D
Mc2
2
D
3
2
kT.
With c1 the r.m.s. velocity of the ions, k Boltzmann’s constant and T absolute
temperature, expression (5.58) can be written in the form
K D 0.815
e#i
mc1
m C M
m
. 5.59
For an electron m − M this expression reduces to
K D 0.815
e
m
#e
c1
. 5.60
Table 5.7 gives some experimentally determined mobilities for negative and
positive ions. The presence of impurities is found to have a profound effect on
the measured mobility. The effect is particularly large in the case of negative
ions when measured in non-attaching gases such as helium or hydrogen for
which the electrons are free if the gases are extremely pure. The ion and elec-
tron mobilities can be used for the determination of conductivity or resistivity
of an ionized gas. In the simplest case when the concentrations of positive
ions and electrons are equal
nC D ne D n,
then the total current density
j D ji C je D nui C uee
where ui and ue are the drift velocities of the ions and electrons respectively.
In terms of mobilities, the current density j and the conductivity become
j D neEKe C Ki

Table 5.7 Mobility of singly charged gaseous ions at O°C and 760 Hg (in
cm/sec/volts/cm) (taken from Cobine4
)
Gas K
KC
Gas K
KC
Air (dry) 2.1 1.36 H2 (very pure) 7900.0
Air (very pure) 2.5 1.8 HCl 0.95 1.1
A 1.7 1.37 H2S 0.56 0.62
A (very pure) 206.0 1.31 He 6.3 5.09
Cl2 0.74 0.74 He (very pure) 500.0 5.09
CCl4 0.31 0.30 N2 1.84 1.27
C2H2 0.83 0.78 N2 (very pure) 145.0 1.28
C2H5Cl 0.38 0.36 NH3 0.66 0.56
C2H5OH 0.37 0.36 N2O 0.90 0.82
CO 1.14 1.10 Ne 9.9
CO2 0.98 0.84 O2 1.80 1.31
H2 8.15 5.9 SO2 0.41 0.41
and
D
j
E
D neKe C Ki. 5.61
Since Ke × Ki, the conductivity is given approximately by
D neKe. 5.62
In the presence of appreciable space charge ne 6D ni the conductivity compo-
nents must be considered separately.
Diffusion
In electrical discharges whenever there is a non-uniform concentration of ions
there will be movement of ions from regions of higher concentration to regions
of lower concentration. The process by which equilibrium is achieved is called
diffusion. This process will cause a deionizing effect in the regions of higher
concentrations and an ionizing effect in regions of lower concentrations. The
presence of walls confining a given volume augments the deionizing effect
as the ions reaching the walls will lose their charge. The flow of particles
along the ion concentration gradient constitutes a drift velocity similar to that
of charged particles in an electric field. Both diffusion and mobility result in
mass motion described by drift velocity caused in one case by the net effect
of unbalanced collision forces (ion concentration gradient) and in the other
case by the electric field.

If we consider a container with gas in which the concentration varies in the
x-direction, then taking a layer of unit area and thickness dx placed perpendicu-
larly to the direction x, the number of particles crossing this area is proportional
to the ion concentration gradient dn/dx. The flow of particles or flux in the
x-direction is
 D D
dn
dx
5.63
The negative sign indicates that n increases and the rate of flow  must
decrease in the direction of flow. The constant D is known as the diffusion
coefficient. From kinetic theory it can be shown that D D u Ð #/3. With u
being the mean thermal velocity, the rate of change of concentration in the
layer dx is
d
dt
ndx D 

 C
d
dx
dx

dn
dt
D D
d2
n
dx2
.
5.64
For the three-dimensional case eqn (5.64) becomes
∂n
∂t
D Dr2
n 5.65
which is the general equation for diffusion.
5.2.8 Relation between diffusion and mobility
In most transport phenomena, both diffusion and mobility will be acting
together. It is therefore important to establish a relation between the diffu-
sion coefficient and mobility. Consider a cloud of singly charged particles
diffusing through the gas. For simplicity let us take again the unidirectional
case with particles diffusing in the x-direction at a rate of flow given by
eqn (5.63). Then the ion velocity is equal to
ui D

ni
D
D
ni
dni
dx
where ni is the ion concentration. Because ni is directly proportional to pi
ui D
D
pi
dpi
dx
D
D
pi
fi
the force acting on the ions in this volume.

Since there are N ions per unit volume, the force exerted on one ion is
fi D
1
N
dpi
dx
D
pi
DN
ui.
An ion subjected to E, the force acting on it opposite to drift motion is
fe D eE D
eu
K
, with u-drift velocity of ion.
In order that there is no net flow in the x-direction the force fi must be
balanced by fe (oppositely directed, ui D u) and fi D fe:
D
K
D
Pi
eN
D
kTni
eni
D
kT
e

ni D
pi
kT

5.66
In general the mobilities of negatively charged ions are higher than those
of positive ones (Table 5.7) and consequently the negative ions will diffuse
more rapidly. If the concentration of the diffusing particles is significant, the
differential rate of diffusion will cause charge separation which will give
rise to an electric field. The action of the field is such that it will tend to
augment the drift velocity of the positive ions and retard that of negative
ions, and the charge separation reaches a state of equilibrium in which the
position and negative ions diffuse with the same velocity. This process is
known as ambipolar diffusion. The average velocity of the diffusing ions may
be obtained by considering the ion motion to be governed by the combined
action of diffusion and mobility in the induced field E.
Then the velocity of the positive ions is given by
uC
D
DC
dnC
nC
dx
C KC
E. 5.67
Similarly the velocity of negative ions is
u
D
D
dn
n
dx
K
E. 5.68
Eliminating E between eqns (5.67) and (5.68), and assuming nC
D n
D n,
dnC
dx
D
dn
dx
D
dn
dx
and uC
D u
D u.
The average velocity of the ions then becomes
u D
DC
K
C D
KC
nKC
C K

dn
dx
5.69

and the ambipolar diffusion coefficient for mixed ions may be written as
Da D
DC
K
C D
KC
KC
C K 5.70
and since from eqn (5.66)
KC
DC D
e
kTC and
K
D D
e
kT
therefore, for the cases when Te D T
× TC
and when Ke D K
× KC
we have
Da
¾
D DC Te
TC
¾
D D KC
Ke
¾
D De
KC
Ke
¾
D
kTe
e
. 5.71
If the electrons and ions are in equilibrium with the gas, that is all particles
are at the same temperature, then we may put DeKi D DiKe and the ambipolar
diffusion coefficient becomes
Da ³
2DiKe
Ke
³ 2Di, 5.72
since Ke × Ki.
Finally, the field E between the space charges can be obtained by eliminating
u from eqns (5.66) and (5.67), giving
E D
D
DC
K
C KC
1
n
dn
dx
. 5.73
Equations (5.71) and (5.72) are commonly used, although both are only
approximated, but they demonstrate that Da increases with Te, that is with
the random electron energy and that if electrons are at the same temperature
as the gas, Da is of the same order as Di so that electrons are slowed much
more than positive ions are accelerated. Diffusion processes are of particular
importance in studying streamer discharge and spark channels.
5.3 Cathode processes – secondary effects
Electrodes, in particular the cathode, play a very important role in gas
discharges by supplying electrons for the initiation, for sustaining and for the
completion of a discharge. Under normal conditions electrons are prevented
from leaving the solid electrode by the electrostatic forces between the
electrons and the ions in the lattice. The energy required to remove an electron
from a Fermi level is known as the work function Wa and is a characteristic

of a given material. There are several ways in which the required energy may
be supplied to release the electrons.
5.3.1 Photoelectric emission
Photons incident upon the cathode surface whose energy exceeds the work
function hv Wa may eject electrons from the surface. For most metals the
critical frequency v0 lies in the u.v. range. When the photon energy exceeds
the work function, the excess energy may be transferred to electron kinetic
energy according to the Einstein relation:
1
2 mu2
e D hv D hv0 5.74
where m is the electron mass, ue its velocity and hv0 is the critical energy
required to remove the electron and hv0 D Wa the work function.
Table 5.8 gives the work functions for several elements. The work function
is sensitive to contamination which is indicated by the spread in the measured
values shown in Table 5.8.
The spread is particularly large in the case of aluminium and metals which
readily oxidize. In the presence of a thin oxide film, it has been shown by
Malter16
that positive ions may gather at the oxide layer without being
neutralized, giving rise to a high field strength leading to augmented secondary
emission. The effect is known as the Malter effect.
Table 5.8 Work function for typical elements15
Element Ag Al Cu Fe W
Wa (eV) 4.74 2.98–4.43 4.07–4.7 3.91–4.6 4.35–4.6
5.3.2 Electron emission by positive ion and excited atom impact
Electrons may be emitted from metal surfaces by bombardment of positive
ions or metastable atoms. To cause a secondary emission of an electron the
impinging ion must release two electrons, one of which is utilized to neutralize
the ion charge. The minimum energy required for a positive ion electron emis-
sion is twice the work function WK C Wp ½ 2Wa, since the ion is neutralized
by one electron and the other electron is ejected. WK and Wp are the respec-
tive kinetic and potential energies of the incident ion. The electron emission
by positive ions is the principal secondary process in the Townsend spark
discharge mechanism.

Neutral excited (metastable) atoms or molecules incident upon the electrode
surface are also capable of ejecting electrons from the surface.
5.3.3 Thermionic emission
In metals at room temperature the conduction electrons will not have sufficient
thermal energy to leave the surface. If we consider the electrons as a gas at
room temperature, then their average thermal energy is
mu2
e
2
D
3kT
2
D 3.8 ð 102
eV,
which is much lower than the work function (Table 5.8). If, however, the metal
temperature is increased to some 1500–2500 K, the electrons will receive
energy from the violent thermal lattice vibrations sufficient to cross the surface
barrier and leave the metal. The emission current is related to the tempera-
ture of the emitter by the Richardson17
relation for thermionically emitted
saturation current density:
Js D
4mek2
h3
T2
exp

Wa
kT

A/m2
5.75
where e and m are the electronic charge and mass respectively, h is Planck’s
constant, k Boltzmann’s constant, T the absolute temperature and Wa the
surface work function.
Putting
A D
4mek2
h3
,
the above expression becomes
Js D AT2
exp

Wa
kT

5.76
which shows that the saturation current density increases with decreasing work
function and increasing temperature. On substitution of the constants m, e, k
and h, A D 120 ð 104
A m2
degr2
. The experimentally obtained values are
lower than predicted by eqn (5.76). This discrepancy is attributed to the wave
nature of the electrons. Although electrons may possess the required escape
energy, some of them may be reflected back into the solid from the surface
atoms or surface contaminants such as adsorbed gases. The effect may be
taken into account by inserting the effective value Aeff D A1 R in the
current density expression (5.76), where R is the reflection coefficient. In the
presence of a strong electric field there will be a reduction in the work function

as the Schottky18
effect, discussed in the next section, and the thermionic
emission will be enhanced.
5.3.4 Field emission
Electrons may be drawn out of a metal surface by very high electrostatic
fields. It will be shown that a strong electric field at the surface of a metal may
modify the potential barrier at the metal surface to such an extent that electrons
in the upper level close to the Fermi level will have a definite probability
of passing through the barrier. The effect is known as ‘tunnel effect’. The
fields required to produce emission currents of a few microamperes are of the
order of 107
–108
V/cm. Such fields are observed at fine wires, sharp points
and submicroscopic irregularities with an average applied voltage quite low
(2–5 kV). These fields are much higher than the breakdown stress even in
compressed gases.
To derive an expression for the emission current let us consider an electron
as it leaves the surface in the direction x as shown in Fig. 5.13. Its electric
field can be approximated as that between a point charge and the equipotential
planar surface. The field lines here are identical to those existing when an
image charge of Ce is thought to exist at a normal distance of x on the
other side of the equipotential metal surface. Applying Coulomb’s law, the
force on the electron in the x-direction is given by
Fx D
e2
4ε02x2
D
e2
16ε0x2
.
Metal
0
∆W
xmas Vacuum or gas
eEx
E
1
2
Conduction band
3 = 1 + 2
Wa
Potential
energy
(eV)
Figure 5.13 Lowering of the potential barrier by an external field.
1. Energy with no field. 2. Energy due to field. 3. Resultant energy

The potential energy at any distance x is obtained by integrating the above
equation from 1 to x.
We1 D
e2
16ε0x
5.77
which gives a parabola shown by curve 1 of Fig. 5.13. The effect of the
accelerating external field when applied at right angles to the cathode surface
gives the electron a potential energy
WE D eEx 5.78
which is a straight line shown by Fig. 5.13 (curve 2). The total energy is then
W D Wa C WE D
e2
16ε0x

eEx 5.79
which is shown by the resultant curve 3 (Fig. 5.13). Thus a marked reduction
W in the potential barrier is obtained. The maximum reduction at xm is
obtained by differentiating eqn (5.79) or
dW
dx
D
e2
16ε0x2
m
eE D 0
xm D
e
16ε0E
.
Inserting this value into eqn (5.79) the lowering in the work function becomes
W D e

eE
4ε0
. 5.80
Hence, the effective value of the work function is
Weff D Wa

eE
4ε0
5.81
and the saturation current due to electron emission using eqn (5.76) in the
presence of field E becomes
Js D AT2
exp

e
kT
Wa

eE
4ε0

5.81

which is known as the Schottky’s equation. If the current density in the absence
of external field is J0 (eqn (5.76)) then rearranging (5.81) we obtain
Js D J0 exp

e
kT
Wa

eE
4ε0

D J0 exp

B
p
E
T

. 5.82
To obtain emission current J significantly higher than J0, E must be of
the magnitude of 10 MV/cm or higher. In practice a significant field emis-
sion current may be observed at somewhat lower fields. The effect has been
explained by Fowler and Nordheim19
who derived an expression for field
emission on the basis of wave mechanics. These authors have shown that a
few electrons in a metal will have an energy slightly above the Fermi level and
thus will have a greater probability to penetrate the potential barrier ‘tunnel
effect’. The Fowler–Nordheim equation has the form
j D CE2
exp

D
E

5.83
where C and D are constants involving atomic constants. Equation (5.83)
shows that field emission is independent of temperature, but this is valid only
at low temperatures. At higher temperatures both thermionic and field emission
will occur simultaneously.
5.3.5 Townsend second ionization coefficient g
According to eqn (5.43) a graph of log I against gap length should yield a
straight line of slope ˛ if for a given pressure of p, E is kept constant. In
his early measurements of current in parallel plate gaps Townsend4
observed
that at higher voltages the current increased at a more rapid rate than given
by eqn (5.43) or (5.55). Figure 5.14 shows the kind of curves obtained by
plotting log I against electrode separation at a constant pressure. To explain
this departure from linearity Townsend postulated that a second mechanism
must be affecting the current. He first considered liberation of electrons in the
gas by collision of positive ions, and later the liberation of electrons from the
cathode by positive ion bombardment according to the mechanism discussed
earlier. On these assumptions he deduced the equation for the current in the
self-sustained discharge. Other processes responsible for the upcurving of the
(log I d) graph, Fig. 5.14, include the secondary electron emission at the
cathode by photon impact and photoionization in the gas itself
Following Townsend’s procedure we consider the case for a self-sustained
discharge where the electrons are produced at the cathode by positive ion
bombardment.
Let n D number of electrons reaching the anode per second, n0 number of
electrons emitted from the cathode by (say) u.v. illumination, nC number of

log I0
log
I
Slope = α
E
P P
1
E
P 1
E
2
P
E
2

P
E
3
P
E
3

Electrode spacing
Figure 5.14 Variation of gap current with electrode spacing in uniform field
gaps
electrons released from the cathode by positive ion bombardment, number
of electrons released from the cathode per incident positive ion.
Then
n D n0 C nCe˛d
and
nC D [n n0 C nC].
Eliminating nC
n D
n0e˛d
1 e˛d
1
.
or for steady state current
I D I0
n0e˛d
1 e˛d
1
. 5.84
Townsend’s original expression was of the form
I D I0
˛ ˇ e˛ˇd
˛ ˇe˛ˇd
5.85
where ˇ represents the number of ion pairs produced by a positive ion trav-
elling a 1-cm path in the field direction and ˛, d, I and I0 have the same
significance as in eqn (5.84). Townsend’s original suggestion for secondary

ionization in the gas by positive ion impact does not work, because ions
rapidly lose energy in elastic collisions according to eqn (5.35) and ordinarily
are unable to gain sufficient energy from the field to cause ionization on
collision with neutral atoms or molecules.
5.3.6 Secondary electron emission by photon impact
The case where the secondary emission arises from photon impact at the
cathode may be expressed by the equation:12
I D I0
˛e˛d
˛ *ge˛Fd
5.86
where is the number of photons produced by an electron in advancing 1 cm
in the direction of the field, F is the average absorption coefficient for photons
in the gas, g is a geometrical factor representing the fraction of photons that
reach the cathode, and * is the fraction of the photons producing electrons at
the cathode capable of leaving the surface.
In practice both positive ions and photons may be active at the same time
in producing electrons at the cathode. Furthermore, metastable atoms may
contribute to the secondary emission at the cathode. Which of the particular
secondary mechanisms is predominant depends largely upon the experimental
conditions in question. There may be more than one mechanism operating in
producing the secondary ionization in the discharge gap and it is customary
to express the secondary ionization by a single coefficient and represent the
current by eqn (5.84), bearing in mind that may represent one or more of
the several possible mechanisms D I C ph C . . ..
Experimental values of can be determined from eqn (5.84) by measure-
ment of the current in the gap for various pressures, field strength and gap
length and using the corresponding values of ˛. As would be expected from
the considerations of the electron emission processes, the value of is greatly
affected by the nature of the cathode surface. Low work function materials
under the same experimental conditions will produce higher emission. The
value of is relatively small at low values of E/p and will increase with
E/p. This is because at higher values of E/p there will be a larger number of
positive ions and photons of sufficiently high energy to eject electrons upon
impact on the cathode surface.
Llewellyn Jones and Davies20
have studied the influence of cathode surface
layers on the breakdown characteristic of air and on the corresponding values
of . Their data are included in Table 5.9 which shows a wide variation in
the minimum breakdown voltage Vm and the accompanying variation in the
values of . Influence of to breakdown strength is restricted to the ‘Townsend
breakdown mechanism’, i.e. to low-pressure breakdown only as can be shown
by the various breakdown criteria to be discussed in the next section.

Table 5.9
Gas Cathode Vm E/p
(volts)

V
cm.torr

Air contaminated Copper amalgam 460 720 0.004
with Hg vapour Mercury film on aluminium 390 885 0.014
Mercury film on nickel 390 885 0.014
Mercury film on staybrite steel 390 585 0.006
Air Oxidized aluminium 416 905 0.01
Oxidized nickel 421 957 0.01
Hydrogen Aluminium 243 200 0.1
(electrode Aluminium deposited on nickel 212 200 0.15
treated by Nickel 289 180 0.075
glow discharge) Nickel deposited on aluminiun 390 245 0.015
Commercial aluminium 225 200 0.125
Aluminium on staybrite steel 205 210 0.15
Staybrite steel 274 190 0.075
Steel deposited on aluminium 282 190 0.075
5.4 Transition from non-self-sustained discharges to
breakdown
5.4.1 The Townsend mechanism
As the voltage between electrodes in a gas with small or negligible electron
attachment increases, the electrode current at the anode increases in accordance
with eqn (5.84)
I D I0
e˛d
1 ead
1
or, introducing eqn (5.44) and E D V/d
I
I0
D
epd
Ð f

V
pd

1

epd
Ð f

V
pd

1

until at some point there is a sudden transition from the dark current I0 to a
self-sustaining discharge. At this point the current I becomes indeterminate

and the denominator in the above equation vanishes, i.e.
e˛d
1 D 1.
If the electron attachment is taken into account (section 5.10), this equation
becomes
˛
˛ *
[e˛*d
1] D 1
or approximately
e˛*d
D e˛d
D 1 5.87
since
e˛d
× 1 and ˛ × *
where ˛ D ˛ * represents the effective ionization coefficient defined earlier
in this chapter. The electron current at the anode equals the current in the
external circuit. Theoretically the value of the current becomes infinitely large,
but in practice it is limited by the external circuitry and, to a small extent,
by the voltage drop within the arc. Equation (5.87) defines the conditions for
onset of spark20
and is called the Townsend criterion for spark formation or
Townsend breakdown criterion. When e˛d
1 D 1, the number of ion pairs
produced in the gap by the passage of one electron avalanche is sufficiently
large that the resulting positive ions, on bombarding the cathode, are able
to release one secondary electron and so cause a repetition of the avalanche
process. The secondary electron may also come from a photoemission process
(see eqn (5.86)). In either case electron avalanche will have a successor. The
discharge is then self-sustaining and can continue in the absence of the source
producing I0, so that the criterion e˛d
1 D 1 can be said to define the
sparking threshold. For e˛d
1 1 the ionization produced by successive
avalanches is cumulative. The spark discharge grows more rapidly the more
e˛d
1 exceeds unity.
For e˛d
1 1 the current I is not self-sustained, i.e. on removal of the
source producing the primary current I0 it ceases to flow (see Fig. 5.14).
An alternative expression for the Townsend breakdown criterion is obtained
by rewriting expression (5.87) in the form
˛d D ln

1

C 1

D K. 5.88
The right-hand side of this equation, K, can often be treated as being constant,
due to the following phenomena. As mentioned earlier, the electron emission
processes characterized by are greatly affected by cathode surface, as well
as by gas pressure. However, is of very small value 10
2 10
3 and
therefore 1/ is quite a high number. Therefore, K D ln1/ C 1 does not

change too much and is for a Townsend discharge of the order of 8–10. As ˛
is often very strongly dependent upon gas pressure p or field strength E, the
exact value of K is of minor importance and may be treated as a constant for
many conditions of p and E.
5.5 The streamer or ‘Kanal’ mechanism of spark
The growth of charge carriers in an avalanche in a uniform field E0 D V0/d is
described by the exponent e˛d
. This is valid only as long as the electrical field
of the space charges of electrons and ions can be neglected compared to the
external field E0. In his studies of the effect of space charge of an avalanche
on its own growth, Raether observed that when the charge concentration was
higher than 106
but lower than 108
the growth of an avalanche was weakened.
When the ion concentration exceeded 108
the avalanche current was
followed by a steep rise in current and breakdown of the gap followed. Both
the underexponential growth at the lower concentration and rapid growth in the
presence of the high concentration have been attributed to the modification of
the originally uniform field E0 by the space charge field. Figure 5.15 shows
diagramatically the electric field around an avalanche as it progresses along
d
E0
E0
E (x)
−
−
C
+
+
A
Figure 5.15 Diagrammatic representation of field distortion in a gap
caused by space charge of an electron avalanche

the gap and the resulting modification to the original field E0. For simplicity
the space charge at the head of the avalanche is assumed concentrated within
a spherical volume, with the negative charge ahead because of the higher
electron mobility. The field is enhanced in front of the head of the avalanche
with field lines from the anode terminating at the head. Further back in the
avalanche, the field between the electrons and the ions left behind reduced
the applied field E0. Still further back the field between the cathode and the
positive ions is enhanced again. The field distortion becomes noticeable with a
carrier number n 106
. For instance, in nitrogen with d D 2 cm, p D 760 torr,
˛ ³ 7 and E0/p ³ 40 V/torr cm, the field distortion is about 1 per cent, leading
to 5 per cent change in ˛. If the distortion of ¾
D1 per cent prevailed in the entire
gap it would lead to a doubling of the avalanche size, but as the distortion is
only significant in the immediate vicinity of the avalanche head it has still an
insignificant effect. However, if the carrier number in the avalanche reaches
n ³ 108
the space charge field becomes of the same magnitude as the applied
field and may lead to the initiation of a streamer. The space charge fields play
an important role in the mechanism of corona and spark discharges in non-
uniform field gaps. For analytical treatment of space charge field distortion
the reader is referred to reference 12.
In the Townsend spark mechanism discussed in the previous section the
gap current grows as a result of ionization by electron impact in the gas
and electron emission at the cathode by positive ion impact. According to
this theory, formative time lag of the spark should be at best equal to the
electron transit time ti. In air at pressures around atmospheric and above
pd 103
torr cm the experimentally determined time lags have been found
to be much shorter than ti. Furthermore, cloud chamber photographs of
avalanche development have shown22
that under certain conditions the space
charge developed in an avalanche is capable of transforming the avalanche into
channels of ionization known as streamers that lead to rapid development of
breakdown. From measurements of the prebreakdown current growth23
and
the minimum breakdown strength it has been found that the transformation
from avalanche to streamer generally occurs when the charge within the
avalanche head (Fig. 5.15) reaches a critical value of n0 exp[˛xc] ³ 108
or
˛xc ³ 18–20, where xc is the length of the avalanche path in field direction
when it reaches the critical size. If xc is larger than the gap length xc d
then the initiation of streamers is unlikely. Typical cloud chamber photographs
of electron avalanche and streamer development are shown in Figs 5.16(a) to
(d). In (a) the discharge has been arrested before reaching the critical size
¾108
, giving the avalanche the classical ‘carrot’ shape. In (b) the avalanche
has grown beyond the critical size, its head has opened up indicating ionization
around the original avalanche head and a cathode directed streamer starts. This
continues (c, d) till a plasma channel connects cathode and anode. The early
cloud chamber results have led Raether22
to postulate the development of two

(a) (b) (c) (d) (e)
Figure 5.16 Cloud chamber photographs showing development of the
cathode directed streamers (with increasing pulse length): (a) avalanche near
anode; (b) and (c) cathode directed streamer starts; (d) and (e) time period
for plasma channel to connect cathode and anode
types of streamers: (1) the ‘anode directed streamer’ describing the apparent
growth of ionization and of the avalanche head, and (2) the ‘cathode directed
streamer’ describing the additional discharge growth from the avalanche tail.
In later investigations, Wagner24
has obtained streak photographs of
‘avalanche streamer’ development using an image intensifier. In these
experiments the time and space resolved in radiation density which
corresponds to the electron density is monitored. The observed radiation
pattern together with the photocurrent growth is sketched in Figs 5.17(a)
and (b). Region (a)–(b) corresponds to the development of avalanche with
an approximate velocity of 108
cm/sec. The current growth is exponential.
Beyond (b), after the avalanche has reached the critical size, there is an
increase in the velocity of the avalanche head by about a factor of 10. In many
cases almost simultaneously a second luminous front is observed proceeding
towards the cathode with the same velocity as the anode directed growth. The
current growth in this region is faster than exponential.
The observed short time lags together with the observations of discharge
development have led Raether and independently Meek25
and Meek and
Loeb26
to the advancement of the ‘streamer’ or ‘Kanal’ mechanism for spark
formation, in which the secondary mechanism results from photoionization of
gas molecules and is independent of the electrodes.
In the models developed by Raether and Meek it has been proposed that
when the avalanche in the gap reaches a certain critical size the combined
space charge field and externally applied field lead to intense ionization and
excitation of the gas particles in front of the avalanche head. Instantaneous
recombination between positive ions and electrons releases photons which

120 130 140
140 150 160 170 180 190
150 160 170 n sec
n sec
20nsec
a
b
b′
t
t
t
Figure 5.17 Image intensifier photographs, and a photocurrent oscillogram
showing the development of cathode directed streamers in N2 (20 per cent
CH4 p D 88.5 torr. (a) and (b) progress of streamers after switching off
external voltage; (b0
) photocurrent oscillogram corresponding to (b)24
in turn generate secondary electrons by the photoionization process. These
electrons under the influence of the electric field in the gap develop into
secondary avalanches as shown in Fig. 5.18. Since photons travel with the
velocity of light, the process leads to a rapid development of conduction
channel across the gap.
On the basis of his experimental observations and some simple assumptions
Raether33
developed an empirical expression for the streamer spark criterion
of the form
˛xc D 17.7 C ln xc C ln
Er
E
5.89
where Er is the space charge field strength directed radially at the head of
avalanche as shown in Fig. 5.19, E is the externally applied field strength.

A
C
Figure 5.18 Secondary avalanche formation by photoelectrons
A
C
E
E
Er
Xc
Figure 5.19 Space charge field Er around avalanche head
The resultant field strength in front of the avalanche is thus E C Er while
in the positive ion region just behind the head the field is reduced to a value
E Er. It is also evident that the space charge increases with the avalanche
length e˛x
.
The condition for the transition from avalanche to streamer assumes that
space charge field Er approaches the externally applied field Er ³ E, hence
the breakdown criterion (eqn (5.89)) becomes
˛xc D 17.7 C ln xc. 5.90
The minimum breakdown value for a uniform field gap by streamer mechanism
is obtained on the assumption that the transition from avalanche to streamer
occurs when the avalanche has just crossed the gap d. Then Raether’s empir-
ical expression for this condition takes the form
˛d D 17.7 C ln d. 5.91
Therefore the breakdown by streamer mechanism is brought about only when
the critical length xc ½ d. The condition xc D d gives the smallest value of ˛
to produce streamer breakdown.

A similar criterion equation for the transition from avalanche to streamer has
been developed by Meek.25
As in Raether’s case the transition is assumed to
take place when the radial field about the positive space charge in an electron
avalanche attains a value of the order of the externally applied field. Meek25
has shown that the radial field produced by positive ions immediately behind
the head of the avalanche can be calculated from the expression
Er D 5.3 ð 107 ˛e˛x

x
p
1/2
volts/cm 5.92
where x is the distance (in cm) which the avalanche has progressed, p is
the gas pressure in torr and ˛ is the Townsend coefficient of ionization by
electrons corresponding to the applied field E. As in Raether’s model the
minimum breakdown voltage is assumed to correspond to the condition when
the avalanche has crossed the gap of length d and the space charge field Er
approaches the externally applied field. Substituting into eqn (5.92) Er D E
and x D d and rearranging gives
˛d C ln
˛
p
D 14.5 C ln
E
p
C
1
2
ln
d
p
. 5.93
This equation is solved by trial and error using the experimentally determined
relation between ˛/p and E/p. Values of ˛/p corresponding to E/p at a
given pressure are chosen until the equation is satisfied.
Table 5.10 compares Meek’s calculated and the measured values Vb for air
according to eqn (5.93). At small d, the calculated values Vb are higher than the
measured ones. The reverse is true at large d. In general, however, the deviation
between theory and experiment should be regarded as not very large, in view
of the various simplifying assumptions made by Meek,25
especially those in
order to determine the charge density and the tip radius of the avalanche. The
avalanche radius was calculated on the basis of thermal diffusion using the
relationship r D
p
3Dt where D is thermal diffusion coefficient and t the time.
The charge was assumed to be concentrated in a spherical volume which is only
approximately correct. At the charge densities in question, ambipolar diffusion
is likely to be important, but so far has been neglected.
In section 5.4 we have seen that the Townsend criterion for spark formation
is satisfied when the product ˛d reaches a value of 8–10 ˛d D ln1/ C 1 D
8–10. The streamer criterion for spark formation, however, requires a value of
18–20, ˛d D ˛xc D ln 108
³ 20 with xc d. Therefore under certain exper-
imental conditions there will be a transition from the Townsend to streamer
mechanism. This transition is brought about by increased pressure and gap
length and in practice it occurs in the region of pd ½ 1–2 bar cm. The tran-
sition is indicated by a discontinuity in the formative time lag discussed in
section 5.10. The streamer mechanism which relies on photoionization in the

Table 5.10 Comparison of calculated and measured Vb
values for air according to Meek’s model
Gap E/p ˛d Vb Vb
length V/cm calculated measured
cm torr kV kV
0.1 68.4 15.7 5.19 4.6
0.5 48.1 17.7 18.25 17.1
1.0 42.4 18.6 32.20 31.6
2.5 37 19.7 70.50 73
10 32.8 21.5 249 265
20 31.2 22.4 474 510
0.1
10−8
10−7
10−6
10−5
10−4
1.0 10 50
T ∗
K
t
f
(sec)
qcr = ∆V/V (%)
Figure 5.20 Formative time lag in N2 as function of overvoltage.
p D 500 torr, d D 2 cm. TŁ
Townsend mechanism, K streamer mechanism
gas requires a much shorter formative time than the Townsend mechanism in
which the secondary mechanism is cathode dependent and is affected by the
transit time of positive ions. Figure 5.20 compares the formative time lag20
in nitrogen at p D 500 torr and d D 2 cm with the measured values, plotted as
a function of percentage overvoltage D V/V%. At lower overvoltages

the formative time lag follows the Townsend mechanism up to a critical
value crit D Vcrit/V at which the electron amplification within the avalanche
reaches a value e˛d
½ 108
. Curve K has been calculated from the time tA
required to reach the critical size at various overvoltages,
tA D
xcrit
v
D
18
˛v
5.94
where v is the electron drift velocity. Curve TŁ
is obtained from the Townsend
mechanism. No discontinuity is observed and the curve gives a too long forma-
tive time lag for the higher overvoltages.
Table 5.11 gives the critical overvoltages for several of the commonly used
gases together with the corresponding pd values. The sudden change in the
formative time lag usually takes place for values of some 107
sec.
Table 5.11 Critical overvoltages for various
gases27
Gas p d pd crit
(torr) (cm) (torr cm) (%)
H2 500 2 1000 16.6
N2 500 2 1000 18.2
N2 400 3 1200 17
Air 760 1 760 4.5
5.6 The sparking voltage–Paschen’s law
The Townsend criterion, eqn (5.87), enables the breakdown voltage of the gap
to be determined by the use of appropriate values ˛/p and corresponding
to the values E/p without ever taking the gap currents to high values, that
is keeping them below 107
A, so that space charge distortions are kept to
a minimum, and more importantly so that no damage to electrodes occurs.
Good agreement has been found28
between calculated and experimentally
determined breakdown voltages for short or long gaps and relatively low
pressures for which this criterion is applicable.
An analytical expression for breakdown voltage for uniform field gaps as
a function of gap length and gas pressure can be derived from the threshold
eqn (5.87) by expressing the ionization coefficient ˛/p as a function of field
strength and gas pressure. If we put ˛/p D fE/p in the criterion equation

we obtain
efE/ppd
D
1

C 1
or
fE/ppd D ln

1

C 1

D K. 5.95
For uniform field Vb D Ed, where Vb is the breakdown voltage,
efVb/pdpd
D K0
D eK
5.96
or
Vb D Fpd
which means that the breakdown voltage of a uniform field gap is a unique
function of the product of pressure and the electrode separation for a particular
gas and electrode material. Equation (5.96) is known as Paschen’s law, and
was established experimentally in 1889. Equation (5.96) does not imply that
the sparking voltage increases linearly with the product pd, although it is found
in practice to be nearly linear over certain regions. The relation between the
sparking voltage and the product pd takes the form shown in Fig. 5.21 (solid
curve). The breakdown voltage goes through a minimum value Vbmin at a
particular value of the product pdmin.
pd
(pd)2W
(pd)min
(pd)2N
Vb min
Vb (N W)
Vb
N W
Figure 5.21 The sparking voltage–pd relationship (Paschen’s curve)

Let us now examine graphically the relation of the Paschen’s curve
(Fig. 5.21) with the spark criterion eqn (5.88). If the experimental relationship
between the ionization coefficient and the field strength ˛d/p D fE/p for
a given gas is plotted we obtain a curve as shown in Fig. 5.22 (curve 1) with
a limiting value E/p, corresponding to the onset of ionization. Rearranging
the Townsend criterion, eqn (5.88), and remembering that in uniform field
V D Ed, where V is the applied voltage, gives
˛
p
D
K
V
E
p
3
4
S
W
N
2
1
02 01 E
P
E
P
c
Eb
P 2W
Eb
P min
Eb
P 2N
α/p
Figure 5.22 Relation between the Townsend criterion for spark ˛d D k
and the function of ˛/p D E/p
and
tan D
˛/p
E/p
D
K
V
. 5.97
Equation (5.97) gives for constant values of K straight lines of a slope tan
depending upon the value of the applied voltage (V), curves (2, 3, 4) Fig. 5.22.
At low values of V there is no intersection between the line (curve 4) and the
curve ˛/p D fE/p. No breakdown therefore results with small voltages
below Paschen’s minimum irrespective of the value pd in eqn (5.96). At
the higher applied voltage (V), there must exist two breakdown values at a
constant pressure, one corresponding to the small value of gap length d1 and
intersection at N and the other to the longer gap d2 intersection at W. The
point S (tangent) gives the lowest breakdown value or the minimum sparking

voltage. The breakdown voltages corresponding to the points W, N and S are
indicated in the Paschen’s curve in Fig. 5.21.
The existence of the minimum value in the breakdown voltage–gap length
relation may be explained qualitatively by considering the efficiency of the
ionization of electrons traversing the gap with different electron energies.
Neglecting the secondary coefficient for values pd pdmin, electrons
crossing the gap make more frequent collisions with the gas molecules than
at pdmin, but the energy gained between collisions is lower than at pdmin.
Hence, the probability of ionization is lower unless the voltage is increased.
For pd pdmin electrons cross the gap without making many collisions.
The point pdmin corresponds to the highest ionization efficiency.
An analytical expression for the minimum values of Vbmin and pdmin
may be obtained by inserting in the criterion eqn (5.87) for ˛/p the expres-
sion (5.47)
˛ D Ap exp

Bp
E

D Ap exp

Bpd
V

and determining the minimum value of V.
Assuming that the coefficient remains constant, then
d D
eBpd/Vb
Ap
ln

1 C
1

.
Rearranging we obtain
Vb D
Bpd
ln
Apd
ln1 C 1/
5.98
Differentiating with respect to pd and equating the derivative to zero
dVb
dpd
D
B
ln
Apd
ln1 C 1/

B

ln
Apd
ln1 C 1/
2
D 0.
Therefore
ln
Apd
ln1 C 1/
D 1
and
pdmin D
e1
A
ln

1 C
1

. 5.99
Substitution into eqn (5.98) gives
Vbmin D 2.718
B
A
ln

1

C 1

. 5.100

This equation could be used for the calculation of the minimum sparking
constants (Vbmin, pdmin) if the correct values of A and B are used for the
simulation of the real dependency of ˛/p D fE/p in the vicinity of
d2
˛/p
dE/p2
D 0.
In practice, the sparking constants (Vbmin and pdmin) are measured values,
and some of these are shown in Table 5.12. For example, by inserting in
eqn (5.100) the values for the constants A D 12, B D 365 and D 0.02 that
are commonly quoted in the literature, we obtain for the minimum breakdown
voltage for air Vbmin D 325 V which agrees well with the experimental value
quoted in Table 5.12. It should be noted, however, that these values are some-
times strongly dependent upon the cathode material and cathode conditions,
according to eqns (5.94) and (5.95), in which the real value of is significant.
Table 5.12 Minimum sparking constants
for various gases29
Gas pdmin Vbmin
torr cm volts
Air 0.55 352
Nitrogen 0.65 240
Hydrogen 1.05 230
Oxygen 0.7 450
Sulphur hexafluoride 0.26 507
Carbon dioxide 0.57 420
Neon 4.0 245
Helium 4.0 155
The measured minimum sparking voltage in any gas is dependent upon the
work function of the cathode material. A minimum sparking voltage as low
as 64 V has been observed by Cueilleron30
in neon between cesium-coated
electrodes at a gas pressure of 26 torr.
The breakdown voltage for uniform field gaps in air over a wide range of
pressures and gap lengths may be calculated by combining the Schumann’s
relation with the criterion eqn (5.88). Schumann31
has shown that over a
wide but restricted range of E/p, ˛/p may be expressed as
˛
p
D C

E
p

E
p

c
2
5.101

where E and Ec are field strengths, Ec being the limiting value of E at which
effective ionization starts. p is pressure and C is a constant.
Dividing eqn (5.88) by pd and combining with eqn (5.101) we obtain
K
pd
D C

E
p

E
p

c
2
or
E
p
D

E
p

c
C

K/C
pd
and the expression for the breakdown voltage Vb becomes
Vb D

E
p

c
pd C
K
C

pd. 5.102
Inserting the values for the constants Ec and K/c which were determined
by Sohst32
and Schröder33
for homogeneous field gaps at p D 1 bar; 20°C;
Ec D 24.36 (kV/cm); K/C D 45.16 (kV)2
/cm; eqn (5.102) becomes
Vb D 6.72

pd C 24.36pd kV. 5.103
The calculated breakdown voltages, using eqn (5.103) for uniform field gaps
in air for a range of the product pd from 102
to 5 ð 102
(bar cm) are
compared with the available experimental data in Fig. 5.23. The calculated
10−3 10−2
10−1
100
101
102
103
10−3 100 101 102 103
Pressure spacing product pd (bar mm)
Air
Temperature : 20°C
Publication No. 52 de la CEI (1960) [1]
Norme Italienne 42.1 (1963) [2]
Norme C 41050 de l'UTE (1960) [3]
IEEE standard No. 4 (ANSI - C 68 - 1) [4]
British standard 358 (1960) [5]
VDE 0433 [6]
S. Schreier [8]
Seeliger [9]
H. Bertein [7]
Breakdown
voltage
(Crest)
(kV)
Figure 5.23 Paschen curve for air in log–log scale. Temperature 20°C. (ž
calculated VB D 6.72
p
pd C 24.4pd

and the measured data agree well except at the very low values of the product
pd. In this region in which the E/p values are high the Schumann’s quadratic
relationship (eqn (5.102)) no longer holds, but this region is of little practical
interest.
It is often more convenient to use the gas density υ instead of the gas
pressure p in the Paschen’s eqn (5.96), since in the former case account is
taken for the effect of temperature at constant pressure on the mean free path
in the gas. The number of collisions by an electron in crossing the gap is
proportional to the product υd and .
Atmospheric air provides the basic insulation for many practical h.v. instal-
lations (transmission lines, switchyards, etc.). Since the atmospheric condi-
tions (temperature and pressure) vary considerably in time and locations,
the breakdown characteristics of various apparatus will be affected accord-
ingly. For practical purposes, therefore, the breakdown characteristics can
be converted to standard atmospheric conditions (p D 760 torr D 1.01 bar and
t D 20°C D 293 K). Correction for the variation in the ambient conditions is
made by introducing the relative density defined as
υ D
p
760
293
273 C t
D 0.386
p
273 C t
. 5.104
The breakdown voltage at standard conditions multiplied by this factor gives
the breakdown voltage corresponding to the given ambient conditions approx-
imately
Vbυ D υVbυ D 1 5.105
Paschen’s law is found to apply over a wide range of the parameter value up to
1000–2000 torr cm. At higher products, however, the breakdown voltage (in
non-attaching gases) is found to be somewhat higher than at smaller spacing for
the same values of pd. This departure is probably associated with the transition
from the Townsend breakdown mechanism to the streamer mechanism, as the
product pd is increased above a certain value. We have seen that the streamer
breakdown criterion is satisfied at higher values of ˛d than the Townsend
criterion, i.e. the value of the constant K in eqn (5.88) will increase from
about 8–10 to 18–20. At very low pressure deviations from Paschen’s law
are observed when the breakdown mechanism ceases to be influenced by the
gas particles and becomes electrode dominated (vacuum breakdown).
5.7 Penning effect
Paschen’s law is not applicable in many gaseous mixtures. The outstanding
example is the neon–argon mixture. A small admixture of argon in neon
reduces the breakdown strength below that of pure argon or neon as shown in

0
250
500
750
1000
1250
1500
1750
2000
50 100 150 200 250 300 350 400 450
Air A
Ne
Ne + 10−4% A
Ne + 10−3% A
Ne + 10−2% A
Ne + 2 × 10−4% A
Ne + 5 × 10−4% A
P0d (Torr × cm)
V
Figure 5.24 Breakdown voltage curves in neon–argon mixtures between
parallel plates at 2-cm spacing at 0°C
Fig. 5.24. The reason34
for this lowering in the breakdown voltage is that the
lowest excited state of neon is metastable and its excitation potential (16 eV)
is about 0.9 eV greater than the ionization potential of argon. The metastable
atoms have a long life in neon gas, and on hitting argon atoms there is a
very high probability of ionizing them. The phenomenon is known as the
Penning effect.
5.8 The breakdown field strength (Eb)
For uniform field gaps the breakdown field strength in a gas may be obtained
from eqn (5.98) by dividing both sides of this equation by pd, then
Vb
pd
D
Eb
p
D
B
ln
Apd
ln1 C 1/
5.106
We note that for a constant gas pressure p the breakdown field strength (Eb)
decreases steadily with the gap length (d). Furthermore, the field strength
to pressure ratio Eb/p is only dependent upon the product of pd.
Equation (5.106) also shows that the breakdown field strength Eb for a
constant gap length increases with the gap pressures but at a rate slightly
lower than directly proportional, as the pressure also affects the denominator
in the expression.

Qualitatively the decrease in the pressure related breakdown field strength
Eb/p with increasing pd may easily be understood by considering the rela-
tionship between the ionization coefficient ˛ and the field strength
˛
p
D f

E
p

plotted in Fig. 5.25 and applying the Townsend criterion equation to different
values of pd as shown. The breakdown criterion of eqn (5.88) can be
written as
˛
p
pd D k.
Assuming first that this equation is satisfied for a small product pd2 at
Eb/p2 when ˛/p reaches the point A (Fig. 5.25) and then increasing in
value of the product to pd1, the criterion equation will now be satisfied
at a lower value of ˛/pB giving a reduced breakdown strength Eb/p1.
By repeating this procedure for other values of pd we obtain a functional
relationship between the breakdown strength Eb/p and the product pd as
shown in Fig. 5.26. Thus, the curve obtained in Fig. 5.26 is in qualitative
agreement with experimental data.
0 E
p
a
p
A
B
(pd )2
(pd)1 (pd)2
Eb
p 1
Eb
p 2
E
p 0
Figure 5.25 Functional relationship between ionization coefficient ˛ and
breakdown field strength at different values of pd
Calculations of the breakdown strength Eb and the pressure related break-
down field strength E/p using eqn (5.106) yield data that are in agreement
with the experimental values over a limited range of pressures and gap length.
For air a much closer agreement with the experimental data may be obtained

Eb
p
Eb
p = f1 (pd )
(pd )2 (pd )1 pd
E
p
c
Eb
p 2
Figure 5.26 Functional relationship between the breakdown field strength
and the product pd
using the Schumann’s eqn (5.103). Dividing both sides of eqn (5.103) by the
product pd gives
Vb
pd
D
Eb
p
D
6.72

pd
C 24.36
kV
cm bar
. 5.107
The breakdown field strength Eb, when calculated for air at standard temper-
ature and pressure for gap lengths extending from 1 mm to 100 mm using
eqn (5.107), agrees well with experimental values.
5.9 Breakdown in non-uniform fields
In non-uniform fields, e.g. in point-plane, sphere-plane gaps or coaxial cylin-
ders, the field strength and hence the effective ionization coefficient ˛ vary
across the gap. The electron multiplication is governed by the integral of ˛
over the path

˛dx. At low pressures the Townsend criterion for spark takes
the form

exp
d
0
˛dx

1

D 1 5.108
where d is the gap length. The integration must be taken along the line of the
highest field strength. The expression is valid also for higher pressures if the
field is only slightly non-uniform. In strongly divergent fields there will be at
first a region of high values of E/p over which ˛/p 0. When the field falls
below a given strength Ec the integral

˛dx ceases to exist. The Townsend

mechanism then loses its validity when the criterion relies solely on the
effect, especially when the field strength at the cathode is low.
In reality breakdown (or inception of discharge) is still possible if one takes
into account photoionization processes.
The criterion condition for breakdown (or inception of discharge) for the
general case may be represented by modifying the expression (5.90) to take
into account the non-uniform distribution of ˛ or
exp
xcd
0
˛dx D Ncr 5.109
where Ncr is the critical electron concentration in an avalanche giving rise to
initiation of a streamer (it was shown to be approx. 108
), xc is the path of
avalanche to reach this size and d the gap length. Hence eqn (5.109) can be
written as
xcd
0
˛dx D ln Ncr ³ 18 20. 5.109a
Inoization region
0 xc x
Ec
E(x)
Critical field line
E (x)
Figure 5.27 Electric field distribution in a non-uniform field gap
Figure 5.27 illustrates the case of a strongly divergent field in a positive point-
plane gap. Equation (5.109a) is applicable to the calculation of breakdown or
discharge inception voltage, depending on whether direct breakdown occurs or
only corona. The difference between direct breakdown and corona inception
will be discussed in detail in the next section.
For the special case of a coaxial cylindrical geometry in air, an empirical
relation based on many measurements of the critical field strength Ec (corona

inception) for different diameters of the inner conductor (2r) and relative air
density υ was developed by Peek35
of the form:
Ec
υ
D 31.53 C
9.63
p
υr
5.110
where Ec is in kV/cm, r in cm and υ is the relative air density defined by
eqn (5.104). For values of υr 1 cm this expression gives higher values than
experimentally observed. More recently Zaengl et al.36
have developed an
analytical expression based upon eqns (5.109) and (5.101) replacing the Peek’s
empirical eqn (5.110) for calculating the corona inception voltage given as

Ec
υ
2
2

Ec
υ

E0 ln

1
E0

Ec
υ
2

E2
0 D
K/C
υr
. 5.111
0.02 0.05 0.01 1 10 100
d ⋅ r (cm)
25
30
35
40
45
50
55
60
65
70
75
80
85
E
c
/d
(kV/cm)
Eqn. 5.110
d = 2.88
d = 1.91
d = 1.38
d = 0.39
d = 2.4
d = 3.37
d = 4.82
Full line: Evaluated eqn. 5.111
(K/C) = 42 (kV2/cm)
Values according to Schumann (31)
(d=1)
Measurements by Peek (35)
(d = 0.1 − 1); r = 0.1905 (cm) (35)
Measurements by Zaengl (36)
Figure 5.28 Variation of corona inception field strength Ec/υ with υr for
coaxial cylindrical geometry in air
In this expression the constants Ec and K/C have the same significance
as in the Schumann’s eqn (5.103), but the best agreement of the calculated

values with many of the known measured values (a.c. and d.c.) of positive
polarity is obtained with the constant K/C D 42 (kV/cm)2
as compared to
K/C D 45.16 (kV/cm)2
used in eqn (5.103). Figure 5.28 compares the calcu-
lated values Ec/υ plotted as function of the product υr using eqn (5.111)
(solid curve), with the measured values by Peek and those measured by
Schumann. The dotted curve indicates the calculated values obtained using
the original empirical expression of Peek (eqn (5.110)). It is seen that for
the product less than υr 1 the values obtained from the Peek’s empirical
expression are in good agreement with experimental observations, but a devi-
ation is observed for conductors of larger radius, due to the fact that the
original expression was based on measurements on conductors of small size.
Equation (5.111) also shows that the critical field strength Ec for a coaxial
arrangement is independent of the radius of the outer cylinder R. This is true
as long as the field strength ER does not exceed υE0.
5.10 Effect of electron attachment on the breakdown
criteria
In section 5.2 it was shown that there are a number of gases in which the
molecules readily attach free electrons forming negative ions, having a similar
mass as the neutral gas molecules. They are, therefore, unable to ionize neutral
particles under field conditions in which electrons will readily ionize. The
ionization by electron collision is then represented by the effective ionization
coefficient ˛ D ˛ *.
In the presence of attachment, the growth of current in a gap when the
secondary coefficient is included in eqn (5.84) is given by the relation37
I D I0
˛
˛ *
[exp˛ *d */˛]

1
˛
˛ *
[exp˛ *d 1]
5.112
where ˛ and are the primary and secondary ionization coefficients and * is
the attachment coefficient as defined earlier in this chapter. It was also shown
that in a given gas both coefficients ˛ and * are dependent only on the field
strength E and the gas pressure

˛
p
D f

E
p

,

*
p
D f

E
p

For a self-sustained discharge in an attaching gas the denominator in
eqn (5.112) will tend to zero, and as a result we obtain the Townsend criterion

for attaching gases
˛ *d D ln

˛ *
˛
1

C 1

. 5.113
As the difference of ˛ */p for low values of E/p becomes negative,
eqn (5.113) can only be valid for values ˛ *. This means that a critical value
of a pressure-dependent field strength exists for which ˛ D * and E/p !
E/p0. Therefore, in the presence of attachment no breakdown can take place
in accordance with processes so far considered for this or lower values of E/p.
Of special engineering applications interest is sulphur hexafluoride SF6,
which has a large attachment coefficient (and thus exhibits high dielectric
strength) and has been widely applied in gas-insulated power equipment. In
the early 1950s Geballe and Harrison and Geballe and Reeve10,37
studied the
values of ˛ and * for SF6 (see Table 5.6) over a wide range of E/p and found
these fit well into the following linear equation
˛
p
D
˛ *
p
D k

E
p

E
p

0

5.114
where ˛ and * are the ionization and attachment coefficients, and k is obtained
from Figure 5.29, which has a numerical value of
k D 27.7 kV1
or 2.77 ð 102
V1
and
E/p0 D 88.5 kV/(cm bar) or 118 V/(cm Torr) at which ˛
D 0 or ˛ D *.
The linear relationship of eqn 5.114 (Fig 5.29) remains valid within
75
E
P
200
kV
cm bar
If eqn (5.114) is combined with the simplest breakdown criterion eqn (5.88),
which is
˛d D ln

1

C 1

D k
We obtain for the breakdown field strength for SF6 in uniform fields
Eb
p
D

E
p

0
C
K
kpd

kV
cm

5.115
where E/p0 D 88.5 kV/cm bar, k D 27.7 kV1
, and K D 8 . . . 10 for the
Townsend mechanism and 18 . . . 20 for the streamer mechanism. Substituting

−300
0 50 100
−200
−100
0
100
AIR
SF6
27 kV/cm⋅bar 89 kV/cm⋅bar
E /P (kV/cm⋅bar)
(α−η)/
P
=
cm
−1
⋅bar
−1
200
300
400
500
Figure 5.29 ˛ */P ¾ E/P relationship in SF6
for the constants K D 18 gives for a gas pressure p (in bar) and a gap
distance d (in cm):
Eb D 88.5 C
0.65
d
[kV/cm] 5.116
The steep increase of ˛ */p in Fig. 5.29 with pressure dependent field
strength E/p (the gradient k is much larger than in air) accounts for the
strong influence of local field distortions upon the breakdown strength.
For a uniform field gap eqn (5.116) converts to the Paschen dependency
Vb D 0.65 C 88.5 pd[kV] 5.117
This gives good agreement with measured values for the approximate voltage
range
1 kV Vb 250 kV in pd range 0.04 pd 3 bar Ð cm

At higher values of pd, Vb is slightly lower than predicted by eqn (5.117)
and follows the equation
Vb D 40 C 68pd [kV] pd in bar Ð cm 5.118
Zaengl has shown38
that since all the secondary feedback processes repre-
sented by the coefficient can be quite sensitive to gas pressure, to electrode
surface conditions or even impurities, this reduction is largely due to ‘micro-
scopic field effects’ generated by protrusions at the electrodes (called the elec-
trode effect). In uniform field gaps the dielectric strength given by eqns (5.116)
and (5.117) is approximately three times that of air, as shown in Fig. 5.30
120
100
80
60
40
20
0
0.01 0.02 0.05 0.1 0.2 0.5 1 2 5
Pd (bar⋅cm)
(E /P)0 = 27 kV/cm⋅bar
(E /P)0 = 89 kV/cm⋅bar
E
b
/
P
(kV/cm⋅bar)
AIR
SF6
(THEORETICAL)
SF6
EXPERIMENTAL
Figure 5.30 Breakdown field strength of uniform gap in SF6 and air
A number of refined equations for predicting breakdown of SF6 have
been proposed by various research workers, notably by the group working
at CRIEPI, Japan (Nitta and Takuma and Associates).39,40,41
All are based
on the assumption that the streamer mechanism prevails, that is, the criterion
for spark is satisfied when the space charge in the avalanche changes the
avalanche to a streamer. Further information can be found in reference 42.
5.11 Partial breakdown, corona discharges
In uniform field and quasi-uniform field gaps the onset of measurable
ionization usually leads to complete breakdown of the gap. In non-uniform
fields various manifestations of luminous and audible discharges are observed

long before the complete breakdown occurs. These discharges may be transient
or steady state and are known as ‘coronas’. An excellent review of the
subject may be found in a book by Loeb.43
The phenomenon is of particular
importance in h.v. engineering where non-uniform fields are unavoidable.
It is responsible for considerable power losses from h.v. transmission lines
and often leads to deterioration of insulation by the combined action of
the discharge ions bombarding the surface and the action of chemical
compounds that are formed by the discharge. It may give rise to interference
in communication systems. On the other hand, it has various industrial
applications such as high-speed printing devices, electrostatic precipitators,
paint sprayers, Geiger counters, etc.
The voltage gradient at the surface of the conductor in air required to
produce a visual a.c. corona in air is given approximately by the Peek’s
expression (5.110).35
There is a distinct difference in the visual appearance of a corona at wires
under different polarity of the applied voltage. Under positive voltage, a corona
appears in the form of a uniform bluish-white sheath over the entire surface
of the wire. On negative wires the corona appears as reddish glowing spots
distributed along the wire. The number of spots increases with the current.
Stroboscopic studies show that with alternating voltages a corona has about
the same appearance as with direct voltages. Because of the distinctly different
properties of coronas under the different voltage polarities it is convenient to
discuss separately positive and negative coronas.
In this section a brief review of the main features of corona discharges
and their effect on breakdown characteristics will be included. For detailed
treatment of the basic fundamentals of this subject the reader is referred to
other literature sources.43
5.11.1 Positive or anode coronas
The most convenient electrode configurations for the study of the physical
mechanism of coronas are hemispherically capped rod-plane or point-plane
gaps. In the former arragement, by varying the radius of the electrode tip,
different degrees of field non-uniformity can be readily achieved. The point-
plane arrangement is particularly suitable for obtaining a high localized stress
and for localization of dense space charge.
In discussing the corona characteristics and their relation to the breakdown
characteristics it is convenient to distinguish between the phenomena that occur
under pulsed voltage of short duration (impulse corona), where no space charge
is permitted to drift and accumulate, and under long lasting (d.c.) voltages
(static field corona).

Under impulse voltages at a level just above ionization threshold, because
of the transient development of ionization, the growth of discharge is diffi-
cult to monitor precisely. However, with the use of ‘Lichtenberg figures’
techniques,44
and more recently with high-speed photographic techniques,
it has been possible to achieve some understanding of the various discharge
stages preceding breakdown under impulse voltages.
+ + +
(c)
(b)
(a)
Figure 5.31 Schematic illustration of the formation of streamers under
impulse voltage-progressive growth with increasing pulse duration-positive
rod-plane gap
The observations have shown that when a positive voltage pulse is applied
to a point electrode, the first detectable ionization is of a filamentary branch
nature, as shown diagrammatically in Fig. 5.31(a). This discharge is called
a streamer and is analogous to the case of uniform field gaps at higher pd
values. As the impulse voltage level is increased, the streamers grow both
in length and their number of branches as indicated in Figs 5.31(b) and (c).
One of the interesting characteristics is their large number of branches which
never cross each other. The velocity of the streamers decreases rapidly as
they penetrate the low field region. Figure 5.32 shows velocities of impulse
streamers recorded in air in a 2.5-cm gap under two different values of voltage.
The actual mechanism of the transition from streamer to final breakdown is
complex, and several models have been developed4
to explain this transition,
but because of space limitation the reasons will not be discussed here.
When the voltage is applied for an infinitely long time (e.g. under d.c. or
60 Hz) the ionization products will have sufficient time to wander in the gap
and accumulate in space, causing a distortion in the original field.
To study this phenomenon, let us choose the rod-plane gap with the rod tip
of radius of (say) 1 cm as shown in Fig. 5.33 and study the various discharge
modes together with the breakdown characteristics for this arrangement in
atmospheric air. Then if the gap length is small (less than about 2 cm) and

7
6
5
4
3
2
1
0 0.5 1.0 1.5 2.0 2.5
Distance from point (cm)
Propagation
velocity
of
axial
streamers
(10
6
m/sec)
38.7 kV
25.0 kV
Figure 5.32 Streamer velocity in a gap of 2.5 cm under two different
voltages of fast rise in air43
50
0 5
1
2
3
4
d
r
Onset streamers
No ionization
Gap spacing (d) (cm)
d.c.
voltage
(kV)
Glow
Spark
r = 1 cm
Breakdown streamers
10 15 20 25 30 35
100
150
200
4′
Figure 5.33 Threshold curves for various modes of anode corona and for
spark breakdown for a hemispherically capped anode and plate cathode

the voltage is gradually raised no appreciable ionization is detected up to
breakdown. As the gap is increased, the field distribution becomes more
inhomogeneous, and on increasing the voltage at first a transient slightly
branched filamentary discharge appears. These discharges have been shown
to be identical with those observed under impulse voltages and are also called
streamers. Under steady state the streamer develops with varying frequencies,
giving rise to currents that are proportional to their physical length. These
streamers are sometimes called onset streamers or burst pulses.
When the voltage is increased further, the streamers become more frequent,
until the transient activity stops, the discharge becomes self-sustained and a
steady glow appears close to the anode. This glow gives rise to continuous but
fluctuating current. A further increase in voltage increases the luminosity of
the glow both in area and in the intensity. It should be noted that glow corona
develops only in the presence of negative ions. On increasing the voltage still
further, new and more vigorous streamers appear which ultimately lead to
complete breakdown of the gap.
The onsets of the various discharge modes observed, as the gap length is
increased, are illustrated schematically in Fig. 5.33 together with the corre-
sponding discharge characteristics. At the smaller spacing when the voltage
is still reasonably uniform the streamer is capable of penetrating the weaker
field, reaching the cathode and initiating breakdown in the same manner as
in uniform field gaps. This condition is shown by curve 1 of Fig. 5.33. With
the larger spacing above 10 cm, streamers appear that do not cross the gap
(shown by curve 2). Curve 3 represents transition from streamers to steady
glow corona without sparking. At the larger spacings there is a consider-
able spread in the voltage at which breakdown streamers develop preceding
the complete breakdown of the gap. The dashed area represents the region
of uncertain transitions; portion 1 indicates the onset of streamers followed
immediately by transition to spark. If, however, the gap is increased to a point
where glow is established and then reduced keeping the voltage constant, the
glow discharge will have stabilized the gap against breakdown at a voltage
that otherwise would have broken down. If the voltage is then raised, a spark
is induced by glow corona (curve 4), but if it is lowered, a streamer breakdown
is induced. By decreasing the gap further to lower values and increasing the
voltage at the various points the glow-corona sparking voltage characteristic
can be projected backwards as shown by curve 4. Thus if a steady corona
glow is established, the sparking voltage is raised and the lower breakdown
by streamer is suppressed.
5.11.2 Negative or cathode corona
With a negative polarity point-plane gap under static conditions above the
onset voltage the current flows in very regular pulses as shown in Fig. 5.34(b),

which indicates the nature of a single pulse and the regularity with which the
pulses are repeated. The pulses were studied in detail by Trichel45
and are
named after their discoverer as ‘Trichel pulses’. The onset voltage is prac-
tically independent of the gap length and in value is close to the onset
of streamers under positive voltage for the same arrangement. The pulse
frequency increases with the voltage and depends upon the radius of the
cathode, the gap length and the pressure. The relationship between the pulse
frequency and the gap voltage for different gap lengths and a cathode point
of 0.75 mm radius in atmospheric air is shown in Fig. 5.34(a). A decrease in
pressure decreases the frequency of the Trichel pulses.
300
250
200
Trichel
pulse
frequency
(kHz)
150
100
50
0 10 20 30
Voltage (kV)
40 50 60
1 cm
2 cm
5 cm
+
−
Figure 5.34 Trichel pulse frequency–voltage relationship for different gap
lengths in air r D 0.75 mm
Figure 5.35 illustrates the onset voltage of different negative coronas plotted
as a function of electrode separation for a typical example of a cathode of
0.75 mm radius. The lowest curve gives the onset voltage for Trichel pulses
not greatly affected by the gap length. Raising the voltage does not change
the mode of the pulses over a wide voltage range. Eventually at a much
higher voltage a steady glow discharge is observed, but the transition from
Trichel pulses to glow discharge is not sharply defined and is therefore shown
as a broad transition region in Fig. 5.35. On increasing the voltage further,
the glow discharge persists until breakdown occurs. It should be noted that
breakdown under negative polarity occurs at considerably higher voltage than
under positive voltage, except at low pressures; therefore, under alternating

0 1 2 3 4 5 6
Gap length (cm)
20
40
60
80
100
120
d.c.
voltage
(kV)
r = 0.75 mm
(−ve)
+ Spark
Glow
Transition region
Trichel pulses
No ionization
(a) (b)
50 m sec
Figure 5.35 Negative rod-plane breakdown and corona characteristics in
atmospheric air (a) discharge modes, (b) pulse pattern
power frequency voltage the breakdown of non-uniform field gap invariably
takes place during the positive half-cycle of the voltage wave.
5.12 Polarity effect – influence of space charge
It was shown in Fig. 5.33 that in non-uniform field gaps in air the appearance
of the first streamer may lead to breakdown or it may lead to the estab-
lishment of a steady state corona discharge which stabilizes the gap against
breakdown. Accordingly we may have a corona stabilized or direct break-
down. This subject is been discussed in section 5.11. Whether direct or corona
stabilized breakdown occurs depends on factors such as the degree of field
non-uniformity, gas pressure, voltage polarity and the nature of the gas. For
example, in air the corona stabilized breakdown will extend to higher pressures
than in SF6 due to the relatively immobile SF6 ions (Figs 5.36 and 5.37).
Figure 5.36 compares the positive and negative point-plane gap breakdown
characteristics measured in air as a function of gas pressure. At very small
spacing the breakdown characteristics for the two polarities nearly coincide
and no corona stabilized region is observed. As the spacing is increased, the
positive characteristics display the distinct high corona breakdown up to a
pressure of approximately 7 bar, followed by a sudden drop in breakdown
strengths. Under the negative polarity the corona stabilized region extends to
much higher pressures.

0
20
40
60
80
100
120
140
160
180
200
5 10 15 20 25 30 35 40
Pressure (bar)
d.c.
voltage
(kV)
1 mm
1 mm
5 mm
20 mm
20 mm
Gap length
Figure 5.36 Point-plane breakdown and corona inception characteristics in
air: positive point, -- negative point (radius of curvature of point
r D 1 mm)
A practical non-uniform field geometry that is frequently used in the
construction of h.v. apparatus is the coaxial cylindrical arrangement. By
properly choosing the radial dimensions for the cylinders it is possible to
optimize such a system for the maximum corona-free breakdown.
Let us consider a system of two coaxial cylinders with inner and outer radii
ri and ro respectively. Then it can be readily shown that in the interelectrode
space at radial distance r the field strength is given by
Er D
V
r ln
ro
ri
where V is the applied voltage. Since breakdown or corona onset will follow
when the voltage stress at the smaller wire reaches the breakdown stress Eb
we can write the above equation as
Vb D Ebri ln
ro
ri
. 5.119

0 1 2 3 4 5 6
Pressure (bar)
50
100
150
200
V
i
′
V
b
(kV)
(−ve)
(−ve)
(+ve)
(+ve)
d = 20 mm
d
r
d.c.
Vb
Vi
Figure 5.37 D.C. corona inception and breakdown voltage in
SF6 –rod-plane gap d D 20 mm; r D 1 mm46
The maximum breakdown voltage for the system is obtained by differentiating
eqn (5.119) with respect to ri. In eqn (5.119) Eb is the breakdown (or corona
inception) field strength of the system. It was shown earlier that this field
strength depends upon the gas density as well as the radius r D ri of the inner
conductor. Neglecting this dependency, which would hold approximately for
not too small radii ri and/or strongly attaching gases (with a steep increase of
˛/p D fE/p, we may assume that Eb is a constant value. Then, keeping
ro, constant this condition gives the optimal design for the system.
dVb
dri
D Eb

ln
ro
ri
1

D 0
or
ro
ri
D e
and
Vbmax D Ebri. 5.120
Figure 5.38 shows the functional relationship between the breakdown voltage
and the radius of the inner cylinder for a fixed radius ro of the outer cylinder.

1/e = 0.368 ri = r0 ri /r0
0
Radius of inner cylinder
(Vb /Eb)max
Vb /Eb
(Breakdown/corona
conception)
Figure 5.38 Relationship between breakdown voltage and inner radius in a
coaxial cylinder system
The maximum breakdown voltage is also indicated. The dotted curve indicates
quantitatively the corona onset voltage and the solid curve the breakdown
voltage.
At low pressures the breakdown voltage is usually lower when the smaller
electrode is negative. The effect is due to the higher field at the cathode
so that is greater and therefore a lower value is needed for exp
d
o ˛dx to
satisfy the sparking criterion equation. Figure 5.39 shows the direct breakdown
voltage characteristics for nitrogen at low pressures between a wire and coaxial
cylinder. At higher pressures the order of the characteristics is reversed. The
large polarity effect at the higher pressure can be qualitatively explained by
considering the role of the space charge of the prebreakdown current.
If we consider the case of a positive point-plane gap shown in Fig. 5.40(a)
then an ionization by electron collision takes place in the high field region close
to the point. Electrons because of their higher mobility will be readily drawn
into the anode, leaving the positive space charge behind. The space charge
will cause a reduction in the field strength close to the anode and at the same
time will increase the field further away from it. The field distortion caused
by the positive space charge is illustrated in Fig. 5.40(b). The dotted curve
represents the original undistorted field distribution across the gap while the
solid curve shows the distorted field. The high field region is in time moving
further into the gap extending the region for ionization. The field strength at
the tip of the space charge may be high enough for the initiation of a cathode-
directed streamer which subsequently may lead to complete breakdown. With

1300
1200
1100
1000
900
800
700
600
500
400
300
200
0 1 2 3 4 5 6 7 8 9 10 11
P (mm Hg)
V
1
2
Figure 5.39 Breakdown voltage curves for nitrogen between a wire and a
coaxial cylinder (radii 0.083 and 2.3 cm respectively); curve 1 refers to a
positive wire, curve 2 to a negative wire
(a) (b)
E (x)
Without space charge
With space charge
x
Figure 5.40 (a) Space charge build-up in positive point-plane gap.
(b) Field distortion by space charge
the negative point (Fig. 5.40) the electrons are repelled into the low field
region and in the case of attaching gases become attached to the gas molecules
and tend to hold back the positive space charge which remains in the space
between the negative charge and the point. In the vicinity of the point the
field is grossly enhanced, but the ionization region is drastically reduced.

The effect is to terminate ionization. Once ionization ceases, the applied field
sweeps away the negative and positive ion space charge from the vicinity of the
point and the cycle starts again after the clearing time for the space charge.
To overcome this retarding action of the ions a higher voltage is required,
and hence negative breakdown voltage is higher than the positive breakdown
voltage in gaps with marked asymmetrical fields.
Mathematically at any given time the voltage across the gap is given by the
field integral

Ex dx D V. Integration of the space charge distorted field in
Figs 5.40 and 5.41 respectively shows immediately that
VbCpoint Vbpoint.
(a) (b)
x
E(x)
Without space charge
With space charge
Figure 5.41 (a) Space charge build-up in negative point-plane gap.
(b) Field distortion by space charge
5.13 Surge breakdown voltage–time lag
For the initiation of breakdown an electron must be available to start the
avalanche. With slowly rising voltages (d.c. and a.c.) there are usually
sufficient initiatory electrons created by cosmic rays and naturally occurring
radioactive sources. Under surge voltages and pulses of short duration,
however, the gap may not break down as the peak voltage reaches the lowest
breakdown value Vs unless the presence of initiatory electrons is ensured
by using artificial irradiation. Vs is a voltage which leads to breakdown of
the gap after a long time of application. With weak irradiation the peak value
may have to be greatly increased so that the voltage remains above the d.c.
value Vs for long intervals of time. Figure 5.42 illustrates the breakdown on
a step-function voltage pulse; Vp represents the peak value of a step voltage
applied at time t D 0 to a gap that breaks down under Vs after a long time.

V(t)
Vs
t
t
ts
t0
tf
Vp
Figure 5.42 Time lag components under a step voltage. Vs minimum static
breakdown voltage; Vp peak voltage; ts statistical time lag; tf formative time
lag
The time which elapses between the application of voltage to a gap sufficient
to cause breakdown and the breakdown is called the time lag t. It consists of
two components: one is the time which elapses during the voltage application
until a primary electron appears to initiate the discharge and is known as the
statistical time lag ts; and the other is the time required for the breakdown
to develop once initiated and is known as the formative time lag tf.
The statistical time lag depends upon the amount of preionization in the
gap. This in turn depends upon the size of the gap and the radiation producing
the primary electrons. The appearance of such electrons is usually statistically
distributed. The techniques generally used for irradiating gaps artificially, and
thereby reducing the statistical time lag, include the use of u.v. light, radio-
active materials and illumination by auxiliary sparks. The statistical time will
also be greatly reduced by the application of an overvoltage Vp Vs to
the gap.
The formative time lag tf depends essentially upon the mechanism of
spark growth in question. In cases when the secondary electrons arise entirely
from electron emission at the cathode by positive ions, the transit time from
anode to cathode will be the dominant factor determining the formative time.
The formative time lag increases with the gap length and the field non-
uniformity, but it decreases with the applied overvoltage.
5.13.1 Breakdown under impulse voltages
An impulse voltage is a unidirectional voltage which rises rapidly to a
maximum value and then decays slowly to zero. The exact definition of a
standard impulse voltage was presented in Chapter 2.

When an impulse voltage of a peak value higher than Vs is applied to a gap,
as shown in Fig. 5.43, there is a certain probability but not a certainty that
breakdown will follow. For breakdown it is essential that the spark develops
during the interval of overvoltage [Vt Vs] duration, i.e. the overvoltage
duration must exceed the time lag [t t2 t1]. For a given impulse voltage
waveshape the overvoltage duration will increase with the voltage ampli-
tude Vp.
Vp
Vs
t
t
V(t)
ts
t1 t2
tf
Figure 5.43 Breakdown under impulse voltage
Because of the statistical nature of the time lags, when a given number of
impulses of an amplitude Vp, exceeding the static value Vs, are applied to a
gap only a certain percentage will lead to breakdown. We therefore obtain a
breakdown probability P for each given applied maximum impulse voltage
Vp as a function of Vp. This subject will be discussed in Chapter 8.
5.13.2 Volt–time characteristics
When an impulse voltage of sufficiently high value is applied to a gap, break-
down will result on each voltage application. The time required for the spark
development (time lag) will depend upon the rate of rise of voltage and the
field geometry. Therefore, for each gap geometry it is possible to construct
a volt–time characteristic by applying a number of impulses of increasing
amplitude and noting oscillographically the time lag. A schematic plot of
such a characteristic is shown in Fig. 5.44. In uniform and quasi-uniform field
gaps the characteristic is usually sharply defined and it rises steeply with
increasing the rate of rise of the applied voltage. In non-uniform field gaps,
however, due to larger scatter in the results, the data fall into a dispersion
band as shown in Fig. 5.45. The time to breakdown is less sensitive to the
rate of voltage rise. Hence, quasi-uniform field gaps (sphere–sphere) have
often been used as protective devices against overvoltages in electric power

T5 T2 t
Time
V (t)
(Vb)5
(Vb)2
Voltage
V − t characteristic
5
4
3
2
1
Figure 5.44 Impulse ‘volt–time’ characteristics
Non-uniform field
Uniform field gap
Time (t)
Breakdown
voltage
V
(
t
)
Figure 5.45 Schematic diagram of volt–time characteristics for uniform
and non-uniform field gaps
systems. The volt–time characteristic is an important practical property of
any insulating device or structure. It provides the basis for establishing the
impulse strength of the insulation as well as for the design of the protection
level against overvoltages and will be discussed in Chapter 8.
5.13.3 Experimental studies of time lags
Numerous investigators have studied time lags in the past. In the techniques
generally used either a constant voltage is applied to an irradiated gap and a
spark is initiated by a sudden illumination of the gap from a nearby spark, or
an overvoltage is suddenly applied to a gap already illuminated.

In the former case the time lag is measured from the flash until breakdown
occurs, while in the latter the time lag is measured between the voltage applica-
tion and the gap breakdown. The overvolted conditions may be obtained either
by superimposing a step voltage pulse upon a direct voltage already applied to
the gap or by using an impulse voltage of a suitably short front duration. The
measured time lags for given experimental conditions are usually presented
graphically by plotting the average time lags against the overvoltage. The latter
is defined as the percentage ratio of the voltage in question to the minimum
direct voltage which will cause breakdown. In the case when an impulse
voltage is used on its own, the time lags are plotted against the impulse ratio
defined as the ratio of the applied impulse voltage to the minimum direct
breakdown voltage.
The measured values are affected by factors such as the intensity of the
background irradiation, the nature and the condition of the electrode surface,
the gap length, the electron affinity of the gas, etc.
With a gap illuminated from an intense u.v. source, time lags down
to 108
sec and shorter have been recorded in highly overvolted gaps.47
Figure 5.46 shows time lags of spark breakdown for short gaps with the
0 20 40 60 80 100 120
Time lag ( ×10−9 sec)
20
40
60
80
100
120
140
160
180
200
Overvoltage
(%)
Figure 5.46 Time lag of spark gap as a function of overvoltage for short
gap between spheres with intense u.v. illumination of the cathode in air

cathode irradiated by a quartz mercury lamp, obtained by Brayant and
Newman,47
between spheres in air. Fisher and Benderson48
studied time
lags in air between uniform field electrodes in slightly overvolted conditions
and the results obtained for four gap lengths are shown in Fig. 5.47. These
authors used different gas pressures and found that in the range of pressure
from 760 Torr down to about 200 Torr the results were independent of the gas
pressure.
1.00
0.75
0.50
0.25
0
0 10 20 30 40 50 60 70
Time lag (msec)
Overvoltage
(%)
I
II
III
IV
I − d = 1.4 cm
II − d = 1.0 cm
III − d = 0.6 cm
IV − d = 0.3 cm
Figure 5.47 Time lag as a function of overvoltage for four gap lengths in
air. The curves represent the average data for all pressures between
atmospheric and 200 mm Hg
Long and highly scattered time lags have been observed in strongly elec-
tronegative gases under irradiated conditions. Figure 5.48 compares time lags
observed in SF6 with those obtained in air under similar experimental condi-
tions. It was impossible49
to attribute these long time lags to the shortage
of initiatory electrons. It was suggested that the long time lags are associated
with the complex nature of the growth of spark in the highly electron attaching
gases.
An alternative method for presenting time lags has been developed by
Laue50
and Zuber.51
These authors showed that the time lag in spark gaps
may be represented in the form
n
N
D e

t
0
I1I2ˇ dt
5.121

1
0.8
0.6
0.4
0.3
0.2
0.01
8
5
4
3
0.02
n
/
N
0 5 10 15 20 25 30 35 40 45
e
d
c
b
a
Time lag (µsec)
Figure 5.48 Time lag distribution in SF6 and air. Overvoltages: (a) 10%;
(b) 15%; (c) 20%; (d) 25% for SF6 ; (e) 5.3% for air
where N represents the total number of time lags observed, n is the number of
time lags of length greater than t, ˇ the rate at which electrons are produced in
the gap by irradiation, I1 the probability of an electron appearing in a region
of the gap where it can initiate a spark, and I2 the probability that an electron
at a given field strength will lead to the development of the spark. The factor
I1 is a function of the gap length and the gas density, while I2 is a function
of the applied field. The factor ˇ is dependent on the source of irradiation.
Providing that the primary current in the gap is constant and the applied field
remains constant with respect to time, eqn (5.121) can be written as:
n
N
D ekt
. 5.122
Equation (5.122) gives a linear relation between ln n
N and time t. The method
gives a truer representation of the results in the case of highly scattered results.
References
1. B. Loeb. The Kinetic Theory of Gases. Wiley, New York, 1963, Chapter 2.
2. E.W. McDaniel. Collision Phenomena in Ionised Gases. Wiley, New York, 1964.
3. D. Ramp and P. Englander-Golden. J. Chem. Phys. 43 (1965), p. 1964.
4. J.S. Townsend. Electricity in Gases. Oxford Press, 1914.
5. H. Raether. Z. Phys. 117 (1941), pp. 375, 524.
6. A. von Angel. Ionised Gases (2nd edn), p. 181. Clarendon Press, 1965.
7. F.H. Sanders. Phys. Rev. 44 (1932), p. 667.
8. K. Marsch. Arch. Elektrotechnik 26 (1932), p. 598.

9. R. Geballe and M.A. Harrison. Phys. Rev. 85 (1952), p. 372.
10. G.K. Kachickas and L.H. Fischer. Phys. Rev. 91 (1953), p. 775.
11. M.H. Saba. Phil. Mag. 40 (1920), p. 472.
12. L.B. Loeb. Basic Processes of Gaseous Electronics. University of California Press, 1955.
13. L.B. Loeb. Formation of negative ions. Encyclopedia of Physics, Vol. 16, p. 445. Springer,
Berlin, 1956.
14. P. Langevin. Ann. Chim. Phys. 8 (1905), p. 238.
15. G.L. Weissler. Photoelectric emission from solids. Encyclopedia of Physics Vol. 16, p. 342.
16. L. Matter. Phys. Rev. 49 (1936), p. 879.
17. O.W. Richardson. The Emission of Electricity from Hot Bodies. Longmans Green, London,
1921.
18. W. Schottky. Ann. Phys. 44 (1914), p. 1011.
19. R.H. Fowler and L.W. Nordheim. Proc. Roy. Soc. London 119 (1928), p. 173; 124 (1929),
p. 699.
20. F. Llewellyn Jones and D.E. Davies. Proc. Phys. Soc. B64 (1951), p. 397.
21. J.M. Meek and J.D. Craggs. Electrical Breakdown of Gases. Clarendon Press, Oxford, 1953.
22. H. Raether. Electron Avalanches and Breakdown in Gases. Butterworths, London, 1964.
23. H. Raether. Z. Phys. 112 (1939), p. 464.
24. K.H. Wigner. Z. Phys. 189 (1966), p. 466.
25. J.M. Meek. Phys. Rev. 57 (1940), p. 722.
26. L.B. Loeb and J.M. Meek. The Mechanism of Electric Spark. Stanford University Press,
1940.
27. K. Dehne, W. Khörman and H. Lenne. Measurement of formative time lags for sparks in
air, H2, and N2. Dielectrics 1 (1963), p. 129.
28. D.H. Hale. Phys. Rev. 56 (1948), p. 1199.
29. J.J. Thomson and G.P. Thomson. Conduction of Electricity through Gases. 2 vols. New York:
Dover Publications Inc. 1969 (paper edition).
30. J. Cueilleron. C.R. Acad. Sci. Paris 226 (1948), p. 400.
31. W.O. Schumann. Arch. fur Elektrotechnik 12 (1923), p. 593.
32. H. Sohst. Zeitsch. fur Angew. Physik 14 (1962), p. 620.
33. G.A. Schöder, Zeitsch. fur Angew. Physik 13 (1961), p. 296.
34. F.M. Penning. Physica 1 (1934), p. 1028.
35. F.W. Peek. Dielectric Phenomena in High Voltage Engineering (2nd edn.), McGraw-Hill,
New York, 1920.
36. W.S. Zaengl and N.U. Nyffenegger. Proc. 3rd Int. Conf. on Gas Discharges, 1974, p. 303.
37. R. Geballe and M.L. Reeves. Phys. Rev. 92 (1953), p. 867.
38. W.S. Zaengl. Proc. 10th Symposium on Electrical Insulating Materials, Tokyo, 1977, p. 13.
39. T. Nitta and Y. Sakata. IEEE Trans. EI 89 (1971), pp. 1065–1071.
40. H. Fujinami et al., IEEE Trans. EI 18(4) (1983), pp. 429–435.
41. T. Takuma. IEEE Trans. EI 21(6) (1986), pp. 855–867.
42. W.S. Zaengl. Electronegative Gases, Present State of Knowledge and Application, Future
Prospects. Nordic Symposium on Electric Insulation, June 13–15, 1988, Trondheim,
pp. I1–I39.
43. L.B. Loeb. Electrical Coronas. University of California Press, 1965.
44. E. Nasser. IEEE Spectrum 5 (1968), p. 127.
45. G.W. Trichel. Phys. Rev. 55 (1939), p. 382.
46. R.L. Hazel and E. Kuffel. Trans. IEEE Pas. 95 (1976), p. 178.
47. J.M. Bryant and M. Newman. Trans. AIEE 59 (1940), p, 813.
48. L.H. Fischer and B. Benderson. Phys. Rev. 81 (1951), p. 109.
49. E. Kuffel and R. M. Radwan. Proc. IEE 113 (1966), p. 1863.
50. M. von Laue. Ann. Phys. Lpz. 76 (1925), p. 261.
51. K. Zuber. Ann. Phys. Lpz. 76 (1925), p. 231.

Chapter 6
Breakdown in solid and liquid dielectrics
6.1 Breakdown in solids
Solid insulation forms an integral part of high voltage structures. The solid
materials provide the mechanical support for conducting parts and at the same
time insulate the conductors from one another. Frequently practical insulation
structures consist of combinations of solids with liquid and/or gaseous media.
Therefore, the knowledge of failure mechanisms of solid dielectrics under
electric stress is of great importance.
In gases the transport of electricity is limited to positive and negative charge
carriers, and the destruction of insulating properties involves a rapid growth
of current through the formation of electron avalanches. The mechanism of
electrical failure in gases is now understood reasonably clearly. This is not
the case for solid insulation. Although numerous investigators have studied
the breakdown of solids for nearly a century now, and a number of detailed
theories have been put forward which aimed to explain quantitatively the
breakdown processes in solids, the state of knowledge in this area is still very
crude and inconclusive.
Electrical conduction studies in solids are obscured by the fact that the trans-
port phenomena besides electronic and ionic carriers include also currents due
to the slower polarization processes such as slow moving dipoles (orientation
polarization) and interfacial polarization (see Chapter 7, Section 7.1). Elec-
trical methods are unable to distinguish between the conduction currents and
the currents due to polarization having a longer time constant than the duration
of a particular experiment. At low stresses and normal temperatures conduc-
tion by free electrons and ions in solids is exceptional. Examples in which the
conduction is believed to be of the simple electrolytic type at room tempera-
ture and above are glasses. In this case the conduction–temperature relation
is found to be of the form
D A exp

u
kT

where A and u are empirical constants. Ceramics also develop a significant
conductivity at higher temperatures that may be electronic or ionic.

As the stress in solids is increased and approaches the breakdown stress,
the current is found to increase exponentially, but does not vary so markedly
with time for steady voltage.1
* This increased current at high stresses is
generally believed to result from the injection of carriers from an electrode or
from electron multiplication in the bulk of the material or both. In addition,
if impurities or structural defects are present they may cause local allowed
energy levels (traps) in the forbidden band, and electrons may pass through
the insulator by jumping from one trap to another (hopping effect).
From the electrodes the electrons are believed to be ejected by either
the ‘Schottky’s emission effect’ or the ‘field emission effect’ (tunnelling)
discussed already in Chapter 5. Once injected into the material the electron
multiplication is thought to be analogous to that in a gas discharge. Under
certain strictly controlled experimental conditions the breakdown of solids
may therefore be accomplished by a process similar to gas breakdown. Under
normal industrial conditions, however, the same solid materials are found to
exhibit a wide range of dielectric strength, depending upon the conditions of
the environment and the method of testing. The measured breakdown voltage is
influenced by a large number of external factors such as temperature, humidity,
duration of test, whether a.c., d.c., or impulse voltage is applied, pressure
applied to the electrodes, discharges in the ambient medium, discharges in
cavities and many other factors. The fundamental mechanisms of breakdown
in solids are understood much less clearly than those in gases; nevertheless,
several distinct mechanisms have been identified and treated theoretically.2–4
In this section the presently accepted breakdown mechanisms will be
discussed briefly in a qualitative manner. No conduction mechanism will
be discussed here and the reader is referred to reference 6. Broadly speaking
the mechanism of failure and the breakdown strength changes with the time of
voltage application and for discussion purposes it is convenient to divide the
time scale of voltage application into regions in which different mechanisms
operate, as shown in Fig 6.1.
6.1.1 Intrinsic breakdown
If the material under test is pure and homogeneous, the temperature and
environmental conditions are carefully controlled, and the sample is so stressed
that there are no external discharges. With undervoltages applied for a short
time the electric strength increases up to an upper limit which is called the
intrinsic electric strength. The intrinsic strength is a property of the material
and temperature only. Experimentally the intrinsic strength is rarely reached,
but numerous attempts have been made to measure it for various materials.
To achieve the highest strength the sample is so designed that there is a high
* Superscript numbers are to references at the end of the chapter.

Breakdown in solid and liquid dielectrics 369
Thermal
Erosion
electrochemical
BD
strength
(
E
b
)
0 10−8 10−4 100 104
sec
Intrinsic,
Avalanche,
Electromechanical
Figure 6.1 Mechanisms of failure and variation of breakdown strength in
solids with time of stressing
Low stress
High stress
Figure 6.2 Electrode arrangement used for measuring intrinsic breakdown
in solids
stress in the centre of the solid under test and too low stress at the edges
which cause discharge in the medium as shown in Fig. 6.2.
The intrinsic breakdown is accomplished in times of the order of 108
sec
and has therefore been postulated to be electronic in nature. The stresses
required for an intrinsic breakdown are well in excess of 106
V/cm. The
intrinsic strength is generally assumed to be reached when electrons in the
insulator gain sufficient energy from the applied field to cross the forbidden
energy gap from the valence to the conduction band. The criterion condition is
formulated by solving an equation for the energy balance between the gain of
energy by conduction electrons from the applied field and its loss to the lattice.
Several models have been proposed in an attempt to predict the critical value
of the field which causes intrinsic breakdown, but no completely satisfactory
solution has yet been obtained. The models used by various workers differ
from each other in the proposed mechanisms of energy transfer from conduc-
tion electrons to the lattice, and also by the assumptions made concerning the

distribution of conduction electrons. In pure homogeneous dielectric materials
the conduction and the valence bands are separated by a large energy gap, and
at room temperature the electrons cannot acquire sufficient thermal energy
to make transitions from valence to conduction band. The conductivity in
perfect dielectrics should therefore be zero. In practice, however, all crystals
contain some imperfections in their structures due to missing atoms, and more
frequently due to the presence of foreign atoms (impurities). The impurity
atoms may act as traps for free electrons in energy levels that lie just below
the conduction band, as illustrated schematically in Fig. 6.3.
Conduction
band
Shallow traps
Deep traps
∆W
W∆W
2(W−∆W)
Figure 6.3 Schematic energy level diagram for an amorphous dielectric
At low temperatures the trap levels will be mostly filled with electrons
caught there as the crystal was cooled down during its manufacture. At room
temperature some of the trapped electrons will be excited thermally into the
conduction band, because of the small energy gap between the trapping levels
and the conduction level. An amorphous crystal will therefore have some free
conduction electrons.
When a field is applied to a crystal the conduction electrons gain energy
from it, and due to collisions between them the energy is shared by all elec-
trons. For a stable condition this energy must be somehow dissipated. If there
are relatively few electrons such as in pure crystals, most of the energy will
be transferred to the lattice by electron–lattice interaction. In steady state
conditions the electron temperature (Te) will be nearly equal to the lattice
temperature (T).
In amorphous dielectrics the electron interactions predominate, the field
raises the energy of the electrons more rapidly than they can transfer it to the
lattice, and the electron temperature Te will exceed the lattice temperature T.
The effect of the increased electron temperature will be a rise in the number of
trapped electrons reaching the conduction band. This increases the material’s
conduction and as the electron temperature continues to increase a complete
breakdown is eventually reached known as ‘high-temperature breakdown’.

Neglecting for the moment the details of the mechanism of energy transfer
and assuming electronic conduction in solids, for an applied field E the rate
of energy gained by electrons from the field will be a function of the field
strength E and the lattice temperature T. The rate at which this energy is
transferred to the lattice will depend only on T. In addition, both rates will
depend on parameters describing the conduction electrons. If we denote these
parameters collectively by ˛, then for the steady state conditions the energy
equation for conduction electrons may be written as
AE, T, ˛ D BT, ˛ 6.1
where the l.h.s. represents the rate of energy gain by electrons from the
field, and the r.h.s., the rate of energy transfer from electrons to lattice.
Equation (6.1) can be physically satisfied for values of E below a certain
critical value Ec, and this value has been considered by several workers as the
intrinsic critical field. The value of Ec can be found by identifying correctly the
parameters ˛ describing the conduction electrons and then solving eqn (6.1)
for the critical field strength Ec.
For a pure homogeneous dielectric Fröhlich developed the so-called ‘high
energy’ breakdown criterion, based on the assumption that the dielectric is
destroyed by an infinitely large multiplication of electrons in the conduc-
tion band. In this model the critical field strength (Ec) in the energy balance
eqn (6.1) is obtained by first identifying the parameter ˛ with the electron
energy (We) such that the balance equation is satisfied and then calculating
the critical field strength.
The functional relationship between the parameters in eqn (6.1) is shown
schematically in Fig. 6.4, which shows the average rate of energy gain from
the field for various field strengths and the rate of energy loss to the lattice.
For the critical field criterion, eqn (6.1) becomes
AEc, T, I D BT, I 6.2
where I is the ionization energy corresponding to the transition of an electron
from a valence band to a conduction band. From Fig. 6.4 it is seen that for
an electron to remain accelerated and thus lead to instability at any given
field it should find itself with an energy which brings it above the curve B
so that it gains energy more rapidly than it loses. Equation (6.2) enables us
to determine the critical field strength Ec that is required to cause collision
ionization from valence to conduction band. For field strength exceeding Ec the
electrons gain energy more rapidly from the field than they lose to the lattice
and breakdown will result. The above mechanism applies to pure solids in
which the equilibrium is controlled by collisions between electrons and the
lattice vibrations.
Fröhlich and Paranjape5
have extended this model to amorphous materials
in which the concentration of conduction (or trapped) electrons is high enough

Rate
of
energy
gain
or
loss
d
W
/d
t
B
A(E2T, We)
A(E1T, We)
A(Ec′T, We)
W
W1
W2
W ′ I
Figure 6.4 The average rate of energy gain AE, T, We from an applied
field for various field strengths and the average rate of energy loss to lattice
BWL, T
to make electron–electron collisions the dominant factor. In this case it is
necessary to calculate the electron temperatureTe which will be higher than
the lattice temperature T.
The energy balance eqn (6.1) will then take the form
AE, Te, T D BTe, T. 6.3
This relationship is plotted schematically in Fig. 6.5 in which the family of
curves plotted for various values of E represents the l.h.s. of the equation and
Rate
of
energy
gain
or
loss
A
E = E2
E = Ec
E = E1
T1 T2
Tc Tc
B
Electron temperature
Figure 6.5 Rate of energy gain and loss for h.t. intrinsic breakdown model

the single curve represents the r.h.s. The intersections give possible solutions
for the various electron temperatures.
For the analytical expressions for the critical field strength (Ec) for both of
the above two models the reader should refer to reference 6.
To date there has been no direct experimental proof to show whether an
observed breakdown is intrinsic or not, except for plastic materials such as
polyethylene and so conceptually it remains an ideal mechanism identified as
the highest value obtainable after all secondary effects have been eliminated.
6.1.2 Streamer breakdown
Under certain controlled conditions in strictly uniform fields with the elec-
trodes embedded in the specimen, breakdown may be accomplished after the
passage of a single avalanche. An electron entering the conduction band of
the dielectric at the cathode will drift towards the anode under the influence
of the field gaining energy between collisions and losing it on collisions. On
occasions the free path may be long enough for the energy gain to exceed the
lattice ionization energy and an additional electron is produced on collision.
The process is repeated and may lead to the formation of an electron avalanche
similar to gases. Seitz7
suggested that breakdown will ensue if the avalanche
exceeds a certain critical size and derived an expression for a single avalanche
breakdown strength. The concept is similar to the streamer theory developed
by Raether, and Meek and Loeb for gases discussed earlier.
6.1.3 Electromechanical breakdown
Substances which can deform appreciably without fracture may collapse when
the electrostatic compression forces on the test specimen exceed its mechan-
ical compressive strength. The compression forces arise from the electrostatic
attraction between surface charges which appear when the voltage is applied.
The pressure exerted when the field reaches about 106
V/cm may be several
kN/m2
. Following Stark and Garton,8
if d0 is the initial thickness of a spec-
imen of material of Young’s modulus Y, which decreases to a thickness of
d (m) under an applied voltage V, then the electrically developed compressive
stress is in equilibrium with the mechanical compressive strength if
ε0εr
V2
2d2
D Y ln

d0
d

6.4
or
V2
D d2 2Y
ε0εr
ln

d0
d

where ε0 and εr are the permittivity of free space and the relative permittivity
of the dielectric.
Differentiating with respect to d we find that expression (6.4) has a
maximum when d/d0 D exp[1/2] D 0.6. Therefore, no real value of V
can produce a stable value of d/d0 less than 0.6. If the intrinsic strength
is not reached at this value, a further increase in V makes the thickness
unstable and the specimen collapses. The highest apparent strength is then
given by
Ea D
V
d0
D 0.6

Y
ε0εr
1/2
6.5
This treatment ignores the possibility of instability occurring in the lower
average field because of stress concentration at irregularities, the dependence
of Y on time and stress, and also on plastic flow.
6.1.4 Edge breakdown and treeing
In practical insulation systems, the solid material is stressed in conjunction
with one or more other materials. If one of the materials is, for example, a
gas or a liquid, then the measured breakdown voltage will be influenced more
by the weak medium than by the solid.
A cross-section of a simplified example is shown in Fig. 6.6 which
represents testing of a dielectric slab between sphere-plane electrodes. Ignoring
the field distribution, i.e. assuming a homogeneous field, if we consider an
Puncture due to
ambient discharge
Puncture not due to
ambient discharge
Solid
Ambient
dA
X
B
d1
d2
∈1
∈2
Figure 6.6 Breakdown of solid specimen due to ambient discharge-edge
effect
elementary cylindrical volume of end area dA spanning the electrodes at
distance x as shown in Fig. 6.5, then on applying the voltage V between
the electrodes, according to Section 4.3.1 a fraction V1 of the voltage appears

across the ambient given by
V1 D
Vd1
d1 C

ε1
ε2

d2
6.6
here d1 and d2 represent the thickness of the media 1 and 2 in Fig. 6.6 and ε1
and ε2 are their respective permittivities. For the simple case when a gaseous
dielectric is in series with a solid dielectric stressed between two parallel
plate electrodes, the stress in the gaseous part will exceed that of the solid by
the ratio of permittivities or E1 D εrE2. For the case shown in Fig. 6.6, the
stress in the gaseous part increases further as x is decreased, and reaches very
high values as d1 becomes very small (point B). Consequently the ambient
breaks down at a relatively low applied voltage. The charge at the tip of
the discharge will further disturb the applied local field and transform the
arrangement to a highly non-uniform system. The charge concentration at the
tip of a discharge channel has been estimated to be sufficient to give a local
field of the order of 10 MV/cm, which is higher than the intrinsic breakdown
field. A local breakdown at the tips of the discharge is likely, therefore, and
complete breakdown is the result of many such breakdown channels formed
in the solid and extending step by step through the whole thickness.
The breakdown event in solids in general is not accomplished through the
formation of a single discharge channel, but assumes a tree-like structure
as shown in Fig. 6.7 which can be readily demonstrated in a laboratory by
applying an impulse voltage between point-plane electrodes with the point
embedded in a transparent solid, e.g. plexiglass. The tree pattern shown in
Fig. 6.7 was recorded by Cooper9
with a 1/30 µsec impulse voltage of the
same amplitude. After application of each impulse the channels were observed
with a microscope and new channels were recorded. Not every impulse will
produce a channel. The time required for this type of breakdown under alter-
nating voltage will vary from a few seconds to a few minutes.
The tree-like pattern discharge is not limited specifically to the edge effect
but may be observed in other dielectric failure mechanisms in which non-
uniform field stresses predominate.
6.1.5 Thermal breakdown
When an insulation is stressed, because of conduction currents and dielectric
losses due to polarization, heat is continuously generated within the dielec-
tric. In general, the conductivity () increases with temperature, conditions of
instability are reached when the rate of heating exceeds the rate of cooling
and the specimen may undergo thermal breakdown. The situation is illustrated
graphically in Fig. 6.8 in which the cooling of a specimen is represented by the
straight line and the heating at various field strengths by curves of increasing

1
1
1
1
2
3
4
5
6
7
8
8
8
8
8
8
6
3
5
8
4 8
9
10
10
11
11
11
11
13
15
12
12 13
13
13 14
14
15
16
16
18
17
Figure 6.7 Breakdown channels in plexiglass between point-plane
electrodes. Radius of point D 0.01 in; thickness 0.19 in. Total number of
impulses D 190. Number of channels produced D 16; n point indicates end
of nth channel. Radii of circles increase in units of 102
in
Heat
gain
or
loss
Heat gain
Field 3
Heat gain
Field 1
Field 2
Newton's law
of cooling
Specimen temperature
T2
T1
Ambient
Figure 6.8 Thermal stability or instability under different applied fields

slope. Field (1) is in equilibrium at temperature T1, field (2) is in a state of
unstable equilibrium at T2 and field (3) does not reach a state of equilibrium
at all. To obtain the basic equation for thermal breakdown let us consider a
cube of face area A m2
within dielectric. Assume that the heat flow in the
x-direction is as shown in Fig. 6.9, then the
heat flow across face 1 D KAdT
dx (K-thermal conductivity).
heat flow across face 2 D KAdT
dx C KA d
dx

dT
dx

x.
1
2
dT
dx ∆x
A (m2)
x
Heat
Direction of
heat flow
Figure 6.9 Heat input and output, cubical specimen
The second term represents the heat input into the block.
Hence
heat flow/volume K d
dx

dT
dx

D div K grad T.
The conservation of energy requires that heat input into the element must be
equal to the heat conducted away, plus the heat used to raise the temperature
T of the solid or
heat generated D heat absorbed C heat lost to surroundings,
i.e.
Cv
dT
dt
C div K grad T D E2
6.7
where Cv is the thermal capacity of the dielectric, is the electrical conduc-
tivity and in the case of alternating voltage the heat is generated primarily as
a result of dipole relaxation and the conductivity is replaced by ωε0ε
00
r where
ε0 represents permittivity of free space and ε
00
r the imaginary component of
the complex relative permittivity of the material.

Calculation of the critical thermal situation involves the solution of
eqn (6.7). In solving it, one assumes that a critical condition arises and
the insulation properties are lost, when at some point in the dielectric the
temperature exceeds a critical temperature Tc. The solution gives the time
required to reach Tc for a given field and boundary condition. The equation
cannot be solved analytically for the general case since Cv, K and may be
all functions of temperature (T) and may also depend upon the applied field.
We consider two extreme cases for the solution of eqn (6.7).
Case 1. This assumes a rapid build-up of heat so that heat lost to surround-
ings can be neglected and all heat generated is used in raising the temperature
of the solid. We obtain an expression for ‘impulse thermal breakdown’ and
eqn (6.7) reduces to
Cv
dT
dt
D E2
.
To obtain the critical field Ec, assume that we apply a ramp function field.
Then
E D

Ec
tc

t
and
E2
D Cv
dT
dE
dE
dt
.
For the conductivity, we can assume
D 0 exp

u
kT

.
0 is here the conductivity at ambient temperature T0. Substituting for and
rearranging, we get
Ec
0
tc
Ec
0
Cv
E2
dE D
Tc
T0
exp
u
kT

dT.
For the case when
u × kT
and
Tc T0 Tc critical temperature
the solution of the r.h.s. is
Tc
T0
exp
u
kT

dT ! T2
0
k
u
exp

u
kT0

and that of the l.h.s. is
Ec
0
tc
Ec
0
Cv
E2
dE !
1
3
tc
0
Cv
E2
c.
Therefore
Ec D
3CvkT2
0
0utc
0.5
exp

u
2kT0

. 6.8
It is seen that reaching the critical condition requires a combination of crit-
ical time and critical field and that the critical field is independent of the
critical temperature Tc due to the fast rise in temperature.
Case 2 concerns minimum thermal voltage, i.e. the lowest voltage for
thermal breakdown. For this case we assume a thick dielectric slab that is
constrained to ambient temperature at its surfaces by using sufficiently large
electrodes as shown in Fig. 6.10.
Electrode
Electrode
T0
T0
+V/2
−V/2
V = 0
Temperature
distribution
Tm
Vx ′ Tx
x
Figure 6.10 Arrangement for testing a dielectric for minimum thermal
breakdown voltage
On application of voltage, after some time, a temperature distribution within
the dielectric will be established with the highest temperature at the centre
(T1), that at the surface remaining at ambient temperature. On increasing the
voltage to a new higher value, an equilibrium will be established at a higher
central temperature (T2). If the process is continued, a thermal runaway will
eventually result as shown in Fig. 6.11.
To calculate the minimum thermal voltage, let us consider a point inside the
dielectric distance x from the centre, and let the voltage and temperature at
that point be Vx and Tx respectively. For this case we assume that all the heat
generated in the dielectric will be carried away to its surroundings through
the electrodes. Neglecting the term CvdT/dt, eqn (6.6) becomes
E2
D
d
dx

K
dT
dx

.

Temperature
T
T2
T1
T0
V1
V2
V1
V2
V3
V4
Runway
Equilibrium temperature
Time
0
Figure 6.11 Temperature–time relationship for slow thermal stressing
under various applied voltages
Using the relations of E D j and E D ∂V/∂x (j-current density), and inserting
in the above equation, we obtain
j
∂V
∂x
D
d
dx

K
dT
dx

.
Integrating to an arbitrary point x in the dielectric
j
Vx
0
dV D
x
0
d
dx

K
dT
dx

dx
jVx D K
dT
dx
or
Vx
dV
dx
D K
dT
dx
.
Substituting for D 0 exp[u/kT], and integrating from the centre of the
dielectric to the electrode,
Vc/2
0
VxdV D
K
0
Tc
T0
exp
u
kT

dT
V2
c D 8
K
0
Tc
T0
exp
u
kT

dT 6.9
Equation (6.9) gives the critical thermal breakdown voltage, where Tc is the
critical temperature at which the material decomposes and the calculation

assumes that Tc corresponds to the centre of the slab. The voltage is inde-
pendent of the thickness of the specimen, but for thin specimens the thermal
breakdown voltage becomes thickness dependent and is proportional to the
square root of the thickness tending asymptotically to a constant value for
thick specimens. Under alternating fields the losses are much greater than
under direct fields. Consequently the thermal breakdown strength is generally
lower for alternating fields, and it decreases with increasing the frequency
of the supply voltage. Table 6.1 shows thermal breakdown values for some
typical dielectrics under alternating and direct voltages at 20°C. These results
correspond to a thick slab of material.
The thermal breakdown is a well-established mechanism, therefore the
magnitude of the product ε tan υ which represents the loss is a very essential
parameter for the application of insulation material.
Table 6.1 Thermal breakdown voltages for some typical dielectrics 20°C
Material Thermal voltage
in MV/cm
d.c. a.c.
Crystals: Mica muscovite 24 7–18
Rock salts 38 1.4
Quartz: Perpendicular to axis 12 000 –
Parallel to axis 66 –
Impure – 2.2
Ceramics: H.V. steatite – 9.8
L.F. steatite – 1.5
High-grade porcelain – 2.8
Organic Capacitor paper – 3.4–4
materials: Ebonite – 1.45–2.75
Polythene – 3.5
Polystyrene – 5
Polystyrene at 1 MHz – 0.05
Acrylic resins 0.3–1.0
6.1.6 Erosion breakdown
Practical insulation systems often contain cavities or voids within the dielectric
material or on boundaries between the solid and the electrodes. These cavities

are usually filled with a medium (gas or liquid) of lower breakdown strength
than the solid. Moreover, the permittivity of the filling medium is frequently
lower than that of the solid insulation, which causes the field intensity in the
cavity to be higher than in the dielectric. Accordingly, under normal working
stress of the insulation system the voltage across the cavity may exceed the
breakdown value and may initiate breakdown in the void.
Figure 6.12 shows a cross-section of a dielectric of thickness d containing
a cavity in the form of a disc of thickness t, together with an analogue circuit.
In the analogue circuit the capacitance Cc corresponds to the cavity, Cb corre-
sponds to the capacitance of the dielectric which is in series with Cc, and Ca
is the capacitance of the rest of the dielectric. For t − d, which is usually the
case, and assuming that the cavity is filled with gas, the field strength across
Cc is given by the expression
Ec D εrEa
where εr is the relative permittivity of the dielectric.
A
d
t
Cb
Cc Vc
Ca Va
Figure 6.12 Electrical discharge in cavity and its equivalent circuit
For the simple case of a disc-shaped dielectric in solid shown in Fig. 6.12,
the discharge inception voltage applied across the dielectric can be expressed
in terms of the cavity breakdown stress. Assuming that the gas-filled cavity
breakdown stress is Ecb, then treating the cavity as series capacitance with the
healthy part of the dielectric we may write
Cb D
ε0εrA
d t
Cc D
ε0A
t
.
The voltage across the cavity is
Vc D
Cb
Cc C Cb
Va D
Va
1 C
1
εr

d
t
1
.

Therefore the voltage across the dielectric which will initiate discharge in the
cavity will be given by
Vai D Ecbt 1 C
1
εr

d
t
1

. 6.10
In practice a cavity in a material is often nearly spherical, and for such a case
the internal field strength is
Ec D
3εrE
εrc C 2εr
D
3E
2
6.11
for εr × εrc, where E is in the average stress in the dielectric, under an
applied voltage Va when Vc reaches breakdown value VC
of the gap t, the
cavity may break down. The sequence of breakdowns under sinusoidal alter-
nating voltage is illustrated in Fig. 6.13. The dotted curve shows qualitatively
the voltage that would appear across the cavity if it did not break down. As Vc
reaches the value VC
, a discharge takes place, the voltage Vc collapses and the
gap extinguishes. The voltage across the cavity then starts increasing again
until it reaches VC
, when a new discharge occurs. Thus several discharges
may take place during the rising part of the applied voltage. Similarly, on
decreasing the applied voltage the cavity discharges as the voltage across it
reaches V
. In this way groups of discharges originate from a single cavity
and give rise to positive and negative current pulses on raising and decreasing
the voltage respectively. For measurements of discharges refer to Chapter 7.
Va
Vc
i
t
V −
V +
Figure 6.13 Sequence of cavity breakdown under alternating voltages
When the gas in the cavity breaks down, the surfaces of the insulation
provide instantaneous cathode and anode. Some of the electrons impinging
upon the anode are sufficiently energetic to break the chemical bonds of the
insulation surface. Similarly, bombardment of the cathode by positive ions

may cause damage by increasing the surface temperature and produce local
thermal instability. Also channels and pits are formed which elongate through
the insulation by the ‘edge mechanism’. Additional chemical degradation may
result from active discharge products, e.g. O3 or NO2, formed in air which
may cause deterioration. Whatever is the deterioration mechanism operating,
the net effect is a slow erosion of the material and a consequent reduction of
the breakdown strength of the solid insulation.
When the discharges occur on the insulation surface, the erosion takes place
initially over a comparatively large area. The erosion roughens the surface
and slowly penetrates the insulation and at some stage will again give rise to
channel propagation and ‘tree-like’ growth through the insulation.
For practical application it is important that the dielectric strength of a
system does not deteriorate significantly over a long period of time (years).
In practice, however, because of imperfect manufacture and sometimes poor
design, the dielectric strength (e.g. in cables) decreases with the time of voltage
application (or the life) and in many cases the decrease in dielectric strength
(Eb) with time (t) follows the empirical relationship
tEn
b D const 6.12
where the exponent ‘n’ depends upon the dielectric material, the ambient
conditions, and the quality of manufacture. Figure 6.14 illustrates the case for
40
20
10
5
10−1
100
102
104
t (h)
E
D
(kV/mm)
n = 20
n = 12
n = 8
1a 10a 100a
Figure 6.14 Lifetime (t) stress relationship of polyethylene m.v. cables
determined by different manufacturers10

several m.v. polyethylene cables produced by different manufacturers. The
breakdown strength has been plotted against time on a log–log scale.
In earlier years when electric power distribution systems used mainly
paper-insulated lead-covered cables (PILC) on-site testing specifications called
for tests under d.c. voltages. Typically the tests were carried out at 4 to 4.5 V0.
The tests helped to isolate defective cables without further damaging good
cable insulation. With the widespread use of extruded insulation cables of
higher dielectric strength, the test voltage levels were increased to 5–8 V0. In
the 1970s premature failures of extruded dielectric cables that were factory
tested under d.c. voltage at specified levels were noted.26
Hence on-site
testing of cables under very low frequency (VLF), ¾0.1 Hz, has been adopted.
The subject has recently been reviewed by Gnerlich10
and will be further
discussed in Chapter 8.
6.1.7 Tracking
Tracking is the formation of a permanent conducting path, usually carbon,
across a surface of insulation and in most cases the conduction path results
from degradation of the insulation. For tracking to occur the insulation must
contain some organic substance.
In an outdoor environment insulation will in time become covered with
contaminant which may be of industrial or coastal origin. In the presence of
moisture the contamination layer gives rise to leakage current which heats
the surface and causes interruption in the moisture film; small sparks are
drawn between the separating moisture films. This process acts effectively
as an extension to the electrodes. The heat resulting from the small sparks
causes carbonization and volatilization of the insulation and leads to formation
of permanent ‘carbon track’ on the surface. The phenomenon of tracking
severely limits the use of organic insulation in the outdoor environment. The
rate of tracking depends upon the structure of the polymers and it can be
drastically slowed down by adding appropriate fillers to the polymer which
inhibit carbonization.
Moisture is not essential to tracking. The conducting path may arise from
metallic dust; for example, in oil-immersed equipment with moving parts
which gradually wear and deposit on the surface.
6.2 Breakdown in liquids
The general state of knowledge on the electrical breakdown in liquids is less
advanced than is in case of gases or even solids. Many aspects of liquid
breakdown have been investigated over the last decades, but the findings
and conclusions of the many workers cannot be reconciled and so produce

a general theory applicable to liquids, as the independent data are at vari-
ance and sometimes contradictory. The principal reason for this situation is
the lack of comprehensive theory concerning the physical basis of the liquid
state which would form the skeleton structure in which observations could be
compared and related.
Comprehensive reviews of the published data on the subject have been made
periodically and the more recent ones include the reviews of Lewis,11
Shar-
baugh and Watson,12
Swann,13
Kok,14
Krasucki,15
Zaky and Hawley,16
and Gallagher.17
The work falls broadly into two schools of thought. On the
one hand there are those who attempt to explain the breakdown of liquids on
a model which is an extension of gaseous breakdown, based on the avalanche
ionization of the atoms caused by electron collision in the applied field. The
electrons are assumed to be ejected from the cathode into the liquid by either
a field emission, in which case they are assumed to tunnel out through the
surface aided by the field, or by the field enhanced thermionic (Schottky’s)
effect. This type of breakdown mechanism has been considered to apply to
homogeneous liquids of extreme purity, and does not apply to commercially
exploited liquid insulation. Conduction studies in highly pure liquids showed
that at low fields the conduction is largely ionic due to dissociation of impu-
rities and increases linearly with the field strength. This conduction saturates
at intermediate fields. At high field, as we approach breakdown, the conduc-
tion increases more rapidly and tends to be unstable. It is believed that this
increased current results from electron emission at the cathode by one or both
of the above mechanisms, and possibly by field aided dissociation of molecules
in the liquid.
It has long been recognized that the presence of foreign particles in liquid
insulation has a profound effect on the breakdown strength of liquids. In one
approach it has been postulated14
that the suspended particles are polarizable
and are of higher permittivity than the liquid. As a result they experience an
electrical force directed towards the place of maximum stress. With uniform
field electrodes the movement of particles is presumed to be initiated by surface
irregularities on the electrodes, which give rise to local field gradients. The
accumulation of particles continues and tends to form a bridge across the gap
which leads to initiation of breakdown.
The impurities can also be gaseous bubbles of lower breakdown strength
than the liquid, in which case on breakdown of the bubble the total
breakdown of the liquid may be triggered. A mathematical model for
bubble breakdown has been proposed by Kao.18
6.2.1 Electronic breakdown
Both the field emission and the field-enhanced thermionic emission mech-
anisms discussed earlier have been considered responsible for the current

at the cathode. Conduction studies in insulating liquids at high fields show
that most experimental data for current fit well the Schottky-type equation
(eqn (5.81)19–Chapter 5
) in which the current is temperature dependent. Break-
down measurements carried out over a wide range of temperatures, however,
show little temperature dependence. This suggests that the cathode process is
field emission rather than thermionic emission. It is possible that the return of
positive ions and particularly positively charged foreign particles to the cathode
will cause local field enhancement and give rise to local electron emission.
Once the electron is injected into the liquid it gains energy from the applied
field. In the electronic theory of breakdown it is assumed that some electrons
gain more energy from the field than they lose in collisions with molecules.
These electrons are accelerated until they gain sufficient energy to ionize
molecules on collisions and initiate avalanche.
The condition for the onset of electron avalanche is obtained by equating
the gain in energy of an electron over its mean free path to that required for
ionization of the molecule.
eE( D chv 6.13
where E is the applied field, ( the electron mean free path, hv the quantum
of energy lost in ionizing the molecule and c an arbitrary constant.
Typical strengths for several highly pure liquids are included in Table 6.2.
Table 6.2 Electric strength of
highly purified liquids
Liquid Strength (MV/cm)
Hexane 1.1–1.3
Benzene 1.1
Good oil ¾1.0–4.0
Silicone 1.0–1.2
Oxygen 2.4
Nitrogen 1.6–1.88
The electronic theory satisfactorily predicts the relative magnitude of break-
down strength of liquids, but the observed formative time lags are much longer
than predicted by electronic theory.18
6.2.2 Suspended solid particle mechanism
Solid impurities may be present in the liquid either as fibres or as dispersed
solid particles. Let us consider a spherical particle of radius r and permittivity

ε to be suspended in dielectric liquid of permittivity εliq. Then in a field the
particle will become polarized and it will experience a force given by
Fe D εliq.r3 ε εliq.
ε C 2εliq.
E grad E. 6.14
This force is directed towards a place of maximum stress if ε εliq. but for
bubbles ε εliq., it has the opposite direction. The force given by eqn (6.14)
increases as the permittivity of the suspended particle (ε) increases, and for a
conducting particle for which ε ! 1 the force becomes
Fe D F1 D r3
E grad E. 6.15
Thus the force will urge the particle to move to the strongest region of the
field.
In a uniform field gap or sphere gap of small spacing the strongest field is
in the uniform region.
In this region grad E is equal to zero so that the particle will remain in
equilibrium there. Accordingly, particles will be dragged into the uniform field
region. If the permittivity of the particle is higher than that of the medium,
then its presence in the uniform field region will cause flux concentration
at its surface. Other particles will be attracted into the region of higher flux
concentration and in time will become aligned head to tail to form a bridge
across the gap. The field in the liquid between the particles will be enhanced,
and if it reaches critical value breakdown will follow.
The movement of particles by electrical force is opposed by viscous drag,
and since the particles are moving into the region of high stress, diffusion
must also be taken into account. For a particle of radius r slowly moving with
a velocity v in a medium of viscosity ,, the drag force is given by Stokes
relation
Fdrag D 6-r,vx 6.16
Equating the electrical force with the drag force (Fe D Fdrag) we obtain
vE D
r2
E
6-,
dE
dx
6.17
where vE is the velocity of the particle towards the region of maximum stress.
If the diffusion process is included, the drift velocity due to diffusion will be
given by the equation
vd D
D
N
dN
dx
D

kT
6-r,

dN
Ndx
. 6.18
The relation on the r.h.s. of the equation follows from the Stokes–Einstein
relation D D kT/67-r,, where k is Boltzmann’s constant and T is the absolute

temperature. Equating vE with vd gives
r2
6-r,
E
dE
dx
D

kT
6-r,rN

dN
dx
. 6.19
This introduces breakdown strength dependence in time on concentration of
particles N, their radii and the liquid viscosity. The critical value of transverse
field Ex, the equilibrium value above which breakdown will occur sooner
or later, can be obtained from integration of eqn (6.19).
r2
E2
2
EDEx
EDE1
D

kT
r
ln N
NDNx
NDN1
Nx
N1
D exp r3
fE2
x E2
1g
2kT
. 6.20
If the increase in the electrostatic energy when the particles drift towards
a place of maximum stress is much smaller than their kinetic energy, i.e.
r3

E2
x E2
1

− 2kT, the life of the insulation is infinite. The criterion
for breakdown resulting from movement of particles towards the high stress
region corresponds to the condition
r3

E2
x E2
1

D 2kT. 6.21
If we consider the case where the initial non-uniformity of field is caused by
a hemispherical hump on the electrode, discussed earlier in Chapter 4, and
assume that an applied field E0 will lead to breakdown after a long time of
application, then the maximum stress at the tip of the sphere is 3E0, or in
general the maximum stress is gE0, where g is a geometrical factor. Then
eqn (6.21) can be written as
r3
[g2
1]E2
0 D 1
4
kT. 6.22
For g D 3 we obtain
r3
E2
0 D 1
4 kT. 6.23
A more complete theory gives a relation which takes into account the permit-
tivities and is of the form
ε εliq.
ε C 2εliq.
r2
E2
0 D
1
4
kT.
Equation (6.23) gives a breakdown strength E0 after a long time as a function
of the size of the suspended impurities. This relationship has been checked
experimentally and reasonable agreement has been obtained with calculations.

100
50
20
10
20 50 100
E
0
(kV/mm)
2r (dia.) Å
Figure 6.15 Breakdown strength E0 after a long duration of time as a
function of the diameter 2r of foreign particles of high permittivity, with
T D 300 K18
Figure 6.15 shows a plot of eqn (6.23) for a range of sizes up to 50 A in
radius at temperature T D 300 K, for the case where εliq. − ε.
6.2.3 Cavity breakdown
Insulating liquids may contain gaseous inclusions in the form of bubbles. The
processes by which bubbles are formed include:
(i) gas pockets on the electrode surface,
(ii) changes in temperature and pressure,
(iii) dissociation of products by electron collisions giving rise to gaseous prod-
ucts,
(iv) liquid vaporization by corona-type discharges from points and irregular-
ities on the electrodes.
The electric field in a spherical gas bubble18
which is immersed in a liquid
of permittivity εliq. is given by:
Eb D
3E0
εliq. C 2
6.24
where E0 is the field in the liquid in the absence of the bubble. When the field
Eb becomes equal to the gaseous ionization field, discharge takes place which

will lead to decomposition of the liquid and breakdown may follow. Kao18
has developed a more accurate expression for the ‘bubble’ breakdown field
strength which is of the form
E0 D
1
ε1 ε2

2-2ε1 C ε2
r
-
4

Vb
2rE0

1
1/2
6.25
where is the surface tension of the liquid, ε1 and ε2 are the permittivities
of the liquid and the bubble respectively, r is the initial radius of the bubble
(initially spherical, which is assumed to elongate under the influence of the
field), and Vb is the voltage drop in the bubble. This expression indicates that
the critical electric field strength required for breakdown of liquid depends
upon the initial size of the bubble which is affected by the external pressure
and temperature. A strong dependence of liquid breakdown strength upon the
applied hydrostatic pressure has been observed experimentally.22
Commercial insulating liquids cannot readily be subjected to highly elabo-
rated purification treatment, and the breakdown strength will usually depend
upon the nature of impurities present.
6.2.4 Electroconvection and electrohydrodynamic model of
dielectric breakdown
The importance of electroconvection in insulating liquids subjected to high
voltages was not appreciated until recently. Most of the work comes from
Felici and his coworkers.19,20,21,22
In highly purified dielectric liquids
subjected to high voltage, electrical conduction results mainly from charge
carriers injected into the liquid from the electrode surface. The resulting space
charge gives rise to Coulomb’s force, which under certain conditions causes
hydrodynamic instability yielding convecting current. It follows that whenever
conduction in a fluid is accompanied by a significant space charge formation,
convection motion is very likely to occur. Lacroix et al.19
have studied the
conditions under which turbulent motion sets in. Using parallel plate electrodes
and controlled injection current, they showed that the onset of instability is
associated with a critical voltage. They observed that as the applied voltage
is increased near the critical voltage the motion at first exhibits a structure
of hexagonal cells. With a further increase in voltage the motion becomes
turbulent. Thus the interaction between electric field and space charge gives
rise to forces creating an eddy motion of the liquid. It has been shown that
at voltages close to breakdown the speed of this motion approaches a value
given by
p
ε/0/E where ε is the permittivity of the liquid, 0 the specific mass
and E the electric field strength. In liquids the ratio of this speed to ionic drift
velocity (KE), K being the mobility, M D
p
ε/0/K, is always larger than
unity and the ratio sometimes is very much larger than unity (see Table 6.3).
M is considered to play a dominant role in the theory of electroconvection.

Thus, the charge transport will be largely by liquid motion and not by ionic
drift. The key condition for the instability onset is that local flow velocity
uD
p
ε/0/E exceeds the ionic drift velocity (u KE).
Table 6.3 M D
p
ε/0/K
Medium Ion Relative M number
permittivity
Methanol HC
33.5 4.1
Ethanol Cl
25 26.5
Nitrobenzene Cl
35.5 22
Propylene carbonate Cl
69 51
Transformer oil HC
2.3 ¾200
Air N.T.P. O2

1.0 2.3 ð 102
The experimental values for M for various fluid media and common ions
obtained by Lacroix et al.19
are included in Table 6.3. The table also contains
the value for air at NTP. It is seen that in this case M − 1 and the rate of
electroconvection is negligible. Experiments show that electroconvection is
prevalent in all experimental settings in dielectric liquids subjected to elec-
tric fields irrespective of the gap geometries, provided the applied voltage is
high enough. This is true even in thoroughly deionized liquids because of the
adequate supply of ions by the high field processes at the electrodes.
Cross et al.,23
have studied electric stress-induced motion in transformer
oil under d.c. and 60 Hz stresses. Using high-speed schlieren photography,
they found that the turbulent motion was due to injection of positive charges
from one electrode. This was confirmed for both d.c. and 60 Hz stresses.
They also observed that the delay time in the onset of instability is related
to the condition for the injection or creation of charges at the electrode
surface. The time delay was found to decrease rapidly with increasing the
field strength ranging from a few seconds at 106
V/m to a few milliseconds
at 6 ð 106
V/m. Also as the temperature of the liquid increased, the time
delay for the given field decreased. Under 60 Hz voltage the time delay
was found to reach a minimum value approximately 4 msec, which is to
be expected. A 60 Hz wave requires 4.17 msec to reach the peak. From
these observations and calculations Cross et al. concluded that under these
conditions instability occurs when the injection strength, which is the ratio of
the space charge field to the applied field, reaches a large enough value for
a critical voltage to develop across the space charge layer within one half-
cycle period. The lowest value of the critical voltage occurs where space
charge limited conditions prevail at the injecting electrode.

6.3 Static electrification in power transformers
Static electrification (SE) in transformers is an interfacial phenomenon, which
involves oil, paper and transformer board. Its physical mechanism involves
a source of charge and region of excessive charge accumulation. Extensive
investigations about this phenomena have been made during recent years.25
When oil is forced through the tank and coolers, it acquires an electrostatic
charge, i.e. it contains an equal number of positive and negative ions. When
the oil passes the paper and solid insulation in the windings, the insulation
becomes negatively charged and the oil positively charged with the charge
separation occurring at the oil-insulated interface (Fig. 6.16). The earliest
reports on this phenomenon were in the 1970s from Japan, where number
Conservator
Radiator
Bushing
Core
Partial discharge
(meter)
Acoustic
sensor(s)
Windings
Pump
Clamps
Figure 6.16 Schematic of flow electrification density in transformers

of h.v. large transformer failures occurred. And later quite a few SE-related
incidents were also reported in the USA and other countries. It is believed
that transformers of large rating (e.g. 100 MVA) are most likely affected
by SE because they possess greater amounts of insulation and require larger
oil flow volumes than transformers of smaller ratings. As different oils have
different electrostatic charging tendencies (ECT), oil additives might be a way
to reduce oil ECT. As an alternative to the additive, used oil can be regener-
ated because new oil exhibits a lower ECT than aged oil. On the other hand,
operation practices are also of great importance. SE incidents can be caused
by poor operating practices such as increasing forced oil cooling capacity
beyond manufacturer’s recommendations, or having more forced oil cooling
in operation than the load on the transformer justifies.
References
1. D.M. Taylor and T.J. Lewis. J. Phys. D4 (1971), p. 1346.
2. A. von Hippel. Ergebn. Exakt. Naturw. 14 (1935), p. 79.
3. H. Fröhlich. Proc. Roy. Soc. 160 (1937), p. 230; A188 (1947), pp. 521, 532.
4. R. Stratton. Progress in Dielectrics, Vol. 3, p. 235. Haywood, London, 1961.
5. H. Fröhlich and B.V. Paranjape. Proc. Phys. Soc. London B69 (1956), p. 21.
6. J.J. O’Dwyer. The Theory of Electrical Conduction and Breakdown in Solid Dielectrics.
Clarendon Press, Oxford, 1973.
7. F. Seitz. Phys. Rev. 73 (1979), p. 833.
8. K.H. Stark and C.G. Garton. Nature, London 176 (1955), p. 1225.
9. R. Cooper. Int. J. of Elec. Eng. Education 1 (1963), p. 241.
10. H.R. Gnerlich. Field Testing of HV Power Cables Under VLF Voltages. IEEE Electrical
Insulation Magazine, Vol. 11, No. 5, 1995, p. 13.
11. T.J. Lewis. Progress in Dielectrics, Vol. 1. Haywood, London; Wiley, New York, 1959.
12. A.H. Sharbaugh and P.K. Watson, Progress in Dielectrics, Vol. 4, 1962.
13. D.W. Swann. Brit. J. Appl. Phys. 135 (1962), p. 208.
14. J.A. Kok. Electrical Breakdown in Insulating Liquids. Philips Tech. Library, 1961.
15. Z. Krasucki. Breakdown of commercial liquid and liquid solid dielectrics, in High Voltage
Technology (Alston), p. 129. Oxford University Press, 1968.
16. A.A. Zaky and R. Hawley. Conduction and Breakdown in Mineral Oils. Pergamon Press,
Oxford, 1973.
17. T.J. Gallagher. Simple Dielectric Liquids. Clarendon Press, Oxford, 1975.
18. K.C. Kao. Trans. AIEEE Elec. Ins. Vol. E1-11 (1976), pp. 121–128.
19. J.C. Lacroix, P.A. Hen and E.J. Hopfinger. J. Fluid Mech. 69 (1975), p. 539.
20. N.J. Felici. Direct Current 2 (1971), p. 147.
21. N.J. Felici and J.C. Lacroix. J. Electrost. 5 (1978), p. 135.
22. J.K. Nelson. Dielectric Fluids in Motion. IEEE Electrical Insulation Magazine, Vol. 10,
No. 11 994, pp. 16–28.
23. J.D. Cross, M. Nakans and S. Savanis. J. Electrost. 7 (1979), p. 361.
24. CIGRE Report, JWG 12/15.13, Task Force 02, August 1997.
25. Peyraque et al. Static Electrification and Partial Discharges induced by Oil Flow in Power
Transformers. IEEE Transactions Dielectric and Elec. Insul. Vol. 2, No. 1, 1995, pp. 40–45.
26. G.S. Eager, Jr, et al. Effect of DC Testing Water Tree Deteriorated Cables and a Primary
Evaluation of VLF as alternative. IEEE Transaction on Power Delivery, Vol. 7, N, July 1992.

Chapter 7
Non-destructive insulation test
techniques
This chapter is dedicated to test techniques, which provide information about
the quality of insulation systems which form part of an equipment or apparatus.
The tests as described here take advantage of well-known or desirable electrical
properties of either a specific dielectric material or an insulation system as
formed by a combination of different (fluid and/or solid) materials. Although
also mechanical or chemical tests are often applied to assess the insulation
quality, such tests are not taken into account.
Tests related to electrical properties are usually based on measurements
of insulation resistance or (d.c.) resistivity as well as capacitance and loss
factors, which are dependent on the frequency of the a.c. voltages applied.
As the techniques for the measurement of d.c. resistances are well known to
electrical engineers, they are neglected. Another group of non-destructive tests
on insulation systems is based on the detection and quantification of ‘partial
discharges’ or PDs, a measurement technique already applied over the past
five decades, but still a topic of research and increasing application. This latter
topic is strongly tied with Chapter 5, gaseous discharges and gas breakdown.
The tests related to electrical properties are often assumed to be quite simple
and ‘standard’. This is true if only tests with d.c. and power frequency are
considered. During recent years, however, it was recognized that the dynamic
electrical properties are very essential to quantify or at least to indicate the
ageing phenomena of insulation and thus to use the results as an essential diag-
nostic tool for equipment already in service for a long time. The individual,
partly very specific methods used to quantify the changes of the dielectric prop-
erties can only be mentioned and not be described in detail in this chapter. It is
essential to introduce a short description of the basic of ‘dielectric relaxation’
processes. We therefore, start, this chapter with an introduction to ‘dynamic
electrical properties’.
7.1 Dynamic properties of dielectrics
In contrast to Chapters 5 and 6, which primarily dealt with the maximum
resistance to destructive breakdown of gaseous, liquid and solid dielectrics
in high electrical fields, we will now examine the situation occurring when

such materials are exposed to much lower field stresses, thereby avoiding
any destructive or non-linear effects. Gases are generally not referred to as
‘materials’, because the distance between the adjacent molecules is so large
and the number of atoms or molecules per unit volume is so low, that they are
not able to withstand mechanical forces. However, gases are ‘dielectrics’ in
the sense of (electrical) insulators, and are used to prevent the flow of current.
Inherent in any dielectric material within an electric field are the well-known
effects of ‘dielectric polarization’, which are well documented in the literature
(see, e.g., A.K. Jonscher, 1983,1Ł
W. von Münch, 19872
). In understanding
these effects, it is useful to review some of the fundamental aspects of dielec-
tric polarization.1
At an atomic level, all matter consists of negative and positive charges
balancing each other in microscopic as well as in more macroscopic scales, in
the absence of any unipolar charge having been deposited within the matter
before. Macroscopically, some local space charge may be present, but even
in that condition an overall charge neutrality exists. While such local space
charges may have been produced by, e.g., thermal excitation or through the
absorption of light, an equal number of positively charged ions and detached
electrons will be present, and these processes of ionization and recombination
are usually in equilibrium.
As soon as the matter is stressed by even a very weak ‘macroscopic’ or
external electric field as, e.g., generated by a voltage across some electrodes
between which the dielectric is deposited, very different kinds of dipoles
become excited even within atomic scales. Local charge imbalance is thus
‘induced’ within the neutral species (atoms or molecules) as the ‘centres of
gravity’ for the equal amounts of positive and negative charges, šq, become
separated by a small distance d, thus creating a dipole with a ‘dipole moment’
p D qd, which can also be related to the ‘local’ electric field E acting in close
vicinity of the species. The relation between the dipole moment, p, and the
electric field, E, is given by p D ˛E, where ˛ is the ‘polarizability’ of the
material under consideration. Note that p, d and E are vectors, which is not
marked here. As the distance d will be different for different materials, so
is their polarizability. Due to chemical interactions between dissimilar atoms
forming molecules, many molecules will have a constant distance d between
the charge centres thus forming ‘permanent dipole moments’, which, however,
are generally distributed irregularly within the matter as long as no external
field is applied. (Note that any kind of ‘permanent polarization’ such as that
occurring in electrets or ferroelectrics is not considered here.) The macroscopic
effect of the ‘polarizability’ of individual materials is ultimately manifested
in a general relation between the macroscopic polarization P and the number

Non-destructive insulation test techniques 397
of the polarized species N per unit volume of the matter. These relationships
are quite well known, but not treated here.
The following highlights of the polarization processes should be sufficient
to gain an understanding of the main effects producing polarization. Electronic
polarization is effective in every atom or molecule as the centre of gravity
of the electrons surrounding the positive atomic cores will be displaced by
the action of the electric field. This process is extremely fast and thus effec-
tive up to optical frequencies. Ionic polarization refers to matter containing
molecules which will form ions, which are not separated by low electric fields
or working temperatures. Dipolar polarization belongs to matter containing
molecules with permanent dipole moments, the local distribution of which is
governed statistically due to action of thermal energies. Under the influence
of E, the dipoles will only partly be oriented so that a linear dependency of P
with E can still be assumed. Ionic and dipolar polarization are also quite fast
effects and can follow a.c. frequencies up to GHz or MHz. Interfacial polar-
ization is effective in insulating materials composed from different dielectric
materials such as oil-impregnated paper. The mismatch of the products of
permittivity and conductivity for the different dielectrics forces moveable
charges to become attached on interfaces. This phenomenon is quite often
very slow and in general active up to power frequencies. Finally, polarization
by hopping charge carriers,1
a mechanism more recently postulated,3,4
may occur.3,4
This type of polarization process is based on the well-known
hopping processes of electronic charges in amorphous and disordered non-
metallic solids, in which direct current conduction generally takes place by
the hopping of ions. A prerequisite of d.c. conduction is the presence of a
continuous connected network of hopping sites, so that the charges are able
to traverse the physical dimension of the dielectric. If, however, the matter
is very strongly disordered, the normal concept of band conduction by free
charge carriers must be replaced by very localized sites, which are surrounded
by very high potential wells which cannot be surpassed by electrons.
In summary, dielectric polarization is the result of a relative shift of positive
and negative charges in the matter under consideration. This shift is produced
by an electric field, provoking either ‘induced polarization’ of individual atoms
and/or ions, an orientation of any permanent dipoles, the build-up of charges
at interfaces between quite different dielectrics, or the creation of dipoles at
localized hopping sites. During all of these processes, the electric field is
therefore not able to force the charges to escape from the matter, which would
lead to electric conduction.
For any matter, which is isotropic and homogeneous at least in macroscopic
scales, we may therefore write the following general relation between the
(macroscopic) polarization P and the field E as:
P D ε0 E C higher order terms in E. 7.1

Here, ε0 is the permittivity of free space (D8.85419 ð 1012
(As/Vm)) and
is the susceptibility of the matter, which is a dimensionless number and with
a value of zero for vacuum and/or free space. From equation (7.1) we see that
accounts for all kinds of polarization processes acting within the dielectric.
Note, that the vectors E and P have the same direction in isotropic materials.
The additional higher order terms in equation (7.1) can be neglected under the
assumption that the dielectric response of the material remains linear, i.e. as
long as the magnitude of the exciting electric field is not too large.
Dynamic properties of dielectrics can be defined and also measured in the
frequency or time domain. We will start to define the properties in the time
domain and proceed to the frequency domain definitions.
7.1.1 Dynamic properties in the time domain
In any vacuum-insulated electrode arrangement, the ‘dielectric displacement’
or ‘dielectric flux density’ (or ‘electrical induction’) D is proportional to the
applied electric field E. The relation between the two quantities is
D D ε0E
or, if the electric field is generated by a time-varying voltage,
Dt D ε0Et 7.2
where ε0 D 8.85419 ð 1012
As/Vm is the permittivity of free space or
vacuum, a number with dimensions converting the unit for electric field (V/m)
to that of area charges (As/m2
). As E is a vector, D is also a vector usually
assumed to exist within the space in which the electrostatic field is present. One
should note, however, that the electric displacement D represents the (positive
and negative) electric charges per unit area as deposited at the surface of the
electrodes which are the origin – sources and sinks – of all electric field lines.
The origin of D and E is usually provided by a voltage source connected to
the electrodes of the electrode arrangement under consideration. If the voltage
is time-dependent as already assumed in eqn (7.2), both D and E are of
identical time dependency with no time delay between their magnitudes. The
so-called ‘displacement current’ released from the voltage source as necessary
to maintain the area charge density at the electrodes is then only governed
by dQ/dt, if Q is the sum or integral of all charges deposited on each of the
electrodes.
If the vacuum is replaced by any kind of isotropic dielectric material, the
displacement is obviously increased by the (macroscopic) polarization P of
this dielectric, which was already defined in eqn (7.1), resulting in:
Dt D ε0Et C Pt. 7.3

From equation (7.1) and the explanations previously given, we know that P
will be a vector in the direction of E, as isotropic materials are assumed.
(Further on, we can therefore avoid to indicate this by bold letters.) The time
dependency of P, however, will not be the same as that of E, as the different
polarization processes will have different delays with respect to the appearance
of E. This delay is obviously caused by the time-dependent behaviour of the
susceptibility . . .
The time delay between Et and Pt may best be understood with the
following considerations. Let us assume, that the (macroscopic) electric field
E within the matter is excited by an ideal voltage step at time t D 0 and
that its time evolution is marked by E0. The dielectric material is then char-
acterized by its susceptibility t as a response in the time domain. This
parameter covers the formation and evolution of the different kinds of polariza-
tion processes including those that develop within extremely short times (e.g.
electronic polarization) as well as those which are much slower or even very
slow (e.g. interfacial polarization). For t 0, the magnitude of susceptibility
is still zero.
Figure 7.1 illustrates the situation. For this special case of excitation, P
becomes
P0t/E0 D ε0 t1t 7.4
where t and P0t represent ‘step response (SR) functions’. The factor 1t
is used to indicate the unit step.
E0
εE0γ (t→∞)
P0 (t) or χ (t)
χ (0′)
0 t
0′
Figure 7.1 Polarization of a dielectric material subjected to an electrical
step field of magnitude E0

As known from general circuit theory, the time-dependent polarization Pt
can be calculated for any other time-dependent excitation Et of the system
since the special solution for the SR is already known. This can be done using
‘Duhamel’s integral’ or by convolution in the time domain. As not everybody
may be familiar with Duhamel’s integral, we first display it in general terms.
If at is a general force acting on a system or network and at D 0 for
t 0, and if bt is the effect of this force at any place of the system, and if
b1t is the known effect at that place for the unit step of at D 1t, then the
following result applies for t ½ 0:
bt D
d
dt
t
0
azb1t z dz D b10at C
t
0
az
db1t z
dt
dz. 7.5
Identifying now ‘a’ with ‘E’, ‘b’ with ‘P’ and ‘b1’ with the step response
function of eqn (7.4), for an arbitrary time-dependent electrical field Et we
now obtain the polarization Pt as
Pt D ε0 0Et C ε0
t
0
E
d t
dt
d. 7.6
In this equation, the derivative of the susceptibility t appears. This function
is given by
ft D d t/dt 7.7
and is the dielectric response (or relaxation) function of the dielectric material.
This function is, as shown by Fig. 7.1, a monotonically decaying function
for time scales accessible to usual measurements. Thus eqn (7.6) may be
rewritten as
Pt D ε0 0Et C ε0
t
0
Eft d. 7.6a
The first term of this equation is zero, since when a true ideal step excitation
is assumed, 0 D 0. For actual measurements, however, the fast or very fast
polarization phenomena cannot be defined and then this term will be finite.
This fact can be taken into account by finding the actual value of this term
for a delayed time instant t D 00
(Fig. 7.1) and then quantifying this value
by 00
. This represents a value effective for high or very high frequencies.
More detailed explanation of this is given in section 7.1.3.
The polarization Pt is not an observable magnitude by itself, but it
produces a main part of the displacement current in a test object. Up to
now, we have not yet considered any conductivity of the dielectric, which
is not involved in polarization. As already postulated by Maxwell in 1891,
an electrical field Et applied to a dielectric generates a current density jt,

which can be written as a sum of conduction and total displacement current:
jt D 0Et C
dDt
dt
. 7.8
Here 0 represents the ‘pure’ or effective d.c. conductivity of the material.
Using the relations (7.3), (7.6a) and (7.8), the current density can now be
expressed as:
jt D 0Et C ε01 C 00

dEt
dt
C ε0
d
dt
t
00
ft E d
D 0Et C ε01 C 00

dEt
dt
C ε0f00
Et C ε0
t
00
dft
dt
E d.
7.9
The factor 1 C 00
in these equations is the real part of the relative permit-
tivity εr for ‘high’ frequencies, which is already effective at a time instant
00
as defined before. These equations are thus the basis for the experimental
measurement of the dielectric response function ft.
Determination of the dielectric response function from polarization and
depolarization (relaxation) currents
Together with a known geometry, the current densities can easily be converted
to currents in test equipment or a test cell, as shown later. Therefore we may
still use eqn (7.9) for further considerations related to currents.
If a step voltage at time t D 0 is applied and maintained for a long time
(minutes, hours), a polarization (or charging, absorption) current can be moni-
tored. The amplitude of this current will change by orders of magnitude with
time. In accordance with eqn (7.9), in which the time-variable field must be
replaced by a step-like ‘charging’ field of magnitude Ec, this current is due to
a current density of
jpol.t D 0Ec C ε01 C 00
Ecυt C ε0Ecft. 7.10
Here, υt is the delta function which will produce an extremely large current
pulse coincident with the sudden increase of the voltage. The magnitude of this
current pulse will, in general, not be measured. This second term is thus related
to the displacement currents due to the sum of vacuum capacitance of the test
object and the capacitance related to the ‘high-frequency’ susceptibility of the
dielectric used. The first term is due to the d.c. conductivity of the material
and determines the current after a (more or less) long time, for which the last
term, which quantifies the response function completely, becomes negligible.
Polarization current measurements can finally be stopped if the current
becomes stable. Immediately afterwards, the depolarization (or discharging,
desorption) current can be measured by a subsequent short-circuiting of the

ipol (t)
idepol (t)
t
0
Uc
Tc tc
Figure 7.2 Principle of relaxation current measurement
sample, see Fig. 7.2. According to the superposition principle and neglecting
immediately the second term in eqn (7.10), we get
jdept D ε0Ec[ft ft C Tc] 7.11
where Tc is the time during which the step voltage was applied to the test
object. The second term in this equation can be neglected, if measurements
with large values of Tc have been made for which the final value of the polar-
ization current was already reached. Then the depolarization current becomes
directly proportional to the dielectric response function.
An example of a recently performed relaxation current measurement is
shown in Fig. 7.3. The data were obtained during investigations concerning
the dielectric response of oil-paper (transformer or pressboard) insulation
with different moisture content (m.c.). Preparation and test conditions of
the samples can be found within the original publication.5
In this case,
all measurements started 1 s after voltage application ipol and short-circuit
idepol respectively. From the selected results it can be seen that the final value
of the polarization current will be reached only for the higher moisture content
samples in spite of the fact that the measurements lasted up to 200 000 s, i.e.
about 56 hours. Representation of such results in log–log scale is paramount
due to the large change of the quantities.
The effect of depolarization currents is illustrated by the recharging of h.v.
capacitors. When such capacitors are stressed for a long time with d.c. voltage
and then briefly discharged through short-circuiting they will recharge to quite
a high voltage when the short-circuit is removed. This ‘return’ or ‘recovery’

1E+0 1E+1 1E+2 1E+3 1E+4 1E+5
Time (s)
1E−9
1E−10
1E−11
1E−12
1E−13
1E−14
Current
(A)
2 mm pressboard, θ = 20 °C, Uc = 200 V, Tc = 200000 s
unaged (m.c.: 4.0%)
unaged (m.c.: 2.5%)
unaged (m.c.: 1.0%)
unaged (m.c.: 0.2%)
ipol
idepol
Figure 7.3 Relaxation currents of unaged samples with different moisture
contents
voltage effect is due to depolarization currents. In the case of h.v. capacitors,
it is dangerous and may cause severe accidents.
Specific response functions
The relaxation currents from Figs 7.2 and 7.3 decay monotonically, but do
not follow simple relationships which can be expressed by adequate analytic
functions. Nevertheless, many researchers have attempted to represent experi-
mental data through such functions which can sometimes be related to at least
idealized physical models of the polarization processes. We will here only
mention some analytic functions to which reference is often made. For further
studies see reference 1.
The simplest response is related to Debye and is represented by
ft D B expt/
where B is an amplitude for t D 0 and is a single relaxation time. Only some
pure and simple liquids will follow this response at least within certain time
regions. This response is also typical for ‘interfacial polarization’, i.e. a series
combination of two different dielectrics between which a barrier is present
adjacent to a bulk conducting material. The barrier attracts the charges and
appears as a capacitance, whereas the bulk behaves like a series resistance.

Another extreme is provided by the ‘general response’ expression1
which
also involves the somewhat simpler ‘power law’ expression. Such a response
is expressed by
ft D
B

t

n
C

t

m
and will appear in the log–log scale as the superposition of two straight
decaying lines. The simpler power law expression neglects one term in the
denominator to form only a single straight line dependency. The polariza-
tion processes in polyethylene for instance can in general be simulated by
such a behaviour. The processes can then be related to the physical process
of diffusion or injected charges. A special form of an electric network, the
uniform distributed R-C line, can simulate such a response. In section 7.1.3
some further hints to simulating networks are provided.
7.1.2 Dynamic properties in the frequency domain
The dielectric properties of dielectrics can also be measured and quantified
in the frequency domain, i.e. with a.c. voltages as a function of frequency.
The transition to the frequency domain from the time domain can be executed
by means of Laplace or Fourier transformation. This is shown by means of
eqn (7.9) in the following revised form, in which the response for the total
current within a test specimen for an ideal voltage step starting at t D 0 is
considered:
jt D 0Et C ε0
dEt
dt
C ε0
d
dt
t
0
ft E d. 7.12
With jt ) jp; Et ) Ep; E0
t ) pEp; ft ) Fp; and by
considering the convolution of the last term in this equation we get, for the
present, formally with p being the Laplace operator:
jp D 0Ep C ε0pEp C ε0pFpEp.
As p for the given conditions is the complex frequency, jω, we can reduce
the equation to
jω D Eω[0 C jωε01 C Fω]. 7.13
Thus it becomes obvious that the dielectric response function ft is the time
domain of the frequency dependent susceptibility ω, which is defined as
the Fourier transform of the dielectric response function ft:
ω D Fω D 0
ω j 00
ω D
1
0
ft expjωt dt. 7.14

From this equation in the frequency domain the following relationships for
the susceptibility in time and frequency domain apply:
t ! 1 , 0
ω ! 0 and t ! 0 , 0
ω ! 1.
Now, in the frequency domain the polarization can be written as:
Pω D ε0 ωEω. 7.15
Equation (7.8) expressed in the frequency domain becomes:
jω D 0Eω C jωDω. 7.16
Using eqns (7.2) and (7.14) the current density is rewritten as:
jω D f0 C iωε0[1 C 0
ω j 00
ω]gEω
D f0 C ε0ω 00
ω C iωε0[1 C 0
ω]gEω. 7.17
The displacement can now be expressed by the relative but complex dielectric
permittivity of the material εrω with the relation:
Dω D ε0εrωEω D ε0[1 C 0
ω j 00
ω]Eω 7.18
where:
εrω D ε0
rω jε00
r ω D 1 C 0
ω j 00
ω. 7.19
For a practical determination of the dielectric response from eqn (7.17), a
bridge or any other instrument cannot distinguish between the current contri-
bution of d.c. conductivity and that of dielectric loss. This means that the
effective measured relative dielectric permittivity Q
εrω is different from the
relative permittivity εrω defined in eqns (7.17) and (7.18). If the effective
relative dielectric permittivity Q
εrω is defined from the following relation
jω D jωε0 Q
εrωEω. 7.20
Therefore:
Q
εrω D ε0
iω i[ε00
r ω C 0/ε0ω]
D 1 C 0
ω j[ 00
ω C 0/ε0ω]. 7.21
Then the dissipation factor tan υ (see section 7.2) will be:
tan υω D
ε00
r ω C 0/ε0ω
ε0
rω
. 7.22
The real part of eqn (7.21) defines the capacitance of a test object, while
the imaginary part represents the losses. Both quantities are dependent on

frequency, which sometimes is not realized. We display, therefore, the results
of frequency-dependent measurements as made on the test samples which
were used in Fig. 7.3 for the measurements of relaxation currents. The results,
again reproduced from literature,5
are shown in Figs 7.4 and 7.5. Figure 7.4
shows the capacitance of the specimens and Fig. 7.5 their dissipation factors
over a frequency range of nearly 8 decades. The measurements were taken
1E−4 1E−3 1E−2 1E−1 1E+0 1E+1 1E+2 1E+3 1E+4
Frequency (Hz)
400
300
200
C
′
(pF)
2mm pressboard, θ = 20 °C
unaged (m. c.: 4.0%)
C′
Figure 7.4 Real part of the complex capacitance of pressboard samples in
dependence on frequency
1E−4 1E−3 1E−2 1E−1 1E+0 1E+1 1E+2 1E+3 1E+4
Frequency (Hz)
1E−3
1E−2
1E−1
1E+0
tan
δ
2 mm pressboard, θ = 20 °C
unaged (m. c.: 2.5 %)
unaged (m. c.: 1.0 %)
unaged (m. c.: 0.2 %)
tan δ
unaged (m. c.: 4.0 %)
Figure 7.5 Dissipation factor tan υ of pressboard samples in dependence on
frequency

using a ‘dielectric spectrometer’ at an a.c. voltage of about 3 volts. Note that
the influence of moisture is much more significant at the lower frequencies.
Its influence at power frequencies (50/60 Hz), where the measurements are
usually made, is not as pronounced.
Measurements in the frequency domain become very lengthy if many indi-
vidual values for very low frequencies are considered. At least three cycles of
an a.c. voltage are in general necessary to quantify the amplitudes and phase
shift between voltage and currents. Therefore, about 3000 seconds are neces-
sary to get a single value of C and tan υ for a frequency of 1 MHz. Since the
results of relaxation current measurements can be converted in the frequency
domain and vice versa, both methods complement each other.
7.1.3 Modelling of dielectric properties
Modelling or simulating dielectric properties through the use of equivalent
electrical circuits has been practised for decades.6
As an introduction, we
will derive one of the models which can immediately be detected from the
relaxation currents as treated in sections 7.1.1 and 7.1.2 respectively. For
convenience, first a formal transition from current densities and electric fields
to currents and voltages as applied to the terminals of a system shall be made.
Let us rewrite eqn (7.17) first:
jω D f0 C ε0ω 00
ω C jωε0[1 C 0
ω]gEω 7.17a
and assume an ideal plate capacitor, in which the area of the plates is A
and the gap distance is d. Without any dielectric within the gap, the ‘vacuum
capacitance’ of this object becomes Cvac D ε0A/d. As the dielectric shall have
the d.c. conductivity 0, the d.c. resistance becomes R0 D d/0A. As the
voltage applied to the terminals is V D d.E and the current is I D jA, we can
introduce these equations into eqn (7.17a) and obtain:
Iω D Uω

1
R0
C ωCvac
00
ω C jωCvac[1 C 0
ω]

. 7.23
This equation represents an equivalent circuit comprising a single resistor R0 in
parallel with a single capacitor Cvac. But in addition, two other elements are in
parallel: another resistor of magnitude 1/[ωCvac
00
ω] producing additional
losses and a second capacitor, whose vacuum capacitance is multiplied by
0
ω. Both additional circuit elements are strongly dependent on frequency,
as 00
ω as well as 0
ω depend on ω, and the lossy term 1/[ωCvac
00
ω]
decreases additionally by 1/ω. This f dependency can be simulated either by
a multiple series connection of parallel RC circuits or a formal conversion of
the two elements into one series RC circuit, in which again both elements are
frequency dependent. This single RC circuit can then again be split up in a

multiple arrangement of parallel RC elements, which represent as a whole the
frequency dependence of the terms of the dielectric. We display the second
type of equivalent circuit in Fig. 7.6.
Cvac R0
Cn
C2
C1
Rn
Ci
Ri
R2
R1
Figure 7.6 Equivalent circuit to model a linear dielectric
That this type of circuit will completely simulate both types of relax-
ation currents can well be recognized from the behaviour of this circuit in
time domain, an example of which was shown in Fig. 7.3: the depolarization
currents and thus also the response function ft is a monotonic decaying func-
tion as well as the polarization currents. As every RiCi element produces – for
charging as well as for discharging the circuit – an exponential (increasing or
decreasing) current, the depolarization current is treated as a superposition
of such exponentials. For this model, all circuit parameters can therefore be
determined by measured quantities! The d.c. resistance R0 can be approximated
from the applied step voltage Uc and the difference between polarization and
depolarization currents for the highest available time values. The individual
elements Ri, Ci with the corresponding time constants i D RiCi can then be
determined by fitting the depolarization current with the equation
idepol.t D
n

iD1
Ai expt/i, 7.24
where
Ai D Uc[1 expTc/i]/Ri for i D 1 . . . n 7.25
and Tc is the duration of the time, during which the sample was charged. If the
measured currents are not recorded within a very short time of the application
of the step voltage, or the short-circuit after polarization, then Cvac must be
replaced by a capacitance effective at the time instant at which the currents
are available. For more information see the relevant literature.5,7
The basic idea in applying this simple model is to facilitate all further
calculations with respect to the frequency domain or even the calculation of
a ‘polarization spectrum’ belonging to a very special procedure of recovery
voltage measurements, which will be briefly explained in section 7.1.4.

To complete this section, some additional hints are provided. For the equiv-
alent circuit of Fig. 7.6, the complex capacitance Cω can be calculated
according to eqn (7.23) from its complex admittance, Yω as:
Cω D
Yω
iω
D C? Hz C
1
jωR0
C
n

jD1
Ci
1 C jωRiCi
7.26
Instead of Cvac some larger capacitance C? Hz is taken into account as explained
before. The real and imaginary parts of Cω are then given as:
C0
ω D C? Hz C
n

iD1
Ci
1 C ωRiCi2
7.27
and
C00
ω D
1
ωR0
C
n

iD1
ωRiC2
i
1 C ωRiCi2
. 7.28
Finally, tan υω can be written as:
tan υω D
1
ωR0
C
n

iD1
ωRiC2
i
1 C ωRiCi2
C? Hz C
n

iD1
Ci
1 C ωRiCi2
. 7.29
7.1.4 Applications to insulation ageing
As already mentioned in the introduction to this chapter, the application of
measurements related to the dynamic properties of dielectrics has recently
undergone an increase in use. The background and motivation of such appli-
cations is due to the need for utilities to employ insulation monitoring and
diagnostic systems with the goal of reducing costs, assessing the performance
of existing apparatus, and maintaining and enhancing safety and reliability.
It is now generally believed that ‘condition-based monitoring’ or ‘condition
maintenance’ will replace the ‘periodic maintenance’ practices applied to date.
One of the primary drives in this direction is the large amount of aged and
expensive equipment forming the backbone of most modern power systems.
The increasingly popular move towards a competitive electricity supply market
coupled with the ageing infrastructure calls for new techniques to retain aged
plant in service as long as possible.
It is not possible to discuss this new philosophy in more detail in this
text. Many publications including those available through CIGRE or IEEE
outline the continued development of condition monitoring over the past ten

years. Here, only some hints on the application of the more recently developed
methods can be provided with special relevance to the dynamic properties of
dielectrics and insulation systems. One should note, however, that the applica-
tion of ‘dielectric measurements’ can be traced back to the last century and that
an overview about the history and the state of the art of such measurements
up to about 1990 may be found in the literature.11
Well-established, conventional methods
Ageing effects in electric insulation are always caused by changes of the
chemical structure of the dielectric matter. These changes produce mechan-
ical degradation (which usually cannot be detected within sealed equipment),
release chemical ‘ageing’ products, and alter electrical properties leading to
in-service electrical breakdown. Examples of conventional methods used for
in-service monitoring of power transformers with standard oil-paper insulation
are: Dissolved Gas Analysis (DGA);8
Oil Parameter Analysis;9
measurement
of C tan υ at power frequency (see section 7.2); measurement of insulation
resistance; and measurement of the ‘polarization index’.
New methods
New methods for transformer in-service monitoring include: Furan Analysis
and HPLC (High Performance Liquid Chromatography) to quantify chemical
ageing products and the Dielectric Response Analysis (DRA), the fundamen-
tals of which have been treated above. The methods related to DRA are briefly
explained below.
Dielectric response analysis (DRA)
This method is based on the measurements of polarization as well as de-
polarization currents and is thus sometimes described as the ‘PDC method’.
The background and the measurement procedure are described at the end of
section 7.1.1. It should, however, again be noted that the dielectric response
function ft, if quantified for times larger than several milliseconds, also
provides access to the frequency-dependent dielectric parameters starting from
ultra-low frequencies up to power frequencies.
Recovery voltage
The measurement of recovery or return voltages is another method to quantify
the dielectric response of materials. The principle of the measurement can be
traced back to the last century. With reference to Fig. 7.7, it can be explained
as follows: a constant voltage Uc charges the test object for 0 t t1; after a
relatively short period between t1 t t2 during which the sample is short-
circuited, the test object is left in open-circuit condition. Then for times t ½ t2

a recovery voltage urt caused by residual polarization is built up across the
test object, resulting in repolarization. If the voltmeter recording urt has
an extremely high input impedance, the test object remains charged until it
discharges through its internal resistance. If the dielectric response function of
the test object is known, the time dependence of recovery voltage for t ½ t2
can be derived from eqn (7.9).1,5,11
The advantage of this method is that it
includes a self-calibration with respect to the capacitance of the test object,
but for numerical evaluations the response function has to be measured as
shown previously.
ur(t)
0 t1 t2 tr t
Uc
Tc Td
Vr max
Sr0
Figure 7.7 Principle of recovery voltage measurement
The ‘polarization spectrum’
The so-called ‘polarization spectrum’11
is a quantity derived from a special
measurement procedure of recovery voltages. With reference to Fig. 7.7, it is
determined from the peak values (at time instants tr) of many recovery volt-
ages, which vary depending on charging duration Tc. These recovery voltage
maxima are then presented as a function of increasing charging duration.
Normally the charging time Tc is chosen to be twice as long as the discharge
duration, Td. It can be shown that the peak of the ‘polarization spectrum’ will
be reached for a very dominant time constant of a Debye process already
mentioned in section 7.1.1. This measurement method, for which a commer-
cial equipment exists, has often been applied in practice during recent years
with the goal of identifying the moisture content of the pressboard within
power transformers. The interpretation of the results, however, is in general
very difficult. Misleading results can be obtained if the moisture content is
quantified in accordance with instructions provided by the manufacturer.
7.2 Dielectric loss and capacitance measurements
In section 7.1 it was explained that all kinds of dielectrics or insulation mate-
rials and systems can be characterized by its inherent polarization phenomena,

which in the frequency domain can be expressed by a capacitance C and a
magnitude of power dissipation (dielectric loss) as quantified by the dissipa-
tion or loss factor tan υ. Whereas these quantities within a wide frequency
range are of utmost interest for new materials or even for the quality control
of well-known insulation materials on receipt after delivery, nearly every high-
voltage equipment prior to delivery to the customer will undergo a test related
to ‘C tan υ’ for a final quality control. Such tests are in general only made
with a frequency for which the equipment is designed. The reasons for this are
as follows. First, too high losses at least during a.c. test voltages may cause
thermal breakdown, see Chapter 6, section 6.1. Secondly, the manufacturer of
a specific equipment knows the typical magnitudes which can be tolerated and
which provide information about the quality of the newly manufactured equip-
ment. Finally, such tests are in general made in dependence of the test voltage
applied; both magnitudes, capacitance and dissipation factor shall be essen-
tially constant with increasing voltage, as insulation systems are linear systems
and any ‘tip-up’ of the tan υ with voltage level, called ‘ionization knee’, is a
preliminary indication of ‘partial discharges’ (discussed in section 7.3) within
the system.
The measurement of these dielectric properties with power frequency
belongs, therefore, to standard testing procedures, for which ancillary
principles are used. New measurement equipment is in general based on these
principles, but nowadays is supplemented with microprocessor control and
evaluation supports or software.
The various laboratory techniques for electrical insulation measurement
have earlier been reviewed by Baker13
and for detailed descriptions the reader
is advised to refer to that publication. Additional information about the earlier
techniques is also provided by Schwab.14
7.2.1 The Schering bridge
Still one of the most commonly used methods for measuring ‘loss tangent’,
tan υ, and capacitance with high precision is the high-voltage Schering bridge,
originally patented by P. Thomas in 1915 and introduced to h.v. measurements
by H. Schering in 1920.15,p.212
The basic circuit arrangement is shown in
Fig. 7.8.
The bridge measures the capacitance CX and loss angle υ (i.e. tan υ) of a
capacitor or any capacitance of a specimen by comparing it with a gas-filled
standard capacitor CN which has very low and nearly negligible loss over a
wide frequency range and can be built for test voltages up to the megavolt
range (see Chapter 3, section 3.5.4). The ‘X’ h.v. arm of the bridge consists
of a sample, the dielectric loss and capacitance of which are to be measured.
On account of the dielectric loss the current through the capacitor leads the
voltage by an angle (90 υ) which is only slightly less than 90°. This current

H.V. source
(a)
(d)
(b)
(c)
Rx
R3
R4
C4
G
Cx
CN
I
III
II
IV
Figure 7.8 The high-voltage Schering bridge
produces a voltage drop of usually less than 100 V across the variable resistor
R3 of the low-voltage part of the bridge. The elements R4 and C4 of this
part are necessary to balance the bridge. As seen from the circuit, the balance
conditions are represented by considering that for the frequency applied a
pure capacitance Cx is connected in series with a hypothetical resistance RX,
the power dissipated in the resistance simulating the power loss in the actual
capacitor.
For the Schering bridge, the balance conditions are always derived for this
series RX CX equivalent circuit. The derivation is shown below. For any
series equivalent circuit the dissipation factor tan υ is defined by the following
equation:
tan υs D ωRsCs. 7.30
The balance conditions obtained when the indicator (null detector) ‘G’ shows
zero deflection in Fig. 7.8 are:
Zab
Zbc
D
Zad
Zdc
,
where
Zad D Rx j
1
ωCx
, Zab D i
1
ωCN
,
Zbc D
R4[j1/ωC4]
R4 j1/ωC4
, Zdc D R3

By separation of the real and imaginary terms we get:
Cx D CN
R4
R3
, 7.31
RX D R3
C4
CN
7.32
Substituting CN in eqn (7.31) from eqn (7.32) and multiplying by ω we obtain
the dissipation factor according eqn (7.30):
ωCxRx D ωC4R4 D tan υX. 7.33
If the results shall be expressed in terms of a parallel equivalent circuit for
the test object, for which a parallel arrangement of a capacitance CP and a
resistor RP is assumed, the following equations shall be taken into account.
For a parallel equivalent circuit, the dissipation is
tan υp D
1
ωRPCP
. 7.34
On the condition that the losses in the two circuits must be equal, the quantities
of series and parallel equivalent circuits may be converted to each other by:
Cp D
Cs
1 C tan υs2
; 7.35
Rp D Rs

1 C
1
1 C tan υs2

7.36
In practice R3 is a variable resistance and is usually in the form of a four-
decade box. Its maximum value is limited to about 10 000 0 in order to keep
the effects of any stray capacitance relatively small. R4 is made constant and
in general realized as a multiple of (1000/1) 0 thus making possible a direct
reading of tan υ, provided a constant value of the frequency is indicated. C4
is variable. To exclude from Zbc and Zdc and the galvanometer branches any
currents due to inter-capacity between the h.v. and l.v. arms, except those
flowing through Zab and Zad, the bridge is fully double screened as indicated
in Fig. 7.8, in which only one screen is sketched. The l.v. branches are usually
protected with spark gaps against the appearance of high voltages in the event
of failure of ZI or ZII.
In Fig. 7.8 the network is earthed at the l.v. end of the transformer supplying
the high voltage, and by this also the bridge is earthed at (c). Under balance
conditions, both sides of the null detector G are at the same potential, but
the shield is earthed. Therefore partial stray capacitances appear across the
branches III and IV, and depending upon the length of the leads to CN and
Cx, these partial capacitances can assume values over a wide limit. These

capacitances can be measured and thus their influence on the dissipation factor
can be calculated. If CI is the partial capacitance of branch I, and CII that of
branch II, the calculation shows
tan υ D RωC4 C CII R3ωCI. 7.37
This procedure is time consuming and inconvenient, and there are methods
available to overcome this effect. The principle of the method may easiest be
described by the ‘Wagner earth’ as shown in Fig. 7.9 applied to the Schering
bridge, but introduced by K.W. Wagner for another bridge already in 1911.17
In this arrangement an additional arm Z is connected between the l.v. terminal
of the four-arm bridge and earth. Together with the stray capacitance of the
h.v. busbar to earth the arrangement becomes equivalent to a six-arm bridge
and a double balancing procedure is required which can be achieved either by
using two detectors or a switch arrangement which enables the detector to be
switched on into either sets of arms. At balance the terminals of the detector are
at earth potential and capacitances between the terminals and screens having
no potential difference between them do not affect the balance conditions. Both
the detector and the l.v. leads must be screened. The capacitances between the
leads and screens are in parallel with the impedance Z and as such do not
contribute to the balance conditions.
CN
Rx
Cx
R3
R4
C4
G G
Figure 7.9 Bridge incorporating ‘Wagner earth’
This method, however, is rarely used today, as operational amplifiers for
automatic balancing of ‘Wagner earth’ may be used. The basic circuit is shown
in Fig. 7.10. Although the bridge may well be earthed at (c), the potentials
of the screens are shifted to the potential of the detector branch by a high-
quality amplifier with unity voltage gain. The shields of the leads to Cx and
CN are not grounded, but connected to the output of the amplifier, for which
operational amplifiers can conveniently be used. The high input impedance
and very low output impedance of the amplifier do not load the detector
branch and keep the screen potential at any instant at an artificial ‘ground’.

H.V. source
(b)
+1
(a)
(d)
(c)
R3
R4
RX
C4
CX
G
CN
Figure 7.10 Automatic ‘Wagner earth’ (dividing screen technique)
A second screen, which is earthed, may be added to prevent disturbances by
neighbouring voltage sources.
Measurement of large capacitance
When the capacitance to be measured is large, a variable resistance R3 in
eqn (7.31) capable of passing large currents would be required. To maintain
a high value of R3 it may be shunted by another resistor A as shown with
the simplified circuit in Fig. 7.11. An additional resistor B is put in series
with R3 to protect it from excessive currents should it accidentally be set to a
very low value. With this arrangement it can be shown1
that the specimen’s
capacitance and loss tangent become respectively:
Cx D CN

R4
R3

1 C

B
A

C

R3
A

7.38
and
tan υ D ωCNR4

B
R3

. 7.39
For more sophisticated circuits see reference 14.
The Schering bridge principle is suitable for measurements at frequencies
up to some 100 kHz, if the circuit elements are properly designed. Common
Schering bridges for power frequencies may be used at frequencies up to about

(a)
(d)
(c)
Rx
Cx
R3
B
A
Figure 7.11 Shunt arrangement for measurement of large capacitance
(compare with Fig. 7.8)
10 kHz only. At higher frequencies it becomes necessary to use a special
high-frequency Schering bridge or substitution and resonance methods.13
Measuring uncertainty is in general as low as 105
, if the bridge ratio R3/R4
is 1. With other ratios the uncertainty is typically 104
. The uncertainty of the
standard capacitor used as reference will contribute to these values. The range
of measurements is in the interval of 1 pF–100 µF.
7.2.2 Current comparator bridges
The shortcomings of the Schering bridge, among which also the possible
changes of the circuit elements with temperature and ageing can be mentioned,
stimulated quite early the search for improved forms of bridge circuits, which
have been based on ‘inductive coupling’ or ‘ampere-ratio arms’. But only in
the late 1950s has the technology of such new circuits been successful in
demonstrating the advantages of such new circuits, in which the current in
the test piece and that in the standard capacitor is compared by means of a
magnetic toroidal core on which two uniformly distributed coils with different
numbers of turns are wound such as to force zero flux conditions. Although the
first ideas to apply this principle may be traced back to Blumlein in 1928, the
fundamental circuit arrangement for the measurement of capacitance and losses
in h.v. capacitors was introduced by Glynne.18
A prerequisite of applying the
‘ampere-ratio arm’ principle was the availability of new magnetic material with
a very high initial permeability with which such ‘current comparator bridges’
could be made.19
The basic circuit of the ‘Glynne bridge’ is shown in Fig. 7.12. The main
part of the bridge circuit consists of a three-winding current comparator which

is carefully and heavily shielded against magnetic stray fields and protected
against mechanical vibrations. Thus, the particular merit in this arrangement
is that there is no net m.m.f. across windings 1 and 2 at balance conditions.
Furthermore, the stray capacitance across the windings and that of the screened
l.v. leads does not enter in the balance expression since there is no voltage
drop on the comparator windings apart from their d.c. resistance. This enables
long leads to be used without ‘Wagner earth’. The sensitivity of the bridge is
higher than that of the Schering bridge.
H.V. source
Null
detector
In
I2
Cn
Cx
Rx
Ix
C
R
1 2
Figure 7.12 Current comparator (Glynne) bridge
The balance is indicated by zero voltage induced in the detector coil and
corresponds to the conditions when IXN1 D INN2 where N1 and N2 are the
number of turns in series with the sample CX and the standard capacitor
CN respectively, and IX and I2 are the corresponding currents flowing in
CX and NN. Again a series equivalent circuit is assumed for the specimen
under test.
For a current IN in the standard capacitor the voltage developed across the
R-C arm is given by:
V D
INR
1 C jωCR
The portion of current I2 in coil 2 is
I2 D
IN
1 C jωCR
,
and for a unity total applied voltage
I2 D
1
[fR/1 C jωCR C 1/jωCNg1 C jωCR]
D
jωCN
1 C jωCN C CR
,

therefore:
Cx D CN

N2
N1

7.40
and
tan υ D ωRCN C C. 7.41
The capacitance and phase angle balance are obtained by making N1, N2 and
R variable. The uncertainty and sensitivity of this type of bridge is better than
that indicated for the Schering bridge; the working frequency range is about
50 Hz to 1 kHz.
Nowadays, bridges with fully automatic self-balancing are preferred espe-
cially if only unskilled personnel are used or series measurements have to
be performed within a production process. There are many solutions avail-
able. Older methods used servo-motor-driven potentiometers controlled from
a feedback loop. The capability of electronic circuits provides many different
solutions,20
the most recent of which takes advantage of microcomputer
control. Figure 7.13 displays such a typical circuit as published by Osvàth
and Widmer in 1986.21
Raw balancing of the bridge is realized by relays, the
fine balancing and loss factor compensation are made by electronic circuits.
The whole system is controlled by a microprocessor, which simulates the
manual procedure of balancing. Newest developments of this circuit provide
continuous automatic balancing within less than 1 second as well as manual
balancing.50
The sensitivity and uncertainty of this type of bridge is equal to
the best h.v. C tan υ bridges.
+
NX NN
Nb
Ni
90°
D/Ac
D/Atgδ
IX
IN
Ik
α
β
∼
Ie
V
uV
CX
CN
Figure 7.13 Automatic high-voltage current comparator bridge

7.2.3 Loss measurement on complete equipment
It is often required to measure the dielectric loss on specimens one side of
which is permanently earthed. There are two established methods used for such
measurement. One is the inversion of a Schering bridge, shown in Fig. 7.14, with
the operator, ratio arms and null detector inside a Faraday cage at high potential.
The system requires the cage to be insulated for the full test voltage and with
suitable design may be used up to the maximum voltage available. There are,
however, difficulties in inverting physically the h.v. standard capacitor and it
becomes necessary to mount it on a platform insulated for full voltage.
Cx
CN
C4
G
Rx
R3
R4
Figure 7.14 High-voltage bridge with Faraday cage
Cx CN
G
Figure 7.15 Fully screened bridge
An alternative method, though limited to lower voltages, employs an artifi-
cial earth which differs in potential from a true earth by the voltage developed
across each of the l.v. arms as shown in Fig. 7.15. The artificial earth screen

intercepts all the field from high potential to earth except in the specimen. It
thus requires screening of the h.v. lead and presents difficulties at voltages in
excess of about 5 kV.
7.2.4 Null detectors
The null detector G for ancient bridges was simply a vibration galvanometer
of high mechanical Q factor. Although their application is well justified, the
sensitivity to mechanical noise (if present) and the limited electrical sensitivity
present some disadvantages. Since a few decades more sensitive electronic null
detectors are commonly used. The possible high sensitivity, however, cannot
be utilized in general, as noise voltages from the circuit, or electromagnetically
induced voltages from the stray fields of the h.v. circuit, disturb the balance.
This electronic null detector reads the voltage across the detector branch.
As the balance equations of the bridge are only valid for a particular fixed
frequency, the unavoidable harmonic content of the high input voltage of the
bridge results in higher harmonic voltages across the null detector, for which
the bridge is not balanced. A very pronounced pass-band characteristic is
therefore necessary to attenuate these harmonics.
A very much improved balance is possible using electronic null detectors,
which are also sensitive to the phase. Bridges may only slowly converge,
i.e. the magnitude of the detector branch voltage may only slightly change
within the individual settings of R3 and C4 in the Schering bridge or R in
the transformer ratio-arm bridge. In the use of phase-sensitive null detectors,
the balance condition is indicated in terms of magnitude and phase. With a
reference voltage in phase with the (high) source voltage, these values describe
Lissajou figures at the screen of a CRO used for the display. In this way the
balancing procedure is always known and the balance is obtained much faster.
7.3 Partial-discharge measurements
What is a ‘partial discharge’? Let us use the definition given in the Interna-
tional Standard of the IEC (International Electrotechnical Commission) related
to partial discharge measurements, see reference 31:
Partial discharge (PD) is a localized electrical discharge that only partially
bridges the insulation between conductors and which may or may not occur
adjacent to a conductor.
This definition is supplemented by three notes, from which only notes 1 and
2 shall be cited:
NOTE 1 – Partial discharges are in general a consequence of local electrical
stress concentrations in the insulation or on the surface of the insulation.

Generally such discharges appear as pulses of duration of much less than
1 µs. More continuous forms may, however, occur, as for example the so-
called pulse-less discharges in gaseous dielectrics. This kind of discharge
will normally not be detected by the measurement methods described in
this standard.
NOTE 2 – ‘Corona’ is a form of partial discharge that occurs in gaseous
media around conductors which are remote from solid or liquid insulation.
‘Corona’ should not be used as a general term for all forms of PD.
No further explanations are necessary to define this kind of phenomena: PDs
are thus localized electrical discharges within any insulation system as applied
in electrical apparatus, components or systems. In general PDs are restricted
to a part of the dielectric materials used, and thus only partially bridging the
electrodes between which the voltage is applied. The insulation may consist
of solid, liquid or gaseous materials, or any combination of these. The term
‘partial discharge’ includes a wide group of discharge phenomena: (i) internal
discharges occurring in voids or cavities within solid or liquid dielectrics;
(ii) surface discharges appearing at the boundary of different insulation mate-
rials; (iii) corona discharges occurring in gaseous dielectrics in the presence of
inhomogeneous fields; (iv) continuous impact of discharges in solid dielectrics
forming discharge channels (treeing).
The significance of partial discharges on the life of insulation has long
been recognized. Every discharge event causes a deterioration of the material
by the energy impact of high energy electrons or accelerated ions, causing
chemical transformations of many types. As will be shown later, the number
of discharge events during a chosen time interval is strongly dependent on
the kind of voltage applied and will be largest for a.c. voltages. It is also
obvious that the actual deterioration is dependent upon the material used.
Corona discharges in air will have no influence on the life expectancy of an
overhead line; but PDs within a thermoplastic dielectric, e.g. PE, may cause
breakdown within a few days. It is still the aim of many investigations to
relate partial discharge to the lifetime of specified materials. Such a quantita-
tively defined relationship is, however, difficult to ensure. PD measurements
have nevertheless gained great importance during the last four decades and a
large number publications are concerned either with the measuring techniques
involved or with the deterioration effects of the insulation.
The detection and measurement of discharges is based on the exchange
of energy taking place during the discharge. These exchanges are manifested
as: (i) electrical pulse currents (with some exceptions, i.e. some types of glow
discharges); (ii) dielectric losses; (iii) e.m. radiation (light); (iv) sound (noise);
(v) increased gas pressure; (vi) chemical reactions. Therefore, discharge detec-
tion and measuring techniques may be based on the observation of any of
the above phenomena. The oldest and simplest method relies on listening to

the acoustic noise from the discharge, the ‘hissing test’. The sensitivity is,
however, often low and difficulties arise in distinguishing between discharges
and extraneous noise sources, particularly when tests are carried out on factory
premises. It is also well known that the energy released by PD will increase the
dissipation factor; a measurement of tan υ in dependency of voltage applied
displays an ‘ionization knee’, a bending of the otherwise straight depen-
dency (see section 7.2). This knee, however, is blurred and not pronounced,
even with an appreciable amount of PD, as the additional losses generated
in very localized sections can be very small in comparison to the volume
losses resulting from polarization processes. The use of optical techniques is
limited to discharges within transparent media and thus not applicable in most
cases. Only modern acoustical detection methods utilizing ultrasonic trans-
ducers can successfully be used to localize the discharges.2225
These very
specialized methods are not treated here. Summaries of older methods can be
found in the book of Kreuger.26
More recent developments may be found in
reference 45.
The most frequently used and successful detection methods are the elec-
trical ones, to which the new IEC Standard is also related. These methods
aim to separate the impulse currents linked with partial discharges from any
other phenomena. The adequate application of different PD detectors which
became now quite well defined and standardized within reference 31, presup-
poses a fundamental knowledge about the electrical phenomena within the
test samples and the test circuits. Thus an attempt is made to introduce the
reader to the basics of these techniques without full treatment, which would
be too extensive. Not treated here, however, are non-electrical methods for
PD detection.
7.3.1 The basic PD test circuit
Electrical PD detection methods are based on the appearance of a ‘PD (current
or voltage) pulse’ at the terminals of a test object, which may be either a
simple dielectric test specimen for fundamental investigations or even a large
h.v. apparatus which has to undergo a PD test. For the evaluation of the funda-
mental quantities related to a PD pulse we simulate the test object, as usual, by
the simple capacitor arrangement as shown in Fig. 7.16(a), comprising solid or
fluid dielectric materials between the two electrodes or terminals A and B, and a
gas-filled cavity. (A similar arrangement was used in Chapter 6, see Fig. 6.12.)
The electric field distribution within this test object is here simulated by some
partial capacitances, which is possible as long as no space charges disturb this
distribution. Electric field lines within the cavity are represented by Cc and
those starting or ending at the cavity walls form the two capacitances C0
b and
C00
b within the solid or fluid dielectric. All field lines outside the cavity are
represented by Ca D C0
a C C00
a. Due to realistic geometric dimensions involved,

and as Cb D C0
bC00
b/C0
b C C00
b, the magnitude of the capacitances will then
be controlled by the inequality
Ca × Cc × Cb. 7.42
A
A
S
B
Vs
Vb
Cb
Cc
Rc
Vc
ic (t)
Ca
B
(a) (b)
Cb
′
Ca
′
Ca
′′
Cb
′′
ic (t)
Cc
Figure 7.16 Simulation of a PD test object. (a) Scheme of an insulation
system comprising a cavity. (b) Equivalent circuit
This void will become the origin of a PD if the applied voltage is increased,
as the field gradients in the void are strongly enhanced by the difference in
permittivities as well as by the shape of the cavity. For an increasing value
of an a.c. voltage the first discharge will appear at the crest or rising part of a
half-cycle. This discharge is a gas discharge (see Chapter 5) creating electrons
as well as negative and positive ions, which are driven to the surfaces of the
void thus forming dipoles or additional polarization of the test object. This
physical effect reduces the voltage across the void significantly. Within our
model, this effect is causing the cavity capacitance Cc to discharge to a large
extent. If the voltage is still increasing or decreasing by the negative slope of
an a.c. voltage, new field lines are built up and hence the discharge phenomena
are repeated during each cycle (see Fig. 6.13 in Chapter 6). If increasing d.c.
voltages are applied, one or only a few partial discharges will occur during the
rising part of the voltage. But if the voltage remains constant, the discharges
will stop as long as the surface charges as deposited on the walls of the void
do not recombine or diffuse into the surrounding dielectric.
These phenomena can now be simulated by the equivalent circuit of this
scheme as shown in Fig. 7.16(b). Here, the switch S is controlled by the
voltage Vc across the void capacitance Cc, and S is closed only for a short time,
during which the flow of a current ict takes place. The resistor Rc simulates
the time period during which the discharge develops and is completed. This

discharge current ict, which cannot be measured, would have a shape as
governed by the gas discharge process and would in general be similar to a
Dirac function, i.e. this discharge current is generally a very short pulse in the
nanosecond range.
Let us now assume that the sample was charged to the voltage Va but the
terminals A, B are no longer connected to a voltage source. If the switch S
is closed and Cc becomes completely discharged, the current ict releases a
charge υqc D CcυVc from Cc, a charge which is lost in the whole system as
assumed for simulation. By comparing the charges within the system before
and after this discharge, we receive the voltage drop across the terminal υVa as
υVa D
Cb
Ca C Cb
υVc 7.43
This voltage drop contains no information about the charge υqc, but it is
proportional to CbυVc, a magnitude vaguely related to this charge, as Cb
will increase with the geometric dimensions of the cavity.
υVa is clearly a quantity which could be measured. It is a negative voltage
step with a rise time depending upon the duration of ict. The magnitude of
the voltage step, however, is quite small, although υVc is in a range of some
102
to 103
V; but the ratio Cb/Ca will always be very small and unknown
according to eqn (7.42). Thus a direct detection of this voltage step by a
measurement of the whole input voltage would be a tedious task. The detection
circuits are therefore based upon another quantity, which can immediately be
derived from a nearly complete circuit shown in Fig. 7.17. The test object,
Fig. 7.16(a), is now connected to a voltage source V, in general an a.c. power
supply. An impedance Z, comprising either only the natural impedance of
the lead between voltage source and the parallel arrangement of CK and Ct
or enlarged by a PD-free inductance or filter, may disconnect the ‘coupling
capacitor’ CK and the test specimen Ct from the voltage source during the
short duration PD phenomena only. Then CK is a storage capacitor or quite
a stable voltage source during the short period of the partial discharge. It
V
Z
i (t), q
Vs Ct
Cs
Cb
Cc
CK
Figure 7.17 The PD test object Ct within a PD test circuit

releases a charging current or the actual ‘PD current pulse’ it between CK
and Ct and tries to cancel the voltage drop υVa across Ct ³ Ca C Cb. If
CK × Ct, υVa is completely compensated and the charge transfer provided
by the current pulse it is given by
q D

it D Ca C CbυVa 7.44
With eqn (7.43), this charge becomes
q D CbυVc 7.45
and is the so-called apparent charge of a PD pulse, which is the most funda-
mental quantity of all PD measurements. The word ‘apparent’ was introduced
because this charge again is not equal to the amount of charge locally involved
at the site of the discharge or cavity Cc. This PD quantity is much more real-
istic than υVa in eqn (7.43), as the capacitance Ca of the test object, which is
its main part of Ct, has no influence on it. And even the amount of charge as
locally involved during a discharge process is of minor interest, as only the
number and magnitude of their dipole moments and their interaction with the
electrodes or terminals determine the magnitude of the PD current pulse.
The condition CK × Ca¾
DCt is, however, not always applicable in prac-
tice, as either Ct is quite large, or the loading of an a.c. power supply becomes
high and the cost of building such a large capacitor, which must be free of any
PD, is not economical. For a finite value of CK the charge q or the current
it is reduced, as the voltage across CK will also drop during the charge
transfer. Designating this voltage drop by υVŁ
a, we may compute this value by
assuming that the same charge CbυVc has to be transferred in the circuits of
Figs 7.16(b) and 7.17. Therefore
υVaCa C Cb D υVŁ
Ca C Cb C CK. 7.46
Introducing eqn (7.43) as well as eqn (7.45), we obtain
υVŁ
D
Cb
Ca C Cb C CK
υVc D
q
Ca C Cb C CK
. 7.47
Again, υVŁ
is a difficult quantity to be measured. The charge transferred from
CK to Ct by the reduced current it is, however, equal to CKυVŁ
; it is related
to the real value of the apparent charge q which then can be measured by an
integration procedure, see section 7.3.3. If we designate this measured quantity
as qm, then
qm D CKυVŁ
D
CK
Ca C Cb C CK
q ³
CK
Ca C CK
q

or
qm
q
¾
D
CK
Ca C CK
³
CK
Ct C CK
. 7.48
The relationship qm/q indicates the difficulties arising in PD measurements
for test objects of large capacitance values Ct. Although CK and Ct may be
known, the ability to detect small values of q will decrease as all instruments
capable of integrating the currents it will have a lower limit for quantifying
qm. Equation (7.48) therefore sets limits for the recording of ‘picocoulombs’
in large test objects. During actual measurements, however, a calibration
procedure is needed during which artificial apparent charge q of well-known
magnitude is injected to the test object, see section 7.3.7.
A final, critical note is made with reference to the definition of the apparent
charge q as given in the new IEC Standard 60270.31
The original text of this
definition is:
apparent charge q of a PD pulse is that unipolar charge which, if injected
within a very short time between the terminals of the test object in a specified
test circuit, would give the same reading on the measuring instrument as
the PD current pulse itself. The apparent charge is usually expressed in
picocoulombs.
This definition ends with:
NOTE – The apparent charge is not equal to the amount of charge locally
involved at the site of the discharge and which cannot be measured directly.
This definition is an indication of the difficulties in understanding the physical
phenomena related to a PD event. As one of the authors of this book has been
chairman of the International Working Group responsible for setting up this
new standard, he is familiar with these difficulties and can confirm that the
definition is clearly a compromise which could be accepted by the interna-
tional members of the relevant Technical Committee of IEC. The definition is
correct. It relates to a calibration procedure of a PD test and measuring circuit,
as already mentioned above. The ‘NOTE’, however, is still supporting the
basically wrong assumption that a certain amount or number of charges at the
site of the discharge should be measured. As already mentioned: it is not the
number of charges producing the PD currents, but the number of induced
dipole moments which produce a sudden increase in the capacitance of the
test object. With section 7.1, this phenomenon is much more plausible.
7.3.2 PD currents
Before discussing the fundamentals of the measurement of the apparent charge
some remarks concerning the PD currents it will be helpful, as much of the

research work has been and is still devoted to these currents, which are difficult
to measure with high accuracy. The difficulties arise for several reasons.
If V is an a.c. voltage, the main contribution of the currents flowing within
the branches CK and Ct of Fig. 7.17 are displacement currents CdV/dt, and
both are nearly in phase. The PD pulse currents it with crest values in the
range of sometimes smaller than 104
A, are not only small in amplitude,
but also of very short duration. If no stray capacitance in parallel to CK were
present, it would be the same in both branches, but of opposite polarity. For
accurate measurements, a shunt resistor with matched coaxial cable may be
introduced in the circuit as shown in Fig. 7.18. The voltage across the CRO
(or transient recorder) input is then given by Vmt D it C iZ0R/R C Z0.
Only if the capacitance of the test object is small, which is a special case, will
the voltages referring to the PD currents it be clearly distinguished from the
displacement currents itt.
V
Z
R CRO
(ik + it)
it
Vm
ik
Ck
Z0
Z0
Ct
i (t)
Figure 7.18 Measurement of PD current it – low sensitivity circuit
Improvements are possible by inserting an amplifier (e.g. active voltage
probe) of very high bandwidth at the input end of the signal cable. In this
way the signal cable is electrically disconnected from R. High values of R,
however, will introduce measuring errors, which are explained with Fig. 7.19.
A capacitance C of some 10 pF, which accounts for the lead between Ct
and earth as well as for the input capacitance of the amplifier or other stray
capacitances, will shunt the resistance R and thus bypass or delay the very
high-frequency components of the current it. Thus, if it is a very short
current pulse, its shape and crest value are heavily distorted, as C will act
as an integrator. Furthermore, with R within the discharge circuit, the current
pulse will be lengthened, as the charge transfer even with C D 0 will be
delayed by a time constant RCtCK/Ct C CK. Both effects are influencing
the shape of the original current pulse, and thus the measurement of it is a
tedious task and is only made for research purposes.

V
Z
C
To CRO
Ck
Vm (t)
Ct
i (t)
R
Figure 7.19 Measurement of PD currents – high sensitivity circuit
All measured data on current shapes published in many papers are suffering
from this effect. One may, however, summarize the results by the following
statements. Partial discharge currents originated in voids within solids or
liquids are very short current pulses of less than a few nanoseconds duration.
This can be understood, as the gas discharge process within a very limited
space is developed in a very short time and is terminated by the limited
space for movement of the charge carriers. Discharges within a homogeneous
dielectric material, i.e. a gas, produce PD currents with a very short rise time
5 nsec and a longer tail. Whereas the fast current rise is produced by the fast
avalanche processes (see Chapter 5), the decay of the current can be attributed
to the drift velocity of attached electrons and positive ions within the dielectric.
Discharge pulses in atmospheric air provide in general current pulses of less
than about 100 nsec duration. Longer current pulses have only been measured
for partial discharges in fluids or solid materials without pronounced voids, if
a number of consecutive discharges take place within a short time. In most of
these cases the total duration of it is less than about 1 µsec, with only some
exceptions e.g. the usual bursts of discharges in insulating fluids.
All these statements refer to test circuits with very low inductance and
proper damping effects within the loop CK Ct. The current it, however,
may oscillate, as oscillations are readily excited by the sudden voltage drop
across Ct. Test objects with inherent inductivity or internal resonant circuits,
e.g. transformer or reactor/generator windings, will always cause oscillatory
PD current pulses. Such distortions of the PD currents, however, do not change
the transferred charge magnitudes, as no discharge resistor is in parallel to CK
or Ct. If the displacement currents itt or iKt are suppressed, the distorted
PD currents can also be filtered, integrated and displayed.
7.3.3 PD measuring systems within the PD test circuit
In sections 7.3.1 and 7.3.2 the evolution of the PD current pulses and measure-
ment procedures of these pulses have been broadly discussed. To quantify the
‘individual apparent charge magnitudes’ qi for the repeatedly occurring PD

pulses which may have quite specific statistical distributions, a measuring
system must be integrated into the test circuit which fulfils specific require-
ments. Already at this point it shall be mentioned that under practical envi-
ronment conditions quite different kinds of disturbances (background noise)
are present, which will be summarized in a later section.
Most PD measuring systems applied are integrated into the test circuit in
accordance with schemes shown in Figs 7.20(a) and (b), which are taken from
the new IEC Standard31
which replaces the former one as issued in 1981.32
Within these ‘straight detection circuits’, the coupling device ‘CD’ with its
input impedance Zmi forms the input end of the measuring system. As indicated
in Fig. 7.20(a), this device may also be placed at the high-voltage terminal
side, which may be necessary if the test object has one terminal earthed.
Optical links are then used to connect the CD with an instrument instead of
a connecting cable ‘CC’. Some essential requirements and explanations with
reference to these figures as indicated by the standard are cited here:
the coupling capacitor Ck shall be of low inductance design and should
exhibit a sufficiently low level of partial discharges at the specified test
voltage to allow the measurement of the specified partial discharge magni-
tude. A higher level of partial discharges can be tolerated if the measuring
system is capable of separating the discharges from the test object and the
coupling capacitor and measuring them separately;
the high-voltage supply shall have sufficiently low level of background
noise to allow the specified partial discharge magnitude to be measured at
the specified test voltage;
high-voltage connections shall have sufficiently low level of background
noise to allow the specified partial discharge magnitude to be measured at
the specified test voltage;
an impedance or a filter may be introduced at high voltage to reduce back-
ground noise from the power supply.
The main difference between these two types of PD detection circuits
is related to the way the measuring system is inserted into the circuit. In
Fig. 7.20(a), the CD is at ground potential and in series to the coupling capac-
itor Ck as it is usually done in praxis. In Fig. 7.20(b), CD is in series with the
test object Ca. Here the stray capacitances of all elements of the high-voltage
side to ground potential will increase the value of Ck providing a somewhat
higher sensitivity for this circuit according to eqn (7.48). The disadvantage is
the possibility of damage to the PD measuring system, if the test object fails.
The new IEC Standard defines and quantifies the measuring system char-
acteristics. The most essential ones will again be cited and further explained
below:

Z
U~
Ca
Ck
CD
CD
Alternative position for CD
CC
OL
Ck
MI
Zmi
(a) Coupling device CD in series with the coupling capacitor
Z
U~
Ck
Ca
CD
CC
MI
Zmi
(b) Coupling device CD in series with the test object
U~ high-voltage supply
Zmi input impedance of measuring system
CC connecting cable
OL optical link
Ca test object
Ck coupling capacitor
CD coupling device
MI measuring instrument
Z filter
Figure 7.20 Basic partial discharge test circuits – ‘straight detection’
The transfer impedance Zf is the ratio of the output voltage amplitude
to a constant input current amplitude, as a function of frequency f, when
the input is sinusoidal.
This definition is due to the fact that any kind of output signal of a measuring
instrument (MI) as used for monitoring PD signals is controlled by a voltage,
whereas the input at the CD is a current.

The lower and upper limit frequencies f1 and f2 are the frequencies at
which the transfer impedance Zf has fallen by 6 dB from the peak pass-
band value.
Midband frequency fm and bandwidth f: for all kinds of measuring
systems, the midband frequency is defined by:
fm D
f1 C f2
2
7.49
and the bandwidth by:
f D f2 f1; 7.50
The superposition error is caused by the overlapping of transient output
pulse responses when the time interval between input current pulses is less
than the duration of a single output response pulse. Superposition errors
may be additive or subtractive depending on the pulse repetition rate n of
the input pulses. In practical circuits both types will occur due to the random
nature of the pulse repetition rate.
This rate ‘n’ is defined as the ratio between the total number of PD pulses
recorded in a selected time interval and the duration of the time interval.
The pulse resolution time Tr is the shortest time interval between two
consecutive input pulses of very short duration, of same shape, polarity and
charge magnitude for which the peak value of the resulting response will
change by not more than 10 per cent of that for a single pulse. The pulse
resolution time is in general inversely proportional to the bandwidth f of
the measuring system. It is an indication of the measuring system’s ability
to resolve successive PD events.
The integration error is the error in apparent charge measurement which
occurs when the upper frequency limit of the PD current pulse amplitude-
spectrum is lower than (i) the upper cut-off frequency of a wideband
measuring system or (ii) the mid-band frequency of a narrow-band
measuring system.
The last definition of an ‘integration error’ will need some additional
explanation. PD measuring systems quantifying apparent charge magnitudes
are band-pass systems, which predominantly are able to suppress the high
power frequency displacement currents including higher harmonics. The lower
frequency limit of the band-pass f1 and the kind of ‘roll-off’ of the band-
pass control this ability. Adequate integration can thus only be made if the
‘pass-band’ or the flat part of the filter is still within the constant part of the
amplitude frequency spectrum of the PD pulse to be measured. Figure 7.21,

A band-pass of the measuring system
B amplitude frequency spectrum of the PD pulse
C amplitude frequency spectrum of calibration pulse
f1 lower limit frequency
f2 upper limit frequency
A A
f2
f1
B C
B, C
−6 dB
Figure 7.21 Correct relationship between amplitude and frequency to
minimize integration errors for a wide-band system31
again taken from the new standard, provides at least formal information about
correct relationships. More fundamental information may be found within
some specific literature.28,29
Now we can proceed to explain the basic types of PD instruments to see
how the requirements can be fulfilled.
7.3.4 Measuring systems for apparent charge
The following types of measuring systems all comprise the already mentioned
subsystems: coupling device (CD), transmission system or connecting cable
(CC), and a measuring instrument (MI), see Fig. 7.20. In general the trans-
mission system, necessary to transmit the output signal of the CD to the input
of the MI, does not contribute to the measuring system characteristics as both
ends are matched to the characteristics of both elements. The CC will thus
not be considered further.
The input impedance Zmi of the CD or measuring system respectively will
have some influence on the waveshape of the PD current pulse it as already

mentioned in the explanation of Fig. 7.19. A too high input impedance will
delay the charge transfer between Ca and Ck to such an extent that the upper
limit frequency of the amplitude frequency spectrum would drop to unaccept-
able low values. Adequate values of Zmi are in the range of 100 0.
In common with the first two measuring systems for apparent charge is a
newly defined ‘pulse train response’ of the instruments to quantify the ‘largest
repeatedly occurring PD magnitude’, which is taken as a measure of the ‘speci-
fied partial discharge magnitude’ as permitted in test objects during acceptance
tests under specified test conditions. Sequences of partial discharges follow in
general unknown statistical distributions and it would be useless to quantify
only one or very few discharges of large magnitude within a large array of
much smaller events as a specified PD magnitude. For further information
on quantitative requirements about this pulse train response, which was not
specified up to now and thus may not be found within in earlier instruments,
reference is made to the standard.31
Wide-band PD instruments
Up to 1999, no specifications or recommendations concerning permitted
response parameters have been available. Now, the following parameters are
recommended. In combination with the CD, wide-band PD measuring systems,
which are characterized by a transfer impedance Zf having fixed values of
the lower and upper limit frequencies f1 and f2, and adequate attenuation
below f1 and above f2, shall be designed to have the following values for
f1, f2 and f:
30 kHz f1 100 kHz;
f2 500 kHz; 7.51
100 kHz f 400 kHz.
The response of these instruments to a (non-oscillating) PD current pulse is in
general a well-damped oscillation as shown below. Both the apparent charge
q and – with some reservation – the polarity of the PD current pulse can be
determined from this response. The pulse resolution time Tr is small and is
typically 5 to 20 µs.
Figure 7.22 shows the typical principle of such a system. The coupling
devices CD (Fig. 7.20) are passive high-pass systems but behave more often
as a parallel R-L-C resonance circuit (Fig. 7.22(a)) whose quality factor is
relatively low. Such a coupling impedance provides two important qualities.
At first, a simple calculation of the ratio output voltage V0 to input current
Ii in dependency of frequency (Dtransfer impedance Zf) would readily
demonstrate an adequate suppression of low- and high-frequency currents in
the neighbourhood of its resonance frequency. For a quality factor of Q D 1,

this attenuation is already 20 dB/decade and could be greatly increased close
to resonance frequency by increasing the values of Q. Secondly, this parallel
circuit also performs an integration of the PD currents it, as this circuit is
already a simple band-pass filter and can be used as an integrating device.
Let us assume that the PD current pulse it would not be influenced by the
test circuit and would be an extremely short duration pulse as simulated by a
Dirac function, comprising the apparent charge q. Then the calculation of the
output voltage V0t according to Fig. 7.22(a) results in:
V0t D
q
C
e˛t

cos ˇt
˛
ˇ
sin ˇt

7.52
Ii
ik or it
i (t)
(a)
R L C
V0
V0(t)
V0(t) VCRO(t)
MI
(b)
ii (t)
T/2
t
τ τ
t
∼q
t
Figure 7.22 Principle of ‘wide-band’ PD measuring system. (a) Simplified
equivalent circuit for the CD and amplifier. (b) Typical time-dependent
quantities within (a) (T D period of power frequency; ³ pulse resolution
time Tr)
where
˛ D
1
2RC
; ˇ D
1
LC
˛2 D ω0 1 ˛2LC.
This equation displays a damped oscillatory output voltage, whose amplitudes
are proportional to q. The integration of it is thus performed instantaneously
t D 0 by the capacitance C, but the oscillations, if not damped, would heavily
increase the ‘pulse resolution time Tr’ of the measuring circuit and cause

‘superposition errors’ for too short time intervals between consecutive PD
events (see definitions above). With a quality factor of Q D 1, i.e. R D
p
L/C,
a very efficient damping can be achieved, as then ˛ D ω0/2 D 1f0. For a
resonance frequency f0 of typically 100 kHz, and an approximate resolution
time of Tr
¾
D t D 3/˛, this time becomes about 10 µsec. For higher Q values,
Tr will be longer, but also the filter efficiency will increase and therefore a
compromise is necessary. The resonance frequency f0 is also influenced by the
main test circuit elements Ck and Ca, as their series connection contributes
to C. The ‘RLC input units’ must therefore be changed according to spec-
imen capacitance to achieve a bandwidth or resonance frequency f0 within
certain limits. These limits are postulated by the bandwidth f of the addi-
tional band-pass amplifier connected to this resonant circuit to increase the
sensitivity and thus to provide again an integration. These amplifiers are typi-
cally designed for lower and upper limit frequencies of some 10 kHz and some
100 kHz respectively, and sometimes the lower limit frequency range may also
be switched from some 10 kHz up to about 150 kHz to further suppress power
frequencies. In general the fixed limit frequencies are thus within a frequency
band in general not used by radio stations, and higher than the harmonics of
the power supply voltages. The band-pass amplifier has in general variable
amplification to feed the ‘CRO’ (reading device!) following the amplifier with
adequate magnitudes during calibration and measurement. For a clearer under-
standing the time-dependent quantities (input a.c. current with superimposed
PD signals, voltages before and after amplification) see Fig. 7.22(b).
Finally, the amplified discharge pulses are in general displayed by an
(analogue or digital) oscilloscope superimposed on a power frequency elliptic
timebase, as shown in Fig. 7.23. The magnitude of the individual PD pulses
is then quantified by comparing the pulse crest values with those produced
during a calibration procedure, see section 7.3.7. With this type of reading by
individual persons it is not possible to quantify the standardized ‘pulse train
response’ which quantifies the ‘largest repeatedly occurring PD magnitude’.
Correct readings are, however, possible by applying additional analogue peak
detection circuits or digital peak detection software prepared to follow the
specified pulse train response.
The pattern on the CRO display can often be used to recognize the origin
of the PD sources. (Instead of a simple CRO display digital acquisition
of PD quantities and up-to-date methods for evaluation are used now, see
section 7.3.8.) A typical pattern of Trichel pulses can be seen in Fig. 7.23(a).
Figure 7.23(c) is typical for the case for which the pulse resolution time of the
measuring system including the test circuit is too large to distinguish between
individual PD pulses.
It was clearly shown that even the response of such ‘wide-band PD instru-
ments’ provided no more information about the original shape of the input
PD current pulse as indicated in Fig. 7.22(b) and confirmed by the pattern

(a)
(b)
(c)
Figure 7.23 Elliptical display. (a) Point plane (‘Trichel pulses!’). (b) Void
discharges at inception. (c) Void discharges at twice inception voltage
of the Trichel pulses in Fig. 7.23(a). Figure 7.24 further confirms this state-
ment. Here, two kinds of recorded responses – Figs 7.24(a) and (b) – of two
consecutive calibration pulses (‘double pulse’) are shown within a time scale
of microseconds. A comparison of both recorded responses shows their differ-
ences with respect to a (positive) short and lengthened input pulse, which has
some significant influence on the peak value of the undershoot after the first
excursion of the response which indicates the polarity of the input signal.
Polarity detection by digital PD acquisition systems may thus be difficult.
Narrow-band PD instruments
It is well known that radio transmission or radiotelephony may be heavily
disturbed by high-frequency interference voltages within the supply mains to
which receivers are connected or by disturbing electromagnetic fields picked
up by the aerials.27
It was also early recognized that corona discharges at
h.v. transmission lines are the source of such disturbances. The measurement

10 µs
Time t
Uout
Uout
(a) Short-duration input pulse
10 µs
Time t
(b) Lengthened input pulse
Figure 7.24 Output voltage signals Uout of a wide-band PD detector with
f D 45 . . . 440 kHz for two different input pulses
of ‘radio noise’ in the vicinity of such transmission lines is thus an old and
well-known technique which several decades ago triggered the application of
this measurement technique to detect insulation failures, i.e. partial discharges,
within h.v. apparatus of any kind.
The methods for the measurement of radio noise or radio disturbance have
been subjected to many modifications during the past decades. Apart from
many older national or international recommendations, the latest ‘specifica-
tions for radio disturbance and immunity measuring apparatus and methods’
within a frequency range of 10 kHz to 1000 MHz are now described in the
CISPR Publication 16-1.30
As defined in this specification, the expression
‘radio disturbance voltage (RDV)’, earlier termed as ‘radio noise’, ‘radio influ-
ence’ or ‘radio interference’ voltages, is now used to characterize the measured
disturbance quantity.

Narrow-band PD instruments, which are now also specified within the
new IEC Standard31
for the measurement of the apparent charge, are very
similar to those RDV meters which are applied for RDV measurements in the
frequency range 100 kHz to 30 MHz. The PD instruments are characterized
by a small bandwidth f and a mid-band frequency fm, which can be varied
over a wider frequency range, where the amplitude frequency spectrum of the
PD current pulses is in general approximately constant. The recommended
values for f and fm for PD instruments are
9 kHz f 30 kHz; and
50 kHz fm 1 MHz. 7.53
It is further recommended that the transfer impedance Zf at frequencies of
fm š f should already be 20 dB below the peak pass-band value.
Commercial instruments of this type may be designed for a larger range of
mid-band frequencies; therefore the standard provides the following note for
the user. ‘During actual apparent charge measurements, mid-band frequencies
fm 1 MHz should only be applied if the readings for such higher values
do not differ from those as monitored for the recommended values of fm.’
This statement denotes that only the constant part of the PD current amplitude
frequency spectrum is an image of the apparent charge. As shown below in
more detail, the response of these instruments to a PD current pulse is a
transient oscillation with the positive and negative peak values of its envelope
proportional to the apparent charge, independent of the polarity of this charge.
Due to the small values of f, the pulse resolution time Tr will be large,
typically above 80 µs.
The application of such instruments often causes some confusion for the
user. A brief description of their basic working principle and their use in PD
measurements will help make things clearer. Figure 7.25 displays the relevant
situation and results.
In general, such instruments are used together with coupling devices
providing high-pass characteristics within the frequency range of the
instrument. Power frequency input currents including harmonics are therefore
suppressed and we may assume that only the PD current pulses converted
to PD voltage pulses are at the input of the amplifying instrument,
which resembles closely a selective voltmeter of high sensitivity (or a
superheterodyne-type receiver) which can be tuned within the frequency
range of interest. Such a narrow-band instrument is again a quasi-
integration device for input voltage pulses. To demonstrate this behaviour,
we assume (Fig. 7.25(a)) an input voltage v1t D V0 expt/T, i.e. an
exponentially decaying input pulse which starts suddenly with amplitude V0
(see Fig. 7.25(b)). The integral of this pulse,
1
0 v1t dt, is V0T and is thus
a quantity proportional to the apparent charge q of a PD current pulse. The

complex frequency spectrum of this impulse is then given by applying the
Fourier integral
V1jω D
1
0
v1t expjωt dt D
V0T
1 C jωT
D
S0
1 C jωT
7.53
and the amplitude frequency spectrum jV1iωj by
jV1jωj D
V0T
1 C ωT2
D
S0
1 C ωT2
7.54
where S0 is proportional to q. From the amplitude frequency spectrum,
sketched in Fig. 7.25(c), it is obvious that the amplitudes decay already to
Tunable narrow-band
amplifier
∆f
f
t
T
0
V2 (t)
V1 (t)
V0
V0 T
3 dB
log ω
ωc = 1/T
20 dB/decade
Area V0 T
V1 (t)
G(iω)
fm
(a)
(b)
(c)
log V1(j ω)
Figure 7.25 Narrow-band amplifiers: some explanations to the impulse
response. (a) Block diagram. (b) Input voltage V1 t, see text. (c) Amplitude
frequency spectrum from V1 t. (d) Idealized transfer function of
narrow-band amplifier. (e) Computed impulse response according to
eqn (7.56) for f D 150 kHz and f ¾
D 9 kHz

(d)
(e)
G(j ω)
G0
∆ω
f(ω) = ω.t0
ωm − ∆ω/2 ωm ω
ωm + ∆ω/2
τ
−200 −160 −120 − 80 −40 0
t0
40 80 120 160 200
msec
3 dB or more than about 30 per cent for the angular frequency of ωc 1/T.
This critical frequency fc is for T D 0.1 µsec only 1.6 MHz, a value which
can be assumed for many PD impulses. As the indication of a narrow-band
instrument, if tuned to fm, will be proportional to the relevant amplitude
of this spectrum at fm the recommendations of the new standard can well be
understood. If the input PD current pulse is, however, distorted by oscillations,
the amplitude frequency spectrum would also be distorted by maxima and
minima which can then be recorded by tuning fm.
If the narrow-band instrument is tuned to the constant part of the spec-
trum which is proportional to q, we may also assume a Dirac pulse or delta
function of magnitude V0T D S0 to calculate its output voltage V2t. As

the spectrum of a Dirac pulse is constant for all frequencies, the response
v2t is then proportional to S0 at any frequency fm. The impulse response
of the instrument is then of course dependent upon the exact (output/input
voltage) transfer function Gjω of the system; we may, however, approx-
imate the actual band-pass characteristic by an idealized one as shown in
Fig. 7.25(d), with a mid-band angular frequency ωm, an angular bandwidth
ω and the constant amplitude or ‘scale factor’ G0 within ωm š ω/2. For
such ideal band-pass systems and especially narrow-band amplifiers the phase
shift =ω may well be assumed to be linear with frequency as indicated, at
least within the band-pass response. With this approximation no phase distor-
tion is assumed, and t0 (see Fig. 7.25(d)) is equal to the delay time of the
system. The impulse response with S0 as input pulse appearing at t D 0 can
then be evaluated47,48
from
v2t D
1
1
ωmCω/2
ωmω/2
S0G0 cos[ωt t0] dω 7.55
This integral can easily be solved; the result is
v2t D
S0G0ω
1
si

ω
2
t t0

cos ωmt t0 7.56
where six D sinx/x.
Equation (7.56) shows an oscillating response whose main frequency is
given by fm D ωm/21, the amplitudes are essentially given by the six func-
tion which is the envelope of the oscillations. A calculated example for such
a response is shown in Fig. 7.25(e). The maximum value will be reached for
t D t0 and is clearly given by
V2 max D
S0G0ω
1
D 2S0G0f 7.57
where f is the idealized bandwidth of the system. Here, the two main
disadvantages of narrow-band receivers can easily be seen: first, for ω − ωm
the positive and negative peak values of the response are equal and therefore
the polarity of the input pulse cannot be detected. The second disadvantage is
related to the long duration of the response. Although more realistic narrow-
band systems will effectively avoid the response amplitudes outside of the first
zero values of the sin x/x function, the full length of the response, with
as defined by Fig. 7.25(e), becomes
D
2
f
D
41
ω
, 7.58
being quite large for small values of f, due to the actual definition of the
‘pulse resolution time Tr’ as defined before. This quantity is about 10 per cent
smaller than , but still much larger than for wide-band PD detectors.

Simple narrow-band detectors use only RLC resonant circuits with high
quality factors Q, the resonance frequency of which cannot be tuned. Although
then their responses are still quite similar to the calculated one (eqn 7.56),
we show such a response for a ‘double pulse’ in Fig. 7.26, taken from a
40 µs
Time t / temps t
Uout
Figure 7.26 Response of a simple narrow-band circuit with f D 10 kHz;
fm D 75 kHz
commercial PD instrument. As the time scale is shown and data for the
frequencies applied are provided, no further explanations are necessary. High-
quality tunable detectors apply the heterodyne principle. Responses for such
instruments can be taken from a RDV meter in front of the ‘psophometric
weighting circuit’ (see Fig. 7.29) as shown in Fig. 7.27. Here again a situation
(Fig. 7.27(b)) is displayed for which superposition errors occurs.
Radio disturbance (interference) meters for the detection of partial
discharges
As instruments such as those specified by the International Special Committee
on Radio Disturbance (Comité International Spécial des Perturbation
Radioélectrique, CISPR) of IEC30
or similar organizations are still in
common use for PD detection, the possible application of an ‘RDV’ or ‘RIV’
meter is still mentioned within the new standard.31
New types of instruments
related to the CISPR Standard are often able to measure ‘radio disturbance
voltages, currents and fields’ within a very large frequency range, based on
different treatment of the input quantity. Within the PD standard, however,
the expression ‘Radio Disturbance Meter’ is only applied for a specific
radio disturbance (interference) measuring apparatus, which is specified for
a frequency band of 150 kHz to 30 MHz (band B) and which fulfils the
requirements for a so-called ‘quasi-peak measuring receivers’.
In Fig. 7.28 a block diagram of such a simple RIV meter is sketched and
compared with the principle of a narrow-band PD instrument as described and

Repetition rate
n = 55 × 103
(Hz)
Repetition rate
n = 104
(Hz)
50 ms/div.
50 ms/div.
Input pulses
Input pulses
(b)
(a)
Figure 7.27 Measured pulse response of an actual narrow-band detector
(RIV meter). Signals taken from the intermediate frequency amplifiers for
repetitive input signals (a) with adequate and (b) inadequate time distances.
Bandwidth f ¾
D 9 kHz
discussed before. The main difference is only the ‘quasi-peak’ or ‘psopho-
metric weighting circuit’ which simulates the physiological noise response of
the human ear. As already mentioned within the introduction of this section,
forthcoming PD instruments will be equipped with a similar, but different
circuit with a ‘pulse train response’ quantifying the ‘largest repeatedly occur-
ring PD magnitudes’. Within the block diagram of Fig. 7.28, the simplified
coupling device as indicated by a resistance shunted by the inductance L forms
a transfer impedance Zm with a high-pass characteristic which for RDV meters

RIV-meter
Narrow-band PD-detector
From Ck or Ct
Narrow-band
amplifier Psophometric
weighting
Indicating
meter
Er
Ii
L V1
V1/Ii
1 µs
t
f
fm
∆f
6 dB
t
f
∼ 100 kHz
∼ 200 µs for ∆f = 9 kHz
V2 max
V2 (t)
Amplification
Figure 7.28 Block diagram of a quasi-peak RIV meter including weighting
circuit compared with PD narrow-band PD detector
have standardized values. Based on the derivations as already made for the
calculation of eqn (7.57) we can now easily quantify the differences of both
types of meter.
The quasi-peak RDV meters are designed with a very accurately defined
overall pass-band characteristic fixed at f D 9 kHz. They are calibrated in
such a way that the response to Dirac type of equidistant input pulses providing
each a volt–time area of 0.316 µVs at a pulse repetition frequency N of
100 Hz is equal to an unmodulated sine-wave signal at the tuned frequency
having an e.m.f. of 2 mV r.m.s. as taken from a signal generator driving the
same output impedance as the pulse generator and the input impedance of the
RIV meter. By this procedure the impulse voltages as well as the sine-wave
signal are halved. As for this repetition frequency of 100 Hz the calibration
point shall be only 50 per cent of V2 max in eqn (7.57), the relevant reading of
the RDV meter will be
ERDV D
1
2
p
2
2S0G0f D
S0G0f
p
2
7.59
As G0 D 1 for a proper calibration and f D 9 kHz, S0 D 158 µVs, the
indicated quantity is S0f/
p
2 D 1 mV or 60 dB µV, as the usual reference
quantity is 1 µV. RDV meters are thus often called ‘microvolt meters’!

This response is now weighted by the ‘quasi-peak measuring circuit’ with a
specified electrical charging time constant 1D1 ms, an electrical discharging
time constant 2D160 ms and by an output voltmeter, which, for conventional
instruments, is of moving coil type, critically damped and having a mechanical
time constant 3D160 ms. This procedure makes the reading of the output
voltmeter dependent on the pulse repetition frequency N. This non-linear func-
tion fN as available from reference 30 or 31 is shown in Fig. 7.29 and is
only accurate if the input pulses are equidistant and of equal amplitudes! It
can be seen that for N 1000 the function fN would saturate to a value
of 2, for which, however, superposition errors occur.
01
1
10
1 10 100 1000
N [1/s]
f
(
N
)
/
f
(100)
Figure 7.29 Variation of CISPR radio interference meter reading with
repetition frequency N, for constant input pulses
With this function fN we can now finalize the reading of an RIV meter by
taking the transfer impedance Zm of our CD in eqn (7.59) into account, which
converts input PD currents into input voltages v1t. For RDV meters, this
transfer impedance, the real value of which jZmj is constant for the frequency
range under consideration, the quantity S0 in eqn (7.59) may then be written as
S0 D

v1t dt D jZmj

i1t dt D jZmj q, 7.60
where q is the measured charge quantity for an impulse current i1t. Now
eqn (7.59) becomes
ERDV D
G0
p
2
qf jZmj fN 7.61
With this equation conversion factors between the measured charge q and the
indicated voltage by an RDV meter can be calculated. For N D 100 equidis-
tant pulses of equal magnitude fN D 1, f D 9 kHz, correct calibration
G0 D 1 and a reading of 1 mV (DERDV) or 60 dB, charge magnitudes of 1

(or 2.6) nC for jZmj D 150 (or 60) 0 can be calculated. These relationships
have also been confirmed experimentally.34,35
Instead of eqn (7.61) the new
standard31
displays in Annex D a reading in which the first term of eqn (7.61)
is generalized, namely
URDV D
qfZmfN
ki
where
N D pulse repetition frequency,
fN D the non-linear function of N (see Fig. 7.29),
f D instrument bandwidth (at 6 dB),
Zm D value of a purely resistive measuring input impedance
of the instrument,
ki D the scale factor for the instrument Dq/URDV
As, however, the weighting of the PD pulses is different for narrow-band
PD instruments and quasi-peak RDV meters, there is no generally applicable
conversion factor between readings of the two instruments. The application
of RDV meters is thus not forbidden; but if applied the records of the tests
should include the readings obtained in microvolts and the determined apparent
charge in picocoulombs together with relevant information concerning their
determination.
Ultra-wide-band instruments for PD detection
The measurement of PD current pulses as briefly treated in section 7.3.2
belongs to this kind of PD detection as well as any similar electrical method
to quantify the intensity of PD activities within a test object. Such methods
need coupling devices with high-pass characteristics which shall have a pass-
band up to frequencies of some 100 MHz or even higher. Records of the
PD events are then taken by oscilloscopes, transient digitizers or frequency
selective voltmeters especially spectrum analysers. For the location of isolated
voids with partial discharges in cables a bandwidth of about some 10 MHz
only is useful, whereas tests on GIS (gas-insulated substations or apparatus)
measuring systems with ‘very high’ or even ‘ultra-high’ frequencies (VHF
or UHF methods for PD detection) can be applied. This is due to the fact
that the development of any partial discharge in sulphur hexafluoride is of
extremely short duration providing significant amplitude frequency spectra up
to the GHz region. More information concerning this technique can be found
in the literature.54,55,56
As none of these methods provides integration capabilities, they cannot
quantify apparent charge magnitudes, but may well be used as a
diagnostic tool.26

7.3.5 Sources and reduction of disturbances
Within the informative Annex G of the IEC Standard31
sources and sugges-
tions regarding the reduction of disturbances are described in detail. A citation
of some of the original text together with some additional information is thus
adequate.
Quantitative measurements of PD magnitudes are often obscured by inter-
ference caused by disturbances which fall into two categories:
Disturbances which occur even if the test circuit is not energized. They may
be caused, for example, by switching operations in other circuits, commu-
tating machines, high-voltage tests in the vicinity, radio transmissions, etc.,
including inherent noise of the measuring instrument itself. They may also
occur when the high-voltage supply is connected but at zero voltage.
Disturbances which only occur when the test circuit is energized but which
do not occur in the test object. These disturbances usually increase with
increasing voltage. They may include, for example, partial discharges in
the testing transformer, on the high-voltage conductors, or in bushings (if
not part of the test object). Disturbances may also be caused by sparking of
imperfectly earthed objects in the vicinity or by imperfect connections in
the area of the high voltage, e.g. by spark discharges between screens and
other high-voltage conductors, connected with the screen only for testing
purposes. Disturbances may also be caused by higher harmonics of the test
voltage within or close to the bandwidth of the measuring system. Such
higher harmonics are often present in the low-voltage supply due to the
presence of solid state switching devices (thyristors, etc.) and are transferred,
together with the noise of sparking contacts, through the test transformer or
through other connections, to the test and measuring circuit.
Some of these sources of disturbances have already been mentioned in the
preceding sections and it is obvious that up to now numerous methods to
reduce disturbances have been and still are a topic for research and develop-
ment, which can only be mentioned and summarized here.
The most efficient method to reduce disturbances is screening and filtering,
in general only possible for tests within a shielded laboratory where all elec-
trical connections running into the room are equipped with filters. This method
is expensive, but inevitable if sensitive measurements are required, i.e. if the
PD magnitudes as specified for the test objects are small, e.g. for h.v. cables.
Straight PD-detection circuits as already shown in Fig. 7.20 are very sensi-
tive to disturbances: any discharge within the entire circuit, including h.v.
source, which is not generated in the test specimen itself, will be detected
by the coupling device CD. Therefore, such ‘external’ disturbances are not
rejected. Independent of screening and filtering mentioned above, the testing

transformer itself should be PD free as far as possible, as h.v. filters or induc-
tors as indicated in Fig. 7.20 are expensive. It is also difficult to avoid any
partial discharges at the h.v. leads of the test circuit, if the test voltages are
very high. A basic improvement of the straight detection circuit may there-
fore become necessary by applying a ‘balanced circuit’, which is similar to a
Schering bridge. In Fig. 7.30 the coupling capacitor CK and test specimen Ct
form the h.v. arm of the bridge, and the l.v. arms are basically analogous to
a Schering bridge. As CK is not a standard capacitor but should be PD free,
the dissipation factor tan υK may also be higher than that of Ct, and therefore
the capacitive branch of the l.v. arm may be switched to any of the two arms.
The bridge can then be adjusted for balance for all frequencies at which tan
υK D tan υt. This condition is best fulfilled if the same insulation media are
used within both capacitors. The use of a partial discharge-free sample for CK
of the same type as used in Ct is thus advantageous. If the frequency depen-
dence of the dissipation factors is different in the two capacitors, a complete
balance within a larger frequency range is not possible. Nevertheless, a fairly
good balance can be reached and therefore most of the sinusoidal or transient
voltages appearing at the input ends of CK and Ct cancel out between the
points 1 and 2. A discharge within the test specimen, however, will contribute
to voltages of opposite polarity across the l.v. arms, as the PD current is
flowing in opposite directions within CK and Ct.
V
Ck Ct ; tan δt
tanδk
1 2
To detector
Figure 7.30 Differential PD bridge (balanced circuit)
Polarity discrimination methods take advantage of the effect of opposite
polarities of PD pulses within both arms of a PD test circuit. Two adequate
coupling devices CD and CD1 as shown in Fig. 7.31 transmit the PD signals
to the special measuring instrument MI, in which a logic system performs the
comparison and operates a gate for pulses of correct polarity. Consequently
only those PD pulses which originate from the test object are recorded and
quantified. This method was proposed by I.A. Black.37,38

Zmi (Zmi)1
Z
U∼
Ca
Ca1
or
(Ck)
CD (CD)1
CC CC
MI
Figure 7.31 Polarity discrimination circuit
Another extensively used method is the time window method to suppress
interference pulses. All kinds of instruments may be equipped with an elec-
tronic gate which can be opened and closed at preselected moments, thus
either passing the input signal or blocking it. If the disturbances occur during
regular intervals the gate can be closed during these intervals. In tests with
alternating voltage, the real discharge signals often occur only at regularly
repeated intervals during the cycles of test voltage. The time window can be
phase locked to open the gate only at these intervals.
Some more sophisticated methods use digital acquisition of partial discharge
quantities, to which the final section 7.3.8 is devoted.
7.3.6 Other PD quantities
The measurement of the ‘apparent charge q’ as the fundamental PD quantity is
widely acknowledged and used today, and only the ‘largest repeatedly occur-
ring magnitudes’ of this kind are usually specified. Individual charge magni-
tudes qi are different, however, as well as the number of partial discharges
recorded within a selected reference time interval. But the deterioration process
within an insulation system is certainly a result of all discharges and is not
limited to the maximum values only. Much research work has been related
to the measurement of all single PD impulses and to the evaluation of the
results on a statistical basis. Such measuring systems are known as PD pulse
analysers and depending on the performance of the detection and analysing
systems, the number of pulses, the pulse intervals or the amplitudes of the
individual pulses may be recorded and stored (section 7.3.8).
Such additional quantities related to PD pulses, although already mentioned
in earlier standards, will be much more used in future and thus their definitions
are given below with brief comments only:

(a) The phase angle =i and time ti of occurrence of a PD pulse is
=i D 360ti/T 7.62
where ti is the time measured between the preceding positive going transition
of the test voltage through zero and the PD pulse. Here T is the period of the
test voltage.
(b) The average discharge current I is the sum of the absolute values
of individual apparent charge magnitudes qi during a chosen reference time
interval Tref divided by this time interval, i.e.:
I D
1
Tref
jq1j C jq2j C . . . C jqij 7.63
This current is generally expressed in coulombs per second or in amperes.
By this definition a quantity is available which includes all individual PD
pulses as well as the pulse repetition rate n. The measurement of this quantity
is possible based upon either linear amplification and rectification of the PD
discharge currents, by processing the output quantities of the apparent charge
detectors by integration and averaging or by digital post processing. This
average discharge current has not been investigated extensively up to now,
although early investigations show quite interesting additional information
about the impact on the lifetime of insulation.31
(c) The discharge power P is the average pulse power fed into the termi-
nals of the test object due to apparent charge magnitudes qi during a chosen
reference time interval Tref, i.e.:
P D
1
Tref
q1u1 C q2u2 C . . . C qiui 7.64
where u1, u2, . . . , ui are instantaneous values of the test voltage at the instants
of occurrence ti of the individual apparent charge magnitudes qi. This quantity
is generally expressed in watts. In this equation the sign of the individual
values must be strictly observed, which is often difficult to fulfil. Narrow-band
PD instruments are not able to quantify the polarity of PD events and even
the response of wide-band instruments may not be clear, see Fig. 7.26. In the
vicinity of the test voltage zero PD pulses and instantaneous voltage are often
different in polarity!36,45
As discharge energy is directly related to discharge
power, this quantity is always directly related to insulation decomposition.46
(d) The quadratic rate D is the sum of the squares of the individual apparent
charge magnitudes qi during a chosen reference time interval Tref divided by
this time interval, i.e.:
D D
1
Tref
q2
1 C q2
2 C . . . C q2
m 7.65

and is generally expressed in (coulombs)2
per second. Although this quantity
appears to have no advantages compared to the measurement of the maximum
values of q only32
, some commercially available, special instruments record
this quantity.
7.3.7 Calibration of PD detectors in a complete test circuit
The reasons why any PD instrument providing continuously variable sensi-
tivity must be calibrated in the complete test circuit have mainly be explained
within sections 7.3.1 to 7.3.3. Even the definition of the ‘apparent charge q’
is based on a routine calibration procedure, which shall be made with each
new test object. Calibration procedures are thus firmly defined within the
standard.31
A calibration of measuring systems intended for the measurement of the
fundamental quantity q is made by injecting short duration repetitive current
pulses of well-known charge magnitudes q0 across the test object, whatever
test circuit is used. For an example, see Fig. 7.32. These current pulses are
generally derived from a calibrator which comprises a generator producing step
voltage pulses (see ‘G’) of amplitude V0 in series with a precision capacitor
C0. If the voltages V0 also remain stable and are exactly known, repetitive
calibration pulses with charge magnitudes of q0=V0C0 are injected. A short
rise time of 60 ns is now specified for the voltage generator to produce current
pulses with amplitude frequency spectra which fit the requirements set by the
bandwidth of the instruments and to avoid integration errors if possible.
Z
U∼
Ca
Co
Cs
Ck
CD MI
CC
Zmi
G
Figure 7.32 The usual circuit for the calibration of a PD measuring
instrument MI within the complete test circuit. For identification of circuit
elements see text and Fig. 7.20

Whereas further details for the calibration procedures shall not be discussed
here, the new philosophy in reducing measuring errors during PD tests will
be presented.52
It has been known for some time that measuring uncertainties in PD measure-
ments are large. Even today, PD tests on identical test objects performed with
different types of commercially available systems will provide different results
even after routine calibration performed with the same calibrator. The main
reasons for this uncertainty are the different transfer impedances (bandwidth)
of the measuring systems, which up to 1999 have never been well defined
and quantified. The new but not very stringent requirements31
related to this
property will improve the situation; together with other difficulties related to
disturbance levels measuring uncertainties of more than about 10 per cent
may, however, exist. The most essential part of the new philosophy concerns
the calibrators, for which – up to now – no requirements for their performance
exist. Tests on daily used commercial calibrators sometimes display deviations
of more than 10 per cent of their nominal values. Therefore routine type, and
performance tests on calibrators have been introduced with the new standard.
At least the first of otherwise periodic performance tests should be traceable
to national standards, this means they shall be performed by an accredited
calibration laboratory. With the introduction of this requirement it can be
assumed that the uncertainty of the calibrator charge magnitudes q0 can be
assessed to remain within š5 per cent or 1 pC, whichever is greater, from its
nominal values. Very recently executed intercomparison tests on calibrators
performed by accredited calibration laboratories showed that impulse charges
can be measured with an uncertainty of about 3 per cent.42
7.3.8 Digital PD instruments and measurements
Between 1970 and 1980 the state of the art in computer technology and related
techniques rendered the first application of digital acquisition and processing of
partial discharge magnitudes.39,40,41
Since then this technology was applied
in numerous investigations generally made with either instrumentation set
up by available components or some commercial instruments equipped with
digital techniques. One task for the working group evaluating the new IEC
Standard was thus concerned with implementing some main requirements for
this technology. It is again not the aim of this section to go into details of
digital PD instruments, as too many variations in designing such instruments
exist. Some hints may be sufficient to encourage further reading.
Digital PD instruments are in general based on analogue measuring systems
or instruments for the measurement of the apparent charge q (see section 7.3.4)
followed by a digital acquisition and processing system. These digital parts
of the system are then used to process analogue signals for further evaluation,
to store relevant quantities and to display test results. It is possible that in

the near future a digital PD instrument may also be based on a high-pass
coupling device and a digital acquisition system without the analogue signal
processing front end. The availability of cheap but extremely fast flash A/D
converters and digital signal processors (DSPs) performing signal integration
is a prerequisite for such solutions.
10.0
5.00
0.0
1.3E2
1.3E3
6.5E1
6.5E2
0.0E0
0.0E0
[N]
[N]
0 180 360
[ deg ]
0 180 360
[ deg ]
0.0 5.00 10.0
[ pC ]
[pC]
Figure 7.33 The pattern of a phase-resolved PD measurement for a moving
metal particle within a GIS. Further details see text (courtesy FKH, Zurich,
Switzerland)
The main objective of applying digital techniques to PD measurements is
based on recording in real time at least most of consecutive PD pulses quanti-
fied by its apparent charge qi occurring at time instant ti and its instantaneous
values of the test voltage ui occurring at this time instant ti or, for alter-
nating voltages, at phase angle of occurrence =i within a voltage cycle of

the test voltage. As, however, the quality of hard- and software used may
limit the accuracy and resolution of the measurement of these parameters, the
new standard31
provides some recommendations and requirements which are
relevant for capturing and registration of the discharge sequences.
One of the main problems in capturing the output signals of the analogue
front end correctly may well be seen from Figs 7.24 and 7.26, in which three
output signals as caused by two consecutive PD events are shown. Although
none of the signals is distorted by superposition errors, several peaks of each
signal with different polarities are present. For the wideband signals, only the
first peak value shall be captured and recorded including polarity, which is
not easy to do. For the narrow-band response for which polarity determination
is not necessary, only the largest peak is proportional to the apparent charge.
For both types of signals therefore only one peak value shall be quantified,
recorded and stored within the pulse resolution time of the analogue measuring
system. Additional errors can well be introduced by capturing wrong peak
values which add to the errors of the analogue front end.
−980
−820
−660
−500
−340
−180
−20
140
300
460
620
780
940
7
71
136
201
266
331
0
2
4
6
8
10
12
n
q
phi
Figure 7.34 An example of a = q n diagram. On-site PD measurements
performed on an h.v. cable, heavy partial discharges at a terminator
(courtesy Presco AG, Weiningen, Switzerland)
Further aims of PD instruments are related to post-processing of the recorded
values. Firstly, the so-called ‘=i qi ni’ patterns as available from the
recorded and stored data in which ni is the number of identical or similar PD
magnitudes recorded within short time (or phase) intervals and an adequate
total recording duration can be used to identify and localize the origin of the

PDs based on earlier experiencesee, e.g., 49,51,53,57,59,67
and/or even to estab-
lish physical models for specific PD processes.60 and cited references
If recorded
raw data are too much obscured by disturbances, quite different numerical
methods may also be applied to reduce the disturbance levels.61,62
We end this chapter with two records of results from PD tests made with
digital PD instrument. In Figs 7.33 and 7.34 typical test results of phase
resolved PD measurement for a moving metal particle within a GIS and on-site
PD measurements performed on HV cable (at a terminator) are shown.
For further reading about PD measurements and their applications see refer-
ences 63 to 66.
References
1. A.K. Jonscher, Dielectric Relaxation in Solids. Chelsea Dielectrics Press, London, 1983.
2. Waldemar von Münch, Elektrische und magnetische Eigenschaften der Materie.
B.G. Teubner, Stuttgart, 1987.
3. A.K. Jonscher and R.M. Hill Physics of Thin Films, Vol. 8, 1975, pp. 169–249.
4. N.F Mott and E.A. Davies Electronic Processes in Non-cristalline Materials. Oxford Univer-
sity Press 1979.
5. Der Houhanessian, Vahe. Measurement and Analysis of Dielectric Response in Oil-Paper
Insulation Systems. Thesis Swiss Federal Institute of Technology, Diss. ETH No. 12, 832,
1998.
6. Bernhard Gross. Dielectric relaxation functions and models. J. Appl. Phys. 67(10), May 1990,
pp. 6399–6404.
7. Der Houhanessian, Vahe and Zaengl, Walter S. Application of relaxation current measure-
ments to on-site diagnosis of power transformers. 1997 IEEE Annual Report, Conference on
El. Insulation and Dielectric Phenomena, Minneapolis, October 1997, pp. 45–51.
8. J. Jalbert and R. Gilbert Decomposition of Transformer Oils: A New Approach for the Deter-
mination of Dissolved Gases. IEEE Trans. PD Vol. 12, pp. 754–760, 1997.
9. IEC Standard 61181 (1993). Impregnated insulating materials – Application of dissolved gas
analysis (DGA) to factory tests on electrical equipment.
10. E. Ildstad, P. Thärning and U. Gäfvert Relation between voltage return and other methods
for measurement of dielectric response. 1994 IEEE Symposium on El. Insulation, June 1994,
Baltimore, USA.
11. A. Bognar, L. Kalocsai, G. Csepes, E. Nemeth and J. Schmidt Diagnostic tests of high
voltage oil-paper insulating systems (in particular transformer insulation) using DC dielec-
trometrics. CIGRE 1990 Session, paper 15/33-08.
12. A. Van Roggen An Overview of Dielectric Measurements. IEEE Trans. on El. Insulation
Vol. 25, 1990, pp. 95–106.
13. W.P. Baker. Electrical Insulation Measurements. Newnes International Monographs on Elec-
trical Engineering and Electronics, 1965.
14. A.J. Schwab. High Voltage Measurement Techniques. M.I.T Press, 1972.
15. See Chapter 4 in L. Schnell (ed.). Technology of Electrical Measurements. John Wiley and
Sons Ltd, 1993, or: B. Hague. Alternating-Current Bridge Methods (5th ed). Pitman Sons,
London, 1962.
16. W. Brueckel. The Commutable Schering-Bridge of Tettex AG. TETTEX Information No. 1
(1983).

17. K.W. Wagner. Zur Messung dielektrischer Verluste mit der Wechselstrombrücke (To the
measurement of dielectric losses with the a.c. bridge). Elektrotechnische Zeitschrift Vol. 32
(1911), p. 1001.
18. W.B. Baker. Recent developments in 50c/s bridge networks with inductively coupled ratio
arms for capacitance and loss-tangent measurements. Proceedings of IEE Part A, Vol. 109
(1962), pp. 243–247.
19. P.N. Miljanic, N.L. Kusters and W.J.M. Moore. The development of the current comparator,
a high-accuracy a.c. ratio measuring device. AIEE Transactions, Pt I (Communication and
Electronics), Vol. 81 (Nov. 1962), pp. 359–368.
20. O. Peterson. A self-balancing high-voltage capacitance bridge. IEEE Trans. on Instr. and
Meas. Vol. IM13 (1964), pp. 216–224.
21. P. Osvath and S. Widmer. A High-Voltage High-Precision Self-Balancing Capacitance and
Dissipation Factor Measuring Bridge. IEEE Trans. on Instr. and Meas. Vol. IM35, (1986),
pp. 19–23.
22. L. Lundgaard, M. Runde and B. Skyberg. Acoustic diagnosis of gas insulated substations;
a theoretical and experimental basis. IEEE Trans. on Power Delivery Vol. 5 (1990),
pp. 1751–1760.
23. P. Moro and J. Poittevin. Localization des décharges partielles dans le transformateurs par
détection des ondes ultrasonores emises. Rev. Générale de I’Electricité 87 (1978), pp. 25–35.
24. E. Howells and E.T. Norton. Detection of partial discharges in transformers using emission
techniques. Trans. IEEE PAS97 (1978), pp. 1538–1549.
25. R.T. Harrold. Acoustic waveguides for sensing and locating electrical discharges in
H.V. power transformers and other apparatus. Trans. IEEE PAS98 (1979), pp. 449–457.
26. F.H. Kreuger. Discharge Detection in High Voltage Equipment. Heywood, London, 1964;
Elsevier, New York, 1964, Temple Press Books, 1964.
27. NEMA Publication No. 107 (1940). Methods of Measuring Radio Noise.
28. L. Satish and W.S. Zaengl. An effort to find near-optimal band-pass filter characteristics for
use in partial discharge measurements. European Transactions on Electrical Power Engi-
neering, ETEP 4, No. 6 (1994), pp. 557–563.
29. W.S. Zaengl, P. Osvath and H.J. Weber. Correlation between the bandwidth of PD detec-
tors and its inherent integration errors. 1986 IEEE Intern. Symp. on Electr. Insul. (ISEI),
Washington DC. June 1986, Conference Rec. pp. 115–121.
30. CISPR 16-1 (1993). Specifications for radio disturbance and immunity measuring apparatus
and methods – Part 1: Radio disturbance and immunity measuring apparatus.
31. IEC Standard 60270 (Third edition, 2000). Partial Discharge Measurements. International
Electrotechnical Commission (IEC), Geneva, Switzerland.
32. Partial Discharge Measurements. IEC Standard 270, 1981.
33. H. Polek. Kompensator-Messbrücke für Kapazitäts- und Verlustfaktormessung mit
Registrierung. Elektrotechnische Zeitschrift ETZ, Vol. 76 (1955), pp. 822–826.
34. E.M. Dembinski and J.L. Douglas. Calibration of partial-discharge and radio interference
measuring circuits. Proc. IEE 115 (1968), pp. 1332–1340.
35. R.T. Harrold and T.W. Dakin. The relationship between the picocoulomb and microvolt for
corona measurements on hv transformers and other apparatus. Trans. IEEE PAS92 (1973),
pp. 187–198.
36. Th. Praehauser. Lokalisierung von Teilentladungen in Hochspannungs-apparaten. Bull. SEV.
63 (1972), pp. 893–905.
37. British Patent No. 6173/72. Improvements in or relating to High Voltage Component Testing
Systems.
38. I.A. Black. A pulse discrimination system for discharge detection in electrically noisy envi-
ronments. 2nd Int. High Voltage Symposium (ISH), Zurich, Switzerland, 1975, pp. 239–243.
39. R. Bartnikas. Use of multichannel analyzer for corona pulse-height distribution measurements
on cables and other electrical apparatus. Trans. IEEE IM-22 (1973), pp. 403–407.

40. S. Kärkkainen. Multi-channel pulse analyzer in partial discharge studies. 2nd Int. High
Voltage Symposium (ISH), Zurich, 1975, pp. 244–249.
41. K. Umemoto, E. Koyanagy, T. Yamada and S. Kenjo. Partial discharge measurement system
using pulse-height analyzers. 3rd Int. Symp. on High Voltage Engg. (ISH), Milan, 1979,
Report 43.07.
42. K. Schon and W. Lucas. Intercomparison of impulse charge measurements. EU Synthesis
Report on Project SMT4-CT95-7501, 1998.
43. B. Kübler. Investigation of partial discharge measuring techniques using epoxy resin samples
with several voids. IEEE Intern. Symp. on El. Insulation, 1978, 78 CH 1287-2EI (see also
Ph.D. Thesis, Techn. University Braunschweig, Germany, 1978).
44. J. Carlier et al. Ageing under voltage of the insulation of rotating machines: influence of
frequency and temperature. CIRGRE-Rapport No. 15-06, 1976.
45. R. Bartnikas and E.J. McMahon. Corona measurement and interpretation. Engineering
Dielectrics Vol. 1, ASTM STP 669, 1979.
46. F. Viale et al. Study of a correlation between energy of partial discharges and degradation
of paper-oil-insulation. CIGRE Session 1982, report 15-12.
47. K. Kuepfmueller. Die Systemtheorie der elektrischen Nachrichten-uebertragung. S. Hirzel
Verlag, Stuttgart, 1968.
48. A. Papoulis. The Fourier Integral and its Applications. McGraw-Hill, 1962.
49. A.G. Millar et al. Digital acquisition, storage and processing of partial discharge signals. 4th
Int. Symp. on High Voltage Engg. (ISH), Athens 1983, Report 63.01.
50. Presco AG, Weiningen, Switzerland. Capacitance and Dissipation Factor Measuring Bridge
TG-1Mod. Company brochure 1999.
51. L. Satish and W.S. Zaengl. Artificial neural networks for recognition of 3-d partial discharge
pattern. IEEE Trans. on Dielectrics and El. Insulation Vol. 1 (1994), pp. 265–275.
52. Calibration procedures for analog and digital partial discharge measuring instruments. Electra
No. 180 (Oct. 1998), pp. 123–143.
53. Partial discharge measurement as a diagnostic tool. Electra No. 181 (Dec. 1998), pp. 25–51.
54. Insulation co-ordination of GIS: Return of experience, on site tests and diagnostic techniques.
Electra No. 176 (Feb. 1998), pp. 67–97.
55. J.S. Pearson, O. Farish, B.F. Hampton et al. PD diagnostics for gas insulated substations.
IEEE Trans. on Dielectrics and El. Insulation Vol. 2 (1995), pp. 893–905.
56. B.F. Hampton et al. Experience and progress with UHF diagnostics in GIS. CIGRE Session
1992, Report 15/23-03.
57. E. Gulski et al. Experiences with digital analysis of discharges in high voltage components.
IEEE Electrical Insulation Magazine Vol. 15, No. 3 (1999), pp. 15–24.
58. G.C. Stone. Partial discharge part XXV: Calibration of PD measurements for motor and
generator windings – why it cant’t be done. IEEE Electrical Insulation Magazine Vol. 14,
No. 1 (1998), pp. 9–12.
59. International Conference on Partial Discharge. University of Kent in Canterbury, UK, Sept.
1993. Conference Proceedings No. 378, IEE 1993.
60. K. Wu, Y. Suzuoki, T. Mizutani and H. Xie. A novel physical model for partial discharge
in narrow channels. IEEE Trans. on Dielectrics and Electr. Insulation Vol. 6 (1999),
pp. 181–190.
61. Beierl et al. Intelligent monitoring and control systems for modern AIS and GIS substations.
CIGRE Session 1998, paper 34–113.
62. U. Köpf and K. Feser. Noise suppression in partial discharge measurements. 8th Int. Symp.
on HV Engineering, Yokohama, Japan (1995), paper No. 63.02.
63. IEEE Committee Report. Digital techniques for partial discharge measurements. IEEE Trans.
on Power Delivery Vol. 7 (1992), pp. 469–479 (with 105 references).
64. IEEE Committee Report. Partial discharge testing of gas insulated substations. IEEE Trans.
on Power Delivery Vol. 7 (1992), pp. 499–506 (with 25 references).

65. Special Issue: The Volta Colloquium on Partial Discharge Measurements. IEEE Trans. on
Electrical Insulation Vol. 27, No. 1 (Feb. 1992), with 13 original papers related to PD.
66. Special Issue: Partial Discharge Measurement and Interpretation. IEEE Trans. on Dielectrics
and Electrical Insulation Vol. 2, No. 4 (Aug. 1995), with 19 original papers related to PD.
67. M. Hoof, B. Freisleben and R. Patsch. PD source identification with novel discharge parame-
ters using counterpropagation neural networks. IEEE Trans. on Dielectrics and El. Insulation
Vol. 4 (1997), pp. 17–32.

Chapter 8
Overvoltages, testing procedures and
insulation coordination
Power systems are always subjected to overvoltages that have their origin in
atmospheric discharges in which case they are called external or lightning
overvoltages, or they are generated internally by connecting or disconnecting
the system, or due to the systems fault initiation or extinction. The latter
type are called internal overvoltages. This class may be further subdivided
into (i) temporary overvoltages, if they are oscillatory of power frequency or
harmonics, and (ii) switching overvoltages, if they are heavily damped and
of short duration. Temporary overvoltages occur almost without exception
under no load or very light load conditions. Because of their common origin
the temporary and switching overvoltages occur together and their combined
effect has to be taken into account in the design of h.v. systems insulation.
The magnitude of the external or lightning overvoltages remains essentially
independent of the system’s design, whereas that of internal or switching
overvoltages increases with increasing the operating voltage of the system.
Hence, with increasing the system’s operating voltage a point is reached
when the switching overvoltages become the dominant factor in designing
the system’s insulation. Up to approximately 300 kV, the system’s insula-
tion has to be designed to withstand primarily lightning surges. Above that
voltage, both lightning and switching surges have to be considered. For ultra-
h.v. systems, 765 kV and above switching overvoltages in combination with
insulator contamination become the predominating factor in the insulation
design.1Ł
For the study of overvoltages occurring in power systems, a thor-
ough knowledge of surge propagation laws is needed which can be found in
a number of textbooks2,3
and will not be discussed here.
8.1 The lightning mechanism
Physical manifestations of lightning have been noted in ancient times, but the
understanding of lightning is relatively recent. Franklin carried out experiments
on lightning in 1744–1750, but most of the knowledge has been obtained over
the last 50 to 70 years. The real incentive to study lightning came when elec-
tric transmission lines had to be protected against lightning. The methods

Overvoltages, testing procedures and insulation coordination 461
include measurements of (i) lightning currents, (ii) magnetic and electromag-
netic radiated fields, (iii) voltages, (iv) use of high-speed photography and
radar.
Fundamentally, lightning is a manifestation of a very large electric discharge
and spark. Several theories have been advanced to explain accumulation of
electricity in clouds and are discussed in references 4, 5 and 6. The present
section reviews briefly the lightning discharge processes.
In an active thunder cloud the larger particles usually possess negative
charge and the smaller carriers are positive. Thus the base of a thunder cloud
generally carries a negative charge and the upper part is positive, with the
whole being electrically neutral. The physical mechanism of charge separation
is still a topic of research and will not be treated here. As will be discussed
later, there may be several charge centres within a single cloud. Typically the
negative charge centre may be located anywhere between 500 m and 10 000 m
above ground. Lightning discharge to earth is usually initiated at the fringe of
a negative charge centre.
To the eye a lightning discharge appears as a single luminous discharge,
although at times branches of variable intensity may be observed which termi-
nate in mid-air, while the luminous main channel continues in a zig-zag path
to earth. High-speed photographic technique studies reveal that most lightning
strokes are followed by repeat or multiple strokes which travel along the path
established by the first stroke. The latter ones are not usually branched and
their path is brightly illuminated.
The various development stages of a lightning stroke from cloud to earth
as observed by high-speed photography is shown diagrammatically in Fig. 8.1
Cloud
Stepped
leader
Ground
Dart leader Dart leader
Return stroke
100 ms 100 ms 100 ms
0.03 sec 0.03 sec
Return
stroke
1000 ms
1000 ms
20 000 ms
Current measured
at ground
Time
Figure 8.1 Diagrammatic representation of lightning mechanism and
ground current3

together with the current to ground. The stroke is initiated in the region of the
negative charge centre where the local field intensity approaches ionization
field intensity (¾
D30 kV/cm in atmospheric air, or ¾10 kV/cm in the presence
of water droplets).
During the first stage the leader discharge, known as the ‘stepped leader’,
moves rapidly downwards in steps of 50 m to 100 m, and pauses after each
step for a few tens of microseconds. From the tip of the discharge a ‘pilot
streamer’ having low luminosity and current of a few amperes propagates into
the virgin air with a velocity of about 1 ð 105
m/sec. The pilot streamer is
followed by the stepped leader with an average velocity of about 5 ð 105
m/sec
and a current of some 100 A. For a stepped leader from a cloud some
3 km above ground shown in Fig. 8.1 it takes about 60 m/sec to reach the
ground. As the leader approaches ground, the electric field between the leader
and earth increases and causes point discharges from earth objects such
as tall buildings, trees, etc. At some point the charge concentration at the
earthed object is high enough to initiate an upwards positive streamer. At
the instance when the two leaders meet, the ‘main’ or ‘return’ stroke starts
from ground to cloud, travelling much faster (¾50 ð 106
m/sec) along the
previously established ionized channel. The current in the return stroke is in
the order of a few kA to 250 kA and the temperatures within the channel
are 15 000°C to 20 000°C and are responsible for the destructive effects
of lightning giving high luminosity and causing explosive air expansion.
The return stroke causes the destructive effects generally associated with
lightning.
The return stroke is followed by several strokes at 10- to 300-m/sec inter-
vals. The leader of the second and subsequent strokes is known as the ‘dart
leader’ because of its dart-like appearance. The dart leader follows the path of
the first stepped leader with a velocity about 10 times faster than the stepped
leader. The path is usually not branched and is brightly illuminated.
A diagrammatic representation of the various stages of the lightning stroke
development from cloud to ground in Figs 8.2(a) to (f) gives a clearer appre-
ciation of the process involved. In a cloud several charge centres of high
concentration may exist. In the present case only two negative charge centres
are shown. In (a) the stepped leader has been initiated and the pilot streamer
and stepped leader propagate to ground, lowering the negative charges in the
cloud. At this instance the striking point still has not been decided; in (b) the
pilot streamer is about to make contact with the upwards positive streamer
from earth; in (c) the stroke is completed, a heavy return stroke returns to
cloud and the negative charge of cloud begins to discharge; in (d) the first
centre is completely discharged and streamers begin developing in the second
charge centre; in (e) the second charge centre is discharging to ground via
the first charge centre and dart leader, distributing negative charge along the
channel. Positive streamers are rising up from ground to meet the dart leader;

(a) Charge centres in cloud;
pilot streamer and stepped
leader propagate earthward;
outward branching of streamers
to earth. Lowering of charge
into space beneath cloud.
(b) Process of (a) almost
completed; pilot streamer
about to strike earth.
(c) Heavy return streamer;
discharge to earth of
negatively charged space
beneath cloud.
(d) First charge centre completely
discharged; development
of streamers between charge
centres within cloud.
(e) Discharge between two
charge centres; dart leader
propagates to ground along
original channel; dart leader
about to strike earth; negative
charge lowered and distributed
along stroke channel.
(f) Heavy return streamer discharge
to earth of negatively charged
space beneath cloud.
Figure 8.2 Schematic representation of various stages of lightning stroke
between cloud and ground6
(f) contact is made with streamers from earth, heavy return stroke proceeds
upwards and begins to discharge negatively charged space beneath the cloud
and the second charge centre in the cloud.
Lightning strokes from cloud to ground account only for about 10 per
cent of lightning discharges, the majority of discharges during thunderstorms

take place between clouds. Discharges within clouds often provide general
illumination known as ‘sheath lightning’.
Measurements of stroke currents at ground have shown that the high current
is characterized by a fast rise to crest (1 to 10 µsec) followed by a longer
decay time of 50–1000 µsec to half-time. Figure 8.3 gives the probability
distribution of times to crest for lightning strokes as prepared by Anderson.7
There is evidence that very high stroke currents do not coincide with very
short times to crest. Field data3,20
indicate that 50 per cent of stroke currents
including multiple strokes have a rate of rise exceeding 20 kA/µsec and 10
per cent exceed 50 kA/µsec. The mean duration of stroke currents above half
value is 30 µsec and 18 per cent have longer half-times than 50 µsec. Thus
for a typical maximum stroke current of 10 000 A a transmission line of surge
impedance (say) Z D 400 and assuming the strike takes place in the middle
of the line with half of the current flowing in each direction Z ¾
D 200
the lightning overvoltage becomes V D 5000 ð 400 D 2 MV. Based on many
investigations the AIEE Committee8
has produced the frequency distribution
of current magnitudes, shown in Fig. 8.4, which is often used for performance
calculations. Included in Fig. 8.4 is a curve proposed by Anderson.7
100
80
60
40
20
0
0 1 2 3 4
Time to crest (ms)
5 6
Probability
of
exceeding
abscissa
(%)
Figure 8.3 Distribution of times to crest of lightning stroke currents (after
Anderson7

The data on lightning strokes and voltages has formed the basis for estab-
lishing the standard impulse or lightning surge for testing equipment in labo-
ratories. The standard lightning impulse waveshape will be discussed later in
this chapter.
8.1.1 Energy in lightning
To estimate the amount of energy in a typical lightning discharge let us assume
a value of potential difference of 107
V for a breakdown between a cloud

0.05
10 20 40 60 100 200
0.1
0.2
0.5
1
2
5
10
20
40
60
80
90
95
98
99
1 2 4 6 10 20
2
Probability
of
exceeding
abscissa
(%)
Stroke current (kA)
1
1
Figure 8.4 Cumulative distributions of lightning stroke current magnitudes:
1. After AIEE Committee.8
2. After Anderson6
and ground and a total charge of 20 coulombs. Then the energy released is
20 ð 107
Ws or about 55 kWh in one or more strokes that make the discharge.
The energy of the discharge dissipated in the air channel is expended in several
processes. Small amounts of this energy are used in ionization of molecules,
excitations, radiation, etc. Most of the energy is consumed in the sudden
expansion of the air channel. Some fraction of the total causes heating of
the struck earthed objects. In general, lightning processes return to the global
system the energy that was used originally to create the charged cloud.
8.1.2 Nature of danger
The degree of hazard depends on circumstances. To minimize the chances of
being struck by lightning during thunderstorm, one should be sufficiently far
away from tall objects likely to be struck, remain inside buildings or be well
insulated.
A direct hit on a human or animal is rare; they are more at risk from
indirect striking, usually: (a) when the subject is close to a parallel hit or
other tall object, (b) due to an intense electric field from a stroke which can

induce sufficient current to cause death, and (c) when lightning terminating on
earth sets up high potential gradients over the ground surface in an outwards
direction from the point or object struck. Figure 8.5 illustrates qualitatively
the current distribution in the ground and the voltage distribution along the
ground extending outwards from the edge of a building struck by lightning.9
The potential difference between the person’s feet will be largest if his feet are
separated along a radial line from the source of voltage and will be negligible
if he moves at a right angle to such a radial line. In the latter case the person
would be safe due to element of chance.
Voltage
0 Distance
(a)
(b)
Figure 8.5 Current distribution and voltage distribution in ground due to
lightning stroke to a building (after Golde9
)
8.2 Simulated lightning surges for testing
The danger to electric systems and apparatus comes from the potentials that
lightning may produce across insulation. Insulation of power systems may be
classified into two broad categories: external and internal insulation. External
insulation is comprised of air and/or porcelain, etc., such as conductor-to-
tower clearances of transmission lines or bus supports. If the potential caused

0
1
1 2 3 4 5
Gap spacing (m)
6 7
2
3
4
5
6
Voltage
(MV)
5(−) ms
5(+) ms
CFO(−)
CFO(+)
8 m
9.5 m
2(+)
3(+)
2(−) 3(−)
Figure 8.6 Impulse (1.2/50 µsec) flashover characteristics of long rod gaps
corrected to STP (after Udo10
)
6
5
4
3
2
1
0 10 20 30 40 50
Number of insulators (254 × 146 mm)
Voltage
(MV)
+CFO
−CFO
Negative
Positive
µs µs
2
2
3
3
10
10
5
5
Figure 8.7 Impulse (1.2/50 µsec) flashover characteristics for long insulator
strings (after Udo10
)

by lightning exceeds the strength of insulation, a flashover or puncture occurs.
Flashover of external insulation generally does not cause damage to equip-
ment. The insulation is ‘self-restoring’. At the worst a relatively short outage
follows to allow replacement of a cheap string of damaged insulation. Internal
insulation most frequently consists of paper, oil or other synthetic insulation
which insulates h.v. conductors from ground in expensive equipment such as
transformers, generators, reactors, capacitors, circuit-breakers, etc. Failure of
internal insulation causes much longer outages. If power arc follows damage
to equipment it may be disastrous and lead to very costly replacements.
The system’s insulation has to be designed to withstand lightning voltages
and be tested in laboratories prior to commissioning.
Exhaustive measurements of lightning currents and voltages and long expe-
rience have formed the basis for establishing and accepting what is known
as the standard surge or ‘impulse’ voltage to simulate external or lightning
overvoltages. The international standard lightning impulse voltage waveshape
is an aperiodic voltage impulse that does not cross the zero line which reaches
its peak in 1.2 µsec and then decreases slowly (in 50 µsec) to half the peak
value. The characteristics of a standard impulse are its polarity, its peak value,
its front time and its half value time. These have been defined in Chapter 2,
Fig. 2.23.
Extensive laboratory tests have shown that for external insulation the light-
ning surge flashover voltages are substantially proportional to gap length and
that positive impulses give significantly lower flashover values than negative
ones. In addition, for a particular test arrangement, as the applied impulse crest
is increased the instant of flashover moves from the tail of the wave to the
crest and ultimately to the front of the wave giving an impulse voltage–time
(‘V–T’) characteristic as was discussed in Chapter 5, Fig. 5.45. Figures 8.6
and 8.7 show typical impulse sparkover characteristics for long rod gaps
and suspension insulators obtained by Udo10
at various times to flashover.
These figures include the critical or long time flashover characteristics (CFO)
occurring at about 10 µsec on the wave tail as well as the characteristics corre-
sponding to shorter time lags near the wave crest. Data for both polarities are
shown. The values plotted in Figs 8.6 and 8.7 have been corrected to standard
atmospheric conditions.
8.3 Switching surge test voltage characteristics
In power transmission systems with systems voltages of 245 kV and above,
the electrical strength of the insulation to switching overvoltages becomes
important for the insulation design. A considerable amount of data on break-
down under switching surges is available. However, a variety of switching
surge waveshapes and the correspondingly large range of flashover values

make it difficult to choose a standard shape of switching impulses. Many tests
have shown that the flashover voltage for various geometrical arrangements
under unidirectional switching surge voltages decreases with increasing the
front duration of the surge, reaching the lowest value somewhere in the range
between 100 and 500 µsec. The time to half-value has less effect upon the
breakdown strength because flashover almost always takes place before or at
the crest of the wave. Figure 8.8 illustrates a typical relationship for a crit-
ical flashover voltage per metre as a function of time to flashover for a 3-m
rod-rod gap and a conductor-plane gap respectively.11
It is seen that the stan-
dard impulse voltages give the highest flashover values, with the switching
surge values of crest between approx. 100 and 500 µsec falling well below the
corresponding power frequency flashover values.
0.6
0.4
0.2
0
1 10 100 1000
3
p.f.
1
2
Time to flashover (µs)
MV/m
Figure 8.8 Relationship between vertical flashover voltage per metre and
time to flashover (3 m gap). 1. Rod-rod gap. 2. Conductor-plane gap.
3. Power frequency
The relative effect of time to crest upon flashover value varies also with
the gap spacing and humidity.21
Figure 8.9 compares the positive flashover
characteristics of standard impulses and 200/2000 µsec with power frequency
voltages for a rod-rod gap plotted as flashover voltage per metre against gap
spacing.11
We observe a rapid fall in switching surge breakdown strength
with increasing the gap length. This drastic fall in the average switching surge
strength with increasing the insulation length leads to costly design clearances,
especially in the ultra-h.v. regions. All investigations show that for nearly all
gap configurations which are of practical interest, positive switching impulses
result in lower flashover voltage than negative ones. The flashover behaviour
of external insulations with different configurations under positive switching
impulse stress is therefore most important. The switching surge voltage break-
down is also affected by the air humidity. Kuffel et al.22
have reported that
over the range from 3 to 16 g/m3
absolute humidity, the breakdown voltage

0.6
0.4
0.2
MV/m
0 2 4 6
1
2
3
D
H
Gap spacing (m)
Figure 8.9 Relationship between flashover voltage (MV/m) and gap length
for 1: 1.2/50 µsec impulses, 2: 200/2000 µsec switching surges and 3: power
frequency voltages
of positive rod gaps increases approximately 1.7 per cent per 1 g/m3
increase
in absolute humidity.
For testing purposes the standard switching surge recommended by IEEE
St-4-1995 Publication12
and IEC Publication 60-113
1998-11 has a front
time T2 D 2500 µsec š 20 per cent and half-time value T2 D 2500 µsec š 60
per cent. The general designation for a standard switching impulse is given as
250/2500 µsec. The front is counted from the actual beginning of the impulse
till the peak value is reached. Full characteristics of a standard switching test
surge have been defined in Chapter 2, Fig. 2.24.
It was shown in Chapter 5, section 5.9 that in non-uniform field gaps the
shape of both electrodes affects the formation and propagation of streamers and
directly influences the flashover values. This explains the different flashover
values observed for various insulating structures, especially under switching
surges. Much of the laboratory flashover data for large gaps under switching
surges have been obtained for rod-plane gaps. Subsequently, several attempts
have been made to relate data for other structures to rod-plane gap data.
Several investigators14,15
have shown that the positive 50 per cent switching
surge voltage of different structures in air in the range from 2 to 8 m follow
the expression
V50 D k 500 d0.6
kV 8.1
where d is the gap length in metres and k is gap factor relating to the electrodes
geometry. For rod-plane gaps the factor k is accepted as unity. Thus, the ‘gap
factor’ k represents a proportionality factor of the 50 per cent flashover voltage
of any gap geometry to that of a rod-plane gap for the same distance or
k D
V50
V50 rod-plane
8.2

Expression (8.1) applies to data obtained under the switching impulse of
constant time to crest. A more general expression which gives minimum
strength and applies to longer times to crest has been proposed by Gallet
and Leroy16
as follows:
V50 D
k3450
1 C
8
d
kV 8.3
where k and d have the same meaning as in expression (8.1).
In expression (8.2) only the function V50 rod-plane is influenced by the
switching impulse shape, while the gap factor k depends only on the gap
geometry and hence upon the field distribution in the gap. The parameters
influencing the gap factor k have been fully discussed by Schneider and
Weck.17
These authors have measured the gap factor k for different gap
geometries and spacings using a large three-dimensional electrolytic tank and
modelling scaled down gaps. Their data are included in Table 8.1. The corre-
sponding geometric configurations are shown in Fig. 8.10(a) to (f).
Half sphere
d
6m
60 mmf
60 mmf
0.2 mf
0.2 mf
60 mmf
60 mmf
60 mmf
(a) Rod plane (b) Rod structure (c) Vertical rod-rod
12.5 m
25 m
25 m
6 m 1m × 1.5m
1m × 1.5m
16m × 16 m
d
6 m
2 m
Half sphere
8 m
12.5 m
d
(d) Conductor plane (e) Conductor-crossarm end (f) Conductor-rod
f.p. 1 d1
2and
4 m
f.p. 2
25 m
d2
Figure 8.10 Configuration (gap factor)
Expressions (8.1) and (8.3) together with data presented in Table 8.1 can be
used in estimating required clearances in designing e.h.v. and u.h.v. structures.
Refinements to these expressions are being introduced as more data become
available.

Table 8.1 Geometric gap factor for various structures
Configuration Figure d D 2 m d D 3 m d D 4 m d D 6 m
k k k k
Rod-plane (a) 1 1 1 1
Rod-structure (b) 1.08 – 1.07 1.06
Rod-rod vertical
H D 2 m (c) 1.27 1.26 1.21 1.14
Conductor-plane (d) 1.08 – 1.14 1.15
Conductor-cross
arm end (e) 1.57 1.68 1.65 1.54
Conductor-2m rod (f) 1.47 – 1.40 1.25
Conductor-4m rod (f) 1.55 – 1.54 1.40
8.4 Laboratory high-voltage testing procedures and
statistical treatment of results
Practical high voltage insulation systems comprise various types of dielectrics,
e.g. gases, liquids, solids or any combination of these. The result, following
the application of a voltage stress to insulation, individually and also collec-
tively, is a discharge or withstand, and has a random nature. Hence the
parameters characterizing the behaviour of the insulation must be handled
statistically.
Test methods and procedures adopted for the determination of the parame-
ters characterizing insulation behaviour generally involve the repeated appli-
cation of dielectric stress and the appropriate evaluation of the results. The
aim of the statistical evaluation of the test methods is to establish proce-
dures for relevant interpretation of the parameters characterizing the insu-
lation behaviour and to determine confidence limits for the data obtained.
Hence a brief treatment of the statistical methods generally used will be
presented.
The documents addressing this issue are the IEEE Standard12
and the IEC
Publication 60-1 1989-11.13
8.4.1 Dielectric stress–voltage stress
A voltage stress when applied to a piece of insulation is completely defined
when the applied voltage Vt is known during the time of stress (tO, tM).
Trying to correlate the behaviour of the insulation to even a slightly different

value of Vt requires accurate knowledge of the physical processes occurring
inside the insulation.
8.4.2 Insulation characteristics
The main characteristic of interest of an insulation is the disruptive discharge
which may occur during the application of stress. However, because of the
randomness of the physical processes which lead to disruptive discharge, the
same stress applied several times in the same conditions may not always
cause disruptive discharge. Also, the discharge when it occurs may occur at
different times. In addition, the application of the stress, even if it does not
cause discharge, may result in a change of the insulation characteristics.
8.4.3 Randomness of the appearance of discharge
Randomness of the appearance of discharge can be modelled by considering a
large number of stress applications, a fraction p of which causes discharge, D,
and the remaining fraction q D 1 p being labelled as withstand, W. The
value of p depends on applied stress, S, with p D pS being the ‘probability
of discharge’ and it represents one of the characteristics of the insulation.
Recognizing that the time to discharge will also vary statistically, the prob-
ability of discharge will become a function of both the stress, S, and the
time t.
pV D pt, S 8.4
8.4.4 Types of insulation
Insulations are grouped broadly into:
(i) Self-restoring (gases) – no change produced by the application of stress
or by discharge, hence the same sample can be tested many times.
(ii) Non-self-restoring (liquids) – affected by discharge only, the same sample
can be used until discharge occurs.
(iii) Affected by applied stress, insulation experiences ageing and in testing it
becomes necessary to introduce a new parameter related to the sequential
application of stress.
8.4.5 Types of stress used in high-voltage testing
For design purposes it is sufficient to limit the knowledge of the insulation
characteristics to a few families of stresses which are a function of time Vt
e.g. switching surge of double exponential with time to crest T1 and to half

value T2 and the variable crest value V (see definitions in Chapter 2 for
lightning and switching surges). For testing purposes, the family is further
restricted by using fixed times T1 and T2, hence only one variable is left (V).
The same applies to both types of surges. The behaviour of the insulation
is then defined by the discharge probability as a function of crest voltage
p D pV.
The most commonly used distribution function is the normal (Gaussian)
distribution which has a particular shape (bell shape), plotted in Fig. 8.11.
The equation for the normal distribution density function is
pf D
1

p
2
efkfav2/22
8.5
f
Probability
density
function
p
(
f
)
α/2 α/2
zα /2 zα/2
1 − α
Figure 8.11 Gaussian (normal) distribution curve with confidence limits
where fk is the kth value of the variable, fav is the average value and is
the standard deviation. When the applied voltage, V, becomes the variable the
Gaussian distribution function used takes the form
pV D
1

p
2
eVV502
/22

8.6
where V50 is the voltage which leads to 50 per cent probability of discharge.

The knowledge of V50 and allows us to calculate the value of the proba-
bility pV for any applied voltage.
Also shown In Fig. 8.11 are the confidence limits A and B. The confidence in
our results when expressed in per cent is shown by the area (1 ˛) between the
limits ˛/2 and C˛/2. A more convenient form of the normal distribution
is the cumulative distribution function, the integral of eqn (8.12), which has
the form
PV D
1

p
2
1
1
eVV502/22
dx 8.7
A plot of this function is included in Fig. 8.12. When plotted on the probability
scale a straight line results as shown in Fig. 8.13. In this figure are plotted the
cumulative frequency PWS of withstand voltage, the PFO of flashover voltage
and the parameter z, explained below, versus the breakdown voltage of a 1-m
rod gap under positive switching impulse voltage in atmospheric air. We note
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Breakdown voltage (kV)
Cumulative
probability
P
(
V
)
V50
Figure 8.12 Gaussian cumulative distribution function

0.01
0.05
0.1
0.2
0.5
1
2
5
10
20
30
40
50
60
70
80
90
95
98
99
99.8
99.9
99.9
420 440 460 480 500 520 540 560 580 600
Voltage (kV)
0.01
0.05
0.1
0.2
0.5
10
20
30
40
50
60
70
80
90
95
98
99
99.9
99.8
99.9
1
2
5
− 3
− 2
− 1
0
+ 1
+ 2
+ 3
z
Pfo
Pws
Figure 8.13 Breakdown voltage distribution plotted on probability scale
that there are three vertical scales, two non-linear giving directly the PWS
(l.h.s.), the PFO (r.h.s.) and further to the right a linear scale given in units
of dimensionless deviation z. The parameter z is convenient for analysis of
normal distribution results. Equation (8.7) is rewritten in the form
Pz D
1

p
2
z
1
ez2
/2
dz 8.8

where
z D
V V50

As noted earlier the distribution of flashover of the gap is characterized by
two parameters:
(i) V50, called the critical flashover (CFO),
(ii) , called the standard deviation.
Both can be read directly from the best fit line drawn through the exper-
imentally determined points. Note, that CFO corresponds to z D 0 and is
given by the difference between two consecutive integers of z. In practice the
voltage range over which the probability of flashover is distributed is
CFO š 3 8.9
ž (CFO 3) is known as the statistical withstand voltage (SWV) and repre-
sents the point with flashover probability 0.13 per cent;
ž (CFO C 3) is known as the statistical flashover voltage (SFOV) and repre-
sents the point with flashover probability 99.87 per cent.
The SWV and SFOV are used in insulation coordination and will be
discussed later. For a complete description of insulation parameters, the
time to breakdown must also be considered. The times to breakdown are
represented by
Pt D
1

p
2
t
0
ett2/22
dt 8.10
where
t D mean time to breakdown,
D standard deviation.
An example of the distribution of times to breakdown is included in Fig. 8.14.
In this example the range
t š D t š z
is shown by a straight line but not at the extremities. Nevertheless the method
is often used to represent distribution of times to breakdown because of its
simplicity.
Another frequently used distribution function for representing breakdown
voltage probability is the Weibull function of the form:
PV D 1 0.5[1CVV50m
/n ln 2]
8.11

0.01
0.05
0.1
0.2
0.5
1
2
5
10
20
30
40
50
60
70
80
90
95
98
99
99.8
99.9
99.9 0.01
0.05
0.1
0.2
0.5
1
2
5
10
20
30
40
50
60
70
80
90
95
98
99
99.8
99.9
99.9
420 200 300 400 500
z = + 1
z
z = 0
z = −1
Time to breakdown (µs)
P
Figure 8.14 Distribution of times to breakdown
where
PV D the probability of flashover,
V D the applied voltage,
V50 D the applied voltage which gives 50 per cent probability flashover,
D the standard deviation.
In the Weibull function n is not known but it determines the voltage V50 n
below which no flashover occurs, or PV D 0 for V V50 n. For air n

lies in the range 3 n 4. The value 3 is usually used resulting in
m D
ln
ln 0.84
ln 0.5
ln
n 1
n
D 3.4 8.12
The adaptation of the Weibull function to normal distribution using the above
values for n and m gives PV D 0.5 for V D V50 and PV D 0.16 for V D
V50 . Both the Gaussian and the Weibull functions give the same results
in the range 0.01 PV 0.99.
8.4.6 Errors and confidence in results
In the determination of a parameter two types of error are present:
(i) error associated with the statistical nature of the phenomena and the
limited number of tests (εS),
(ii) error in the measurement (εM).
The statistical error is expressed by means of two confidence limits C per
cent. The total error is given by
εT D

ε2
M C ε2
S 8.13
The various IEC recommendations specify the permissible measurement accu-
racy as 3 per cent. Hence, a statistical error of, say, 2 per cent will increase
the total error by a factor of 1.2, while a statistical error of 1.5 per cent will
increase the total error by 1.1.
The outcome of a test procedure and the analysis of the results is usually
an average of a parameter z with C per cent confidence limits zA and zB (see
Fig. 8.11). For a normal distribution the probability density of a function for
a frequency of occurrence can be represented graphically in terms of area as
shown in Fig. 8.11 (1 ˛).
8.4.7 Laboratory test procedures
The test procedures applied to various types of insulation are described
in national and international standards as already mentioned before.12,13
Because the most frequently occurring overvoltages on electric systems
and apparatus originate in lightning and switching overvoltages, most
laboratory tests are conducted under standard lightning impulse voltages
and switching surge voltages. Three general testing methods have been
accepted:

1. Multi-level method.
2. Up and down method.
3. Extended up and down method.
1. Multi-level test method
In this method the procedure is:
ž choose several test voltage levels,
ž apply a pre-specified number of shots at each level (n),
ž count the number (x) of breakdowns at each voltage level,
ž plot pV (xj/n) against V (kV),
ž draw a line of best fit on a probability scale,
ž from the line determine V50 at z D 0 or PV D 50 per cent,
ž and at z D 1 or D V50% V16%
P(V)
kV
Figure 8.15 Probability of breakdown distribution using the multi-level
method
The recorded probability of breakdown, xj/n, is the number which resulted
in breakdown from the application of n shots at voltage Vj. When xj/n is
plotted against Vj on a linear probability paper a straight line is obtained as
shown in Fig. 8.15.
The advantage of this method is that it does not assume normality of
distribution. The disadvantage is that it is time consuming, i.e. many shots
are required.
This test method is generally preferred for research and live-line testing
(typically 100 shots per level, with 6–10 levels).
2. Up and down method
In this method a starting voltage (Vj) close to the anticipated flashover value is
selected. Then equally spaced voltage levels (V) above and below the starting

voltage are chosen. The first shot is applied at the voltage Vj. If breakdown
occurs the next shot is applied at Vj V. If the insulation withstands, the
next voltage is applied at Vj C V. The sequential procedure of testing is
illustrated in Fig. 8.16.
Figure 8.17 illustrates the sequence with nine shots applied to the insulation
under test. The IEC Standard for establishing V50 (50 per cent) withstand
voltage requires a minimum n D 20 voltage applications for self-restoring
Pick starting
voltage V1 = Vi
Pick ∆V
Apply one shot
at Vi
Breakdown Withstand
Apply next shot
at Vi − ∆V
Apply next shot
at Vi + ∆V
Figure 8.16 Schematics of the sequential up and down procedure
∆V
Vo
1
2
3
4
5
6
7
8
9
breakdown
withstand
Vo = lowest level at which a shot is applied
Figure 8.17 Example illustrating the application of nine shots in the
sequential up and down method. X D breakdown; O D withstand

insulation. To evaluate the V10 (10 per cent) withstand voltage for self-
restoring insulation with the up and down method with one impulse per group
also requires a minimum of n D 20 applications.
In practice the points, expressing the probability of withstand, are plotted
against the voltage Vj on a probability scale graph as was shown in Fig. 8.13.
The best straight line is then plotted using curve fitting techniques. The 50
per cent and 10 per cent discharge voltages are obtained directly from the
graph. This method has the advantage that it requires relatively few shots
and therefore is most frequently used by industry. The disadvantage is that it
assumes normality and is not very accurate in determining . Alternatively,
the V10 can be obtained from the V50 using the formula
V10 D V501 1.3z D V50 Ð 0.96 8.14
From the sequentially obtained readings (Fig. 8.17), the values of V50 and
can be also calculated analytically as follows.
In the example chosen (Fig. 8.17): total number of shots n D 9, total number
of breakdowns nb D 4, total number of withstands nw D 5, and lowest level
at which a shot is applied D V0.
In calculating V50% and ,
if nb nw then ni D number of withstands at level j
if nw nb then nj D number of breakdowns at level j
(always use the smaller of the two). The expressions are:
V50 D V0 C V

A
N
š
1
2

)

ni D nbi use negative sign
ni D nwi use positive sign
8.15
D 1.62AV

NB A2
N2
C 0.029

8.16
where
N D
k

iD0
niw or
k

iD0
nib
A D
k

iD0
iniw or
k

iD0
inib
B D
k

iD0
i2
niw or
k

iD0
i2
nib
with i referring to the voltage level, niw to the number of withstands and nib
the number of breakdowns at that level.

3. The extended up and down method
This method is also used in testing self-restoring insulation. It can be used to
determine discharge voltages corresponding to any probability p. A number
of impulses are applied at a certain voltage level. If none causes discharge,
the voltage is increased by a step V and the impulses are applied until at
least one causes breakdown, then the voltage is decreased. For an example of
the extended up and down method procedure see Fig. 8.18.
0
7
0
7
0
7
1
1
0
7
1
2
1
3
0
7
0
7
0
7
1
4
1
2
1
2
0
7
1
7
940
970
1000
1030
1060
Figure 8.18 Example of the extended up and down method
The number n is determined such that a series of n shots would have 50 per
cent probability of giving at least one flashover. The 50 per cent probability
of discharge is given by
0.5 D 1 1 pn
or
n D 0.5 D ln 1 p 8.17
from which p becomes a discrete value. The value n D 7 impulses per voltage
level is often used as it allows the determination of 10 per cent discharge
voltage without the necessity to use . Substituting n D 7 into eqn (8.17)
gives p D 0.094 or approximately 10 per cent.
The IEC switching withstand voltage is defined as 10 per cent withstand,
hence the extended up and down method has an advantage. Other advantages
include: discharge on test object is approximately 10 per cent the number of
applied impulses rather than 50 per cent as applicable to the up and down

method. Also the highest voltage applied is about V50 rather than V50 C 2. In
the up and down method the V10% may also be obtained from:
V10 D V50 1 1.13z D V50 Ð 0.96 8.18
In today’s power systems for voltages up to 245 kV insulation tests are still
limited to lightning impulses and the one-minute power frequency test. Above
300 kV, in addition to lightning impulses and the one-minute power frequency
tests, tests include the use of switching impulse voltages.
8.4.8 Standard test procedures
1. Proof of lightning impulse withstand level
For self-restoring insulation the test procedures commonly used for withstand
establishment are:
(i) 15 impulses of rated voltage and of each polarity are applied, up to two
disruptive discharges are permitted,
(ii) in the second procedure the 50 per cent flashover procedure using either
the up and down or extended up and down technique as described earlier.
From the up and down method the withstand voltage is obtained using
eqn (8.18). In tests on non-self-restoring insulation, three impulses are applied
at the rated withstand voltage level of a specified polarity. The insulation is
deemed to have withstood if no failure is observed.
2. Testing with switching impulses
These tests apply for equipment at voltages above 300 kV. The testing
procedure is similar to lightning impulses using 15 impulses. The tests are
carried out in dry conditions while outdoor equipment is tested under positive
switching impulses only. In some cases, when testing circuit isolators or circuit
breakers which may experience combined voltage stress (power frequency
and switching surge) biased tests using combined power frequency and surge
voltages are used. The acceptable insulating capability requires 90 per cent
withstand capability.
8.4.9 Testing with power frequency voltage
The standard practice requires the insulation to perform a one-minute test with
power frequency at a voltage specified in the standards. For indoor equipment,
the equipment is tested in dry conditions, while outdoor equipment is tested
under prescribed rain conditions for which IEC prescribes a precipitation rate
of 1–1.5 mm/min with resistivity of water.

8.4.10 Distribution of measured breakdown probabilities
(confidence in measured P(V))
We apply at a level Vi, n shots and obtain x breakdowns. The outcome is
breakdown or withstand, that is
n - shots
x - breakdown
therefore 1 shot
breakdown
withstand
hence, the distribution of PV is binomial around the expected value x/n. This
distribution depends on x, n and q (the breakdown around which Px, n, + is
centre d) as shown in Fig. 8.19.
Binom. distrib. of P(V )
around x/n
P(V )
= Pi
x
n
I
B D (kV)
Figure 8.19 Binomial distribution of PV around the expected value x/n
Pi is distributed around the value the point gives. For example, if we get:
V D 500 kV; P(FO) D 78 per cent, we do not really know that it is 78 per
cent but we do know that it is distributed around 78 per cent.
The binomial distribution of P around x/n is given by
Px, n, + D
n
x
+x
1 +nx
where
n
x
D
n!
x!n x!

+ D true value of the most likely outcome (value around which the distribution
is centred).
We do not know + but we can replace it with the expected value x/n as
was shown in Fig. 8.19.
Hence
Px, n, + D
n
x
+x
1 + 8.19
with x/n D 0.5, Px is symmetrical around x/n but at extremities x/n D 1
per cent and 99 per cent). Px is skewed as seen in Figs 8.20(a) and (b).
To obtain these distributions, leave x/n as the expected value and then vary
x to obtain the corresponding Px.
P(x)
P(x)
skewed skewed
symmetric
0.0 0.5 1.0
x
n
0.4 0.5 0.6 x
n
(a) P(x) symmetric around
(b) Skewed
x/n=0.5
Figure 8.20 Relation between Px and x/n: (a) Symmetric around
x/n D 0.5. (b) Skewed

Example
n D 5; x D 2, therefore x/n D 0.4: find Px D 4, using eqn (8.19)
Px D 4 D
5!
4!5 4!
0.44
1 0.454
D 7.7%
Using the eqn (8.19) we find that as n increases with x/n being constant we have
greater confidence in Px as seen in Figs 8.21(a) and (b), while for small values of n
the results are spread.
0
x =
(a)
(b)
P (x)
P (x)
50
n = 100
100
0
x = 5
n = 10
10
Area = 1
Area = 1
Figure 8.21 Effect of number of shots on Px distribution: (a) n D 100;
(b) n D 10
8.4.11 Confidence intervals in breakdown probability (in measured
values)
The normalized value of the variable x in the binomial distribution is
X n+
n+1 +
8.20
For a given level of confidence 1 ˛ , where ˛ is the level of significance as
shown in Fig. 8.11, the confidence interval at a measured point is given by
Za/2
x n+
n+1 +
CZa/2 8.21

The probability of breakdown with a confidence level 1 ˛ is given by
PV D
x
n
š Za/2

x
n

1
x
n

n
8.22
Using this expression it can be shown on the linear probability scale that the
confidence in the measured values of breakdown is at maximum at x/n D 0.5
and progressively decreases as the extreme values of breakdown probability
are approached. Za/2 is obtained from tables of statistics or for convenience
from the graph directly.
Example
n D 10; x D 5 for a confidence level of 90 per cent
˛ D 1 0.9 D 0.1
using statistical tables,25
we obtain for
a
2
D 0.05, Za/2 D 1.64
hence
PV D 1
2
š 1.64

1
2
1
2

10
The confidence limit at near PV D 50 per cent is much smaller than the
confidence limit for PV approaching 1 per cent or 99 per cent. The solution
is a non-linear distribution of n, that is we need to take many shots near the
limits and a few in the middle. Confidence expressed in terms of kV is more
convenient than confidence in probability as shown in Fig. 8.22.
P(V )
V50
∆V
∆Z
kV
Z
Z = 1
Z = 0
Z = −1
50%
23%
Figure 8.22 Confidence expressed in kV

Using the same example as before
23% PV50 77%
to determine z D z1 z2
z1 is determined from the value of P(0.50)
z2 is determined from the value of P(0.23)
from Fig. 8.22
for Fz D 50% D 0.5, z1 D 0.0
for Fz D 23% D 0.77, z2 D 0.74
therefore
z D z1 z2
D 0 0.74
D 0.74
the standard deviation is the run from z D 0 to z D 1 therefore
rise D slope
D z/V
D 1/
and therefore V D z
Thus the confidence in V is
V D V š V
D V š z 8.23
confidence in
V50 D V50 š 0.74 8.24
8.5 Weighting of the measured breakdown probabilities
Weights can be assigned to various data points to the measured breakdown
probability and the number of impulses applied at each level.
8.5.1 Fitting of the best fit normal distribution
On probability paper the normal distribution best characterizing the data points
will appear as the best fit straight line. An example of this is shown in
Fig. 8.23.

450 500 550 600 650
V1
Breakdown voltage (kV)
ZI
zi
Z(v)
z = +3
z = +2
z = +1
z = +0
z = −1
z = −2
z = −3
Figure 8.23 Best fit normal distribution drawn through measured flashover
probability points
In order to obtain this best fit straight line, it is necessary to minimize the
deviation of the data points around the line. The root mean square deviation
for the case shown in Fig. 8.23 is given by

1
m

iD1
wi zi .i2 8.25
where xi is the value of the measured breakdown probability on the probit
scale at the voltage level Vi, xi is the probit scale value of the breakdown
probability as given by the best fit straight line for the same voltage level, and
w1 is the weighting coefficient assigned to the measurement, xi. The expression
given in eqn (8.25) is in terms of the dimensionless deviation z. This can be
rewritten using
zi D
Vi V50

8.26
to obtain
1
m

iD1
wi
Vi V50

.i
2
8.27

Minimizing this expression is equivalent to minimizing

iD1
wi
Vi V50

.i
2
8.28
The minimum value of the above expression occurs when the quantity

iD1
wi Vi V50 .i2
8.29
is at its minimum. The best fit straight line which is in fact the normal distri-
bution best representing the breakdown probability can now be obtained by
setting
∂
∂V50

iD1
wi Vi V50 .i2
D 0 8.30
and
∂
∂

iD1
wi Vi V50 .i2
D 0 8.31
and solving for V50 and . These values are found by carrying out the partial
differentiation of eqns (8.30) and (8.31). This gives the following two simul-
taneous equations

iD1
wi vi

iD1
wiV50

iD1
wi.i D 0 8.32
and

iD1
wiVi.i V50

iD1
wi.i

iD1
wi.2
i D 0 8.33
which can be solved to obtain
V50 D

iD1
wiVi

iD1
wi.i

iD1
wi
8.34
and
D

iD1
wiVi

iD1
wi.i

iD1
wi

iD1
wiVi.i

iD1
wi.i
2

iD1
wi

iD1
wi.2
i
8.35
Thus values for V50 and are obtained.

8.6 Insulation coordination
Insulation coordination is the correlation of insulation of electrical equip-
ment with the characteristics of protective devices such that the insulation
is protected from excessive overvoltages. In a substation, for example, the
insulation of transformers, circuit breakers, bus supports, etc., should have
insulation strength in excess of the voltage provided by protective devices.
Electric systems insulation designers have two options available to them:
(i) choose insulation levels for components that would withstand all kinds of
overvoltages, (ii) consider and devise protective devices that could be installed
at the sensitive points in the system that would limit overvoltages there. The
first alternative is unacceptable especially for e.h.v. and u.h.v. operating levels
because of the excessive insulation required. Hence, there has been great incen-
tive to develop and use protective devices. The actual relationship between the
insulation levels and protective levels is a question of economics. Conventional
methods of insulation coordination provide a margin of protection between
electrical stress and electrical strength based on predicted maximum over-
voltage and minimum strength, the maximum strength being allowed by the
protective devices.
8.6.1 Insulation level
‘Insulation level’ is defined by the values of test voltages which the insulation
of equipment under test must be able to withstand.
In the earlier days of electric power, insulation levels commonly used were
established on the basis of experience gained by utilities. As laboratory tech-
niques improved, so that different laboratories were in closer agreement on
test results, an international joint committee, the Nema-Nela Committee on
Insulation Coordination, was formed and was charged with the task of estab-
lishing insulation strength of all classes of equipment and to establish levels for
various voltage classification. In 1941 a detailed document18
was published
giving basic insulation levels for all equipment in operation at that time.
The presented tests included standard impulse voltages and one-minute power
frequency tests.
In today’s systems for voltages up to 245 kV the tests are still limited
to lightning impulses and one-minute power frequency tests, see section 8.3.
Above 300 kV, in addition to lightning impulse and the one-minute power
frequency tests, tests include the use of switching impulse voltages. Tables 8.2
and 8.3 list the standardized test voltages for 245 kV and above ½300 kV
respectively, suggested by IEC for testing equipment. These tables are based
on a 1992 draft of the IEC document on insulation coordination.

Table 8.2 Standard insulation levels for Range I (1 kV Um 245 kV)
(From IEC document 28 CO 58, 1992, Insulation coordination Part 1:
definitions, principles and rules)
Highest voltage Standard Standard
for equipment power frequency lightning impulse
Um short-duration withstand voltage
kV withstand voltage kV
(r.m.s. value) kV (peak value)
(r.m.s. value)
3.6 10 20
40
7.2 20 40
60
12 28 60
75
95
17.5 38 75
95
24 50 95
125
145
36 70 145
170
52 95 250
72.5 140 325
123 (185) 450
230 550
145 (185) (450)
230 550
275 650
170 (230) (550)
257 650
325 750
245 (275) (650)
(325) (750)
360 850
395 950
460 1050

Table 8.3 Standard insulation levels for Range II (Um 245 kV) (From
IEC document 28 CO 58, 1992, Insulation coordination Part 1: definitions,
principles and rules)
Highest Longitudinal Standard Phase-to-phase Standard
voltage for insulation lightning (ratio to the lightning
equipment C kV impulse phase-to-earth impulse
Um kV (peak value) withstand voltage peak withstand
(r.m.s. value) Phase-to-earth value) voltage kV
kV (peak value)
(peak value)
300 750 750 1.50 850
950
750 850 1.50 950
1050
362 850 850 1.50 950
1050
850 950 1.50 1050
1175
420 850 850 1.60 1050
1175
950 950 1.50 1175
1300
950 1050 1.50 1300
1425
525 950 950 1.70 1175
1300
950 1050 1.60 1300
1425
950 1175 1.50 1425
1550
765 1175 1300 1.70 1675
1800
1175 1425 1.70 1800
1950
1175 1550 1.60 1950
2100
CValue of the Impulse component of the relevant combined test.
Note: The introduction of Um D 550 kV (instead of 525 kV), 800 kV (instead of 765 kV), 1200 kV, of a value between 765 kV
and the associated standard withstand voltages, are under consideration.

8.6.2 Statistical approach to insulation coordination
In the early days insulation levels for lightning surges were determined by
evaluating the 50 per cent flashover values (BIL) for all insulations and
providing a sufficiently high withstand level that all insulations would with-
stand. For those values a volt–time characteristic was constructed. Similarly
the protection levels provided by protective devices were determined. The two
volt–time characteristics are shown in Fig. 8.24. The upper curve represents
the common BIL for all insulations present, while the lower represents the
protective voltage level provided by the protective devices. The difference
between the two curves provides the safety margin for the insulation system.
Thus the
Protection ratio D
Max. voltage it permits
Max. surge voltage equipment withstands
8.36
kV
A
B
time
A: protecting device
B: device to be protected
safety margin
Figure 8.24 Coordination of BILs and protection levels (classical approach)
This approach is difficult to apply at e.h.v. and u.h.v. levels, particularly for
external insulations.
Present-day practices of insulation coordination rely on a statistical approach
which relates directly the electrical stress and the electrical strength.11
This
approach requires a knowledge of the distribution of both the anticipated
stresses and the electrical strengths.

The statistical nature of overvoltages, in particular switching overvoltages,
makes it necessary to compute a large number of overvoltages in order to
determine with some degree of confidence the statistical overvoltages on a
system. The e.h.v. and u.h.v. systems employ a number of non-linear elements,
but with today’s availability of digital computers the distribution of overvolt-
ages can be calculated. A more practical approach to determine the required
probability distributions of a system’s overvoltages employs a comprehen-
sive systems simulator, the older types using analogue units, while the newer
employ real time digital simulators (RTDS).24
For the purpose of coordinating the electrical stresses with electrical
strengths it is convenient to represent the overvoltage distribution in
the form of probability density function (Gaussian distribution curve as
shown in Fig. 8.11) and the insulation breakdown probability by the
cumulative distribution function (shown in Fig. 8.12). The knowledge of these
distributions enables us to determine the ‘risk of failure’. As an example, let
us consider a case of a spark gap for which the two characteristics in Figs 8.11
and 8.12 apply and plot these as shown in Fig. 8.25.
Overvoltage
distribution
1
2
Insulation
break down
probability
pb(Vk)
pb(V )
p0(Vk)du
p0 (u )
R
A B V
Vk
Risk of
failure
Figure 8.25 Method of describing the risk of failure. 1. Overvoltage
distribution–Gaussian function. 2. Insulation breakdown
probability–cumulative distribution)
If Va is the average value of overvoltage, Vk is the kth value of over-
voltage, the probability of occurrence of overvoltage is p0Vk du, whereas
the probability of breakdown is PbVk or the probability that the gap will
break down at an overvoltage Vk is PbVkp0Vk du. For the total voltage

range we obtain for the total probability of failure or ‘risk of failure’
R D
1
0
PbVkp0Vk du. 8.37
The risk of failure will thus be given by the shaded area under the curve R.
In engineering practice it would become uneconomical to use the complete
distribution functions for the occurrence of overvoltage and for the withstand
of insulation and a compromise solution is accepted as shown in Figs 8.26(a)
and (b) for guidance. Curve (a) represents probability of occurrence of over-
voltages of such amplitude Vs that only 2 per cent (shaded area) has a
chance to cause breakdown. VS is known as the ‘statistical overvoltage’. In
Fig. 8.26(b) the voltage Vw is so low that in 90 per cent of applied impulses,
breakdown does not occur and such voltage is known as the ‘statistical with-
stand voltage’ Vw.
P0(V )
Pb(V )
Reference probability
2%
Vs
Statistical (max) overvoltage V Statistical withstand voltage V
1.0
0.1 Vw
Reference
probability
90%
(a) (b)
Figure 8.26 Reference probabilities for overvoltage and for insulation
withstand strength
In addition to the parameters statistical overvoltage ‘VS’ and the statis-
tical withstand voltage ‘VW’ we may introduce the concept of statistical
safety factor 4. This parameter becomes readily understood by inspecting
Figs 8.27(a) to (c) in which the functions PbV and p0Vk are plotted for
three different cases of insulation strength but keeping the distribution of
overvoltage occurrence the same. The density function p0Vk is the same in
(a) to (c) and the cumulative function giving the yet undetermined withstand
voltage is gradually shifted along the V-axis towards high values of V. This
corresponds to increasing the insulation strength by either using thicker insu-
lation or material of higher insulation strength. As a result of the relative shift
of the two curves [PbV and p0Vk] the ratio of the values Vw/Vs will vary.
This ratio is known as the statistical safety factor or
Vw
Vs
D 4 8.37

P0 (V )
P0 (V )
Pb (V )
(a) (b)
(c) (d)
Vs = Vw Vs Vw
Vs Vw
Vw = 1.2Vs
Vw = 1.4Vs
V
R1 = ∫ Pb (Vs), P0(Vk)
0
R2
R3
R1
R
R2
R3
Vs
Vw
γ =
γ1
γ2
γ3
γ
Figure 8.27 The statistical safety factor and its relation to the risk of
failure (R)
In the same figure (d) is plotted the relation of this parameter to the ‘risk of
failure’. It is clear that increasing the statistical safety factor (4) will reduce
the risk of failure (R), but at the same time will cause an increase in insulation
costs. The above treatment applies to self-restoring insulations. In the case of
non-self-restoring insulations the electrical withstand is expressed in terms of
actual breakdown values. The statistical approach to insulation, presented here,
leads to withstand voltages (i.e. probability of breakdown is very small), thus
giving us a method for establishing the ‘insulation level’.
8.6.3 Correlation between insulation and protection levels
The ‘protection level’ provided by (say) arresters is established in a similar
manner to the ‘insulation level’; the basic difference is that the insulation
of protective devices (arresters) must not withstand the applied voltage.
The concept of correlation between insulation and protection levels can be
readily understood by considering a simple example of an insulator string
being protected by a spark gap, the spark gap (of lower breakdown strength)
protecting the insulator string. Let us assume that both gaps are subjected to
the same overvoltage represented by the probability density function p0V,
Fig. 8.28. The probability distribution curves for the spark gap and the
insulator string are presented by PgV and PiV respectively in Fig. 8.28.

Gap
breakdown
probability
Margin
of safety
Risk of failure of protective gap
Risk of failure R
Risk of failure of protected gap
1.0
Gap
Pg (V )
p0 (V )
Pi (V )
Overvoltage
Insulation
0.5
1.0
0.5
0 V
Pp(V )
Figure 8.28 Distribution functions of breakdown voltages for protective gap
and protected insulation both subjected to an overvoltage p0 V
The statistical electrical withstand strength of the insulator string is given by
a curve identical to Fig. 8.26. The probability of breakdown of this insulation
remains in the area R which gives ‘risk of failure’. Since the string is protected
by a spark gap of withstand probability PgV, the probability that the gap
will operate (its risk of failure) is obtained from integrating the product
PgVp0V dV. In Fig. 8.28 this probability is denoted (qualitatively) by
PPV. As is seen the probability is much higher than the probability of
insulation damage or failure R. In the same figure is shown the traditional
margin of safety corresponding to the voltage difference between the 50 per
cent flashover values of the protecting gap and the protected gap.
For overvoltages of the highest amplitude (extreme right of Fig. 8.28) the
probability curves of insulation failure and that of protective spark gap break-
down overlap. In reality such cases will not arise. Figure 8.28 is simplified
in that it contains information pertaining to the amplitude of the overvoltage,
and ignores the effect of time of voltage application on the breakdown of both
the protective gap and the insulation. In practice, the protective gap will in
general break down before the insulation and will cause a reduction (to a safe
limit) in overvoltage reaching the protected insulation.

8.7 Modern power systems protection devices
8.7.1 MOA – metal oxide arresters
The development of MOA (metal oxide arresters) represented a breakthrough
in overvoltage protection devices. It became possible to design arresters
without using gaps which were indispensable in the conventional lightning
arresters, which utilized non-linear resistors made of silicon Carbide (SiC)
and spark gaps. Figure 8.29 shows a block diagram of the valve arrangements
in the two types of arrester.
In (a) the elements and the spark gaps are connected in series. In (b) the
elements are stacked on top of each other without the need for spark gaps.
H.V.
SiC elements
Spark gaps
H.V.
MO elements
(a)
(b)
Figure 8.29 Block diagram of valve arrangements in (a) SiC, (b) MOA

An ideal lightning arrester should: (i) conduct electric current at a certain
voltage above the rated voltage; (ii) hold the voltage with little change for
the duration of overvoltage; and (iii) substantially cease conduction at very
nearly the same voltage at which conduction started.25
In Fig. 8.29(a) the
three functions are performed by the combination of the series spark gaps and
the SiC elements. In the (b) case the metal oxide valve elements perform all
three functions because of their superior non-linear resistivity.
The volt–current characteristics for the two types of arresters can be repre-
sented by the following equations:
For SiC valves: I D kVa
where a D 4–6 8.38
For ZnO valves: I D kVb
where k D const, b D 25–30 8.39
400
200
100
80
10
10−7
10−6
10−5
10−4
10−3
10−2
10−1
100
101
102
103
20
40
60
Volts/mm
Amps/cm2
25°C
125°C
ZnO
SiC
50
75
100
Figure 8.30 Normalized volt–ampere characteristic of zinc oxide and
silicon carbide valve elements
Typical volt–current characteristics for the valve elements used in the two
types of arresters are plotted in Fig. 8.30. The metal oxide varistors, which
consist of compacted and sintered granules of zinc oxide with a small amount
of other carefully selected metal oxide additives (Bi2O3, MnO, Cr2O3, Sb2O3)
to improve the V–I non-linearity, were first introduced in the electronics
industry in 1968 by Matsushita Electric Industrial Co. in Japan. The ZnO grains
have a low resistivity, while the additives (oxides) which form the boundaries
between the grains provide high resistance. The two are strongly bonded when
sintered at high temperature. Figure 8.31 shows the microstructure of a metal
oxide varistor.
Subsequently these were developed for use as a substitute for SiC valve
blocks in surge arresters by General Electric Co.26
From Fig. 8.30 it can be
seen that for a change in current from 103
to 102
A/cm2
, the voltage increase

ZnO grain
Boundary layer
New fine-grain
(b) Microstructure of conventional element
(a) Microstructure of new element
ZnO grain
Boundary layer
Spinel
Spinel
5 µm
5 µm
Figure 8.31 Cross-section, showing the microstructure of ZnO elements.
(a) Latest type (advanced). (b) Older conventional type (courtesy of
Mitsubishi Elec. Co.)
for ZnO is only 56 per cent.25
With such a high degree of non-linearity it is
entirely feasible to use these elements without series gaps in an arrester with
a current of only tens of µA at operating voltage.
The elements are manufactured in the form of discs of several sizes. The disc
voltage rating has been increasing with the improvement in the manufacturing
technology and the microstructure composition, e.g. Fig. 8.32 compares the
V–I characteristics of an older type ZnO element with that of a new type,
both developed by Mitsubishi.27
It is noted that the voltage rating per unit valve has been approximately
doubled. For higher voltage and current ratings the discs are arranged in series
and in parallel. Figure 8.33 shows a schematic structure of a three-column
arrangement of the arrester valves in an advanced MOA compact structure
manufactured by Mitsubishi.27

Current density (A/cm2)
Conventional
Varistor voltage
Voltage
(p.u.)
10−6 10−5 10−4 10−3 10−2 10−1 101 102 103
100
0.5
1
2
3
4
X
New
Figure 8.32 Comparison of volt–current characteristics of (a) advanced
MOA with (b) that of an older type MOA (courtesy of Mitsubishi Co.)
Insulator
Conductor
ZnO elements
Figure 8.33 Schematic structure of a three column series arrangement of
elements in advanced MOAs

In Fig. 8.34 is shown part of an assembled advanced 500 kV MOA. The
percentages indicate the reduction in size by replacing the older type MOA with
the advanced MOA elements whose V–I characteristics are shown in Fig. 8.32.
Advanced
MOA
Insulating
spacer
Shield
ZnO
elements
60%
92%
Conventional
MOA
500 kV
Figure 8.34 Part of an assembled 500 kV MOA Arrester. (courtesy of
Mitsubishi Co.)
In this construction the individual surge arresters are interconnected
by means of corona-free stress distributors. The modular design and the
lightweight construction allow easy on-site erection and in the event of any
units failing the individual unit may be readily replaced.
The advantages of the polymeric-housed arresters over their porcelain-
housed equivalents are several and include:
ž No risk to personnel or adjacent equipment during fault current operation.
ž Simple light modular assembly – no need for lifting equipment.

ž Simple installation.
ž High-strength construction eliminates accidental damage during transport.
ž The use of EPDM and/or silicon rubber reduces pollution flashover
problems.
Thus the introduction of ZnO arresters and their general acceptance by
utilities since late 1980s, and in 1990s in protecting high voltage substations,
has greatly reduced power systems protection problems.
In the earlier construction the valve elements were mounted within a ceramic
housing. The metal oxide elements were surrounded by a gaseous medium
and the end fittings were generally sealed with rubber O-rings. With time
in service, especially in hostile environments, the seals tended to deterio-
Terminal
block
Thermal
barrier
Polymeric
housing
Metal oxide
elements
Heat sink
Weather seal
Cap
Stainless steel
terminal
assembly
Figure 8.35 Cross-section of a polymer-housed arrester (courtesy of
Bowthorpe EMP)

rate allowing the ingress of moisture. In the 1980s polymeric-housed surge
arresters were developed. Bowthorpe EMP (UK)28
manufactures a complete
range of polymeric-housed arresters extending from distribution to heavy duty
station arresters for voltages up to 400 kV. In their design the surface of the
metal oxide elements column is bonded homogeneously with glass fibre rein-
forced resin. This construction is void free, gives the unit a high mechanical
strength, and provides a uniform dielectric at the surface of the metal oxide
column. The housing material is a polymer (EPDM)–Ethylene propylene diene
monomer–which is a hydrocarbon rubber, resistant to tracking and is partic-
ularly suitable for application in regions where pollution causes a problem.
A cross-section detailing the major features of a polymeric-housed arrester is
given in Fig. 8.35.
The ZnO elements are separated by aluminium blocks which serve as
heat sinks. To achieve higher voltages and higher current ratings a modular
construction with the individual units mounted in series–parallel arrangement
is shown in Fig. 8.36.
Figure 8.36 Construction of a series–parallel polymeric-housed arrester.
(courtesy of Bowthorpe EMP)

References
1. Hydro-Quebec Symposium on Extra-High Voltage Alternating Current, Oct. 1973.
2. L.V. Bewley. Travelling Waves on Transmission Systems. Dover Publications, New York NY,
1963.
3. Electrical Transmission and Distribution. Westinghouse Electric Co., Pittsburgh, Penn., 1964.
4. W.W. Lewis. The Protection of Transmission Systems against Lightning. Dover Publications
Inc., New York, 1965.
5. R.H. Golde (ed.). Lightning, Vol. I: Physics of Lightning; Vol. II: Lightning Protection.
Academic Press, London/New York/San Francisco, 1977.
6. B.F.J. Schonland. Progressive lightning, IV. The discharge mechanism. Proc. Roy. Soc.
Series A, 164 (1938), p. 132.
7. E.H.V. Transmission Lines. Reference Book General Electric Company, Edison Electrical
Institute, New York, 1968, p. 288.
8. AIEE Committee Report. Method for Estimating Lightning Performance of Transmission
Lines. Trans. AIEE Part III, 69 (1950), p. 1187.
9. R.H. Golde. A plain man’s guide to lightning protection. Electronics and Power, March
1969.
10. T. Udo. Sparkover characteristics of long gaps and insulator strings. Trans. IEEE PAS 83
(1964), p. 471.
11. W. Diesendorf. Insulation Coordination in High Voltage Electric Power Systems. Butter-
worths, 1974.
12. IEEE St-4-1995, Standard Techniques for High Voltage Testing. IEEE Inc. publication.
13. International Electrotechnical Commission, IEC Intern. Standard 61-1, 1989-11, High Voltage
Test Techniques, Pt. 1: General Definitions and Test Requirements.
14. L. Paris. Influence of air gap characteristics on line to ground switching surge strength. Trans.
IEEE PAS 86 (1967), p. 936.
15. L. Paris and R. Cortina. Switching surge characteristics of large air gaps and long insulator
strings. Trans. IEEE PAS 87 (1968), p. 947.
16. G. Gallet and G. Leroy. Expression for switching impulse strength suggesting the highest
permissible voltage for AC systems. IEEE-Power, Summer Power Meeting, 1973.
17. K.H. Schneider and K.H. Week. Electra No. 35 (1974), p. 25.
18. Standard Impulse, Basic Insulation Levels. A Report of the Joint Committee on Coordination
of Insulation AIEE, EEI and NEMA. EEI Publication No. H-9, NEMA Publication #109,
AIEE Transactions, 1941.
19. Dielectric Stresses and Coordination of Insulation. Brown Boveri Publication No. CH-A0500
20E No. 4, 1972.
20. K. Berger. Method und Resultate der Blitzforschung auf dem Monte San Salvatore bel
Lugano in den Jahren 1963–1971.
21. W. Büsch. The effect of humidity on the dielectric strength of long air gaps of UHV-
configurations subjected to positive impulses. Ph.D. thesis, ETH, Zurich, 1982 (see also:
W. Büsch. Trans. IEEE PAS 97 (1978), pp. 2086–2093).
22. J. Kuffel, R.G. van Heswijk and J. Reichman. Atmospheric influences on the switching
impulse performance of 1-m gaps. Trans. IEEE PAS 102(7), July 1983.
23. International Electrotechnical Commission. IEC Intern. Standard IEC 71-2 1996 ‘Insulation
Coordination Part 2. Application Guide’.
24. R. Kuffel, J. Giesbrecht, T. Maguire, R.P. Wierckx and P. McLaren. ‘RTDS’ A Fully Digital
Power System Simulator Operating in Real Time, Proceedings of the First International
Conference on Digital Power System Simulators, pp. 19–24, April 1995.
25. E.C. Sakshang et al. A new concept in Station Design. IEEE Trans. PAS, Vol. 96, No. 2,
1977, pp. 647–656.

26. A. Sweetana et al. Design, development and testing of 1,200 kV and 550 kV gapless surge
arresters. IEEE Trans. PAS Vol. 101, No. 7, 1982, pp. 2319–2327.
27. Private communication.
28. Bowthorpe EMP Catalogue 030 1992.

Chapter 9
Design and testing of external insulation
Conventional air-insulated substations represent a large majority of installed
high-voltage substations. They range in voltage from distribution levels to
765 kV systems. The external insulation generally utilized in these outdoor
substations takes the form of insulators (posts, suspension and pin types) and
housings. These types of apparatus are generally broken into classifications
based on manufacturing and materials.
The two broad categories of insulators are ceramic and polymeric. Ceramic
insulators include those constructed from porcelain and glass. Polymeric insu-
lators, often referred to as non-ceramic insulators (NCIs), are made up of
various designs, usually incorporating a fibre glass core encapsulated in rubber
housings which afford protection to the fibre glass core from electrical stresses
and moisture. Both categories are described in more detail in section 9.5.
In addition to the general review of the design and use of outdoor insulation,
this chapter also presents a synopsis of the physical mechanism of insulator
contamination flashover. Following this, methods used in the evaluation and
testing of insulators operating in contaminated environments are discussed,
from the laboratory and in-service perspectives. Finally, methods of mitigating
contamination-related flashovers of in-service insulators are reviewed.
9.1 Operation in a contaminated environment
The environment in which an insulator is installed can have a significant impact
on the unit’s performance. When insulators are situated in areas where they
are exposed to contamination, their performance can deteriorate significantly.
This is likely the single greatest challenge encountered in the design and oper-
ation of substation insulation. In order to provide some insight into this topic,
the processes of contamination accumulation and the flashover mechanism of
polluted insulators will be presented, together with some remedies to control
the problem.
Although the problem of contamination of insulators has been recognized
for over 50 years most studies have been carried out within the last three
decades, i.e. since the advent of compact transmission systems with reduced
insulation. Contamination flashover has become the most important and often

the limiting factor in the design of high-voltage outdoor insulation and hence
became a subject of extensive studies. A number of empirical and theoretical
models for the flashover mechanisms have been proposed.1,2,3,4,5,6,7
Contamination flashover (FO) requires both soluble salts and moisture. To
a large extent differences in insulator behaviour arise due to the variety of
environments and complex wetting mechanisms. Hence the performance of
insulators in contaminated environments is best assessed by tests done under
natural conditions.
However, for practical reasons artificial tests which can be performed in h.v.
laboratories are required (see section 9.3). Artificial tests assess insulators on a
relative basis, because they cannot account for the effect of the characteristics
of shape on collection of contamination and self-cleaning properties.
The various sources of pollution that affect power system insulation include:
ž Sea salt – salt from sea water is carried by winds up to 15–30 km inland
or further.
ž Industrial products which contain soluble salts.
ž Road salts.
ž Bird excrement.
ž Desert sands.
9.2 Flashover mechanism of polluted insulators under
a.c. and d.c.
Insulators in service become covered with a layer of pollution. When the
surface is dry the contaminants are non-conducting; however, when the insu-
lator surface is wetted by light rain, fog, or mist, the pollution layer becomes
conducting with the following sequence of events:
ž conducting layer build-up,
ž dry band formation,
ž partial arcing,
ž arc elongation,
ž eventual arc spanning the whole insulator followed by flashover.
The pollution layer in general is not uniform. When conduction starts, the
currents are in the order of several milliamps, resulting in heating of the
electrolyte solution on the insulator surface. The leakage current begins to
dry the pollution layer and the resistivity of the layer rises in certain areas.
This leads to dry band formation, usually in areas where the current density
is highest. The dry band supports most of the applied voltage. The air gap
flashes over, with the arc spanning the dry band gap which is in series with the
wet portion of the insulator. The arc may extinguish at current zero and the

Design and testing of external insulation 511
insulator may return to working conditions. Dry band formation and rewetting
may continue for many hours.
The current coinciding with the occurrence of dry band breakdown is in
the order of 250 mA. The current at this stage is in surges, and the voltage is
unaffected.
9.2.1 Model for flashover of polluted insulators(1,2,6)
Let us assume a uniform pollution layer with resistance r k/mm as shown
in Fig. 9.1. When the arc is burning in series with the pollution layer, the
voltage across the insulator with an arc partially bridging the insulator will be
given by:
V D VarcI, x C IL xr 9.1
where the function VarcI, x relates the arc voltage to the current I and the
arc length x. In general, for a given resistance r the curve relating V to x/L
has the form shown in Fig. 9.2.
Arc
I
X L-X
V (kV)
Figure 9.1 Model of a single arc developing on a polluted surface (uniform
pollution layer r, k/mm)
For an applied voltage Va, x/L may have values no greater than x/La.The
curve has a maximum critical voltage Vc, and for voltages equal to or greater
than Vc, x/L may have values up to unity. When the applied voltage Va is less
than Vc, x/L cannot increase to unity and flashover cannot occur. Numerous
empirical relations have been1,4,5
proposed to solve eqn (9.1). For example,
for vertical 33 kV and above
Vc D 0.067r1/3
L2/3
a L1/3
s kV(r.m.s.) 9.2
where La is the minimum arc length (mm) to the bridge insulator and Ls is
the leakage path (mm) on the insulator surface.

V
Vc
Va
1.0
(X/L)crit
X/La
0
X/L
Figure 9.2 Voltage versus x/L for an arc in series with a pollution layer of
a fixed resistance per unit length
For a cap and pin type insulator string,
Vc D 0.671/3
Nr1/3
L
2/3
S 1/3
kV(r.m.s) 9.3
where is a constant and N is the number of insulators in the string.
For a given insulator type
Ls / La
or
La D kLs with k D constant
Therefore eqn (9.3) can be written as
VC D 0.0672/3
kr1/3
Ls 9.4
9.3 Measurements and tests
Assessments of the performance of insulators is based on laboratory and field
tests which include:
(i) measurement of insulator dimensions;
(ii) measurement of pollution severity;
(iii) tests on polluted insulators.

9.3.1 Measurement of insulator dimensions
In order to effectively assess the degree of contamination present on an insu-
lator surface the dimensions of the insulator must be taken into account. The
relevant dimensions include the leakage path Ls, and the surface gradient
expressed in kV/Ls (Ls in mm). For a definition of Ls see Fig. 9.3. The inden-
tations X and Y are assumed filled with a conducting material.
The insulator surface area is required to determine the equivalent salt
density deposit (ESDD) in mg/cm2
(usually mean area based on maximum
and minimum areas).
B
Y
X
A
Ls
Figure 9.3 Evaluation of insulator dimensions for a typical shape.
Ls D leakage path length.
The parameter relating resistance R of a polluted insulator in air to the
surface resistivity is known as the form factor F, defined as:
F D
R

9.5
F D
LS
0
dLS
2a
9.6
where a is the radius corresponding to the path element dLs.

In the laboratory, the resistance R can be measured with a low-voltage
bridge and the average resistivity is determined from eqns (9.5) and (9.6).
The average value of r is obtained from:
r D
R
LS
9.7
Therefore can be related to the minimum flashover voltage.
An example of the process of evaluating the profile for a simple insulator
is included in Fig. 9.4.
9.3.2 Measurement of pollution severity
In general, the severity of the contamination present on the surface of insulators
is classified according to ESDD. This information is used to designate various
severity zones based on the characteristics of pollution present in the service
environment where the insulators are used. The classifications are shown in
Table 9.1. They are used as a guideline for choosing the leakage distance as
a function of system voltage for a particular environment. It should be noted
that the data in the table is meant for ceramic insulators, but in the absence
of parallel information for composite insulators, it is also often used as a
guideline for application of non-ceramic insulators.
Determination of equivalent salt (NaCl) deposit density
When an insulator is recovered from service, swabs are taken from the surface
using a pre-cleaned cloth according to the following procedure:
ž The conductivity of distilled water with the clean cloth submersed is
measured.
ž Swabs are taken from the top and bottom surfaces of the insulator indepen-
dently.
ž The cloth is rinsed and the conductivity is remeasured and the increase is
noted.
ž The solution is transferred into a standard volumetric flask.
ž The solution is diluted with distilled water to a volume of 0.5 or 1 litre.
ž The conductivity of the diluted solution is measured at two different temper-
atures and the conductivity corresponding to a temperature of 20°C is
calculated through interpolation.
ž The ESDD in mg/cm3
is calculated using the following expression[15]
ESDD D
0.42vol. in ml
Area in cm2
20°C1.039

Surface area 674 cm2
Form factor 3.09
Leakage 179 mm
Radius,
mm
Length along insulator, mm
80
60
40
20
0
80
60
40
20
0
674.3210559
3.091535268
179.301749
10732.16199
3.091535268
179.301749
Sums
44
80
0.44
0.000227273
0.01
44
0
44
79.99
219.56
0.113409091
4.99
44
0
44
75
367.0817348
0.166175525
7.810249676
47
−6
50
70
550
0.181818182
10
55
−10
60
70
791.4395113
0.184473984
12.08304597
65.5
−11
71
75
640.8732324
0.115467453
8.602325267
74.5
−7
78
70
791.945863
0.142686521
10.63014581
74.5
7
71
62
854.0155151
0.199059615
13.03840481
65.5
11
60
55
528.3029434
0.168463949
9.433981132
56
8
52
50
114.0394669
0.04384447
2.236067977
51
2
50
49
508.3827298
0.230141571
10.81665383
47
6
44
40
220
0.113636364
5
44
0
44
35
367.0817348
0.166175525
7.810249676
47
−6
50
30
550
0.181818182
10
55
−10
60
30
791.4395113
0.184473984
12.08304597
65.5
−11
71
35
640.8732324
0.115467453
8.602325267
74.5
−7
78
30
791.945863
0.142686521
74.5
7
71
22
854.0155151
0.199059615
10.63014581
13.03840481
65.5
11
60
15
528.3029434
0.168463949
9.433981132
56
8
52
10
114.0394669
0.04384447
2.236067977
51
2
50
9
508.3827298
0.230141571
10.81665383
47
6
44
0
dl*avgr
dl/avgr
dl
Avg rad
delta rad
Rad (mm)
Dist (mm)
Area (cm2)
Form factor
Leakage (mm)
Figure 9.4 Spreadsheet-based example for the evaluation of the relevant
dimensions of a simple insulator

Table 9.1 IEC 815 contamination severity table
Pollution level Examples of typical Min. leakage
(max. ESDD) environments distance
I – Light (0.06 mg/cm2
) Areas without industries and
low density of houses
equipped with heating plants.
Areas with low density of
industries or houses but
subjected to frequent winds
and/or rainfall.
16 mm/kV
Agricultural areas (use of
fertilizers can lead to a higher
pollution level).
Mountainous areas.
Note: All these areas shall be
situated at least 10 km to
20 km from the sea and shall
not be exposed to winds
directly from the sea.
II – Medium (0.20 mg/cm2
) Areas with industries not
producing particularly
polluting smoke and/or with
average density of houses
equipped with heating plants.
Areas with high density of
houses and/or industries but
subjected to frequent winds
and/or rainfall.
20 mm/kV
Areas exposed to wind from
the sea but not too close to
the coast (at least several km
distant).
III – Heavy (0.60 mg/cm2
) Areas with high density of
industries and suburbs of
large cities with high density
of heating plants producing
pollution.

Pollution level Examples of typical Min. leakage
(max. ESDD) environments distance
Areas close to the sea or in
any case exposed to relatively
strong winds from the sea.
25 mm/kV
IV – Very heavy
(0.60 mg/cm2
)
Areas generally of moderate
extent, subjected to
conductive dusts and to
industrial smoke producing
particularly thick conductive
deposits.
Areas generally of moderate
extent, very close to the coast
and exposed to sea-spray or
to very strong and polluting
wind from the sea.
31 mm/kV
Desert areas, characterized by
no rain for long periods,
exposed to strong winds
carrying sand and salt, and
subjected to regular
condensation.
Note. The table is based on ceramic and glass insulators (ref. IEC 815). Its use for composite insulators is still to be verified.
9.3.3 Contamination testing
Contamination testing of insulators can be realized under field or laboratory
conditions. Both of these are discussed below.
Field tests
Insulators are exposed to natural pollution at testing stations while subjected
to operating voltage. During testing their performance is monitored throughout
the measurement of parameters such as frequency of flashover and the levels,
durations, and repetition rates of leakage current bursts. Subsequently they are
removed and tested in the laboratory to establish flashover levels. Field test
sites also provide an opportunity to characterize the pollution by removing it
from the insulator and submitting it to chemical analysis. Typical pollutants
which may be found on insulators include NaCl, CaSO4, MgCl2 and CaCO4.

Laboratory tests
There are two laboratory testing procedures in common use. They are known
as the salt fog and the clean fog tests.
(a) Salt fog test
The ‘salt fog method’ reflects the contamination mechanism prevalant along
coastal areas; it is largely followed by European countries and Japan.
In this method a clean dry insulator is energized at its highest working
voltage and is exposed to a salt fog produced by standardized nozzles. The
highest fog salinity (kg of NaCl/m3
) which the insulator can withstand for three
out of four one-hour tests is used to characterize the insulator’s performance.
Alternatively the salinity of the fog is fixed at 80 kg/m3
of solution and the
voltage is raised from 90 per cent of flashover voltage in steps of 2 to 3 per
cent each 5 minutes until flashover.
The standards in salinity vary in values from country to country. For
example, Table 9.2 lists values from the Italian specification.
Table 9.2 Example of salinity
standards
Severity of Withstand salinity
pollution at site required
kg/m3
Light 20–40
Moderate 40–80
Heavy 80–160
Very heavy 100–200
(b) Clean fog test
Clean fog tests reflect the contamination mechanism occurring in industrial
areas. It is the most widely used method. Both methods are accepted as
standards.12,13,14
Studies show that when wetting conditions are properly
controlled the most critical conditions for insulator flashover occur when the
surface is totally contaminated and efficiently wetted by fog or dew with
minimum washing.
9.3.4 Contamination procedure for clean fog testing
The insulator is dipped into a slurry consisting of kaolin, water and NaCl
to give sufficient conductivity. The kaolin provides the mechanical matrix

binding the conductive salt to the insulator. The insulator is allowed to dry, and
is then tested. Alternatively, the contaminants may be deposited by spraying.
This method gives poorer uniformity than dipping.
The deposit density should be recorded for various parts of the insulator.
The clean fog method is applicable to suspension or post type ceramic, as
well as polymer insulators. Figure 9.5 shows the relationship between ESDD
and NaCl concentration of the slurry applied to the insulator under test.
0.001
100.0
10.0
NaCl
(g/l)
1.0
0.001
ESDD (mg/cm2)
Kaolin - 40 g/l
0.001 0.001
Figure 9.5 Relationship between ESDD and contaminant NaCl
concentration
9.3.5 Clean fog test procedure
The dried precontaminated insulators are energized under constant voltage,
and wetted by clean fog. The test voltage is maintained until FO or WS
takes place. For WS the voltage is maintained for at least twice the time
required for FO. Time to maximum wetting rate is determined by monitoring
the leakage current. The maximum leakage current is measured after about
25–30 minutes of wetting. This is found to be the time required to reach
maximum wetting. The critical flashover voltage is then determined using the
up and down method.
Other important parameters which effect the recorded flashover voltage
include the short-circuit current of the testing transformer8
and the rate of
fog temperature rise. The recommended minimum values are a short-circuit
current of 5 A and a fog temperature rise of 0.8°C/min.16

9.3.6 Fog characteristics
The characteristics of the fog introduced to provide insulator wetting have an
effect on the experimental results. The fog is characterized by:
1. Droplet size distribution.
2. Fractional liquid volume (1.8–6.2 g/m3
).
3. Fog temperature. (fog temperature rise).
1 and 2 are interrelated and contribute to wetting and washing of the exposed
insulator surface by droplet impingement. For contamination testing it is essen-
tial that the natural washing conditions be simulated. Depending upon the
source, both droplet size and the fractional liquid volume of natural fog vary
over a wide range. At a fractional volume below 1.8 g/m3
it is difficult to
control temperature and density due to heat loss. The fog temperature must be
significantly higher than ambient in order to ensure effective wetting through
condensation on the insulator surface.
9.4 Mitigation of contamination flashover
There are a number of methods used to reduce or eliminate the possibility of
contamination flashover of ceramic insulators. These include:
ž use of insulators with optimized shapes,
ž periodic cleaning,
ž grease coating,
ž RTV coating,
ž resistive glaze,
ž replacement of ceramic insulators with composite.
9.4.1 Use of insulators with optimized shapes
The shape and leakage distance of insulators can be varied to address
environmental conditions. Generally the shapes are aerodynamically optimized
to gather as little pollution as possible and to enhance self-cleaning through
wind and rain. Standardized variation in shape parameters is available
for service areas characterized by different environmental contamination
processes. Special designs of varying shed profiles, diameters, spacings,
leakage distance, etc. are available.
9.4.2 Periodic cleaning
In many installations high-pressure water systems, or corn and CO2 pellet
blasting, are utilized to periodically clean surface contaminants off insulators.
Of these, high-pressure water cleaning is predominant and by far the cheapest.

Corn and CO2 pellet blasting are far more effective for cleaning cement-like
deposits that are difficult to remove. These procedures are generally applied on
a repetitive basis linked to the pollution composition, severity, and deposition
mechanism as well as local wetting conditions
9.4.3 Grease coating
Coating of insulator surfaces with petroleum gels or hydrocarbon greases is
utilized in areas of heavy contamination. These coatings produce hydrophobic
surfaces and the surface layer is able to encapsulate the contaminants into its
bulk. The former characteristic was covered in the section on non-ceramic
insulators, while the latter prevents the contaminants from going into solution
upon initiation of the surface wetting mechanism. This approach has proven
effective and has been in use for many years. As with washing, this is a
maintenance-based solution, which must be periodically repeated. Usually the
old grease is removed before new grease can be applied. In most instances,
both the application of the new and removal of the old grease are manual
operations. The process is slow and requires circuit outages.
9.4.4 RTV coating
Room temperature vulcanizing (RTV) silicone coatings are being applied with
increasing frequency on both substation and line insulators. RTV coatings
are applied over porcelain insulators and bushings to provide hydrophobic
surfaces (described in detail in the section on non-ceramic insulators). Current
information based on service experience and laboratory testing shows that
these coatings perform well and will last for a number of years. The
lifetime depends upon the coating composition, the application thickness
and of course the pollution severity. RTV coatings are popular in that they
represent a longer-term solution, which does not require replacement of the
insulators.19
They can be applied over existing insulators after adequate
cleaning. A further advantage is that they can be applied to insulators on live
circuits. Their mechanism of resistance to contamination flashover is based on
surface hydrophobicity maintenance and contamination encapsulation. These
processes are similar to those described in subsequent sections dealing with
polymer insulators. As is the case with non-ceramic insulators, RTV coatings
can rapidly deteriorate in the presence of electrical discharges, so care must be
taken at higher voltage levels to ensure that the insulators are free of corona.
9.4.5 Resistive glaze insulators
In areas of heavy contamination resistive glaze insulators are often used to
alleviate contamination flashover. Resistive glaze insulators utilize a special-
ized glaze, which is partially conductive. The glaze is formulated so as to

provide steady state power frequency current flow along the insulator surface.
Its use results in a uniform electric field distribution and surface heating. Both
of these contribute to superior contamination performance. Surface heating
inhibits wetting through condensation and aids in the drying process, whilst
the more uniform electric field distribution acts to control dry band flashover.
The conduction current of the glaze is generally designed to be approximately
1 mA and results in an insulator surface that is several degrees warmer than
the ambient surroundings.
The improvement in contamination flashover performance through the use
of resistive glaze was first demonstrated in the 1940s.17
Since that time resis-
tive glaze insulators have met with mixed success when applied in service.
The technology has been commercially available since the 1950s. The 1970s
saw production of both suspension and post type resistive glaze insulators.
While they both provided excellent contamination flashover resistance, they
suffered from glaze corrosion at the junction point where electrical contact
was made between the metal portions of the insulator and the glaze. The
problem was particularly severe in the case of suspension units where there is
a high current density at the glaze/pin junction. The glaze corrosion resulted
in a break in the conductive path between the insulator’s line and ground
end. This prevented the flow of resistive current and thereby eliminated the
improvement in contamination performance. Manufacture of resistive glaze
suspension insulators was halted after several years, but application of the
technology to the production of post insulators and bushings continued. Over
the years, the corrosion problem on post insulators has been studied and perfor-
mance improved. Currently, resistive glaze post insulators and bushings are
successfully utilized in many installations worldwide where the environmental
conditions are severe. Recently, there has been renewed interest in resistive
glaze suspension insulators. The process utilized in producing these units has
undergone significant improvement and work performed by manufacturers
suggests that the problems associated with severe glaze corrosion have been
successfully addressed.18
9.4.6 Use of non-ceramic insulators
Application of non-ceramic insulators as a solution to contamination flashover
problems has been growing since the early 1980s. Their application for this
purpose and a number of others is given in the section on non-ceramic
insulators.
9.5 Design of insulators
The two basic types of insulators in use are:

1. Ceramic insulators. The material includes porcelain or toughened glass, the
connection is provided by a zinc-coated iron pin and cap-clevis-ball-socket;
cement is used for mounting.
Typical examples are shown in Figs 9.6(a) to (c); all are suspension type,
but are of different shape. Of special interest is the insulator in Fig. 9.6(c)
which has a significantly longer leakage distance of 17.3 inches compared
with 11.5 in the previous two figures. This design is known as the antifog
insulator. All are standard ANSI classes.
The type shown in Fig 9.6(c) is also used in d.c. applications and in cases
where the pollution is very severe.
2. Non-ceramic insulators (NCI). Composite, both suspension and post type, is
displacing the earlier ceramic types, especially at the lower voltage levels.
These were first introduced in the 1960s and 1970s with extensive applica-
tions only within the last two decades. Examples illustrating general design
features are shown in Figs 9.7(a) and (b). The basic materials are polymers
with metal end fittings. The inner part is a fibre glass reinforced resin
rod which provides the mechanical strength. The outer sheds consist of
polymer material, nowadays made of synthetic rubber EPDM or silicon
rubber. Other materials such as epoxies have also been tried in the past,
but the use is limited to indoor applications because of epoxy tracking and
erosion under polluted-wet conditions.
The limiting factors as far as electrical withstand of insulators is concerned
are the environmental conditions, particularly pollution. In dry and clean condi-
tions there are seldom flashover problems experienced with outdoor insulators.
Pollution sets the practical limits for insulator size and design.
The two basic categories of insulators will now be described in more detail.
9.5.1 Ceramic insulators
Insulators are made of ceramic materials which include porcelain and glass.
Their initial use precedes the construction of power systems. They were first
introduced as components in telegraph networks in the late 1800s.
There are a number of basic designs for ceramic insulators, examples were
shown in Figs 9.6(a) to (c). Porcelain is used for the production of cap and pin
suspension units, solid and hollow core posts, pin type, multi-cone and long
rod insulators, and bushing housings. Glass, on the other hand, is used only for
cap and pin suspension and multi-cone posts. Porcelain and glass insulators
are well established, as might be expected based on their long history of use.
Currently these types of insulators comprise by far the majority of in-service
units. Continuous improvements in design and manufacturing processes have
resulted in insulators, which are both reliable and long lasting. Porcelain units
are coated with a glaze to impart strength to the surface. Today’s glass insula-
tors are predominantly manufactured from thermally toughened glass, which

(a)
(c)
(b)
Figure 9.6 Typical constructions of ceramic type suspension insulators.
(a) Standard. (b) Open profile (self-cleaning). (c) Anti-fog and for d.c.
applications

prevents crack formation. Both of the materials have inert surfaces, which
show very good resistance to surface arcing, and both are extremely strong in
compression.
The manufacturing process for electrical porcelain is complex and involves
numerous steps. With glass insulators, the manufacturing process is less
complex, but still requires tight control. Failures of porcelain and glass
insulators can usually be traced back to the manufacture, material or
application of the units. If adequate caution and control in these areas is
not maintained, the likelihood of an inferior product increases. However, as
previously mentioned, when well made, both porcelain and glass insulators
are highly reliable. The majority of bushings and lightning arresters installed
in today’s substations are contained within porcelain housings. Porcelain
housings are, in essence, hollow core post insulators.
Forged steel and fitting galvanized
and joined to rod by swaging process
Fibreglass reinforced resin rod
Injection moulded rubber
weathersheds and rod covering
Sheds of alternating diameters
prevent bridging by ice, snow and
cascading rain
(a)
Figure 9.7 Examples of non-ceramic insulators. (a) Suspension type.
(b) Post type

End fitting joined to rod
by swaging process
Moulded rubber
weathersheds and
protective covering
Malleable iron end fitting
bonded to EPDM rubber
covering: outer
surfaces galvanized
for corrosion protection
Fibre glass reinforced
epoxy resin rod
(b)
9.5.2 Polymeric insulators (NCI)
Polymeric or non-ceramic insulators were first introduced in 1959. They were
made from epoxy and when used outdoors or in contaminated environments,
they were susceptible to problems associated with u.v. degradation, tracking,
and erosion. NCIs were produced by various manufacturers through the 1960s
and 1970s. Those early designs were primarily of the suspension/dead-end and
post type. Certain fundamental aspects of the early designs formed the basis
of today’s production units. They utilized a pultruded fibre glass core as the
strength member. The fibre glass core afforded protection against the environ-
ment through encapsulation in a rubber housing. The mechanical connections

at the insulator ends were made using a variety of means. Some designs
used glued fittings, others had a wedge type attachment, still other manufac-
turers utilized crimping. In all cases metal end fittings were attached to the
fibre glass rod to give the insulator the mechanical strength the applications
required.
Early advocates of NCIs claimed that they achieved up to 90 per cent
weight reduction when compared to their ceramic equivalents. They also had
superior resistance to shock loads due to conductor or hardware failure on
adjacent spans. Another area in which they showed promise was their ability
to withstand vandalism. Significant portions of ceramic insulator failures are
due to vandalism involving shooting. When a bullet hits a ceramic unit, it
breaks or shatters. NCIs do not fail immediately when shot because their
components are not brittle. There are instances reported where NCIs have
remained in service without problems for many years after being shot. A final
advantage claimed by manufacturers and users of early NCIs was that they
could be designed with extremely high leakage lengths which could be easily
optimized to differing environmental conditions.
Early experience with NCIs was confined to short lines and trouble spots.
The trouble spots were generally associated either with areas of environ-
mental contamination or gunshot damage. The initial experience with these
applications proved somewhat disappointing. A host of problems not previ-
ously experienced with ceramic units were encountered. Amongst these were
tracking and erosion, u.v. damage, chalking and crazing, hardware separa-
tion, corona splitting, and water penetration. Many of these were associated
with the use of inappropriate housing materials and manufacturing techniques,
poor quality fibre glass rods, modular sheds and poor sealing between the rod,
housing and end fittings. These operating problems resulted in a significant
number of outages and line drops. Based on the initial field performance, NCIs
saw limited use and therefore production.
By the 1980s the technology had evolved sufficiently to address the concerns
generated through the early field experience. Understanding of the early failure
mechanisms combined with improvements in materials and manufacturing
technology resulted in the development of the NCIs available today. Generally,
today’s NCIs are characterized by a one-piece shed or housing structure. This
one-piece external housing is obtained through single stage moulding or post
assembly vulcanization. Tracking and erosion performance as well as resis-
tance to u.v. has increased markedly. Most industry standards include tracking
and erosion tests, and most of the insulators in production today utilize a track-
free high-temperature vulcanized elastomer housing. The importance of sealing
the exterior of the insulator against moisture has been well recognized, and is
addressed in most current designs. Present experience with these insulators is
beginning to indicate failure rates approaching that of ceramic units.

Today NCIs are utilized as standard products in many of the world’s power
delivery systems. Their main areas of application include distribution and
transmission systems rated up to 345 kV. There is limited use above 345 kV
all the way up to 765 kV.
Shed material
In today’s manufacture of NCIs the most commonly used shed and
housing materials are hydrocarbon and silicone elastomers. The hydrocarbon
elastomers include ethylene-propylene rubbers such as ethylene-propylene
monomer (EPM), ethylene-propylene diene monomer (EPDM), and a co-
polymer of ethylene-propylene and silicone (ESP). The silicone elastomers
include both high temperature and room temperature vulcanizing silicones.
Both these families of materials utilize aluminatrihydrate (ATH) as a filler
which enhances the materials’ tracking performance. The silicone and
hydrocarbon elastomer housing materials have been developed to the stage
where the tracking and u.v. degradation encountered with older designs
are no longer a concern. Both materials are utilized on distribution and
transmission systems. The EPR materials have shown good performance in
clean environments, whereas the silicone-based materials function well in both
clean and contaminated applications.
One of the key characteristics affecting the contamination performance of
NCIs is surface hydrophobicity. Hydrophobicity is a characteristic ability of
a surface to ‘bead’ water which is deposited on it. As previously explained,
contamination flashover of external insulation involves dry band arcing which
develops due to heating and evaporation of electrically continuous liquid paths
formed from the dissolution of surface contaminants in a layer of mois-
ture present on the insulator surface. When a surface has a high degree of
hydrophobicity, water deposited on it forms individual beads or droplets. This
droplet formation inhibits the occurrence of leakage currents and the associated
dry band arcing process. Simply put, an insulator with a highly hydrophobic
surface will be characterized by significantly better contamination flashover
performance than an identical one with a non-hydrophobic surface.
Most polymer insulator housings are hydrophobic when the insulators are
first installed. Exposure to surface discharges, corona and certain chemicals
(including water) reduces the hydrophobicity of polymer surfaces. With EPR-
based housings, exposure to the operating environment results in the reduction
and eventual permanent elimination of surface hydrophobicity. This is one of
the more significant differences between the two housing materials. Unlike
the EPR compounds, silicone housings have the ability to recover a highly
hydrophobic surface state after it has been lost. In the silicone materials used,
high and low weight molecular chains constantly break down and recombine.
The material’s initial hydrophobic state is due to the presence of the low molec-
ular weight oils on the surface. The process of losing hydrophobicity involves

the removal of these oils. In service this occurs primarily through exposure
to surface arcing which can be present when the insulators are applied in
areas of severe contamination. Typically even under extremely severe condi-
tions, the duration of conditions that cause surface arcing is limited to tens
of hours. When the arcing abates, the surface again becomes coated with the
low molecular weight oil and the hydrophobicity is regained. This process
of hydrophobicity regeneration takes somewhere between several hours and
several days. The number of times that the process can repeat is not known,
but given the thickness of the bulk material used, it is expected that the process
can go on for the expected life of the insulators.
Fibre glass core
The mechanical strength of NCIs is provided through the use of a fibre glass
core. For strain, dead-end and solid core post designs, the fibre glass rod
is generally manufactured using a pultrusion process. These pultruded rods
contain axially aligned electrical grade glass fibres in a resin matrix. Two types
of resin are in common use. Epoxy resin is generally believed to give better
performance, while polyester resin is a lower cost alternative. Potential prob-
lems associated with these types of pultruded rod include axial cracking due
to poor handling or manufacturing procedures, and stress corrosion cracking
otherwise termed brittle fracture. Brittle fracture is a process which culminates
in the physical parting of the insulator under low mechanical loading and is
therefore of significant concern. It is not fully understood, and is currently the
focus of a significant amount of research.11
End fitting attachment and moisture ingress
Since their inception, several methods of attaching end fittings to solid core
NCIs have been utilized. Some of the original designs had end fittings that
were glued, while others used a wedge method of connection. The end fitting
fulfils two very important requirements of an NCI. First, it has to be able
to support mechanical loading of the insulator with no slippage. Second, it
must be designed so as to ensure that moisture cannot reach the fibre glass
core through the interface that exists where the end fitting is joined to the
insulator. The importance of the first function is obvious; however, if long-
term performance is to be achieved, the second requirement is more critical.
Most of the end fittings used in today’s designs are either swaged or crimped.
This type of connection has proven to give the best performance from both the
strength and the sealing aspects. Moisture sealing is achieved in three ways.
RTV or some other sealant is applied over the end fitting/housing interface,
the end fitting is installed using an interference or friction fit over the housing,
or the housing material can be extruded over a portion of the end fitting during
the moulding phase of the manufacturing process. The last of these appears

to be the most effective, and the first has proven least reliable in preventing
moisture ingress.
Hollow core NCIs
Hollow core NCIs (HCNCIs) are made of fibre glass filament tubes impreg-
nated with glass epoxy resin. The housing is then generally extruded over
the fibre tube, this extrusion process can result in the manufacture of the
weathersheds, or alternatively, the weathersheds can be fitted over the housing
and vulcanized. In substations they are used primarily as housings for light-
ning arresters, and for transformer, circuit breaker and wall bushings. There
are also some applications where they are used as station post insulators for
supporting buswork, switches, and other electrical equipment. When compared
to conventional ceramic bushings and insulators they offer several advan-
tages. Amongst these are light weight, superior contamination performance
and increased reliability under earthquake conditions. For bushings, their use
represents an important safety enhancement in that unlike porcelain bushings
they do not fail explosively when internal power arcs occur.
9.6 Testing and specifications
All insulators are tested according to standard procedures outlined in various
national and international publications. Ceramic and glass insulators are
mechanically and electrically proof tested prior to shipment. In the case
of NCIs, prior to leaving the factory each production piece is subject to
mechanical but not to electrical proof testing. The primary reason for this
difference is that ceramic and glass units are generally made of a number of
smaller units in series. For example, a 230 kV station post would generally
comprise two smaller posts bolted together to give the clearances required for
230 kV. This allows for piecewise testing of individual components. With NCIs
a 500 kV insulator is manufactured as a single piece. Performing electrical
tests on each unit would require significant time and investment in a sizeable
h.v. test facility. In addition to mechanical and electrical proof tests, the raw
materials used in the production of ceramic, glass, and polymer insulators are
tested as a control on the production process.
With regards to qualification and application testing, the most widely used
standards are those issued by IEC, ANSI, IEEE, CSA, and CEA. Porcelain,
glass, and polymeric insulators are subjected to both electrical and mechan-
ical tests. Depending upon the type of insulator, the electrical tests include
wet and dry power frequency flashover, lightning impulse flashover, steep
front impulse flashover, power arc, and RIV/corona tests. Mechanical tests

include tension, thermal mechanical cycling, torsion, cantilever, and elec-
trical–mechanical testing. Contamination performance tests are also performed
on these insulators in accordance with the techniques discussed earlier.
Contamination flashover performance tests were described in earlier sections
of this chapter. The two test methods mentioned earlier were both developed
for ceramic and glass insulators. These methods and variations upon them are
also being applied to NCIs, and at the same time other tests to characterize
the operation of polymer insulators in contaminated environments are being
developed.
9.6.1 In-service inspection and failure modes
Insulators are often periodically inspected to ensure their continued integrity.
For regular porcelain, this includes visual inspection and in the case of suspen-
sion units, an in-service electrical test. For glass, monitoring usually comprises
only a visual inspection. These simple actions are usually sufficient to detect
any impending problems with ceramic or glass insulators. In-service moni-
toring of NCIs presents a greater challenge. Methods such as IR thermography,
radio noise detection, corona observation, and electric field monitoring have
proven somewhat effective as diagnostic tools for assessing the in-service
condition of NCIs. The difference in approaches to NCI and ceramic/glass
monitoring are due to the different failure mechanisms characterizing each
type of insulator. Damage to ceramic insulators is generally noticeable due to
surface cracks. In suspension units, there can be hidden electrical punctures
through the insulator in the area between the cap and the pin. These are not
visually detectable, but can be detected using a simple field instrument. With
glass insulators, any significant physical damage usually results in destruction
of the shed as the units are made of tempered glass. With NCIs the insulator
can be seriously damaged inside with no indication on its exterior. The mode
of failure predominant with NCIs involves mechanical or electrical failure due
to rod breakage or surface or internal tracking. Because of this, monitoring of
NCIs is more complex than monitoring ceramic or glass insulators. Up-to-date
information on different approaches to the monitoring of NCIs in-service is
summarized in reference 14.
References
1. L.L. Alston and S. Zolendziowski. Growth of Discharges on Polluted Insulation. Proc. IEE
Vol. 110, No. 7, July 1963, pp. 1260–1266.
2. H. Boehme and F. Obenaus. Pollution Flashover Tests on Insulators in the Laboratory and
Systems and the Model Concept of Creepage–path–flashover. CIGRE paper No. 7, June
1969, pp. 1–15.
3. B.F. Hampton. Flashover Mechanism of Polluted Insulation. Proc. IEE Vol. 111, No. 5, May
1964, pp. 985–990.

4. H.H. Woodson and A.J. Mcleroy. Insulators with Contaminated Surfaces, Part H: Modelling
of Discharge Mechanisms. IEEE Trans. on Power Apparatus and Systems Nov./Dec. 1970,
pp. 1858–1867.
5. F.A.M. Rizk. Mathematical Models for Pollution Flashover. Electra Vol. 78, 1981,
pp. 71–103.
6. D.C. Jolly. Contamination Flashover Part II, Flat Plate Model Tests. IEEE Trans. Vol. PAS-
90, No. 6, Nov. 1972, pp. 2443–2451.
7. R. Sudararajan and R.S. Gorur. Dynamic arc modeling of pollution flashover of insulators
under d.c. voltage. IEEE Trans. on Elec. Insul. Vol. 28, No. 2, April 1993, pp. 209–219.
8. F.A. Chagas. Flashover Mechanism and Laboratory Evaluation of Polluted Insulators under
d.c. Voltages. Ph.D. Thesis, Elec. Eng. Dpt., University of Manitoba, Canada, 1996.
9. ANSI/IEEE Std. 987-1985. IEEE Guide for Application of Composite Insulators.
10. IEEE Std. 1024-1988. IEEE Recommended Practice for Specifying Composite Insulators.
11. ANSI/IEEE Std. 987–Final draft balloted in 1999: IEEE Guide for Application of Composite
Insulators, to be published in 2000.
12. IEEE Std. 4-1995. IEEE Standard Techniques for High-Voltage Testing.
13. IEC Publication 507 (1997). Artificial Pollution Tests on High Voltage Insulators (a.c.).
14. IEC Publication 1245 (1993). Artificial Pollution Tests on High Voltage Insulators (d.c.).
15. W.A. Chisholm, P.G. Buchan and T. Jarv. Accurate measurement of low insulator contami-
nation levels IEEE Trans. on PD Vol. 9, July 1994, p. 1552.
16. J.N. Edgar, J. Kuffel and J.D. Mitz. Leakage Distance Requirements for Composite
Insulators Designed for Transmission Lines. Canadian Electrical Association Report CEA
No. 280 T 621, 1993.
17. S.T.J. Looms. Insulators for High Voltages. Peter Peregrinus Ltd, London, 1998.
18. R. Matsuoka, M. Akizuki, S. Matsui, N. Nakashima and O. Fuji. Study of Performance of
Semi-conducting Glaze Insulators. NGK Review Overseas Edition No. 21, Dec. 1997.
19. Round Robin Testing of RTV SILIKON Rubber Coating for Outdoor Insulation. IEEE Trans.
on PD Vol. 11, No. 4, Oct 1996, p. 1889. Paper prepared by IEEE Dielectric and Insulation
Soc. Committee S-32-33.

Index
A.C. transmission systems 1
A.C. peak voltage measurement 111
A.C. to D.C. conversion 10
Active peak-reading circuits 117
A/D converter 179–80
A/D recorder 183
Ammeter, in series with
high ohmic resisters 96
Amorphous dielectric 370
Anode coronas 349, 351
Anti-fog insulator 524
Aperture uncertainty 195
Apparent charge 433
Arithmetic mean value 9
Attachment coefficient () 306
cross-section 306
negative ion formation 304
Avogadro’s number 282
Band pass filter 421
Basic insulation level (BIL) 492
Belt-driven generator 24
Bernoulli–l’Hopital’s rule 142
Best fit normal distribution 490
Binomial distribution 485
Biphase, half-wave rectifier 13
Boltzmann’s constant 282
Boltzmann–Maxwell distribution 285
Boundary element method (BEM) 270
Boyle and Mariotte Law 281
Breakdown and corona inception:
in coaxial cylindrical systems 213
voltage 204
Breakdown criteria 345
Breakdown field strength (Eb) 340
potential relationships 342
Breakdown in solids 367
in liquids 385
in non-uniform fields 326
strength of insulating materials 201
tests on solid dielectric plate materials
233
under impulse voltage 360
Breakdown probability 485, 487, 489
Breakdown voltage curves for N2 358
Breakdown voltage (Vb), expression:
for air 338, 342
for SF6 347–8
Breakdown voltage, of:
rod gaps 94
uniform field gaps 92
Bubble breakdown 391
Bushing, capacitor 235–6, 238, 241
simple, arrangement 236
Capacitance of spark gaps 62–3
Capacitor, reservoir 11
Cathode corona 352
processes 316
Capacitors:
compressed gas 122–4
high voltage 118
Carbon track 385
Cascade circuits 13–14
Cascaded transformers 21, 37, 38, 39
Cavity breakdown 383, 390
Ceramic insulator 523
Charge coupled device (CCD) 177
Charge density 25
Charge Q 11–12
Charge simulation method (CSM) 254
Charge transferred 12

534 Index
Charging resistors 61
voltage 52
Chubb–Fortescue methods:
fundamental circuit 110–11
voltage and current relation 112
CISPR Publication 16-1 438
Clean fog testing 518–9
Cloud chamber photographs 328
Coaxial cable:
cylindrical and spherical fields
209–10
with layers of different permittivities
231
spherical terminations 212
Cockroft–Walton voltage doubler 14
Collision cross section 288
Collisions, elastic 283
Confidence interval 487
Contamination flashover 520
test 517
Corona discharges 348
Crest voltmeter for a.c. measurement
113
with discharge error compensation
117
Cross-section, effective 288
Current comparator bridge 417–9
Current density 312
Damped capacitor voltage divider
162–4
Damping resistors 32
D.C. cascade circuit with transformers
21
D.C. corona inception and breakdown in
air 204
Deionization 302
Deltatron 22
Dielectric loss 411
Dielectric polarization 396
Dielectric refraction 232–3
Dielectric response analysis 410
Differential non-linearity 191
Differential PD bridge 449
Diffusion 313
and mobility relationship 314
coefficient 314, 331, 388
equation 314
Digital transient recorder 175
Dimensioning of circuit elements 57–8
Diode 10, 13
Dirichlet boundaries 247, 251
Discharge in cavity 382
Discharge resistors 53, 64
Disruptive discharge 78
of sphere gaps 83–7
Dissipation factor 406
Divider for D.C. 107
Drag force 388
Drift velocity 311, 391–2
Duhamel’s integral 136
Dynamic error 183
EBS tube 178
Eddy motion 391
Edge breakdown 374
Effective ionisation, coefficient 300,
325
Efficiency of impulse generator 55–7
Electric power 1
Electric strength of highly purified
liquids 387
Electric stress distribution:
cylindrical conductors in parallel
221–2
in sphere-to-sphere arrangement 216,
218
Electroconvection 391
Electrodynamic model of breakdown
391
Electromechanical breakdown 373
Electron affinity 304
Electron attachment 306, 345
avalanche 297
emission by photon impact 323
Electrostatic charging tendency (ECT)
394
Electrostatic fields 201
generators 24
voltmeters 94–6
Elliptical display 437
Energy balance equation 372
Energy functional 250–51

Index 535
Energy level diagram 370
Energy in lightning 464
Energy transfer, collision 292–3, 301
Engetron 22
Environment, contaminated 509
Epoxy disc spacer 234
Equivalent salt deposit density (ESDD)
514
Erosion breakdown 381
Exciting winding 33
Fast digital transient recorder 175
Felici generator 28
Field computations by CSM with surface
charges 268–9
Field distortions by:
conducting particles 221
space charge 326
Field distribution in non-uniform field
gaps 343
Field efficiency factor 202, 203, 214
Field emission 319
Field sensor 107
Field stress control 201
Fields in:
homogeneous, isotropic materials
205
multidielectric materials 225
Finite difference method (FDM) 242
Finite element method (FEM) 246
Flashover characteristics of:
long rod gaps 467–8
insulator strings 467–8
Force, electrostatic voltmeter 94–5
Form factor 513
Formative time lag in N2 332
Four terminal network of measuring
system 132–3
Fourier:
series 133
transform 134–5
Fowler–Nordheim equation 321
Free path 287
Free paths, distribution of 290
Frequency 9
Frequency domain 404
Frequency response of measuring system
133
Front-chopped impulses 50–51, 188
Front oscillation 189
Gap factors (k) 471–2
Gaussian probability distribution curve
474–5
Gay–Lussac’s Law 281
Generating voltmeters 107
principle of 108
G.I.S. (gas insulated substations) 44,
66, 183, 212
Grading electrode 106
Grading rings 106
Grease coating 521
Half wave rectifier 11, 13
Harmonic currents 41
High energy breakdown criterion 371
High temperature breakdown 370
High voltage bridge with Faraday cage
420
High voltage capacitors 118
HVAC 9
HV output:
loaded 16
open circuit 14
HVDC technology 2, 9
voltages and power transmitted 4
Ideal rectifiers 16
IEC Publication 60 9
IEEE Standard 4 9
Image intensifier photographs 329
Impedance of cable 131
supply 41
Impedance, internal of diodes 11
Impulse generator:
outdoor construction 67
indoor construction 69
Impulse generators, design and
construction of 66
Impulse thermal breakdown 378
voltages 48

536 Index
Impulse voltage generators 52
wave components 55
withstand level 493–4
Inductance of h.v. reactor 45
nominal (Ln) 45–6
Insulation aging 409
Insulation coordination 492
level 492
Integral non-linearity 194
Intrinsic breakdown 368
electric strength 369
Ionisation constants 299–300
cross-sections 295, 297
processes 294
Kanal 326, 328
Kaolin 518–9
Laplace transform circuit, of impulse
generator 53–4
Laplace’s equation 245, 258
Laplacian field 246
Leakage path length 513
Lifetime stress relationship of
polyethylene 384
Lightning current 461, 464–5
Lightning impulse voltage, definition of
50
Lightning mechanism 460
Lightning overvoltages 49, 460
Lightning stroke between cloud and
ground, development 463
Lightning strokes 49, 461
Line charge, finite 258, 263
Loss measurement 411
Lossless transmission line 143, 163
Low voltage arm of the divider 171,
174
Main stroke 462
Malter effect 317
Marx generator 61
Matching impedance for signal cable
172–3
Mean free path 287
of atom 289
of electron 289
of ions 309
Mean free time 311
Mean molecular velocities 286
Measurement system, computation of
139–40
Metastables 301
Minimum breakdown voltage 334,
336–7
Minimum sparking constants 337
Mixed resistor-capacitor dividers
156–7
MOA (metal oxide arrester) 500
Mobility 308, 391
of electron 313
of single charged gaseous ions 313
Multi-level test method 480
Multiplier circuit 13
Multiphase rectifiers 13
Multistage generator 60, 61, 64
Narrow band amplifiers 440–41
Narrow band PD detector circuits 437
Nearby earth objects, effect of 89
Non-ceramic insulator 522
Non-destructive insulation test 77, 395
Normalised amplitude frequency spectra
of impulse voltage 135
Numerical methods 241
Operational amplifiers in crest
voltmeters 117
Optimum, number of stages 19
Oscillating switching impulses 66
Output load 11, 16
Output voltage 15
Overvoltages 3, 460
Partial discharge currents 428–9
equivalent circuit 423–4
measurements 421
Paschen’s curve for air 338
Law 333, 339
Passive rectifier circuits for peak voltage
measurement 113

Index 537
Peak measurement 184
Peak reverse 11
Peak voltage, measurement of 78, 109
Peak voltmeter for impulse voltages
116
Peek’s equation 344
Penning effect 339
Phase resolved PD measurement 454
Photoelectric emission 317
Photoexcitation 301
Photoionisation 301
Pilot streamer 462
Point-plane breakdown and corona
inception in air 355
Poisson’s equation 245, 258
Polarity effect of 354
Polluted insulator 510
Polution severity 514, 516
Polymer-housed arrestor 505
Polymeric insulator 526
Positive ion, emission by 317
Potential barrier 319
Potential coefficients 258, 263, 267
Potential distribution along gas capacitor
123
Potential related to point charge 260
Potentials 14
Power rating 31
Probability density function 496
Protection level 495, 498
Quality factor 42, 421, 436
Radio disturbance voltage (RDV)
438–9, 443
Radio interference voltage (RIV) 438,
443, 445–6
Rate of energy gain 371–2
Reactances of transformer windings
39–40
Reactor, prototype 47
Real time digital simulator (RTDS) 496
Recombination 302
Recombination, coefficient of 303
in air 304
Recovery voltage 402, 410–11
Rectifier circuits 10
Rectifier half wave single phase 10
regulation 19
Reference measuring system 91
Relative air density (RAD) 89–90, 339
Relaxation current 401–3
Resistance, time dependent 71
Resistive glaze insulator 521
Resistor, shielded 99–100
Resistor h.v.:
equivalent network 101
standard 99–100
voltage dividers 149, 173
Resistors:
wavefront 52
wavetail 52
Resonance frequency 40
Response function of transfer network
133
Response time of impulse measuring
system 137–8
Response time of resistor dividers
151–2, 166, 170
Return stroke 461
Return voltage 402
Ripple 12, 16, 17, 19
Risk or failure 496, 499
R.M.S. (voltage) 30
Rod gaps 93
Rod-to-plane electrode configuration
202
Rogowski’s profile 207
Rotating barrel generators 28
RTV coating 521
Saha’s equation 302
Sames generator 29
Sampling error 182, 185–6
Saturation current density 318
Schering bridge 412–13
Schottky’s equation 321, 386
Secondary avalanches 330
Secondary electron emission 323
Self-cleaning insulator 524
Self restoring insulation 468

538 Index
Series resonant circuits 40, 42–4
with variable test frequency 45–6
Simulation of dielectric boundary by
discrete charges 266
Single stage generator circuits 52–3
Space charge 354
Space charge field
in negative point-plane gap 359
Sparking voltage-Paschen’s Law
333–4
Sphere gap, clearances around 83
horizontal 81
peak voltage 84–7
vertical 80
Sphere gaps 79
Sphere shank 80
‘Stacked’ capacitor units 124
Standard capacitor for 1000 kV 124
Standard deviation 474
Standard insulation levels 493–4
Static electrification 393
Static error 179
Statistical overvoltage 496
safety margin 495
time lag 360
withstand voltage 484, 493–4
Straight PD detection circuits 431
Stray capacitances of standard capacitor
125
Streamer breakdown 373
Streamer spark criterion 329, 331
Streamer velocity 350–51
Streamer or “Kanal” mechanism of spark
326
Streamers under impulse voltage 350
Stress control by floating screens 235
Surge impedance 1, 163, 165, 172
Suspended solid particle mechanism
387
Switching impulse, standard 51–2
Switching impulse voltages, circuits for
generating 64–5
Switching impulses, oscillating 66
Switching overvoltages 460
Switching surge voltage characteristics
468
Synchronisation and tripping 70
Tandem accelerator 27
Tank of transformer 34, 37
Taylor’s series 243
Tertiary exciting windings 37
Testing transformers 32
single unit 33–4
with mid-point potential 34–5
Testing voltages 5, 479, 484
Testing with lightning impulses 5, 484
A.C. voltages 5, 484
D.C. voltage 6
switching impulses 6, 484
very low frequency 7
Thermal breakdown 369, 375
impulse 378
minimum 379
Thermal capacity 377
Thermal instability 376
Thermal ionisation 302
Thermionic emission 318
Thermal voltage, minimum 380–81
Time delay, built in 70
Time delay of lead L 143–4
Time domain 398
Time lag 359
experimental studies 362–3
overvoltage relationship 364–5
Time to crest 65
Time to peak 51
Tolerances on sphere gaps 81
Townsend breakdown mechanism 324
Townsend criterion for spark 325
Townsend criterion for spark for non
uniform field 342
Townsend first ionisation coefficient
295
Townsend second ionisation coefficient
321
Tracking 385
Transfer characteristics of measuring
system 132
Transformer 11, 32, 35
Transformer current 12
Transformation of a square grid from W
to Z plane 207–8
Transient digital recorder 176
Transmission voltage 2

Index 539
Travel time L 163
Treeing 374
Trichel pulses 353
Trigatron 72–3
Trigatron characteristics 73–4
Trigger electrode 72
Tripping 70, 74
Tunnel effect 319
Ultra-wide-band PD detection 447
Uniform field gaps 92, 206
breakdown voltage 92
Up and down method 480, 483
USR for capacitor voltage divider
160–62
USR and definition of response time
137, 145–6
USR for low value resistor divider 168
USR for mixed dividers 157–8, 166
USR for resistor dividers 152, 169
Van de Graaf generator 24
Velocities, distribution of 284, 286
Virtual front time 50
Virtual time to half value 50
Voltage dividers 96, 130, 147, 149,
156, 159, 163, 171
Voltage dividing systems, impulse
measurements 129
Voltage doubler 13
Voltage multiplier 13, 21
Voltage regulator transformer 42
Voltage stresses 3, 472
Volt–ampere characteristics 501, 503
Volt–time characteristics 361–2
Vr.m.s. 30, 95
Wagner earth 415–16
Wave shaping network 52
Weibull function 477
Wide band PD detection circuits 434–5
Wimshurst machine 24
Work function for typical elements 317
ZnO element 501–2, 504

High_Voltage_Engineering_2E.pdf

More Related Content

Similar to High_Voltage_Engineering_2E.pdf (20)

More from Gollapalli Sreenivasulu (20)

Recently uploaded (20)

High_Voltage_Engineering_2E.pdf