ANN Approach for Fault Classification in Induction Motors using Current and Voltage Signal

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 04 Issue: 11 | Nov -2017 www.irjet.net p-ISSN: 2395-0072
© 2017, IRJET | Impact Factor value: 6.171 | ISO 9001:2008 Certified Journal | Page 1953
ANN Approach for Fault Classification in Induction Motors Using
Current and Voltage Signal
Praful D. Thamke1, Dr. Mrs. Anjali U. Jawadekar2
1Department of Electrical Engineering, S.S.G.M.C.E. shegaon, Maharashtra (444203), India
2 Associate Professor, Department of Electrical Engineering, S.S.G.M.C.E. shegaon, Maharashtra (444203), India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract – In many industries, three-phase Induction
motors especially squirrel cage Induction motor plays an very
important role as an prime mover. So failure of such motor
may leads to increase the productionlossesandalsomayleads
to shut down the entire industries. Hence to prevent such
failure, continuous maintenance schedule is required.
Condition monitoring and Fault classification has great
importance in the industrial line. In this paper, Fault
classification using Artificial Neural Network is proposed.
Motor phase currents and voltage recorded under various
fault conditions were analyzed by using negative sequence
current and Swing angle. The multilayer perceptron feed-
forward ANN were used for the classification purpose.
Key Words: Broken-bar, inter-turn short circuit,
Negative sequence current, swing angle, ANN, Induction
Motor.
1. INTRODUCTION
During 1888, when Nicola tesla was invented of an
Induction motor which is simple and rugged for converting
electrical energy to mechanical energy. But the use of such
types of motor was quite limited due the speed torque
control. But when the power controlled switch was come to
in phase which has the capabilities to implement the field
oriented control system for torque-speed control, the
Induction motor was the mostpopularinevery powersector
areas, Such as industrial, tractional and also agricultural
areas. Now days, accordingly, the induction motor specially
squirrel cage Induction motor are the main prime mover at
many industrial process. So failure of such motorsmayleads
to increase the production loss due to the shut-down of the
plant. Hence maintenance schedules are provided to reduce
such types of failure and to prevent from unwanted
interruption on motor.
Accordingly, the online fault detection systems becomevery
popular tool to increase the efficiency and reliability of the
industrial sectors. Thus online fault classification and
diagnosis becoming very importance in electrical machine
protection since the greatly improve reliability, availability
and maintainability in a wide range of application.
Artificial Neural Network is the most robust technique for
the classification purpose. In this, the negative sequence
current and swing angle used to classify the healthy
condition, rotor broken-bar fault condition and StatorInter-
turn short circuit condition. Symmetrical undaunted motors
powered by symmetrical multiphasevoltagesources haveno
negative sequence currents flowing in the motor. The
internal failure of motor will break thatsymmetry anditgive
rise to a negative sequence current which maybeincreaseas
the severity of fault get increase [3]. And also due to these
internal failure the air gap magnetic field get disturb. So that
these field get oscillate around its original axis duetothisthe
angular phase shift between thecurrentandvoltage whichis
known as swing angle is get change[4]. For the different
types of fault condition the values of negative sequence
current and swing angles are different which can be used to
classify the situation of the motor.
In this paper, Negative sequence current and swing angle is
used to classify the healthy, rotor broken-bar, and stator
inter-turn short-circuit condition of motor with the help of
FFANN. ANN data applied in this work proposed to evaluate
the performance of networks and also present the results
obtained from experimental data.
2. ARTIFICIAL NEURAL NETWORK
ANN is a system based on the biological neural network
such as a brain. The brain has approximately 109 neuron,
which are communicated through junction called synapses.
Each neuron receives thousands of connection
approximately 1012 with other neuron constantly receiving
the incoming signal to reach the cell body. If the resultant
sum of the signal surpasses a certain threshold,a response is
sent signal through axon.
In a similar way Artificial Neural Network is an efficient
computing system whose central theme is borrowed from
the analogy of biological neural networks. ANN has large
collection of units whichareinterconnectedpatternsothatit
allows communication between them. These units are
known as neurons. Every neuron is interconnected with the
other neuron by connection link. Every connection link has
weight. Weight has some important information about the
input signal for particular solution of the network. To
perform specific task, network is adjusted the weight by
comparing the output and the target, until the network
output is matches with the target.
Neural networks have been trained to perform complex
functions in various fields of application including pattern
recognition, identification and classification. In this paper
multilayer perceptron feed-forward network were used for

training purpose, for that pattern recognition neural
network toolbox is used for classification purpose.
1.1 Multi-layer Perceptron’s
A multi-layer perceptron called feed-forward neural
networks with hidden layers contains at least two layers of
functional units shown in figure 1.
Fig -1: Multilayer Perceptron feed-forward network
In MLP FFN least one layer contains hidden units, which
do not communicate with the environment. If the number of
hidden units is appropriately chosen multi layer perceptron
are universal approximators, i.e. they can solve, at least
theoretically, any association problem.
3. NEGATIVE SEQUENCE CURRENT AND SWING
ANGLE
When the motor is on healthy condition, it consists of
symmetrical condition. If the failure occurred then such
motor is categories into unsymmetrical condition. And
according to power system, the analysis of unsymmetrical
circuit can be achieved by usingsymmetrical components i.e.
positive, negative and zero sequence component and
accordingly most of the investigation of fault diagnosis of
induction motor were done by using Negative sequence
current and negative and zero sequence impedance. In this
paper Negative sequence current called are used to classify
the machine behaviour. The expression for negative
sequence Voltage and current is defined as:
Vns(t)= (va(t)+αvb(t)+α2vc(t))
(1)
Ins(t)= (ia(t)+αib(t)+α2ic(t)) (2)
Where, α= exp(j2π/3) is the sequence component
transformation operator.
Vns and Ins are the Negative sequence of stator voltage and
current respectively. va, vb, vc and ia, ib, ic are the measured
quantity of stator voltage and current of phase a, b and c
respectively.
For the classification of the different condition, the direct
measured signals are not suitable. The signal must have
some critical information about the condition of motor,
which describe healthy, rotor broken-bar and stator inter-
turn short circuit of the motor. Therefore the peak value
obtained per cycle of swing angle and absolute value of real
part of negative current signal is used, the expressionforthe
absolute value of real part of the negative sequence current
and phase shift between current and voltage() is define as:
r= abs(real(Ins(t))) (3)
= Ins(t)- Vns(t) (4)
Where the swing angle (Δ) is define as
Δ= max - min (5)
The quantity r and Δ is used as an input layer to ANN to
train the network.
4. EXPERIMENTAL SETUP
A 2-HP, 4 Pole, 415v, 50Hz, three phase squirrel cage
Induction motor was tested under full-Load condition. The
Motor used has 36 slots and 24 coils. Each phase comprises
with 8 coils which carries 300 turns. As a load a 3HP, 240v,
DC shunt generator was used. For capturing the phase
current and voltage signals, 16-bit USB 2.0-based DAQ
Adlink module is used. To sense current, current probes of
input range of 0.01Amps-5 Amps and output range of 400
mV/A, were used and 230/6v, step-down transformers are
used to sense the phase voltage. The experimental setup is
shown in figure 2.
Fig -2: Experimental Setup
4.1. For Healthy Condition:
A balance three phase power suppliesisenergizedto
three phase Induction motor. The motor is operated in
healthy condition. The stator phase voltage and current are
shown in figure 3.

Fig -3: Voltage and current signal at full-load for healthy
condition.
4.2. Rotor Broken-Bar Fault Condition:
As the squirrel cage Induction motor is rugged in
construction, the bar defect generally occurred in a large
machines. This defect comes in machine due to high
temperature and large centrifugal force during transient
operation of motor, such as continuous start up, this may
due to frequent start- stop duty cycle. And also the bar get
damage due to poor end ring joints forms during
manufacturing. Once the defect occurred, rotor gets
overheated in the cage, which produces high centrifugal
forces.
In this paper the Rotor Broken bar achieved by cutting
5 bar of the rotor. In this the current and voltage signal are
shown in figure 4.
Fig -4: Voltage and current signal at full-load for Rotor
Broken-bar
4.3. Stator Inter-turn Short Circuit
Stator faults generally occurred due to the winding
insulation failure. This may due to overloading, frequent
motor start and stop, coil vibration and due to transient
voltage stress. Such faults may leads to generate heat in a
defective region of a winding which causes the fault rapidly
progress to more severs forms such as phase to ground and
phase to phase faults.
In this, A phase has been tapped where eachtapping
is made after 10 turns, near to the star point (neutral). For
Inter-turn fault 20 turn of phase A gets shorted. The current
and voltage signal for stator with 20-turn short circuited are
shown in figure 5.
Fig -5: Voltage and current signal at full-load for stator
Inter-turn fault.
5. RESULTS:
Stator voltage and currents are captured from motor
terminals under healthy and faulty conditions, at full load
conditions. From the voltage and current signal, negative
sequence current Ins(t) and voltage Vns(t) are obtained by
using equation (1) and (2). And to definerand equation (3)
and (4) were used. The curve for peak values of r and Δ
obtained for each cycle for healthy, Rotor broken bar and
Stator inter-turn short circuit for seven second are shownin
figure 6 and 7 respectively.
Fig -6: rpeak for Healthy, broken rotor-bar and stator inter-
turn short circuit

Fig -8: Δpeak for Healthy, broken rotor-bar and stator
inter-turn short circuit.
5.1. Fault Classification using ANN
An Artificial neural network with its patternrecognitiontool
can be effectively employed for the classification of fault of
three phase induction motor. In present study two layer
feed-forward artificial neural networks is used and trained
with supervised learning algorithm. FFANN consists of one
hidden layer and one output layer. In Input layer, it consists
of 2 neurons; i.e. rpeak and Δpeak. Output layer consists of
three neurons representing healthy, rotor fault and stator
fault respectively. For training, Learning rate =0.8,
Momentum=0.5,and for constrained output, Tansigmoid
Transfer function is used, the samples are divided into three
subset which are training subset, testing subset and
validation subset. For training subset contains 80% of
samples are set, for Testing is 15 % and for validation 05%.
With these assumptions variationofpercentageaccuracyfor
classification of faults of induction motor with respect to
number of processing elements in hidden layer is obtained.
Table 1 and Figure 9 and shows % accuracy with the
variation of processing elements.
Table -1: Percentage accuracy of classification
No. of
Processing
Element
Percentage accuracy
Healthy Rotor Fault Stator Fault
1 85% 60% 100%
2 100% 80% 100%
3 100% 100% 100%
Fig -9: percentage accuracy using ANN
6. CONCLUSIONS
In This paper, an application of neural network for healthy,
broken rotor bar and stator inter-turn fault detection of
three phase induction motor using stator current data are
presented. The major component for analysis is used to
extract pattern and data reduction. Stator current and
voltage signals are recorded, statistical parametercomputed
for healthy and faulty conditions. Further analyzed by
negative sequence current and swing angle which are fed as
an input to ANN. The results shows that the ANN with major
principal components was the most efficient regarding the
criteria of accuracy in fault detection with three processing
elements in hidden layer.
REFERENCES
[1] V. N. Ghate and S. V. Dudul “Cascade Neural-Network-
Based Fault Classifier forThree-PhaseInductionMotor,”
IEEE transactions on industrial electronics,vol.58,no.5,
may 2011, pp. 1555-1562.
[2] J.H. Jung, J.J Lee, B.H. Kwon, “Online Diagnosis of
Induction Motors Using MCSA” IEEE Transactions on
Industrial Electronics, vol.53, no 6,pp1842-1852,
December 2006..
[3] G.B.Kliman,W.J.Premerlani, R.A.Koeg1&D.Hoevveler,“A
New Approach to On-Line Turn Fault Detection in AIC
Motors” 0-7803-3544-919$65 .00 0 1996IEEE,pp.687-
693.
[4] A. U. Jawadekar, S. Paraskar, S. Jadhav & G. Dhole,
“Artificial neural network-based induction motor fault
classifier using continuous wavelet transform,” Systems
Science & Control Engineering: An Open Access Journal,
2014 Vol. 2, 684–690

ANN Approach for Fault Classification in Induction Motors using Current and Voltage Signal

More Related Content

What's hot (20)

Similar to ANN Approach for Fault Classification in Induction Motors using Current and Voltage Signal (20)

More from IRJET Journal (20)

Recently uploaded (20)

ANN Approach for Fault Classification in Induction Motors using Current and Voltage Signal