First-Order_Process_Time__Delay_2002.pdf

Mechatronics
Control of a First-Order Process + Dead Time
K. Craig
1
Control of a First-Order Process
with Dead Time
Σ
V DT s
Ke
s 1
− τ
τ +
C
+
-
Controller
E M
B
The most commonly used model to describe the dynamics
of chemical processes is the First-Order Plus Time Delay
Model. By proper choice of τDT and τ, this model can be
made to represent the dynamics of many industrial
processes.

Mechatronics
K. Craig
2
• Time delays or dead-times (DT’s) between inputs and
outputs are very common in industrial processes,
engineering systems, economical, and biological systems.
• Transportation and measurement lags, analysis times,
computation and communication lags all introduce DT’s
into control loops.
• DT’s are also used to compensate for model reduction
where high-order systems are represented by low-order
models with delays.
• Two major consequences:
– Complicates the analysis and design of feedback
control systems
– Makes satisfactory control more difficult to achieve

Mechatronics
K. Craig
3
• Any delay in measuring, in controller action, in actuator
operation, in computer computation, and the like, is called
transport delay or dead time, and it always reduces the
stability of a system and limits the achievable response time
of the system.
( ) ( )
i
o i DT DT
q (t) input to dead-time element
q (t) output of dead-time element q t u t
=
= = − τ − τ
( )
( )
DT DT
DT DT
u t 1 for t
u t 0 for t <
− τ = ≥ τ
− τ = τ
( ) ( ) ( )
DTs
DT DT
L f t u t e F s
−τ
 
− τ − τ =
 
Dead
Time
qi(t) qo(t)
Laplace Transform

Mechatronics
K. Craig
4
Qi
(s) Qo
(s)
Amplitude
Ratio
Phase
Angle
Dead Time
Frequency
Response
qi
(t) qo
(t)
DT
−ω τ
DT
τ DTs
e
− τ
1.0
0D
φ

Mechatronics
K. Craig
5
• Dead-Time Approximations
– The simplest dead-time approximation can be obtained graphically
or by taking the first two terms of the Taylor series expansion of
the Laplace transfer function of a dead-time element, τDT.
– The accuracy of this approximation depends on the dead time
being sufficiently small relative to the rate of change of the slope
of qi(t). If qi(t) were a ramp (constant slope), the approximation
would be perfect for any value of τDT. When the slope of qi(t)
varies rapidly, only small τDT's will give a good approximation.
– A frequency-response viewpoint gives a more general accuracy
criterion; if the amplitude ratio and the phase of the approximation
are sufficiently close to the exact frequency response curves of
for the range of frequencies present in qi(t), then the approximation
is valid.
( ) DTs
o
DT
i
Q
s e 1 s
Q
−τ
= ≈ − τ ( ) ( ) i
o i DT
dq
q t q t
dt
≈ − τ

Mechatronics
K. Craig
6
Dead-Time Graphical Approximation
tangent line
DT
τ
( )
o i DT
q q t
= − τ
( ) i
o i DT
dq
q q t
dt
= − τ
qi(t)
qi
t

Mechatronics
K. Craig
7
– The Pade approximants provide a family of
approximations of increasing accuracy (and
complexity):
– In some cases, a very crude approximation given by a
first-order lag is acceptable:
( ) DTs
o
i DT
Q 1
s e
Q s 1
−τ
= ≈
τ +
k
2 2
s
2
s
s k
2
2 2
s
s s 2
1
e 2 8 k!
e
s
e
s s 2
1
2 8 k!
−τ
−τ
τ
τ
 
−
 
τ τ  
− + + +
= ≈
τ
 
 
τ τ  
+ + + +
"
"

Mechatronics
K. Craig
8
• Pade Approximation:
– Transfer function is all pass, i.e., the magnitude of the
transfer function is 1 for all frequencies.
– Transfer function is non-minimum phase, i.e., it has
zeros in the right-half plane.
– As the order of the approximation is increased, it
approximates the low-frequency phase characteristic
with increasing accuracy.
• Another approximation with the same properties:
k
s
2
s
s k
2
s
1
e 2k
e
s
e 1
2k
−τ
−τ
τ
τ
 
−
 
 
= ≈
τ
 
+
 
 

Mechatronics
K. Craig
9
10
-1
10
0
10
1
10
2
10
3
10
4
10
5
-400
-350
-300
-250
-200
-150
-100
-50
0
frequency (rad/sec)
phase
angle
(degress)
Dead-Time Phase-Angle Approximation Comparison
( )
o dt
i dt
Q 2 s
s
Q 2 s
− τ
=
+ τ
( )
( )
( )
2
dt
dt
o
2
i dt
dt
s
2 s
Q 8
s
Q s
2 s
8
τ
− τ +
=
τ
+ τ +
dts
dt
e 1
−τ
= ∠ − ωτ
τdt = 0.01
Dead-time Approximation Comparison

Mechatronics
K. Craig
10
• Observations:
– Instability in feedback control systems results from an
imbalance between system dynamic lags and the
strength of the corrective action.
– When DT’s are present in the control loop, controller
gains have to be reduced to maintain stability.
– The larger the DT is relative to the time scale of the
dynamics of the process, the larger the reduction
required.
– The result is poor performance and sluggish responses.
– Unbounded negative phase angle aggravates stability
problems in feedback systems with DT’s.

Mechatronics
K. Craig
11
– The time delay increases the phase shift proportional to
frequency, with the proportionality constant being
equal to the time delay.
– The amplitude characteristic of the Bode plot is
unaffected by a time delay.
– Time delay always decreases the phase margin of a
system.
– Gain crossover frequency is unaffected by a time delay.
– Frequency-response methods treat dead times exactly.
– Differential equation methods require an approximation
for the dead time.
– To avoid compromising performance of the closed-loop
system, one must account for the time delay explicitly,
e.g., Smith Predictor.

Mechatronics
K. Craig
12
Smith Predictor
G(s)
D(s) e s
−τ
G s e s
( )[ ]
1− −τ
Smith
Predictor
- -
+ +
Σ Σ
yr y
~
( )
D s
s
s
s
r
y D(s)G(s)e D(s)G(s)
e
y 1 D(s)G(s)e 1 D(s)G(s)
−τ
−τ
−τ
= =
+ +

s
D(s)
D(s)
1 (1 e )D(s)G(s)
−τ
=
+ −

Mechatronics
K. Craig
13
• D(s) is a suitable compensator for a plant whose transfer
function, in the absence of time delay, is G(s).
• With the compensator that uses the Smith Predictor, the
closed-loop transfer function, except for the factor e-τs, is
the same as the transfer function of the closed-loop system
for the plant without the time delay and with the
compensator D(s).
• The time response of the closed-loop system with a
compensator that uses a Smith Predictor will thus have the
same shape as the response of the closed-loop system
without the time delay compensated by D(s); the only
difference is that the output will be delayed by τ seconds.

Mechatronics
K. Craig
14
k
2 2
s
2
s
s k
2
2 2
s
s s 2
1
e 2 8 k!
e
s
e
s s 2
1
2 8 k!
−τ
−τ
τ
τ
 
−
 
τ τ  
− + + +
= ≈
τ
 
 
τ τ  
+ + + +

• Implementation Issues
– You must know the plant transfer function and the time
delay with reasonable accuracy.
– You need a method of realizing the pure time delay that
appears in the feedback loop, e.g., Pade approximation:

Mechatronics
K. Craig
15
t
time
out
output
Sum
Step
1
s2
Plant
in
Input
121.7
Gain
s+3
s+18.23
Controller
Clock
Basic Feedback Control System with Lead Compensator
Example Problem

Mechatronics
K. Craig
16
BUT with Time Delay τ = 0.05 sec
t
time
out_delay
output
Transport
Delay
Sum
Step
1
s2
Plant
in
Input
121.7
Gain
s+3
s+18.23
Controller
Clock

Mechatronics
K. Craig
17
t
time
out_delay_SP
output
Transport
Delay
tau^2/8.s -tau/2s+1
2
tau^2/8.s +tau/2.s+1
2
Time Delay
Sum2
Sum1
Sum
Step
1
s2
Plant
1
s2
Plant
1
s2
Plant
in
Input
121.7
Gain
s+3
s+18.23
Controller
Clock
BUT with Time Delay τ = 0.05 sec
AND Smith Predictor

Mechatronics
K. Craig
18
0 0.5 1 1.5 2 2.5 3
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
time
response
time (sec)
No Time Delay
Time Delay τ = 0.05 sec
Time Delay τ = 0.05 sec
with Smith Predictor
System Step Responses

Mechatronics
K. Craig
19
• Comments
– The system with the Smith Predictor tracks reference
variations with a time delay.
– The Smith Predictor minimizes the effect of the DT on
stability as model mismatching is bound to exist. This
however still allows tighter control to be used.
– What is the effect of a disturbance? If the disturbances
are measurable, the regulation capabilities of the Smith
Predictor can be improved by the addition of a
feedforward controller.

Mechatronics
K. Craig
20
• Minimum-Phase and Nonminimum-Phase Systems
– Transfer functions having neither poles nor zeros in the
RHP are minimum-phase transfer functions.
– Transfer functions having either poles or zeros in the
RHP are nonminimum-phase transfer functions.
– For systems with the same magnitude characteristic, the
range in phase angle of the minimum-phase transfer
function is minimum among all such systems, while the
range in phase angle of any nonminimum-phase transfer
function is greater than this minimum.
– For a minimum-phase system, the transfer function can
be uniquely determined from the magnitude curve
alone. For a nonminimum-phase system, this is not the
case.

Mechatronics
K. Craig
21
Frequency (rad/sec)
Phase
(deg);
Magnitude
(dB)
Bode Diagrams
-6
-4
-2
0
From: U(1)
10-2
10-1
100
-200
-150
-100
-50
0
To:
Y(1)
G1(s)
G2(s)
A small amount
of change in
magnitude
produces a small
amount of
change in the
phase of G1(s)
but a much
larger change in
the phase of
G2(s). T1 = 5
T2 = 10
Consider as an example the following two systems:
( ) ( )
1 1
1 2 1 2
2 2
1 Ts 1 Ts
G s G s 0 T T
1 T s 1 T s
+ −
= =
+ +

Mechatronics
K. Craig
22
– These two systems have the same magnitude
characteristics, but they have different phase-angle
characteristics.
– The two systems differ from each other by the factor:
– This factor has a magnitude of unity and a phase angle
that varies from 0° to -180° as ω is increased from 0 to ∞.
– For the stable minimum-phase system, the magnitude and
phase-angle characteristics are uniquely related. This
means that if the magnitude curve is specified over the
entire frequency range from zero to infinity, then the
phase-angle curve is uniquely determined, and vice versa.
This is called Bode’s Gain-Phase relationship.
1
1
1 Ts
G(s)
1 Ts
−
=
+

Mechatronics
K. Craig
23
– This does not hold for a nonminimum-phase system.
– Nonminimum-phase systems may arise in two different
ways:
• When a system includes a nonminimum-phase element or
elements
• When there is an unstable minor loop
– For a minimum-phase system, the phase angle at ω = ∞
becomes -90°(q – p), where p and q are the degrees of
the numerator and denominator polynomials of the
transfer function, respectively.
– For a nonminimum-phase system, the phase angle at ω
= ∞ differs from -90°(q – p).
– In either system, the slope of the log magnitude curve
at ω = ∞ is equal to –20(q – p) dB/decade.

Mechatronics
K. Craig
24
– It is therefore possible to detect whether a system is
minimum phase by examining both the slope of the
high-frequency asymptote of the log-magnitude curve
and the phase angle at ω = ∞. If the slope of the log-
magnitude curve as ω → ∞ is –20(q – p) dB/decade and
the phase angle at ω = ∞ is equal to -90°(q – p), then
the system is minimum phase.
– Nonminimum-phase systems are slow in response
because of their faulty behavior at the start of the
response.
– In most practical control systems, excessive phase lag
should be carefully avoided. A common example of a
nonminimum-phase element that may be present in a
control system is transport lag: dts
dt
e 1
−τ
= ∠ − ωτ

Mechatronics
K. Craig
25
Time (sec.)
Amplitude
Step Response
0 2 4 6 8 10 12
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
1.2
1.4
From: U(1)
To:
Y(1)
Unit Step Responses
2
1
s s 1
+ +
2
s 1
s s 1
+
+ +
2
s 1
s s 1
− +
+ +

Mechatronics
K. Craig
26
Time (sec.)
Amplitude
Step Response
0 2 4 6 8 10 12
-0.2
0
0.2
0.4
0.6
0.8
1
1.2
1.4
From: U(1)
To:
Y(1)
2
1
s s 1
+ +
2
s
s s 1
+ +
2
s 1
s s 1
+
+ +
Unit Step Responses

Mechatronics
K. Craig
27
Time (sec.)
Amplitude
Step Response
0 2 4 6 8 10 12
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
1.2
1.4
From: U(1)
To:
Y(1)
Unit Step Responses
2
s 1
s s 1
− +
+ +
2
s
s s 1
−
+ +
2
1
s s 1
+ +

Mechatronics
K. Craig
28
• Nonminimum-Phase Systems: Root-Locus View
– If all the poles and zeros of a system lie in the LHP,
then the system is called minimum phase.
– If at least one pole or zero lies in the RHP, then the
system is called nonminimum phase.
– The term nonminimum phase comes from the phase-
shift characteristics of such a system when subjected to
sinusoidal inputs.
– Consider the open-loop transfer function:
( )
( )
K 1 2s
G(s)H(s)
s 4s 1
−
=
+

Mechatronics
K. Craig
29
-1 -0.5 0 0.5 1 1.5 2
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
Real Axis
Imag
Axis
Root-Locus Plot
( )
( )
K 1 2s
G(s)H(s)
s 4s 1
−
=
+
Angle Condition:
( )
K(2s 1)
G(s)H(s)
s(4s 1)
K(2s 1)
180 180 2k 1
s(4s 1)
− −
∠ = ∠
+
−
= ∠ + ° = ± ° +
+
K(2s 1)
0
s(4s 1)
−
∠ = °
+
1
K
2
=
or

Mechatronics
K. Craig
30
• Dynamic Response of a First-Order System with a
Time Delay
– The transfer function of a time delay combined with a
first-order process is:
– Consider the case with: K =1, τ = 10, τDT = 5, and a
unit step input at t = 0.
– Simulate the step response with:
• An exact representation of a time delay
• A first-order Pade approximation of a time delay
• A second-order Pade approximation of a time delay
– Simulate the frequency response for the same cases.
DTs
Ke
s 1
−τ
τ +

Mechatronics
K. Craig
31
Time (sec.)
Amplitude
Step Response
0 14 28 42 56 70
-0.2
0
0.2
0.4
0.6
0.8
1
1.2
From: U(1)
To:
Y(1)
No Time Delay
Exact Time Delay
1st-Order Approx.
2nd-Order Approx.
Note the inverse response
and the double inverse
response in the plots
using the time delay
approximations. How
does this relate to RHP
zeros?

Mechatronics
K. Craig
32
Frequency (rad/sec)
Phase
(deg);
Magnitude
(dB)
Bode Diagrams
-80
-60
-40
-20
0
From: U(1)
10-2
10-1
100
101
102
-500
-400
-300
-200
-100
0
To:
Y(1)
Exact Time Delay
No Time Delay
2nd-Order Approx.
1st-Order Approx.
Magnitude Plot is same
for all cases.

Mechatronics
K. Craig
33
• Empirical Model
– The most common plant test used to develop an
empirical model is to make a step change in the
manipulated input and observe the measured process
output response.
– Then a model is developed to provide the best match
between the model output and the observed plant
output.
– Important Issues:
• Selection of the proper input and output variables.
• In step-response testing, we first bring the process to a
consistent and desirable steady-state operating point, then
make a step change in the input variable.
• What should the magnitude of the step change be?

Mechatronics
K. Craig
34
1. The magnitude of the step input must be large enough so that
the output signal-to-noise ratio is high enough to obtain a
good model.
2. If the magnitude of the step input is too large, nonlinear
effects may dominate.
• Clearly there is a trade off.
– By far the most commonly used model for control-
system design purposes, is the 1st-order plus time
delay model.
– The three process parameters can be estimated by
performing a single step test on the process input.
DTs
Ke
s 1
−τ
τ +

Mechatronics
K. Craig
35
Process
time time
DT
τ
tangent at
steepest slope
K is the long-term change in process output
divided by the change in process input.
Estimate time constant from a semi-log
plot of first-order response.

Mechatronics
K. Craig
36
– If the process is already in existence, experimental step
tests allow measurement of τDT and τ.
– At the process design stage, theoretical analysis allows
estimation of these numbers if the process is
characterized by a cascade of known 1st-order lags.
– Approximate the dead time with a 1st-order Pade
approximation:
– Consider the open-loop transfer function:
DT
DT
2 s
2 s
− τ
+ τ
DT
DT
s
DT
2 s
K
2 s
Ke
G
s 1 s 1
−τ
 
− τ
 
+ τ
 
≈ =
τ + τ +

Mechatronics
K. Craig
37
– The closed-loop system transfer function is:
– The characteristic equation of the closed-loop system
is:
– For what value of K will this system go unstable?
C G
V 1 G
=
+
( ) ( )
DT
DT
2
DT DT DT
1 G(s) 0
2 s
K
2 s
1 0
s 1
s 2 K s 2 K 1 0
+ =
 
− τ
 
+ τ
 
+ =
τ +
τ τ + τ + τ − τ + + =

Mechatronics
K. Craig
38
– The Routh Stability Criterion predicts that for stability:
– The gain value for marginal stability can be found
precisely from the Nyquist criterion since we know the
frequency response of a dead time exactly. For
marginal stability, we require that (B/E)(iω) go
precisely through the point –1 = 1∠180°. The phase
angle part of the requirement can be stated as:
– This fixes (for a given τ τDT) the frequency ω0 at which
(B/E)(iω) passes through the point –1 = 1∠180°.
DT
1 K 2 1
 
τ
− +
 
τ
 
1
0 DT 0
tan−
−π = −ω τ − ω τ

Mechatronics
K. Craig
39
( )
( )
2
0
B K
i 1.0
E 1
ω = =
ω τ +
– This equation has no analytical solution. Once ω0 is
found numerically, the gain K for marginal stability is
obtained by requiring that:
– A table shows results for a range of the most common
values encountered for τDT / τ in modeling complex
systems.

Mechatronics
K. Craig
40
2.26
2.03
1.0
2.43
2.22
0.9
2.64
2.45
0.8
2.92
2.74
0.7
3.29
3.13
0.6
3.81
3.67
0.5
4.59
4.48
0.4
5.89
5.80
0.3
8.50
8.44
0.2
16.4
16.4
0.1
K
ω0τ
τDT / τ

Mechatronics
K. Craig
41
– The steady-state error is typical of proportional control.
Design values of K must be less than those for marginal
stability.
– A design criterion sometimes used in industrial process
control is quarter-amplitude damping, wherein each cycle of
transient oscillation is reduced to ¼ the amplitude of the
previous cycle.
– A useful approximation for this behavior is a gain margin of
2.0 for the frequency response.
– If we apply this to the table of results for, say, τDT / τ = 0.2,
we get a design gain value of 4.25, giving large steady-state
errors.
– For this reason, processes of this type often use integral or
proportional + integral control, which reduces steady-state
errors without requiring large K values.

Mechatronics
K. Craig
42
• Exercise:
– For the closed-loop system below, evaluate the step
response using:
• τDT = 1 sec
• τ = 5 sec
• K = 8.5, 4.25, 2.13, 1.06
Σ
V DTs
Ke
s 1
− τ
τ +
C
+
-
E
B

Mechatronics
K. Craig
43
0 1 2 3 4 5 6 7 8 9 10
-0.5
0
0.5
1
1.5
2
time (sec)
Response
First-Order + Time Delay Closed-Loop Response: K = 8.5, 4.25, 2.13, 1.06
K = 8.5
K = 4.25
K = 2.13
K = 1.06

Mechatronics
K. Craig
44
• Consider Integral Control of a First-Order Process
plus a Dead Time
– Proportional control was found to be difficult since
loop gain was restricted by stability problems to low
values, causing large steady-state error.
– Integral control gives zero steady-state error ( for both
step commands and/or disturbances) for any loop gain
and is thus an improvement.
– The values of K for marginal stability are given in the
following table.
– Compared with proportional control, both loop gain and
speed of response (ω0 for a given τ) are lower.
However, we do not depend on it to reduce steady-state
error.

Mechatronics
K. Craig
45
1.14
0.86
1.0
1.25
0.92
0.9
1.39
0.99
0.8
1.57
1.07
0.7
1.81
1.18
0.6
2.15
1.31
0.5
2.65
1.48
0.4
3.49
1.74
0.3
5.16
2.16
0.2
10.2
3.11
0.1
K
ω0τ
τDT / τ

Mechatronics
K. Craig
46
• Check Time-Domain Response
– Run simulations on the system for KI = 1.14 (marginal
stability) and for KI = 0.57 (gain margin of 2.0).
– Check response of C to both step inputs in V and U.
– Note the well-damped response with zero steady-state
error for both step commands and disturbances for KI =
0.57.
1
s+1
Process
Ki
s
Integral Control
Disturbance U
Unit Step at t = 25 sec
Dead Time
C
Controlled Variable
Command V
Unit Step at t = 0

Mechatronics
K. Craig
47
0 5 10 15 20 25 30 35 40 45 50
-1
-0.5
0
0.5
1
1.5
2
2.5
3
time (sec)
Response
Integral Control: First-Order + Time Delay Closed-Loop Response: Ki = 1.14, 0.57
KI = 1.14
KI = 0.57

First-Order_Process_Time__Delay_2002.pdf

More Related Content

Similar to First-Order_Process_Time__Delay_2002.pdf (20)

Recently uploaded (20)

First-Order_Process_Time__Delay_2002.pdf