MKEM1732: Assignment 2

Adaptive Feedback Controller Design based on

Gradient and Stability Approach
For a Chemical Process
Ling Nai Ho (MKE141067), Ibrahim Albool (MKE141040)
Faculty of Electrical Engineering,
Universiti Teknologi Malaysia

Abstract— Adaptive control refers to modification of the control The aim of this assignment is to design two controllers for a
law used by the controller to cope with the drastic change in chemical process based on the gradient and stability
parameters of the system being controlled due to changes in approaches. The performance of both will be compared via
environmental conditions and in system itself. The main aim of Matlab Simulink simulation to show the robustness and the
adaptive control process is to generate an actuating signal in such
adaptation ability of the controllers. Series of analysis have
a way that optimal performance can be maintained regardless of
system changes. This paper deals with application of model been done by varying the time constant, τ and time delay,
reference adaptive control scheme and the system performance is and also the amplitude of the reference input signal, r.
compared with Gradient and Stability approach. A first order
reference model and chemical plant has been taken as the system The paper is organized as follows where the basic theory
for analysis with a PI controller. Comparison is done between and formula of PI Controller is briefly described in section II,
different values of time constant and time delay in the system inMRAC theory and its structure in section III whereas gradient
both MIT rule and Lyapunov method to evaluate the adaptation and stability approaches are described in section IV and V
performance of the designed controller.
respectively. The adaptive feedback controller design scheme
as well as time delay approximation are described in section
Keywords — Model reference adaptive control, gradient approach, VI. The simulation results and analysis are as shown in
stability approach, PI Controller, Adaptive feedback controller.
section VII. Finally, the results are concluded at the end of
the paper.
I. INTRODUCTION II. PI CONTROLLER

A control system is in the broadest sense, an

interconnection of the physical components to provide a An adaptive Proportional-Integral (PI) controller is
required to control a chemical process. The general formula
desired function, involving some type of control action with it.
adopted for the PI-controller is given as below:
The requirement of high performance control system for
industrial applications has produced great research efforts for
the application of modern control theory and, in particular,
adaptive control. As compared to fixed gain PID controllers
Adaptive Controllers are very effective to handle the
situations where the parameter variations and environmental
changes are frequent. The controller parameters are adjusted where TI is integration time constant. If the controller is
to give a desired closed-loop performance. The adaptive tuned to be slow and TI is large, the controller will first act like
controller maintains constant dynamic performance in the a Proportional controller, however the steady-state deviation
presence of unpredictable and immeasurable variations. w i l l s l o wl y g o to zero after that.
Adaptive control changes the control algorithm coefficients in On the other hand, as the integration starts to take effect,
real time to compensate for variations in the environment or in In PI- control, the steady-state deviation will finally go to
zero. If the controller is tuned to be fast and TI is small, then
the system itself. It also varies the system transfer function
both terms (P and I) will affect the control signal all the way
according to situation.
from the beginning. The system becomes faster, but the
In this paper we focus on design of a PI controller with
output signal might oscillate and eventually overshoot.
adjustable parameters and a mechanism for adjusting the
Thus, a proper P and I values need to be selected and
parameters by using the gradient approach (MIT) and the adjust properly for better transient performance. This can be
stability approach (Lyapunov Method). done through adaptation method as described in section V.

1
 MKEM1732: Assignment 2

III. MODEL REFERENCE ADAPTIVE CONTROL (e=y- ) by designing a controller that has one or more
adjustable parameters such that a certain cost function is
The general structure of the Model Reference Adaptive minimized so that the output of the closed-loop system (y) to
follow the output of the reference model (ym).
Control (MRAC) system is shown in figure 1 below. The
basic MRAC system consists of 4 main components:
V. STABILITY METHOD (LYAPUNOV)
i) Plant to be controlled
ii) Reference model to generate desired closed loop
The Lyapunov stability theory can be used to describe the
output response
algorithms for adjusting parameters in Model Reference
iii) Controller that is time-varying and whose
Adaptive control system. MRAC can be designed such that
coefficients are adjusted by adaptive mechanism
the globally asymptotic stability of the equilibrium point of
iv) Adaptive mechanism that uses „error‟ (the difference
the error difference equation is guaranteed. To do this, the
between the plant and the desired model output) to
Lyapunov Second Method is used. It requires an appropriate
produce controller coefficient
Lyapunov function to be chosen, which could be difficult.
This approach has stability consideration in mind and is
also known as the Lyapunov Method.

VI. ADAPTIVE FEEDBACK CONTROLLER DESIGN SCHEME

The chemical process is described by the transfer functions

and a controller by the form as indicated below where the time
delay TD and the time constant τ, of the process are unknown.

Plant: ( ) (1)
Fig. 1: General Structure of MRAC system

Regardless of the actual process parameters, adaptation in Ref. Model: ( ) (2)

MRAC takes the form of adjustment of some or all of the
controller coefficients so as to force the response of the Controller: ( ) (3)
resulting closed-loop control system to that of the reference
model. Therefore, the actual parameter values of the controlled
A. Time Delay Approximation
system do not really matter. Each components in the MRAC
structure is as described below:
A time delay element is represented by a non-linear transfer
function. Since many control design methods require linear
Reference Model: It is used to specify the ideal response of
systems, it is often necessary to approximate a time delay by a
the adaptive control system to external command. It should
linear transfer function. Therefore, a suitable approximation is
reflect the performance specifications in control tasks. The
needed to represent the time delay term in the system.
ideal behavior specified by the reference model should be
There are many approximation methods available but for
achievable for the adaptive control system.
this assignment, we have selected Padéapproximation for the
time delay approximation as shown below:
Controller: It is usually parameterized by a number of
adjustable parameters. In this paper two parameters θ1 and θ2
are used to define the controller law. (4)

Adaptation Mechanism: It is used to adjust the parameters (5)

in the control law. Adaptation law searches for the parameters
such that the response of the plant which should be same as
The above approximations will be used throughout this
the reference model. paper in designing the controller.
IV. GRADIENT METHOD (MIT RULE)
B. Gradient Approach Design
This rule is developed in Massachusetts Institute of
technology and is used to apply the MRAC approach to any The control objective is to adjust the controller parameters,
practical system. In this rule the cost function or loss function θ1 and θ2, so that e(t) is minimized. To do this, a cost function,
is defined as ( ) ⁄ where e is the error (y- ) and θ J(θ) is chosen and minimized.
is the adjustable parameter. We aim to minimize the error ( ) ( )

2
 MKEM1732: Assignment 2

Let and . Substitute into the Equation (3), * + (21)

we obtain the control law
The MRAC gradient approach is then simulated using
(7)
Mathlab Simulink block with the reference input of a square
wave signal with amplitude 1 and 3 with frequency of
After the time delay approximation using Eq.(4) and (5), 0.0033Hz as well as and Both the
rewriting the Eq.(1) and (2) as follows: output of the system responses (y and ) as well as ( and
) are displayed in section VII.
Plant: ( ) (8)
( )
C. Stability Approach Design
Ref. Model: ( ) (9) For the design using Lyapunov method, sequences of steps
have been established as shown below:
Rewriting the Eq.(8) in y and u,
Step 1:
( )
(10)
Derive a differential equation for error, into
Substituting the u from Eq.(7) into Eq.(10), we get ̇ ̈ containing the parameter,
From Eq.(11), replacing A= , B= , C= and
( ) , the equation now becomes,
( )
(11)
( )

̈ ( ) ̇ ( ) ̇ (22)
Similarly, for the reference model, we get

( ) ̈ ̇ ̇ (23)
(12)
Then, substituting Eq.(22) and (23) into ̈ ̈ ̈ , we get
Computing the error, ,
̈ ̇ ( ) ̇ ( )
( ) ( )
( )
(13) ( ) ̇ ( ) (24)
(( )

Thus, by taking the derivative of both and ,we have Step 2:

( )
(14) Find a suitable Lyapunov function, V (e, θ) - usually in a
( ) ( )
quadratic form (to ensure positive definiteness).
( )
( )
(15) The Lyapunov function, V ( ̇ ) is based on
( )
Eq.(24) above where V becomes,
In this case we need to do some approximation: i.e. perfect
model following, y = . Therefore, we then have, ̇

* + (16) Where is positive definite. The

derivative of V then becomes,

* + (17) ̇ ̇ ( ) ̇ ( ̇ ̇ ̇)
( ̇ ̇ ) ( ̇
and ̇ ̇) ( ̇ ̇ )

* + (18) where for stability ̇

* + (19) Step 3:

Derive an adaptation mechanism based on V(e,θ) such that e

Finally, we get goes to zero.
( )* + (20) ̇ ̇ ̇
̇ (25)
and

3
 MKEM1732: Assignment 2

̇
̇ ̇ (26)
̇ ̇
̇ ̇ (27)
̇
̇ ̇ (28)

Summing Eq.(25) and (26) we get

̇
̇ ̇( )

( ̇ )( ) Fig. 3: System Output ( ,y) with τ =1 and

Similarly, summing Eq.(27) and (28) we get

̇ ( ̇ ̇ ̇ )( )

( ̇ )( )

Thus, rewriting the equation for k and , we get

( ̇ ) ( ̇ )
Fig. 4: System Output ( ,y) with τ=2 and

( ̇ )

VII. RESULT AND DISCUSSION

A. Gradient Approach Result

The Simulink block diagram for simulation of chemical

plant using Gradient method is as shown in figure 2 below.

Fig. 5: System Output ( ,y) and with τ=1 and

Fig. 2: Simulink Block Diagram for plant simulation using Gradient method
Fig. 6: System Output ( ,y) with τ=1 and
The simulation is done with reference input of square
wave signal with amplitude of 1 with period of 300s and pulse From Fig.3 to 6, we can clearly see that the controller
width of 80%. The adaptation gain and is set to be 0.02 designed using MIT Rule is able to adapt to any changes made
and 0.0015 respectively.The analysis is done by fixing τ = 1 to the system by varying the time delay, τ and constant, .
and 2 while varying the and vice versa as highlighted in The controller has successfully forces the system output to
Fig.2 above.

4
 MKEM1732: Assignment 2

track the reference output closely as shown but with a 3.5

Lyapunov Method-Plant Output (Td=2, tau=2)
ym
small overshoot. 3
y

In addition, the controller parameters ( ) designed

2.5
using Gradient Method is also able to converge to a steady
2
value after some time but with small fluctuations as shown in

Amplitude
1.5
Fig.7 below.
1

0.5

-0.5
0 500 1000 1500 2000 2500 3000 3500 4000
Time,t(s)

Fig. 10: System Output ( ,y) with τ=2 and

3.5
Lyapunov Method-Plant Output (Td=2.5, tau=1)
ym
y
3

2.5

Amplitude
Fig. 7: Controller parameters ( ) with τ=1 and 1.5

0.5
B. Stability Approach Result
0

The Simulink block diagram for simulation of chemical -0.5

0 500 1000 1500 2000 2500 3000 3500 4000
Fig. 11: System Output ( ,y) and with τ=1 and
Time,t(s)
plant using Lyapunov method is as shown in figure 8 below.
3.5
Lyapunov Method-Plant Output (Td=3.0, tau=1)
ym
y
3

2.5
Amplitude

1.5

0.5

0
0 500 1000 1500 2000 2500 3000 3500 4000
Time,t(s)
Fig. 12: System Output ( ,y) with τ=1 and

Fig. 8: Simulink Block Diagram for plant simulation using Lyapunov method From Fig.9 to 12, we can clearly see that the controller
designed using Lyapunov is able to adapt to any changes
The simulation is done with reference input of square made to the system by varying the time delay and constant.
wave signal with amplitude of 3 with period of 1200s and The controller has successfully forces the system output to
pulse width of 80%. The adaptation gain and is set to be track the reference output perfectly as shown.
0.003 and 0.001 respectively.The analysis is done by fixing τ In addition, the controller parameters ( ) designed
= 1 and 2 while varying the and vice versa as highlighted using Lyapunov Method is also able to converge quickly to a
in Fig.8 above. steady value as shown in Fig.13 below.

Lyapunov Method-Theta(Td=2, tau=1)

Lyapunov Method-Plant Output (Td=2, tau=1) 0.6
3.5
ym 31
y 32
3 0.4

2.5 0.2

2
Amplitude

0
Amplitude

1.5
-0.2
1
-0.4
0.5
-0.6
0

-0.8
-0.5 0 500 1000 1500 2000 2500 3000 3500 4000
0 500 1000 1500 2000 2500 3000 3500 4000
Time,t(s)
Time,t(s)

Fig. 9: System Output ( ,y) with τ=1 and Fig. 13: Controller parameters ( ) with τ=1 and

5
 MKEM1732: Assignment 2

C. Comparison between Gradient and Stability Approach

This shows that when the system is injected with a
The MRAC designed using the Gradient Method does not reference signal of amplitude = 1, both controllers
guarantee stability to the resulting closed-loop system. In are able to give a stable output with slow transient
contrast, the Stability approach/ Lyapunov method has proven response at beginning.
to be more stable compared to MIT Rule as shown in Fig.14
below where the comparisons between their responses are However, when the amplitude is increased to 3
plotted. MIT rules does not give satisfactory result with a
high overshoot as compared to Lyapunov method.
This is shown in Fig. 14 earlier where the
comparisons between the two controllers are being
made. MIT rule shows high overshoot while
Lyapunov Method is still able to track perfectly.

In terms of electronics and control point of

view, an accumulation or saturation of error could
result in the control actuator which might
eventually spoil the actuator in long term operations.
Thus, it is concluded that controller design using
Gradient Method is said to be less stable as
compared to Lyapunov method as expected.
Fig. 14: Comparisons between performance of MIT and Lyapunov method.

iii) As shown in Fig.7 and 13 earlier, the controller

Based on the observations from above designs, we can
parameters and of Lyapunov converge quickly
deduce the differences of each design as below:
as compared to MIT rule with small fluctuations
which does not converge when amplitude of
i) The adjustable controller parameter for
reference signal is increased.
Lyapunov method are simpler as compared to
Gradient Method as summarized in table 1 below. VIII. CONCLUSION
Table 1: Comparison of adjustable parameters for both design
approaches. In conclusion, this paper has examined the comparisons in
Lyapunov Method Gradient Method performance between both Gradient and Stability approaches
for Model Reference Adaptive Control Design and it is proven
( )
[ ] that Lyapunov method is more preferable since it ensures a
stable closed-loop system while gradient approach doesn‟t
( )
[ ] take any consideration on the stability of the system.

ii) The performance of MRAC designs is also

depended on the amplitude of the reference input IX. REFERENCES
signal where increase in amplitude will affect the
result of the system. Simulation with amplitude of 1 [1] Parks, P. C. (1966). Liapunov redesign of model reference adaptive
control systems. Automatic Control, IEEE Transactions on, 11(3), 362-
with good response is as shown in Fig.15 below 367.
with output of MIT rules plotted against the [2] Feron, E., Apkarian, P., & Gahinet, P. (1996). Analysis and synthesis
Lyapunov Method. of robust control systems via parameter-dependent Lyapunov functions.
Automatic Control, IEEE Transactions on, 41(7), 1041-1046..
[3] Sapiee, M. R., Selamat, H., Noordin, A., & Jahari, A. N. (2008,
November). PI Controller Design Using Model Reference Adaptive
Control Approaches For A Chemical Process. In Proceedings of 2008
Student Conference on Research and Development (SCOReD 2008)
(pp. 155-1).
[4] Swarnkar, P., Jain, S., & Nema, R. K. (2010). Effect of adaptation gain
on system performance for model reference adaptive control scheme
using MIT rule. World Academy of science, engineering and
technology, 70, 621-626.
[5] Pankaj, S., Kumar, J. S., & Nema, R. K. (2011). Comparative analysis
of MIT rule and Lyapunov rule in model reference adaptive control
scheme. Innovative Systems Design and Engineering, 2(4), 154-162.
[6] Dr Hazlina Selamat, “Model Reference Adaptive Control (MRAC)”
MEM1732 Lecture Notes, Malaysia: Universiti Teknologi Malaysia,
2014.
Fig. 15: System Output with Amplitude =1.