You are on page 1of 13

ISA 2002 Technical Conference Paper

ISA 2002
21-24 October
McCormick Place South
Chicago, IL

Copyright © 2002 by ISA —The Instrumentation, Systems, and Automation Society. All rights
reserved. Not for resale. Produced in the United States of America.

ISA
67 Alexander Drive
P.O. Box 12277
Research Triangle Park, North Carolina 27709
Phone: (919) 549-8411
Fax: (919) 549-8288
Email: info@isa.org
ISANetwork: http://www.isa.org
Fuzzy Logic Control of a Multivariable Coupled Process
Dr. José de Jesús Rodríguez, M.C. Rafael Ruiz León
Instituto Tecnológico y de Estudios Superiores de Monterrey
Programa de Maestría en Automatización
Sucursal de Correos J, Monterrey, N.L. 64849 México
Tel.:(8) 358-2000 Ext. 5486 FAX: (8)328-4077
email: jjrodrig@campus.mty.itesm.mx

Abstract.
This paper reports an implementation of a Fuzzy Logic controller reported in [11]
based on error and the changes of the error. This Coupled Multivariable process shows a
coupling of 2X2. This controller is compared against a PI controller adjusted by a
heuristic criterion and applied to a coupled process that is decoupled by a Partial
Decoupler [7], [1]. The design of the Partial Decoupler is based in the same plant used in
[11].

I. INTRODUCTION.
Our objective is to evaluate the performance of a Fuzzy Control algorithm,
comparing it with the performance of a Partial process Decoupler plus PI controller of a
2X2 variable coupled process. Prior works ([4], [3]) points out, as general benefits of
fuzzy control, the smooth performance during the control response and the time saved, by
the fact that no process identification is needed, since the fuzzy logic control does not
need the exact mathematic model of the controlled plant. As a disadvantage, it requires a
considerable amount of time for the tuning of the fuzzy controller since there is no
methodology for this task. This can be easily solved having an experienced operator and
utilizing his/her knowledge of the process in the fuzzy control algorithm. The conclusion
of the bibliographical research of the multivariable process fuzzy controller [3], [4], [7],
[9] is the remarkable importance of research for an application of a fuzzy strategy over a
2X2 multivariable coupled process and its comparison with the control strategies of
Relative gains and the Partial Decoupler, both controlled by PI tuned by a proposed
manual strategy.

The Performance of the strategies is compared using the following criteria:

· Transient performance.
· The achieved decoupling of the strategy
· Reset Time
· Control Effort.
· Robustness.

II. PROCESS USED FOR SIMULATION.


For the control algorithm application was choose a 2 in – 2 out process used in
[11]. The process is defined by the below equations:
é 12.8e - s 18.9e -4 s ù
ê ú
G ( s ) = ê16.7 s-+7 s1 21s +-31s ú (1)
ê 6 .6 e 19.4e ú
êë10.9 s + 1 14.4 s + 1 úû

The block diagram of the system is showed in the figure 1

Fig. 1.- 2 X 2 variables process

III. PROCESS DECOUPLER.


Equations for The Partial Process Decoupler for a 2X2 and for the coupled process are
showed in Equations 2 :

G12(z) G (z)
D12(z) = - D21(z) = - 21 (2)
G11(z) G22(z)

12.8e - s 18.9e -4 s
G11 ( s ) = G12 ( s ) =
16.7 s + 1 21s + 1

6.6e -7 s 19.4e -3s


G21 ( s ) = G22 ( s ) =
10.9 s + 1 14.4 s + 1

The above equations need to be passed to the discrete form using the Z transform.
For such task we choose a sample time of T = 3. Thus, for the modified Z transform we
have:
q
m = 1-
T

Now, we can calculate the Decoupler equations. Developing the operations we reach:

1.1884( z - 0.83557)( z + 0.4654)


D12 ( z) = - (3)
z( z - 0.86688)( z + 0.45684)

0.30325( z + 0.43526)(z - 0.81194)


D21( z) = - (4)
z 2 ( z - 0.7594)
Once calculated the partial decoupler equations we proceed to apply the control.
For such objective we perform a unit step test for each of the inputs in open loop over the
decoupler – process system, as is showed in the Figure 2, where the unit step is applied in
the u input. The results of those tests are showed in the Figs 3 and 4, where we can see
the performance of the decoupler.

Fig. 2 Block Diagram Partial Decoupler - System

7
6
5
Plant Response

4
3 Output 1
2
Output 2
1
0
-1
0 10 20 30 40 50 60 70 80 90 100

Time(sec)

Fig. 3 Unit step test applied to the input 1 with Partial Decoupler

10
9
8
7
Plant Response

6
5
4
3 Output 1
2 Output 2
1
0
-1
0 10 20 30 40 50 60 70 80 90 100
Time (sec)
Fig. 4 Unit step test applied to the input 2 with Partial Decoupler

An identification of first order with dead time was done based on the results of applying
the step test on the inputs and this lead us to the following results:

Salida 1 Salida 2
K 6.4 9.7
qo 1 3.45
t 7.78 5.55
Where K is the output gain, qo is the dead time and t is the time constant for the identified
process. Once those parameters are calculated is possible to figure out the PI’s that will
be used for control the process.
With the identification of the process is possible to do the tuning of the controllers via
the different techniques offered by the integrals criteria or using the genetic algorithms.
In an attempt of improve the tuning and the performance of the controllers a manual
tuning was done looking for a methodology for do such task. The following is the used
methodology:

Heuristic Method for tuning the PI Controller.

· Apply a unit step to both inputs, with the controllers just having the proportional gain.
The proportional gain constant is incremented from a very small value and is
increased depending upon the plant response. The proportional constant is tuned such
way both outputs do not present oscillations in their response. Having just the
controller in proportional action will result in a steady stay error from the reference,
but the important fact is to have a stable response of the plant. For the preset case of
study (2X2 process) sometimes can present difficulties to move the proportional
constants and have a stable response due to the coupling, the type and the percentage
of the coupling.
· In order to eliminate the steady stay error of the plant we aggregate the integral gain,
and in the same way that we did for the proportional gain, we start with a very small
value (It depends of the system constants and gains. If the plant has big gains the
integral effect should be a small number, i.e. if the plant gain is of the order of 100,
the integral constants should start around 1/100) and both integral constants are
increases until we eliminate the error from the reference.
· The increments for the integral and proportional actions is not at the same time, is
required to watch out for the system response in order to figure out where the system
requires more effect from the constants, i.e. Increase the constants such way both
plant responses have approximately the same steady stay error form the reference.

Note: This tuning even though has very close similarities to the Ziegler-Nichols
tuning method, the above steps were applied and developed during the simulations
successfully, base on a “try and error” methodology and can be take as a variation of the
well know method ZN. This manual tuning gave us a much better results during the
tuning of the PI controllers.

The tuning made base on the proposed methodology gave the following
parameter:

Kc Kc/Ti
Controlad
or
PI 1 0.13 0.05
PI 2 0.07 0.016
The performance of the tuned controllers on this way is showed in the Figs 5 and
6 that is an acceptable performance.

1.2

0.8
Plant Response

0.6
Output 1
0.4
Output 2
0.2

-0.2
0 10 20 30 40 50 60 70 80 90 100
Time (sec)

Fig. 5 Adjusted Controllers performance (input 1). Process Decoupler.

1.2
1
Plant Response

0.8
0.6
Salida 1
0.4
Salida 2
0.2
0
-0.2
0 20 40 60 80 100
Time (sec)

Fig. 6 Adjusted Controllers performance (input 2). Process Decoupler.

The block diagram of the simulation, done using the MatrixX software is showed
in the figure 7.

Fig. 7 Blocks from the System Build for the Process Decoupler
IV. FUZZY CONTROLLER.
The fuzzy controller that is used for the comparison with the process decoupler is
developed in [11]. Is a controller with a knowledge base of 16 rules. The membership
functions for the inputs and the outputs are showed in the figures 8 and 9. The reader may
note that the membership functions, for the inputs, are basically binary. In this case,
having two inputs with two possible memberships, it ends up to 16 rules. The rules for
the controller are showed in the table I. The applied tuning in [11] for the controller is as
below:
The normalizations that had the best performance of the fuzzy controller are:
· e1 and e2: both were from –0.35 to 0.35
· De1: from –0.041 to 0.041
· De2: from –0.07 to 0.07

The adjustments of the normalizations for the manipulations increment changes are:
· For the manipulation num. 1 the increments are: from –0.149 to 0.149
· For the manipulation num. 2 the increments are: from –0.075 to 0.075

Fig. 8 Member Functions for the manipulation’s increments 1 and 2.

Fig. 9 Member Functions for the Error and the Error Change.
TABLE I
CONTROL ACTION FOR EACH OF THE MANIPULATIONS.
e1 De1 e2 De2 DX1 DX2
N N N N AC1N1 AC2N1
N N N P AC1N2 AC2P1
N N P N AC1N2 AC2P1
N N P P AC1N3 AC2P3
N P N N AC1P1 AC2N2
N P N P AC1C AC2C
N P P N AC1C AC2C
N P P P AC1N1 AC2P2
P N N N AC1P1 AC2N2
P N N P AC1C AC2C
P N P N AC1C AC2C
P N P P AC1N1 AC2P2
P P N N AC1P3 AC2N3
P P N P AC1P2 AC2N1
P P P N AC1P2 AC2N1
P P P P AC1P1 AC2P1

V. Robustness test.
The robustness test was done based on changes in the plant constants. Such
changes were applied incrementing every parameter a 50%, from its original value,
turning the plant as is showed in equation 5:

é 19.2e -1.5 s 28.35e -6 s ù


ê 25.05s + 1 31.5s + 1 ú
ê ú
G( s) = ê ú (5)
ê 9.9e -10.5 s 29.1e - 4.5 s
ú
ê ú
ë 16.35s + 1 21.6s + 1 û

The Partial Decoupler was built based on the original system’s dynamic given by
equations 2 and is expected that its decoupling performance become less effective under
such plant changes. The results of the unitary step tests done over the Partial Decoupler –
Plant in open loop are showed in the figures 10 and 11, which show the diminished
performance of the Partial Decoupler on the modified plant, these results may be
compared to the ones showed in Figs. 3 and 4.

10
8
Plant Response

6
4 Output 1
2
Output 2
0
-2
0 20 40 60 80 100 120 140
T ime (sec)
Fig. 10 Unit step test applied to system input 1 Partial Decoupler-Plant in
open loop
16
14
12

Plant Response
10
8 Output 1
6
4 Output 2
2
0
-2
-4
0 20 40 60 80 100 120 140
Time (sec)

Fig. 11 Unit step test applied to system input 2 Partial Decoupler-Plant in open
loop

The comparison, in performance was made between the PI Controller (using the
manual tuning) controlling the Decoupled Process and the Multivariable Fuzzy Controller
with 16 rules. Comparisons were made, way of simulation using the MatrixX, with the
original parameters for the plant as showed in equations 2 and with the 50% change with
a plant as showed in equations 5.

1.4

1.2

0.8 Output 1 P.D.


Plant Output

0.6 Output 2 P.D.


0.4 Output 1 FUZZY
0.2
Output 2 FUZZY
0

- 0.2

- 0.4
0 50 100 150 200 250

T ime (sec)

Fig. 12 Robust test: Unit step test applied to plant’s input 1

1.4
1.2
1 Output 1 P.D.
Plant Output

0.8
Output 2 P.D.
0.6
Output 1 FUZZY
0.4
Output 2 FUZZY
0.2
0
-0.2
0 50 100 150 200 250
Time (sec)

Fig. 13 Robust test: Unit step test applied to plant’s input 2


In the Figs. 12 and 13 is showed the performance of both strategies controlling the
modified plant. The graphic’s nomenclature is:
· Output 1 D.P. Plant’s Output 1 using PI controlling the Decoupled Process.
· Output 2 D.P. Plant’s Output 2 using Partial Decoupler.
· Output 1 FUZZY Plant’s Output 1 using 16 Rules Fuzzy Controller.
· Output 2 FUZZY Plant’s Output 1 using 16 Rules Fuzzy Controller.

We may do the followings remarks from the figures 12 and 13 :

Robust Tests applying the unitary step in the plant’s input 1:


· The answer of the 16 Rule Fuzzy Controller is smoother and the oscillations are also
smoother than the PI controlling the Decoupled Process.
· The over-impulse is around 23.58% for the PI controlling the Decoupled Process and
8.44% for the 16 Rule Fuzzy Controller.
· The decoupling oscillations during the transitory are less violent for the 16 Rule
Fuzzy Controller.
· The maximum coupling for the PI controlling the Decoupled Process was 37.21% and
29% for the 16 Rule Fuzzy Controller.

Note:
The results from the unit step test applied to the original (unmodified plant)
decoupled process are not showed, but the achieved decoupling of the output number 1
over the number 2 was 10.7% while the decoupling achieved for the modified plant @
50% was 34.8%

Robust Test applying the unitary step in the plant’s input 2:


· The 16 Rule Fuzzy Controller showed a smoother answer to the input and both
showed oscillations.
· For the PI controlling the Decoupled Process the over-impulse is 37.03% and 19.3%
for the 16 Rule Fuzzy Controller.
· The decoupling oscillations during the transitory for the PI controlling the Decoupled
Process were smaller than the 16 Rule Fuzzy Controller, but the oscillations for the
Fuzzy Controller were less violent.
· The maximum coupling showed by the PI controlling the Decoupled Process was
12% and 21.93% for the 16 Rule Fuzzy Controller.

Note:
The results from the unit step test applied to the original (unmodified plant)
decoupled process are not showed, but the achieved decoupling of the output number 2
over the number 1 was 1.34% while the decoupling achieved for the modified plant @
50% was 15.86%

VI. CONTROL EFFORT COMPARISON.


In the Figures 14 and 15 we can see the manipulations comparison between the
Fuzzy and the PI controlling the Decoupled Process. The graphic’s nomenclature is:
Man. 1 D.P. PI Controller’s Manipulation for the Plant’s output 1 using the PI
controlling the Decoupled Process
Man. 2 D.P. PI Controller’s Manipulation for the Plant’s output 2 using the PI
controlling the Decoupled Process
Man 1 FUZZY Manipulation for the output 1 of the 16 Rule fuzzy Controller.
Man 2 FUZZY Manipulation for the output 2 of the 16 Rule fuzzy Controller.

0.25

0.2
Manipulation Magnitude

0.15
Man. 1 P.D.
0.1
Man. 2 P.D.
0.05
Man. 1 FUZZY
0 Man. 2 FUZZY
-0.05

-0.1
0 10 20 30 40 50 60 70 80 90 100
Time (sec)

Fig. 14 Control Manipulations from unit test step applied to plant’s input 1.
0.15
Manipulations Magnitude

0.1
0.05 Man. 1 P.D.
0 Man. 2 P.D.
-0.05 Man. 1 FUZZY
-0.1 Man. 2 FUZZY
-0.15
-0.2
0 10 20 30 40 50 60 70 80 90 100
Time (sec)

Fig. 15 Control Manipulations from unit test step applied to plant’s input 2.

VII. Results and Conclusions.

Table II shows a full numeric comparison between the two controllers:

TABLE II
COMPARISON TABLE FOR THE DIFFEREN PLANT TUNING

PI Fuzzy
D.P. 16 R.
TS1 10 36
D Ref. 1 % ST1 4.8 0.67
Ie2 0.55 2.74
% A2 3.96 14.8
TS2 24 27
D Ref. 2 % ST2 4.47 4.45
Ie1 0.34 2.42
% A1 2.62 7.6
Table II Nomenclature:
· D Ref.1 Step tests to the input 1 keeping reference input 2 at zero.
· D Ref.2 Step tests to the input 2 keeping reference input 1 at zero.
· TS Establishment time @ 5%.
· %ST Percentage of over-shout showed by the output response.
· Ie Output Error Integration.
· %A Maximum percentage of coupling showed during the transient response.

As we can seen in the table II the reset times are shorter for the PI controlling the
Decoupled Process but the oscillations are not present for the Fuzzy controller.
The Control effort of the Fuzzy Controller is smoother than the PI controlling the
Decoupled Process.
Even if the reset times are around the same order, the oscillations for the Fuzzy
Controller are almost null while is not the case for the PI controlling the Decoupled
Process.
In the research of a tuning strategy for the Fuzzy controller, the intention was to make the
controller to be a Decoupler and at the same time to do the control itself, but for such task
the PI controlling the Decoupled Process using the controllers tuned manually showed the
best behavior.
The proposed methodology for manual tuning, even if it is intuitive, provided a procedure
to achieve a satisfactory tuning. This was tested against other techniques, which may be
found in [1], with better results.
The decoupling obtained, by the fuzzy controller, in steady state is 100% but during the
transitory the decoupling showed a deviation of about 20% on the other variable.
The robustness tests run with the 50% changes to the plant parameters showed that the
fuzzy controller is more robust than the PI controlling the Decoupled Process because the
oscillations were smaller.
One final conclusion the authors wish to point out is: The 16 rules Fuzzy controller
developed showed good results against the PI controlling the Decoupled Process,
however, this fuzzy controller was not as good as far as decoupling is concern. Further
work is needed to by using other Fuzzy control strategies an compare its performance to
find out which is best for the application on the control of multivariable coupled
processes.
BIBLIOGRAPHY

[1] Deshpande, P. B. & Ash, R. H. Computer Proces Control. Instrument Society of


America,
2da. edición, 1988.
[2] Rodríguez , J. J. Material de apoyo para el curso y el laboratorio de
instrumentación y
control. ITESM, Departamento de Térmica, Fluidos y Control, 1985.
[3] Viljamaa, P. & Koivo, H. N. Tuning of multivariable Fuzzy Logic Controler.
Tampere
University of Technology, Department of Electrica. Engineering, Control
Engineering
Laboratory, 1993.
[4] Boverie, S. & Demaya B. & Titli A. Fuzzy Logic Control compared with other
automatic
control aproaches. Laboratoire MIRGAS, France, 1991.
[5] Chalmers, S. J. A design methodology for Fuzzy Logic Controllers. School of
Electronic and
Electrical Engineering, The Robert Gordon University, Aberdeen, 1994.
[6] Albertos, P. & Martínez M. Fuzzy Controllers Design: a methodology. DISCA,
Universidad
Politécnica de Valencia, 1993.
[7] R.K. Wood & M.W. Berry. Terminal composition control of a binary distillation
column.
Chemical Engineering Science, 1973, Vol.28,pp. 1707-1717.
[8] Aguiñaga Márquez Esaú. Optimización de controladores PID en tiempo discreto
mediante
Algoritmos Genéticos. Tesis de Maestría, ITESM, Monterrey, N.L. México, Enero
1997.
[9] Hanger Hernández Lucía Elizabeth. Diseño de un controlador basado en Lógica
Difusa
para un proceso multivariable. Tesis de Maestría, ITESM, Monterrey, N.L. México,
Mayo
1994.
[10] Driankov, D., H. Hellendoorn, M. Reinfrank, An introduction to fuzzy Control.
Springer-
Verlag, Berlin Heidelberg, 1993.
[11] Ruiz L. R. Diseño de un Controlador Difuso Multivariable en Base al
Error(e) y el Cambio del Error(De). Tesis de Maestría en ciencias con especialidad
en Automatización, ITESM Campus Monterrey. 1997

You might also like