You are on page 1of 6

19th Mediterranean Conference on Control and Automation TuCT3.

2
Aquis Corfu Holiday Palace, Corfu, Greece
June 20-23, 2011

A Simplified Control Method for Multivariable Stable Nonsquare


Systems with Multiple Time Delays
Cristina I. Pop, Robin De Keyser, and Clara Ionescu

Abstract—Time delays in multivariable nonsquare systems nonsquare form, with more inputs than outputs [2, 9, 10].
challenge the control action, firstly by limiting the degree of Such an example of nonsquare system with multiple time
freedom and secondly because of the interaction that exists delays is the Shell control problem [2, 9], with three
between various input-output channels. Different decoupling manipulated variables and two controlled variables. For
techniques associated with multivariable Smith Predictors have
these special types of processes, traditional control methods
been proposed for the control of such plants. The design of the
primary controller has also been intensively studied. The implied squaring down the system and designing the
method proposed in this paper is a straightforward alternative decentralized multivariable controllers. However, this step
to previously designed control structures for multivariable results in poor closed loop performance because of the
stable nonsquare time delay processes. The authors show that information neglected when squaring down [2].
in this way the task of designing the controller is much Seshagiri and Chidambaram [2] extend the MIMO Smith
simplified, while the overall performance of the closed loop
Predictor structures to nonsquare processes represented by
system remains the same. Further, the control design is suitable
for a general class of stable multivariable processes, rather first order transfer functions and time delays. The
than the limited applicability of the original method to decoupling is done only in steady state by using the pseudo-
processes that can be described by first order time delay inverse of the steady-state gain matrix; the final controller,
transfer functions. Setpoint tracking, disturbance rejection and consisting of a matrix of PIs, is then computed using the
robustness tests are presented in comparison with the original Davison method [12]. Robustness issues are usually tackled
method to prove the efficiency of the proposed simplified
by filter design methods. Chen, He and Qi [11], start from
method.
the previously designed MIMO Smith Predictor for the same
Keywords: Multivariable Smith Predictor, IMC controller, type of nonsquare stable processes, but claim that an IMC
robustness, stable nonsquare processes approach smoothes the design burden for the final PI matrix
of controllers and also leads to better robustness.
I. INTRODUCTION The method proposed in this paper is based on previous
control techniques designed for multivariable stable
T IME delays occur frequently in process control loops,
especially in chemical plants, such as distillation
columns, oil fractionators, reactors [1,2,3]. Most of these
nonsquare processes with multiple time delays [2, 11].
Particularly, the method detailed by Chen, He and Qi [11] is
modified, by simplifying the design burden of the controller
chemical plants are multivariable in nature, a fact that adds
without affecting the overall performance of the closed loop
to the complexity of controlling such processes [1].
control system. The proposed method offers similar overall
Although the problem of controlling SISO time delay
performance as in [11] regarding setpoint tracking,
processes has been long introduced [4], for MIMO time
disturbance rejection and robustness, with the advantage of a
delay processes the solutions proposed revolve around a
more straightforward design of the primary controller, as
multivariable Smith predictor [5, 6, 7].
well as a general applicability.
A key element in all these MIMO Smith Predictors is the
The paper is structured as follows. In Section II, the
decoupling of the process. For example, Wang, Zou and
original control method and the modified control method
Zhang [1], propose a technique to design the decoupling
proposed by the authors are presented. Section III presents
matrix of the system using a frequency domain approach.
the comparative simulation results between the original
Later, the internal model control (IMC) method has been
method and the method proposed in this paper. Simulations
used for the same decoupling matrix [8].
concerning setpoint tracking, disturbance rejection and
A great majority of chemical processes also exhibit a
robustness against modeling errors are presented. The final
section of the paper contains the concluding remarks.
Cristina I. Pop is with the Technical University of Cluj-Napoca,
Automation Department, Str. Memorandumului, no. 28, 400114 Cluj-
Napoca, Romania (phone: +40744-640-068; e-mail: Cristina.Pop@ II. SIMPLIFIED CONTROL METHOD USING SMITH DELAY
aut.utcluj.ro). COMPENSATOR
Robin De Keyser is with Ghent University, Department of Electrical
energy, Systems and Automation, Technologiepark, 913, 9052 Gent, A. Original Control Strategy [11]
Belgium (e-mail: Robain.DeKeyser@UGent.be).
Clara Ionescu is with Ghent University, Department of Electrical For a general process with „m‟ inputs and „n‟ outputs, the
energy, Systems and Automation, Technologiepark, 913, 9052 Gent, transfer function matrix is given as in [2, 11]:
Belgium (e-mail: ClaraMihaela.Ionescu@UGent.be).

978-1-4577-0123-8/11/$26.00 ©2011 IEEE 382


(6)
(1)

Approximation of each term can be done using


th
where gij represent transfer functions from the j input to the graphical identification methods or using genetic algorithms
ith output, is the corresponding time delay and s is the as in [11]. This approximation further facilitates the
Laplace variable. The model of the process is assumed to be controller design steps.
equal to the process transfer function matrix: Due to the static decoupling achieved in (5), the non-
diagonal terms in the decoupled process transfer
function matrix would be zero in steady state conditions;
(2) consequently, only the diagonal terms in (6) will be further
used in the design of the controller, with each diagonal term
corresponding to a certain process output.
A simple method to decouple the multivariable system is Consequently, the primary controller in the Smith delay
to use the Moore-Penrose pseudo-inverse [13] of the steady compensation structure in Fig. 1 a) is given by:
state gain matrix:
(7)

(3) with computed using the Internal Model Controller


(IMC) design method [11].
The design of the final controller is based on two steps:
The Moore-Penrose pseudo-inverse is then computed firstly an IMC controller is designed, and next the GC(s)
based on the following equation [13]: controller is obtained.
The design of the IMC controller is done using:
(4)
(8)
where is the pseudo-inverse, and (…)H is the Hermitian
matrix of the steady state matrix given in (3). where f(s) is the IMC filter and is the invertible part
To compute the controller, the next step is to decouple the of the element . The IMC filter is designed as:
process:
(9)
(5)
with n such that the final IMC controller is proper and
denotes the time constant of the filter associated to each
in which all elements are weighted sums of the original output.
transfer functions in (2). Due to the static Computing the IMC controller based on equation (8) and
decoupling, in steady state the transfer function matrix using (6), results in simple PI controllers:
GD(s=0) will be equal to the unit matrix. The next step is to
approximate the diagonal elements in the decoupled transfer
(10)
function of the process. Chen, He and Qi [11] assume that
the approximation of each element is done with simple first
order transfer functions [11]:

Fig. 1. Smith delay compensator for MIMO nonsquare processes [11]

978-1-4577-0123-8/11/$26.00 ©2011 IEEE 383


Fig. 2. Closed loop control scheme using delay compensator and IMC controller (see text for symbol explanation)

Since the terms are first order transfer functions, rather than directly in the controller structure, as Chen, He
the IMC filters in (9) are chosen to be of first order to make and Qi [11] suggest in equation (11). The modified closed
the final PI controllers proper. loop control structure is given in Fig. 2.
Next, the final controller is obtained based on (7), where The proposed scheme shown in Fig. 2 is obtained by
GC(s) denotes the diagonal matrix of PI controllers derived analytically transforming the control structure in Fig. 1. The
for each particular output: primary controller in Fig. 1 is decomposed in the PI
controllers Gdc(s) and the Moore-Penrose pseudo-inverse
(11) . Further, the Moore-Penrose pseudo-inverse is used as a
pre-compensator in series with the process - Gp(s), the model
The closed loop used is the Smith delay compensator - Gm(s) - and the fast model - , resulting in the block
given in Fig.1, in which the GC(s) controller is in fact the diagram given in Fig. 2.
one derived in (11) and GF(s) are feedback filters used to To increase the closed loop robustness, feedback filters are
increase robustness [11]. added, with the time constant taken as half of the largest
time delay corresponding to the output variable loop [2, 14].
B. Simplified Control Strategy
The closed loop control scheme is given in Fig. 2, where
The original control method proposed by Chen, He and Qi Gp(s) is the process, Gm(s) is the model of the process, , is
[11] describes the necessary steps required to compute a PI
the pre-compensator, GF(s) is the feedback filter, is
controller matrix for the control of nonsquare stable time
delay processes. Although the approach is simplified as the transfer function of all elements , is the
compared to [2] and provides for significantly better results, transfer function matrix without the time delays and
a first problem that arises is the limited applicability. is the process model without the time delays (the fast
Chen, He and Qi [11] assume that the decoupled process model as mentioned above).
diagonal elements can be approximated by first order
transfer functions with time delays, as given in (6) above. III. SIMULATION RESULTS
The PI controller matrix and the design steps described in To demonstrate the effectiveness of the proposed control
Section IIA remain valid as long as the assumption stands. In method, the shell control problem is considered, with the
different situations, the diagonal elements of the decoupling process transfer function matrix given by:
matrix may result in oscillatory behavior, thus the
approximation in (6) fails, as well as the design of the PI
controller matrix. (12)
To overcome this problem and also to simplify the design
of the final controller, this paper proposes a slightly different
approach in which the IMC controllers designed are used The process model is equal to the process transfer
directly as the final controllers. Thus, the approximation in function matrix, that is: . The steady state
(6) is no longer required and the final form of the controller gain matrix of the process model is:
is not restricted to the particular case of PI controllers.
The basic idea is that instead of computing a final
controller matrix consisting of PI regulators, the IMC (13)
controllers derived in (9) can be directly used in the closed
loop controller structure. To make this possible and maintain The Moore-Penrose Pseudo-inverse is given by:
the advantages of the Smith delay compensator in Fig. 1[11],
the decoupling of the process is achieved as in (5), with the
Moore-Penrose pseudo-inverse used as pre-compensator,

978-1-4577-0123-8/11/$26.00 ©2011 IEEE 384


(21)
(14)
(22)

The decoupled process model is given by: For the original control method, the closed loop in Fig. 1
is used [11], with the primary controller GC(s), derived in
(15) (20). For the alternative, simplified method proposed in this
paper the closed loop control scheme of Fig. 2 is used, where
the term IMC denotes the diagonal matrix of the two IMC
with
controllers derived in (16) and (17) respectively.
In the nominal case, servo and interaction responses of the
system for a unit step change in the reference of y1, are
presented in Fig. 3. Fig. 4 shows the disturbance rejection
results, considering input disturbances [2, 11] on the first
input – u1 - channel.

1
The approximations of the diagonal terms in the
decoupled process given in (15) are [11]: 0.8
y1

Servo and interaction responses


0.6

0.4

y2
To design the IMC controllers corresponding to 0.2

and to the filter time constants are chosen as


0
[11]: λ1=20 and λ2=15. A different choice of the filter time
constants would result in a different behavior. However,
-0.2
since it is the purpose of this paper to demonstrate that the Original method
Simplified method
simplified proposed control structure yields the same results -0.4
as the original control method, the filter time constants are 0 50 100 150 200 250
Time (s)
300 350 400 450 500

maintained according to the original choice [11]. Then,


Fig. 3. Servo and interaction responses for a unit step change of the y1
using (9), the IMC controllers are obtained as: reference

(16) Comparing the results from Fig. 3, Fig. 4 a) and b), we


conclude that – in nominal conditions - the simplified
(17) method proposed in this paper has exactly the same
properties as the original one and provides the same results.
Next, the individual PI controllers are derived based on However, to test the robustness of the proposed method,
[11]: we consider situations in which the gain factors are varied
with +20%, the time constants are varied -20% and finally
(18) the time delays are varied +20%. In all situations the
modeling error is considered for each gain, time constant or
(19) time delay associated to the corresponding output.
Fig. 5 depicts the results of servo and interaction
with the final controller matrix obtained based on (11): responses, considering a +20% gain estimation error, for a
step change in the y1 reference (Fig. 5a)) and in the y2
reference (Fig. 5b)).
Fig. 6 depicts the results for changes of -20% in the time
(20) constants of the process, for a step change in the reference
for y1 (Fig. 6a)) and in the y2 (Fig. 6b)).
From Fig. 5 and Fig. 6 we conclude again that the
proposed simplified method provides the same closed loop
To increase robustness, the feedback filters are chosen to performance as the original method.
be:

978-1-4577-0123-8/11/$26.00 ©2011 IEEE 385


y1 response to a unit step disturbance
4 1.2
Original method
3.5 Simplified method 1

y1
3 0.8
Servo and interaction responses

Servo and interaction responses


2.5 y1 0.6

2 0.4
y2
1.5 0.2

1 0

Original method
0.5 -0.2 Simplified method

0 -0.4
0 100 200 300 400 500 600 0 50 100 150 200 250 300 350 400
Time (s) Time (s)

a) a)
y2 response to a unit step disturbance
4 1.2
Original method
3.5 Simplified method
1

3
Servo and interaction responses

Servo and interaction responses


0.8
y2
2.5 y2
0.6
2

0.4
1.5
y1
0.2 Original method
1
Simplified method

0.5 0

0 -0.2
0 100 200 300 400 500 600 0 50 100 150 200 250 300 350 400
Time (s) Time (s)

b) b)
Fig. 4. Responses for a unit step disturbance acting on the u1 input channel Fig. 5. Servo and interaction responses for a +20% gain estimation error and
a) y1 response b) y2 response a unit step change in a) y1 reference b) y2 reference

Finally, the last case scenario presented to demonstrate the The method proposed in this paper has no limitations of
effectiveness of the proposed method implies an error of this kind; thus the final primary controller is not restricted to
+20% in the estimated value of the time delays associated to a PI form.
the corresponding output. Fig. 7 shows the results for a
reference step change in y1 and in y2 respectively.
1.2
From all the results presented in the above simulations, it
can be concluded that the IMC controllers together with the 1
modified control structure given in Fig. 2 lead to the same
results as the original control structure in Fig. 1. Hence, the 0.8
Servo and interaction responses

y1
simplified method proposed in this paper offers the same 0.6
closed loop performance as the original one, without the
further design steps needed to compute the final matrix of PI 0.4

controllers as suggested in the original control method. y2


0.2
Additionally, the proposed control structure in Fig. 2
eliminates some unnecessary steps in the design of the final 0

primary controller. In the original method, the authors


-0.2
presented a method to derive a matrix of PI controllers, Original method
Simplified method
assuming that the decoupled process can be approximated -0.4
0 50 100 150 200 250 300 350 400
using simple first order transfer functions and time delays as Time (s)
in (6). a)

978-1-4577-0123-8/11/$26.00 ©2011 IEEE 386


[7] N. F. Jerome, W. H. Ray, “High-performance multivariable control
1.2 strategies for systems having time delay”, AIChE Journal, vol. 32, no.
6, pp. 914–931, 1986
1
[8] Q.G. Wang, Y. Zhang, M.S. Chiu, “Decoupling internal model control
for multivariable systems with multiple time delays”, Chemical
Engineering Science, 57, pp. 115–124, 2002
Servo and interaction responses

0.8
[9] C. Valchos, D. Williams, J. B. Gomm, “Solution to the shell control
problem using genetically tuned PID controllers”, Control
0.6 y2 Engineering Practice, 10, pp. 151–163, 2002
[10] B. A. Ogunnaike, W. H. Ray, “Process dynamics, modeling and
0.4
control”, Oxford University Press, 1992
[11] J. Chen, Z.-F. He, X. Qi, “A New Control Method for MIMO First
y1
Order Time Delay Non-Square Systems”, Journal of Process Control,
0.2 Original method to appear in 2011
Simplified method [12] E. J. Davison, “Multivariable tuning regulators: The feed forward and
0 robust control of a general servomechanism problem” IEEE
Transactions on Automatic Control, 21(1), pp. 35–47, 1976
[13] S. Skogestad, I. Postlethwaite, Multivariable feedback control:
-0.2
0 50 100 150 200 250 300 350 400 Analysis and design, New York, John Wiley &Sons, 2005
Time (s) [14] J. E. Normey-Rico, C. Bordon, E. F. Camacho, “Improving the
b) robustness of dead time compensating PI controllers”, Control
Fig. 6. Servo and interaction responses for a -20% time constant estimation Engineering Practice, vol. 5, pp. 801-810, 1997
error and a unit step change in a) y1reference b) y2 reference
1.2
IV. CONCLUSIONS
1
This paper introduces a control method based on a general
multivariable Smith predictor structure, in which the primary 0.8
y1
controllers are derived based on the IMC approach. The Servo and interaction responses 0.6
proposed method is derived from a previous approach [11].
0.4
In our paper, we claim that the original version is
y2
unnecessarily complicated in designing the final MIMO 0.2

controller. 0
The method proposed in this paper is a more
straightforward, easy to design and to implement approach -0.2

than the original one. The control structure proposed can be -0.4 Original method

easily determined analytically based on the original version. Simplified method


-0.6
The simulations show that the simplified control method 0 100 200 300 400 500 600
Time (s)
proposed in this paper has no drawbacks when compared to
the original one, neither in terms of setpoint tracking, a)
disturbance rejection nor robustness. Additionally, the 1.2
method proposed in this paper is not restricted only to
processes that can be described by first order transfer 1

functions with time delays, as compared to the original


Servo and interaction responses

version. 0.8
y2

0.6
REFERENCES
[1] Q. G. Wang, B. Zou, Y. Zhang, “Decoupling Smith Predictor design 0.4
for multivariable systems with multiple time delays”, Chemical
y1
Engineering Research and Design, vol.78, pp. 565–572, 2000
[2] R. A., Seshagiri, M. Chidambaram, “Smith delay compensator for 0.2 Original method
multivariable non-square system with multiple time delays”, Simplified method
Computers and Chemical Engineering, 30(8), pp. 1243-1255, 2006 0
[3] J.E. Normey-Rico, E.F. Camacho, “Dead time compensators: a
survey”, Control Engineering Practice, vol. 16, pp. 407- 428, 2008
-0.2
[4] O. J. M. Smith, “Closer control of loops with dead time”, Chemical 0 100 200 300 400 500 600
Engineering Progress, 53, pp. 217-219, 1957 Time (s)

[5] G. Alevisakis, D. Seborg, “An extension of the Smith predictor b)


method to multivariable linear systems containing time delays”, Fig. 7. Servo and interaction responses for a +20% delay time estimation
International Journal of Control, 17, pp. 541–551, 1973 error and a unit step change in a) y1reference b) y2 reference
[6] B. A. Ogunnaike, W. H. Ray, “Multivariable controller design for
linear systems having multiple time delays”, AIChE Journal, vol. 25,
no. 6, pp. 1043–1057, 1979

978-1-4577-0123-8/11/$26.00 ©2011 IEEE 387

You might also like