Professional Documents
Culture Documents
Design of Controllers For Decoupled TITO Systems Using Different Decoupling Techniques
Design of Controllers For Decoupled TITO Systems Using Different Decoupling Techniques
Abstract—This paper deals with the design of controllers for settling time and the overshoot. The design of this controller
TITO decoupled systems. The main objective of these controllers for decoupled Multiple Input Multiple Output systems using
is to ensure some time response specifications such as the settling the simplified decoupling technique was proposed in [10] and
time and the overshoot for each output independently. The [11]. The controller synthesis is formulated as an optimization
controller parameters are obtained by solving a nonconvex problem which takes into consideration the desired closed-
optimization problem. In order to obtain the global solution and loop performances. The methodology of fixing the desired
then an optimal control law a global optimization method was closed loop characteristic equation was detailed in [12].
applied to resolve this problem. In order to decouple TITO Given that the controller parameters are determined by
processes, two different decoupling methods have been
minimizing a nonconvex optimization problem, the use of a
considered.
global optimization method is recommended. In this work, the
Keywords— time response specifications; TITO systems; Generalized Geometric Programming (GGP) method will be
decoupling systems; nonconvex optimization problem; global exploited in order to solve this problem. The main idea of this
optimization. method involves transforming a nonconvex optimization
problem to a convex one by means of variable
transformations.
I. INTRODUCTION
In the industrial field most systems have multiple inputs The outline of this paper is as follows. Section II presents
and multiple outputs [1]. Owing to the interactions between the decoupling techniques used in this work. Section III
inputs and outputs, multivariable systems present usually proposes the controllers design method. In order to show the
difficulties when designing controllers. Thus, in order to efficiency of the proposed controllers some simulation results
surmount this problem, control engineers had resort to are proposed in section VI.
decoupling system techniques. The study of these techniques
has attracted the attention of many researchers over the years II. DECOUPLING SYSTEM METHODS
[2], [3]. Since each decoupling technique has advantages and Let consider the following Two Input Two Output (TITO)
limitations, the choice of one of these methods is relatively process G(s) described by:
complicated. Ideal decoupling, which is not often
implemented, facilitates the focus on the controller transfer
matrix. Simplified decoupling is very used in practice. Its ªG (s) G12 ( s ) º
G ( s ) = « 11 . (1)
G 22 ( s ) »¼
main advantage lies in the simplicity of its elements. Inverted
decoupling is also used in practice. It combines the advantages ¬ G 21 ( s )
of both ideal and simplified techniques. In some works a The process G(s) is represented in Fig. 1. The elements
comparison between different decoupling techniques was G11(s), G12(s), G21(s) and G22(s) are supposed to be known.
performed. References [5] and [6] compared ideal and
From configuration represented in Fig.1, the system
simplified decoupling methods. They concluded that
outputs are given by:
simplified decoupling is more robust than ideal decoupling.
Both simplified and inverted decoupling techniques were
described in [6]. A comparative survey between simplified, Y1 ( s ) = G11 ( s )U 1 ( s ) + G12 ( s )U 2 ( s )
ideal and inverted decoupling was made in [7]. In this research ® . (2)
simplified and inverted decoupling methods will be used. In ¯Y2 ( s ) = G 21 ( s )U 1 ( s ) + G 22 ( s )U 2 ( s )
previous work [8], we have presented the design of a fixed
low-order controller for Single Input Single Output (SISO)
systems with some time response performances such as the
G12 ( s )G 21 ( s ) (9)
T2 ( s ) = G 22 ( s ) − .
G11 ( s )
The control signals are given by:
G12 ( s )
°U 1 ( s ) = V1 ( s ) − G ( s ) V 2 ( s )
° 11 (10)
® .
°U ( s ) = V ( s ) − 21 ( s ) V ( s )
G
°̄ 2 2
G 22 ( s )
1
B. Inverted Decoupling
Fig. 2 Block diagram of the simplified decoupling By considering this decoupling technique, the transfer
matrix T(s) will be given by:
The variables V1(s) and V2(s) represent the controller
outputs, U1(s) and U2(s) are the control signals and Y1(s) and
Y2(s) are the process outputs. ªG ( s ) 0 º (11)
T ( s ) = « 11 .
By using equations (1) and (3), equation (4) can be ¬ 0 G 22 ( s ) »¼
expressed by:
From expressions (5) and (11) we find the following
relation [7]:
1117
The objective, now, is the design of the controllers C1(s)
ª G11 (s)G22 ( s) −G12 ( s)G22 (s) º and C2(s) which will ensure the desired closed loop
« specifications.
ªU1 (s) º « G11G22 (s) − G12 (s)G21 (s) G11G22 (s) − G12 (s)G21 (s) »
«U ( s) » = « »
¬ 2 ¼ −G21 ( s)G11 (s) G11 (s)G22 (s) » (12) III. CONTROLLERS DESIGN METHOD
« G G (s) − G ( s)G (s) G11G22 (s) − G12 (s)G21 (s) »¼
¬ 11 22 12 21
A. Problem Statement
ªV1 (s) º
«V ( s) » As mentioned previously, a TITO process can be
¬ 2 ¼ represented by two SISO systems using a decoupling method.
A SISO process can be defined by a transfer function Ti(s) for
Then, the expressions of the control signals can be deduced i=1, 2 as shown in Fig. 4. The transfer function is given by:
and given by [7]:
N i (s)
Ti ( s ) =
G12 ( s ) Di ( s )
°U 1 ( s ) = V1 ( s ) − U 2 ( s ) G ( s ) (15)
° 11 (13) m m −1
n m s + n m −1 s + " + n0
® . = .
°U ( s ) = V ( s ) − U ( s ) 21 ( s )
G d l s l + d l −1 s l −1 + " + d 0
°̄ 2 2 1
G 22 ( s )
Hence, The controller transfer function is defined by:
U 1 ( s ) = V1 ( s ) + U 2 ( s ) D12 ( s ) (14)
® .
¯U 2 ( s ) = V 2 ( s ) + U 1 ( s ) D 21 ( s ) Ci ( s ) =
Bi ( s ) (16)
Ai ( s )
From equation (14) we can deduce the control system with where
the decoupling structure depicted in Fig. 3. This representation
was introduced by Wade and called “inverted decoupling”
[13]. We remind that the transfer functions of the decoupler, in ° Ai ( s ) = s t + at −1 s t −1 + " + a1 s + a 0
. (17)
the case of inverted decoupling, are the same as those used ® r r −1
°̄ Bi ( s ) = br s + br −1 s + " + b1 s + b0
with simplified decoupling.
For a fixed low order controller, the following relation
must be verified: r ≤ t < l − 1.
So as to obtain a null steady state error, a polynomial Fi(s)
is introduced:
Fi ( s ) = f q s q + f q −1 s q −1 + " + f1 s + f 0 . (18)
Fi ( s ) Bi ( s ) N i ( s ) (19)
H i (s) = .
Ai ( s ) Di ( s ) + Bi ( s ) N i ( s )
Fig. 3 Block diagram of the inverted decoupling
The decoupling methods applied to TITO systems allow us Then, the closed loop characteristic equation will be
to obtain the following configuration which is composed by expressed by:
two SISO systems as shown in Fig. 4. Each system is
constituted by a controller Ci(s) and a transfer function Ti(s) δ i ( s ) = Ai ( s ) Di ( s ) + Bi ( s ) N i ( s )
for i=1, 2. (20)
= δ n s n + δ n −1 s n −1 + " + δ 1 s + δ 0
where n = l + t .
1118
B. Controllers Design
( ) ( )
T
F3 = q − δ W q −δ
We note by x the following controller parameters vector:
and W is a weighting matrix.
It has been reported in [12] that the coefficients of lower
x = [ b0 a t −1 ] .
T
" br a0 " (21)
powers of s in the transfer function are the most associated to
the step response. Then, the weighting matrix elements must
We mention that each controller Ci(s) possess be selected while respecting this hypothesis.
r + t + 1 parameters. The controller parameters can be obtained by solving
The coefficient vectors of the closed-loop characteristic the following nonconvex optimization problem:
polynomial δ and the desired closed-loop polynomial δ are
given respectively by: min f ( x ). (26)
x
The resolution of this nonconvex optimization problem by
δ = [δ 0 δn ]
T
δ 1 " δ n −1 (22) a local approach may lead to a local solution. Consequently, in
T order to obtain an optimal control law, the use of a global
δ = ª¬δ 0 δ 1 " δ n −1 δ n º¼ . (23)
optimization method is suggested. In this work, the GGP
method is exploited. This method was explained in [10], [11],
The coefficient vector of the closed-loop characteristic [14]-[17].
polynomial δ can be expressed in function of x by: IV. SIMULATIONS RESULTS
1119
Setpoints and outputs
3.5
° A(s) = s + a1s + a0
2
(29) 3
Setpoint 1
® . Output 1
°̄ B(s) = b1s + b0
Setpoint 2
2.5 Output 2
Amplitude
2
f ( X ) = c1 x1 x2 + c2 x1x3 + c3 x1x4 + c4 x2 x3 + c5 x2 x4 + c6 x3 x4 0
0 10 20 30 40 50 60 70
Time (s)
2 2 2 2
+c x + c x + c x + c x − c11x1 − c12 x2 − c13 x3 (30)
7 1 8 2 9 3 10 4 Fig. 5 Evolution of set points and outputs using the inverted decoupling
−c14 x4 + c15 Control signals
Amplitude
these six terms have positively signed terms, and then the 0.5
xi = exp( yi ). (31)
A new objective function f(X,Y) which convex is obtained: -0.5
0 10 20 30 40 50 60 70
Time (s)
f ( X ,Y ) = c1 exp( y1 )exp( y2 ) + c2 exp( y1 )exp( y3 ) + c3 exp( y1 )exp( y4 ) Fig. 6 Evolution of control signals using the inverted decoupling
1120
96.016s3 + 272.119s2 +180.423s + 36.408 It is noted that the resulting transfer matrix of the inverted
C3 (s) = − decoupling presents two transfer functions with a low degree
s5 +14.629s4 + 83.022s3 +196.464s2 + 208.364s +111.114 thus the designed controllers have, also, lower degrees.
(38)
V. CONCLUSION
Assume that our purpose is the synthesis of a controller In this work, the design of controller for decoupled TITO
while ensuring the same time specifications for the second systems was presented. The main objective of this controller is
output as in the case of simplified decoupling. to guarantee some time response specifications. The controller
The minimization of the objective function by means of parameters are obtained by solving nonconvex optimization
the GGP method leads to the following controller: problem by using a global optimization method. Two
decoupling techniques have been applied to decouple
11.98s3 + 88.717s2 + 72.67s + 20.255 multivariable system. Simulation results show that it is possible
C4 (s) = − to use one of these methods in the design of the proposed
s5 +10.428s4 + 54.42s3 +102.119s2 +101.851s + 39.016
(39) controller.
A practical implementation of the proposed controllers will
be the subject of future research.
The evolutions of the output signals and the control signals
are plotted in Fig. 7 and Fig. 8 respectively. REFERENCES
Setpoints and outputs
3.5
Setpoint 1
[1] S. Skogestad and I. Postlethwaite, Multivariable Feedback Control:
3 Output 1 Analysis and Design, John Wiley and Sons, New York, 2005.
2.5
Setpoint 2 [2] B. Ogunnaike and W. Harmor, Process Dynamics, Modelling and
Output 2
Control. Oxford University Press, 1994.
Amplitude
2
[3] Q.W. Wang, Decoupling Control, Lecture Notes in Control and
1.5 Information Sciences, Springer-Verlag, 2003.
1
[4] W. L. Luyben, “Distillation decoupling,” AIChE Journal, vol. 16, pp.
198–203, 1970.
0.5
[5] D.E. Seborg, F. E. Thomas and A. M. Duncan, Process Dynamics and
0 Control. John Wiley and Sons, New York, 2010.
0 10 20 30 40 50 60 70
Time (s) [6] F.G. Shinskey, Process Control Systems: Application, Design, and
Adjustment, McGraw-Hill, New York, 1988.
Fig. 7 Evolution of set points and outputs using the simplified decoupling
[7] E. Gagnon, A. Pomerleau and A. Desbiens, “Simplified, ideal or
inverted decoupling ?,” ISA Transactions, vol. 37, pp. 265-276, 1998.
[8] M. Ben Hariz, F. Bouani and M. Ksouri, “Robust controller for
Control signals
2 uncertain parameters systems,” ISA Transactions, vol. 51, pp. 632-640,
Control 1 2012.
Control 2
1.5 [9] L. Jin and Kim Y.C. “Fixed, low-order controller design with time
response specifications using non-convex optimization,” ISA
Transactions, vol. 47, pp. 429–438, 2008.
Amplitude
1
[10] M. Ben Hariz, W. Chagra and F. Bouani, “Controllers design for MIMO
0.5
systems with time response specifications,” International Conference on
Control, Decision and Information Technologies (CoDIT), 2013 pp.
573-578.
0
[11] M. Ben Hariz, W. Chagra and F. Bouani, “Synthesis of Controllers for
MIMO Systems with Time Response Specifications,” International
-0.5
0 10 20 30 40 50 60 70 Journal of System Dynamics Applications, vol. 3, pp. 25-52, 2014.
Time (s)
[12] Y.C. Kim, L.H. Keel and S.P. Bhattacharyya, “Transient response
Fig. 8 Evolution of control signals using the simplified decoupling control via characteristic ratio assignment,” IEEE Trans. on Automatic
Control, vol. 48, pp. 2238–2244, 2003.
[13] Wade H.L. “Inverted decoupling: a neglected technique,” ISA
From Fig. 7, we observe that the designed controllers meet Transactions, vol. 36, pp. 3–10, 1998.
the required time specifications (an overshoot about 2% and a [14] C. Maranas and C. Floudas, “Global optimization in generalized
geometric programming,” Computers and Chemical Engineering, vol.
settling time less or equals to 10s for the first output and an 21, pp. 351-369, 1997.
overshoot about 5% and a settling time less or equals to 8s for [15] J. Tsai, M. Lin and Y. Hu, “On generalized geometric programming
the second output). problems with non positive variables,” European Journal of Operational
It is noted that, in the case of simplified decoupling, the Research, vol. 178, pp. 10-19, 2007.
output signals present a remarkable interaction when changing [16] R. Porn, K. Bjork, and T. Westerlund, “Global solution of optimization
set points compared to those obtained by applying the inverted problems with signomial parts,” Discrete optimization, vol. 5, pp. 108-
decoupling. It is observed, also, that control signals obtained in 120, 2007.
the case of simplified decoupling present peaks with high [17] J. Tsai, “Treating free variables in generalized geometric programming
amplitude when varying set points in comparison with those problems,” Computers and Chemical Engineering, vol. 33, pp. 239-243,
2009.
gotten by applying the inverted decoupling.
1121