Professional Documents
Culture Documents
www.elsevier.com/locate/jprocont
a
School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore 639798, Singapore
b
School of Control Science and Engineering, Shandong University, Jinan, Shandong 250061, PR China
Received 21 January 2006; received in revised form 17 April 2006; accepted 24 April 2006
Abstract
In terms of relative gain and relative frequency, the effective transfer function for independent controller design for multi-input multi-
output processes is provided in this paper. Differing from existing equivalent transfer functions, the proposed effective transfer function
provides both gain and phase information for decentralized controller design in a simple and straightforward manner. The interaction
effects for a particular loop from all other closed loops are directly incorporated into the effective transfer functions in four ways. Con-
sequently, the decentralized controllers can be independently designed by employing the single loop tuning techniques. This design
method is simple, straightforward, easy to understand and implement by field engineers. Several multivariable industrial processes with
different interaction modes are employed to demonstrate the effectiveness and simplicity of the method.
2006 Elsevier Ltd. All rights reserved.
Keywords: Decentralized control; Independent design; PID; Effective relative gain array; Relative frequency array; Effective transfer function; Gain and
phase margins
0959-1524/$ - see front matter 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jprocont.2006.04.001
774 Q. Xiong, W.-J. Cai / Journal of Process Control 16 (2006) 773–784
In recent years, Wang et al. [10] used a modified Ziegler– outputs, G(s) and Gc(s) are process transfer function matrix
Nichols method to formulate a set of equations to solve for and decentralized controller matrix with compatible
the parameters of individual controllers, in which the loop dimensions, expressed by
interactions are taken into account. However, the existence 2 3
of the solutions is not guaranteed. In addition, the exten- g11 ðsÞ g12 ðsÞ g1n ðsÞ
sion for higher dimensional systems seems difficult because 6 g ðsÞ g ðsÞ g2n ðsÞ 7
6 21 22 7
of the complicated and nonlinear computation. Another GðsÞ ¼ 6 7
4 5
more direct approach to deal with this problem is to build
the equivalent processes for each loop [11]. However, the gn1 ðsÞ gn2 ðsÞ gnn ðsÞ
knowledge of other controllers is still required. One possi-
and
ble solution is based on the assumption of ideal or infinite
bandwidth control [12]. Nevertheless, the ideal control does 2 3
gc1 ðsÞ 0 0
not exist in reality. This may result in unsatisfactory con- 6 7
trol system performance. Huang et al. [13] proposed a 6 0 gc2 ðsÞ 0 7
Gc ðsÞ ¼ 6
6
7;
new independent design approach by deriving controller 4 7
5
independent equivalent open-loop transfer functions. If
0 0 gcn ðsÞ
the open loop transfer functions have integration modes,
a simplified complementary sensitive function for each loop
respectively.
is attained by IAE optimal index. Based on these approxi-
Let
mated transfer functions, the decentralized controller is
tuned by a model-based method. The results show that this gij ðjxÞ ¼ k ij g0ij ðjxÞ;
approach is simple and effective for low dimensional multi-
variable processes. For high dimensional processes this where kij and g0ij ðjxÞ are steady state gain and normalized
design has to be more conservative due to the inevitable transfer function of gij(jx), i.e., g0ij ð0Þ ¼ 1, respectively. The
modeling errors encountered in formulation. interaction among individual loop is described by ERGA,
In this paper, a novel equivalent transfer function in the main result of ERGA is summarized as follows [14].
terms of effective relative gain array (ERGA) for multivar- Define eij of a particular transfer function as
Z xc;ij
iable processes is proposed. By considering four combina-
tion modes of gain and phase changes for a particular eij ¼ k ij jg0ij ðjxÞj dx;
0
loop when other loops are closed, this equivalent transfer
function can effectively approximate the dynamic interac- where xc,ij for i, j = 1, 2, . . . , n are the critical frequency of
tions among loops. Consequently, the design of decentral- the transfer function gij(jx) and j•j is the absolute value
ized controller for MIMO processes can be converted to of •. In order to calculate eij, the critical frequency can
the design of single loop controllers. The method is simple, be defined in two ways:
straightforward, easy to understand and implement. Sev-
eral multivariable industrial processes with different inter- 1. xc,ij = xB,ij, where xB,ij for i, j = 1, 2, . . . , n is the band-
action characteristics are employed to demonstrate the width of the transfer function g0ij ðjxÞ and determined
effectiveness and simplicity of the design method compared by the frequency where the magnitude plot of frequency
with the existing methods. response reduced to 0.707 time, i.e.,
jgij ðjxB;ij Þj ¼ 0:707jgij ð0Þj:
2. Preliminaries
2. xc,ij = xu,ij, where xu,ij for i, j = 1, 2, . . . , n is the ultimate
Consider an open loop stable multivariable system with frequency of the transfer function g0ij ðjxÞ and deter-
n inputs and n outputs as shown in Fig. 1, where ri, i = mined by the frequency where the phase plot of fre-
1, 2, . . . , n, are the reference inputs; ui, i = 1, 2, . . . , n, are quency response across p, i.e.,
the manipulated variables; yi, i = 1, 2, . . . , n, are the system arg½gij ðjxu;ij Þ ¼ p:
For the frequency response of gij(jx) as shown in Fig. 2, mission ratio for loop yi uj, the bigger the eij value is, the
eij is the area covered by gij(jx) up to xu,ij. Since jg0ij ðjxÞj more dominant of the loop will be.
represents the magnitude of the transfer function at various Similar to the definition of relative gain [15], the effective
frequencies, eij is considered to be the energy transmission relative gain, /ij, between output variable yi and input var-
ratio from the manipulated variable uj to the controlled iable uj is define as the ratio of two effective energy trans-
variable yi. mission ratio:
Express the energy transmission ratio array as eij
2 3 /ij ¼ ð1Þ
e11 e12 e1n ^eij
6e 7 where ^eij is the effective energy transmission ratio between
6 21 e22 e2n 7
E¼6 7: output variable yi and input variable uj when all other
4 5
loops are closed. When the effective relative gains are calcu-
en1 en2 enn
lated for all the input/output combinations of a multivari-
To simplify the calculation, we approximate the integration able process, it results in an array, ERGA, which can be
of eij by a rectangle area, i.e., calculated by
2 3
eij k ij xu;ij i; j ¼ 1; 2; . . . ; n: /11 /12 /1n
6 7
Then, the effective energy transmission ratio array is given 6 /21 /22 /2n 7
U ¼ E ET ¼ 6 7
6 7:
as: 4 5
E ¼ Gð0Þ X; /n1 /n2 /nn
where the operator is the Hadamard product, and
2 3
k 11 k 12 k 1n
6k Remark 1. The introduction of energy transmission ratio is
6 21 k 22 k 2n 7
7
Gð0Þ ¼ 6 7 to mathematically represent the effectiveness of a control
4 5 loop which is affected by two key factors, i.e., the steady
k n1 k n2 k nn state gain of the transfer function reflecting the effect of the
and manipulated variable uj to the controlled variable yi, and
2 3 the response speed reflecting the sensitivity of the con-
xu;11 xu;12 xu;1n trolled variable yi to the manipulated variable uj and,
6x xu;22 xu;2n 7 consequently, the ability to reject the interactions from
6 u;21 7
X¼6 7 other loops. Since ERGA is a relative measure, using the
4 5
multiplication of the two parameters to approximate the
xu;n1 xu;n2 xu;nn
energy transmission ratio in /ij can simplify the calculation
are the steady state gain array and the critical frequency while captures the key elements in a multivariable control
array, respectively. Since eij is an indication of energy trans- system.
The ERGA is used to determine the best variable paring p þ \grii ðjx
^ u;ii Þ p þ \grii ðjx
^ u;ii Þ
d^ii ¼ ¼ cij : ð6Þ
in [14]. In the following sections, we will employ this inter- ^ u;ij
x xu;ij
action measure to develop effective transfer functions
Notice that
(ETFs) under decentralized control structure.
1. grii ðsÞ is usually low order transfer functions, their contri-
3. Effective transfer function
bution to the phase change at low frequency range are
small and can be equivalently represented by the addi-
Suppose that the best loop configuration has been deter-
tional time delay term.
mined and the best pair is diagonally placed in the transfer
2. In many decentralized control system designs, such as
function matrix as shown in Fig. 1. Similar to the open
gain and phase margin method, an individual loop is
loop gain, we let the effective energy transmission ratio,
tuned around the critical frequency region of each con-
^eij , when all other loops are closed be ^eij ¼ g^ij ð0Þx ^ u;ij
trol loop. Accurate estimation of overall variation is
i; j ¼ 1; 2; . . . ; n, where g^ij ð0Þ and x
^ u;ij are the steady state
required around the critical frequency, not who contrib-
gain and ultimate frequency between output variable yi
ute to the change.
and input variable uj when all other loops are closed,
respectively. Then, from Eq. (1)
By letting \grii ðjx
^ u;ii Þ \grii ðxu;ii Þ, we can make further
gij ð0Þxu;ij simplification to Eq. (6) as
g^ij ð0Þx
^ u;ij ¼ : ð2Þ
/ij
d^ii x
^ u;ii d ii xu;ii ;
By the definition of RGA, we have
which results by considering Eq. (4)
gij ð0Þ
g^ij ð0Þ ¼ ; ð3Þ xu;ii
kij d^ii d ii ¼ cii d ii : ð7Þ
^ u;ii
x
where kij is the relative gain.
Substitute Eq. (3) into (2) and rearrange to result This is the practical formula which will be used to derive
/ij xu;ij the ETFs. Even though Eq. (7) is less accurate than Eq.
¼ cij ; ð4Þ (6), several simulation results have showed that the control
kij x^ u;ij
system performances are comparable by the two approxi-
where cij represents the critical frequency change of loop mations, but Eq. (7) is much more straightforward and eas-
i–j when other loops are closed, defined as relative critical ier explainable and understandable than Eq. (6).
frequency. When the relative frequencies are calculated for Since it is necessary that the controlled system possesses
all the input/output combinations of a multivariable pro- integrity property; that is, the overall control system
cess, it results in an array, i.e., relative frequency array remained to be stable regardless put in and/or taken out
(RFA). of other control loops, g^ii ð0Þ and d^ii in ETF must take differ-
Since control loop transfer functions when other loops ent values for different combination of kii and cii. For the
closed will have similar frequency properties with when four different combinations of kii and cii, g^ii ðsÞ may take dif-
other loops open if it is well paired [16], we can let the ferent modes shown in Figs. 3–6, and are discussed below:
ETFs have the same structures as the corresponding open Case 1: kii 6 1,
cii 6 1
loop transfer functions but with different parameters In this case, k1ii 1 P 0 and (cii 1) 6 0. According
^
g^ii ðsÞ ¼ g^ii ð0Þgrii ðsÞed ii s ; ð5Þ to Eqs. (3) and (7), we have g^ii ð0Þ P gii ð0Þ, and d^ii 6 d ii .
where grii ðsÞ is defined by
• g^ii ð0Þ P gii ð0Þ, this means that the magnitude of the fre-
grii ðsÞ ¼ g0ii ðsÞed ii s quency response when the other loops closed is not less
than that of when the other loops open. Since the retalia-
and d^ii is the time delay of the ETF. tory effect from the other loops magnifies the main effect
As the change in ultimate frequency of a control loop is of ui on yi, we need to reduce the controller gain to assure
generally affected by changes in both time constant and system stability. In this case, the gain is by Eq. (3)
time delay when other loops are closed, and they are g^ii ð0Þ ¼ gii ð0Þ=kii :
exchangeable by linear approximation, it is reasonable to
change only time delay to reflect the phase changes. • d^ii 6 d ii , this means that the time delay when the other
In Eq. (5), g^ij ð0Þ can be determined by using Eq. (3), loops closed is not bigger than that of when other loops
while by the definition of the ultimate frequency, open. The reduced time delay will increase the phase
margin. However, by considering the control system
d^ii x
^ u;ii þ \grii ðjx
^ u;ii Þ ¼ d ii xu;ii þ \grii ðxu;ii Þ ¼ p: integrity, the time delay needs to be kept as before, i.e.,
• g^ii ð0Þ P gii ð0Þ, same as in Case 1, • g^ii ð0Þ < gii ð0Þ, this means that the magnitude of the fre-
g^ii ð0Þ ¼ gii ð0Þ=kii : quency response when the other loops closed is smaller
^ than that of when the other loops open. Even if the
• d ii > d ii , this means that the time delay when the other retaliatory effect from other loops acts in opposition to
loops closed is bigger than that of when the other loops the main effect of ui on yi, we cannot enlarge the control-
open. The enlarged time delay will reduce the phase ler gain for better performance due to the system integ-
margin. In this case, the time delay is determined by rity consideration. Hence, the gain should be
Eq. (7) unchanged, i.e.,
d^ii ¼ cii d ii : g^ii ð0Þ ¼ gii ð0Þ:
778 Q. Xiong, W.-J. Cai / Journal of Process Control 16 (2006) 773–784
To solve such a complex problem, recursive solution is By selecting A = a2, B = a1 and C = 1, the open loop
required by first assigning initial controllers, then finding transfer function becomes
the equivalent loop transfer functions and designing g^ii ð0Þ d^ii s
controllers again. This process is continuous until a stable gc;i ðsÞ^
gii ðsÞ ¼ k e :
s
solution is obtained. To simplify the problem, both
detuning and independent methods proposed so far assume Denoting the gain and phase margin specifications as Am,i
that all other closed loops are under perfect control when and Wm,i, and their crossover frequencies as xg,i and xp,i,
designing the controller for a particular loop and consider respectively, we have
only the gain change. In the proposed method, the changes
are considered for both gain and frequency. Especially, Eq. arg½gc;i ðjxg;i Þ^
gii ðjxg;i Þ ¼ p;
(3) focuses on the gain impact while Eq. (7) contributes to Am;i jgc;i ðjxg;i Þ^
gii ðjxg;i Þj ¼ 1;
time delay portion, i.e., frequency impact. As will be shown jgc;i ðjxp;i Þ^
gii ðjxp;i Þj ¼ 1;
later, it is far more accurate than those existing methods.
Wm;i ¼ p þ arg½gc;i ðjxp;i Þ^
gii ðjxp;i Þ:
Table 2
Decentralized PID controller design
Mode g^ii ðsÞ kp,ii ki,ii kd,ii
kii 6 1, cii 6 1 gii ð0Þ=kii pkii a1;ii pkii pkii a2;ii
ed ii s
a2;ii s2 þ a1;ii s þ 1 2Am;i d ii gii ð0Þ 2Am;i d ii gii ð0Þ 2Am;i d ii gii ð0Þ
gii ð0Þ=kii pkii a1;ii pkii pkii a2;ii
kii 6 1, cii > 1 ecii d ii s
a2;ii s2 þ a1;ii s þ 1 2Am;i cii d ii gii ð0Þ 2Am;i cii d ii gii ð0Þ 2Am;i cii d ii gii ð0Þ
gii ð0Þ pa1;ii p pa2;ii
kii > 1, cii 6 1 ed ii s
a2;ii s2 þ a1;ii s þ 1 2Am;i d ii gii ð0Þ 2Am;i d ii gii ð0Þ 2Am;i d ii gii ð0Þ
gii ð0Þ pa1;ii p pa2;ii
kii > 1, cii > 1 ecii d ii s
a2;ii s2 þ a1;ii s þ 1 2Am;i cii d ii gii ð0Þ 2Am;i cii d ii gii ð0Þ 2Am;i cii d ii gii ð0Þ
780 Q. Xiong, W.-J. Cai / Journal of Process Control 16 (2006) 773–784
Table 6
Controllers for Example 4
Fig. 9. Closed loop responses for Example 3. Controller Proposed Luyben
kp,ii si,ii sd,ii kp,ii si,ii
Loop 1 1.4981 2.8600 0.7150 0.084 33
Fig. 9 shows the closed-loop responses, where the unit Loop 2 24.7566 66.6700 0 5.16 15.5
set-points change in r1 at t = 0, r2 at t = 100 and r3 at Loop 3 0.0292 1 0 0.305 17.0
Loop 4 0.5349 43.4800 10.8700 0.529 11.2
t = 200.
Q. Xiong, W.-J. Cai / Journal of Process Control 16 (2006) 773–784 783
kii > 1 and cii < 1, the main process is selected as corre- function which provides necessary information of gain
sponding equivalent process. For i = 3, 4, kii < 1 and and frequency changes when all other loops are closed.
cii < 1, the equivalent process is calculated as Consequently, the decentralized controllers can be obtained
17:9031es 11:2289e3:79s by simply using single loop design approaches. Simulation
and 2
; results for variety of industrial 2 · 2, 3 · 3 and 4 · 4
sþ1 ð21:74s þ 1Þ processes show that the proposed method provide overall
respectively. The resultant PI/PID controllers by proposed better performance compared to other reported design
method are listed in Table 6 together with method (2) approaches. The advantage of the proposed method is even
based on the diagonal pairing. It should be noticed that dif- more significant when applied to higher dimensional
ferent designs have much different controller parameters processes with complicated interaction modes. The method
due to different pairing schemes. can be easily integrated into an auto-tuning control struc-
Fig. 10 shows the closed-loop responses, where the unit ture combined with some on line parameter identification
set-points change in r1 at t = 0, r2 at t = 400, r3 at t = 800 module and implemented for industrial control systems.
and r4 at t = 1200. Another interesting extension is the block decentralized
Although there are larger overshoots in some loops, the control. This topic is currently under investigation and will
system has better overall performance. This example illus- be reported later.
trates that the proposed method can be easily applied to high
dimensional processes with complicated interaction modes.
In fact, the dimension does not increase the design difficulty. References
[4] A.P. Loh, C.C. Hang, C.K. Quek, V.U. Vasnani, Autotuning of [13] H.-P. Huang, J.-C. Jeng, C.-H. Chiang, W. Pan, A direct method for
multiloop proportional-integral controllers using relay feedback, multi-loop PI/PID controller design, Journal of Process Control 13
Industrial and Engineering Chemistry Research 32 (1993) 1102– (2003) 769–786.
1107. [14] Q. Xiong, W.-J. Cai, M.-J. He, A practical loop pairing criterion for
[5] S.H. Shen, C.C. Yu, Use of relay-feedback test for autotuning of multivariable processes, Journal of Process Control 15 (2005) 741–
multivariable systems, American Institute of Chemical Engineers 747.
Journal 40 (1994) 627–646. [15] E.H. Bristol, On a new measure of interactions for multivariable
[6] W.L. Luyben, Simple method for tuning SISO controllers in process control, IEEE Transactions on Automatic Control 11 (1966)
multivariable systems, Industrial Engineering and Chemical Process 133–134.
Design Development 25 (1986) 654–660. [16] I.L. Chien, H.-P. Huang, J.C. Yang, A simple multiloop tuning
[7] D. Chen, D.E. Seborg, Design of decentralized PI control systems method for PID controllers with no proportional kick, Industrial and
based on Nyquist stability analysis, Journal of Process Control 13 Engineering Chemistry Research 38 (1999) 1456–1468.
(2003) 27–39. [17] H. Cui, E.W. Jacobsen, Performance limitations in decentralized
[8] P. Grosdidier, M. Morari, Interaction measures under decentralized control, Journal of Process Control 12 (2002) 485–494.
control, Automatica 22 (1986) 309–319. [18] Y.-G. Wang, W.-J. Cai, Advanced proportional–integral–derivative
[9] S. Skogestad, M. Morari, Robust performance of decentralized tuning for integrating and unstable processes with gain and phase
control systems by independent design, Automatica 25 (1989) 119– margin specifications, Industrial and Engineering Chemistry Research
125. 41 (2002) 2910–2914.
[10] Q.-G. Wang, T.H. Lee, Y. Zhang, Multi-loop version of the modified [19] T. McAvoy, Interaction Analysis: Principles and Applications,
Ziegler–Nichols method for two input two output process, Industrial Instrument Society of America, 1985.
and Engineering Chemistry Research 37 (1998) 4725–4733. [20] A.P. Loh, V.U. Vasnani, Describing function matrix for multivari-
[11] H.-P. Huang, J.-C. Jeng, Monitoring and assessment of control able systems and its use in multiloop PI design, Journal of Process
performance for single loop systems, Industrial and Engineering Control 4 (1994) 115–120.
Chemistry Research 41 (2002) 1297–1309. [21] N. Doukas, W.L. Luyben, Control of sidestream columns separating
[12] D. Pomerleau, A. Pomerleau, Guide lines for the tuning and ternary mixtures, Instrumentation Technology 25 (1978) 43.
evaluation of decentralized and decoupling controllers for processes
with recirculation, ISA Transactions 40 (2001) 341–351.