Академический Документы
Профессиональный Документы
Культура Документы
www.elsevier.com/locate/jprocont
Received 21 September 2004; received in revised form 7 February 2005; accepted 23 March 2005
Abstract
Utilizing both steady state gain and bandwidth information of the process open loop transfer function elements, this paper pre-
sents a new dynamic loop pairing criterion for decentralized control of multivariable processes. Through defining an effective gain
matrix, the loop pairing procedures of popular relative gain array method is directly extended to the new method which can reflect
dynamic loop interactions under finite bandwidth control. Compared with existing methods, this method is simple, effective, and
easy to be understood and applied by control engineers. Several examples, for which the RGA based loop pairing criterion gives
an inaccurate interaction assessment, are employed to demonstrate the effectiveness of the proposed method.
2005 Elsevier Ltd. All rights reserved.
Keywords: Multivariable process; Decentralized control; Interaction measures; Loop pairing; Effective gain; Relative gain array; Effective relative
gain array
0959-1524/$ - see front matter 2005 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jprocont.2005.03.008
742 Q. Xiong et al. / Journal of Process Control 15 (2005) 741–747
Furthermore, if all n loops are closed, the multi-loop and 40. In such a case, pairing the faster loops (even
system will be unstable for all possible (any) values of with smaller steady state gains) takes advantage of the
controller parameters (i.e., it will be ‘‘structurally mono- time scale decoupling such that seriousness of the
tonic unstable’’), if the NI is negative, i.e. interactions from the slower loop would be reduced.
jGð0Þj
NI ¼ Qn < 0;
i¼1 g ii ð0Þ
3. A modified loop pairing rule
where jG(0)j denotes the determinant of matrix G(0).
The sign of NI, i.e. NI > 0, provides a necessary stability In designing decentralized controllers for multivari-
condition and consequently, constitutes a complemen- able processes, it is desired that the interaction measures
tary tool to the RGA in variable pairing selection. and loop pairing will address the following issues:
One of the main advantages of these methods is that
the interaction depends on only the steady state gains. (1) The interaction measure should consider the finite
This information is easily obtained from simple identifi- bandwidth control, since the assumption of perfect
cation experiments or steady state design models. A po- control is only valid for very low frequency range.
tential weakness of these methods, however, is the same (2) The loop pairing decision should be controller
fact that they only use the steady state gains which are independent such that any controller type could
based on the assumption of perfect loop control to be designed after loop pairing.
determine loop pairing. We use the following example (3) The pairing results in minimal interaction within
to illustrate this point. the interested frequency range not only statically
but also dynamically.
Example 1. Consider a process given by [14] (4) It should be simple and easy to use for practical
2 3 engineers.
5e40s e4s
6 100s þ 1 10s þ 1 7
GðsÞ ¼ 6
4 5e4s
7. In a decentralized control system design, the individ-
5e40s 5
ual loop is tuned around the critical frequency region of
10s þ 1 100s þ 1 the transfer function which is the region around the con-
The RGA is trol system bandwidth. Thus, this is the frequency region
that should be focused upon when considering the effect
0.8333 0.1667 of interactions. Therefore, two factors in the open loop
KðGð0ÞÞ ¼ .
0.1667 0.8333 transfer functions will affect the loop pairing decision:
This result implies the diagonal pairing 1–1/2–2 is a 1. Steady state gain: the steady state gain gij (0) of the
good choice. Hence, according to the RGA based loop transfer function reflects the effect of the manipulated
pairing criterion, the pairing of 1–1/2–2 should be pre- variable uj to the controlled variable yi.
ferred for the smaller interaction to any one loop from 2. Response speed: response speed is accountable for
another closed loop. However, McAvoy et al. used the sensitivity of the controlled variable yi to manip-
DRGA and optimal decentralized PI controllers for var- ulated variable uj and, consequently, the ability to
ious configurations, and found that the diagonal pairing reject the interactions from other loops.
resulted in a poor closed loop performance [14]. The off-
diagonal pairing 1–2/2–1 takes advantage of the fast Since the response speed is proportional to the band-
g21 (s) and g12(s) transfer functions to achieve better width in frequency domain, we can use the bandwidth to
response for y2: reflect both interactions from finite bandwidth control
and pairing loops to result in a fast response. Let
1. there is very little interaction from y1 to y2;
2. the off-diagonal pairing y2 response is significantly gij ðjxÞ ¼ gij ð0Þg0ij ðjxÞ;
better than that of y2 with diagonal pairing; where gij (0) and g0ij ðjxÞ are the steady state gain and
3. the y1 response for the diagonal pairing is somewhat normalized transfer function of gij(jx), i.e. g0ij ð0Þ ¼ 1,
better than that of the off-diagonal pairing when y2 is respectively.
given a step set point change. In order to use both steady state gain and response
speed information for interaction measure and loop
The main reason for the poor performance of the pairing, we now define the effective gain eij for a partic-
diagonal pairing is the dynamic properties of the ular transfer function as
transfer functions. It can be easily seen that the time Z xB;ij
constants and delays of 10 and 4 of the off-diagonal eij ¼ gij ð0Þ jg0ij ðjxÞj dx; ð2Þ
elements are 10 times smaller than diagonal ones of 100 0
744 Q. Xiong et al. / Journal of Process Control 15 (2005) 741–747
interactions, while NI is used as a sufficient condition to Example 3. Consider a process given by [3]
rule out the closed loop unstable pairings. Since we are 2 3
5 2.5e5s
using effective gains (energy) instead of simply, steady 6 4s þ 1 ð2s þ 1Þð15s þ 1Þ 7
state gains, dynamic interactions up to the critical fre- 6 7
GðsÞ ¼ 6 7.
quency can be effectively reflected. Therefore, compar- 4 4e6s 1 5
ing ERGA method with RGA and DRGA, we may 20s þ 1 3s þ 1
expect that:
RGA: k11 = 0.3333 implies off-diagonal pairing.
1. In addition to steady state gains, only bandwidth The bandwidth of each element is xB,11 = 0.25,
need to be calculated in ERGA method, it is far easier xB,12 = 0.066, xB,21 = 0.05 and xB,22 = 0.333. Hence
to calculate than those in DRGA methods. 1.2500 0.1650
2. ERGA combines both steady state gain and band- E¼ ;
0.2000 0.3330
width (effective energy) in measuring the loop interac-
tions, it should provide better pairing results than and ERGA: /11 = 0.9265 implies diagonal pairing.
that of RGA based pairing and comparable with This loop pairing decision was obtained by Grosdi-
DRGA ones. dier and Morari [3] through analyzing both magnitude
3. Since ERGA only uses information of open loop and phase characteristics of the interaction between the
process transfer functions, it is controller type two loops.
independent.
Example 4. Consider a 3 · 3 process [16] given by
2 3
2es 1.5es es
4. Case studies
6 10s þ 1 sþ1 sþ1 7
6 7
6 7
In this section, we use various examples to show the 6 1.5es es 2es 7
GðsÞ ¼ 6 7.
effectiveness of ERGA method in both simplicity and 6 sþ1 sþ1 10s þ 1 7
6 7
correctness. 4 es 2es 1.5es 5
sþ1 10s þ 1 sþ1
Example 2 (Continue with Example 1). For this exam-
ple, k11 = 0.8333 implies diagonal pairing. The RGA of the system is
The four bandwidths are calculated: xB,11 = 0.01,
xB,12 = 0.1, xB,21 = 0.1 and xB,22 = 0.01. 2 3
0.9302 1.1860 0.7442
Hence the effect gain matrix is 6 7
K ¼ 4 1.1860 0.7442 0.9302 5.
" #
0.0500 0.1000 0.7442 0.9302 1.1860
E¼ ;
0.5000 0.0500
Obviously, two possible parings 1–2/2–1/3–3 and 1–3/
and ERGA: /11 = 0.0476. It strongly suggests off-diago- 2–2/3–1 are comparable because all related RGA ele-
nal pairing. ments are close to 1. Therefore, RGA pairing approach
cannot determine which pairing is better.
In order to test the pairing results for this process, The bandwidth matrix of the system is
McAvoy et al. [14] designed two sets of optimal PI con- 2 3
trollers for diagonal and off-diagonal pairings, respec- 0.1 1 1
tively. For each configuration, two controller gains and 6 7
X¼4 1 1 0.1 5.
two integral gains are optimized for a step change in y1
1 0.1 1
followed by a step change in y2 using the same objective
function that was used for the optimal control calcula- The ERGA is then obtained as
tion. Equal weighting is given to the measurements and 2 3
the manipulated variables. In calculating the errors for 0.0554 0.6977 0.2468
y1 and y2 the difference between these measurements 6 7
U ¼ 4 0.6977 0.2468 0.0554 5.
and their set points is used. Since the minimization in-
0.2468 0.0554 0.6977
volved in tuning the PI controllers is non-convex, several
different starting points are used to determine the best This indicates the best paring is 1–2/2–1/3–3 because the
values of the tuning parameters. After all, it is concluded corresponding ERGA elements are much closer to 1
that the DRGA correctly indicates that an off-diagonal than the other option. This was confirmed by the gener-
pairing produces a better overall control system response alized dynamic relative gain (GDRG) approach [16].
than that of the diagonal pairing. However, the proposed method is much simpler.
746 Q. Xiong et al. / Journal of Process Control 15 (2005) 741–747
Example 5. Consider a 4 · 4 industrial reactor/recycle to those produced by the pairing option suggested by
system [17] given by RGA. Nevertheless, the proposed method is much more
2 3.96ð197sþ1Þ 0.536ð258sþ1Þ 9.7
3 direct and the computation is much easier.
0 49.4s2 þ14.1sþ1 83.3s2 þ18.3sþ1 24.3sþ1
6 7
6 0.00111 0.044 0.0152e32s 0.039e20s 7
6 s 46.9s2 þs s s 7 5. Conclusions
GðsÞ ¼ 6
6 15.96ð529sþ1Þ
7.
7
232.2 139.2
6 0 32.2sþ1 10 417s2 þ204sþ1 7.27sþ1 5
7
4 In this paper, both steady state gain and bandwidth
0.582 2.54ð8.11sþ1Þ 0.0462ð45.6sþ1Þe
35s
0.0358e30s of the process were used to provide a simple yet compre-
s 6.25s3 þ5s2 þs 306s3 þ35s2 þs s
hensive description of loop interactions for MIMO
Since interaction analysis via the RGA involves steady processes. The ERGA can be conveniently calculated
state information, for processes that contain pure inte- by control engineers since the bandwidth can be easily
grator elements that have no steady state, the steady obtained from the given transfer function matrix. The
state gain and RGA of integrating processes can be cal- effectiveness of the method was demonstrated by several
culated using a special method proposed by Arkun and examples, for which the RGA based loop pairing
Downs [18]. Let us consider criterion gives an inaccurate interaction assessment,
while the proposed interaction measure and loop pair-
1 ing criterion provides accurate results and they are very
I¼
s easy to be calculated. Furthermore, the ERGA method
is not only an effective tool for loop pairing, but also
and substitute I into the original transfer function very useful in helping to design the decentralized and
matrix to obtain decoupling control systems. The design of the decentral-
K ¼ lim GðsÞ. ized controller, especially, for high dimensional pro-
s!0
cesses using ERGA information as a detuning factor
Finally determine RGA by is currently under investigation, the results will be re-
RGA ¼ lim K ðK 1 Þ
T ported later.
I!1
[13] L. Tung, T. Edgar, Analysis of control–output interactions in [16] H.-P. Huang, M. Ohshima, I. Hashimoto, Dynamic interaction and
dynamic systems, AIChE J. 27 (1981) 690–693. multiloop control system design, J. Process Control 4 (1994) 15–27.
[14] T. McAvoy, Y. Arkun, R. Chen, D. Robinson, P.D. Schnelle, A [17] D. Robinson, R. Chen, T. McAvoy, D. Schnelle, An optimal
new approach to defining a dynamic relative gain, Control Eng. control based approach to designing plantwide control system
Practice 11 (2003) 907–914. architectures, J. Process Control 11 (2001) 223–236.
[15] A. Niederlinski, A heuristic approach to the design of linear [18] Y. Arkun, J. Downs, A general method to calculate input-output
multivariable interacting subsystems, Automatica 7 (1971) 691– gains and the RGA for integrating processes, Comput. Chem.
701. Eng. 14 (1990) 1101–1110.