You are on page 1of 1

Robust Multiloop PID Controller Design: A Successive Semidefinite

Programming Approach
J . Bao,

J . F. Forbes,*
,
and P. J . McLellan

School of Chemical Engineering and I ndustrial Chemistry, TheUniversity of New South Wales, Sydney, New
South Wales, Australia 2052, Department of Chemical and Materials Engineering, University of Alberta,
Edmonton, Alberta, Canada T6G 2G6, and Department of Chemical Engineering, Queens University,
Kingston, Ontario, Canada, K7L 3N6
The probl em of robust mul ti l oop proporti onal -i ntegral -deri vati ve (PI D) control l er tuni ng for
mul ti vari abl e processes i s addressed i n thi s paper. The probl em i s formul ated i n the H

control
framework, and the control l er parameters are determi ned based on both user-speci fi ed
performance and robust stabi l i ty. The PI D setti ngs are computed by sol vi ng a nonl i near
opti mi zati on probl em wi th matri x i nequal i ty constrai nts, usi ng a successi ve semi defi ni te
programmi ng approach. The proposed method i s i l l ustrated by a si mpl e case study that
i nvesti gates robust PI D control of a di sti l l ati on col umn.
1. Introduction
Proporti onal -i ntegral -deri vati ve (PI D) control l ers have
been used extensi vel y i n the process i ndustri es si nce
they are si mpl e and often effecti ve and represent the
basi c bui l di ng bl ocks avai l abl e i n many process control
systems. Despi te thei r wi de spread use and consi derabl e
hi story, PI D tuni ng i s sti l l an acti ve area of research,
both academi c and i ndustri al (e.g., Wang and Cl uett,
1
A strom et al .,
2
and Hovd and Skogestad
3,4
). Duri ng the
past fi ve decades, a comprehensi ve PI D tuni ng l i tera-
ture has devel oped. The fi rst si gni fi cant tuni ng method
was proposed by Zi egl er and Ni chol s.
5
Anal yti cal meth-
ods to obtai n PI D parameters based on si mpl e fi rst- or
second-order transfer functi on model s were devel oped
by Ri vera et al .
6
and Gawthrop and Nomi kos.
7
For more
compl i cated transfer functi ons or transfer matri ces for
mul ti i nput-mul ti output (MI MO) systems model s, nu-
meri cal search procedures that mi ni mi ze di fferent per-
formance objecti ve functi ons were al so proposed (Radke
and I sermann,
8
Zhuang and Atherton,
9
Vega et al .,
10
A strom et al .
2
). A method for autotuni ng ful l y coupl ed
mul ti vari abl e PI D control l ers from decentral i zed rel ay
feedback was presented by Wang et al .
11
Si nce the process model s used for control l er desi gn
are often si mpl i fi cati ons or approxi mati ons, i t i s es-
senti al that the PI D tuni ngs obtai ned by such methods
shoul d tol erate model -pl ant mi smatch. Unfortunatel y,
the above control l er desi gn methods do not deal wi th
the robustness i ssue expl i ci tl y and i n many desi gn
schemes, onl y control performance i s opti mi zed. An H

(robust) PI D control l er synthesi s method was fi rst


presented by Gri mbl e.
12
A geneti c al gori thm was then
proposed to determi ne PI D control l er tuni ng to achi eve
H

opti mal i ty by Chen et al .


13
I n both of these ap-
proaches, the H

norm of the wei ghted sensi ti vi ty


functi on and compl ementary sensi ti vi ty functi on are
mi ni mi zed; however, these desi gn approaches are l i m-
i ted to si ngl e-i nput-si ngl e-output (SI SO) model s.
For control appl i cati on engi neers, mul ti l oop PI D
control l ers can be preferred to the mul ti vari abl e ap-
proach. Many pl ants have ol der, or l egacy, control
systems that do not possess the capabi l i ti es to support
the i mpl ementati on of compl ex mul ti vari abl e control -
l ers. For these pl ants, a mul ti l oop PI D control scheme
does not requi re purchase of addi ti onal control system
hardware, as may be requi red to i mpl ement mul ti vari -
abl e control l ers. Mul ti l oop desi gns al so can have better
fai l ure tol erance than the mul ti vari abl e approach;
however, mul ti l oop control l ers may suffer from control
performance l oss and even the i nstabi l i ty of cl osed-l oop
system when each i ndi vi dual l oop i s tuned by usi ng
SI SO tuni ng methods. Thi s resul ts from the i nteracti ons
among di fferent l oops, whi ch a decentral i zed control
structure cannot deal wi th.
14
The probl em of mul ti l oop robust PI D control l er tuni ng
for MI MO model s i s addressed i n thi s paper. To mi ni -
mi ze the performance l oss due to the restri cti on of a
decentral i zed control l er structure, the control l er pa-
rameters are computed based on the cl osed-l oop system
consi sti ng of the ful l process model and the mul ti l oop
control l er. Thi s l eads to a l ess conservati ve desi gn
compared to the decentral i zed control approaches, whi ch
use the di agonal subsystem as a desi gn model and treat
the known i nteracti ons as uncertai nti es (Morari and
Zafi ri ou,
15
and Samyudi a et al .
16
). Hovd and Skoges-
tad
3,4
provi de a sequenti al method for bui l di ng a mul -
ti l oop desi gn by cl osi ng one l oop at a ti me, as wel l as a
good i ntroducti on to the area of decentral i zed control l er
desi gn. To achi eve robustness, the tuni ng probl em posed
i n thi s paper i s formul ated i n the H

control framework
wi th constrai nts on the control l er structure (i .e., fi xed-
order, PI D, and decentral i zed structure). Thi s approach
al so l eads to a systemati c and uni fi ed tool for mul ti l oop
PI D tuni ng wi th user-speci fi ed performance and robust
stabi l i ty, whi ch needs much l ess heuri sti c user i nterac-
ti on than many of the manual and ti me-consumi ng
approaches menti oned above. A numeri cal opti mi zati on
procedure i s proposed to sol ve the structure-constrai ned
H

probl em wi thi n the framework of semi defi ni te


programmi ng (SDP), si nce the robust mul ti l oop PI D
* Author to whom correspondence shoul d be addressed.
Phone: (780) 492-0873. Fax: (780) 492-2881. E-mai l :
Fraser.Forbes@UAl berta.ca.

The Uni versi ty of New South Wal es.

Uni versi ty of Al berta.

Queens Uni versi ty.


3407 I nd. Eng. Chem. Res. 1999, 38, 3407-3419
10.1021/i e980746u CCC: $18.00 1999 Ameri can Chemi cal Soci ety
Publ i shed on Web 08/11/1999

You might also like