Decentralized and centralized multivariable PI controllers are designe
d for unstable systems described by a transfer function matrix. Decent
ralized PI controllers are designed by using a constant detuning facto
r and centralized PI controllers are designed by the Tanttu and Liesle
hto method [J.T. Tanttu, J. Lieslehto, A comparative study of some mul
tivariable PI controller tuning methods, in: R. Devanathan (Ed.), Inte
lligent Tuning and Adaptive Control Methods, Pergamon, 1991, pp. 357-3
62]. For systems with one-way unstable interaction, both the decentral
ized and centralized PI controllers stabilize the systems. For systems
with two-way unstable interactions, decentralized controllers do not
stabilize the systems, whereas the centralized PI controllers do stabi
lize the systems. The Tanntu and Lieslehto method is extended for desi
gning centralized PID controllers also. (C) 1997 Elsevier Science B.V.