A novel methodology is proposed for the analysis of robust stability of a n
onlinear process under PI (Proportional-Integral) control. The analysis is
based on state-affine empirical models regressed from input-output data. Th
e state model is represented by a set of polynomial matrices nonlinear with
respect to the manipulated variables. This model in combination with a lin
ear PI controller results in a closed loop model that can be shown to lie i
n a polytope of matrices. This allows for the formulation of a Lyapunov sta
bility test in terms of a simple set of LMIs (linear matrix inequalities).
This set of inequalities can be also expanded to account for input saturati
on. The stability analysis produces regions of stability, in terms of the P
I controller parameters, that are significantly larger than the regions pre
viously calculated by a singular value test. The issues of saturation and m
odeling error are also incorporated into the analysis. The technique permit
s also to test the stability of the closed loop system with a gain scheduli
ng PI controller. The conservativeness of the analysis is assessed by compa
rison to closed loop simulations of a highly nonlinear CSTR (continuous sti
rred tank reactor) under PI control. (C) 2001 Elsevier Science Ltd. All rig
hts reserved.