On the Use of Bifurcation Analysis for Robust Controller Tuning for Nonlinear Systems

This paper presents a new approach to robust controller tuning for nonlinear systems. The main idea is to apply numerical bifurcation analysis to the closed-loop process, using the controller tuning parameters, the set points, and parameters describing model uncertainty (parametric as well as unmodeled dynamics) as bifurcation parameters. By analyzing the Hopf bifurcation and saddle-node bifurcation loci, bounds on the controller parameters are identified. These bounds depend upon the type as well as the degree of mismatch that exists between the plant and the model used for controller design.
The method is illustrated by tuning a state feedback linearizing controller for an unstable reactor as well as by comparing a proportional-integral (PI) controller and a state feedback linearizing controller applied to a continuously operated fermenter. The feedback linearizing controller can result in better performance than the PI controller if a very accurate model of the process is known and if the operating conditions vary over a significant range. However, stability properties of systems controlled by feedback linearizing controllers can degrade significantly as the mismatch between the plant and the model increases. This is illustrated in the fermenter example by showing that bounds on the tuning parameter of the feedback linearizing controller are significantly tighter than the ones for the PI controller.