We developed a computationally efficient nonlinear control strategy for constrained multivariable processes. The proposed method is a novel combination of feedback linearizing control and linear model predictive control. First the unconstrained nonlinear system is feedback linearized. Then a constraint mapping procedure is used to transform the actual input constraints into corresponding constraints on the feedback linearized system. The resulting constrained linear system is used to design a linear model predictive controller. Our initial work focused on basic controller design issues and simulation studies using a highly nonlinear chemical reactor model. We developed an analogous controller design method for discrete-time systems that is more amenable to rigorous mathematical analysis. The continuous-time formulation was extended to multivariable processes via application to a free-radical polymerization reactor model.
Funding: National Science Foundation (CTS-9501368) and ExxonMobil Chemical
Student: Michael J. Kurtz (Ph.D.)
Publications: