Directionality and Nonlinearity - Challenges in Process Control

The use of the concepts of directionality and ill-conditionedness is investigated. A lack of consistent terminology is detected, and some clarifications to the terminology are suggested. Robustness issues with respect to directionality are discussed. Control structures in the form of decoupling are discussed, and a more general formulation for dynamic decoupling is formulated. Process nonlinearity, especially in the form of operating region dependent dynamics, is tackled by the use of a nonlinear input/output model of quasi-ARMAX type. The model is applied in the model predictive control framework for control of nonlinear processes, with a pH-process as a case-study. The nonlinear controller is also reformulated for the multivariable case, and applied to control of an ill-conditioned, nonlinear distillation column. The computational demand of the model predictive control formulation is addressed by studying methods to approximate the behaviour of model-predictive controllers by the use of neural networks as function approximators. Also, as a means to decrease the online computational demands, a neurodynamic programming formulation is investigated in the sense that the future cost-to-go function is calculated by use of offline MPC calculations and then approximated.