Nonlinear control of processes with actuator saturations

This dissertation concerns the fundamental problems of windup and process directionality in "input-output linearizing" control of multivariable nonlinear processes with actuator saturation nonlinearities. Two classes of general nonlinear but affine-in-control processes are considered: (a) delay-free processes with full state measurements, and (b) processes with incomplete state measurements and dead-times. The two fundamental problems are addressed by using a model predictive approach. The complete connections between input-output linearizing control and model predictive control are established: input-output linearizing control is simply shortest-prediction-horizon model predictive control. The model predictive control approach leads to nonlinear control laws that can be parameterized into two distinct parts: (i) an input-output linearizing controller that inherently includes an integral windup compensator and (ii) an optimal directionality compensator. The directionality compensator is a quadratic program that is trivially solvable online. In the case that the characteristic (decoupling) matrix of process is diagonal, the optimal directionality compensator is identical to "clipping". For general processes, however, neither direction preservation nor clipping can compensate for process directionality optimally. The concrete connections between the derived control laws, and modified internal model control and model state feedback, are established. When one of the derived control laws is applied to time-invariant linear processes, the resulting linear controller will exactly be a modified internal controller and in a special case, a model state feedback controller. The application and superior performance of the derived control laws and the optimal directionality compensator are demonstrated by numerical simulations of chemical and biochemical processes with input saturation nonlinearities.