| Chemical processes present many challenging control problems including nonlinear dynamics, operating constraints, dead-times and issues associated with measurement and state estimation. The availability of process models (or approximations) allow implementation of advanced control techniques to handle such inherent nonlinearities. This work studies application of model-based and non model-based strategies for nonlinear process control.; Model Predictive Control (MPC) has found widespread acceptance in chemical process industries in the past decade. MPC has the advantage of being able to explicitly handle the constraints imposed on the input and output variables. This makes it suitable for application in drug infusion control which present control challenges similar to chemical processes. A drug infusion system was designed to automate the regulation of vital states such as blood pressure, cardiac output and depth of anesthesia in patients undergoing surgery or under critical care. The primary challenge to drug infusion control is the inter- and intra-patient variability and the lack of physiological models that encompass the wide range of patient responses to drugs, encountered in clinical practice. In this work, a multiple-model predictive control approach (MMPC) was developed to handle constraints such as infusion rates and toxicity limits, and to adapt to patient variability. Results from laboratory experiments implementing MMPC on canine patients were presented.; Gain scheduling is a prevalent method used to design controllers for systems with widely varying nonlinearities. Auto-tuning is a convenient tool often used for tuning PID-based controllers local to a region of operation. This work proposed the integration of auto-tuning methods in a gain scheduling framework. The study focused on its application on chemical processes exhibiting output multiplicities that require operation near regions of infinite process gain. Three different auto-tuned controller designs, namely, interpolation of Tyreus-Luyben tuning parameters, direct and indirect interpolation of auto-tuned, IMC-based PI parameters were considered. Their performances were compared against an exact model-based gain scheduling strategy in simulation studies. |
