| Control methodologies to cope with multivariable process systems are studied in this thesis. Topics on both modern and conventional decentralized control strategies are covered with the major focus on the latter.;In the modern control aspect, the two most common approaches--the LQ optimal control and the eigenvalue assignment design are discussed. Emphasis is placed on investigation of LQG applications to complex real-time processes, and eigenvalue assignment design for improved steady state and robust performance. By careful consideration of practical issues and innovative use of model identification techniques, a strongly interactive and non-linear multivariable pressure tank system is satisfactorily controlled by the LQG scheme. A PI state feedback controller is proposed, and an eigenvalue assignment design for robust performance is discussed.;With respect to decentralized control, various issues including interaction measurement, variable pairing, stability and stability robustness, robust performance, controller design and integrity, are systematically addressed. Major results include: (1) A new interaction measure capable of measuring the absolute interaction and the nature of interaction is developed. (2) A new comprehensive variable pairing criterion, based on the Niederlinski Index, is presented. (3) The use of the RGA as a direct measure of integrity is expanded. (4) A series of stability conditions for decentralized control under independent design and variable pairing are established. (5) A new stability robustness measure to evaluate the effects of model error on process gains is developed. (6) New disturbance directionality indices and robust performance criteria with respect to model uncertainty associated with manipulated variables are developed, and robust design procedures for general multivariable controllers are provided. |
