An expert system environment for robust controller design.

This thesis considers several important issues in the implementation of an expert system for robust controller design. The work focuses on the architecture of the expert system, robust controller design methods, implementation of knowledge bases and issues concerning the integration of these to form a useful robust controller design environment. The main contribution of this thesis is the development of an expert system package to support industrial control engineers in solving their controller design problems. The expert system captures the experiences and skills of human experts, leaving the user of the expert system free to concentrate more on the creative aspects of the controller design. The expert system makes use of recent results on the Hinfinity loop-shaping design procedure (LSDP) of McFarlane and Glover. A two degree of freedom loop-shaping design procedure (TDF-LSDP) of Limebeer and co-workers is also considered to enhance the capability of the expert system. Based on these two LSDP approaches, a systematic controller design procedure has been developed in the expert system for tackling generic control system problems. In addition, the relationships between the closed-loop time/frequency domain performance specifications and the open-loop singular value requirements were also considered. Methods for performing systematic loop shaping design based on the closed-loop time/frequency domain specifcations are incorporated into rule knowledge bases of the expert system. The organization and handling of control related data by a frame system are explicitly described. Several numerical optimization techniques are also examined and incorporated to enhance the robust control system design approach used in the integrated environment. A systematic controller design procedure using Multi-Objective Genetic Algorithm is presented. A Multi-Layer Multi-Objective Genetic Algortihm is developed for designing controllers with varying structures. A number of case studies are carried out to demonstrate the feasibility of the expert system package.