An Expert System can Aid in the Evolution of a Design Methodology

An expert system (CASCADE) has been implemented for the design of multivariable controllers for linear plants. using the linear quadratic Gaussian/loop transfer recovery (LQG/LTR) design methodology [1,2,3]. The development and testing of the knowledge base uncovered several inadequaces in the LQG/LTR theory of design. This paper focuses on the problems associated with the selection of a state weighting matrix for the regulator design (or of the process noise covariance matrix for the filter design) to satisfy constraints on the singular values of the design\´s return ratio (the "balancing" problem [4]). The availability of the expert system allowed rapid generation of designs for example systems, and rapid modifications to the design knowledge base. This provided a close study of the interaction between the effects of nominal plant poles at the origin (system type), state augmentation to control steady state error performance, selection of the constraints placed on the singular values of the return ratio, and the method used to meet those constraints. The paper tracks the evolution of our philosophy toward LQG/LTR design, and summarizes our current approach toward "balancing" the design.