Self-organizing model based expert controller

A self-tuning expert fuzzy controller is developed and applied in real time to a process control problem. The knowledge base consists of rules describing the control law in terms of the process error and the resulting control action. Conditions and conclusions of each rule are fuzzy variables which are described through their continuous membership curves. The inference engine used is the backward chaining process of the Prolog language. To implement the self-tuning property, the membership curve of the controller output is changed according to an error-based performance index. A control supervisor makes this tuning decision as a function of either past or predicted future set-point errors of the control system. If the current process model is considered reliable, then the decision is based on the predicted set-point error. The feasibility of this self-tuning expert controller is demonstrated on the speed control problem for a DC motor load system.<>