ADAPTIVE CONTROL WITH AN EXPERT SYSTEM BASED SUPERVISORY LEVEL by

Adaptive Controllers are susceptible to many commonly occurring implementation problems, and require some form of supervision to maintain good performance. In this thesis a two level adaptive control strategy is presented that uses an expert system to diagnose and correct problems with an adaptive controller. In contrast with other expert supervised adaptive control schemes, where problem treatment is primarily event---driven, the system described here, supports an interactive diagnosis and treatment paradigm. Basically, the interactive supervision paradigm makes it possible for the supervisory system to change some aspect of the adaptive controller's operation, and then base problem diagnosis on the subsequent response of the controller. Since the supervisory system can "experiment" with the adaptive controller, more complex diagnostics can be used, making it possible to respond to a wider set of problems than the event-driven systems are capable of. Topics presented in the thesis may be divided into explanations of the supervisory system architecture, knowledge engineering, and a review of a simulation study. In the discussions of the supervisory architecture, temporal features are emphasized, and it is shown that all of the internal structures of the expert system are tailored to operate in a dynamic environment. In addition, temporal representation features of the knowledge representation language are given, along with a review of the planning capabilities of the supervisory system. Discussions of knowledge engineering work include the definition of the adaptive controller problem domain, and the incorporation of diagnostic and treatment methods into the supervisory system. It is shown that interaction with adaptive controller environment is a necessary capability for diagnosis of certain problem