A memory-based approach to fault detection and diagnosis

Fault detection and diagnosis are functions with enormous importance to advanced intelligent supervisory control systems. In the quest for improved quality and safer operations, we adopt a different approach to fault diagnosis based on the memory-based learning paradigm. The properties of memory-based methods that make them especially appropriate for autonomous systems functioning in environments that are not known in advance and in which the designers will not be able to tune the learning parameters during operation are thoroughly discussed. Some aspects of practical implementations are considered. Finally, we explore a sound approach to dealing with practical fault detection scenarios when the available database is huge.