Knowledge compilation and refinement for fault diagnosis

A knowledge-compilation framework that is based on explanation-based learning and that uses cases, causal models, and measurement and testing knowledge for diagnosis is described. Fault diagnostic cases as well as operationality criteria are used to compile causal diagnostic knowledge into operational diagnostic knowledge. A difference analysis is then performed on that operational knowledge to refine it into heuristic diagnostic knowledge. The implementation of the system and its application to a practical problem involving fault diagnosis in cigarette factories are discussed.<<ETX>>