Causality‐based failure‐driven learning in diagnostic expert systems

It has been recognized that a diagnostic expert system's ability to learn from past experience will improve its diagnostic efficiency as well as make it acquire new heuristics. In this paper, we propose a failure-driven learning scheme by which the expert system automatically updates its compiled knowledge by acquiring new heuristics or refining existing heuristics. A heuristic is refined if it hypothesizes the wrong causal origin during a diagnosis. Using its deep-level knowledge of the process, the expert system draws inductive inferences from causal models to determine why the hypothesis proposed by the heuristic is inconsistent with the current state of the process. The refinement limits the applicability of the heuristic and prevents it from firing if a similar situation were to subsequently arise.