A Comparison of Model-Based and Incremental Case-Based Approaches to Electronic Fault Diagnosis

Abstract

CBR seems well suited to fault diagnosis because diagnostic episodes naturally form cases and much of expert competence seems to be based on reuse of old solutions. However, in many diagnosis problems it is difficult to compile a complete case description in advance, consequently the conventional one-shot case retrieval methodology will not work. In this paper we introduce a set of fault diagnosis problems that have this characteristic and we describe a model-based goal-driven system that produces focused questions that request extra information required for diagnosis. The central contribution in this paper is a description of a CBR system that also has this characteristic of producing focused questions in diagnosis. We describe the information theoretic mechanism that allows the CBR system to do this and we present an evaluation of the CBR system and a comparison of the two systems.