Model-based learning for diagnostic tasks

Despite many years of research, the two problems of knowledge acquisition and maintenance of a growing knowgledge base while preserving consistency are still largely unsolved. This paper introduces an approach which supports the acquisition of diagnostic rules by learning from examples representing the experiences of domain experts. Using a model of the system to be diagnosed, these heuristic rules are consistency checked before they are added to the knowledge-base. Moreover, this model constitutes a base for learning by transforming analogous cases into actual problems when the system is operational.