Supporting model-based diagnosis with explanation-based learning and analogical inferences

This paper introduces two learning approaches used in different phases of a diagnostic system's life cycle. First, an initial knowledge-base is built using an explanation-based learning approach which generates diagnostic rules. A functional model of the object to be diagnosed constitutes the necessary domain knowledge. Later when the system is operational, analogical inferences which utilize taxonomic information continue to improve its diagnostic performance. In this way knowledge which is 'objectivized' by the model can be acquired, greatly improving the performance of a pure model-based diagnosis while preserving the advantages of the model-based approach.