Prototype Diagnosis Expert System with Knowledge Refinement Function

Abstract

When developing an expert system, the difficulty in acquiring knowledge poses a bottleneck. It is essential that the system possess a means to modify its knowledge bases. This study details the construction of a versatile inference system equipped with a rule-base refinement function that is expressed by a network with relations between the hypotheses as compositional elements. As an example for practical use, this paper examines the effectiveness of the system being proposed, using the rule base of an existing expert system to diagnose cracks in damaged bridge slabs. As a result, it was found that, by presenting adequate examples as training samples, the rule base is refined along with a remarkable increase in damage-cause inference accuracy.