1992 年 7 巻 5 号 p. 810-819
In building a large-scale knowledge system, it is quite difficult to assure the completeness and the consistency of knowledge base in advance. Therefore, it is necessary to realize a flexible framework for refining knowledge base and acquiring new knowledge through real operation of knowledge systems. In the domain of fault diagnosis of machine equipments, heuristic diagnostic knowledge, structural knowledge and failure diagnostic cases are available for diagnosis. Heuristic diagnostic knowledge has problems of incompleteness and justification and structural knowledge has a problem of inefficiency. Failure diagnostic cases can be acquired more easily than heuristic knowledge. Therefore, there exists a complementary relation among these knowledge sources. The purpose of this paper is to present a method of refinement of heuristic diagnostic knowledge based on structural knowledge. First, we presents a method of generating operational diagnostic knowledge from failure diagnostic cases, structural knowledge and measuring/testing knowledge under the framework of explanation-based learning. Then, we presents a method of incremental refinement of imperfect knowledge base based on the difference analysis between heuristic knowledge and operational knowledge. These methods have been implemented as a knowledge refinement system on ESP language. By applying it to actual diagnostic cases about cigarette making machine, its usefulness has been shown.
