Second generation expert systems should be based upon an expert´s high level understanding of the application domain and upon specific real world experiences. By having an expert categorize different types of relevant experiences and their components, hierarchies of abstract problems and operator classes are determined on the basis of the expert´s accumulated problem solving experiences. The expert´s global understanding of the domain is integrated with the experiences by a model of expertise. This model postulates problem classes at different levels of abstractions and associated skeletal plans. During a consultation with the expert system previously unseen types of input may be used to delineate a new problem. The application of the expert system can thus be situated in changing environments and contexts. With increasing dissimilarity between the cases that were analyzed during knowledge acquisition and the specific problem that is processed at the time of the application of the system, its performance gracefully degrades by supplying a more and more abstract skeletal plan. More specifically, the search space which is represented by the skeletal plan increases until the competence of the system is exceeded. This paper describes how such a case-oriented expert system is developed for production planning in mechanical engineering.
Abstract. During the course of the development of a Case-Oriented Expert System for situated applications additional cases were needed. The required cases were obtained by having a human expert refit old solutions to new problems and the structural relations between source and target cases were analyzed: A higher degree of reuse of the old cases was found when the expert could apply derivational reasoning and a uniform design rationale (i.e. the solution of the source was generated by the expert himself) than when the expert could only analyze structural relationships (i.e. the source solution was constructed by some one else). Except with very obvious cases, it was also found, that different experts perceive different cases as the most similar source to a given target problem. The results also indicate for user-situated applications of expert systems.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.