Design Guidelines for Expert Systems

Abstract

Knowledge based expert systems are applicable to a wide range of engineering problems ranging from formation to derivation. At the formation end of the spectrum, design, planning and prediction have been identified as generic tasks with similar issues that are dealt with by experts, and need to be formalized for successful expert system implementation. At the derivation end, diagnosis, interpretation and monitoring have been identified as generic tasks with similar subproblems with which experts must cope. At the implementation level, four levels of programming have been identified: logic programming, production system programming, object oriented programming and hybrid programming. The following tentative guidelines are offered to aid an expert system architect in developing a system as efficiently and effectively as possible:

• Define the expert domain and the eventual environment of the implementation.

• Ensure that the domain is well defined and there is a wealth of information and, more importantly, expertise.

• Start at the highest programming level possible.

• For derivation type problems, rule based systems (or hybrid rule based systems) offer a number of advantages.

• For diagnostic problems, the EMYCIN model is particularly adaptable. Monitoring and interpretation problems can also be adapted to the EMYCIN model.

• For formation problems, object oriented code (or hybrid object-oriented systems) offer some distinct advantages.

• Production systems offer a number of advantages to both formation and derivation problems including modularity and portability.