An experiment in the design of modular expert systems with special reference to fault diagnosis

A method for dealing with three of the major problems in hybrid rule-based expert systems is proposed. They are; the problem of managing the complexity of the knowledge base, the problem of graceful degradation and the problem of the logics used to manipulate the rules, which do not fit easily with human expertise. An experimental expert system, Aristotle, is developed in which modules are used as a tool to reduce complexity. Modules group related rules together into components, which interact with the outside world through parameters. The module also provides a suitably sized object on which extra (partial) knowledge can be attached. It is proposed to use partial knowledge to reason, at a general level, about components and groups of components, giving the appearance of a knowledge base, which degrades gracefully. A five-value logic is introduced with the extra logic values irrelevant, do-not-know and unknown. This logic is able to reason about unusual situations without having to explicitly check for them. This thesis demonstrates, that modules, partial rules and the five-value logic, can overcome, to an extent, the problems mentioned earlier.