A new approach to the logic-based expert system

Conventional rule-based expert system based on modus ponens as its rule of inference. Since modus ponens is not a complete rule of inference, it has limited applicabilities. Generally, the inference of a expert system involves rule chaining which is a time consuming procedure. This paper proposes a new inference method employing a table which produces complete solutions fast. The inference table is easily established using resolvents which are resolved from rule clauses. It is shown that this resolving procedure is equivalent to simplification of the logical sum of negated rule clauses. According to the fact that the interpretations which falsify rules of a RBS could never be happened and also they are minterms which are subsumed by the prime implicants of the simplification, the proposed inference table or rules between Boolean variables be easily induced from logged examples. It is also shown that these prime implicants or resolvents are useful to develop a rule verification procedure, which is very systematic and can detect chained errors. Traditional approaches to the RBS verification are generally based on either pairwise comparison of rules or generation test input data which are not applicable to a large system with chained errors