A composite methodology for low maintenance expert systems development

XCON was the first production expert system when placed in service in 1980. It is still an evolving production system, and has grown dramatically from 700 rules in 1980 to 17500 rules in September 1988. It has a very high maintenance requirement which motivated the development of RIME, resulting in the re-implementation of XCON into RIME. RIME prescribes a developed and implemented methodology for the reduction of maintenance. The author compares RIME with traditional development methodology and with another large, volatile system, CASEy, and integrates the contributions of other researchers to derive a composite second generation expert systems development methodology with the goal of producing low-maintenance expert systems. RIME decomposes the control elements into structured, declarative modules; CASEy follows a similar but less explicit methodology to achieve similar results. RIME represents a more declarative solution than did the original XCON, supporting the conclusion that the control elements should be as declarative as possible. The rule bases of RIME and CASEy also show considerable similarity. Both are formed by a layer-per-rule decomposition of the problem until primitive elements are encountered. Both modularize the rule base for easier maintenance. To these concepts, the requirements of completeness and accuracy are made explicit