Currently industry and academia are interested in finding ways to employ artificial intelligence technology in useful applications. Expert systems, a major subdivision of artificial intelligence, have been used successfully in many applications, but are limited because of their generally slow response times. If expert systems could be made to run at higher rates of speed, then very powerful intelligent real-time, on-line systems would result. This research attempts to increase the speed of expert systems in a particular application domain by taking advantage of parallelism. A new parallelized match algorithm is proposed which may be used in a rule-based system shell to improve the run time of monitoring type applications. In monitoring problems, the set of objects to be matched changes rapidly, which causes most current matching algorithms used in rule-based system shells to give poor performance results. This new parallelized match algorithm, called PMA, is implemented in a rule-based system shell called PMCLIPS. For monitoring type applications, PMA used in PMCLIPS works better than the Rete match algorithm used in CLIPS. CLIPS, a rule-based system shell developed by NASA, and the Rete match algorithm are used as starting points in the development of PMCLIPS and PMA. In one rule set test, PMCLIPS ran 6.79 times as fast as CLIPS by effectively using eight processors simultaneously. PMCLIPS, which runs on a single bus shared memory multiprocessor computer, improves its response time by utilizing extra processors. Up to five or six processors may be used effectively, after which little speed is gained per processor added. It is proposed that shorter response times could be achieved with a multiple bus interconnection network machine since bus contention would thus be reduced. As it stands, however, PMA in PMCLIPS may be used more effectively than existing rule-based system shells for many applications which have rapidly changing fact bases.
