Hypercube Expert System Shell Applying Production Parallelism

摘要

Abstract : This research investigation proposes a hypercube design which supports efficient symbolic computing to permit real-time control of an air vehicle by an expert system. Design efforts are aimed at alleviating common expert system bottlenecks, such as the inefficiency of symbolic programming languages like Lisp and the disproportionate amount of computation time commonly spent in the match phase of the expert system match-select-act cycle. Faster processing of Robotic Air Vehicle (RAV) expert system software is approached through 1) fast production matching using the state-saving Rete match algorithm, 2) efficient shell implementation using the C-Programming Language and 3) parallel processing of the RAV using multiple copies of a serial expert system shell. In this investigation, the serial C-Language Integrated Production System (CLIPS) shell is modified to execute in parallel on the iPSC/2 Hypercube. Speedups achieved using this architecture are quantified through theoretical timing analysis, and comparison with serial architecture performance results, with earlier designs' performance results, with best case results and with goal performance. Theses. (RRH)