Abstract

Circuit pack troubleshooting is one of the high-impact processes aimed to increase the quality of manufacturing of electronic systems. We describe a methodology which allows for smooth integration of shallow and deep knowledge (e.g., rules and device topology) as well as for amalgamation of different troubleshooting strategies (e.g., “test” and “replace”). An expert system for troubleshooting a 200-component hybrid circuit pack is constructed using the STAREX approach and installed at one of AT&T’s factories. The prototype expert system in PROLOG consists of eighteen meta-rules and an automatic run-time rule generator. The automatically translated production system in either C5 or C currently uses 11000 rules and has an estimated 80% fault coverage. Our methodology uses a variety of techniques from object-oriented and logic programming paradigms. The concept of malfunctioning hierarchy is extended to obtain a uniform representation of variable diagnostic tree objects with hierarchical run-time binding capabilities in both parameters and topology. The counterpart concept, that of a variable inference engine is proposed and utilized to obtain an object-oriented uniform diagnosis and repair circuit pack knowledge representation.