Alternate path reasoning in intelligent instrument fault diagnosis for gas chromatography

Abstract

Intelligent instrument fault diagnosis is addressed using expert networks, a hybrid technique which blends traditional rule-based expert systems with neural network style training. One of the most difficult aspects of instrument fault diagnosis is developing an appropriate rule base for the expert network. Beginning with an initial set of rules given by experts, a more accurate representation of the reasoning process can be found using example data. A methodology for determining alternate paths of reasoning and incorporating them into the expert network is presented. Our technique presupposes interaction and cooperation with the expert, and is intended to be used with the assistance of the expert to incorporate knowledge discovered from the data into the intelligent diagnosis tool. Tests of this methodology are conducted within the problem domain of fault diagnosis for gas chromatography. Performance statistics indicate the efficacy of automating the introduction of alternate path reasoning into the diagnostic reasoning system.