Fuzzy rules based qualitative modelling: Application to fault diagnosis of dynamic processes

This paper describes a systematic off-line algorithm for constructing a rule-based fault diagnostic system using the signed digraph (SDG) model. In order to overcome certain limitations related to the SDG (Iri and al. 1979), (Kramer and Palowitch 1987) and (Oyeleye and Kramer 1988), we have developped an alternative strategy for elimining spurious interpretations attributed to system compensations and inverse responses from backward loops and forward paths in the process when fault exist. Moreover, this method apply fuzzy logic to represent the rule base, in order to explain fault propagation and to ascertain fault origins. The algorithm was then successfully applied to a lever controlled tank process to extract rules and these rules were evaluated using on-line data to discover the fault origins in an early stage and perform the diagnosis.