Knowledge based Real Time Fault Diagnosis with EFTAS

Abstract This paper presents theory and application of a real-time knowledge based system for fault diagnosis based on the evaluation of fault trees. Fault diagnosis is the art of deriving the underlying causes for a set of symptoms which have occurred with the help of apriori knowledge about process behaviour. The knowledge about the process is structured here in the form of fault trees, which relate causes and effects by AND/OR-elements. The fault trees are evaluated by applying fixed inference rules for each element until the maximum number of unknown states in the fault tree has been determined from the known symptoms. Process dynamics is taken into account by using a specially tailored many-valued logic. For the states which cannot be determined with certainty a probabilistic analysis is done based on conditional probabilties. Then the program EFTAS (=expert fault tree analysis shell) is presented, which incorporates this diagnostic technique. EFTAS works as a real-time analyser of process symptoms transmitted via a serial communications link. Finally, the method is demonstrated by simulation of different faults of a condensate injection seal of a large boiler feed pump in a power station.