Timed fault diagnosis

Time has been shown to be determinant in many diagnosis situations. A way to account for time is to enrich the fault signature matrix with time information. However this requires to anticipate the symptoms occurrence order or provide estimates for the symptom occurrence dates. In this paper, we use the time information present in the behavior model of the system and automatically retrieve this information when a symptom occurs. A conflict-based approach allows us to generate the diagnoses. For this purpose, we extend the classical DX diagnosis generation algorithm based on hitting sets to deal with time and provide an incremental version. The algorithm takes as input time labelled conflicts that are obtained by checking the consistency of observations against a causal behavioral model including explicit time information. Our algorithm accounts automatically for time and outputs time labelled diagnoses that are updated in time. It is illustrated on a two tanks example including delays in which time aspects have a strong impact on diagnosis and fault discrimination.