Diagnosis basfd on temporal band sequences an empirical comparison to statistical approachs

Diagnosis based on Temporal Band Sequences (TBS) is a decision procedure which ensures that no fake alarm will be generated provided that noise, sensors inaccuracies and model parameters are represented by an interval derived from the physical characteristics of components. The question is: will there be any diagnosis at all ?. In this paper we relate an empirical comparison between TBS method and statistical approaches. It shows that the TBS method finds the same set of faults as statistical methods do (Cusum) but with a longer delay. Nonetheless, with statistical methods, false alarms might always occur and faults of small intensity might be missed . On the contrary the TBS method does not generate false alarms and finds fault of small intensity if the observation duration is long enough. A possible explanation is that no derivative has to be computed in the TBS approach.