The Reduction of No Fault Found Incidents Using Posterior Knowledge Integration

Over the past ddecade or so, research effort has been directed at developing methods of identifying faults or conditions in dynamical systems using statistical classifiers based upon historical data. Much effort- and possibly expense- can go into the development of such classifiers which form the basis of diagnostic systems. A st of fault-condition (FC) posterior probabilities is agenerated upon which diagnoses are made for example using maximum A POSTERIORI (MAP) or Bayesian risk weighted decision criteria (e.g. Melsa and Cohn, 1978). In short, this is a STATISTICAL viewrpoint on condition monitoring. When given a ranked set of FC probabilities representing the most likely FC's to have occurred, if the most probable FC is known not to have occurred, then what should further decisions be based upon? Does it make sebse always to choose the next most likely FC or set of FC's.