Multiple-fault diagnosis in intelligent control systems

Fault isolation, diagnosis, and recovery are key components of intelligent control systems. This paper presents some new approaches to the important issue of detecting and isolating faults from multiple causes. By using a new, heuristic, fault likelihood measure and sequential testing of components, fault diagnosis algorithms are presented for which polynomial computation is sufficient, in the average case, to complete the fault diagnosis procedure. Both iterative and noniterative approaches are considered and are compared through simulation studies. Finally, the testing cost associated with the diagnostic and repair effort is shown to be approximately linear in relation to problem size.