Almost all previous work on model-based diagnosis has focussed on persistent faults where the prior probability of component failure is provided by the manufacturer or estimated from fleet-wide service records. However, some of the most difficult to diagnose faults are intermittent. It is very difficult to isolate intermittent faults which occur with low frequency but yet at high enough frequency to be unacceptable. For example, a printer which prints one blank page out of a 1000 or a computer that spontaneously reboots once per day is unacceptable. Accurate assessment of intermittent failure probabilities is critical to diagnosing and repairing equipment. This paper presents an overall framework for estimating component failure probabilities which includes both persistent and intermittent faults. These estimates are constantly updated while the equipment is running. This paper also extends model-based diagnosis to systems where material instead of information in the form of voltages, currents, pressures is conveyed from one component of the system to another.
de Kleer, J.; Do, M. B.; Kuhn, L.; Price, R.; Zhou, R. A framework for continuously estimating persistent and intermittent failure probabilities. 19th International Workshop on Principles of Diagnosis (DX '08); 2008 Sept. 22-24; Blue Mountains, Australia.