A Mathematical Model for Diagnosing System Failures

Abstract

A mathematical model is developed for diagnosing system failures when symptoms are observable. Optimal policies for searching malfunctions yielding minimum expected diagnostic cost are developed, based on the probabilities of various malfunctions conditioned on the set of observable symptoms, the detection probability of each malfunction, and its associated testing cost. The necessary and sufficient conditions satisfied by such policies are derived. The model can be implemented easily on a computer, reducing costs of diagnosis and training diagnosticians.