A POMDP-Based Optimization Method for Sequential Diagnostic Strategy With Unreliable Tests

Abstract

This paper considers the sequential fault diagnosis problem with unreliable tests which exist widely in practice. This problem involves real-time inference of the most likely set of failure sources, i.e., fault state, based on unreliable test outcomes. The purpose of this paper is to optimize test set and diagnostic strategy so as to cut down the test cost while isolating the fault accurately, and a method for optimal diagnosis strategy based on partially observable Markov decision process (POMDP) is presented. The components of the POMDP tailed to optimizing the diagnostic strategy are specified, and the solution to the POMDP-based model, namely the optimal strategy, is obtained to describe the optimal test sequence for fault diagnosis. The performance of the proposed method is evaluated with simulation experiments. All the results indicate that this method performs good in diagnostic efficiency and accuracy, even compared with the strategies of traditional methods.