Title

Advances in system fault modeling and diagnosis

Date of Completion

January 1996

Keywords

Engineering, Electronics and Electrical|Operations Research

Degree

Ph.D.

Abstract

In this thesis, we develop effective techniques for system fault modeling and multiple fault diagnosis of complex, hierarchical systems. First, we present a comprehensive modeling methodology based on multi-signal directed flow graphs that alleviates some of the model validation problems of traditional dependency modeling, while achieving the same diagnostic resolution. We discuss techniques for extracting the fault-test dependency information (a binary dictionary matrix denoting fault-test dependencies, which is necessary for test-sequencing and fault diagnosis) for systems described by a variety of modeling methods, including qualitative models, signed and unsigned directed graph models, fault trees, and multi-signal models.^ Next, by employing concepts from information theory and AND/OR graph search, we develop several test sequencing algorithms for the multiple fault isolation problem. These algorithms provide a trade-off between the degree of suboptimality and computational complexity.^ For large systems, the test-sequencing algorithms that generate a static diagnostic strategy suffer from the following limitations: (a) high storage requirements for the diagnostic strategy, and (b) inability to handle changes in resource/test availability during a diagnostic session. Hence, we develop interactive (on-line) test-sequencing algorithms that suggest the nextbest-test to be applied based on the previous test outcomes and currently available tests. These algorithms incorporate real-world testing and monitoring features, such as precedence constraints and setup operations for tests, system modes, and modular diagnosis.^ Finally, we consider the on-board diagnosis problem of determining the most likely set of multiple failures in a system given all the test outcomes in advance, wherein the tests may have false alarms and missed detections. We develop an efficient near-optimal algorithm based on Lagrangian relaxation for generating the most likely set of multiple failures. Due to the NP-hard nature of the problem, it is not sensible to continue the search procedure until the optimal solution is reached. This premature stopping results in solutions that are not guaranteed to be optimal. Hence, we developed a novel partitioning technique to determine a ranked list of multiple fault candidate sets, which has a greater chance of containing the optimal solution.^ The algorithms developed herein have been successfully applied to several large real-world systems. Computational results indicate that our algorithms: (1) can considerably reduce the probability of false alarm error or RTOK (retest OK), (2) can be used in systems with as many as 1000 faults and tests, and (3) have superior performance when compared to the existing algorithms. ^