Fault Isolation Using Fuzzy Sets

In many real-world systems, isolating faults is complicated by the large number of faults that can occur, the complex interactions or relationships between faults, and the presence of uncertainty. Consequently, automated fault isolation for such systems can become computationaly expensive, thereby making it difficult to provide an accurate diagnosis in real-time. This proposed work will develop an approach to automated fault isolation in the presence of uncertainty that can often provide a diagnosis in real-time. The principal types of uncertainty that will be dealt with in this work are uncertainty in the observed symptoms and uncertainty in the relationships in the fault net. Various methods for utilizing fuzzy logic in incorporating information in the form of natural language rules, degree of severity of the symptom, degree of symptom-fault relationship, continuous transition across thresholds, etc. and the combination of this evidence, will be developed and tested. As a result the diagnostic method will be more robust and accurate than when such uncertainties are ignored. This method will be applied in the problem of diagnosing failures in a space communications system.