Application of a mahalanobis-based pattern recognition technique for fault diagnosis on a chemical process

The paper proposes a Fault Detection and Isolation (FDI) procedure based on a model-free approach and the use of pattern recognition techniques. In particular this paper aims to improve the isolation performance of a Fuzzy Faults Classifier (FFC) previously proposed by the author by modifications of the fuzzification module and by the use of the Mahalanobis distance as metric for identifying the most probable fault. In the paper faults due to the wear and tear of the thrust bearing and to fouling of the compressor stage of an industrial multishaft centrifugal compressor are considered. The presented results show the goodness of the overall procedure in the detections of single as well as multiple faults and its promptness in terms of faults isolation.