Fault detection and isolation of gas turbine: Hierarchical classification and confidence rate computation

Abstract

In this paper, a Fault Detection and Isolation (FDI) system based on an ensemble-based hierarchical classifier is devised to detect and isolate twelve typical turbine faulty scenarios and a healthy scenario. The hierarchical classifier is designed based on eight studied classifiers' best-practice confusion patterns, ranging from classical to state-of-the-art methods. The proposed hierarchical classifier discriminates five faulty classes with almost perfect accuracy while increasing the accuracy of healthy scenario discrimination by 10%, which is particularly important since it significantly reduces the false alarm rate for any form of fault. The overall accuracy of classification is improved by 2%, as well. Along with the classifier, an appropriate notion of confidence rate helps evaluate the classifier's decisions, especially when unlabeled data are involved. This motivation leads to devising a confidence rate concept that suits the devised classifier well. The obtained results show that the introduced confidence rate changes monotonically with classification accuracy. In the case of the high confidence rate of unlabeled data, this property is utilized to label the data with high accuracy. The proposed confidence rate is particularly effective in decision-making in the labeling of challenging mixed-up scenarios. Obtained results show that the classifier's decisions are reliable, with a confidence rate greater than 0.7. The developed FDI system is examined with simulated and real data, where obtained results verify the satisfactory performance of the system.