A Novel Fault Identifying Method with Supervised Classification and Unsupervised Clustering

To satisfy the robust requirement when designing fault identifying method, this paper proposes a novel method to identify sensor fault. Conventional fault identifying method could only classify fault into explicit set. Yet, when a novel faulty pattern occurs, the conventional method can not identify this new pattern and will classify it into a set known ahead of time. For the purpose of robustness of fault identifying method, the supervised classification and the unsupervised clustering are integrated together. Faulty features are extracted with wavelet package decomposition to train the supervised classification algorithm and the unsupervised clustering algorithm. As the supervised classification method, Support Vector Machine (SVM) is utilized to classify those faulty patterns known ahead of time into explicit set for the purpose of identifying faulty pattern. As the unsupervised clustering method, subtractive clustering method is utilized to identify the novel fault pattern. Therefore, SVM and subtractive clustering are integrated to identify sensor fault pattern even when novel faulty pattern emerges. The applicability and effectiveness of the proposed method is illustrated by flow sensor data in an engine fuel providing system. The result shows that the method adopted provides better performance compared with conventional method while satisfying the robust requirement of fault identifying method.