A new approach to fault pattern classification of gasoline engine vibration

This paper presents a new approach to fault pattern classification of gasoline engine vibration based on statistics analysis, rough set and the support vector machines. First, different time domain statistical features are extracted from the resultant subband signals which derived from multiscale analysis of the raw vibration data, to acquire more fault characteristic information. Second, a rough set model is utilized to select the most superior features from the initial feature set. Finally, the selected superior features are input into the support vector machines classifier to accomplish faulty pattern classification. The experimental result show that the proposed method can extract the faulty features with better classification ability and at the same time reduce lots of features in case of assuring the classification accuracy, accordingly a better performance of fault diagnosis is obtained.