Data Fusion and Multi-fault Classification Based On Support Vector Machines

As a new general machine-learning tool based on structural risk minimization principle, Support Vector Machines (SVM) has the advantageous characteristic of good generalization. For this reason, the application of SVM in fault diagnosis field has becomes one growing reach focus. In this paper, data fusion strategy based on multi-class SVMs is proposed to diagnose the gear fault. The fault features extracted from vibration signals with various analysis methods are transferred into the SVM in the feature fusion level. The signal analysis methods include Power Spectrum, Cepstrum, wavelet, etc. Data fusion improves the reliability of the diagnosis results. The SVM is originally designed for two-class classification. In order to satisfy the need of multi-fault classification, one three-class SVMs is built to combine the outputs of the feature fusion levels and to classify the four fault states of the gear. The actually diagnosis results show that the fault classification performance of the multi-class SVMs is evidently powerful and precise.