Adaptive feature extraction algorithms and SVM with optimal parameters on fault diagnosis of bearing

Abstract

Combining adaptive feature extraction algorithms with multi scale entropy (MSE), a synthetic method for fault diagnosis of bearing based on support vector machine (SVM) is proposed. The performances of two new adaptive feature extraction algorithms and Empirical Mode Decomposition(EMD) on the fault feature extraction of vibration signals are first time compared simultaneously. It is shown that multi-scale entropy (MSE) of the optimal mode is introduced for fault classification, and the effect is better when the scale factor is selected reasonably. In addition, as the feature vectors of SVM, the MSE of optimal mode complete fault diagnosis of bearing, under the optimal penalty parameter C and kernel parameter γ of RBF kernel function have been searched by Cross-validation and Grid-search. Finally, data in engineering are used to prove that not only the performance of new adaptive feature extraction algorithm but also the diagnostic unit on SVM are further improved, and the accuracy of diagnosis is raised to 97% over EMD+SVM only.