A Fault Diagnosis Method Based on EMD-SVM with Multi-Feature Fusion via Sound Signals

In recent years, sound signal-based fault diagnosis methods have been investigated. However, in the field of fault diagnosis for nonlinear and non-stationary signals, traditional methods based on single-one features receive low fault diagnosis accuracy. In this paper, a fault diagnosis method that is based on the empirical mode decomposition signal processing method and support vector machine (EMD-SVM) with multi-feature fusion via sound signals is proposed. In the proposed method, multi-frequency scale decomposed signals are first obtained by the EMD, and the several instinct mode functions (IMFs) that best characterize the fault states are selected through the principle of maximum kurtosis. Based on these selected IMFs, fault features are extracted respectively, and the fusion features with the lowest correlation are generated through the analysis of the Pearson correlation coefficient. Subsequently, the fusion features are used as input for the training of the SVM model. Finally, the proposed method is effectively verified by an automobile rearview mirror folding experiment.