Fault Diagnosis of Rolling Bearing Based on Wavelet Packet Decomposition and SVM-LMNN Algorithm

Abstract

Aiming at the effective identification of failure modes of rolling bearings, a support vector machine (SVM) and Levenberg–Marquardt (LM algorithm) fault diagnosis method for rolling bearings is proposed. First, use wavelet packet decomposition to obtain sub-bands, reconstruct the decomposition coefficients, and expand the decomposed sub-band signals to the original signal length; then, use SVM to classify the fault state; finally, input the feature vector into LMNN (LM algorithm Neural network) to realize failure mode recognition. The method is verified by the rolling bearing fault diagnosis experiment. The results show that the SVM-LMNN based on wavelet packet decomposition has a rolling bearing fault diagnosis accuracy rate of up to 99.456%. The method proposed in the study is compared with the instantaneous energy method of the VMD component of the kurtosis criterion and the enveloping spectrum solution diagnosis method, and the higher accuracy is obviously obtained, which proves the feasibility and effectiveness of the proposed method.