Abstract: Aiming at the low accuracy of fault diagnosis due to the difficulty in extracting fault features of rolling bearings under strong noise background, a bearing fault diagnosis method was proposed based on wavelet transform, improved singular value decomposition multistage denoising algorithm and support vector machine model. Firstly, wavelet denoising was used to reduce the initial noise of the original signal and eliminate part of the random noise. Then, the improved singular value decomposition (SVD) was used for secondary denoising of the signal. The denoising was completed through three steps: improving the phase space matrix reconstruction method, putting forward a new method to determine the effective rank order of singular value, and optimizing the onedimensional signal extraction scheme by using kurtosis. Finally, 10 effective features were extracted and simulated in MATLAB with support vector machine. The influence of different features on diagnosis results was analyzed and the comparison between the proposed method and other methods was made. The results show that the multi-stage noise reduction algorithm can reduce the background noise under the working state of bearings and make the characteristic frequency more obvious. At the same time, the fault identification accuracy of the SVM classifier can reach 98.3%, which can effectively identify the location and severity of faults.

Key words: rolling bearing; fault characteristic frequency; wavelet transform(WT); improved singular value decomposition(ISVD); multi-level noise reduction algorithm; support vector machines(SVM); mechanical operation and maintenance