Bearing Fault Diagnosis Based on Wavelet Transform and Auto-Encoder Neural Network

Abstract

Rolling bearing is one of the important parts in rotating machinery. The sensitive feature of signal is an important guarantee for effective diagnosis of bearings. For different data sets, there are currently no consistently validated feature extracting methods. This paper proposes a bearing fault diagnosis method based on wavelet transform and auto-encoder neural network. Firstly, the multi-scale decomposition for the signal is performed by wavelet transform. Then, the reconstructed component of each scale is made Fourier transform, and the obtained frequency spectrum is used as the input of the auto-encoder neural network. Finally, the auto-encoder neural network performs deep learning on the input data to obtain a bearing fault diagnosis model. The 10 state data sets of the rolling bearings are used to verify the performance. The results show that the method can avoid the manual feature extraction and obtain a 98.44% diagnostic accuracy.