Intelligent Fault Diagnosis of Bearing Based on Stacked Denoising Autoencoders

A fault diagnosis method based on stacked denosing autoencoder network is proposed to apply deep learning to equipment fault diagnosis. A deep-layer network model is established, pre-training is performed in a layer-by-layer greedy coding mode to achieve adaptive extraction and mining of high-dimensional deep fault features, and then a back-propagation algorithm is used to supervise fine-tune the model. The method integrates the two steps of feature extraction and state classification, and gets rid of the dependence of traditional machine learning methods on artificially extracted sample features, and effectively overcomes the problems of gradient disappearance and local extremum. Through the rolling bearing data experiment, the method's ability to identify and generalize faults was verified.