Research On Rolling Bearing Fault Diagnosis Based On Deep Learning

The selection of eigenvalues has been the focus of research since machine learning has been put forward.The suitability of the selection of the eigenvalues will directly determine the performance of the final model,so how to extract more appropriate eigenvalues from the data is the key to machine learning.Deep learning can get the eigenvalues which are needed to do the classification through learning the information hidden in the data by self learning.Then the eigenvalues learned by deep learning are used for fault diagnosis and the results will be significantly improved.This article starts from the point of view of fault diagnosis firstly and introduces the research status and existing methods of fault diagnosis firstly.Then introducing the concept of deep learning and carry out research on fault diagnosis method of stacked sparse auto encoder,local cycle mapping stacked sparse auto encoder and cycle convolutional neural network in rotating bearing fault diagnosis.The specific research contents are as follows:(1)Put forward the method of rolling bearing fault diagnosis based on stacked sparse auto encoder.First of all,take a rotation cycle as a sample to build a training set and a test set.Then using the model of stacked sparse auto encoder and combined with the softmax classifier to train the network parameters using the training set.Finally,the performance of the fault diagnosis method is verified by the test set.(2)Put forward the method of rolling bearing fault diagnosis based on local cyclic mapping stacked sparse auto encoder.First,a local cyclic mapping auto encoder network structure for fault diagnosis is constructed for the periodic signal.Then reduce the connection rate between each node and the connected layer through take advantage of the correlation between nodes.Finally,the performance of the fault diagnosis method is verified by the test set.(3)Put forward the method of rolling bearing fault diagnosis based on cyclic convolutional neural network.First,a fault diagnosis method for rolling bearings based on convolution neural network is put forward.Then a cyclic convolution neural network for rolling bearing fault diagnosis is put forward.Finally,the performance of the fault diagnosis method is verified by the test set.