End To End Model And Analysis Of Convolution Neural Network For Fault Diagnosis Of Rolling Bearing

Rolling bearing is one of the important parts of rotating machinery,and its fault will lead to equipment shutdown and even serious safety accidents.Therefore,the fault detection of rolling bearing is particularly important.Based on the deep learning theory,this thesis takes the vibration signal of rolling bearings as the research object.Aiming at the problem that the complex and changeable operation environment of bearings leads to the false judgment of the common fault diagnosis model on the target environment signal,three end-to-end fault diagnosis models are proposed under the conditions of lightweight,noise interference and working condition change,and the diagnostic reliability of the proposed model is verified by experiments on different data sets.The main research contents of this thesis include the following aspects:1.One-dimensional convolutional neural network is used to extract the fault feature of one-dimensional time series vibration signals of rolling bearings.Combined with channel attention mechanism to reduce the interference of invalid features,an endto-end lightweight fault diagnosis model(AMCNN)is proposed.The experiment verifies that the proposed model has a more reliable diagnostic accuracy under a smaller weight scale.2.Based on the end-to-end lightweight fault diagnosis model(AMCNN),the feature adaptation method(FA)is proposed,and an end-to-end anti-noise diagnosis model(AMCNN-FA)is constructed to improve the adaptability of the diagnosis model in the noise environment.The experiment verifies that the proposed method has good diagnostic performance under multiple different degrees of noise interference from multiple data sets.3.Based on the end-to-end lightweight fault diagnosis model(AMCNN),a domain adaptation method combining whole domain adaptation(WDA)and category domain adaptation(CDA)is proposed,and an end-to-end condition migration diagnosis model(AMCNN-WCDA)is established to improve the migration ability of the diagnosis model in the condition changing environment.The experiment verifies that the proposed method has good migration ability under various working conditions of multiple data sets.4.The data visualization analysis method t-SNE is used to conduct twodimensional visualization analysis on the network layer of each model proposed.According to the clustering and classification of samples,the role of attention mechanism and domain adaptive method in the model is verified,and the interpretability of the deep learning model is improved.The visual analysis results show that the attention mechanism improves the clustering robustness and classification accuracy of the model;the domain adaptive method effectively alleviates the error clustering phenomenon of samples under different working conditions and improves the migration diagnosis performance of the model under cross-working conditions.The research in this thesis has a certain reference value for the fault diagnosis of rolling bearings.