Research On The Intelligent Diagnosis Method Of Bearing Fault Based On Deep Learning

With the rapid development of computer science and technology and the Internet,fault diagnosis technology has entered the era of big data.Traditional fault diagnosis methods usually require personnel with professional knowledge backgrounds to collect,analyze,and process data,which makes the diagnostic process costly and error-prone,resulting in low generality and unable to meet the current needs of intelligent diagnosis of rolling bearing faults.Therefore,this paper uses the feature extraction and representation learning ability of deep learning to carry out research on rolling bearing faults,and the main contents are as follows:(1)A sparse structured domain adaptive bearing fault diagnosis method is proposed to address the problems of the large number of model parameters and poor recognition performance under varying loads.By utilizing a sparse structure to construct the network model,the number of model parameters can be greatly reduced,and the network structure can be expanded to extract feature information at different scales,thus improving the diagnostic accuracy of the model.In addition,adaptive batch normalization(Ada BN)is introduced to overcome the sample distribution difference between the source and target domains and enhance the model’s ability to diagnose under varying loads.(2)To enhance the fault diagnosis performance of the network model under strong noise,variable load,and complex environments,a novel bearing fault diagnosis method that combines the improved residual network and the Convolutional Block Attention Module(CBAM)is proposed.The improved residual module is used to enhance the network’s anti-interference ability,while the CBAM attention module generates dual attention weights for channel and spatial information to improve the network’s ability to identify important features,enabling the network to maintain high accuracy under complex fault diagnosis conditions.(3)For the problem that fault data are difficult to obtain in practical engineering,which leads to poor training of network models,an application of migration learning based model in bearing fault diagnosis is constructed.Firstly,the source domain data is used to train the built network model to get a pre-trained model with well-defined structure and parameters;then some parameters of the pre-trained model are migrated by combining migration learning,and a small amount of target domain data is used to fine-tune the model;finally,a network model with better rolling bearing fault diagnosis is obtained.