基于SE-ResNet和迁移学习的轴承故障诊断方法研究OA

In response to the problem of low accuracy and poor generalization in traditional rolling bearing fault diagnosis under small sample conditions,a bearing fault diagnosis method based on SE-ResNet and transfer learning is proposed in this paper.Firstly,the data is preprocessed and transformed into two-dimensional images.Secondly,the attention mechanism of the Squeeze and Excitation network structure is introduced into the residual blocks of the neural network to improve the deep residual network model.Then,the transformed images are fed into the SE-ResNet model for classification of the rolling bearing states in the test samples.Finally,by using transfer learning to utilize a model trained on small samples as a pre-trained model,experiments are conducted with data from different working conditions.This approach is compared to 2D-CNN,ResNeXt-50,AlexNet,and BiLSTMmethods.The results demonstrate that the accuracy of SE-ResNet under single working conditions in CWRU datasets ranges from 87.09%to 99.61%,while the accuracy of the SE-ResNet+transfer learning method ranges from 92.8%to 99.54%and 85.1%to 100%under the working conditions of the two datasets(CWRU、XJTU-SY)in this paper,respectively.These results indicate that the SE-ResNet+transfer learning method exhibits superior accuracy and generalization performance compared to other methods.