Considering the lack of fault samples of rolling bearing in fault diagnosis,a fault diagnosis model of rolling bearing based on S-LSSF is proposed with StyleGan2-ada applied to the field of bearing fault diagnosis.Firstly,the time-domain vibration signal was transformed into a time-frequency image by continuous wavelet transform,which was input into StyleGan2-ada to generate corresponding samples.Then,original samples and generated samples were combined and input into the improved ShuffleNetV2 model.In the process of back propagation,the loss function of LabelSoomthloss was introduced to reduce the influence of wrong labels on the diagnosis performance of the model,and to further avoid over-fitting,and the LeakyReLU function was introduced to the down-sampling unit to solve the vanishing gradient problem.The experimental results show that compared with the original model,the diagnostic accuracy of S-LSSF model is improved by 1.9%,and the average time is reduced by 5 s.Compared with original samples,the accuracy rate,precision rate,recall rate and F1 score of the generated samples are improved by 3.58%,5.71%,6.15%and 6.06%respectively,which verifies the feasibility and generalization of S-LSSF model in bearing fault diagnosis with a small sample size.
关键词
滚动轴承/样式生成对抗网络/连续小波变换/小样本故障诊断
Key words
rolling bearing/style generative adversarial network/continuous wavelet transform/fault diagnosis with small a sample size
维普
万方数据