Bearing Fault Diagnosis Based on Multiscale Residual Attention Domain Adaptation
Rolling bearing samples to be monitored are difficult to diagnose in cross machine tasks.To solve this problem,a fault diagnosis method for rolling bearing based on multiscale residual attention domain adaptation was proposed.This method directly takes the vibration signal of rolling bearings as the input of the multiscale atten-tion residual network module.In order to effectively extract the shared characters of source domain and target do-main,this module introduces multiscale convolution to extract feature information,and a compressed excitation net-work with attention mechanism to solve the problem of data differences and cross layer connection of residual net-works.The domain adaptation part adopts the local maximum mean difference measurement criterion,and selects the publicly available fault dataset of rolling bearings to conduct comparison and ablation tests.The test results show that the rolling bearing fault diagnosis method based on multiscale residual attention domain adaptation a-chieves an average recognition accuracy of 99.1%in cross machine tasks,and has better generalization perform-ance compared to other methods.The study conclusions provide a theoretical basis for the monitoring and diagnosis of rolling bearing faults.
rolling bearingfault diagnosis modeltransfer learningmultiscale convolution kernelattention residual block
万方数据
维普
