Joint ResNeXt Domain Adaptation Diagnosis Method for BearingFault Under Variable Working Conditions
Rolling bearing has cross-domain problem due to different distribution of fault data features caused by changes in actual working conditions,which is difficult to be solved by the conventional fault diagnosis method that takes the independent co-distribution of data as the prerequisite.Therefore,a ResNeXt was designed and built to achieve deep mining of fault sensitive features through a unique network structure of residual connection and grouped convolution.Then,a Joint ResNeXt Domain Adaptation fault diagnosis method was proposed,which uses the Optimal Generalized S-Transform to build a ResNeXt to extract the transferable features of images.Finally,the Joint Maximum Mean Discrepancy was used to adaptively reduce the joint distribution difference among data,and achieve fault diagnosis of bearing under variable operating conditions.Moreover,6 sets of migration tests were car-ried out on rolling bearing under 3 kinds of working conditions.The test results show that the fault diagnosis accura-cy of the Joint ResNeXt Domain Adaptation method reaches 98.29%,which is improved by 21.0%and 5.1%compared to the joint distribution adaptation method(JDA)and the joint distribution adaptation + convolutional neural network method(JDA+CNN)respectively.The study results provide technical reference for the fault diag-nosis of rolling bearing under variable working conditions.
rolling bearingfault diagnosisResNeXtjoint ResNeXt domain adaptationfault diagnosis accuracy
万方数据
维普
