Research On Fault Diagnosis Method Of Rolling Bearings Under Different Working Conditions Based On Federated Multi-representation Domain Adaptation

Rolling bearing is the core component of rotating machinery and widely used in social production.At the same time,rolling bearings is also a component prone to failure.Once a failure occurs,it may lead to serious accidents.However,it is often difficult to collect enough labeled data to build a high-performance diagnostic model for a single user.In addition,due to data privacy and other reasons,multi-users’ data cannot be used in a centralized manner,thus forming data silos.Therefore,it is of great significance to aggregate island data knowledge of rolling bearing and establish a high-performance fault diagnosis model on the premise of ensuring user data privacy.To solve the above problems,two methods of rolling bearing fault diagnosis are proposed.(1)Aiming at the problem of insufficient data for a single user and data barriers between multiple users,a federated learning method for rolling bearings fault diagnosis is proposed.Firstly,the time-domain vibration signals of multi-user rolling bearings are transformed into the local silos data sets by wavelet transform.Everyone has only 10 samples for each type of state.Then,users use their own sample data for building local model,and the deep learning model compression algorithm is proposed to improve the federate parameter transmission strategy,further ensuring the privacy of user data and reducing communication overhead.Subsequently,the local model parameters are compressed and uploaded to the central server.Finally,the server aggregates the multi-user local model parameters to build an effective federated global model.The experimental results show that the proposed fault diagnosis model can effectively solve the problem of rolling bearing fault diagnosis under the same working condition,and the accuracy of fault diagnosis can reach 98.3%on the premise of ensuring user data privacy.The above methods effectively solve the problem of data soils and data privacy problem in the field of rolling bearing fault diagnosis,but they are not suitable for multi-user rolling bearing data acquisition from different working conditions.Therefore,a method that can not only inherit the advantages of the above methods,but also can also be applied to rolling bearing fault diagnosis scenarios under different working conditions is needed.(2)In view of the different distribution of multi-user rolling bearings data,a fault diagnosis method of rolling bearings under different working conditions based on the federal multi representation domain adaptation is proposed.Firstly,the priori labeled public data are used as the source domain and multi-users unlabeled privacy silos data are used as the target domain.Then,the multi representation feature extraction structure is introduced to improve the original residual network,align the multi representation features of the source domain and target domain data,and then build a multi-user local domain adaptation model.Subsequently,continue to use the parameter transfer strategy based on the model compression idea to transfer local model parameters.Finally,a federated global model for rolling bearing fault diagnosis under different working conditions is built on the server side.The experimental results of two bearing datasets show that the proposed method can integrate silos data knowledge without multi-user sharing data,effectively solve the fault diagnosis problem of rolling bearings under different working conditions,and the average fault diagnosis accuracy can reach 97.6%.