Research On Cross-Domain Intelligent Fault Diagnosis Method Of Rolling Bearing Based On Transfer Learning

Rotating machinery plays a vital role in many industries.Rolling bearing is an important part of rotating machinery,which has higher requirements for its operational safety.Therefore,it is necessary to monitor and diagnose the operation process.With the continuous improvement of scientific level,the fault diagnosis approaches based on artificial intelligence technology have been developed by leaps and bounds.However,from a practical application point of intelligent fault diagnosis methods,the existing intelligent diagnosis models still have disadvantages such as the lack of practical deployability to meet the demands in actual application.In order to break the constraints of the above-mentioned practical factors,this thesis established the cross-domain intelligent fault diagnosis framework of rolling bearing,the main contents are as follows:The method theory of transfer learning and three types of typical transfer learning models based on sample instances,feature information and model parameters are introduced.In addition,the reasons for the vibration during bearing operation are revealed from the rolling bearing dynamics model.Combining with the signal characteristics reflected during its operation,the transferability between different fault data sets of rolling bearings and the transferability of intelligent fault diagnosis tasks based on rolling bearings are analyzed.Focusing on the problem that the existing transfer diagnosis model is difficult to dynamically adjust for different fault diagnosis tasks,a deep dynamic domain adaptive network model is proposed.The difference between the marginal probability distribution and the conditional probability distribution between the domains data,the transfer diagnosis model can effectively obtain more source domain fault feature information.In the case study part,the proposed adaptive network model is validated by the transfer diagnosis between different working conditions,as well as the diagnosis task between different types of bearings.Aiming at the problem that the complete fault data required by the existing transfer diagnosis model is difficult and time-consuming to obtain in practical application and the actual deployability of the model is insufficient,a time-varying multi-source online transfer diagnosis model is proposed.In the offline stage,only limited unlabeled target machine operating data is needed to construct the initial diagnosis model.In the online stage,after the model collects online instances of the target machine,parameters will be updated to obtain renew fault feature representations.Furthermore,the proposed transfer diagnostic model with multiple source domains,which could provide more transferable diagnostic knowledge between domains.By comparing with other diagnosis methods,the convenience and the practical deployability of the proposed method are demonstrated.