Fault Diagnosis Method Of Rolling Bearing Based On Deep Transfer Metric Learning

Research on fault diagnosis methods driven by mechanical "big data" has become the focus and hotspot in the field of equipment fault diagnosis.Deep learning is widely used because of its strong automatic feature extraction capabilities and advantages in processing high-dimensional and non-linear big data.However,mechanical big data has the problem of many normal samples and few fault samples.Transfer learning can apply the knowledge of the source domain to different but related target domains,which can effectively solve the problem of few failure samples.To this end,this paper combines the characteristics of deep learning and transfer learning to carry out the research of deep transfer method in rolling bearing fault diagnosis.First of all,to deal with the problem of actual machine failure samples small and inconsistent data distribution under different working conditions,resulting in low diagnostic accuracy of traditional machine learning methods,a deep learning bearing fault diagnosis method based on feature transfer is proposed.The transfer component analysis method(TCA)is used in the auxiliary data set of different working conditions to map the source domain data and the target domain data to the potential Hilbert space,so that the source domain and target domain sample sets are more similar,and the maximum mean deviation embedding method(MMDE)is used.Determine the source domain data that can be transferred,and transfer the source domain samples to the target domain,provide sufficient training samples for deep learning,and solve the problem of fewer actual failure samples.In addition,in order to solve the problem of low diagnostic reliability caused by a single kernel function as a mapping function,a combined kernel function semi-supervised migration component analysis(CFSSTCA)algorithm model is proposed,And use the DBN model to train the source domain samples,and perform fault diagnosis and analysis on the unmarked target domain samples.The diagnosis and analysis of the public bearing data under different working conditions verify the effectiveness and high accuracy of the proposed method CFSSTCADBN,and the accuracy is higher than other methods.Secondly,in order to solve the problem that the data samples at the fault boundary are difficult to distinguish and the physical meaning of the diagnosis is difficult to understand,a bearing fault diagnosis method based on Yu norm deep transfer metric learning is proposed,which is completed using the combined kernel function semisupervised transfer component analysis(CFSSTCA)algorithm.The transfer of data samples under different working conditions,combined with the Yu norm similarity measurement criterion,in the depth measurement network model,the largest difference between the classes is the smallest within the class,while minimizing the top training and test samples of the learning network.The distribution difference verifies that the diagnostic model DCFSSTCAML-YU has a good diagnostic effect in fault diagnosis.Finally,with the rolling bearing vibration test bench as the object,the above two methods were used to diagnose and analyze the bearing fault data of different working conditions.The average recognition rate of CFSSTCADBN for the same working condition reached 92.61%,the highest recognition rate of a single diagnostic task reached 95.56%,and the average recognition rate of DCFSSTCAML-YU for different working conditions It reaches 93.75%,the highest recognition rate of a single diagnostic task reaches 96.56%,and the diagnosis accuracy is higher.It can be seen that the diagnosis method proposed in this paper can effectively diagnose and analyze the mechanical faults under a small sample fault,which enriches the fault diagnosis theory and methods.