Research On Bearing Degration Trend Index Construction And Prediction Method Based On Recurrent Neural Network

Rolling bearing are momentous fundamental components extensively used in the transmission system.However,under the condition of heavy load and high speed for a long time,it’s easy to cause performance degradation and failure.Accurate characterization and prediction of bearing performance degradation trend can strengthen the health management of machinery and equipment,ensure the safety of operation during start-up and avoid the occurrence of accidents.Focusing on bearing performance degradation trend characterization and prediction,this paper studies the bearing degradation process,and explores and analyzes the feature extraction,feature construction and prediction of the degradation trend.The main contents of the paper include:(1)In order to fully consider the influence of the vibration law at the historical moment and the future moment on the bearing degradation status at the current moment,the bidirectional recurrent neural network(BRNN)is proposed to study the degradation trend.The monotonicity index is constructed to screen the features with the best monotonicity among the parameters extracted from several domains.Based on the accelerated degradation data set,the prediction is completed by using the BRNN.Compared with the prediction results of the RNN network,the proposed method is better.(2)Aiming at the efficient use of the detailed information of the data collected in the accelerated degradation test,which is conductive to trend prediction,a method based on MRSVD to extract bearing degradation trend features is proposed.By decomposing the original signal,extracting weighted fusion features that are sensitive to weak faults in the early stage,and strengthening the sensitivity of the features to early weak faults.In order to solve the problem that the gradient does not update the weight of the RNN structure in the late stage of bearing degradation,a bidirectional long-short time memory(Bi-LSTM)network is constructed to predict the performance degradation trend of bearings,and the effectiveness and robustness of the proposed method are verified in bearing acceleration degradation data sets.(3)Aiming at the problems that the bearing degradation information cannot be fully represented by a small number of characteristic parameters,and the LSTM network iteration is slow,and the monitoring signals with a large amount of data cannot be predicted in time,a trend prediction method based on the denoising auto-encoder(DAE)and the bidirectional gated recurrent unit(Bi-GRU)network is proposed.The RMS were taken as the reference,and the multi-feature parameter set was constructed through the correlation index.The effect and significance of multi-step trend prediction were discussed.Compare the proposed method to single-step RNN,the effectiveness and generalization of the method are proved in bearing degradation data sets.