A Research On Fault Diagnosis And Remaining Useful Life Prediction Of Rolling Element Bearings Based On Residual Networks

Rolling element bearing is one of the key components of rotating machinery,its health condition directly affects the safety and reliability of machinery.Therefore,fault diagnosis and remaining useful life prediction of rolling element bearings can not only avoid equipment accidents,ensure its safe and stable operation,but also provide the foundation for a planned maintenance of the machines.This paper uses deep learning techniques for fault diagnosis and remaining useful life prediction of rolling element bearings.The main contents are as follows:(1)A rolling element bearing fault diagnosis technique based on temporal residual network is proposed in this work to overcome the requirement of manual input in the traditional bearing fault diagnosis approach.In this approach,temporal residual block is constructed by adding residual connections to the temporal convolutional network,and three time series residual blocks are stacked to form a temporal residual network model.Fault type signals are classified.The validity and accuracy of the model’s fault diagnosis are verified using the published CWRU bearing fault dataset and a set of gearbox defect data,and the results show that the model can accurately identify bearing and gearbox faults with different damage degrees.(2)Aiming at the problems of complex construction of degradation indicators and low prediction accuracy in the remaining useful life(RUL)prediction methods of rolling element bearings,a RUL prediction technique combining deep residual network(DRN)and gated recurrent unit(GRU)is proposed.The DRN-GRU model is constructed by stacking the DRN and GRU networks.The DRN is used to extract the spatial characteristics of the vibration signal,and then the GRU part is used to capture the timing information of the vibration signal and obtain more comprehensive and deeper level degradation characteristic information,thereby improving the prediction accuracy.The validity and accuracy of the proposed prediction model are validated using the PHM 2012 bearing acceleration degradation dataset.(3)To further improve the prediction accuracy of bearing remaining useful life,the DRN-Bi GRU prediction model is proposed where a bidirectional gated recurrent unit(Bi GRU)is employed to replace the GRU layer of the preceding.The effectiveness of the model is then tested using the PHM 2012 and XJTU-SY bearing degradation datasets.It is showed that the proposed model can yield a relatively high accuracy in bearing RUL prediction.