Research On Residual Life Prediction Method Of Rolling Bearing Based On Deep Learning

The working condition of rolling bearings is closely related to the safe and stable operation of mechanical systems.Therefore,it is extremely necessary to monitor the operating status and predict the remaining life of rolling bearings.In the process of predicting the remaining life of rolling bearings,the quality of vibration signals,characteristic indicators characterizing degradation performance,and residual life prediction models will directly affect the accuracy of the prediction.Therefore,this article conducts research on vibration signal noise processing,multi domain feature set construction methods for characterizing degradation performance,and LSTM prediction models to improve the accuracy of residual life prediction.Firstly,the VMD denoising method for vibration signals was studied,and an adaptive particle swarm optimization algorithm was proposed to optimize the key parameters of VMD:decomposition levels and penalty coefficients.Then,the method proposed in this paper was applied to the simulated and measured signals of rolling bearings.On the simulated signals,it was compared with the EMD method and CEEMD method.The signal-to-noise ratio of this method was increased by 1.56 and 0.57 respectively,and the mean square error was reduced by0.27 and 0.3,respectively.On the measured signal,by comparing the effects of parameter optimization before and after,the results show that the signal-to-noise ratio of this method is improved by 2.4 decibels,and the mean square error value is reduced by 1.95.Secondly,in order to extract effective features that characterize the degradation performance of rolling bearings,the Fisher score method is used to filter the time-domain,frequency-domain,and time-frequency domain feature sets,and the filtered features are fused using kernel parameter optimized KPCA.Finally,this article uses a combination of reliability evaluation and characteristic parameters to determine the remaining life prediction points.Predict the remaining life of rolling bearings using the LSTM prediction model.The method proposed in this paper was applied to the rolling bearing dataset of the University of Cincinnati in the United States,and compared with LSTM and BP neural network prediction methods that did not start predicting point selection.The root mean square error of the proposed method was reduced by 0.0393 and 0.0693,respectively.