Life Prediction Of Rolling Bearing Based On Deep Learning

Rolling bearing is an important connecting part of mechanical equipment.According to statistics,70% of the damage of large mechanical structure is related to bearing damage.Therefore,researching the remaining useful life(RUL)of rolling bearing is conducive to formulating reasonable failure maintenance measures for mechanical equipment in advance and greatly reducing the accidents and losses caused by bearing failure.In this paper,the deep learning model Long Short-Term Memory Neural Network(LSTM)is used to construct the RUL prediction of rolling bearings combined with signal processing and degradation index.Firstly,the whole life test of rolling bearings is designed in this paper.Through the radial force loading device,the whole life vibration signal of rolling bearings can be obtained in a short time.As the original data of this paper,the validity of this method is verified.Then,the noise reduction method combining Complementary Ensemble Empirical Mode Decomposition(CEEMD)with improved wavelet threshold de-noising is used to process the vibration signal of rolling bearing,and CEEMD is used to decompose the vibration to obtain several intrinsic mode functions(IMF)through the one-sided peak of autocorrelation function.The critical point of signal-noise separation of IMF component is determined by proportion.The noise IMF component is de-noised by improved wavelet threshold.The de-noised IMF component and the effective signal IMF component are reconstructed to get the final de-noised signal.Thirteen common time and frequency domain features of vibration signals are extracted from the denoised signals.The features are filtered according to the correlation coefficient method(CC).The dimensionality of the filtered features is reduced by the Kernel Principal Component Analysis(KPCA).Several principal components are obtained according to the contribution rate.Mahalanobis Distance(MD)is used to fuse the principal components to obtain the final degradation index of rolling bearings.Finally,using degradation index as input data,the rolling bearing's RUL is predicted by LSTM,and the KPCA-MD-LSTM model of rolling bearing life prediction is completed.The prediction results of the model are evaluated by the scoring value of evaluation index.In order to prove the validity of this method,the RUL prediction results of KPCA-MALSTM rolling bearing model in this paper are compared with those of LSTM which directly predicts the life of vibration signals,KPCA-MD-LSTM model without noise reduction,traditional PSO-BP neural network model and LS-SVR model,which prove that the prediction results of this model are more accurate.It is also shown that the noise reduction method,the degraded index constructed and the predictive network selected in this paper can improve the accuracy of rolling bearing life prediction.