Research On Prediction Of The Remaining Life Of Rolling Bearing Based On Deep Learning

Rolling bearings in modern industrial equipment are one of the indispensable components,and the wide range of applications determines the necessity of their degradation assessment and remaining life prediction.In order to provide a theoretical basis for enterprises to formulate equipment maintenance plans reasonably,this paper takes rolling bearings as the research object,and conducts research on the core feature extraction,degradation assessment and remaining life model modeling in the process of prediction of the remaining life of rolling bearings.(1)Improve the wavelet threshold noise reduction of the rolling bearing vibration signal,extract the rolling bearing time domain,frequency domain and frequency domain feature information through Fourier transform and empirical mode decomposition,and extract multifrequency scale sample entropy through complementary set empirical mode decomposition combined with correlation coefficients Feature,a feature screening method based on Fisher scores is used to remove insensitive features with low scores,and a multi-domain feature set sensitive to performance degradation of rolling bearings is constructed.(2)The basis function is used to establish the degradation assessment model for the Weibull distribution proportional fault model.The nuclear principal component analysis is used to analyze the extracted multi-domain features of the rolling bearing,and the extracted nuclear principal components are used as the covariates of the model to estimate the model parameters.Finally,the reliability of the bearing is calculated according to the model,and the length after parameter optimization is calculated.The short-term memory neural network is used for reliability prediction.(3)In view of the shortcomings of large amount of artificial feature extraction and large subjective influence by humans,a long and short-term memory neural network prediction model based on deep feature extraction is established to predict the remaining life of rolling bearings;for non-full life cycle data,the length of real-time update of parameters is proposed.Time memory neural network model.(4)After the start prediction point is selected,and the remaining life prediction model parameters are optimized and the simulation is carried out.It is verified that the prediction effect of the remaining life of rolling bearing based on deep feature extraction is better than that of manual feature extraction,and it is proved that the long and short-term memory neural network model with real-time parameter update has good applicability to non-full life cycle data.