Prediction Of Residual Life Of Rolling Bearing Based On Time-frequency Analysis And Gray Level Co-occurrence Matrix

As one of the most commonly used core basic components in mechanical equipment,the health status of rolling bearings can affect the precision and safety of the entire mechanical system.Effectively and accurately predicting the remaining service life(RUL)of rolling bearings can provide guidance for the maintenance of mechanical equipment and save manpower and material resources.In order to predict the RUL of rolling bearings in complex environmental conditions,this paper proposes a residual life prediction method for rolling bearings based on time-frequency analysis and gray level co-occurrence matrix.Currently,most RUL prediction methods for bearings directly use raw data containing noise,resulting in incomplete analysis of degraded features and low prediction accuracy.This article proposes the use of improved EWT for denoising vibration signals.First,the initial dividing points are filtered based on mutual information to reduce the number of bands.Secondly,the optimal component is selected based on kurtosis for signal reconstruction,ultimately achieving denoising of bearing vibration signals.Considering that direct deep learning processing of time-frequency images can lead to issues such as large data volume,long network training time,and high requirements for computer performance.After obtaining the time-frequency map of the reconstructed signal through CWT,this article calculates the grayscale co-occurrence matrix of the time-frequency map and extracts the relevant features of its grayscale co-occurrence matrix as the time-frequency feature vector of the bearing.On the basis of the above,the time-frequency characteristics are input into the convolutional neural network(CNN)to construct the bearing degradation index HI,and the residual life of the bearing is predicted with the degradation index HI based on least squares support vector regression(LSSVM),and other three models are used to compare with the proposed methods,including the long short memory method(LSTM),random forest(RF)and deep neural network(DNN).This article was validated on the PHM dataset.By comparing the decomposed components and the time-frequency maps before and after denoising,it can be seen that the improved EWT can effectively reduce the number of frequency bands.After denoising,the high-frequency parts containing noise in the time-frequency map of the reconstructed signal are significantly reduced,but the low-frequency parts containing fault information are all preserved,proving the effectiveness of the proposed method in removing noise.Secondly,this article constructs a bearing degradation index,CNN-HI,which shows a monotonically increasing trend with the increase of bearing time.Finally,compared with the other three models,the residual life method based on LSSVM bearings proposed in this paper is closer to the actual values and has smaller errors.