Research On Remaining Useful Life Prediction Of Rolling Bearing Based On MFPE-MACNN

Rolling bearing is a key component in most mechanical equipment,and the health status of mechanical equipment is closely related to the health status of rolling bearing.The study on remaining useful life prediction method of rolling bearing can prevent the mechanical equipment from failure in advance,which leads to downtime or casualties,and is essential to improve the efficiency of mechanical equipment.This thesis takes rolling bearing as the research object,takes the whole life cycle vibration signal generated by the operation of rolling bearing as the research data to carry out the data pre-processing research of vibration signal noise reduction and feature extraction,and further researches the rolling bearing life prediction model on this basis.The main research contents of this thesis are as follows.Firstly,for the problem of improving the integrity of signal feature information,a signal noise reduction method based on the ensemble empirical mode decomposition（EEMD）method and wavelet semi-soft thresholding（WSST）method is proposed.The EEMD method is used to decompose the rolling bearing vibration signal into Intrinsic Mode Function（IMF）,the Pearson correlation analysis is used to screen the IMFs,and the WSST method is used to remove the noise components in the IMFs with high Pearson correlation coefficients,and the overall signal noise reduction is completed after the screened IMFs are completed.The simulation comparison test results show that the proposed noise reduction method can ensure the integrity of the vibration signal feature information after noise reduction,and the noise reduction effect is good,the signal-to-noise ratio is improved by 2.2959 on average,and the root mean square error is reduced by 0.3484 on average.Secondly,for the problems of signal feature information mixed with noise components and the low sensitivity of features to the remaining life decline trend,feature enhancement methods and multi-scale fusion permutation entropy feature extraction methods are proposed.To enhance the feature information in the signal,the low-resonance components in the signal are extracted using the resonance sparse decomposition method;the low-resonance components are reconstructed into the short-time sequence matrix using the sliding window slicing processing method to enrich the feature information in the signal;the feature information in the short-time sequence matrix is enhanced using the phase space reconstruction method;the multi-scale permutation entropy value of the short-time sequence matrix is calculated,and the local linear embedding algorithm is used to extract the entropy The multi-scale fusion entropy features are obtained by using the local linear embedding algorithm to reduce the dimensional fusion of the extracted entropy features.It is verified by simulation that the multi-scale fused permutation entropy features have higher resolution in reflecting the remaining lifetime decline trend.Then,a multi-scale attentional convolutional neural network rolling bearing remaining life prediction model is proposed to address the limitations of the one-dimensional convolutional neural network in dealing with complex input sequences.A multi-scale feature extraction module containing convolutional kernels of different sizes is added to the one-dimensional convolutional neural network to improve the ability of capturing detailed features when the model learns features;an attention mechanism is added to the model to improve the expression capability after the fusion of multi-scale features.The experimental validation results show that the multi-scale feature extraction module improves the fitting ability of the model,and the attention mechanism increases the expression ability of the features and improves the prediction accuracy of the model.The average deviation between the predicted and true values of the multiscale attentional convolutional neural network model is within 6 minutes,and it can predict the remaining life of rolling bearings well,with good generalization and practicality.Finally,the experimental validation using XJTU-SY whole life cycle actual measurement data set shows that the noise reduction algorithm,feature extraction method and remaining useful life prediction model proposed in this thesis are consistent with the conception and can reliably and effectively complete the remaining life prediction of rolling bearings.