Research On Useful Life Prediction Of Rolling Bearing Based On Pearson-KPCA Multi-feature Fusion

Rolling bearings are the most widely used important basic components in rotating machinery and equipment,and their operating status is closely related to the overall working performance of the equipment.Therefore,predicting the changing trend and remaining life of rolling bearing performance can effectively avoid dangerous accidents.It can also reduce maintenance costs,which has important practical significance for improving the health management level of rotating machinery.This paper takes the full-life vibration signal of rolling bearings as the research object,adopts data-driven research methods,and gradually develops three aspects of signal noise reduction,characteristic analysis and life prediction,and provides technology for early warning of mechanical equipment failure and formulation of maintenance strategies.support.The main research contents are as follows:Firstly,it is suggested that the bearing vibration signal denoising algorithm is based on the convolutional neural network of self-coding.In the self-encoder method of encoding and decoding,the convolution neural network and deconvolution neural network are respectively combined,and then the built convolution self-coding neural network is used to learn end-to-end mapping from the noisy bearing vibration signal to the pure bearing vibration signal.Hence,to complete the noise reduction of the bearing vibration signal,the pure bearing vibration signal can be derived from the noisy bearing vibration signal.The simulation modulated signal measurement analysis reveals that a magnificent denoising effect has been accomplished by the denoising algorithm based on the convolutional self-coded neural network under the assumption of maintaining maximum efficient signal integrity.Secondly,a feature fusion method based on Pearson correlation analysis is proposed.In view of the traditional life prediction using a single time domain or frequency domain index analysis and prediction,which cannot take into account the local and overall characteristics of the vibration signal.The multi-category features are extracted for this in the time domain,frequency domain,and time-frequency domain that can represent the bearing’s degradation condition.After that,the Pearson correlation analysis is carried out on the extracted multi-domain and multi-category features to obtain the correlation between features and divide the feature groups based on this.At last,The KPCA is used to fuse the features of each group to obtain the feature sets that can better reflect the degradation of the bearing performance.In the next one,based on the long short-term memory neural network and Cox proportional fault model,a rolling bearing life prediction approach is proposed.To predict the pattern of multi-dimensional bearing characteristics,the output deterioration prediction model of rolling bearing based on the long-term short-term memory neural network is developed.In order to evaluate the bearing reliability and failure rate,the Cox proportional fault model based on the Weibull distribution was developed.Then,integrate them as the multidimensional covariate of the Cox proportional fault model by using the expected multidimensional bearing characteristic recession pattern to estimate the bearing’s reliability and failure rate to adjust over time.The prediction methodology has superb prediction precision,robustness,and generalization capability tested via the calculated carrying maximum life cycle data collection.Finally,to validate the noise reduction approach suggested in this article,the bearing fault simulation experiment platform is installed and the vibration signal of its operating phase is obtained.The bearing life cycle test bench is built to obtain the bearing life cycle test data and verify the life prediction method which are proposed in this paper.