Research On Prediction Method Of Remaining Useful Life Of Rolling Bearings

With the integration and intelligence of modern equipment,once someone link fails,it is easy to cause casualties or the entire system to shut down,which could cause heavy economic losses.Therefore,the Prognostics and Health Management of the system based on industrial big data is getting more and more attention.As an important part of most industrial equipment,the health performance of rolling bearings affects the operation of many equipment.If the health of the bearing can be monitored and then the bearing can be replaced or repaired in advance,the useful life of the equipment can be prolonged.In recent years,many scholars have begun to research the remaining useful life prediction method,so as to achieve the purpose of health management of systems.Therefore,predicting the remaining useful life of the rolling bearing can improve the reliability and safety of the systems,and has high practical application value.This subject constructs fault indicators based on the time domain feature s of the bearing vibration signal,and then detects initial degradation points of bearing s to obtain the data of the degradation stage.On this basis,this subject extracts time domain,frequency domain and time-frequency domain features according to the characteristics of bearing vibration data,and then selects features based on the trend and correlation indicators to obtain features with degradation trend.In the process of establishing the remaining useful life prediction method based on health indicators,this subject firstly constructs health indicators in the time domain,frequency domain,and time-frequency domain based on Kernel Principal Component Analysis,and then reconstructs the health indicators based on the sum function of the double exponential function and the modified Hausdorff distance.This subject trains proposed Auto-encoder Correction Model based on reconstructed health indicators.According to the trained model,the health indicators can be corrected in real time to improve the anti-interference of the health indicators.Therefore,the performance of remaining useful life prediction can be improved.After obtaining the corrected health indicators,this subject establishes a multiscale gated recurrent unit prediction model,and then directly predicts the remaining useful life of rolling bearings from multiple scales in the time domain,frequency domain,and time-frequency domain.In addition,this subject adds dropout and step attention mechanisms to the model,to improve the generalization ability and prediction performance of the model.At the end of this paper,this subject uses the XJTU-SY bearing data set and the PHM2012 data set to verify and compare the model.The analysis results show that the method proposed by this subject can predict the remaining useful life well,and has good applicability to different data sets.