Research On Remaining Useful Life Prediction Of Rolling Bearing Based On Ensemble Learning

As an important part of modern mechanical equipment,rolling bearing is also one of the most vulnerable parts of mechanical equipment.Accurate prediction of the remaining useful life of rolling bearing can improve the reliability of equipment and avoid accidents,which is of great significance to the safe operation of mechanical equipment.This paper takes the vibration signal of rolling bearing as the research object,starting from the theoretical research of data-driven,through time-frequency domain analysis,ensemble learning,deep learning and so on,realizes the remaining useful life prediction of rolling bearing.The main research work is as follows:Firstly,aiming at the problems of noise interference and unobvious degradation trend existing in the original features,a remaining useful life prediction method of rolling bearing based on manual feature extraction is proposed.The vibration signals are extracted from time domain,frequency domain and time-frequency domain respectively.Based on the above characteristics,the XGBoost algorithm is used to realize the remaining useful life prediction of rolling bearing,and the results are compared with other classical regression algorithms on the PHM2012 data set.The results show that the XGBoost can better combine the above characteristics,and the prediction accuracy and score are better than other methods.Secondly,in order to solve the problem of a large average error rate of the prediction model in the above research,the remaining useful life prediction of rolling bearing based on automatic feature extraction is proposed.Firstly,the Hilbert envelope spectrum is used to process the original vibration signal,which is more sensitive to the impact signal when the fault occurs.Furthermore,convolution neural network is used for automatic feature extraction and dimension reduction.Finally,through the combination of Bi-directional Gated Recurrent Unit which is more sensitive to the sequence information and attention mechanism which can mine the importance of sequence information,the remaining useful life prediction of rolling bearing is completed.Finally,the experimental results show that,compared with the first method,the average prediction error rate of this method is reduced by 36%,and the overall error rate is better.Finally,the prediction accuracy of the above two methods on different bearings has their own advantages and disadvantages.In order to get a better prediction model,the remaining useful life prediction of rolling bearing based on stacking ensemble is proposed.In this method,support vector regression is introduced as a secondary learner,and the first two methods are integrated as a base learner.Compared with the above two methods,the average error rate is increased by 47.5% and 11.5% respectively,and the score is improved by 0.17 and 0.04,so a more accurate prediction model is obtained.