Research On Prediction Model Of Remaining Useful Life Of Rolling Bearing

For solving the existing model-based prediction suffer from insufficient generalizability and difficulty in modeling,and prediction accuracy based on data-driven prediction approaches is too dependent on data volume,and the problem of low fault tolerance in the actual use of a single prediction model,this thesis establishes a remaining useful life prediction model of rolling bearing of electric generator based on Agent Technology,which is to evaluate the remaining useful life of rolling bearing of electric generator through the comprehensive analysis of the prediction results of multiple agents.The details are as follows:(1)Theory and data preparation before model establishment.In order to predict the remaining useful life of rolling bearing of electric generator,the vibration phenomenon caused by bearing degradation is deeply analyzed in the thesis.In addition,the data set used in this thesis,the division of bearing degradation stage and the principle of life prediction are explored,which provides assistance for the processing of prediction label and the establishment of bearing remaining useful life prediction model.(2)A noise reduction method for rolling bearing vibration data of electric generator based on improved threshold value is proposed.For most current prediction methods,the signal collected by the sensor is not preprocessed,the problems of incomplete analysis of degradation features,and low accuracy of final prediction of bearing life when using data with noise directly are discussed.In this thesis,an improved threshold wavelet denoising method is proposed to pre-process the data.The research shows the proposed noise reduction method can denoise effectively.(3)Establishment of feature extraction network and reconstruction of bearing degradation features.In order to fully mine the degradation features of rolling bearings of electric generator contained in the data and give full play to the value of the data,not only the common statistical indexes in time domain and frequency domain are used as features,but also a feature extraction network based on feedback fuzzy neural network with internal variables is constructed to further extract the degraded features hidden in the data in the process of feature selection.Then,the time and frequency domain features are reconstructed with the features extracted by the network,and the remaining useful life of the bearing is predicted by the new features obtained from the reconstruction.In addition,the use of fuzzy neural network improves the ability of the overall model to deal with dynamic information and fuzzy information.Research shows the reconstructed features have more information to describe the bearing degradation phenomenon,which improves the prediction model’s accuracy.(4)Establishment of rolling bearing remaining useful life prediction model.Aiming at the single prediction model is not fault-tolerant and persuasive in practice,remaining useful life prediction model is built by multi-Agent technology in this thesis: predict separately using multiple forecast agents first,then the predicted value of the prediction agent is integrated by integrating Agents,and the predicted result is output.Research shows the multiAgent prediction model can improve the stability and robustness of single model prediction to some extent.In addition,for improving the model’s ability to perceive "external environment" changes,a feature extraction agent is constructed,which bases on the fuzzy neural network to select an appropriate membership function based on the data changes and enhances the model’s ability to process data.This thesis uses the data in the XJTU-SY dataset to train the model and verify the performance of the model.The results show the multi-agent rolling bearing of electric generator remaining useful life prediction model constructed in this thesis has good prediction effect and anti-risk ability.