Rolling Bearing State Evaluation And Remaining Useful Life Prediction Based On Machine Learning

With the blossom of manufacturing industry by leaps and bounds,the application of rotating machinery equipment is becoming more complicated,the fault diagnosis of rotating machinery has gradually shifted from relying on expert experience to intelligent processing.Bearing is one of the key parts,so how to evaluate the state and predict the life is the core content.To solve this problem,machine learning is introduced in signal processing of rotating machinery,and the research on rolling bearing state evaluation and residual life prediction are carried out.In view of the isolation of different feature selection algorithms in the diagnosis of bearing,a feature selection algorithm based on network variable selection and feature entropy for rolling bearing diagnosis is proposed in this thesis.The algorithm makes full use of complementarity between the feature mean impact value based on network and the feature entropy in feature selection and feature classification.The standard is applied to the screening of sensitive feature sets,and then evaluate the state by the support vector machine.The experimental results show that the method presented in this thesis has higher classification accuracy because of the fusion of different feature selection algorithms.Aiming at the limitation of the bearing state of degradation modeling method,this article embarks from the monitoring data of one dimensional mechanical equipment,in-depth study based on data driven bearing degradation state modeling method.In order to make full use of the influence factor of feature changes in the remaining useful life prediction problem of rolling bearings under limited state data,as well as the correlation between the feature and the time,this thesis proposes a feature selection method based on variable correlation and time correlation.In this model,MIV algorithm is used for feature selection at first,which meets the most demands of regression network for the first selection of variables.In addition,the separability measure of residual features is calculated by the correlation coefficient identification,which implements the second feature selection based on time correlation.Then the bearing degradation curve was obtained through RNN.Finally,particle filter is used to obtain the remaining useful life.The results show that compared with other algorithms,the accuracy of this algorithm is improved obviously.Aiming at the limitation of the artificial feature modeling method,this thesis studies the bearing state of degradation modeling method based on the deep learning,this thesis proposes a bearing degradation modeling method based on the convolutional autoencoding network,on the basis,the algorithm takes into account the evaluation tendency of the bearing degradation state model,and carry on the improvement of the objective function of the simulation.The results show that the algorithm can not only ignore the working condition,but also have good improvements in accuracy and stability.