Study On Remaining Useful Life Of Rolling Bearings Based On Adaptive Feature Selection And Iterative Relevance Vector Machine

Rolling bearings are widely used in aerospace,rail transit and other large-scale machinery because of their convenient assembly and low frictional resistance.They are one of the most important components in rotating machinery and their performance status is directly Affect the operating status of the entire device.Life is one of the important indicators to measure the performance of rolling bearings,but the actual data shows that the service life of rolling bearings is very discrete.Under the same batch conditions,the difference between the minimum life and the maximum life is several tens of times,so it is in operation.Rolling bearing health monitoring and residual life prediction are essential.Accurate residual life prediction can detect rolling bearing damage and deterioration trend as early as possible,provide data support for economically rational maintenance strategy,minimize production accidents and improve economic benefits.In recent years,datadriven machine learning methods have become widely used in the field of mechanical equipment fault diagnosis and life prediction.This paper will also adopt the data-driven method,three key steps around the prediction of residual life of rolling bearings: vibration feature extraction and selection,performance degradation state classification,residual life prediction research,the main research contents are as follows:(1)For the problem of selecting the appropriate vibration signal to characterize the state of the rolling bearing,the vibration characteristics of the rolling bearing are extracted from the time domain,the frequency domain,the time-frequency domain,the information entropy,etc.,and the feature library is constructed to comprehensively describe the state of the rolling bearing.information.An adaptive feature selection method is proposed.By adding white noise features and fusion features,and using correlation,monotonicity and robustness to comprehensively evaluate features,it is possible to automatically determine feature dimensions and screen out sensitive features.set.The problem of optimizing feature subsets from a large number of features is solved.The effectiveness of the proposed method is verified by experimental data,which lays a foundation for the classification of rolling bearing performance degradation and residual life prediction.(2)Aiming at the problem of classification of rolling bearing performance degradation state,an index that can achieve fast classification,relative energy accumulation ratio(REAR)and Bayesian optimization based SVM classification method are proposed.This method takes into account the multi-classification problem in performance degradation and the impact of category data imbalance.The problem that the traditional mesh optimization SVM training time is too long is solved,and the initial prediction time is provided for the prediction of residual life.The validity of the proposed method is verified by experimental data.(3)According to the problem of residual life prediction of rolling bearings,based on the idea of integrated learning and promotion,an iterative RVM prediction model with equal weight residuals and an iterative RVM prediction model with weighted residual optimization are proposed.The experimental data proves that the proposed two models can effectively improve the accuracy and reduce the error compared with the traditional RVM prediction model.(4)Based on LabVIEW and MATLAB language,the proposed vibration feature extraction algorithm,adaptive sequential optimal feature selection algorithm,undersampled Bayesian optimization SVM classification method,weighted residual optimization iterative RVM prediction model,etc.An online monitoring system for rolling bearing operating conditions.The system can realize the collection of vibration and temperature signals,data storage and playback,operational status judgment and residual life prediction.