Study On Degradation Performance Evaluation And Remaining Useful Life Prediction Of Rolling Bearings

Rolling bearings are one of the most important components in electromechanical equipment.Any qualitative failure may bring huge economic losses to the factory,and even casualties.Research on the prediction of the remaining useful life(RUL)of rolling bearings has become an urgent need and consensus in the industry today.The physical model of RUL prediction is difficult and limited due to mechanism analysis,but the datadriven model based on big data technology has the advantages of simple structure and strong universality,and the method based on vibration signal data is the most effective and has attracted much attention in the establishment for RUL prediction models in various fields.In addition,sufficient degradation information is a necessary factor for establishing bearing degradation performance indicators,and appropriate health performance indicators are the most critical step to improve the accuracy of rolling bearing life assessment.For these reasons,based on the vibration signal of rolling bearing,this paper studies the remaining useful life prediction of rolling bearing from the perspective of data driving.The main research content and innovative results of the thesis are as follows:(1)In order to solve the problem that the degradation information is difficult to retain in the construction of health performance indicators,a new feature selection method is proposed,the principal factor correlation index(MFCI),this method first extracts main factors from different bearing degradation samples based on self-encoding neural networks,time-frequency analysis methods,and signal processing methods.Then the correlation of the main factors is analyzed by the maximum information coefficient(MIC),and calculates its MFCI.Finally,the appropriate health performance index is selected according to the MFCI ranking.This method fully considers the correlation between the bearings,it can still select the appropriate feature set effectively without considering the degradation time label,which lays the foundation for constructing a suitable rolling bearing RUL prediction degradation performance index.(2)Aiming at the problems that the traditional physical-driven RUL prediction model is not universal and its complicated construction process,the theory of the data-driven rolling bearing RUL prediction deep network model is studied in this paper.Based on the idea of local and global feature extraction,a multi-scale neural network(Multiscale CNN)is introduced,and a hybrid feature combination based on time-frequency domain analysis and EEMD is proposed.At the same time,to extract deep time series features and further improve the prediction accuracy,combined with the proposed new method of MFCI feature selection,a rolling bearing RUL prediction model construction theory based on Bi-GRU(Bi-GRU)is proposed.Finally,a variety of comparison schemes are designed to analyze and verify the advantages and disadvantages of various combination methods.(3)In order to solve the problem of the excessive dependence of artificial intelligence methods on data,the data augmentation theory of rolling bearing RUL prediction is studied.Combining the above-mentioned MFCI and deep network methods,further mining the mixed degradation information of the enhancement and original data,which proves that the methods described in this article have excellent predictive effects under the same conditions with several other state-of-the-art combined models.Compared with other comparative models,it solves the problem of insufficient degraded samples to a certain extent,and improves the robustness and generalization of the model,reduces the complexity of model construction,which is still applicable under a variety of working conditions.