Research On Rolling Bearing Performance Degradation Assessment And Remaining Life Prediction Based On Data-driven Method

Rolling bearings are one of the most core and fault-prone components in rotating machinery and equipment,and their operating conditions affect the working conditions of the entire mechanical equipment.Accurate assessment and prediction of the health status of rolling bearings in mechanical equipment will help to formulate a scientific maintenance strategy for mechanical equipment.On the other hand,it can effectively avoid downtime and even major disasters caused by cascading equipment failure.The study of the remaining life prediction method has important research value and engineering application significance.By effectively evaluating the running state of the bearing and predicting the future degradation trend,the current performance status of the bearing can be grasped in real time,and it can be predicted in advance before its serious failure occurs,which is helpful to reduce equipment damage to a minimum and improve the overall reliability of the equipment.From the perspective of data-driven,this paper studies the evaluation of rolling bearing performance degradation and remaining life prediction methods.The main research contents are as follows.(1)Aiming at the problem of low recognition accuracy of traditional performance degradation state recognition methods,a recognition method of rolling bearing performance degradation state based on information-theoretic metric learning is proposed.Using variational mode decomposition(VMD)to extract the singular values and relative energies of each intrinsic mode function(IMF)of the bearing vibration signal,and introducing the Information-theoretic metric learning(ITML)algorithm.The distance metric in the feature space is learned by constructing the pair-wise samples constraints,uses the learned metric matrix to transform the original features,and uses K-nearest Neighbor(KNN)classifier to realize the degradation state recognition of bearing performance and the bearing degradation sensitivity is enhanced.Experimental analysis results show that the proposed features have better intra-class aggregation and inter-class dispersion.Compared with the local preserve projection(LPP)and principal component analysis(PCA)methods,the classification accuracy of ITML transform features is 8% and 14.667% higher,respectively.(2)The ITML algorithm is introduced into the bearing performance degradation assessment,and an ITML-FCM rolling bearing performance degradation assessment model is proposed.The pair-wise samples constraint is constructed based on the normal and failure state training samples.The ITML algorithm is used to learn the metric matrix.The FCM algorithm is used to obtain the cluster centers corresponding to the normal and failure states.The degradation index(DI)is constructed based on the membership of the test samples relative to the normal state.At the same time,in order to quantitatively describe the comprehensive quality of degraded indicators,a comprehensive evaluation criterion combining trend,monotonicity,robustness and discreteness is constructed.The experimental results of four sets of bearing data verify the effectiveness and superiority of the proposed ITML-FCM degradation assessment model.In particular,the comprehensive evaluation criteria value obtained by the ITML-FCM model is 8.45% higher than the ED-FCM method,and 19.71% higher than the MD-FCM method on average.(3)Aiming at the problem that the traditional particle filter(PF)has serious particle degradation,resulting in low accuracy in long-term system prediction,an enhanced particle filter algorithm based on adaptive importance probability density selection is proposed.In each iteration of the particle filtering algorithm,the particle state transition probability density function and the probability distribution of the particle swarm at the current time are considered simultaneously.The importance probability density function is obtained by mixing and weighting the two probability density functions in real time,and the weighted values are updated by combining the measured value in each step,which solves the problem of severe particle degradation of the particle filtering algorithm.The simulation and experimental analysis results show that the proposed algorithm has better particle degradation suppression advantages and higher prediction accuracy.The relative error of the remaining lifetime prediction of the proposed algorithm in 20% accounts for 40.51%,which is higher than that of the traditional particle filtering method.22.19% and 24.79% of the support vector regression(SVR)method.