Degradation Modeling And Remaining Useful Life Prediction Of Rail Vehicle Wheels Based On Mixed-effect Model

Prognostics and health management is the significant direction in the field of reliability research of complex mechanical and electrical equipment,which is also the promising direction of the urban rail equipment maintenance in the future.Since the wheel status and performance are significant to the safety and quality of rail vehicles,mastering the degradation of wheels and predicting the remaining useful life have become one of the core contents of rail vehicle prediction and health management.Because of the discreteness of wheel degradation and the differences among the individuals in complex operating environments,the wheel degradation modeling and life prediction are studied from the perspective of data driving.This thesis constructs a longitudinal dataset of wheel degradation based on the understanding of rail vehicle wheels degradation.Besides,the mixed effect model is selected as the research method by using the pre-analysis of the data to solve the characteristics of objectives' data.Then,based on the degradation characteristics of the tread,a series of linear mixed effect models are established and the parameters of these models are solved through EM algorithm.Furthermore,the basic linear mixed effect model is selected referring to the fitting index.The heteroscedasticity and autocorrelation structure are used to adjust the error effect matrix,which is to improve the fitting accuracy of the basic linear mixed effect model.According to the degradation characteristics of the rim,the exponential family and the coupling function are selected to design the structure of the basic generalized linear model.What's more,a series of generalized linear mixed effect models are established by adding random effects.Then,the adaptive Gauss-Hermite method is used to solve the parameters of the model,which is helpful to the optimal model selection.The penalised quasi-likelihood is used to estimate the parameters of the optimal model.Comparing the solution results of the above two parameter estimation methods,the model parameter estimation method is ultimately selected for model parameter estimation.Finally,referring to the tread degradation index of the running wheel,the data are independently transformed by the spectral decomposition of the error effect matrix,and the real-time residual useful life of running wheel is predicted according to Bayesian fusion idea of numerical analysis.Additionally,for the rim index of the running wheel,the real-time residual life prediction of the service wheels is solved based on the Bayesian fusion idea of simulation.The feasibility and effectiveness of the proposed method are proved through the above example verification.And it also provides the scientific reference for state maintenance of the rail vehicle to ensure the safety and reliability of the vehicle operation.