Research On Health Monitoring Of Train Bearings Based On HMM Time Series Analysis

In recent years,with the rapid development of high-speed railway in China,the scale of high-speed railway network continues to expand,how to ensure the safety and reliability of railway vehicles has become a very challenging problem.Running part bearing is the most key component that affects the safety of train operation,and monitoring its health status has always been the focus of the research of railway vehicle operation and maintenance department.At present,how to evaluate the bearing health status is still an urgent problem to be studied.Therefore,based on the bearing data of the whole life cycle,this paper focuses on the evaluation method of bearing health status,which provides technical basis for the establishment of bearing health monitoring standards,and also for railway vehicles.Reliability research provides technical support.In this paper,according to the time series and unlabeled characteristics of bearing related monitoring data during train operation,the hidden Markov model is selected as the research focus,which can not only model the correlation of states in time series,but also train in an unsupervised way.and aiming at the deficiency that the number of hidden states of HMM must be set in advance,Infinite hidden Markov model,the nonparametric version of HMM,is introduced into the field of bearing health monitoring,the model uses the hierarchical sharing principle of hierarchical Dirichlet process and good clustering attributes to infer the number of hidden states,which makes up for the shortcomings of HMM.At the same time,this paper optimizes the related defects of iHMM and the accuracy of the division of the state,establishes an effective model,and divides the health state of the whole life cycle of the bearing into four deterioration grades,and realizes the monitoring of the health state of the bearing.The specific research has the following points:(1)In order to solve the problem that the convergence results of iHMM are sensitive to its hyperparameter setting,Bayesian optimization and Mann-Kendall criterion are used to optimize its hyperparameters.At the same time,considering that the ergodic topology of the traditional iHMM model does not conform to the degradation process of the healthy state of the bearing,this paper constructs its topology is constructed into a left-to-right mode to fit the needs of bearing health monitoring,and the improved model is named LR-iHMM.(2)The actual bearing degradation data has dynamic changes on two time-scale.The single-layer model has insufficient state expression ability and the state division is too direct.Therefore,this paper extends the structure of the LR-iHMM model to two layers,and macroscopically divides the hidden state of the single-layer model,simulates the health state of the bearing with macroscopic state,and improves the model's ability to model bearing data.At the same time,according to the principle that the data characteristics of the same health state have certain similarity,the microscopic state with similar data distribution is merged into a macroscopic state,and the quality of the healthy state division is further improved under unsupervised conditions,forming a Doubly LR-iHMM model.(3)The bearing health monitoring model based on Doubly LR-iHMM is constructed.Based on the actual bearing data,the results show that Doubly LR-iHMM can better model the bearing performance degradation data compared with single layer model.The results of the division of health status are also more reasonable.According to the experimental results,the health status of the bearing life cycle is defined by four degradation levels,and the health monitoring of the bearing is realized.