Health Assessment And Trend Analysis Of Axle Box Bearing Of High Speed EMU

Since the initiation of the 13th Five-year Plan,the grid of high speed railway will add another four vertical and four horizon lines along with the implement of Plan for Mid-long Term Railway Grid,which means the construction of the railway network is widely expanded.As the railway grid extends and the trains are more frequently dispatched,the time for repairing a train will be shorter.Therefore,the security of the train will be paid more attention to.It is necessary to improve the operation and maintenance efficiency of EMUs and enhance the operation and maintenance policies to ensure the operation safety and service quality of trains.As the key part of EEF bogie,the axle box bearing plays a significant role in the security of the train.The data accumulated in the process of running is regarded as the guideline of repair and examination to the train.Existing axlebox bearing evaluation and trend analysis methods mainly have the following drawbacks:the single failure and non-fault classification which easily lead to the phenomenon of "under-repair" and "over-repair";over-reliance on expert experience and fault diagnosis relying on human identification;the utilization of the accumulated data is too insufficient in the train operation process;only using the threshold alarm method which will make the false alarm rate higher and the accuracy rate lower.In response to above problems,this paper presents a four-state health state classification method based on support vector machine and a process-based multi-eigenvalue trend analysis method.The classification and the trend forecast of the axis Box bearing state can be achieved by making full use of the value of the operational data and mining the decision rule.And a reliable assessment and prediction model is built to effectively optimize the repair system of the EMUs.It can not only improve the efficiency of operation and maintenance,but also reduce the cost of the maintenance and ensure the operational safety.The paper firstly divides the status of the axlebox bearings into four categories including health,temperature rise,strong temperature and warmth from the point of the EMU whole life cycle data.By analyzing the characteristics of the operation and maintenance data,a proper date preprocessing scheme is proposed combining with the approaches of data cleaning,dimension reduction and reduction.In order to solve the shortcomings of the traditional algorithms,a modified decision tree method is proposed to improve the health status of DT-SVM algorithm.And an improved AHP-DT-SVM algorithm is presented based on the idea of AHP which can deal with the misclassification of decision tree.Then,the paper utilizes the regression analysis to build the temperature-dependent curve of axlebox bearings under different health conditions of the classification data,and predict the trend of malfunction according to multiple eigenvalues.Finally,extensive experiments validate the efficiency of the proposed method by analyzing the reliability of the AHP-DT-SVM algorithm.And the paper constructs an axlebox bearing health assessment and trend analysis model based on the experimental results,and implements the application to a visual display.