Research On Service State Prediction And Maintenance Decision Of High Speed Turnouts

With the continuous expansion of China’s high-speed railway network,high-speed switches have been widely used.The main function of railway switches is to achieve track conversion and ensure the safe operation of trains.If a fault occurs,it will affect the normal transportation process and even more seriously endanger the safety of passengers and railway staff.However,as the service life of high-speed switches increases,wear and tear will inevitably occur.Therefore,the railway department attaches great importance to the reliability of high-speed switches.To avoid the occurrence of serious malfunction of the turnout,it is necessary to study the degradation rule and maintenance strategy of the turnout.At present,the maintenance mode of high-speed turnout by railway departments is mainly periodic maintenance,which is prone to "under maintenance" and "over maintenance".In this paper,the wear data of the turnout in service process are used to study the prediction of service status and service maintenance strategy.Meanwhile,other analysis methods and calculation tools are used,such as Failure theory of structural component in service,Grey model,BP neural network and Markov chain.The main research contents are as follows:(1)To provide theoretical support for the subsequent work,these efforts need to be implemented: analyzing the type of turnout degradation and its influencing factors,understanding the influence of wear turnout rail quality,in addition,introducing the basic theory of degradation prediction method and data processing method.(2)The turnout degradation model is established.In the selection of the model,considering the small sample of the actual degradation data and the unequal time interval of the degradation detection data,based on the grey model,a grey prediction model with unequal time interval is firstly established.After that,three different calculation methods are compared: traditional gray model,new information gray model and metabolism gray model,which shows that the general non-isochronous prediction model and metabolism non-isochronous grey prediction model are more suitable for the degradation prediction of turnout rail.Finally,via BP neural network to correct the residual error of the non-isochronous grey prediction model,the combined prediction model of grey BP neural network is established.Accuracy test shows that the accuracy of the combined prediction model is significantly higher than that of the non-equidistant grey prediction model: the prediction results of straight sharp rail shows that the accuracy reaches the first level with a average relative residual of 0.27%;the accuracy of non-isochron grey prediction model is second order,and the average relative residual is 3.31%,there are big improvements in both aspects.From the point of curved sharp rail,the prediction accuracy of both models is secondary.However,the average relative residual of the non-isochronous grey prediction model is 4.09%,and that of combined prediction model is 1.65%.The prediction accuracy is significantly improved.(3)Establishing a turnout state prediction model which based on Markov principle.The Markov state prediction based on exponential function model and that based on combination prediction model are completed respectively by taking the straight pointed rail data as an example.Then the predictive maintenance decision is made,and the specific implementation rules of the maintenance decision of the straight sharp rail are presented.(4)The establishment of turnout predictive maintenance system requires two parts:maintenance system design and system function module.The former includes data management,condition prediction and making turnout maintenance plan;by storing turnout data,building data models,predicting turnout status,and making related maintenance decisions,system function module can reduce the work intensity of workers.