Research On Turnout Degradation State And Health Management Method Based On Deep Learning

With the rapid development of the railway industry in my country and even the world,people have higher requirements for the safety,reliability,energy saving and efficiency of railway transportation operations.As one of the key basic equipment of railway signal system,turnout is an important equipment for arrange the routes of the station interlocking system and realize the conversion of train from one track to directly crossing another track.Therefore,it is particularly important to study the degraded state and health management methods of turnout equipment.At present,in terms of railway transportation,the real-time monitoring of turnout equipment by the railway electrical department is mainly realized through the centralized signal monitoring system.The monitoring content mainly includes data such as switch machine operating current,operating voltage,conversion power and gap.At the site,periodic repairs and emergency repairs are adopted,and professional signalers mainly rely on manual experience to diagnose turnout faults.In most cases,it is impossible to achieve advance prediction and effective maintenance management.Aiming at implement effective maintenance of turnout,it is necessary to study the degradation state and health management method.Before the turnout fails,the operation state of the turnout is analyzed,and the degradation level of the turnout is grasped according to the model,and the maintenance work is carried out in advance;when the turnout is in the failure,the intelligent fault location and identification are carried out.Based on this idea,this paper combines the feature extraction based on distance evaluation technology with the MPSO-SVM classifier and applies it to turnout fault identification;for turnout Outdoor diode fault,a new method based on ensemble empirical mode decomposition(EEMD),sample entropy(SE)theory and long-short-term memory neural network combined model is proposed,and applied to the research of turnout degradation state,therefore realize the health management of turnout.The main contents of this paper are as follows:(1)Taking the S700 K point machine as an example,firstly explains the basic structure and working principle of the switch;secondly,it analyzes the principle of data acquisition of the three-phase AC switch machine and the necessity of monitoring the switch power curve to study the real-time operation status of the switch;Analyze the power curves of normal and common faults in detail,and then summarize various fault phenomena and the reasons for this situation.(2)Construction of turnout health management framework and feature extraction.For turnout data,based on the open PHM framework,the health management framework of the turnout system is proposed which mainly includes modules such as data collection and data processing,condition monitoring,fault diagnosis,fault prediction and health management modules,and each module is brief analysis.Taking the action power curve of the S700 K AC switch machine as the original data,the time domain features are extracted in sections and a feature candidate set is formed.The compensation distance evaluation technology is used to reduce the dimension of the extracted feature candidate set,and the sensitive features are selected as the characteristic of the turnout movement.Feature;For time-frequency feature extraction: the ensemble empirical mode decomposition(EEMD)algorithm is used to extract the power signal,and several groups of modal components(IMF)are obtained,and then the sample entropy of each group of modal components is calculated as the feature of turnout action.(3)Using modified particle swarm optimization(MPSO)to optimize support vector machine(SVM)to intelligently identify turnout faults.The disturbance term and momentum term are introduced to improve the particle swarm optimization algorithm and the relevant parameters of the support vector machine are optimized,which is used as a classifier to identify turnout faults.(4)The long short-term memory neural network(LSTM)is used to predict the degeneration state of turnout.The time-frequency characteristics of turnout action extracted from training samples and the health factors constructed are used to train LSTM network.Finally,the test samples are input into the model to predict the degradation state of turnout.Based on the two methods in deep learning,this paper conducts health management research on turnout fault identification and degradation status.The research shows that both methods have certain feasibility.On the one hand,it can accurately identify turnout faults with a recognition rate of 97.78%.on the other hand,it can predict its degraded state with an accuracy rate of 92%.Combining the two aspects,it can provide a strong guarantee for the safe and reliable operation of turnout equipment,and can provide a reference for the development and application of a comprehensive platform for turnout health management.