Research On The Method Of Railway Turnout Fault Diagnosis Based On Machine Learning

In recent years,Chinese railways have developed rapidly.The rapid development of railways urgently needs to improve train maintenance technology.Turnout switching equipment is an important driving device,a key equipment to ensure safe driving and improve railway transportation efficiency,and a hub to realize signal interlocking of infrastructure.However,the large number of turnouts,various types,complex structures,short service life,and diverse installation environments increase the difficulty of turnout maintenance.Most of the current railway turnouts rely on traditional routine preventive testing and manual planned maintenance to complete the operation status detection of the turnout equipment.On-site maintenance personnel could not find the faulty switch in the first time.In addition,manual testing requires maintenance personnel to have extensive work experience,and inexperienced employees are prone to make mistakes in diagnosis.In order to solve this problem,some railway lines are equipped with microcomputer monitoring equipment for turnouts,which can collect signals such as the current and power of the turnouts,and integrate fault diagnosis software based on threshold judgments in the equipment to realize fault alarms.However,the operation of most switch sites is complicated and the working environment is harsh,which often causes the threshold value preset by the microcomputer monitoring equipment to lose its reference value for a period of time.In addition,these thresholds are usually set by maintenance experts,which also prevents the thresholds from being effectively updated.In response to the above problems,this paper proposes a data-driven intelligent fault diagnosis method for turnouts based on the power curve of the turnouts,and completes the design and implementation of the turnout fault diagnosis system based on actual needs.The specific work is divided into the following three aspects:(1)Research on Fault Diagnosis Method of Turnout Based on IFD-IS:Aiming at the problem of low accuracy in the detection of multiple faults in existing turnouts,this paper studies intelligent diagnosis methods based on six common faults in turnouts,and proposes a curve segmentation method based on sliding window and Back Propagation Neural Network(BPNN)ICS-p implements segmental preprocessing of the original turnout motion signal data,and then uses features extraction,selection and dimensionality reduction operations,combined with Support Vector Machine(SVM)classification algorithm,and proposes IFD-IS turnout intelligent fault The diagnosis method realizes the fault diagnosis of the turnout,with the highest diagnosis accuracy rate reaching 98.57%;(2)Research on Diagnosis Method of Convolutional Neural Network Based on Data EnhancementAiming at the problem of insufficient fault samples in the actual operation and maintenance of turnouts,data enhancement methods such as Artificial Minority Over-Sampling Technique(SMOTE)and Deep Convolution Generative Adversarial Networks(DCGAN)are used.The fault samples were expanded and combined with Convolution Neural Network(CNN)to propose a turnout fault diagnosis model.Experiments show that the diagnosis accuracy rate is up to 99%,which effectively improves the diagnosis accuracy rate;(3)Research and Realization of Turnout Fault Diagnosis System Based on QTBased on the IFD-IS switch fault diagnosis model proposed in this paper,the model algorithm is systematically implemented using QT and Python language,with data visualization,curve segmentation,feature extraction,feature selection,fault diagnosis and other related functions,and the operation interface is completed,It is easy to operate by maintenance personnel,and it is convenient for the operator to understand the operation of the switch simply and intuitively.