Turnout Fault Diagnosis Based On Multi-domain Feature Extraction And Multi-classifier

As China's railway operation speed continues to increase,driving mileage continues to increase,and driving density gradually increases,as one of the key equipment,turnouts,the frequency of use and the number of applications have also increased exponentially.Once the turnout fails,if you do not take the necessary measures to maintain and repair,it may affect driving efficiency,or even threaten the safety of personal property.Therefore,the monitoring of the operating status of the turnout and rapid repair and maintenance have extremely important practical significance.In order to adapt to the development trend of intelligent fault diagnosis of turnouts,reduce labor intensity and reduce dependence on experience,this dissertation will feature extraction based on Mallat Wavelet Transform and structural framework based on GWO-SVM classifier,multi-domain feature extraction based on three PSO-PNN classifiers three-for-two voting framework is applied to turnout fault diagnosis.This thesis mainly completes the following work:Firstly,this dissertation takes S700 K as an example to analyze the action principle and power collection principle of turnouts,and summarize the types and causes of turnout failures.Secondly,feature extraction.In this dissertation,the power curve during the operation of the S700 K switch is used as the original data.Based on the analysis of the power curves,feature extraction and dimensionality reduction are performed from the time-frequency domain,time domain,and frequency domain to obtain three types of feature vector sets.For the time-frequency domain: implement Wavelet Transform first,then feature extraction;for the time domain: segment the curve first,then use time-domain statistics for feature extraction,and then reduce the dimension;for frequency domain feature extraction: first segment the curve implement Fast Fourier Transform,then use frequency statistics for feature extraction,and then reduce the dimension.Thirdly,an improved GWO-SVM classifier is adopted to realize turnout fault diagnosis:first use training samples and GWO to optimize SVM parameters;then use test samples and SVM to make predictions.Using PSO-PNN classifiers to realize the fault diagnosis of turnouts: first use three types of feature vector set training samples and Particle Swarm Optimization Algorithm to train three PNN classifiers;then use three types of feature vector set test samples and three well trained The PNN performs three-take two-vote classification,and finally evaluates the classification performance of the three-take two-vote classifier.Finally,the classification effect of the two diagnostic structures is verified by Matlab simulation and the classification performance of the two diagnostic structures is compared.The simulation shows that both diagnostic structures have higher diagnostic accuracy and faster diagnostic speed;the three-take two-vote diagnostic structure based on three PSO-PNNclassifiers has certain fault tolerance;the GWO-SVM classifier has a shorter training time and test time.