Methodologies For Fault Diagnosis Of S700K Switch Machine

The switch system is used to realize the transfer or crossing of trains,and the switch machine provides power for the switch conversion,which has complex electromechanical structure,frequent action and long-term outdoor environment,and once a fault occurs,it will have a serious impact on the safety of railway operation.However,at present,at the railway site,the method of regular maintenance and maintenance gap are mainly adopted to ensure the safe and reliable operation of the switch machine.The identification of fault causes mostly depends on the experience of maintenance personnel,in order to solve the problem of heavy workload and low efficiency.This thesis studied the intelligent fault diagnosis of the switch machine.(1)Feature processing of switch machine fault data.Based on switch machine power signal data,the power signal curves of 8 common faults and their failure causes were analyzed,and for the problem of less fault data of and imbalance between normal and fault data.Algorithms based on Synthetic Minority Over-Sampling Technique(SMOTE),SVMSMOTE,KMeans SMOTE and Generative Adversarial Networks were performed on the fault samples to balance the dataset,the results show that KMeans SMOTE is more effective.For feature extraction,manual feature extraction using time and frequency domain feature parameters was compared with automatic feature learning using autoencoder.(2)Fault diagnosis.The classifier were designed based on Support Vector Machine(SVM),Multilayer Perceptron(MLP)and Convolutional Neural Networks(CNN).In the diagnostic study of SVM and MLP,the feature extraction model and the classifier model need to be designed separately.In contrast,CNN can automatically extract features and unify feature extraction and classification learning into one model,realizing end-to-end learning.The classification effect of SVM and MLP on feature data shows that the autoencoder extracts features better than the feature parameter extraction in time domain and frequency domain.Comparing the classification accuracies of the three classifiers,the CNN model has the highest accuracy with practicality.(3)Design of Fault Diagnosis System.Using the SQLite database to store sample data,and designing a fault diagnosis system for switch machines based on Py Qt5,the system can display the power curves of 8 types of faults,as well as the identification of fault types and fault cause analysis,which can provide help for the maintenance of switch machines.There are 71 figures,18 tables,and 58 references.