Research On Intelligent Fault Diagnosis Method For Railway Turnouts Based On Deep Feature Extraction

With the rapid development of railwey industry,railway key equipment gradually tend to automation and intelligence,equipment failure will affect the normal operation of trains,reduce transport efficiency,and even cause loss of life and property.How to carry out intelligent fault diagnosis of the equipment has become a problem that cannot be ignored in the railway industry.As one of the key equipments of railway turnout is the "steering wheel" of the train,and its health status is directly related to the safety of the train.At present,the main way to realize turnout fault diagnosis is to judge by the constant threshold alarm of turnout action power or current curve.The setting of the threshold value relies heavily on the knowledge and experience of technicians,which is inefficient and prone to serious situations such as fault omission and false alarm.Therefore,many turnout intelligent fault diagnosis methods are proposed in order to avoid the influence of artificial and improve the diagnosis accuracy.In this paper,taking railway turnouts as the research object,the intelligent fault diagnosis method based on deep feature extraction is proposed for the problems of incomplete feature extraction and unbalanced data in the current intelligent fault diagnosis methods,and the effectiveness of the method is verified through experiments.The specific research contents are as follows:(1)Aiming at the problems of traditional fault diagnosis methods such as relying on human experience,incomplete feature extraction and unbalanced data,a turnout fault diagnosis method based on improved Autoencoder and data enhancement is proposed.First,by analyzing the characteristics of health and fault data,an improved smooth denoising and feature extraction fusion Autoencoder(SD-FAE)is proposed,which overcomes the noise caused by electrical characteristics and extracts the deep features of data at the same time.Secondly,aiming at the problem of data imbalance,combined with the synthetic Oversampling technology(SMOTE),data enhancement was performed on the minority fault feature data extracted by SD-FAE.Finally,a Softmax classifier is used to classify faults for balanced and labeled features.In experiments with real data,the average diagnostic accuracy of the proposed method in this paper reaches 99.13%.The results show that the proposed method can directly extract more representative and deeper features,making up for the lack of data imbalance.(2)In view of the fact that some fault samples are very few and the turnout action data cannot be marked due to the launch of new equipment or the short service life of equipment,this paper proposes an unsupervised fault diagnosis method for turnout based on Bidirectional Generic Adversary Networks(BIGAN)and Gaussian Mixture model(GMM).First,using the structural characteristics of BIGAN model,only a small number of samples are used to train BIGAN Unsupervised learning model,implicit data enhancement is done in the feature space,and the mapping relationship between learning samples and their corresponding feature representation is made.By learning the deep features of the data through the obtained encoder,the problem of very few turnout fault samples is solved.Secondly,in order to alleviate the impact of unsupervised diagnosis caused by excessive feature dimensions,t-distributed Stochastic Neighbor Embedding(t-SNE)Nonlinear dimensionality reduction technology is used to reduce feature dimensions.Afterwards,the GMM model was used to achieve unlabeled feature fault diagnosis.Finally,under four unsupervised evaluation indicators,a comparison was made with other existing methods,indicating that the proposed scheme is superior in diagnosis and performance.(3)Based on the above research content,this paper forms the proposed turnout intelligent fault diagnosis model into a system module and designs and implements a non-invasive turnout fault diagnosis system by combining network communication technology.The system is developed based on Python language,which can provide interface visualization,user management,data management,network communication,fault diagnosis and querying fault alarm information for train control system operation and maintenance personnel.