Research On Fault Diagnosis Of S700K Switch Machine Based On Bayesian Network

With the rapid increase of the operating mileage of high-speed railways,S700 K switch machine has been widely used as an important outdoor signal equipment.Its working condition not only affects the train operation safety but also affects the railway transportation efficiency.Therefore,more intelligent fault diagnosis methods are needed to improve the safety and reliability of the equipment.At present,the fault diagnosis of switch machine in China mainly relies on the maintenance experience of the staff to locate the fault.This method of fault identification is inefficient and cannot determine the fault location online in real time.Moreover,because of the complex structure of the switch machine and the harsh and changeable working environment,the fault types of the switch machine are diverse and uncertain,which increases the difficulty of manual fault diagnosis.Therefore,it is particularly important to research the intelligent diagnosis of switch machine.Based on the analysis of the basic working principle,action power curve and fault type of S700 K switch machine,this thesis proposes a fault diagnosis method of S700 K switch machine based on Bayesian network.The main research work of the thesis is as follows:Firstly,based on the basic structure,control circuit and working principle of the S700 K switch machine,the relationship between action power and operation state of switch machine is analyzed.On this basis,according to the different action process of the switch machine,the fault is analyzed and summarized,and the power curves and fault causes of eight common fault types of the switch machine are given.Secondly,the characteristics of each working section are extracted according to different time domain characteristic parameters of the signal.Because the extracted feature data are continuous,which is not suitable for machine learning,a discretization algorithm based on the combination of information entropy and rank importance is proposed to discretize the features.In view of the existence of redundant characteristics in fault characteristics and the different importance of each characteristic to system classification,a reduction algorithm is used to reduce fault characteristics and eliminate redundant characteristics on the basis of discretization.Because the naive Bayesian network classifier based on the conditional independence assumption cannot meet the needs of practical applications,an attribute weighted naive Bayesian network classifier model based on gray association analysis is proposed.The model considers the degree of correlation between different feature nodes and type nodes on the basis of feature reduction,and improves the performance of Bayesian network classifier by using the correlation between them,so as to improve the classification accuracy of the classifier.Finally,in view of the diversity and uncertainty of the faulty feature information of the switch machine,a fault diagnosis method of S700 K switch machine based on Bayesian network classifier is proposed.In this method,signal characteristic extraction technology,rough set theory and Bayesian theory are combined.First,the characteristic information of fault signal is extracted by time-domain signal characteristic extraction technology.Then the discretization algorithm based on the combination of information entropy and row column importance is used to achieve the discretization of fault feature,and on the basis of discretization,the attribute reduction algorithm based on class difference matrix is used to achieve the reduction of fault features.At last,on the basis of the minimum fault characteristic set,the attribute weighted naive Bayes classifier is used to classify the data and achieve the fault diagnosis of switch machine.Taking the action power curve of S700 K switch machine collected by an electric section of Lanzhou Railway Bureau as an example,the effectiveness and accuracy of the proposed algorithm are verified.