| As an important part of the station turnout system,the switch machine is mainly used for the switching,locking and indication of the turnout.The reliability of switch machine is directly related to the transportation efficiency and even traffic safety of railway.According to the field investigation,the switch machine has the characteristics of large number,wide installation range and many fault modes,while the existing research results can only diagnose several common fault modes.Therefore,how to expand the fault diagnosis scope of switch machine and improve the level of intelligent diagnosis is an urgent need for railway development.Based on the basic working principle and working process of the S700 K switch machine,19 working modes of the switch machine including the normal mode are summarized.The analysis shows that the failure of the switch machine at different structural components will lead to the real-time change of the active power value of the switch machine.Therefore,the operating state of the switch machine can be analyzed through the action power curve of the switch machine.Through consulting data and field fault records,18 common fault modes of the S700 K switch machine are summarized,and the operation power curve of the switch machine under each fault mode is obtained by analyzing the fault causes.To this end,this paper has carried out the following research:1.Based on the basic structure and working principle of S700 K switch machine and its monitoring system,the variation law of power curve under common fault mode is analyzed.Based on the basic structure and working principle of S700 K switch machine,combined with literature and railway field fault record data,18 corresponding common fault modes are summarized.On this basis,based on the working principle of S700 K switch machine monitoring system,the change law of power curve corresponding to the above failure modes is analyzed.2.An intelligent fault diagnosis method for switch machine based on gamian angular field(GAF)and convolutional neural network(CNN)is proposed.Firstly,the time length and amplitude of the data are normalized by preprocessing the power curve of the switch machine;Then,based on the gram angle field time series coding method,the pretreated switch machine power curve is transformed into a two-dimensional image;Finally,taking the transformed two-dimensional image as input,a convolution neural network fault diagnosis model is constructed.Aiming at the unknown faults other than18 main faults,an unknown fault open set recognition strategy based on extreme value theory(EVT)is designed to realize the intelligent diagnosis of switch machine faults.The experimental results show that the proposed algorithm has the advantages of high accuracy,multiple fault types and strong applicability.3.Based on the monitoring system of switch machine,using C++ and python programming language,and based on QT framework and flask framework,an intelligent fault diagnosis software for switch machine is designed and implemented.According to the field application requirements,the software interface design and function design are carried out,mainly including online fault diagnosis function,offline fault diagnosis function,data monitoring function and diagnosis model management.Finally,the functional modules are verified by experiments.There are 72 figures,14 tables and 49 references. |
