Research On Application Of Neural Network And Expert System In Rail Switch Fault Diagnosis

With the rapid development of railway transportation in China, railway signal facilityhas been tend to be networking and intelligent in order to ensure the security and efficiency.Computer interlocking system replaces electronic integrated system, but turnout controlcircuit is retianed. So far, the automatic monitoring system for switches has not applied to themaintenance. Thus, Based on a lot of documents and materials and experiences of theproblems of switches from on-site maintenance person, this paper analyses the problems ofswitches with expert system and neural network. Research works in this thesis were focusedon the follows:(1) In the diagnosis of expert system, switch is regarded as the research object,itsstructure and failure mechanism are analyzed. Based on the field data and relevant articles, itintegrates fault trees with frame theory to show the information in the knowledge basethrough the studies of fault tree. And then the thesis combines conditions with bottom eventof minimum cut sets of fault tree with the usage of notations on the basis of regulation.Finally, forward reasoning is adopted. This method is helpful for clients to understand thewhole rational process. Besides, it could make the explanation for explanation mechanismmore clear.(2) In the diagnosis of neural network, the electronic fault of switches that have8kindsof fault types are choosed to design the network structure of back propagation(BP) neuralnetwork and various parameters of training. The relevant results are gained through the testsof neural network after the Matlab training. With the results comparison of expert system andneural network, the former better suits science with hard regulation like switches mechanicalfailures, while the latter has more advantages in handling nonlinear data like switcheselectronic problems. Both of them have strongly complementary. Hence, the paper completefailure diagnoses with their cooperation.(3) Expert system is designed by Visual C++6.0, including diagnostic reasoning, repairdecision making, etc. Database is established Microsoft Access2003, and the neural networkis designed by Matlab. The detailed cases are discussed in the paper to test different diagnosesresults of expert system and neural network for kinds of failures and problems. And the testsuggests the feasibility of two methods for the diagnoses of various failures and accidents,this system can carry out rapid processing for different fault types, to improve the efficiencyof diagnosis.