Research And Design Of Centralized Monitoring System Switch Fault Diagnosis Based On Machine Learning

As the end of 2019,China's railway business mileage has exceeded 139,000 kilometers,and the maintenance of signal infrastructure is facing huge challenges,especially switch equipment with a high failure rate.At present,the signal centralized monitoring system can collect and monitor the switch data information to assist the maintenance personnel to quickly troubleshooting,but for some inexperienced personnel,it may not be able to give the causes of fault in time and accurately,so many intelligent fault diagnosis methods are adopted based on the research of switch fault diagnosis problems,which is of great significance for the maintenance stable operation of switch.In this thesis,a switch fault diagnosis model based on PNN decision-level fusion is proposed,which is applied to a centralized monitoring switch fault diagnosis software system.The following researches and applications are mainly made.Firstly,this thesis analyzes the research status of switch fault diagnosis at home and abroad,and summarizes the signal centralized monitoring system and its collection of switch data information.The widely used ZD6 switch machine is selected as the research object to analyze its mechanical level and circuit level action principle.This thesis also summarizes some common faults.The above content will be used as the theoretical basis for the establishment of fault diagnosis model.Secondly,this thesis samples the key areas of real switch action current data to form a sample set for model training and testing.The corresponding fault diagnosis model is established by the methods of BP network,PNN network and SVM respectively,which are combined with the theoretical knowledge of machine learning.The obtained sample set is used to verify and analyze the diagnostic performance of three models.The conclusion shows that the PNN network has more excellent prediction performance.Then,through the diagnostic performance of three initial models,combined with information fusion technology,the decision-level fusion structure and neural network method are selected to propose a switch fault diagnosis model based on PNN decision-level fusion.In this thesis,the principle and design of the model are described in detail.The sample set is used to verify and analyze its diagnostic performance in the case of a certain initial model failure and intact initial model.The results of the sample for the test have shown that the fused model has a certain fault tolerance and further improvement of accuracy rate.Finally,the focus of thesis is on the basis of fault diagnosis model,relying on the "railway regional interlocking and centralized monitoring system" project and experiment platform,the methods of C# language and Matlab mixed programming,My SQL database are uesd and the fault diagnosis model is applied to the project and experiment platform.By designing six major modules for centralized monitoring of switch fault diagnosis system,the function of switch action current query,switch fault intelligent diagnosis,switch alarm processing and diagnosis model update were implemented,and the development of centralized monitoring switch fault diagnosis software system was finally completed.Deploy it on the experimental platform to test it one by one.The test results show that the system can realize the intelligent diagnosis of switch and improve the efficiency of fault diagnosis.The software system can not only provide theoretical research for the application of artificial intelligence in the centralized monitoring system,but also can be applied to the practical teaching of fault diagnosis or product development.There are 72 figures,21 tables,and 43 references in this thesis.