Research On Non-invasive Operation And Maintenance System Of Vehicle ATC Equipment Based On Machine Learning

In the operation and maintenance of existing on-board ATC equipment,there are problems such as incomplete equipment information monitoring and low efficiency of fault diagnosis.At present,the information monitoring of on-board ATC equipment is mainly collected by data bus,and there is no effective monitoring means for the actual input and output,equipment power supply and other information.In the aspect of fault diagnosis means,it mainly relies on manual data analysis for fault diagnosis and classification,and the accuracy and efficiency of diagnosis and classification are low;in the aspect of fault diagnosis and classification algorithm research,it mainly extracts fault features from manually recorded fault record table or equipment monitoring status table,and there are few fault features and classifiable fault categories that can be extracted.In view of the problems existing in the operation and maintenance of on-board ATC equipment,the research on non-invasive operation and maintenance system of on-board ATC equipment based on machine learning algorithm is carried out.This dissertation proposes a non-invasive electric signal acquisition programme based on fluxgate to solve the problem of incomplete equipment information monitoring.Extract fault features from logbook,proposes a multi algorithm two-out-of-three decision-making fault classification model to solve the problem of low efficiency of fault diagnosis.The main contents of this dissertation are as follows:(1)Based on the analysis of historical fault data,the necessity of adding non-invasive signal acquisition system is demonstrated.The principle of fault feature extraction based on logbook is established,and the initial fault feature set is summarized based on maintenance experience.(2)By using association rule analysis to verify and simplify the feature set,CM Apriori algorithm based on compression matrix is proposed to optimize the running time,which is compared with the classic Apriori algorithm.The test results show that CM Apriori algorithm can save 29% of running time.(3)Based on fluxgate,the non-invasive electric signal acquisition programme is proposed to collect the actual input and output,equipment power supply,safety and nonsafety input and output information of onboard ATC equipment.The test results show that the maximum acquisition error is 0.7m A,and the maximum error percentage is 4.4%,which meets the requirement of 5% acquisition accuracy.(4)A two-out-of-three decision fault classification model is proposed to synthesize three machine learning algorithms: Ge NIe-BN network,CART decision tree and PSOSVM.In the construction of Ge NIe-BN network model,MCMC and K2 structure learning algorithms are compared,and model reasoning is carried out by using Ge NIe environment;in the construction of cart decision tree model,multiple parameters such as the best maximum depth and random seed are adjusted to obtain the best decision tree model;in the construction of PSO-SVM model,PSO optimization algorithm is proposed to find the best parameters.Finally,the three models are compared and analyzed,and the two-out-of-three decision-making fault classification model is proposed to make comprehensive decision for the three algorithms.Three different algorithms participate in the decision-making to solve the common mode security problem in the two-out-ofthree decision-making.The test results show that the two-out-of-three decision-making fault classification model effectively reduces the classification error of each algorithm,the average accuracy of the model is 94.6%,and the improvement effect is significant.This research relies on the national key research program of China Academy of Railway Sciences.Through data test,the non-invasive operation and maintenance system of on-board ATC equipment based on machine learning algorithm proposed in this dissertation has good application effect in on-board ATC equipment monitoring and fault diagnosis classification algorithm,which can provide reference and reference for expected maintenance research.There are 56 pictures,22 tables and 72 references.