Fault Diagnosis And Health Status Prediction Of Switch

As one of the important equipment of the station interlocking system,switch plays an essential role for arranging the route and realizing the direction change of the train.With the rapid development of Chinese railway,the maintenance of switch equipment is also facing more and more challenges.In the railway field,the switch monitoring method mainly depends on manual browsing switch operating current data and power data of Centralized Signal Monitoring system.This approach lies on the artificial experience,which is difficult to meet the demand of complex work environment of rail equipment.Furthermore,this method belongs to "after-fault maintenance”and fails to carry out effective fault protection on the turnout equipment.Only after the fault characteristics are obviously displayed are the maintenance actions taken.This kind of repair method is difficult to form an effective maintenance plan.To solve the above two problems and realize the intelligent diagnosis and prediction of the switch equipment,this paper puts forward a method of degraded state mining and status evaluation based on Kohonen neural network and hidden Markov model.By analyzing the non-fault power data collected in the field,degenerate state data are mined using Kohonen neural network,in addition,the reasons for the formation of the related degenerate state are analyzed.According to the fault of abnormal resistance during conversion process of switch,a hidden Markov model is used to establish the degenerate state model.In order to realize the fault diagnosis and health state prediction of switch equipment,the state classifier model is used to identify and evaluate the degradation state and fault state.The main work of this paper is as follows:(1)FMEA analysis of switch equipment.The failure mode analysis table was obtained by analyzing the basic components and functions of switch equipment.Whilst,the monitoring principle of switch equipment is studied to understand the switch action process.For high frequency faults,the reason of power curve formation is summarized,and a typical fault power data sample base is established.(2)Research on feature extraction method.Aiming at the characteristics of the power signal,the typical characteristic signal during the operation process of switch is extracted by statistical analysis,and the eigenvector space is constructed.The Fisher criterion function is used to select the characteristic signal,and the typical characteristic of the fault is preserved for state recognition.Finally,the eigenvectors of the appropriate dimension are obtained by introducing the method based on local linear embedding to reduce the dimensionality of the screened typical features to act as the inputs for the subsequent degenerate state mining and state recognition models.(3)Degeneration state data mining of switch.Unsupervised learning for the non-fault power data collected in the field is implemented by using Kohonen neural network.Compared to the results of K-means clustering and FCM fuzzy clustering,the typical degradation state data are obtained.By analyzing the degenerated state,the relationship between the degenerated state and the failure of the database in the fault state is given,it is found that there is the highest correlation between the degenerated state and the fault of abnormal resistance during conversion process of switch.In addition,a database of degraded states data is established.(4)Study on fault diagnosis and status evaluation of switch equipment.Aiming at the typical fault and degradation states,a hybrid HMM model is established for the state identification and fault diagnosis of the switch.The current state of the switch is identified to diagnose the fault and evaluate the health status.The final experiment shows that when the 6-dimensional input is performed,the recognition accuracy reaches 95.65%,and the correct rate and training time meet the needs of the railway site.In this paper,Kohonen-HMM method is used to excavate and identify the degenerate states during the switch operation.The experimental results show that the method for mining and recognize the deterioration of switch based on Kohonen-HMM has great feasibility and superiority in status recognition and evaluation of the switch equipment.