Research On Key Technologies In Turnout Failure Prediction And Health Management System

With the vigorous development of high-speed railways today,as one of its main components,the signal system has attracted more and more attention.The normal,stable and efficient operation of the switch system is an important prerequisite for ensuring the safety of the signal system.Since the construction and operation of the railway industry in my country,whether it is purely manual maintenance or the current microcomputer monitoring combined with personal experience to complete the maintenance,the main force of the maintenance and repair work of the turnout system is still completed by the on-site staff through manual observation and comprehensive personal experience.This method is susceptible to factors such as on-site experience and knowledge level of the staff,leading to lag,misjudgment,and omission of judgments in the results,and may lead to over-repair or under-repair.This is inconsistent with the high safety and operational stability requirements of the current rapid development of the railway industry.In response to this type of situation,with the help of the knowledge of the PHM system,this thesis proposes to use curve similarity to evaluate the state of equipment,predict the types of failures that may occur,use multiple information fusion methods to diagnose equipment below the failure threshold,and the two results are compared,and the final judgment result is output.The main research contents are as follows:Firstly,in this thesis,the current research status is analyzed,then the residency research results in the field are summarized,and the characteristics and shortcomings of the existing research are also analyzed.Then the composition and working mechanism of the turnout equipment are explained.The one-time action process is divided into three stages,which are marked as T0,T1,and T2,which lay the foundation for the subsequent processing.combined with the 2016-2019 reading information and equipment account records collected by the microcomputer monitoring reading room of the Nanning Power Department.Summarize the types of faults that occur frequently,analyze the factors that lead to them,and establish a fault library.Secondly,for the situation where the health status evaluation requires high real-time and long-term monitoring,but the existing methods require a large amount of data and are difficult to obtain,it is proposed to complete the evaluation and failure prediction of the equipment status through the similarity.The common switch fault types in Chapter 2 were originally used as the fault curve library of this article,the current and power curves generated by the microcomputer monitoring were used as the research object,then the SURF algorithm is used for feature extraction,which improves the real-time performance of the calculation.Then the Hausdorff distance is used to calculate the similarity between the target curve and the normal state sample curve,and set the corresponding health value.If the device is already in a fault state,then it can calculate the similarity between it and the sample curve of the fault library.The fault corresponding to the curve in the sample library with the highest similarity is considered as its possible fault type.This step is also used as the error correction mechanism of the whole system is compared with the method in Chapter 4.Finally,the method is verified by an example analysis.Then,for the equipment below the fault judgment threshold,a fault diagnosis model of the turnout system through the multi-layer fusion of three-domain information is proposed.Aiming at the time domain,value domain and wavelet domain feature information of the extracted turnout power curve,the KPCA method is used to reduce the dimensionality;the dimensionality-reduced eigenvalues are inputted into the DBN,K-ELM,SOM model,perform the first information fusion,and get the result of preliminary diagnosis;The judgment conclusions of the three intelligent methods are used as the BPA of evidence theory to complete the second information fusion,and the results are compared with the prediction results in Chapter 3,and the final fault judgment is output,which is verified by the actual data on the spot.The effectiveness of the method in this thesis.Finally,the actual software programs of the two models described in this article are written and tested on the spot.The software includes: system interface,data collection,fault library,fault prediction and diagnosis,fault type increase,equipment account display,historical health status review,etc.Build its 3D scene model through Bently software,and realize the real-time correlation between the model and data through the lightweight engine,and complete the visual monitoring of the entire system and the station.