Health Monitoring Of Turnout Tips Based On Multi-sensor Data Fusion

Turnouts are an important infrastructure in the railway system,and they are also a difficult point in the maintenance and repair of the rail system.The track structure in the turnout area is complex,the rail is weak,and the sharp rail is more prone to rail damage than the ordinary section.In addition,because the rail flaw detection vehicle cannot fully grasp the status of the rail in the turnout area,it is a blind spot for automatic detection.Therefore,monitoring the status of turnout rails and detecting rail damage early is an urgent need for the development of railway operation in China.Focusing on the health monitoring of turnout tips,this paper proposes a multi-sensor data fusion turnout tip rail damage monitoring and damage identification method based on acoustic emission,which can give full play to the advantages of various sensors,solve the complex and changeable problem of rail acoustic emission signal noise in turnout area,and achieve the purpose of effectively monitoring the operation status of turnout tip rail.The main research contents of the paper are as follows:(1)Developed a monitoring system that enables a variety of signal acquisitions.Select appropriate sensors,appropriate measurement point layout and monitoring targets,study the applicability and division of labor of each monitoring index in the multi-sensor data fusion monitoring system,use train load trigger to collect data,and collect acoustic emission,vibration and strain signals at the same time.(2)Since it is difficult to simulate the real working state of acoustic emission sources and turnout rails by numerical simulation,a large number of field tests will be carried out to obtain real experimental data.By analyzing the measured lossy and lossless data and combining various signal characteristic parameters,this paper provides a basis for the preparation of the later algorithm model dataset.(3)A research method for rail damage identification based on K-nearest neighbor algorithm is proposed.As a supervised learning method,K-nearest neighbor algorithm is usually used when the ratio of normal data and abnormal data is comparable.The K-nearest neighbor algorithm is used to identify the damage of sharp rails,and the recognition results show that the recognition accuracy of multi-sensor data fusion is significantly improved compared with that of single sensors.(4)Aiming at the situation that there is less measured lossy data and more normal data,this paper proposes two unsupervised learning algorithms based on isolated forest and singleclass SVM to identify rail damage.The algorithm is used to identify the abnormal data of each validation dataset,and the performance of the isolated forest and single-class SVM models is verified.It effectively solves the problem of little or no damage data measured and lossy.The research results provide a reference for the detection and identification of rail damage under multi-sensor data signals,which can effectively solve the problem of effective signal identification and processing under complex signal conditions,and the research results provide a reference for the detection and identification of turnout tip rail damage and the formulation of reasonable maintenance plans,which is of great significance for solving the problem of automatic detection blind zone of turnout rail.