Research On Key Technologies Of Service Status Detection Of Track Fasteners Based On Structured Light Point Cloud

Rail fasteners are extremely important basic components in railway lines.In addition to the basic function of tightly fixing the rail tracks to the sleepers or track panels,the rail fasteners also have the functions of damping and adjusting the track smoothness.Therefore,the track fastener plays a vital role in the normal operation of the railway line.At present,the railway operation department mainly inspect the fasteners by manual inspection,which consumes a huge amount of manpower and material resources.With the increase of railway operation mileage,the increase of operating years and the soaring labor costs,this method is difficult to meet the reality of the railway maintenance.As a large number of railways,especially high-speed railways,are put into operation,the responsibility of China's railway operation and maintenance departments is becoming heavier.Aiming at the demand for efficient and rapid detection by railway operators,a variety of image-based fastener detection methods have emerged in recent years.However,due to the natural limitations of twodimensional images,these methods can only detect whether the track fasteners are damaged or not,cannot measure the tightness and critical variable part size,which do not fully meet the requirements of track fastener inspection.Structured light sensor obtains target position information based on triangulation principle,which has the advantages of high sampling rate and high precision and has been widely used in industrial measurement field.In this paper,structure light sensors are used to build the fastener detection equipment,and the high-resolution point cloud of the track fastener is obtained.Based on the obtained point cloud,the key variable components size measurement,fastener defect inspection and fastener tightness measurement are conducted.The main research work of the thesis is as follows:1.Research on multi-Sensor integrated structured light track fastener inspection system.Firstly,according to the actual conditions of railway lines and the requirements of fastener detection,integrate a variety of sensors and build a fastener inspection system,to obtain high-resolution point cloud of fasteners;then,a calibration method for the structured light sensors is proposed,which calibrates the three-dimensional attitude and relative position of the linear structured light sensors in the system.Finally,an experiment is designed to analyze the accuracy of point cloud from the structured light sensor.2.Methods for extracting track fasteners and measuring geometric dimensions of key fasteners are proposed based on point cloud from structured light sensors.Firstly,according to the characteristics of the point cloud,a fastener extraction method based on single frame point cloud detection and a fastener extraction method based on multiframe point cloud sliding window detection are proposed.Then,according to the structural characteristics of fastener,the coordinate system of fastener model is established based on the extracted point cloud of single fastener,and the key geometric parameters of fastener point cloud are extracted by region growing method.The geometric dimensions of the variable parts of fasteners are obtained according to the key geometric parameters of fasteners and the structural characteristics of different types of fasteners.Finally,the method of extracting fasteners and the method of measuring the geometric dimensions of variable fasteners are experimentally validated in different types of fasteners.3.Two fastener defect detection method are prosed,one is based on decision tree and another is based on deep learning point cloud segmentation.Firstly,a method of real-time fastener defect detection based on decision tree is proposed.A top-down decision tree classifier is designed based on the coordinate system of fastener model and the structural characteristics of fastener to classify fastener defects.Then,a fastener defect detection method based on deep learning point cloud segmentation is proposed.In training data set,point cloud region growth is used to automatically label the point cloud of fastener target parts.In the aspect of deep learning network,the point cloud of fastener is segmented based on PointNet++ point cloud segmentation network,and the defect of fastener is detected according to the segmented result.Finally,the accuracy and efficiency of two kinds of fastener defect detection methods are tested on various types of fasteners.4.In order to meet the need of accurate measurement of fastener tightness,a method of extracting spring point cloud skeleton of fastener is proposed,and other applications of this method are studied.The method takes the coordinates and normal vectors of the point cloud of the fastener spring as input and obtains the center line of the fastener spring.It can transform the three-dimensional discrete point cloud of the fastener spring into a continuous one-dimensional expression,thus accurately grasping the overall structure of the fastener spring.Subsequently,a method for measuring the elasticity of fastener spring and a method for detecting the defect of fastener spring are proposed.The tightness and abnormal deformation of fastener spring are detected respectively.Finally,the accuracy and efficiency of elastic strip tightness detection method are compared with manual measurement on various types of fasteners.