Research On Dynamic Detection Method And Technology Of Track Structure Components Based On Machine Vision

In railway safety management,state of track structure components(rail,fastener and etc.)attaches great importance.Vehicle load and natural conditions play a major role in deformation and damage of track structure components.At present,rapid development of China railway requires advanced dynamic detection technology,which can quickly detect and accurately analyze track structure components,then eliminate safety threat to ensure safety of trains.It has become a common measurement problem that needs to be solved urgently in the development of China’s railway,and it is also the research focus in the field at home and aboard.Based on modern optics,visual detection technology,which has the advantages of non-contact,high efficiency,great intelligence and rich information,combines computer technology,image processing,data analysis and etc.,becomes the trend of dynamic detection in both domestic and foreign railway industries.It also the first choice for dynamic detection of track structure components.The research in this field in China started a little later than abroad,however,based on the strong background of high-speed railway development in China and the urgent need for dynamic detection of track structural components,the research has obvious development advantages.The domestic railway science and technology workers have carried out the dynamic detection of track structural components based on the twodimensional image data in the early stage,initially realized the rapid detection of rail surface scratch and missing fasteners,and formed a relatively complete management system and technical standards.However,in practical application,two-dimensional image dynamic detection of track structure,which has no spatial dimension information data,will cause misjudgment becasue of some interference factors such as the stain,cannot be obtain some important informantion such as rail scratch depth,fastener status ect.,bringing hidden trouble to the train operation safety.In order to solve these problems,this doctoral thesis studies the dynamic detection methods and techniques of rail structural components.A 2D & 3D machine vision dynamic detection method of track structure parts was proposed,which can obtain twodimensional gray and 3D depth information of track structure parts at the same time.Multi-source information fusion processing methods of analysis and detection method based on polarized light was studied,which can enhance the accuracy of the track structure parts and anti-jamming,improve detection accuracy and reliability.The main research contents of this paper are as follows:⑴A rail profile detection method based on polarization imaging was proposed to improve the dimension of information acquisition.The polarization image fusion method was studied to reduce the influence of rail surface characteristics on the image quality of structured light.Through theoretical analysis and field test verification,it is shown that the proposed method can obtain accurate,effective and stable basic data for dynamic inspection of rail structure components by machine vision.⑵The rail surface damage detection method based on multi-level feature fusion was proposed,and the rail surface scratch data set and damage detection network based on multi-source data of rail structural components were established.A rail surface scratch detection algorithm based on multi-layer feature fusion deep learning network was proposed to increase the detection rate of rail surface damage,to reduce the false detection rate,and to realize the measurement of scratch depth.⑶A detection method of rail fasteners was proposed by combining two-dimensional strength and three-dimensional depth information,and a dynamic template matching method was constructed to locate the position of the fasteners from the two-dimensional strength map.A deep learning algorithm was built to segment the corresponding 3D depth map coupler objects.According to the position and geometric prior information of track facilities,the model of fastener tightness was established to realize the automatic detection of fastener tightness.⑷Based on the above methods,a dynamic detection system for rail structural components was built,and some field tests were carried out.The experimental results show that the dynamic detection method based on 3D vision can meet the data acquisition speed of 80km/h.The resolution of scanning direction is 2.86 mm,and the resolution of depth direction is 0.15 mm.Rail surface scratch detection can effectively reduce false alarm,improve detection accuracy,and realize scratch depth measurement.Fastener detection can effectively detect the abnormal state of loose or overtight fasteners,and the detection rate is higher than 90%,which realizes the automatic detection of the tightness of fasteners.Through the above research,the dynamic detection ability of track structural components and the intelligent level of data analysis have been improved,which provides technical and equipment support for ensuring the safe operation of China’s railway.