Research On Algorithm For Recognition And Registration Of Rail Abnormal Contours

Rails are an indispensable element of rail transportation.When trains work at high speed,heavy load,and high density,the rails will be squeezed,worn,and deformed,which will seriously affect the safety of trains.Therefore,it is very necessary to make timely,effective and accurate detection of track wear.The traditional manual detection has the shortcomings of low efficiency,large influence of artificial factors,and very dangerous,and can no longer meet the needs of my country’s rapid development of rail transit.At present,the on-board non-contact rail profile measurement is mainly realized by laser vision technology or laser displacement technology.The sensor placed on the bottom of the vehicle collects the light strip profile consistent with the rail crosssection shape,and then the sampled data is compared with the standard The profile data is registered,and then the rail wear at that location can be compared and calculated.Its high precision and high efficiency are the key research directions of track maintenance development at home and abroad.In this paper,a vehicle-mounted rail dynamic detection system based on a twodimensional laser displacement sensor is focused on two aspects: rail profile preprocessing identification and classification,and high-precision rail wear detection.Aiming at the problem of classification and identification of rail profiles,this paper uses the sampling data of two-dimensional laser sensors to propose a classification and identification method of effective profiles and abnormal profiles based on statistical analysis,which realizes the identification of effective profiles and the identification of different profiles.Identification of silhouette categories.First,compare the profile shapes of the rail profiles sampled from different areas in the line,and select the discontinuity at the rail head and the rail waist and the morphological correlation between the rail head and the rail waist as the identification evidence of the effective contour.Secondly,by using the distribution of rail profile point cloud data under different road conditions,it is possible to identify and classify abnormal profile data in the original data that is difficult to assess the degree of wear.The field test results show that the distribution of the effective contours identified by the method in this paper is roughly the same as the actual line.Under the condition of the driving test speed of 8.3km/h,the average identification time of a single contour is 9.78 ms,and the effective identification accuracy rate reaches 97%.above.In order to obtain high-precision detection results of rail wear,this paper proposes an iterative rotation registration method of rail profile based on differential evolution algorithm,which can improve the robustness and overall accuracy of registration.The study found that the Ramer polygon approximation algorithm can be used to locate the straight line inside the rail head at the inner turning line segment of the rail jaw.Then determine the rotation direction by comparing the relative position of the sampling contour and the standard contour at the same coordinates.Finally,the differential evolution algorithm is used for rotation iteration to achieve accurate registration of the rail profile.The experimental results show that,compared with other existing methods,this method can still meet the high registration accuracy when the sampling data is accompanied by heavy noise,even when the data is missing,which greatly improves the robustness of the detection system.,with strong engineering application value.At the end of this paper,based on 2D laser displacement sensor,a vehicle-mounted rail dynamic detection platform is constructed to collect some real data of the line,and the above algorithm is analyzed and verified,and the feasibility of the algorithm in this paper is demonstrated.