Study On Profile Data Pre-processing Method Of Rail Detection

Nondestructive dynamic detection method of rail profile is the mainstream way of track quality state today,which has great significance for improving the efficiency of railway maintenance and ensuring the safety of traffic.Recognizing the effective rail profiles from the original ones acquired by the rail profile detection system real-time and exactly to access the quality of track is a key challenge to rail inspection.This thesis aims to study a data pre-processing method based on rail profile detection and ensure the accuracy and integrity of measurement data,ultimately improve efficiency of track quality detection.The pre-processing method based on rail profile detection includes two core parts: one is recognizing the effective rail profiles,and another is noise elimination and filling the missing data of effective profiles.Firstly,on the basis of the comparative analysis of profiles located in different areas of the railway,the discontinuity between rail head and rail waist and the match between measured rail waist and standard one are used as the important basis to recognize the effective profiles.We match the rail jaw point and railhead inner line to realize the rough contour registration,then the measured profile is classified precisely by computing the directed Hausdorff distance of rail waist overlap areas and comparing it with statistical threshold.On the basis of profiles with the rail waist,nonlinear fitting was conducted on arcs of rail waist that were subject to radius constraint to extract the coordinates of the center of arcs.the profile's coarse-to-fine matching is implemented by combining with the center of standard and the center of measured rail waist.For some measured profiles containing scattered noise and missing data,the thesis propose a method include the noise elimination and filling the missing data.The denoising algorithm based on piecewise contour can remove profile noise,due to the similar of contiguous profile,and the missing data is filled by matching contiguous rail head.Finally,the experiment results of pre-processing method based on rail profile detection show that the distributions of effective profiles based on our proposed method are accurately coincided with the real conditions.The recognition rate reaches to 93.1%,and the average recognition speed reaches to 10.8ms per frame.The measured system can run at a speed of 83.3 km/h under our experimental setup.Through the process of denoising and filling missing data,the effective profiles had been eliminated the noise and enhanced the data integrity.With the accuracy of rail cross section as calculation index,the average error of the index between normal profile and the measured after data pre-processing is 0.1017 mm,and the results can meet precision requirement of rail dynamic detection.So the pre-processing method based on rail profile detection in this thesis provides a new idea and method for ensuring the accuracy of measured data and improving the efficiency of rail inspection.