Rail Wear Detection Algorithm Based On Geometric Features

The value of rail wear is an important parameter to guarantee the safe. In this paper, a rail abrasion detection algorithm based on geometric features has been presented based on the theoretical method of machine vision.The algorithm can implement fast and accurate calculation and high precision matching of rail wear quantity, and it has an important significance to guide maintenance and repair, and to guarantee the safety of train.Method for describing the geometrical features of rail and geometric features detection based on Hough transform has been researched. The mathematic description of rail contour’s geometric features is from the two aspects of linear and circular. The region of unworn contour line is considered as the target region. And then make the detection of rail line and circle of the target region using Hough transform method. The finished line detection result error is 0.26%, the error of the center coordinates is(0.382, 0.78) of circle detection result. The results show that the Hough transform method for the detection of linear and circular have a certain precision, and laid a theoretical foundation for contour matching.Designing the whole section of rail profile research measurement system, researching Levenberg- Marquardt optimization algorithm and resampling algorithm processing the whole section contour data, through these researches can obtain standardization of rail contour line. The simulation results of L-M optimization algorithm show that the error precision is increased by 0.02, the results verified the reliability and accuracy of the over rail section contour system and the L-M optimization algorithm.The measurement system of whole section of rail profile is built, and adopt the system to detect rail vertical abrasion, side abrasion and total abrasion, and comparing with the measuring data of rail head profile measuring instrument. The experimental results show that the error of vertical abrasion wear is 0.086, the side abrasion wear is 0.0434, the total wear 0.1007. These errors are within the allowable range. Therefore, The L-M algorithm can provide a fast and accurate calculation method for the measurement of full face wear of rail.The preliminary design of the full section rail abrasion detection system based on the Qt platform is designed, including the implementation of image acquisition function and the GUI graphical user interface.