Research On Wheelset Tread Profile Detection Based On Machine Vision

Over the past 40 years of reform and opening up,China Railway,especially China's high-speed railway,has achieved remarkable achievements in rapid development and has become a shiny "national business card".As the speed of railway trains continues to increase and the load capacity continues to increase,the importance of train operation safety issues is increasingly prominent.The wheelset is a key component in the train operation,and the outline size of the wheelset tread is one of the key parameters reflecting the state of the wheelset.Accurate and quick detection of the key dimensions of wheelset treads is essential to improve the safety of train operation.Compared with traditional measurement methods,the wheelset tread profile detection system based on machine vision has the advantages of high accuracy and high efficiency,and has been widely used in the measurement of wheelset tread geometric parameters.This article mainly studies a tread profile detection system which is suitable for the practical scene of depots.This research work was supported by the National Natural Science Foundation of China(No.51935002)."Basic problems and key technology research on dynamic monitoring of wheelset status under the extreme conditions of high speed and heavy load".The project carried out theoretical research,laboratory verification,and on-site verification for the on-site measurement part of the wheelset tread.This article completed the following work: Firstly,the theoretical principle of the wheelset tread profile detection system has been studied and analyzed,and a suitable technical scheme and algorithm route of on-site detection are determined.Then this paper studied the mathematical model of camera calibration and improves the calibration method using planar targets,optimizes both radial distortion and tangential distortion,and conducted verification experiments in the laboratory environment to prove its theoretical feasibility.Secondly,theoretical research was conducted on the coordinate constraints of the binocular vision system,and a mathematical model was constructed.For the algorithm of matching feature points,the differences between several algorithms were analyzed.And the actual experiment was conducted to optimize the most suitable algorithm and study the influence of threshold setting on matching effect.And we determined the most suitable algorithm is the improved algorithm based on SIFT method,which lays the foundation for subsequent image reconstruction.Finally,we installed measurement equipment on site and conducted in-field tests.We measured and analyzed the results of real wheelsets.The data shows that this system can meet the accuracy requirement of ±0.28)8)at rim height.The experiment also showed it has good robustness,which means the system meets the characteristics of practicability,high precision and high efficiency.