轮轨动态位移机器视觉检测方法研究

doi:10.3901/JME.260725

摘要/Abstract

摘要： 轮轨动态位移(Dynamic wheel-rail displacement，DWRD)是衡量铁道车辆轮轨系统交互状态的关键物理量，其检测精度直接影响列车运行安全评估与轮轨维护决策的可靠性。针对现有光学检测技术在复杂运营场景下存在的适应性差、精度不足或实时性低等问题，提出一种“辅助激光标记-轻量化深度学习-双目相机三维协同”的融合检测法。首先，采用蓝紫色单线激光构建轮轨区域高对比度参考标记，解决传统视觉检测中轮轨关键点特征模糊的难题；其次，基于改进ShuffleNet v2架构设计两阶段轻量化网络，分别实现轮轨感兴趣区域(Region of interest，ROI)快速定位与激光—轮轨交点自适应识别，在保证检测精度的同时提升计算效率；最后，结合双目相机三维坐标转换模型，解决单目相机无法感知深度的问题，消除相机视角偏移与轮对姿态变化引发的系统误差，并通过时序位移解算完成DWRD量化。为验证方法有效性，在某地铁线路开展工程应用研究，结果表明：所提方法检测速度可达19 ms/帧，轮辋宽度测量值与设计值的偏差率<0.8%；基于该方法获取的DWRD可精准识别车辆1 Hz左右的小幅蛇行失稳，且能反映线路曲率与DWRD幅值的正相关特性。研究成果可为铁道车辆蛇行运动稳定性评估、脱轨风险预警及轨道状态智能化维护提供技术支撑。

关键词: 铁道车辆, 轮轨动态位移, 机器视觉, 深度学习, 双目相机, 蛇行运动稳定性

Abstract: Dynamic wheel-rail displacement(DWRD) is a key physical quantity for measuring the interaction state of the wheel-rail system of railway vehicles, and its detection accuracy directly affects the reliability of train operation safety assessment and wheel-rail maintenance decision-making. Aiming at the problems of poor adaptability, insufficient accuracy or low real-time performance of existing optical detection technologies in complex operation scenarios, a fusion detection method of “auxiliary laser marking- lightweight deep learning-binocular camera 3D collaboration” is proposed. Firstly, a blue-violet single-line laser is used to construct high-contrast reference marks in the wheel-rail area, which solves the problem of blurred features of key wheel-rail points in traditional visual detection. Secondly, a two-stage lightweight network is designed based on the improved ShuffleNet v2 architecture to realize the rapid localization of the wheel-rail Region of interest(ROI) and the adaptive recognition of laser-wheel-rail intersection points respectively, which improves the computational efficiency while ensuring the detection accuracy. Finally, combined with the binocular camera 3D coordinate conversion model, the system errors caused by camera perspective offset and wheel set attitude change are eliminated, and the DWRD quantification is completed through time-series displacement calculation. To verify the effectiveness of the method, an engineering application study is carried out on a subway line. The results show that：the detection speed of the proposed method can reach 19 ms/frame, and the deviation rate between the measured value and the design value of rim width is less than 0.8%; the DWRD obtained based on this method can accurately identify the small-amplitude hunting instability of the vehicle at about 1 Hz, and can reflect the positive correlation characteristic between track curvature and DWRD amplitude. The research results can provide technical support for the assessment of hunting motion stability of railway vehicles, derailment risk early warning and intelligent maintenance of track status.

Key words: railway vehicle, dynamic wheel-rail displacement, machine vision, deep learning, binocular camera, hunting stability

中图分类号:

U270

叶运广, 王建吉, 李浩乾, 李大地, 魏来, 曾京. 轮轨动态位移机器视觉检测方法研究[J]. 机械工程学报, 2026, 62(16): 419-429.

YE Yunguang, WANG Jianji, LI Haoqian, LI Dadi, WEI Lai, ZENG Jing. Machine Vision Inspection Method for Dynamic Wheel-rail Displacement[J]. Journal of Mechanical Engineering, 2026, 62(16): 419-429.

导出引用管理器 EndNote|Reference Manager|ProCite|BibTeX|RefWorks

链接本文: http://www.cjmenet.com.cn/CN/10.3901/JME.260725

http://www.cjmenet.com.cn/CN/Y2026/V62/I16/419

参考文献

[1] YE Y，LI H，LI F，et al. Online assessment of train hunting stability by monitoring dynamic wheel-rail displacement：Why and how?[J]. Nonlinear Dynamics，2024，112：11993-12017.
[2] 逯万春，姜培斌，凌亮，等. 基于KCF-Hash-Match目标跟踪算法的高速列车车轮横向晃动识别方法[J]. 机械工程学报，2023，59(24)：223-230. LU Wanchun，JIANG Peibin，LING Liang，et al. KCF-Hash-Match target tracking algorithm for identifying wheel lateral sway of high-speed train[J]. Journal of Mechanical Engineering，2023，59(24)：223-230.
[3] WANG P，CHEN R，QUAN S. Development and application of wheelset lateral displacement test system in high speed railway turnout zone[J]. Przeglad Elektrothchniczny：Electrical Review，2012，88(1B)：69-73.
[4] JING G，SIAHKOUHI M，QIAN K，et al. Development of a field condition monitoring system in high speed railway turnout[J]. Measurement，2021，169：108358.
[5] PINTO N，RIBERIO C，GABRIEL J，et al. Dynamic monitoring of railway track displacement using an optical system[J]. Proceedings of the Institution of Mechanical Engineers，Part F：Journal of Rail and Rapid Transit，2015，229(3)：280-290.
[6] LIU K，LIU X. Study on determining the vertical wheel-rail force by measuring rail displacement[C]// Proceedings of the International Conference on Chemical，Material and Food Engineering，International Conference on Chemical，Material and Food Engineering. Atlantis Press，2015：856-859.
[7] ENGINEER L. Laser triangulation sensors measure lateral position of rail bogie wheels[EB/OL]. 2020[2021-12-27]. https://www.dpaonthenet.net/article/179382/Laser-triangulation-sensors-measure-lateral-position-of-rail-bogie-wheels.aspx.
[8] BURSTOW M，PODESTA M，PEARCE J. Understanding wheel/rail interaction with thermographic imaging[C]// 22nd International Symposium on Dynamics of Vehicles on Roads and Tracks，Manchester，2011：1-6.
[9] YAMAMOTO D. Improvement of method for locating position of wheel/rail contact by means of thermal imaging[J]. Quarterly Report of Railway Technical Research Institute，2019，60(1)：65-71.
[10] FIRLIK B，STASKIEWICZ T，SLOWINSKI M. Thermal imaging of the wheel-rail interface[J]. Proceedings of the Institution of Mechanical Engineers，Part F：Journal of Rail and Rapid Transit，2023，237(9)：1195-1204.
[11] KIM M. Measurement of the wheel-rail relative displacement for the active wheelsets steering system using the image processing algorithm[C]//Proceedings of the 13th WSEAS International Conference on Mathematical and Computational Methods in Science and Engineering，World Scientific and Engineering Academy and Society (WSEAS)，2011：242-246.
[12] SKRICKIJ V，SABANOVIC E，SHI D，et al. Visual measurement system for wheel-rail lateral position evaluation[J]. Sensors，2021，21(4)：1297.
[13] 钟莎. 基于Qt的机车轮轨相对位移的图像检测系统设计[D]. 石家庄：石家庄铁道大学，2016. ZHONG Sha. The design for image detection system of wheel-rail relative displacement based on Qt[D]. Shijiazhuang：Shijiazhuang Tiedao University，2016.
[14] 张光跃，马增强，苑佳靖，等. 基于轮轨相对横移的轨距检测方法研究[J]. 光电工程，2020，47(2)：48-55. ZHANG Guangyue，MA Zengqiang，YUAN Jiajing，et al. Track gauge measurement based on wheel-rail lateral relative displacement[J]. Opto-Electronic Engineering，2020，47(2)：48-55.
[15] 秦菲菲. 基于图像处理的列车脱轨系数预测方法的研究[D]. 兰州：兰州交通大学，2022. QIN Feifei. Research on prediction method of train derailment coefficient based on image processing[D]. Lanzhou：Lanzhou Jiaotong University，2022.
[16] CHEN Y，TIAN Y，HE M. Monocular human pose estimation：A survey of deep learning-based methods[J]. Computer Vision and Image Understanding，2020，192：102897.
[17] NEWELL A，YANG K，JIA D. Stacked hourglass networks for human pose estimation[C]//European Conference on Computer Vision. Springer International Publishing，2016：483-499.
[18] XIAO B，WU H，WEI Y. Simple baselines for human pose estimation and tracking[C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)，2018：472-487.
[19] SUN K，XIAO B，LIU D，et al. Deep high-resolution representation learning for human pose estimation[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)，2019：5693-5703.
[20] LUVIZON D，PICARD D，TABIA H. 2D/3D Pose estimation and action recognition using multitask deep learning[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR)，2018：5137-5146.
[21] BAZAREVSKY V，GRISHCHENKO I，RAVEENDRAN K，et al. BlazePose：On-device real-time body pose tracking[J]. Arxiv Preprint Arxiv：2006.10204，2020.
[22] SHI D，SABANOVIC E，RIZZETTO L，et al. Deep learning based virtual point tracking for real-time target-less dynamic displacement measurement in railway applications[J]. Mechanical Systems and Signal Processing，2022，166：108482.
[23] YE Y，GAO H，HUANG C，et al. Computer vision for hunting stability inspection of high-speed trains[J]. Measurement，2023，220：113361.
[24] ADARSH P，RATHI P，KUMAR M. Yolo v3-tiny：Object detection and recognition using one stage improved model[C]//20206th International Conference on Advanced Computing and Communication Systems (ICACCS). New York：IEEE，2020：687-694.
[25] ZHANG X，ZHOU X，LIN M，et al. ShuffleNet：An extremely efficient convolutional neural network for mobile devices[C]//Proceedings of the 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)，Salt Lake City，Jun 18-22，2018. New York：IEEE，2018：6848-6856.
[26] HE K，ZHANG X，REN S，et al. Deep residual learning for image recognition[C]//Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)，Las Vegas，Jun 27-30，2016. New York：IEEE，2016：770-778.
[27] ZHOU M，STOFFL L，MATHIS M，et al. Rethinking pose estimation in crowds：Overcoming the detection information-bottleneck and ambiguity[EB/OL]. arXiv.org，2023. https://doi.org/10.48550/arXiv.2306.0787.
[28] 中国国家市场监督管理总局，中国国家标准化管理委员会. GB/T 5599—2019. 机车车辆动力学性能评定及试验鉴定规范[S]. 北京：中国标准出版社，2019. State Administration for Market Regulation，Standardization Administration of the People’s Republic of China. GB/T 5599—2019. Specification for dynamic performance assessment and testing verification of rolling stock[S]. Beijing：Standards Press of China，2019.