Real-time monitoring method for bolt loosening in railway hump retarders based on top-down strategy
-
摘要:
为实现铁路驼峰减速器在持续冲击振动、气候变化及列车通过性遮挡等复杂工况下螺栓松动状态的自动化精准监测，提出了一种基于Top-down策略的铁路驼峰减速器螺栓松动实时监测方法。采用“检测-定位-解算”三级协同流程，通过RTMDet模型实现对螺栓表面区域的精准定位，再利用RTMPose模型在检测框内识别螺栓6个角点，进而提出一套包括几何有效性判定、透视畸变校正和基于对称性约束的角度归一化算法，完成旋转角度的定量计算与状态判断；为验证该方法有效性，在铁路编组站现场布设高清摄像设备，构建包含晴天、阴天、雨雪、雾霾等多种天气条件下的、早晚及夜间不同时段的、局部油污污染条件下的、多角度条件的螺栓图像数据集，开展对比试验。研究结果表明：本文方法准确率为95.1%，召回率为93.8%，螺栓角点定位的均方根误差为1.25像素，旋转角度测量的决定系数为0.952，平均单帧处理时间为28.3 ms；本文方法在不同安装角度、遮挡及光照变化条件下均表现出良好的鲁棒性与泛化能力，角度计算误差低于1°。本研究为解决复杂铁路环境下螺栓松动的高精度、实时视觉监测提供了可行的技术方案，可应用于编组站减速器、道岔等关键铁路设备的智能运维与状态监测系统。
Abstract:To achieve automated and accurate monitoring of bolt loosening states in railway hump retarders under complex operating conditions, including continuous impact vibration, weather variations, and occlusion caused by passing trains, a real-time monitoring method for bolt loosening in railway hump retarders based on a top-down strategy was proposed. A three-stage collaborative process comprising detection, localization, and calculation was adopted. Accurate localization of bolt surface regions was achieved using the RTMDet model. Six corner points of each bolt were then identified within the detection box using the RTMPose model. An algorithm including geometric validity determination, perspective distortion correction, and angle normalization based on symmetry constraints was further proposed. The algorithm was used to quantitatively calculate the rotation angle and determine the bolt state. To verify the effectiveness of the method, high-definition camera equipment was deployed at a railway marshalling yard. A bolt image dataset was constructed under various weather conditions, including sunny, cloudy, rainy, snowy, and hazy conditions, and at different times of day, including morning, evening, and night. The dataset also included images with localized oil contamination and from multiple viewing angles. Comparative experiments were performed to evaluate the proposed method. Research results show that the proposed method achieves an accuracy of 95.1% and a recall of 93.8%. For bolt corner-point localization, the method achieves a root-mean-square error of 1.25 pixels and a coefficient of determination of 0.952 for rotation-angle estimation, with an average processing time of 28.3 ms per frame. The method demonstrates good robustness and generalization ability under varying installation angles, occlusion conditions, and illumination conditions, with angular errors below 1°. The proposed method provides a feasible technical solution for high-accuracy, real-time visual monitoring of bolt loosening in complex railway environments. This approach can be applied to intelligent operation and maintenance systems and condition monitoring systems for critical railway equipment, such as retarders and turnouts in marshalling yards.
-
图 3 现场监测立杆布设与相机视角
Figure 3. On-site monitoring pole deployment and camera viewing angle
图 4 基于Top-down策略的铁路驼峰减速器螺栓松动实时监测方法技术路线
Figure 4. Technical roadmap of the real-time monitoring method for bolt loosening in railway hump retarders based on top-down strategy
图 5 面向螺栓目标检测的RTMDet网络整体架构
Figure 5. Overall architecture of the RTMDet network for bolt target detection
图 6 面向螺栓关键点检测的RTMPose网络整体架构
Figure 6. Overall architecture of RTMPose network for bolt critical point detection
图 7 上海局某站平面布局及试验点
Figure 7. Layout and test points at a certain station in Shanghai Bureau
图 12 基于COCO评估指标的模型性能收敛曲线
Figure 12. Model performance convergence curves based on COCO evaluation metric
表 1 不同算法关键指标的性能表现
Table 1. Comparison of key performance metrics for various algorithms
算法 平均检测时间/ms 参数量/106 准确率/% 召回率/% RMSE/像素 决定系数R2 1 12.5 42.3 58.5 2 45.2 12.8 75.6 72.1 6.85 0.812 3 38.7 18.5 78.9 81.3 4.31 0.865 4 42.1 21.3 86.2 87.5 1.98 0.892 5 40.5 19.7 89.7 88.2 4.05 0.901 本文算法 28.3 9.4 95.1 93.8 1.25 0.952 
下载: 导出CSV
表 2 不同算法在3组典型场景下的螺栓松动检测与角度测量性能对比
Table 2. Comparative performance of algorithms for bolt loosening detection and angle measurement under three typical scenarios
算法 松动判定 初始角度/（°） 终止角度/（°） 角度变化量/（°） 绝对误差/（°） 相对误差/% 检测时间/ms 典型数据1 理论值 是 10.12 26.47 16.35 算法1 是 算法2 是 11.25 28.03 16.78 0.43 2.63 47.63 算法3 是 9.23 25.67 16.44 0.09 0.55 36.24 算法4 是 10.84 27.15 16.31 0.04 0.24 44.95 算法5 是 9.95 25.78 15.83 0.52 3.18 42.78 本文算法 是 9.98 26.50 16.52 0.17 1.04 26.42 典型数据2 理论值 是 21.35 17.28 -4.07 算法1 算法2 是 19.87 19.45 -0.42 3.65 89.68 48.91 算法3 是 22.41 16.35 -6.06 1.99 48.89 40.35 算法4 是 20.12 18.73 -1.39 2.68 65.85 43.16 算法5 是 22.67 16.05 -6.62 2.55 62.65 38.49 本文算法 是 21.04 17.29 -3.75 0.32 7.86 30.87 典型数据3 理论值 否 24.63 24.63 0.00 算法1 算法2 是 26.15 25.92 -0.23 0.23 41.05 算法3 否 23.89 24.95 1.06 1.06 37.62 算法4 否 23.75 25.41 1.66 1.66 39.84 算法5 否 25.33 23.87 -1.46 1.46 43.72 本文算法 否 24.54 25.27 0.73 0.73 27.15
下载: 导出CSV
-
[1] 国家铁路局. 2024年铁道统计公报[R/OL]. 2025-06-06. https://www.nra.gov.cn/xwzx/zlzx/hytj/202506/t20250606_348988.shtml .National Railway Administration. 2024 railway statistical bulletin[R/OL]. 2025-06-06.https://www.nra.gov.cn/xwzx/zlzx/hytj/202506/t20250606_348988.shtml .[2] 仇帅, 夏胜利. 长大货车驼峰溜放中车辆底部与轨面最小距离计算方法及应用[J]. 铁道运输与经济, 2025, 47(2): 206-213, 224.QIU Shuai, XIA Sheng-li. Calculation method of minimum distance between rail surface and bottom of oversize freight car during hump sliding and its application[J]. Railway Transport and Economy, 2025, 47(2): 206-213, 224. [3] 屠志平, 马天宇, 安岩. 驼峰车辆减速器制动性能影响因素分析[J]. 铁道通信信号, 2024, 60(6): 35-41.TU Zhi-ping, MA Tian-yu, AN Yan. Analysis of influencing factors on braking performance of hump car retarder[J]. Railway Signalling & Communication, 2024, 60(6): 35-41. [4] 王前选, 王锐锋, 李虎, 等. 轨道车辆螺栓松动量与预紧力视觉检测方法研究[J]. 铁道科学与工程学报, 2023, 20(9): 3511-3524.WANG Qian-xuan, WANG Rui-feng, LI Hu, et al. Visual detection method of rail vehicle bolt looseness and pre-tightening force[J]. Journal of Railway Science and Engineering, 2023, 20(9): 3511-3524. [5] 刘艾莎, 王勇, 李金龙, 等. 基于关键点的列车螺栓松动状态检测算法研究[J]. 现代电子技术, 2025, 48(14): 178-186.LIU Ai-sha, WANG Yong, LI Jin-long, et al. Research on key point based detection algorithm of loose state of train bolts[J]. Modern Electronics Technique, 2025, 48(14): 178-186. [6] AELA P, CAI J F, JING G Q, et al. Vision-based monitoring of railway superstructure: A review[J]. Construction and Building Materials, 2024, 442: 137385. doi: 10.1016/j.conbuildmat.2024.137385 [7] SHEN S, LU H, SADOUGHI M, et al. A physics-informed deep learning approach for bearing fault detection[J]. Engineering Applications of Artificial Intelligence, 2021, 103: 104295. doi: 10.1016/j.engappai.2021.104295 [8] TANG S N, MA J T, YAN Z Q, et al. Deep transfer learning strategy in intelligent fault diagnosis of rotating machinery[J]. Engineering Applications of Artificial Intelligence, 2024, 134: 108678. doi: 10.1016/j.engappai.2024.108678 [9] GE L F, SADHU A. Domain adaptation for structural health monitoring via physics-informed and self-attention-enhanced generative adversarial learning[J]. Mechanical Systems and Signal Processing, 2024, 211: 111236. doi: 10.1016/j.ymssp.2024.111236 [10] 张万鹏, 江珂, 马祯, 等. 铁路线路环境安全技防系统集成应用方案研究[J]. 中国铁路, 2025(2): 159-165.ZHANG Wan-peng, JIANG Ke, MA Zhen, et al. Research on the integrated application scheme for railway line environmental safety technical protection systems[J]. China Railway, 2025(2): 159-165. [11] 刘亿, 李平, 封博卿, 等. 基于北斗技术的铁路工务基础设施监测应用总体框架研究[J]. 中国铁路, 2025(3): 17-26.LIU Yi, LI Ping, FENG Bo-qing, et al. Research on overall application framework of railway track maintenance infrastructure monitoring based on Beidou technology[J]. China Railway, 2025(3): 17-26. [12] 马小平, 王旭, 贾利民, 等. 基于数字孪生和MBSE的铁路防灾监测仿真推演系统设计[J]. 中国铁道科学, 2024, 45(3): 168-181.MA Xiao-ping, WANG Xu, JIA Li-min, et al. Design of railway disaster prevention monitoring simulation system based on digital twin and MBSE[J]. China Railway Science, 2024, 45(3): 168-181. [13] ZHANG Y, SUN X W, LOH K J, et al. Autonomous bolt loosening detection using deep learning[J]. Structural Health Monitoring, 2020, 19(1): 105-122. doi: 10.1177/1475921719837509 [14] 杜晓辉, 潘科欣, 刘博. 基于深度学习与图像配准的螺栓松动检测[J]. 兵器装备工程学报, 2023, 44(7): 9-17.DU Xiao-hui, PAN Ke-xin, LIU Bo. Bolt looseness detection based on deep learning and image registration[J]. Journal of Ordnance Equipment Engineering, 2023, 44(7): 9-17. [15] 董华军, 姚佳岐, 何晨阳, 等. 地铁车底松动紧固件视觉检测算法研究[J]. 铁道科学与工程学报, 2024, 21(9): 3775-3786.DONG Hua-jun, YAO Jia-qi, HE Chen-yang, et al. Visual detection algorithm of loose fasteners under the subway[J]. Journal of Railway Science and Engineering, 2024, 21(9): 3775-3786. [16] GONG H, DENG X J, LIU J H, et al. Quantitative loosening detection of threaded fasteners using vision-based deep learning and geometric imaging theory[J]. Automation in Construction, 2022, 133: 104009. doi: 10.1016/j.autcon.2021.104009 [17] ZHOU X Y, ZHUO J C, KRÄHENBÜHL P. Bottom-up object detection by grouping extreme and center points[C]//IEEE. 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New York: IEEE, 2019: 850-859. [18] DAI J F, QI H Z, XIONG Y W, et al. Deformable convolutional networks[C]//IEEE. 2017 IEEE International Conference on Computer Vision (ICCV). New York: IEEE, 2017: 764-773. [19] DUAN K W, BAI S, XIE L X, et al. CenterNet++ for object detection[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2024, 46(5): 3509-3521. doi: 10.1109/TPAMI.2023.3342120 [20] LYU C Q, ZHANG W W, HUANG H A, et al. RTMDet: An empirical study of designing real-time object detectors [PP/OL]. V2. arXiv (2022-12-16). https://arxiv.org/abs/2212.07784 .[21] JIANG T, LU P, ZHANG L, et al. RTMPose: Real-time multi-person pose estimation based on MMPose [PP/OL]. V2. arXiv (2023-07-23). https://arxiv.org/abs/2303.07399 .[22] GE Z, LIU S T, WANG F, et al. YOLOX: Exceeding YOLO series in 2021[PP/OL]. V2. arXiv(2021-08-06). https://arxiv.org/abs/2107.08430 .[23] REZATOFIGHI H, TSOI N, GWAK J, et al. Generalized intersection over union: A metric and a loss for bounding box regression[C]//IEEE. 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New York: IEEE, 2019: 658-666. [24] GE Z, LIU S T, LI Z M, et al. OTA: Optimal transport assignment for object detection[C]//IEEE. 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New York: IEEE, 2021: 303-312. [25] ZHANG S F, CHI C, YAO Y Q, et al. Bridging the gap between anchor-based and anchor-free detection via adaptive training sample selection[C]//IEEE. 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New York: IEEE, 2020: 9759-9768. [26] BOCHKOVSKIY A, WANG C Y, LIAO H Y M. YOLOv4: Optimal speed and accuracy of object detection[PP/OL]. V1. arXiv (2020-04-23). https://arxiv.org/abs/2004.10934 .[27] REDMON J, FARHADI A. YOLOv3: An incremental improvement[PP/OL]. V1. arXiv(2018-04-08). https://arxiv.org/abs/1804.02767 .[28] LIU Z, MAO H Z, WU C Y, et al. A ConvNet for the 2020s[C]//IEEE. 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New York: IEEE, 2022: 11976-11986. [29] DING X H, ZHANG X Y, HAN J G, et al. Scaling up your kernels to 31x31: Revisiting large kernel design in CNNs[C]//IEEE. 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New York: IEEE, 2022: 11963-11975. [30] HOWARD A G, ZHU M L, CHEN B, et al. MobileNets: Efficient convolutional neural networks for mobile vision applications[PP/OL]. V1. arXiv(2017-04-17). https://arxiv.org/abs/1704.04861 .[31] JIANG Y Q, TAN Z Y, WANG J Y, et al. GiraffeDet: A heavy-neck paradigm for object detection[C]//ICLR. 10th International Conference on Learning Representations. Washington DC: ICLR, 2022: 1-8. [32] WU Y X, HE K M. Group normalization[C]//Springer. Computer Vision-ECCV 2018. Cham: Springer, 2018: 3-19. [33] WEI X Y, LIU M H, LING Z, et al. Approximate convex decomposition for 3D meshes with collision-aware concavity and tree search[J]. ACM Transactions on Graphics, 2022, 41(4): 1-18. [34] LOSHCHILOV I, HUTTER F. Decoupled weight decay regularization[C]//ICLR. 7th International Conference on Learning Representations (ICLR 2019). Washington DC: ICLR, 2019: 1-10. [35] CHENG T H, SONG L, GE Y X, et al. YOLO-World: Real-time open-vocabulary object detection[C]//IEEE. 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New York: IEEE, 2024: 16901-16911. [36] HE T, ZHANG Z, ZHANG H, et al. Bag of tricks for image classification with convolutional neural networks[C]//IEEE. 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New York: IEEE, 2019: 558-567. [37] CHEN X Q, YANG C Z, MO J, et al. CSPNeXt: A new efficient token hybrid backbone[J]. Engineering Applications of Artificial Intelligence, 2024, 132: 107886. doi: 10.1016/j.engappai.2024.107886 [38] 成佳明, 靳慧, 郑子健, 等. 基于Top-down范式的施工现场钢结构焊缝坡口特征实时检测方法[J]. 东南大学学报(自然科学版), 2023, 53(6): 1100-1110.CHENG Jia-ming, JIN Hui, ZHENG Zi-jian, et al. Real-time detection method for weld feature of construction site steel structure based on top-down paradigm[J]. Journal of Southeast University (Natural Science Edition), 2023, 53(6): 1100-1110. -
计量
- 文章访问数: 7
- HTML全文浏览量: 4
- PDF下载量: 2
- 被引次数: 0
