Abstract: Currently, most trains are equipped with dedicated cameras for capturing pantograph videos. Pantographs are core to the high-speed-railway pantograph-catenary system, and their failure directly affects the normal operation of high-speed trains. However, given the complex and variable real-world operational conditions of high-speed railways, there is no real-time and robust pantograph fault-detection method capable of handling large volumes of surveillance video. Hence, it is of paramount importance to maintain real-time monitoring and analysis of pantographs. Our study presents a real-time intelligent detection technology for identifying faults in high-speed railway pantographs, utilizing a fusion of self-attention and convolution features. We delved into lightweight multi-scale feature-extraction and fault-detection models based on deep learning to detect pantograph anomalies. Compared with traditional methods, this approach achieves high recall and accuracy in pantograph recognition, accurately pinpointing issues like discharge sparks, pantograph horns, and carbon pantograph-slide malfunctions. After experimentation and validation with actual surveillance videos of electric multiple-unit train, our algorithmic model demonstrates real-time, high-accuracy performance even under complex operational conditions.

基于自注意力和卷积特征融合的高铁受电弓实时智能故障检测

[1]BochkovskiyA, WangCY, LiaoHYM, 2020. YOLOV4: optimal speed and accuracy of object detection. arXiv: 2004.10934.

[2]ChenRC, LinYZ, JinT, 2022. High-speed railway pantograph-catenary anomaly detection method based on depth vision neural network. IEEE Transactions on Instrumentation and Measurement, 71:1502710.

[3]DengQH, ChenYG, ZhangYP, 2022. Research on wear detection of pantograph slide plate based on high speed 3D structured light detection. Engineering and Technological Research, 7(12):15-19 (in Chinese).

[4]DingXH, ZhangXY, MaNN, et al., 2021. RepVGG: making VGG-style ConvNets great again. IEEE/CVF Conference on Computer Vision and Pattern Recognition, p.13728-13737.

[5]DosovitskiyA, BeyerL, KolesnikovA, et al., 2021. An image is worth 16×16 words: transformers for image recognition at scale. International Conference on Learning Representations.

[6]FangYT, MaJE, 2023. High-speed railway transport technology. Journal of Zhejiang University-SCIENCE A, 24(3):173-176.

[7]GuoJY, HanK, WuH, et al., 2022. CMT: convolutional neural networks meet vision transformers. IEEE/CVF Conference on Computer Vision and Pattern Recognition, p.12165-12175.

[8]HaoK, ChenGK, ZhaoL, et al., 2022. An insulator defect detection model in aerial images based on multiscale feature pyramid network. IEEE Transactions on Instrumentation and Measurement, 71:3522412.

[9]HeKM, ZhangXY, RenSQ, et al., 2016. Deep residual learning for image recognition. IEEE Conference on Computer Vision and Pattern Recognition, p.770-778.

[10]HeT, ZhangZ, ZhangH, et al., 2019. Bag of tricks for image classification with convolutional neural networks. IEEE/CVF Conference on Computer Vision and Pattern Recognition, p.558-567.

[11]HuTY, MaHM, LiuH, et al., 2022. Self-attention-based machine theory of mind for electric vehicle charging demand forecast. IEEE Transactions on Industrial Informatics, 18(11):8191-8202.

[12]KaradumanG, AkinE, 2022. A new approach based on predictive maintenance using the fuzzy classifier in pantograph-catenary systems. IEEE Transactions on Intelligent Transportation Systems, 23(5):4236-4246.

[13]KrizhevskyA, SutskeverI, HintonGE, 2017. Imagenet classification with deep convolutional neural networks. Communications of the ACM, 60(6):84-90.

[14]LiD, PanX, FuZZ, et al., 2022. Real-time accurate deep learning-based edge detection for 3-D pantograph pose status inspection. IEEE Transactions on Instrumentation and Measurement, 71:5001012.

[15]LinTY, DollárP, GirshickR, et al., 2017. Feature pyramid networks for object detection. IEEE Conference on Computer Vision and Pattern Recognition, p.936-944.

[16]LiuWJ, ZhaoJ, WangSF, 2021. Pantograph slide thickness detection method research based on machine vision. Electronic Measurement Technology, 44(24):128-133 (in Chinese).

[17]MoXF, WangKL, PanCQ, et al., 2022. Intelligent defect-detection technology of pantograph pan based on the image from railway 5C device. China Railway, (2):148-155 (in Chinese).

[18]MuZH, QinY, YuCC, et al., 2023. Adaptive cropping shallow attention network for defect detection of bridge girder steel using unmanned aerial vehicle images. Journal of Zhejiang University-SCIENCE A, 24(3):243-256.

[19]NiXF, MaZJ, LiuJW, et al., 2022. Attention network for rail surface defect detection via consistency of intersection-over-union (IoU)-guided center-point estimation. IEEE Transactions on Industrial Informatics, 18(3):1694-1705.

[20]PanXR, GeCJ, LuR, et al., 2022. On the integration of self-attention and convolution. IEEE/CVF Conference on Computer Vision and Pattern Recognition, p.805-815.

[21]SimonyanK, ZissermanA, 2015. Very deep convolutional networks for large-scale image recognition. International Conference on Learning Representations, p.1-14.

[22]TanBK, WangD, ShiJL, et al., 2024. Temperature field prediction of steel-concrete composite decks using TVFEMD-stacking ensemble algorithm. Journal of Zhejiang University-SCIENCE A, 25(9):732-748.

[23]TanP, MaJE, ZhouJ, et al., 2016. Sustainability development strategy of China’s high speed rail. Journal of Zhejiang University-SCIENCE A, 17(12):923-932.

[24]TanP, LiXF, XuJM, et al., 2020. Catenary insulator defect detection based on contour features and gray similarity matching. Journal of Zhejiang University-SCIENCE A, 21(1):64-73.

[25]TanP, LiXF, WuZG, et al., 2021. Multialgorithm fusion image processing for high speed railway dropper failure-defect detection. IEEE Transactions on Systems, Man, and Cybernetics: Systems, 51(7):4466-4478.

[26]TanP, LiXF, DingJ, et al., 2022. Mask R-CNN and multifeature clustering model for catenary insulator recognition and defect detection. Journal of Zhejiang University-SCIENCE A, 23(9):745-756.

[27]VaswaniA, ShazeerN, ParmarN, et al., 2017. Attention is all you need. Proceedings of the 31st International Conference on Neural Information Processing Systems, p.6000-6010.

[28]WangWH, XieEZ, SongXG, et al., 2019. Efficient and accurate arbitrary-shaped text detection with pixel aggregation network. IEEE/CVF International Conference on Computer Vision, p.8439-8448.

[29]WeiXK, JiangSY, LiY, et al., 2020. Defect detection of pantograph slide based on deep learning and image processing technology. IEEE Transactions on Intelligent Transportation Systems, 21(3):947-958.

[30]WenYT, ChengJX, RenYX, et al., 2024. Complex defects detection of 3-D-printed lattice structures: accuracy and scale improvement in YOLO V7. IEEE Transactions on Instrumentation and Measurement, 73:5013209.

[31]WooS, ParkJ, LeeJY, et al., 2018. CBAM: convolutional block attention module. The 15th European Conference on Computer Vision, p.3-19.

[32]WuY, FuHR, BianXC, et al., 2023. Impact of extreme climate and train traffic loads on the performance of high-speed railway geotechnical infrastructures. Journal of Zhejiang University-SCIENCE A, 24(3):189-205.

[33]XieSN, GirshickR, DollárP, et al., 2017. Aggregated residual transformations for deep neural networks. IEEE Conference on Computer Vision and Pattern Recognition, p.5987-5995.

[34]YaoH, LiuYH, LiX, et al., 2022. A detection method for pavement cracks combining object detection and attention mechanism. IEEE Transactions on Intelligent Transportation Systems, 23(11):22179-22189.

[35]ZhengZH, WangP, LiuW, et al., 2020. Distance-IoU loss: faster and better learning for bounding box regression. Proceedings of the AAAI Conference on Artificial Intelligence, p.12993-13000.

[36]ZhouN, ZhangWH, LiRP, 2011. Dynamic performance of a pantograph-catenary system with the consideration of the appearance characteristics of contact surfaces. Journal of Zhejiang University-SCIENCE A, 12(12):913-920.