摘要
弓网系统作为电力机车的关键供能系统,其运行状态直接决定了电力机车的受流质量,影响列车运行的安全和效率。为了解决传统弓网检测方法效率低、实时性差等问题,文章设计了一种基于深度学习的弓网状态在线检测系统方案,采用英伟达(NVIDIA)的Xavier SoC模块进行图像处理,通过YOLO v4实现弓网检测并添加自适应图像增强模块,优化前后目标检测的mAP分别为0.950和0.961(IOU阈值为0.5);实现基于ViT轻量级注意力模型的吊弦状态分类,平均准确率为97.69%;使用NVIDIA的推理加速器TensorRT加速后,检测模型和分类模型的推理时间分别为31.0 ms和2.2 ms。系统具有较高的鲁棒性与实用性,可为后续弓网异常在线检测功能拓展提供理论依据和设计参考。
The pantograph-catenary system is a key part of power supply for electric locomotive,and its operating status determines the current receiving quality of electric locomotive,as well as safety and efficiency of trains.In order to solve the problems of traditional method for inspection of pantograph-catenary status,such as low efficiency and poor performance in real time inspection,this paper designed an online pantograph-catenary status inspection system solution based deep learning.The solution adopted NIDIA Xavier SoC module to perform image processing,and realizes pantograph-catenary inspection by YOLO v4 and adaptive image enhancement module was also added.The mAP of inspection target before and after optimization was 0.950 and 0.961(with the IOU threshold of 0.5)respectively.Classification of catenary dropper status based on ViT lightweight class attention model was realized at an average accuracy rate of 97.69%.After acceleration by using NVIDIA TensorRT accelerator,the inference time of inspection model and classification model were 31.0 ms and 2.2 ms respectively.The system has high robustness and practicability,which can provide theory basis and design reference for online inspection function of pantograph-catenary abnormality in the future.
作者
周兆安
李树枝
ZHOU Zhaoan;LI Shuzhi(CRRC Qingdao Sifang Rolling Stock Co.,Ltd.,Qingdao,Shandong 26611l,China)
出处
《机车电传动》
北大核心
2022年第5期135-143,共9页
Electric Drive for Locomotives
