基于改进YOLOv5的阀冷系统主循环泵电机故障检测方法  

Fault Detection Method for Main Circulation Pump Motor in Valve Cooling System Based on Improved YOLOv5

作　　者：颜大涵 谢雪花 王键 辛妍丽 林泽康 唐文虎[2] YAN Dahan;XIE Xuehua;WANG Jian;XIN Yanli;LIN Zekang;TANG Wenhu(Shantou Power Supply Bureau of China Southern Power Grid Co.,Ltd.,Shantou 515000,China;School of Electric Power Engineering,South China University of Technology,Guangzhou 510641,China;School of Automation,Guangdong Polytechnic Normal University,Guangzhou 510665,China)

机构地区：[1]南方电网广东汕头供电局,广东汕头515000 [2]华南理工大学电力学院,广东广州510641 [3]广东技术师范大学自动化学院,广东广州510665

出　　处：《电工技术》2023年第15期27-32,37,共7页Electric Engineering

基　　金：广东电网科技项目“基于模型驱动+数据驱动的换流阀冷却系统损耗优化及其设备运行状态监测研究”(编号GDKJXM20200761)。

摘　　要：主循环泵作为柔直换流站阀冷系统的核心设备,对于维护阀冷系统安全稳定运行具有重要作用。为了对主循环泵电机的红外图像进行精准定位与状态识别,提出了一种基于改进YOLOv5的主循环泵电机故障检测方法。首先,使用全新的卷积网络ConvNeXt作为YOLOv5的主干网络,提高网络的检测精度;同时,将定位损失函数替换为有效交并比损失函数(EIOU Loss),提高网络在训练过程中的收敛精度;然后,对改进YOLOv5网络使用数据增强、标签平滑、指数移动平均、迁移学习等策略进行训练,提高网络训练效率;最后,设置实验对所提改进方法进行验证,结果表明该方法能有效提高模型的检测精度,模型最终的均值平均精度(mAP)值达到95.82%,可使电机的故障平均检测精度提高至94.34%。As the core equipment of a valve cooling system of flexible DC converter station,the circulating pump plays an important part in maintaining the safe and stable operation of the valve cooling system.In order to accurately locate and recognize the defects from infrared images of a motor of circulating pump,a motor fault detection method based on improved YOLOv5 is proposed in this paper.Firstly,the new convolutional network ConvNeXt is used as the backbone network of YOLOv5 to improve the recognition accuracy.At the same time,the localization loss function is replaced with Efficient Intersection over Union Loss(EIOU Loss)to improve the convergence accuracy of the network in the training process.Then,the improved YOLOv5 network is trained using data augmentation,label smoothing,exponent moving average and transfer learning strategies to increase network training efficiency.Finally,experiments are set up to validate the proposed improvement methods.The experiment results show that the network improvement methods could effectively increase the detection accuracy of the model.The mean of average precision(mAP)of the final model reaches 95.82%,and the average fault detection accuracy improves to 94.34%.

关 键 词：主循环泵 故障检测 红外图像 YOLOv5 ConvNeXt 

分 类 号：TM3071[电气工程—电机]