基于Mask-RCNN的牵引供电设备红外智能诊断  

Intelligent infrared diagnosis method for traction power supply equipment based on Mask-RCNN

作　　者：林珊 邓树 谌小莉 王林 程宏波 LIN Shan;DENG Shu;CHEN Xiaoli;WANG Lin;CHENG Hongbo(Guangzhou Metro Design&Research Institute Co.,Ltd.,Guangzhou,Guangdong 510010,China;School of Electrical and Automation Engineering,East China Jiaotong University,Nanchang,Jiangxi 330013,China)

机构地区：[1]广州地铁设计研究院股份有限公司,广东广州510010 [2]华东交通大学,电气与自动化工程学院,江西南昌330013

出　　处：《机车电传动》2022年第4期55-61,共7页Electric Drive for Locomotives

基　　金：江西省重点研发计划项目(20202BBEL53008);江西省自然科学基金重点项目(2021ACB204004)。

摘　　要：牵引供电设备的状态诊断存在红外图像人工处理效率低下,智能化程度不高的问题,因此提出一种基于Inception-V3和Mask-RCNN的双层网络模型,通过Inception-V3网络实现电力设备类型识别,在此基础上,利用Mask-RCNN模型实现不同设备结构区域的自动划分,并根据划分的结构区域坐标提取不同区域的最高温度,构造温度特征量,依据设备类型调用不同的判据对设备状态进行自动诊断。试验结果表明,利用双层改进网络模型进行电力设备结构划分的整体mAP值可达0.907 2,进行设备故障诊断的效率比人工提高95.41%,模型精度高、识别效果好,无需依赖故障样本,提升了设备红外图像处理诊断的效率、降低了工作人员的劳动强度。Aiming at the low efficiency of the manual infrared image processing and the low intelligence degree in the state diagnosis of traction power supply equipment, a two-layer network model based on Inception-V3 and Mask-RCNN was proposed in this paper.In this diagnosis method, the first step was to identify power equipment types through Inception-V3 network;on this basis, Mask-RCNN was used to realize automatic division of different equipment structural regions;according to the coordinates of the divided structural regions, the highest temperatures of different regions were extracted, the temperature characteristic quantities were constructed, and the equipment status was automatically diagnosed by invoking different criteria according to the type of equipment. The experimental results show that the overall m AP value of power equipment structural division by using the double-layer improved network model can reach0.907 2, and the efficiency of equipment fault diagnosis can be improved by 95.41% compared with manual processing. The model featuring a high accuracy and good recognition effect, works independent of fault samples, which improves the efficiency of infrared image processing in equipment diagnosis and reduces the labor intensity.

关 键 词：牵引供电设备 深度学习 红外检测 故障诊断 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TM507[自动化与计算机技术—计算机科学与技术]