面向数字孪生的工业设备数字化编码与识别  

Digital Coding and Recognition of the Industrial Equipment for Digital Twins

作　　者：薛正博 王建国 冯勇 李英娜 XUE Zheng-bo;WANC Jian-guo;FENG Yong;Ll Ying-na(Yurnan Prowincial Key Laboratory of Computer Technology Application;China Copper Corporation Limited)

机构地区：[1]云南省计算机技术应用重点实验室 [2]中国铜业有限公司

出　　处：《化工自动化及仪表》2025年第2期218-226,共9页Control and Instruments in Chemical Industry

基　　金：云南省重大科技专项计划(批准号:202202AD080006)资助的课题。

摘　　要：数字孪生技术实现的前提是虚实结合,为使现实中的工业设备与其作业中的虚拟信息相结合,需要对工业设备进行数字化编码并制定设备标识图像分割识别的统一方法,为工作人员提供相关信息从而控制和维护设备。采用版本号为3,纠错等级为L的标准QR二维码对工业设备进行数字编码,设计并使用MA U-Net(多重注意力U型网络)图像分割神经网络对设备标识图像进行分割,并在训练过程中加入特殊的图像增强方法以提升图像分割效果,再通过二维码解码器解码标识图像获取设备编号等基础信息,进而查询数据库获取工业设备信息,提升工厂作业效率。在数据集上进行了多次实验,结果表明所提方法提高了工业设备标识图像的分割精度和识别准确率,基于分割精度的3个评估指标PA、IoU、Dice指数分别达到了97.22、71.66和80.95,高于对比的其他方法。二维码的识别准确率达到了96.32%,误差小于5%。证明所提方法适用于工厂作业场景,可以辅助数字孪生工厂的建设。The combination of virtuality and reality becomes the premise of implementing digital twin technology.In order to combine industrial equipment in reality with virtual information in its operations,digitally coding industrial equipment and developing the unified method for equipment identification,image segmentation and recognition becomes necessary so as to provide the relevant information for workers to control and maintain the equipment.In addition,having the standard QR code of Version 3 and error correction level“L”adopted to encode industrial equipment was implemented,including designing and applying MA U-Net(Multi-Attention U-Net)image segmentation neural network to segment the equipment identification image and add special image enhancement methods during the training process to improve the image segmentation effect;and then,basing on the QR code decoder,the identification image was decoded to obtain basic information such as equipment number,query the database to obtain specific information of industrial equipment so as to improve the efficiency of factory operations.Multiple experiments conducted on the dataset shows that,the proposed method can improve the segmentation accuracy and recognition accuracy of industrial equipment identification images and three evaluation indicators based on segmentation accuracy,PA,IoU,and Dice index,can reach 97.22,71.66,and 80.95 respectively,higher than the methods compared.The recognition accuracy of the QR code can reach 96.32%,with an error of less than 5%.It demonstrates that the proposed method complies with the factory scenarios and can assist the construction of digital twin factories.

关 键 词：数字孪生工厂 数字化编码 图像分割与识别 深度学习 

分 类 号：TP274[自动化与计算机技术—检测技术与自动化装置]