焊接轨迹跟踪控制中的深度视觉研究进展  

Progress in Vision Technology in Welding Trajectory and Control

作　　者：王仕仙[1,2] 陈绪兵[1] WANG Shixian;CHEN Xubing(School of Mechanical and Electrical Engineering,Wuhan Institute of Technology,Wuhan 430205,China;College of Post and Telecommunication of Wuhan Institute of Technology,Wuhan 430074,China)

机构地区：[1]武汉工程大学机电工程学院,湖北武汉430205 [2]武汉工程大学邮电与信息工程学院,湖北武汉430074

出　　处：《武汉工程大学学报》2023年第4期378-383,417,共7页Journal of Wuhan Institute of Technology

基　　金：湖北省中央引导地方科技发展专项(2019ZYYD010)。

摘　　要：建立深度视觉的焊缝跟踪控制系统,能准确快速识别焊缝轨迹,实现焊缝实时跟踪,维持焊炬在整个焊接过程中始终与焊缝精确实时对中,在自动焊接领域应用潜力巨大。总结了传统图像焊缝识别方法和深度图像焊缝识别方法的发展现状,归纳分析了传统图像识别方法、焊接面积预测方法、焊缝图像识别改进算法、多样本训练学习法、神经网络法等方法的关键技术及优缺点。阐述了视觉焊缝跟踪技术的进展,经过图像采集和深度学习处理,实现了机器人运动轨迹的实时偏差修正,可以有效地实时提取初始焊缝,提高了自动焊接质量与效率。建立适用场合广、鲁棒性好、精度高、效率好、安全机制完善的深度视觉焊缝跟踪系统是未来进一步研究的方向。Establishing a deep vision weld seam tracking control system has enormous application potential in the field of automatic welding.It can accurately and quickly identify weld seam trajectory,achieve real-time weld seam tracking,and maintain accurate real-time alignment of the welding torch with the weld seam throughout the entire welding process.The development status of traditional image weld seam recognition methods and deep image weld seam recognition methods was summarized.Then,the advantages and disadvantages of key technologies were analyzed,including traditional image recognition methods,welding area prediction methods,improved weld seam image recognition algorithms,diverse training learning methods,neural network methods,and so on.The progress of visual weld seam tracking technology was elaborated.Through image acquisition and deep learning processing,a real-time deviation correction of robot motion trajectory was achieved,which effectively extracts initial weld seams in real-time and improves the quality and efficiency of automatic welding.It is a future direction of research to establish a deep vision weld seam tracking system with wide applicability,good robustness,high accuracy,good efficiency,and complete safety mechanisms.

关 键 词：深度视觉 焊缝识别 轨迹跟踪 图像处理 自动焊接 

分 类 号：TP242.6[自动化与计算机技术—检测技术与自动化装置]