摘要
针对冷金属过渡焊(CMT)熔滴轮廓特征,联合目标检测和图像分割算法设计了多任务结构网络,以监视熔滴方位并提取出精确的外轮廓,实现对CMT熔滴过渡过程的图像信息数据化,从而反映CMT送丝过程的熔滴行为。将高速摄像机拍摄的原图像直接输入网络,R-Net实现对熔滴位置的定位,此过程精度可达到97.31%,并在此基础上直接生成初步掩码。多任务结构网络主体部分对初步掩码进行精细化处理并结合初步分割的前景图,可实现对熔滴轮廓的提取,轮廓拟合精度可达98.8%。此外,利用以上所设计的网络,可得到CMT完整焊接周期内熔滴形貌的变化,实现对CMT熔滴过渡过程的可视化和数据化。
A multi-task structured network was designed for the cold metal transfer weld(CMT)weld droplet contour feature by combining the target detection and image segmentation algorithms,which could monitor the weld droplet position orientation and extract the accurate outer contour to realize the datamation of image information for the CMT weld droplet transition process,thus reflecting the weld droplet behavior of the CMT wire feeding process.The original images taken by the high-speed camera were input directly into the network,and the positioning of the weld droplet was realized by using R-Net with an accuracy of 97.31%and a preliminary mask was generated directly on this basis.The preliminary mask was refined by the main part of the multi-task structured network and combined with the preliminary segmented foreground map,the weld droplet contour could be extracted,with a fitting accuracy of 98.8%.In addition,using the above-designed network,the change of weld droplet morphology during the complete welding cycle of CMT can be obtained to visualize and datamation the CMT weld droplet transition process.
作者
管森
邢彦锋
张小兵
曹菊勇
GUAN Sen;XING Yanfeng;ZHANG Xiaobing;CAO Juyong(School of Mechanical and Automobile Engineering,Shanghai University of Engineering Science,Shanghai 201620,China)
出处
《机床与液压》
北大核心
2023年第20期71-76,共6页
Machine Tool & Hydraulics
基金
上海市自然科学基金项目(20ZR1422600)。
关键词
CMT
熔滴过渡
多任务结构网络
轮廓提取
CMT
Weld droplet transition
Multi-task structured network
Contour extraction
