基于深度学习的热态环件轧制过程几何状态视觉测量方法  

Visual measurement method of geometric state of hot ring rolling process based on deep learning

作　　者：汪小凯[1,2,3] 董杰[1,2,3] 华林[1,2,3] 韩星会[1,2,3] 武国庆 WANG Xiao-kai;DONG Jie;HUA Lin;HAN Xing-hui;WU Guo-qing(Hubei Key Laboratory of Advanced Technology for Automotive Components,Wuhan University of Technology,Wuhan 430070,China;Hubei Collaborative Innovation Center for Automotive Components Technology,Wuhan University of Technology,Wuhan 430070,China;Hubei Engineering Center of Material Green Precision Forming Technology and Equipment,Wuhan University of Technology,Wuhan 430070,China)

机构地区：[1]武汉理工大学现代汽车零部件技术湖北省重点实验室,湖北武汉430070 [2]武汉理工大学汽车零部件技术湖北省协同创新中心,湖北武汉430070 [3]武汉理工大学材料绿色精密成形技术与装备湖北省工程中心,湖北武汉430070

出　　处：《塑性工程学报》2022年第11期8-14,共7页Journal of Plasticity Engineering

基　　金：国家重点研发计划(2019YFB1704502);国家自然科学基金资助项目(U2037204,52175362)。

摘　　要：为解决高温、强光、汽雾以及氧化皮飞溅等恶劣工况下热态环件轧制过程几何状态的在线测量难题,提出了一种基于深度学习的热态环件轧制几何状态视觉测量方法。以深度学习为基础,结合热态环件轧制过程高温发亮的特点,利用YOLOv5算法的卷积神经网络进行训练,智能捕捉相机视野有效区域,大幅提高了图像处理算法的运算速度。为解决环件部分边缘被遮挡的问题,采用Linemod-2D算法,根据环件边缘梯度特征匹配外圆轮廓及位置状态,并基于卡尺工具边缘检测原理获取有效边缘特征点,利用迭代重加权最小二乘法进行拟合,得到了环件几何状态。实验结果表明,本方法鲁棒性较好,测量误差小于0.5 mm,算法平均耗时60 ms。To solve the difficulty of on-line measurement of geometric state under bad working conditions such as high temperature, strong light, steam fog and oxide skin splash during the hot ring rolling process, a vision measurement method of geometric state of hot ring rolling based on deep learning was proposed. Based on deep learning, combined with the characteristic of high temperature brightness in the hot ring rolling process, the convolution neural network of YOLOv5 algorithm was used for training, and the effective field of the camera′s view was intelligently captured, which greatly improves the operation speed of image processing algorithm. To solve the problem of partial occlusion of the ring′s edge, Linemod-2 D algorithm was used to match the contour and position state of the outer circle according to the edge gradient feature of ring. The effective edge feature points were obtained based on the edge detection principle of caliper tool, and the geometric state of ring was obtained by the iterative reweight least squares method. The experimental results show that the proposed method has good robustness, the measurement error is less than 0.5 mm, and the algorithm takes 60 ms on average.

关 键 词：热态环件 YOLOv5算法 Linemod-2D算法 卡尺工具算法 最小二乘法 深度学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]