基于3D视觉与深度学习的焊接钢管端部引熄弧板切割点空间定位方法  

Spatial Positioning Method for Cutting Points of Tab Plate at the End of Welded Steel Pipes based on 3D Vision and Deep Learning

作　　者：石玮 周坤[1] 刘福海[1] 张焱[1] 杨战利[1] SHI Wei;ZHOU Kun;LIU Fuhai;ZHANG Yan;YANG Zhanli(Harbin Welding Institute Limited Company,CAM,Harbin 150023,China)

机构地区：[1]中国机械总院集团哈尔滨焊接研究所有限公司,黑龙江哈尔滨150023

出　　处：《电焊机》2024年第4期36-42,共7页Electric Welding Machine

基　　金：哈尔滨焊接研究院有限公司技术创新发展基金(202216215)。

摘　　要：针对目前焊接钢管端部引熄弧板切割点空间定位精度低、速度慢和环境适应性差的问题,提出了一种基于3D视觉技术和深度学习的高效、高精度的切割点空间定位方法。首先,利用3D视觉技术获取钢管端部的点云数据,通过三维点云处理方法提取有效点云和过滤点云中的离群点,并采用二维图像处理方法将其投影为二维图像;然后通过深度学习中目标检测方法确定二维图像中钢管端部与引熄弧板连接边缘区域;最后计算分析钢管端部与引熄弧板在二维图像中的交点信息并将交点信息映射至原始三维点云,完成钢管端部引熄弧板切割点的空间定位。实验测试了896个不同的钢管端部带有引、熄弧板点云数据(正样本)和104个不带有引熄弧板数据(负样本),统计测试结果:正样本管端和引熄弧板连接边缘定位准确率达到99.56%、误检率为0.15%,负样本检出率为0.29%。使用本文方法测试294次钢管端部引熄弧板切割点空间位置,统计测试结果与实际切割点位综合误差,结果均在±2 mm以内,钢管端部引熄弧板切割点空间定位平均耗时为5.8 s。测试结果表明,该方法可以有效解决上述问题。A new cutting point spatial positioning method is proposed to address the current issues of low accuracy,slow speed,and poor environmental adaptability of the cutting point space positioning of the end quenching arc plate of welded steel pipes.This method first extracts effective point clouds and filters outliers from point clouds using a three-dimensional point cloud processing method,and projects them into a two-dimensional image using a two-dimensional image processing method;Then,the object detection method in deep learning is used to determine the edge area of the connection between the end of the steel pipe and the arc extinguishing plate in the two-dimensional image;Finally,calculate and analyze the intersection information of the steel pipe end and the run on/off tab in the two-dimensional image and map the intersection Information mapping to the original three-dimensional point cloud to complete the spatial positioning of the cut point of the run on/off tab at the steel pipe end.Experimental tests were conducted on 896 different steel pipe ends with point cloud data(positive samples)and 104 data without arcing plates(negative samples).The statistical test results showed that the accuracy of edge positioning between the positive sample tube end and the arcing plate connection reached 99.7%,the false detection rate was 0.0%,and the negative sample detection rate was 0.96%.The method used in this article was used to test the spatial position of the cutting point of the steel pipe end quenching arc plate 294 times.The comprehensive error between the test results and the actual cutting point position was statistically analyzed,and the results were all within±2 mm.The average time for spatial positioning of the cutting point of the steel pipe end quenching arc plate was 5.8 seconds.The test results indicate that this method can effectively solve the above problems.

关 键 词：引熄弧板 点云处理 图像处理 深度学习 空间定位 

分 类 号：TG409[金属学及工艺—焊接]