【作者】 王涛；

【导师】 王志江；

【作者基本信息】 天津大学 ， 材料科学与工程， 2021， 硕士

【摘要】 焊接自动化、智能化是焊接技术发展的必然趋势。部分生产应用中焊接工件存在非标准化和定制化的特点,研究焊缝的识别与定位,有助于推动标准件焊接自动化向非标准件焊接自动化发展,实现覆盖领域更广、更智能的自动化焊接。目前,基于视觉的焊缝识别与定位的大部分研究应用中,需要先验知识或人为介入操控获取焊件信息,缺乏一定的自主性。文中主要探究基于双目立体视觉传感的非标准件角焊缝识别与定位的方法,建立相应的图像处理系统,解决角焊缝位置和结构信息提取的问题。探究了角焊缝焊件结构成像的特点,针对焊件表面特征较少且不明显以及特征相似度高易造成误匹配等问题,提出了采用边缘特征结合BRIEF描述符的方法对焊件进行立体匹配。在匹配过程中,采用一种多重BRIEF描述符匹配方法,利用不同BRIEF描述符的随机性涵盖不同焊件特性以有效剔除误匹配;通过极线约束提高匹配正确率和减少匹配时间,并结合视差滤波和统计滤波进一步剔除误匹配。通过与SURF和ORB等特征提取与匹配方法对比,该立体匹配方法匹配率高、能获得更多边缘信息、重构效果好,但匹配时间比SURF方法略高。基于角焊缝结构特点,提出了采用聚类分析和SVD平面拟合的方式对点云进行分类,并拟合得到焊件两个构件表平面的方程。通过求取两表平面方程交线获得焊缝位置信息;通过边缘拟合,求解边缘与焊缝交点的方式,获取焊缝始末点位置。对提取焊缝始末点位置的误差进行了总结分析,定量评估其准确性。结果显示,在x、y、z各个方向上的平均误差分别为0.5 mm、0.8 mm、0.9 mm,均在1 mm以内,基本满足预期要求。基于测试结果,指出了对文中提出方法精度影响的因素,包括特征点匹配结果、边缘检测结果完整性、边缘平整度和背景干扰等。基于焊接环境的复杂性和多样性,设计了不同背景复杂度、焊件材料、焊件角度等情况,对文中提出的角焊缝识别与定位方法的鲁棒性进行了测试。结果表明,该方法适用于多种情况变化下的角焊缝识别与定位。更多还原

【Abstract】 Welding automation and intelligence are the promising trends of welding technology.Since some weldments are non-standard or customized in production and application,the recognition and positioning of weld seams for such weldments are helpful to promote the automatic welding from the field of structured environment to non-structured environments,which could make the automatic welding with more intelligences and wider applications.In the research and application of recognition and positioning of weld seams based on vision technology in non-structured environment,prior knowledge or human manipulation are used to obtain weldment information,which makes it difficult to be applied in fully automated cases.This work mainly aims at exploring recognition and positioning of fillet welds in non-structured environment based on binocular stereo vision sensing,and establishing an image processing system to position the fillet weld and to detect its structure information.The images of fillet welds were investigated,which include few and indistinct surface features of the weldments,and are with pixels of similar features,therefore,less features and large mismatches in stereo matching may become a big trouble.To address the issues,the distinct edge information was taken as features and the BRIEF(Binary Robust Independent Elementary Features)descriptor was selected in the stereo matching,in which(a)a multiple-BRIEF-descriptor stereo matching algorithm was proposed,which takes advantage of the different features represented by different BRIEF descriptors,with randomness in generation,to effectively eliminate mismatches;(b)the epipolar constraint was used to improve the matching accuracy and to reduce the matching time;(c)disparity filtering and statistical filtering were employed to further eliminate mismatches.Compared with other matching methods with different descriptors,such as SURF(Speeded Up Robust Feature),ORB(Oriented FAST and Rotated BRIEF),the proposed multiple-BRIEF-descriptor stereo matching method is with a higher matching accuracy,more edge information,a better reconstruction result,but consumes a slightly more time than the method with the descriptor of SURF.As per the structural characteristics of fillet welds,cluster analysis and SVD(Singular Value Decomposition)plane fitting were proposed to classify the point cloud to two sets,and to obtain two plane equations for the two surfaces of the weldment by fitting the two sets of the point cloud.The intersection of the two planes was recognized as the weld seam.The intersections of fitted edges and the weld seam were solved to obtain the start and end positions of the weld seam.The positioning errors of the start and end positions were summarized and quantitatively analyzed.The average errors in the x,y,and z directions were 0.5 mm,0.8 mm,and 0.9 mm,respectively,which were all within 1 mm.The errors could basically meet the expectations.With the analysis of the test results,the factors that affect the accuracy of the proposed method were summarized,including the matching accuracy of feature points,the completeness of the detected edges,the straightness of edges and background interferences,etc.Facing the complexity and diversity of the welding environment,welding conditions with the changes in background complexity,materials,orientation angles,etc.were designed to test the robustness of the proposed recognition and positioning method for fillet welds.The experimental results showed that the proposed method is applicable for most conditions.更多还原