基于机器视觉的焊接工件识别与焊接轨迹校正方法研究

发布时间：2018-05-24 10:06

本文选题：焊接机器人 + 机器视觉　；参考：《华南理工大学》2015年硕士论文

【摘要】：由于人工焊接存在工作环境恶劣、劳动强度大、效率低、焊接质量得不到保证等问题,当前焊接机器人已经在许多工业领域得到了应用。然而焊接机器人一般采用示教再现方式工作,为确保这种工作方式能在具体焊接环境中实施,需要解决两个关键问题:第一正确识别焊接工件以确定示教程序。第二对前工序中由人工点焊定位导致的定位误差进行自动补偿。这两个问题是焊接机器人应用的突出技术难点,成为制约焊接机器人技术推广应用的瓶颈。为此,本课题采用机器视觉技术,以解决焊接工件识别和焊接轨迹自动校正的问题。本课题的研究获得了广东省科技计划项目(编号:2011A091101001,工业机器人核心技术研究及典型产品产业化)和企业横向项目(集装箱后端生产线全自动装配和焊接机器人应用,南方中集东部物流装备制造有限公司)的资助。本文研究了手眼系统的标定方法。针对Eye-to-Hand和Eye-in-Hand两种不同的焊接机器人和视觉系统安装方式,分别采用不同的手眼系统标定方法。获得摄像机的内外参数,计算摄像机坐标系、机器人末端坐标系及世界坐标系之间的转换矩阵,从而实现图像坐标系和世界坐标系的转换。本文研究了图像预处理技术,包括灰度转换、滤波去噪、阈值处理、形态学运算及边缘检测。通过对图像预处理,滤去噪声,增强目标信息,使其具有一定的鲁棒性。在此基础上,对焊接工件的特征进行了分析和提取,结合这些特征,设计和训练了高斯混合模型分类器、多层感知神经网络分类器及支持向量机分类器,并对特征向量和分类器参数进行了优化,最终确定最优的特征向量和分类器参数,以实现焊接工件的准确识别。本文在传统模板匹配技术基础上提出了基于几何形状的金字塔分层匹配算法,提取图像几何特征,并对几何特征进行了分层。计算灰度区域的图像质心,对焊接工件进行定位。以像素点的平移矩阵和旋转矩阵为基础,根据模板匹配检测到的偏移量和旋转角度,计算出实际轨迹,从而校正焊接轨迹。本文设计了基于机器视觉的焊接机器人实验平台,并进行了焊接工件的分类、检测、识别定位和焊接轨迹校正实验。实验结果表明上述的理论和算法都能满足焊接机器人对时间和精度的要求。本课题研究的成果目前已用于南方中集东部物流装备制造有限公司的集装箱后端锁座和铰链的焊接,不仅焊缝质量良好,而且能满足焊接生产线对时序的要求。
[Abstract]:Due to the problems of poor working environment, high labor intensity, low efficiency and unguaranteed welding quality in manual welding, welding robots have been applied in many industrial fields. However, welding robots generally work in teaching and reproducing mode. In order to ensure that the working mode can be implemented in a specific welding environment, two key problems need to be solved: first, the welding workpiece is correctly identified to determine the teaching procedure. Second, the positioning error caused by manual spot welding in the former procedure is automatically compensated. These two problems are the prominent technical difficulties in the application of welding robot and become the bottleneck restricting the application of welding robot technology. Therefore, machine vision technology is used to solve the problem of welding workpiece identification and automatic correction of welding track. The research of this subject has obtained the project of Guangdong province science and technology plan (number: 2011A091101001, industrial robot core technology research and typical product industrialization) and enterprise horizontal project (automatic assembly and welding robot application of container back-end production line), Southern Zhongji East Logistics equipment Manufacturing Co., Ltd. The calibration method of hand-eye system is studied in this paper. Aiming at two different installation modes of welding robot and vision system, Eye-to-Hand and Eye-in-Hand, different calibration methods of hand-eye system are adopted. The internal and external parameters of the camera are obtained, and the transformation matrix between the camera coordinate system, the robot terminal coordinate system and the world coordinate system is calculated, and the transformation between the image coordinate system and the world coordinate system is realized. In this paper, image preprocessing techniques including gray conversion, filtering and denoising, threshold processing, morphological operation and edge detection are studied. By image preprocessing, noise is filtered, and target information is enhanced to make it robust. On this basis, the features of welded workpieces are analyzed and extracted. Combined with these features, Gao Si hybrid model classifier, multi-layer perceptual neural network classifier and support vector machine classifier are designed and trained. The eigenvector and classifier parameters are optimized and the optimal eigenvector and classifier parameters are finally determined in order to realize the accurate identification of the welded workpiece. In this paper, based on the traditional template matching technology, a pyramid hierarchical matching algorithm based on geometric shape is proposed, which extracts the geometric features of images and delaminate the geometric features. The image centroid of gray area is calculated and the welding workpiece is located. Based on the translation matrix and rotation matrix of pixels, the actual track is calculated according to the offset and rotation angle detected by template matching, and the welding trajectory is corrected. In this paper, a welding robot experimental platform based on machine vision is designed, and the experiments of welding workpiece classification, detection, identification, location and welding trajectory correction are carried out. The experimental results show that the above theory and algorithm can meet the requirements of time and precision of welding robot. The research results of this paper have been applied to the welding of container rear end locking seat and hinge in South Zhongji East Logistics equipment Manufacturing Co., Ltd., which not only have good weld quality, but also meet the requirements of welding production line.
【学位授予单位】：华南理工大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP242;TP391.41

【参考文献】