基于线结构光视觉的焊缝余高和熔宽检测

发布时间：2018-06-01 19:46

本文选题：线结构光 + RANSAC算法　；参考：《广东工业大学》2017年硕士论文

【摘要】：目前,对于焊缝的抛光打磨,工业现场主要依靠人工通过肉眼观察焊缝外观以及使用千叶片或者砂轮等来进行打磨作业。这种作业方式不仅效率低下,且其抛磨精度和稳定性比较差。机器视觉技术的发展为解决焊缝抛磨提供了一种新的思路,尤其是线结构光视觉在焊缝检测中的应用,能够很好的解决上述问题。在焊缝抛磨中最重要的参数即余高以及熔宽,因此本文对焊缝检测所涉及到的关键技术进行了初步研究。本文首先从系统层面介绍了整个线结构光视觉系统以及其原理。根据实际工况需求,对线结构光光路进行了设计并对CMOS相机进行选型;考虑到焊缝材料特征,选择了650nm波长的激光发生器,采用了窄带滤光片来滤除环境光干扰;通过以上视觉元件的结构,结合机器人末端尺寸,对视觉传感器夹具进行了设计,并最终搭建了焊缝余高和熔宽检测实验系统。为了建立相机中像素坐标与三维空间的关系,对视觉系统进行了标定。首先建立了相机的数学模型,并利用张氏标定法对相机进行了标定,获得了相机的内参;利用平面棋盘作为标定参照物;对线结构光进行了标定,并得到了线结构光平面方程;在上述内参模型的基础上利用两步法对机器人手眼关系进行了标定,并得到了手眼关系矩阵。应用随机抽样一致性(RANSAC)算法对焊缝图像进行了处理,提取出焊缝余高以及熔宽信息。首先对采集到的焊缝图像进行预处理,得到二值化图像;通过边缘提取得到激光条纹边缘图像,采用改进了的平均法得到单像素激光中心线;通过RANSAC算法可得到其数学模型,并动态设定感兴趣区域(ROI);在此ROI中,再次进行预处理以及RANSAC算法处理,可得到只包含焊缝余高信息的单像素激光中心线,从而得到焊缝特征点及余高和熔宽信息;为实现机器人抛磨,定义了机器人抛磨位姿。为实现视觉信息与机器人的交互,基于MFC以及开源算法库OpenCV开发了焊缝余高和熔宽检测系统软件。软件系统采用模块式开发的方法,分别对图像采集、图像处理、焊缝数据处理、视觉系统标定以及可视化等功能模块进行了开发;为验证系统的可行性,最后进行了焊缝检测实验。
[Abstract]:At present, for the polishing and grinding of weld seam, the industrial field mainly relies on manual observation of weld appearance with naked eye and the use of thousands of blades or grinding wheels for grinding. This operation is not only inefficient, but also its grinding accuracy and stability is poor. The development of machine vision technology provides a new way to solve weld grinding, especially the application of line structured light vision in weld seam detection, which can solve the above problems well. The most important parameters in weld polishing are residual height and weld width. Therefore, the key technologies involved in weld inspection are studied in this paper. This paper first introduces the whole line structured light vision system and its principle from the system level. According to the actual working conditions, the optical path of the line structure is designed and the CMOS camera is selected, considering the characteristics of the weld material, the laser generator of 650nm wavelength is selected and the narrow band filter is used to filter the environmental light interference. According to the structure of the vision component and the size of the robot, the fixture of the vision sensor is designed, and the experiment system of weld residual height and weld width detection is built. In order to establish the relationship between pixel coordinates and 3D space, the vision system is calibrated. Firstly, the mathematical model of the camera is established, and the camera is calibrated by the method of Zhang's calibration, and the inner parameters of the camera are obtained, the plane chessboard is used as the calibration reference, the linear structured light is calibrated, and the plane equation of the linear structured light is obtained. Based on the above model, a two-step method is used to calibrate the hand-eye relationship of the robot, and the hand-eye relation matrix is obtained. The random sampling consistency algorithm (RANSAC) is used to process the weld image and extract the residual height and weld width information. Firstly, we preprocess the weld image to get binary image; get the edge image of laser stripe by edge extraction, get the laser center line of single pixel by the improved average method; get the mathematical model by RANSAC algorithm. In this ROI, pretreatment and RANSAC algorithm are used again to get the single pixel laser centerline which only contains the information of residual height of the weld, thus obtaining the characteristic point of the weld and the information of the residual height and the width of the weld. In order to realize robot grinding, the position of robot grinding is defined. In order to realize the interaction between visual information and robot, the software of weld residual height and weld width detection system is developed based on MFC and open source algorithm library OpenCV. In order to verify the feasibility of the system, the function modules of image acquisition, image processing, weld data processing, visual system calibration and visualization are developed. Finally, the weld test was carried out.
【学位授予单位】：广东工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TG441.7;TP391.41

【参考文献】