机器视觉激光焊接缺陷检测算法研究

发布时间：2017-12-27 02:30

本文关键词：机器视觉激光焊接缺陷检测算法研究　出处：《深圳大学》2017年硕士论文　论文类型：学位论文

【摘要】：机器视觉激光焊接具有焊接质量高、精度高、速度快等优点,实现了机械自动化,现已经被广泛应用到各行各业。如汽车行业对零部件采用激光焊接降低了车身的重量和生产成本、制船行业对船板采用激光焊接在焊后基本没有变形等。随着机器视觉激光焊接的发展,通过机器视觉实现对焊后焊件表面的自动化缺陷检测也变得尤为重要。机器视觉缺陷检测系统一般都具有快速性和实时性,但是在不牺牲速度的前提下提高准确性一直是攻克的难点。这里通过引入Hough变换检测圆(CHT,Circular Hough Transform)做缺陷检测,其主要是定位缺陷所在位置,通过引入Hough变换使得产品缺陷检测的准确性和快速性得到了很大的提高。本课题来源于某公司“基于机器视觉的激光自动焊接设备”项目。本文主要做了如下工作:1)将CHT应用到基于机器视觉的激光焊接焊后焊件表面缺陷检测领域中。对工业相机(CCD,Charge-coupled Device)采集到的图像先用Canny算子进行边缘检测,然后用CHT做焊接位置定位,最后累计焊接位置处坏的像素点的个数,并与设定的缺陷阈值进行比较由此判断产品焊接质量的好坏。本算法与用模版匹配做缺陷检测的算法进行对比分析,误判率和漏判率分别降低了5%和3%左右。2)将改进随机Hough变换(RHT,Random Hough Transform)应用到基于机器视觉的激光焊接焊后焊件表面缺陷检测领域中。RHT检测圆与CHT相比速度得到了提高,但是由此会引入大量的无效累积,因此对RHT进行如下改进:判断随机采样的三点是否共线并对其之间的距离进行限定,这样避免了之后大量的无效计算;当判断候选圆是否为真实圆时,先判断边缘点是否在候选圆的内外切正方形之间然后再计算边缘点是否在候选圆上,避免了大量计算;采用求弦的中垂线对圆参数求解,提高了运行速度。该算法先在matlab上仿真运行然后被应用到机器设备上,验证了其准确性与有效性。改进RHT算法的准确性与RHT、CHT的准确性基本相同,改进RHT相比RHT算法速度提高了100ms左右,改进RHT算法相比CHT速度提高了600ms左右。
[Abstract]:Machine vision laser welding has many advantages, such as high quality, high precision, fast speed and so on. It has realized mechanical automation and has been widely used in all walks of life. Such as the automotive industry parts by laser welding reduces the body weight and production cost of the ship to ship industry, using laser welding after welding without deformation. With the development of machine vision laser welding, it is also very important to realize automatic defect detection on the surface of welding parts by machine vision. The machine vision defect detection system generally has fast and real time, but it is difficult to improve the accuracy without sacrificing speed. Here we introduce the Hough transformation to detect the circle (CHT, Circular Hough Transform) to do defect detection. It mainly locate the location of the defect. By introducing the Hough transform, the accuracy and rapidity of product defect detection have been greatly improved. This project comes from a company "laser automatic welding equipment based on machine vision". The main work of this paper is as follows: 1) the application of CHT to the surface defect detection field of laser welded parts of laser welding based on machine vision. The industrial camera (CCD, Charge-coupled Device) to the first image edge detection using Canny operator, and then use CHT to do the welding position, welding end cumulative number of pixels at the position of the bad, and compared the judgment of product welding quality and defect threshold setting. The algorithm is compared with the template matching algorithm for defect detection, and the error rate and the missed rate are reduced by 5% and 3% respectively. 2) the improved random Hough transform (RHT, Random Hough Transform) is applied to the surface defect detection field of laser welded parts of laser welding based on machine vision. RHT circle detection compared with CHT speed has been improved, but this will introduce invalid accumulated, thus improving the RHT as follows: three to determine the point of random sampling are collinear and the distance between the limit, so as to avoid the invalid after a lot of calculation; when judging the candidate circle for true circle. First determine whether the candidate edge points in the circle tangent square between inside and outside and then calculate the edge point in the candidate circle, to avoid a lot of calculation; using the string perpendicular to solve circle parameters, improves the running speed. The algorithm is first simulated on the MATLAB and then applied to the machine equipment to verify its accuracy and effectiveness. The accuracy of the improved RHT algorithm is basically the same as that of RHT and CHT. Compared with the RHT algorithm, the speed of the improved RHT algorithm is increased by about 100ms. Compared with the improved RHT algorithm, CHT speed has increased 600ms.
【学位授予单位】：深圳大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TG441.7;TP391.41

【参考文献】