CASIA OpenIR  > 复杂系统管理与控制国家重点实验室  > 先进机器人
基于DLP投影的焊接机器人焊缝识别和轨迹检测的技术研究
杨磊
Subtype博士
Thesis Advisor梁自泽
2019-05-22
Degree Grantor中国科学院大学
Place of Conferral中国科学院大学
Degree Discipline控制理论和控制工程
Keyword焊接机器人 视觉测量 结构光视觉 焊缝识别 点云分割 轨迹检测 路径规划 位姿模型
Abstract

焊接技术是制造业中极其重要的加工手段,而焊接机器人在现代制造业中应用越来越广泛。目前,“示教-再现”模式和离线编程是工业机器人主要的两种操作模式。“示教-再现”模式能够适应复杂的工艺过程和不同的焊接对象,是焊接机器人最为常用的编程方法,但对于复杂轨迹的三维焊缝而言,示教过程存在费时、费力、示教精度差、可重复性差等缺点,难以满足快速、精准、柔性的新型制造模式的需求。离线编程模式借助CAD技术能够实现工业机器人的路径示教,在结构化的工业生产场景中能够提高示教效率,但在非结构化的作业环境中,离线编程模式难以实现虚拟环境的准确实时构建,无法适应动态多样性的焊接需求。因此,研究自主示教编程模式,提高三维路径示教的效率和精度是目前智能化焊接机器人的主要研究方向,其中,焊接路径的自主检测是实现焊接机器人自主示教模式的首要前提。本文对焊接机器人焊接路径的焊缝类型识别、焊缝特征提取、焊缝路径检测、焊缝精确定位等关键技术开展研究,论文的主要工作如下:

1.  针对焊接机器人的三维测量需求,设计了基于数字光处理(Digital Light ProcessingDLP)投影仪光栅投影系统,其作为焊接机器人的主动视觉测量系统;通过DLP投影仪的图案编程产生不同类型的编码图案,结合立体视觉系统,构造能够适应不同工况的视觉测量系统,满足焊接机器人在不同环境下三维测量的需求;对于视觉系统的各个关键单元进行了布局优化配置及部件选型,使之适应工业机器人末端有限的安装空间。

2. 针对弱对比度环境下不同类型的平面对接窄焊缝图像的检测识别问题,设计了一种焊缝特征提取算子来替代传统的边沿检测算子,通过对整幅焊缝图像的扫描,实现焊缝中心位置的提取;在此基础上,设计了基于焊缝形状的特征向量,实现不同焊缝图像的特征提取,并提出了一种基于自适应神经模糊系统的焊缝类型检测方法,实现不同类型焊缝的在线检测和识别。

3. 针对中厚板焊缝特征的鲁棒、准确提取问题,基于焊缝的三维特征,提出了一种基于点云分割的焊缝特征提取算法,通过对焊接工件的三维重建、点云滤波、点云分割、特征提取等步骤,克服了弱纹理、弱对比度、金属表面的反射、工件表面上的缺陷,如锈蚀、磨屑和划痕等因素的影响,实现不同类型中厚板焊缝特征信息的自动提取。

4. 针对三维空间复杂焊缝的位置姿态模型的描述问题,提出了一种基于三次平滑样条的路径拟合方法,通过粗糙度罚函数的引入来消除待拟合数据中的噪声影响,保证焊缝路径的平滑和连续;通过对焊缝的三维点云数据的处理,在各焊缝路径点处建立离散的焊缝动坐标系,构建用以描述焊缝的完整姿态模型。

5. 针对焊接机器人系统中工件焊缝的高精度三维测量问题,基于由粗到精的规划思想,提出了一种“粗提取-细定位”相结合的焊缝特征提取思想,利用DLP投影仪分别产生格雷码图案和激光条纹图案,构造焊接机器人的全局视觉和局部视觉,通过对焊缝位置信息的粗提取和精确扫描定位,实现多种类型焊缝信息的精确提取。同时,提出了一种基于核线性滤波算法的焊缝特征提取算法,基于焊缝粗提取的结果,通过对焊缝路径的近距离扫描,来对焊缝信息进行精确定位,并修正焊缝的数学模型,提高焊缝的提取精度。

6. 总结本文的研究工作,并提出了下一步的研究计划。

Other Abstract

Welding technology is an extremely important processing means of industry manufacturing, and the welding robots are widely used in the modern manufacturing industry. At present, the "teaching-playback" mode and the off-line programming mode are two main operation modes of industrial robots. The "teaching-playback" mode is able to adapt to complex technological process and different welding objects. It is the most commonly used programming method for welding robots. However, for the 3D weld seam with complex trajectories, the teaching process exists some shortcomings, such as time-consuming, laborious, poor teaching precision and poor repeatability, so it is difficult to meet rapid, precision and flexible new manufacturing mode. The off-line programming mode can realize 3D path teaching of industrial robots with CAD technology. It can improve teaching efficiency in the structured industrial production environment. However, in the un-structured working environment, the off-line programming mode is difficult to achieve accurate and real-time reconstruction of virtual working environment, and cannot meet the dynamic, complex, multifarious welding requirements. Therefore, studying the automatic teaching mode and improving the efficiency and precision of 3D path teaching are the main research directions of the intelligent welding robots. The automatic identification and detection of weld seam are the pre-requisites for realizing the autonomous 3D path teaching of welding robots. In this paper, the key technologies of the autonomous path teaching of welding robots are studied, including seam type recognition, seam extraction, 3D seam path planning and precise positioning of weld seam. The paper’s main work is as follows:

Firstly, aiming at the 3D measurement requirements of welding robot, the grating projection system based on DLP projector is designed, which is used as the active vision measurement system of welding robots. Through the pattern programming of DLP projector, different types of coding patterns could generated. Combined with stereo vision system, an active vision measuring system is constructed to adapt to different working conditions and meet the 3D measurement requirements of welding robots in different environments. And layout optimization and component selection for each key unit of the designed vision system are done to adapt to the limited installation space of the end of the industrial robots.

Secondly, aiming at the detection and recognition problem of different types of narrow butt joints under poor contrast, an operator of feature extraction is designed to replace the traditional edge detection operators and realize seam extraction under poor contrast. By scanning the whole seam image, the location of seam center is extracted. On this basis, the feature vectors based on weld shape are designed to realize the feature extraction of different seam images. And a seam type detection method based on ANFIS model is proposed to realize the on-line detection and recognition of different weld seams.

Thirdly, aiming at the problem of robust and accurate feature extraction of different thick welds, based on 3D features of different weld seams, a 3D seam extraction algorithm based on point cloud segmentation is proposed. Through 3D reconstruction of welding workpieces, point cloud filtering, point cloud segmentation and feature extraction, the proposed algorithm could overcome the influence of weak texture, poor contrast, reflections from metallic surfaces, and imperfections on the work piece such as rust, mill scale and scratches which are not consistent from part to part and realize automatic 3D seam extraction of different types of thick plates.

Fourthly, aiming at the description problem of the position model and orientation model of the complex weld seam in 3D space, a path fitting method based on the cubic smoothing spline is proposed. The noise of the data to be fitted could be eliminated by introducing the roughness penalty function to ensure the smoothness and continuity of the weld seam path. A discrete dynamic coordinate system is established at each weld path point by processing the 3D point cloud data of the welding workpiece to construct a complete orientation model to describe different weld seams.

Fifthly, aiming at the problem of high-precision 3D measurement requirement of weld seams in welding robot system, based on the idea of coarse to fine planning, a weld seam extraction framework based on "coarse-fine " idea is proposed. Gray code patterns and laser stripe patterns are generated respectively by pattern programming of DLP projector, and the global vision and local vision of welding robot are constructed respectively. Through the coarse extraction and fine positioning of weld seam, the precision seam extraction of various types of weld seam can be achieved. Meanwhile, in order to achieve fast and precision extraction of various types of weld seam, a feature extraction algorithm based on kernel linear filtering algorithm is proposed. Based on the results of coarse seam extraction, by close-range scanning of the weld seam path, the seam path could be precisely positioned, and the mathematical model of the weld seam is corrected to improve the precision of seam extraction.

Finally, the research work of this paper is summarized, and the future research plans are proposed.

Pages152
Language中文
Document Type学位论文
Identifierhttp://ir.ia.ac.cn/handle/173211/23819
Collection复杂系统管理与控制国家重点实验室_先进机器人
Corresponding Author杨磊
Recommended Citation
GB/T 7714
杨磊. 基于DLP投影的焊接机器人焊缝识别和轨迹检测的技术研究[D]. 中国科学院大学. 中国科学院大学,2019.
Files in This Item:
File Name/Size DocType Version Access License
基于DLP投影的焊接机器人焊缝识别和轨迹(9406KB)学位论文 开放获取CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[杨磊]'s Articles
Baidu academic
Similar articles in Baidu academic
[杨磊]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[杨磊]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.