毕业生知识库

      复杂环境下的自主抓取作业是工业机器人智能化的一个研究热点。目前，智能化的工业机器人应能够感知和理解外部环境，在复杂的工业环境中自主智能地进行操作规划，完成抓取任务。而铝电解槽大修主要采用人工方式将金属工件搬运到阴极钢棒和爆炸钢片之间，再用焊接使其连接起来，焊接环境恶劣、工作强度大、效率低，并且难以保证焊接质量和一致性。因此，在铝电解槽环境中，设计一种代替人工进行工件的抓取并完成焊接的机器人成为亟待解决的问题。铝电解槽为强磁环境，焊接连接工件为不同尺寸、不精确摆放、相互遮挡以及存在反光干扰的平面型金属工件，且铝电解槽工位内有多种障碍物，爆炸钢片在焊接过程中受到高温影响发生动态变化，不同工位外侧障碍物存在差异。这些难点对铝电解槽机器人的抓取作业提出了挑战。针对以上问题，本文对铝电解槽机器人视觉自主抓取作业开展研究，并给出了解决方案，论文的主要研究工作总结如下：

      1. 根据铝电解槽大修抓取、焊接作业要求和铝电解槽狭窄环境的特点，设计了铝电解槽双臂焊接机器人；基于分层递阶控制思想和模块化方法设计了控制系统，实现了对机器人的高精度、稳定控制。

      2. 提出了一种基于单目结构光视觉的简单背景下凸多边形金属工件的抓取位姿检测方法，针对金属工件表面存在反光、划痕、锈蚀的现象并且金属工件与背景的灰度值差别明显的情况，采用自适应灰度差法，通过边缘像素点扫描、噪声初步去除、直线拟合，实现了光照变化下的凸多边形金属工件边缘检测，结合结构光视觉系统实现了平面型工件深度信息的测量和最大拟合平面的计算，解决了铝电解槽环境中凸多边形金属工件的视觉定位与测量问题。

      3. 提出了一种基于先验知识与图割法相结合的铝电解槽环境中平面型金属工件的图像分割方法。铝电解槽中焊接工件均为平面型工件，根据不同的工位环境采用不同形状的焊接工件。针对平面型金属工件在铝电解槽复杂环境中的不同情况，采用单一形状、非线性形状、模糊连接度等先验知识，以图割法作为图像分割的框架，实现了复杂环境中不同尺寸、不精确摆放、相互遮挡以及反光的多种金属目标工件的图像分割。

      4. 提出了一种基于概率的机械臂避障路径规划方法。针对铝电解槽环境中障碍物复杂的现状，采用基于图像的视觉控制方法趋近抓取目标工件，通过铝电

解槽作业环境的三维建模和基于线段相交测试的碰撞检测，提出了一种基于概率的障碍物空间约束下的趋向目标节点生成算法，结合二次Bezier曲线轨迹插补算法，完成了狭窄环境中机械臂避障路径的规划，实现了工件的平稳、高精度搬运。目标工件的趋近抓取实验及抓取作业路径规划实验验证了算法的有效性。

       Autonomous intelligent grasping in cluttered scene is one of the key research areas for industry intelligence. Intelligent robot requires the ability to sense and understand the environment, plan a series of manipulation in complicated environment and accomplish the grasping task. The method for aluminium electrolytic cells’ repairment is mainly to carry the workpiece to connect the cathode steel bar and the explosive steel sheet, then to weld the seam between them by human beings. The environment in aluminium electrolytic cells is terrible, the work intensity is high, and it is difficult to ensure the welding quality and consistency. Thus, there is an urgent need for a special robot working in aluminium electrolytic cells to carry the workpieces to the destination and weld the seam. The environment of aluminium electrolytic cells is magnetically strong. The welding workpieces are various in shape, imprecisely placed, occluded by each other and reflective because they are the planar metal workpieces. There are many obstacles in the aluminium electrolytic cells, and the inclination for the surface of the explosive steel sheet changes dynamically because of the high temperature in the process of welding. There are also some small differences in different cells. All these difficulties are great challenges for the grasping in the aluminium electrolytic cells. According the problems mentioned above, the research of autonomous grasping based on vision in aluminium electrolytic cells’ repairment is conducted in this paper, and the methods are proposed. The main content of this paper is in the following.

      Firstly, based on the characteristics of the transportation and welding in the aluminium electrolytic cells’ repairment and their narrow space, the dual arm welding robot is designed and the control system which is based on hierarchical control and modularity method is planned to ensure a high accurate and stable robot system.

      Secondly, the inspect method of the grasping posture measurement for the metal polygon workpieces in simple environment is discussed. There exists reflection, scratch and rust on the surface of the metal workpiece, and the gray difference between the metal workpiece and the background is obvious, thus the self-adaptive gray difference method is used to detect the edge of the workpiece by the edge pixels’ scan, noise reduction and line fitting no matter of the light variation on the surface of the metal

workpiece. Meanwhile, the structured light vision system is used to measure the deep information of the workpiece and fit the maximum plane of the planar workpiece, and the vision localization and measurement for the convex polygon metal workpiece in the field of the aluminium electrolytic cells’ repairment is solved.

      Thirdly, the image segmentation for metal workpieces in cluttered scene based on prior knowledge is proposed. The workpieces for the repairment of the aluminium electrolytic cells are various planar workpieces according to different situations. For different conditions of the cluttered scene, such as the reflection, the occlusion and complicated background, the shape prior, non-linear shape prior and fuzzy connectedness prior knowledge are applied to solve the image segmentation combined with the framework of GraphCuts method for the workpieces with different sizes, imprecisely placed and occlusion.

      Fourthly, the collision avoidance path planning in narrow space is designed. The target workpiece is grasped by approaching it according to the image-based visual control. According to the cluttered scene in the aluminium electrolytic cells, the 3D environment model is constructed by the structured light vision system and the collision checking is executed by line segments' intersection detection. The probabilistic obstacle-constraint reaching-target path nodes generation method is proposed combined with the second-order Bezier curve interpolation method to accomplish the obstacle avoidance path planning, and reach the goal of grasping and carrying the workpiece stably and accurately. The experiments of approaching grasping and transportation of the workpieces validate the effectiveness of the obstacle avoidance path planning algorithm in the aluminium electrolytic cells.