Robotic grinding of welds on freeform surfaces poses an increasing challenge to automatic generation of grinding trajectory while conventional teaching-playback mode and off-line programming method are ineffective. This paper proposes a novel feature-guided trajectory generation method based on point cloud data to perform an efficient grinding process for welds on a freeform surface. The 3D contour of the workpiece was measured by a laser profile scanner. Parent curve of each scanning line was fitted by means of moving average filter, and then, the weld feature points were reliably extracted out of the scattered point cloud through two stages of feature recognition. To achieve the movement guidance of the manipulator, B-spline fitting method was conducted to generate a smooth 3D curve which was discretized into actual tool contact points by an optimized interpolation algorithm and computed the tool postures by cross multiply algorithm. By using robotic force control, the desired force was planned for every tool contact point in order to compensate the error of the processing path. Verification shows that the maximum root mean square root error of recognition of the proposed algorithm is less than 0.7 mm and the computational time is saved by 65.12% in comparison with the reverse engineering method.

在传统的示教回放模式和离线编程方法无效的情况下，对自由曲面上的焊缝进行自动打磨对自动生成打磨轨迹提出了越来越高的挑战。本文提出了一种基于点云数据的新型特征导向轨迹生成方法，以对自由曲面上的焊缝执行高效的磨削过程。工件的3D轮廓通过激光轮廓扫描仪测量。通过移动平均滤波器拟合每条扫描线的父曲线，然后通过两个阶段的特征识别将焊接特征点可靠地从散射点云中提取出来。为了实现机械手的运动指导，进行了B样条拟合，生成了一条平滑的3D曲线，通过优化的插值算法将其离散化为实际的刀具接触点，并通过交叉乘法算法计算了刀具姿态。通过使用机械手力控制，为每个工具接触点计划了所需的力，以补偿加工路径的误差。验证表明，与逆向工程方法相比，该算法识别的最大均方根误差小于0.7 mm，计算时间节省了65.12％。