Machine Learning for Robot-Assisted Industrial Automation of Aerospace Applications

摘要

In this paper, a multiple circular contours extraction method is applied to estimate the geometric primitives of multiple circles in the three-dimensional space for robot-assisted industrial automation for manufacturing processes. The primary objective is to establish an accurate reference frame which is a major requirement in the robot-assisted riveting process for floating aircraft fasteners used in aerospace structures assembly. The reference frame that constitutes the positions, the diameters, and orientations of the fastening holes is used to minimize the positioning and alignment errors between the floating fasteners and the workpiece. The method can be divided into the following steps. Firstly, a Maximum Likelihood Estimation Sample Consensus (MLESAC) method is used to fit the plane on the point cloud and to classify the data into inliers (coplanar points) and outliers (noisy points due to strong reflections of laser on shiny metallic surfaces). Secondly, after downsampling the inliers, the data are rotated using the Rodrigues formula such that the normal direction of the estimated plane and the z-axis direction are parallel. Thirdly, the Delaunay triangulation is constructed on the rotated inliers and a confidence interval is estimated to classify the points that are located at the circular boundaries of the holes from the inliers. Fourthly, a hierarchical clustering approach is applied to partition the classified point cloud into three data sets belonging to one major hole and two minor holes. Finally, the convex hull is constructed on the clustered data sets and three circular profiles are fitted. The method is applied on a noisy experimental data and the repeatability of the outputs is discussed thoroughly. In our evaluation, the point cloud is acquired by a laser stripe sensor placed on a liner rail, which is attached on an end effector of a 6 Degrees-of-Freedom (DoF) KUKA robotic arm. The method is used to successfully automate the riveting of the fastener components on an aerospace structure.