Autonomous Location of the Welding Workspace in the Box Girder

Abstract

Automatic welding of box girders can greatly increase production efficiency. To locate the welding workspace, it is necessary to reconstruct a map of the box girder and extract features from it. In this paper, we propose a robust LiDAR-inertial odometry (LIO) to build the map and a feature extraction method to locate the webs. For the LIO, we implement map loading and map update function based on FAST-LIO2. The map loading function sets prior values for map building, which will improve its accuracy and matching success rate. And we propose a new data structure, Mark-ikd-Tree, to update the map, which can ensure the freshness of it. For the point cloud feature extraction method, we propose a density-based planar extraction method for the webs in box girders. This method can detect the webs quickly, accurately and without missing. Finally, after obtaining the coordinates of the webs, we estimate the welding workspace by the relative position relationship of the webs and the weld. We validate our method in the experiment section. The positioning error of the webs is around 2 cm.