From Model to Reality: A Robust Framework for Automatic Generation of Welding Paths

Current programming methods for welding robots mainly rely on manual teaching or offline programming, making it difficult to adapt to the flexible production mode of small batches and multiple categories. To this end, a robotic welding path automatic generation framework is proposed in this article. The framework performs nonrigid registration between point clouds sampled from computer-aided design (CAD) models of workpieces with point clouds captured by self-designed hybrid vision sensors. By doing so, the welding paths extracted from CAD models are transformed into actual welding paths. In addition, the WeldNet network is proposed to automatically identify weld types and key points, and the interested welding area is automatically extracted based on the point cloud segmentation network PointROINet. Combined with the coded structured light vision model, the 3-D coordinates of weld key points are obtained, thereby enabling fast and accurate registration of weld point clouds. Experimental results demonstrate that the proposed framework can efficiently and robustly generate welding paths for spatial curve butt welds, lap welds, and fillet welds before welding. https://github.com/Yunkai-Ma/PathGeneration