3D edge detection aimed to the automatic programming of welding robots
Miguel Angel Villanueva Portela-CA1; Ricardo Emiro Ramirez Heredia1
1 Universidad Nacional de Colombia, Cra 30 #45-03, Edificio 407 oficina 201. Bogotá,11001000, Colombia
doi:10.20906/CPS/COB-2015-0523
Resumo
Programming a welding robot is a challenging and time-consuming task, different kinds of tools have been developed to facilitate this process. One of these new methods is called intuitive robotic welding programming, in this method the operator hand-guides the robot to the desired welding positions. Other methods use virtual user interfaces to do the same, teach the desired welding positions. This research aims to facilitate the programming process related to welding robots. In this stage, it is carried out through the automated record of position and orientation of straight edges belonging to an assembly that are going to be welded. This process is developed around the processing of a 3D point cloud. Based on the Random Sample Consensus (RANSAC) algorithm, statistical outliers removal and pass through filters and the function planeWithPlaneIntersection of the point cloud library (PCL). An algorithm to process a 3D point cloud was development. This processing algorithm takes as an input the 3D point cloud captured from the working assembly to identify and record the straight edges that will be welded. The target edges reside on T joins with square groove and/or in inside corner join with no preparation. The information from the straight edges is going to be implemented in the next stage of this research as an input of the path and trajectory planner. The proposed method in this work is able to detect straight edges of a working assembles which contain T joins with square groove and/or in inside corner join with no preparation that will be welded. The proposed algorithm was tested on metallic and non-metallic mock-ups. The process relies in artificial vision 3D sensor, in this case a Microsoft Kinect was implemented. The decision of employing mock-ups rather than real working assembles was made due to the low resolution the Kinect sensor has. The benefits of the proposed system span from the easy of use to the quick recognition of the geometric features required in the path and trajectory planning process r
Palavras-chave: 3D Vision; Welding Robot; Point Cloud Processing; Geometric Features identification