Method for 3D measurement and evaluation of joint geometries for adaptive robotic arc welding in the automotive industry

Arc welding is one of the main manufacturing processes in the production of body in white structures in the automotive industry. Typically, robots are used to perform the welding process automatically, as shown in Figure 1. However, a main drawback still is the need for precise fixturing of workpieces to enable a correct process execution with the once generated robot program [1]. Moreover, there might be geometric deviations between the manufactured parts to be welded due to changing conditions at the suppliers productions or the usage of different production technologies like casting or pultrusion. Therefore, the single parts of a welding assembly in the body in white production have to be manufactured with tight tolerances to enable reproducible joint geometries and constant welding process results. To overcome these drawbacks, some approaches focus on seam tracking based on 2D laser sensors for real-time program adaptation [2], [3]. However, the adaptation strategy is based on heuristics and only works in a specific process window for small geometric part deviations and for relatively smooth surface contours. Hence, the sensor-based adaptation has to be reconfigured manually for each new workpiece type, which is a time-consuming task. Moreover, fixtures prevent visibility during the process execution making this approach unfeasible in many welding applications. Other approaches use computer vision to reconstruct the real workpiece geometry and automatically generate robot programs for each new part [4], [5]. Geometry reconstruction however needs complex computations and precise measurements. Another approach focuses on offline program generation based on the workpiece CAD model in combination with 3D optical measurements for detection of geometric deviations of the welding assembly [6]. However, it only adapts the robot path and not the welding process, e.g. to account for changing gaps at the joint geometry. Therefore, the authors have presented a method in their previous work to describe joint geometries on workpiece CAD models including gap information for subsequent adaptive robot program planning [7]. In this paper, we present a novel approach developed as part of the project ARENA2036 for evaluation of joint geometries on welding products based on 3D measurements and definition of weld features on the workpieces' CAD models. A matching process is performed between each part of an assembly CAD model and a measured 3D point cloud to determine relative part positions. An updated CAD model is generated and used for automatic weld feature detection. Different joinability indices are presented for automatic evaluation of the joint geometry with regard to an optimal execution of the welding process. Experimental validation is performed with an industrial robot equipped with a structured light sensor employing a stereo camera and welding gun as shown in Figure 1. This paper is structured as follows: In Section 2, the method for 3D measurement and evaluation of joint geometries is presented. In Section 3, the proposed approach is validated by experiments on real body in white workpieces. Finally, conclusions are presented in Section 4.