Processing of point clouds from 3D scans of welded joints for the weld detection and weld quality analysis as input for a reliable fatigue assessment

3D laser scans are increasingly used for the analysis of the weld quality. Based on these scans, it is possible to determine the weld profiles relevant for the fatigue strength, i.e. the weld angle and notch radii or other features such as spatter or undercuts. These methods are currently being developed and demonstrated in scientific investigations. However, in order to enable the applicability of 3D laser scans for quality assessment in practice, methods are still needed to automatically and reliably identify the weld seam and the weld direction from 3D scans. With this information, the weld profile in 2D cross-sections can be evaluated. Current studies have shown that the detection of the weld seam is the major challenge for weld quality assessment based on local geometrical parameters, such as toe angle and notch radii. In this paper, different algorithms are presented that enable an automated weld seam detection. Three approaches are investigated: algorithm-based methods that are based on curvature of the surface, methods based on the random sample consensus (RANSAC) algorithm or artificial neural networks (ANN). All methods are presented in detail and applied to a different weld geometries.