PSI - Issue 75

Xiru Wang et al. / Procedia Structural Integrity 75 (2025) 85–93 Author name / Structural Integrity Procedia 00 (2019) 000 – 000 3 the statistical distribution of the parameters. Based on the mentioned definition of ‘incorrect local weld toe’ according to ISO 5817:2023 Annex B, Ref. 5052 (weld toe radius ) all specimens were found to have a quality level of C63. According to the same definition Ref. 505 (weld toe angle ) the specimens with start-stop position are in the quality levels C and < D (< D means < 90°), while the specimen without start-stop points refer to quality level B. Fatigue tests were performed at the welded specimen with a width of 120 mm under 4-point bending with an R ratio of 0.1. The fatigue tests results and the corresponding quality levels of the different specimens are given in Figure 1 (c). The load levels of the fatigue tests were converted from bending to tension load and from R=0.1 to R=0.5 by linear-elastic fracture mechanics analysis, details are given by Wang et al. (2025). The fatigue test series met the requirements of FAT90, even if single specimen refer according to ISO5817:2023 Annex B to FAT63 in terms of Ref. 5052 and Ref. 505. Thus, all specimens are within a similar scatter band, even if they refer to different quality levels. Further attention is paid why this is the case.

87

Fig. 1. (a) Definition of weld toe radius and weld toe angle according to ISO 5817:2023, (b) Evaluation of geometrical parameter from 3D-Scans for quality control, (c) Fatigue test results of welded joints with start-stop positions (Wang et al. , 2025)

3. Reverse Engineering Approach As mentioned, an assessment of SCF of welded joints based on geometrical parameters leads to uncertainties compared to a direct assessment based on 3D-scans (reverse engineering approach). For this reason, the 3D-Scans were stepwise translated into Finite Element Models, see Figure 2 (a): After digitalization with SLP system, outliners are automatically detected and the 3D-scan point cloud is aligned according to the Random Sample Consensus (RANSAC) algorithm (see Baumgartner, Schubnell and Augustine (2025) for details). In the next step, a continuous surface model was generated from discrete 3D-point cloud using the Poisson Surface Reconstruction (Kazhdan, Bolitho and Hoppe, 2006). The surface model was transferred into a volume model in STL format and meshed with the commercial preprocessor ANSA©. To reduce the calculation effort and automate the transfer of 2D-cross section into Finite Element mesh, the open-source software tool CalFEM (2025) based on Python was used as pre-processor. In both cases, 2D- and 3D-FEA, the commercial solver ABAQUS© is utilized. To use the same element size in 2D- and 3D-FEA two-dimensional solid tetrahedron-elements CPE6 (3 node linear elements) with a minimum element size of 0.15 mm were used, illustrated in Figure 2 (b) and (c). TIE connectors are used in 2D-FEA to limit the number of the elements in the model, see (Schubnell, Aydogan and Jung, 2023). The Finite Element model was validated with strain determined from DIC measurements in a distance of 10 mm from the weld toe.