PSI - Issue 79

Martin Sladký et al. / Procedia Structural Integrity 79 (2026) 421–432

431

For this dataset, applying the thickness correction based on the actual wall thicknesses of the individual configu rations resulted in a modest reduction in scatter and a notable decrease in the average level of conservatism, without producing any non-conservative predictions among the specimens considered in the evaluation of the comparison parameters. These findings suggest the potential for revising the currently imposed limits on FAT class increases asso ciated with decreasing wall thickness for the hot-spot stress-based approach. Table 2 summarizes the resulting values of ∆ λ P 50% and T σ obtained in this investigation, alongside those corresponding to the conventional estimations. In contrast to the previously discussed fatigue life estimation methods, the notch stress-based approach inherently accounts for the thickness e ff ect, thereby eliminating the need for any additional thickness correction to the FAT class, as noted by Hobbacher and Baumgartner (2024). However, the resulting S–N data shown in Figure 7 exhibit an unexpected trend, as they tend to separate into two approximately parallel S–N curves corresponding to the hollow section-based and plate-based configurations. Considering the close similarity among the individual configurations in terms of wall thickness, load asymmetry ratio, and testing environment, the observed separation is likely attributable either to the varying volumes of the critically loaded weld seam or to di ff erences in the stress-carrying capability of hollow section-based configurations compared to plate-based ones. Certain insight may be drawn from the observation that a similar separation of the S–N data also occurred in the absolute values derived from the hot-spot stress-based approach. However, this separation nearly disappeared once the S–N data were normalized to the corresponding FAT classes. This e ff ect can be primarily attributed to the higher FAT class FAT193 assigned to the hollow section-based configurations in accordance with Zhao and Packer (2000), compared with the FAT classes FAT110 and FAT99 assigned to the plate-based configurations, as recommended by Hobbacher and Baumgartner (2024) and Rennert et al. (2020). This discrepancy, which aligns well with the presented experimental results, partially arose from the higher limit wall thickness of 10 mm defining the threshold for the FAT class increase in the correction proposed by Rennert et al. (2020), compared with 4 mm specified in the correction proposed by Zhao and Packer (2000). In addition, the thickness correction formulas, corresponding to Equations 1 and 2, di ff er in the rate of FAT class increase with decreasing wall thickness. This increase is more pronounced for the correction proposed by Zhao and Packer (2000), which results in a greater FAT enhancement for hollow section-based joints. Another noteworthy observation was that all hollow section-based configurations exhibited higher fatigue en durance than their plate-based counterparts, despite the former incorporating only load-carrying welds and the latter exclusively non-load-carrying welds. This finding suggests that, at least within this dataset, the distinction between load-carrying and non-load-carrying welds played only a secondary role in determining fatigue endurance and further underscores the need for continued investigation into the fatigue performance of thin-walled welded joints. This study evaluated the fatigue life estimation performance of nominal, hot-spot, and notch stress-based ap proaches for thin-walled welded joints using a dataset of ten configurations comprising 85 individual specimens. None of the evaluated approaches produced non-conservative fatigue life estimates for a significant number of specimens. When the data were assessed as a unified dataset, all three approaches exhibited a comparable average level of conservatism, with the nominal stress approach being the least conservative. However, the hot-spot stress-based approach provided the most consistent predictions, reflected by the lowest scatter, and, like the nominal stress-based approach, exhibited a relatively uniform S–N data distribution about the best-fit S–N curve. In contrast, the notch stress-based S–N data tended to separate into two approximately parallel S–N curves corresponding to plate-based and hollow section-based configurations. At least for this dataset, the predictive performance of the hot-spot stress-based approach could be substantially improved by applying thickness corrections derived from the actual wall thickness of each configuration, instead of the prescribed limit wall thickness. This adjustment resulted in a modest reduction in scatter and a noticeable decrease in the average level of conservatism, without producing any non-conservative predictions for the evaluated specimens. Future research will investigate the causes of the S–N data separation observed for the notch stress-based approach and focus on developing a framework for their incorporation into fatigue life estimation. Moreover, using an expanded dataset, subsequent studies should verify the proposed extension of the wall thickness ranges over which thickness corrections for the hot-spot stress-based approach are calculated from the actual wall thickness. 5. Conclusion