PSI - Issue 79

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

429

Notch stress-based approach

9

6

CHS lap CHS fillet

1 Normalized stress range ∆ λ [-] ∆ λ P 50% = 2.370 T σ = 1:2.615 m= 5 2 4

CHS-Plate fillet S CHS-Plate fillet G RHS fillet OPB RHS fillet IPB Plate fillet T Ben Plate fillet T Ten Plate-RHS fillet Plate fillet L Best-fit S-N curve Design S-N curve

0.7

10 4

10 5

10 6

10 7

Fatigue life in cycles N [-]

Fig. 7. S–N data for the notch stress-based approach, normalized to the FAT class FAT500, with white-filled markers denoting specimens excluded from the evaluation of parameters listed in the annotation box.

level of conservatism, and the scatter range index T σ , as defined by Radaj et al. (2006), which quantifies the degree of scatter. Due to the limited size of the evaluated dataset, attempts to fit the data using the natural S–N curve slopes yielded visually unconvincing results, especially for the notch stress-based approach. Therefore, the slopes of the fitted S–N curves were fixed at m = 5, following the same rationale as that adopted for the design S–N curves. The resulting comparison parameters for each prediction approach are summarized in Table 2.

Table 2. Parameters characterizing the average level of conservatism and scatter for the individual fatigue life estimation approaches, evaluated assuming a fixed S–N slope of m = 5.

∆ λ P 50% for 10 6 cycles [-]

Fatigue life estimation approach

Scatter range index T

σ [-]

nominal stress-based approach hot-spot stress-based approach notch stress-based approach

2.016 2.236 2.370

1:2.519 1:1.566 1:2.615

hot-spot stress-based approach employing FAT classes determined from the actual wall thicknesses

1.613

1:1.400

None of the employed prediction approaches produced non-conservative fatigue life estimates for the specimens included in the evaluation of the comparison parameters ∆ λ P 50% and T σ , with the exception of a few Plate fillet L specimens for the nominal and notch stress-based approaches. For this dataset, all approaches exhibited a broadly comparable level of conservatism, with the nominal stress-based approach tending to be the least conservative and the notch stress-based approach the most conservative on average. A closer examination, however, revealed that both the nominal and notch stress-based results were subject to pronounced scatter, resulting in substantial variation in the level of conservatism across individual configurations. Assessment of the investigated prediction approaches in terms of scatter revealed that the results for individual configurations within the nominal stress-based approach were distributed fairly evenly about the best-fit S–N curve. A similarly uniform scatter was observed for the hot-spot stress-based approach; however, its magnitude was substan-