PSI - Issue 38

Jinchao Zhu et al. / Procedia Structural Integrity 38 (2022) 621–630 Author name / Structural Integrity Procedia 00 (2021) 000 – 000

629

9

The estimated fatigue strength at 2 million cycles and a survival probability of 97.7% for the four cases are 86.2 MPa, 88.4 MPa, 108.1 MPa and 101.3 MPa, respectively, see Fig. 10. Although the difference in estimated fatigue strengths between Case 1 and Case 2 is small, it is observed that the estimated strength is higher in Case 2. This observation, which may seem counterintuitive, is explained by the skewness of the PDF of K t in Case 2 (see Fig 7a). That is, the mean value of K t in Case 2 is lower than the deterministic K t computed in Case 1. This explains the slight increase of fatigue life when considering the geometrical uncertainties in the analysis. It is also observed that the fatigue strengths obtained from Case 3 and Case 4 are higher than the fatigue strengths from Case 1 and Case 2. This is expected since applying r ref = r actual + 1 mm instead r ref = 1 mm yields a lower K t . Furthermore, it is also noted that the NS method with r ref = r actual + 1 mm, FAT 200 and stochastic variability in K t (Case 4) yields a more conservative result compared to the same NS method using a deterministic K t (Case 3). Therefore, Case 4 predicts a detrimental influence of the stochastic variabilities in weld geometry.

Fig. 10. SN-curves at 97.7% survival probability of four studied cases.

4. Concluding remarks A numerical comparison in terms of predicted fatigue life between 4 NS methods has been investigated: Case 1: NS method with r ref = 1 mm and FAT 225. The stochastic variability in toe angle, leg length and toe radius along the weld are neglected. Instead, mean values for toe angle and leg length as well as a fixed 1 mm toe radius are used to compute a deterministic SCF using FEA. Case 2: Stochastic NS method with r ref = 1 mm and FAT 225. The stochastic variability in SCF is computed using Monte-Carlo simulations based on the probability distribution of toe angle and leg length. Case 3: NS method with r ref = r actual + 1 mm and FAT 200. The method uses the actual toe radius r actual enlarged by 1 mm in the computation of the SCF. Similar to Case 1, the stochastic variability in toe angle, leg length and toe radius along the weld are neglected. Instead, assumed mean values are used and a deterministic SCF is computed. Case 4: Stochastic NS method with r ref = r actual + 1 mm and FAT 200. Similar to Case 2, the method includes the stochastic variability in the SCF in the fatigue life assessment. In addition to the probability distribution of toe angle and leg length, the probability distribution of actual toe radius is also taken into account.

The following conclusions can be drawn based on the studied non-load carrying cruciform joint:

• The stochastic NS method with r ref = 1 mm and FAT 225 (Case 2) yields roughly the same predicted fatigue life as when deterministic (mean) values for the leg length and toe angle are used (Case 1). • The NS method with r ref = 1 mm and FAT 225 (Case 1) is substantially more conservative, in terms of predicted fatigue life, than the NS method with r ref = r actual + 1 mm and FAT 200 (Case 3).