PSI - Issue 57

Yixuan Hou et al. / Procedia Structural Integrity 57 (2024) 73–78 Author name / Structural Integrity Procedia 00 (2019) 000 – 000

77

5

fatigue lifetime. In Fig. 4, the finite element simulation results using two regenerated surface profiles are shown. It can be found that even if the applied stress (250 MPa) is below the yield stress of the material (753 MPa), the MPS at the critical notch tip are extremely high, which are resulted from the sharp notch geometries. Besides, Fig. 4 (b) indicates that the deepest notch is not the critical notch, because the deepest notch is relatively blunt, therefore the stress concentration is not severe. In [16], the depth of critical notch is considered as an important notch geometry parameter which affects the fatigue lifetime. In this study, the critical notch depths of two real surface profiles used for GAN training datasets are set as notch depth thresholds to filter the regenerated surface profiles. Based on the notch depth filtration strategy, 61 surface profiles are filtered from the 80 regenerated surface profiles. To this end, 300 FE simulations are conducted with these surface profiles, using python scripts to reproduce the fatigue scatter. Fig. 5 shows the estimated fatigue lifetime using 61 filtered surface profiles. To reduce the uncertainty in fatigue lifetime-based design, a regression probabilistic Weibull fatigue model is introduced to determine the statistical probability of estimated results, the probabilistic S-N curves are shown in Fig. 5. The mean fatigue lifetime is estimated by 50% survival probability curve, while the lower and upper boundaries are described by 5% and 95% survival probability curves for less conservative estimation. Compared with experiment data taken from literature [16], the estimated fatigue lifetime shows good conformation.

Fig. 4. FE simulation results using surface profiles: a) Experimental surface profile extracted from [16], (σ max =250 MPa); b) regenerated surface profile using GAN, (σ max =250 MPa).

σ a [Mpa]

5% 50% 95%

Estimated data

Experiment data

N [Cycles]

Fig. 5. Probabilistic stress-life curve of estimated fatigue lifetime.