Comparison of experimental and numerical fatigue life of austenitic stainless steel components at 300 °C with idealized and scanned weld geometries

• All calculated fatigue life predictions were validated with experimental fatigue life data. • With the presented approach fatigue life prediction was more precise and less conservative compared to other approaches. • With this methodology, minimal scatter was achieved without retrospectively modifying material models or fatigue curves. • Gradient based FGF, showed the best performance compared to commonly used and other advanced fatigue damage parameters. Accurately predicting the fatigue life of welded components remains a challenging task. This work compares different geometric approaches, such as scanned or idealised weld geometries, using energy-based and strain-based fatigue damage parameters. Gradient effects have also been successfully considered and investigated. In addition to the analytical calculation using the damage parameters, the fatigue life prediction is based on elastic–plastic material models of the base materials and the weld metal, which have been implemented in the numerical simulation. The material models are based on data determined by testing unnotched specimen in LCF, HCF and VHCF regime. This work answers the question of how accurate the fatigue life prediction of welded components can be without assuming material model parameters, retrospective adaptation of the material model or the fatigue life curves. Furthermore, an improved methodology is presented to reduce conservatism without reducing safety compared to conventional guidelines. The use of the fictitious 1 mm radius gave reliable results (Mdn. of log. deviation 0.418) with low scatter (SD. 0.174) in the LCF, HCF regime and over 2·10 6 cycles. On the other hand, the implementation of the real weld geometry increased the scatter of the results (SD. 0.38), but remained within an acceptable range for the more performant fatigue damage parameters presented in this work (Mdn. of log. deviation −0.15). Locations of crack initiation could be predicted, at 11 out of 13 specimens with failure, using the numerical simulation of the scanned weld geometry.