PSI - Issue 31

N.D. Bibbo et al. / Procedia Structural Integrity 31 (2021) 75–79 N. D. Bibbo et al. / Structural Integrity Procedia 00 (2019) 000–000

78 4

Fig. 1. Notch stress approach applied at weld toes and weld roots; figure taken from, Sonsino et. al. (2012) 5. Test specimen assessment The various test specimens assessed can be seen in Fig. 2. All finite element models where provided by the author of, Pedersen (2016). All test specimens are steel tube welded connections and are tested under bending, torsion, proportional bending-torsion and non-proportional bending-torsion loads. Non-proportionality is introduced via a phase-shift of the bending-torsion load application. The SN-curves utilized for the Findley and MWCM methods are the FAT-225 curve for uniaxial bending, Hobbacher (2016) and the FAT-167 curve for uniaxial torsion, Sonsino (2008). The weld stresses are assessed according to the notch stress approach and the critical region of each weld is marked in red in Fig. 2. For more information on loading conditions, see Pedersen (2016).

SonsinoTP Amstutz

Witt Yousefi

Bäckström

SonsinoTT

Siljander

Fig. 2. Test specimen geometries; figure taken from Pedersen (2016).

6. Results and discussion The seven different experimental sets based on five different test specimens where assessed for number of cycles to failure for both the Findley and MWCM approaches. Results obtained are compared to the experimentally observed number of cycles to failure obtained from the relevant literature. In Fig. 3, the results for Findley are plotted in a) and for MWCM in b). Predictions from Findley and MWCM are conservative if the number of cycles calculated to failure are lower than the number of cycles to failure observed from experimentation. Therefore, any data points below the black dashed mid-line are non-conservative predictions in both plots. For Findley, the number of non-conservative predictions was 32 and for MWCM only 11. Both damage criterions perform worse in the lower cycle fatigue region where the load amplitudes are high. Linear regression was performed on both data sets. Ideally the lower 95% confidence band should at least lie flush with the black dashed mid-line, as the FAT SN-curves are based on a 97.7% survival rate.