A stocky tubular tension-torsion specimen geometry was optimized to characterize the effect of the stress state (stress triaxiality and Lode angle parameter) on metals ductility, at low stress triaxialities. Biaxial tests (proportional and non-proportional) were performed on 36NiCrMo16 steel and 2024-T351 aluminum alloy. Strain fields were measured by stereo-correlation of digital images during the tests. Loading paths to fracture (evolution of the equivalent plastic strain, the stress triaxiality and the Lode angle parameter at the critical point) were determined. The evolution of aluminum ductility with respect to the stress triaxiality measured from tension-torsion tests differed substantially from that obtained by Bao and Wierzbicki in 2004. Indeed, the latter suggested a minimal ductility under shear, while the tension-torsion technique revealed a maximal ductility under shear. Non-proportional loading paths were shown to have an influence on ductility, by means of tests consisting in a pre-compression, pre-tension or pre-torsion, followed by a proportional loading sequence under combined tension-torsion. SEM observations of metallographic sections from biaxial interrupted tests, a real-time monitoring of the surface strain and damage during in-situ torsion tests in the SEM, and a crack propagation test coupled with in-situ X-ray synchrotron laminography brought evidences of localization phenomena at different scales, and of the growth of some cavities, even under pure shear, by contrast with the total collapse predicted by unit cell models. This growth may be due to the significant axial elongation measured under pure torsion (Swift effect). Shear localization was identified as the main coalescence mechanism, which justifies the choice of the Hosford-Coulomb fracture initiation criterion. Used in conjunction with a non-linear damage indicator, it accounts for the measured ductilities, even under possibly non-proportional loadings.
