Torsional damage analysis of metallic materials using simulation-guided defect assessment

Abstract

This study presents an integrated assessment of torsional damage in selectively laser-melted 316L stainless steel, AlSi10Mg aluminium alloy, and Ti–6Al–4V titanium alloy, combining mechanical testing, finite-element interpretation, post-fracture laser-confocal microscopy, and instance segmentation of fracture-surface discontinuities. The torsion series comprised 2, 10, and 3 independent specimens, with mean maximum torques of \(56.5 \pm 0.7\) , \(25.4 \pm 2.0\) , and \(76.3 \pm 4.2\) N m, and corresponding conditional torsional stresses of \(659.5 \pm 6.4\) , \(227.3 \pm 17.2\) , and \(955.3 \pm 50.9\) MPa. Tensile tests provided additional context, with 0.2% proof stresses of \(552.8 \pm 33.8\) , \(161.8 \pm 2.3\) , and \(911.6 \pm 183.9\) MPa, and ultimate tensile strengths of \(680.8 \pm 46.2\) , \(251.9 \pm 3.2\) , and \(1127.1 \pm 37.3\) MPa. Post-fracture microscopy yielded 75 original images and 1125 tiles. A physical-specimen split (9/3/3 train/validation/test) prevented data leakage. Preliminary YOLOv8n-seg diagnostics gave a maximum mask F1 of 0.68 and AP \(_{50}\) of 0.629, but these are exploratory due to missing per-specimen predictions. Finite-element results are treated qualitatively, as model records lack complete post-yield calibration, mesh-convergence history, and synchronized torque–twist data. The workflow separates measured quantities, numerical interpretation, and image-based evidence, identifying the data required for a fully validated torsional constitutive model.