Leakage-Safe Machine Learning for Hydrogen Embrittlement Detection in 316L Stainless Steel: A Region-Held-Out Evaluation of Texture and Deep Features in SEM MicrographsNew

This paper addresses hydrogen embrittlement (HE) detection in 316L stainless steel using scanning electron microscopy (SEM) micrographs. The authors highlight that conventional image-level train-test splits can cause data leakage when multiple images originate from the same specimen region. They propose a region-held-out evaluation protocol to ensure realistic performance assessment. The study compares traditional texture features against deep learning approaches for classifying HE-induced microstructural changes, providing insights into model generalization across unseen specimen regions.