SKM 2023 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
MM: Fachverband Metall- und Materialphysik
MM 15: Topical Session: Fundamentals of Fracture – Atomistic Studies of Fracture
MM 15.3: Vortrag
Dienstag, 28. März 2023, 11:00–11:15, SCH A 216
Revealing Atomistic Fracture in BCC Iron by Active Learning — Lei Zhang1, Gabor Csanyi2, Erik van der Giessen1, and •Francesco Maresca1 — 1University of Groningen, Netherlands — 2University of Cambridge, UK
Fracture is multi-scale process that originates from atomic-scale bond rupture and dislocation activity. A clear understanding of these processes is often lacking, because classical interatomic potentials (IAPs) for Molecular Dynamics simulations yield contradicting results that agree only partially with experiments. This is also due to the limited flexibility of their functional form, which is inadequate to describe the complex potential energy surface associated with fracture processes. Emerging machine learning (ML) IAPs allow near-quantum accuracy, based on density functional theory (DFT), but are orders of magnitude faster than DFT. In this work, we develop an active learning algorithm that enables the prediction of atomistic fracture mechanisms via the Gaussian Approximation Potential (GAP) approach. An existing DFT database for ferromagnetic bcc iron is first enriched with configurations that are relevant for the fracture process. Next, the active learning approach is applied to four crack systems, in which the maximum predicted per-atom error is reduced to 10 meV. The predicted critical stress intensity factors are compared with theory estimates, and the learning efficiency of the approach is analysed. Our work provides an active learning strategy for improving ML-IAPs for fracture, while revealing for the first time the atomic scale mechanisms that initiate fracture in iron with quantum accuracy.