SKM 2023 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

MM: Fachverband Metall- und Materialphysik

MM 37: Topical Session: Defect Phases II

MM 37.6: Talk

Thursday, March 30, 2023, 12:15–12:30, SCH A 216

Learning chemistry dependence of grain boundary segregation energies — •Christoph Dösinger¹, Daniel Scheiber², Oleg Peil², Vsevolod Razumovskiy², and Lorenz Romaner¹ — ¹Montanuniversität Leoben, Department of Materials Science, Leoben, Austria — ²Materials Center Leoben Forschung GmbH, Leoben, Austria

The grain-boundary segregation energy (E_seg) is the central quantity for describing the process of grain-boundary segregation which influences fracture. Usually, to obtain highly accurate values for E_seg, density functional theory is employed, which incurs high computational costs. This makes it impractical to do a thorough study of segregation to multiple grain-boundaries for a range of solutes. To reduce the number of calculations needed for such a complete description, we apply machine learning methods to density functional theory data. By using separate sets of descriptors for the local atomic environment and the solute types, we fit a model based on gaussian process regression. This approach is evaluated on a comprehensive data-set for E_seg in tungsten. The tests indicate that the model has the ability to extrapolate to solutes which are not contained in the training data.