Regensburg 2025 – wissenschaftliches Programm

Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

MM: Fachverband Metall- und Materialphysik

MM 13: Topical Session: Defects of Defects

MM 13.3: Vortrag

Dienstag, 18. März 2025, 14:45–15:00, H10

Data-driven modelling of vacancy segregation to grain-boundaries — •Christoph Dösinger, Oliver Renk, and Lorenz Romaner — Montanuniversität Leoben, Department of Materials Science, Roseggerstraße 12, A-8700 Leoben, Austria

Both vacancies and grain-boundaries (GB) are important defects in materials. The vacancies can interact with the GBs which might lead to a formation of voids, as a result this might start the formation of pores or cracks. From atomistic simulations it is known that vacancies can be attracted to GBs, which indeed may act as sinks for the vacancies. However, such simulations, especially if performed using ab-initio, methods can be tedious and costly. In this work we apply machine learning (ML) methods to predict the segregation energies of vacancies to GBs, which give a measure how strongly a vacancy is attracted to specific sites at different GBs. For this ML approach each segregation site is described by its local environment which can be encoded by using for example Steinhardt or SOAP parameters. Together with the site-specific segregation energies a regression model, in our case a Gaussian Process, is trained. Previously we have shown that this approach can be used to predict the segregation of solutes to grain-boundaries. This method for prediction the segregation of vacancies is tested and applied to GBs in tungsten, for which a complete data-set is available for 15 different GBs (Σ3−Σ43). By using this diverse set of GBs, it will be possible to predict the GB segregation of vacancies for general GBs or polycrystalline materials.

Keywords: Grain-Boundary segregation; Vacancy; ab-initio; Machine Learning