(V) Thermodynamic and data-driven modeling of grain boundary segregation in metallic alloys

Modeling of grain boundary segregation phenomena is an important discipline of integrated computational materials design. Several computational methods, including in particular atomistic, thermokinetic or mechanical models are available to model grain boundary excess and to assess the related impact on key material properties. With the increased availability of material databases, also data-driven modeling approaches are being applied nowadays to learn trends in the periodic table from physics-based simulations.

We present simulations of grain boundary segregation employing atomistic, thermokinetic and data-driven computational methods. We focus on representative metallic alloys, including refractory metals, steels and coinage metals, and show how a multiscale framework connecting density functional theory evaluation of segregation energies with thermodynamic simulation methods can predict chemistry of grain boundaries from composition and processing parameters. Validation examples with several experimental methods including Auger electron spectroscopy, atom probe tomography and high resolution transmission electron microscopy are presented. As a next step we also present recent activities dealing with machine learning of segregation energies where we discuss the critical role of feature engineering on the basis of different physical parameters including cohesive energies, solution energies, Steinhardt/SOAP parameters and others. The performance of different approaches for investigating segregation in transition metals will be discussed.