Thermodynamic modelling routinely guides the research and development (R&D) of new materials and the CALPHAD (CALculation of PHAse Diagrams) method is one of the approaches of choice because of its accuracy achieved with a modest computational effort.
This method uses experimental data to optimize thermodynamic models of materials. However, the fact that CALPHAD is a data-driven method might limit its applicability in the R&D of the materials for the next generation of technologies that are often found in chemistry/structure spaces that are yet to be explored experimentally.
A strategy often used to fill gaps in the experimental data series is to use ab initio modelling based on density functional theory (DFT). However, DFT calculations are still too computationally expensive to be routinely applied to develop CALPHAD models of large families of materials.
Here, the project follows the paradigm of combining high-throughput DFT calculations and machine learning (ML) in order to address data scarcity. A remarkable recent progress has been the development of ML universal interatomic potentials applicable to any chemical compositions. In this project, the team will investigate whether these ML models predict the energies of intermetallic systems with similar accuracy as DFT and could become viable data sources for CALPHAD modelling.
| Thermo-Calc Software AB, Sweden. |