Predicting Solidification Shrinkage of Metals with Machine-Learned Force Fields

Shrinkage during solidification of metals causes imperfections in cast metal products. Several compositionally complex brass alloys developed by our group exhibit less solidification shrinkage than conventional yellow brass, but precise experimental measurement is difficult because it requires working with metals in the liquid state. Our aim is to determine the effect that alloy composition has on volumetric change between the liquid and the solid phases using a computational approach. Our project investigates the low solidification shrinkage of compositionally-complex alloys using machine-learned force fields (MLFF) for the simulations. Machine-learned force fields are computationally efficient and maintain the accuracy of first principles DFT calculations. We use data collected from MLFF simulations to create thermal expansion curves of disordered face-centered cubic crystal lattice and liquid phase structures of alloys to predict shrinkage during the liquid-to-solid phase transition. These results are validated using corresponding experimental casting shrinkage measurements. To build on our method’s initial successes, we are exploring its use for the minimization of solidification shrinkage via alloy composition.