Berlin 2024 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
MM: Fachverband Metall- und Materialphysik
MM 62: Developement of Calculation Methods III
MM 62.5: Talk
Thursday, March 21, 2024, 16:45–17:00, C 264
Machine-learned interatomic potential for microstructure formation in Ni-rich NiAl systems — •Adam Fisher1, Julie B. Staunton1, Huan Wu2, and Peter Brommer1 — 1University of Warwick, Coventry, UK — 2TWI Ltd, Cambridge, UK
Precipitates in nickel-based superalloys form during heat treatment on a time scale inaccessible to direct molecular dynamics simulation, but can be explored using kinetic Monte Carlo (kMC) studies. This requires reliable values for the barrier energies separating distinct atomic configurations. We have previously described a method to find and validate barriers in this system and found that classical potentials such as embedded-atom method (EAM) fail to reproduce the correct ordering of barriers. Modern machine-learned interatomic potentials (MLIPs) have been shown to have an accuracy near that of density functional theory (DFT) at a fraction of the cost. In this work, we fit an atomic cluster expansion (ACE) MLIP for nickel-rich NiAl systems using ACE hyper-active learning (ACEHAL), training on a series of structures, from cubic unit cells of Ni and Ni3Al to large (>100w atoms) NiAl solid solution cells. This is complemented by HAL runs on saddle point configurations, which improve the description of energy barriers. The MLIP barriers are then validated and compared to several traditional interatomic potentials.
Keywords: Ni-based superalloys; Kinetic Monte Carlo; Machine Learned Interatomic Potentials; Density Functional Theory; Atomic Cluster Expansion