| Abstract Scope |
Nickel-based superalloys derive their exceptional high-temperature properties from coherent γ' precipitates, whose formation is controlled by Al and Ti concentrations. To enable detailed atomistic investigations of γ-γ' phase formation mechanisms, we developed machine learning interatomic potentials (ML-IAPs) for the Ni-Al-Ti ternary system using extensive ab initio molecular dynamics (AIMD) datasets from Quantum Espresso calculations. The training data encompasses diverse structural configurations including solid solutions, ordered phases, defects, and finite-temperature snapshots from liquid to solid states. ML-IAPs were trained using MACE and neural network architectures through FitSNAP, with validation against DFT energies, forces, and stresses. The developed potential demonstrates excellent accuracy and is validated through large-scale molecular dynamics simulations of solidification processes and γ-γ' phase transformations, successfully capturing S-L interface motion, γ' precipitate growth, and morphological evolution. This ML-IAP framework provides quantum mechanical accuracy at computational efficiency suitable for studying microstructural evolution across multiple length and time scales. |