Aluminum-based alloys offer exceptional mechanical performance due to their low density, high specific strength, and strong resistance to oxidation and corrosion. In this work, we develop a scalable and transferable machine-learning interatomic potential (MLIP) capable of accurately predicting thermodynamic, mechanical, and microstructural properties across a broad concentration space of Al-Mg-Zr alloys. The MLIP is trained using an active-learning workflow that combines ab initio molecular dynamics, Bayesian uncertainty quantification, and kernel ridge regression, enabling efficient exploration of diverse atomic environments. Additionally, we model an Al/Al3Zr grain-boundary system using experimentally observed orientation relationships and calculate the stress-strain behavior. This framework provides a computationally efficient strategy for exploring the phase space of Al-based alloys and guiding the design of materials with optimized mechanical properties.
Aluminum-based alloys offer exceptional mechanical performance due to their low density, high specific strength, and strong resistance to oxidation and corrosion. In this work, we develop a scalable and transferable machine-learning interatomic potential (MLIP) capable of accurately predicting thermodynamic, mechanical, and microstructural properties across a broad concentration space of Al-Mg-Zr alloys. The MLIP is trained using an active-learning workflow that combines ab initio molecular dynamics, Bayesian uncertainty quantification, and kernel ridge regression, enabling efficient exploration of diverse atomic environments. Additionally, we model an Al/Al3Zr grain-boundary system using experimentally observed orientation relationships and calculate the stress-strain behavior. This framework provides a computationally efficient strategy for exploring the phase space of Al-based alloys and guiding the design of materials with optimized mechanical properties.
