Improved Al-Mg alloy surface segregation predictions with a machine learning atomistic potential

Abstract

Various industrial/commercial applications use Al-Mg alloys, yet the Mg added to Al materials, to improve strength, is susceptible to surface segregation and oxidation, leaving behind a softer and Al-enriched bulk alloy. To better understand this process and provide a systematic methodology for investigating dopants that can mitigate corrosion, we have developed a robust atomistic deep neural net potential (DNP) using a dataset generated with first-principles density-functional theory (DFT). The potential, validated systematically against DFT values, has been shown to have a high fidelity in calculating different elemental and intermetallic Al-Mg systems' properties. Our calculations predict a linear trend in the formation energy of the Al-Mg alloy and its density as a function of temperature, consistent with experimental literature. Employing the DNP within a hybrid Monte Carlo and molecular dynamics (MC/MD) approach, we predict anisotropic surface segregation for Al-Mg alloys such that (111)(100)(110), with (111) surfaces displaying the lowest segregation enthalpies and Mg enrichment. Furthermore, we model the segregation tendencies by adapting a recently introduced isotherm model for grain boundary segregation. Our results show that this model describes the MC/MD segregation profiles with higher fidelity than the McLean and Fowler-Guggenheim isotherm models.