Machine-learning assisted coarse-grained potential development for cold-sprayed multi component alloys

Cold-sprayed deposition of multicomponent systems such as high entropy alloys offer an exciting avenue for developing new corrosion and wear resistant materials. However, the large compositional space makes it an onerous task to experimentally investigate their mechanical properties. As an alternative, an accurate and high-speed simulation method can be invaluable for efficiently exploring these systems. To address this, we devise a coarse-grained molecular dynamics simulation method for multicomponent systems that predicts their mechanical properties with accuracy comparable to that of an all-atom simulation while requiring a fraction of the computational cost. In our approach, the all-atom system is simplified to a collection of coarse-grained quasi particles with fewer degrees of freedom. The interaction potential for these quasi-particles is then learned through a machine-learning algorithm. Simulations of tensile tests using this new interaction potential demonstrate that, compared to an existing coarse-graining method, our method reproduces the stress-strain behavior more accurately in both the elastic and plastic regions.