SurfaceScience21 – wissenschaftliches Programm

Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

O: Fachverband Oberflächenphysik

O 109: Poster Session VIII: Poster to Mini-Symposium: Machine learning applications in surface science III

O 109.7: Poster

Donnerstag, 4. März 2021, 13:30–15:30, P

Using Neural Evolution algorithm to generate disordered High Entropy Alloys structures — •Conrard Giresse TETSASSI FEUGMO¹, Kevin Ryczko^2,3, Abu Anand⁴, Chandra Singh⁴, and Isaac Tamblyn^1,2,3 — ¹National Research Council Canada — ²Department of Physics , University of Ottawa — ³Vector Institute for Artificial Intelligence, Toronto, Ontario, Canada — ⁴Department of Materials Science and Engineering , University of Toronto

A new inverse design approach using pair distribution functions and atomic properties have been implemented. The generative model combines artificial neural networks (ANNs) and genetic algorithms (GAs) to build high disordered crystal structures. The method was introduced by Ryczko et. al. [J. Phys. Chem. C 124, 26117 (2020).] to optimize the doping of graphene-based three-terminal devices for valleytronic applications. Models have been optimized for multicomponent alloy systems such as High Entropy Alloys (HEAs) and structures have been compared to the Special quasi-random (SQSs). Unlike the SQSs, the average optimization time increase slow with the size of the system (ration 1.4). Moreover, the model is able to generate structures with more than 8000 atoms in a few hundred seconds. Finally selected generated structures have been using to compute properties such as the elastic constants, the bulk modulus, and the Poisson ratio, and the results are similar to the SQSs.