Efficient surrogate models for materials science simulations: Machine learning-based prediction of microstructure properties

• DOI: 10.1016/j.mlwa.2024.100544
• Publication Date: 2024-03-11T17:02:15Z
• Authors: Binh Duong Nguyen; Pavlo Potapenko; Aytekin Demirci; Kishan Govind; Sébastien Bompas; Stefan Sandfeld
• Keywords: Robustness

Determining, understanding, and predicting the so-called structure–property relation is an important task in many scientific disciplines, such as chemistry, biology, meteorology, physics, engineering, materials science. Structure refers to spatial distribution of, e.g., substances, material, or matter general, while property a resulting characteristic that usually depends non-trivial way on details of structure. Traditionally, forward simulations models have been used for tasks. Recently, several machine learning algorithms applied these fields enhance accelerate simulation surrogate models. In this work, we develop investigate applications six techniques based two different datasets from domain science: data two-dimensional Ising model formation magnetic domains representing evolution dual-phase microstructures Cahn-Hilliard model. We analyze accuracy robustness all elucidate reasons differences their performances. The impact including knowledge through tailored features studied, general recommendations availability quality training are derived this.