High-entropy alloys (HEAs) are multicomponent alloys with five or more elements. The composition and the ordered microstructure of HEAs can result in alloys with high mechanical properties, such as high strength, high ductility, and high fracture toughness. Due to these properties, HEAs have great potential in state-of-the-art applications, e.g. race cars, aircraft, etc. However, considering the large design space offered by HEAs, experimentally and computationally, it is really challenging to develop new HEAs. In recent years machine learning and deep learning based models have been used as surrogates for computationally expensive molecular simulations methods to accelerate the materials design. Here, we have developed conventional neural networks (CNN) based deep learning models as surrogates for molecular dynamics (MD) simulations. To train the CNN model, MD simulations data was generated and used by performing simulations on FeNiCrCoAl, FeNiCrCoCu and FeNiCrCoMn alloys using the embedded atom method (EAM) and second nearest-neighbor modified embedded atom method (2NN MEAM) interatomic potentials. The CNN model was integrated with different optimization algorithms (particle swarm optimization, genetic algorithms, and Monte Carlo tree search), to further efficiently explore the HEA design space. For each of the alloys, the alloy composition was optimized to generate new alloys with improved mechanical properties. Our study provides a pathway for accelerating the design of new multicomponent materials with target properties that are sensitive to the composition and order of the component in their microstructures.

Deep learning based surrogate models for accelerated design of high entropy alloys

Fall 2022 abstracts, posters, and presentations will launch in mid-September.

ACS Fall 2022

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.