Intelligent Design and Simulation of High-Entropy Alloys via Machine Learning and Multiobjective Optimization...

Area of Science:

• Materials Science and Engineering
• Computational Materials Science
• Alloy Design

Background:

• High-entropy alloys (HEAs) exhibit unique properties but their complex compositions hinder performance investigation.
• Understanding the physical mechanisms governing HEA performance is crucial for their application.
• Current methods face challenges in efficiently exploring the vast compositional space of HEAs.

Purpose of the Study:

• To introduce a novel computational framework, High-Entropy Alloys Design and Simulations (HEADS), for accelerated HEA development.
• To integrate machine learning (ML), molecular dynamics (MD), and multiobjective optimization algorithms (MOOA) for HEA design.
• To optimize critical mechanical properties, specifically elastic modulus (EM) and ultimate tensile strength (UTS), of HEAs.

Main Methods:

• Utilized ML for initial prediction of phase structures based on HEA composition.
• Employed MD simulations to generate mechanical property data (EM, UTS).
• Developed deep neural network (DNN) models for multitask regression to predict HEA performance and integrated them into MOOA for optimization.

Main Results:

• The HEADS framework accurately predicts HEA phase structures and mechanical properties.
• DNN models successfully fitted MD simulation data, creating a reliable performance prediction model.
• MOOA, using the DNN model as a fitness function, optimized EM and UTS for HEAs, validated with FeNiCrCoCuAlMg alloy.

Conclusions:

• The HEADS framework offers a robust strategy for accelerating the discovery and development of high-performance HEAs.
• This integrated approach provides new insights into optimizing HEA properties for specific engineering applications.
• The framework's flexibility in assigning weights for EM and UTS allows for tailored material design based on application requirements.