Machine Learning Accelerated, High Throughput, Multi-Objective Optimization of Multiprincipal Element Alloys.

Area of Science:

• Materials Science
• Computational Materials Science
• Alloy Design

Background:

• Multiprincipal element alloys (MPEAs) exhibit exceptional properties but face challenges in cost-effective design due to vast compositional spaces.
• Identifying MPEAs with specific desired properties remains a significant hurdle in materials development.

Purpose of the Study:

• To present a highly efficient design strategy for MPEAs by integrating molecular dynamics (MD) simulation, machine learning (ML) algorithms, and genetic algorithms (GA).
• To accelerate the discovery of high-performance MPEA materials with desired properties.

Main Methods:

• Utilized MD simulations to generate initial data for training ML models.
• Developed ML algorithms to predict alloy properties like stiffness and critical resolved shear stress (CRSS).
• Employed a multi-objective GA coupled with the ML model to optimize alloy compositions.

Main Results:

• An ML model trained on 54 MD simulations accurately predicted stiffness and CRSS for CoNiCrFeMn alloys with low relative errors (2.77% and 2.17%).
• Achieved a 12,600-fold reduction in computation time compared to traditional methods.
• Identified 100 optimal CoNiCrFeMn alloy compositions with simultaneously high stiffness and CRSS, verified by 100,000 ML-accelerated predictions.

Conclusions:

• The integrated MD-ML-GA strategy offers a highly efficient and precise method for MPEA design.
• This approach significantly accelerates the discovery of novel MPEAs with tailored properties.
• The methodology is adaptable for identifying MPEAs with other principal elements, broadening its applicability.