Evolutionary design of machine-learning-predicted bulk metallic glasses.

Area of Science:

• Materials Science
• Computational Materials Science
• Alloy Design

Background:

• Discovering new alloys with desirable properties is challenging due to vast composition spaces.
• Traditional search methods are often infeasible, limiting the exploration of novel material compositions.
• Prior knowledge constraints can hinder the discovery of truly innovative materials.

Purpose of the Study:

• To apply genetic algorithms for efficient exploration of alloy composition space.
• To evolve alloy candidates towards excellent glass-forming ability (GFA).
• To identify commercially viable novel glass-forming alloys, optimizing for properties like Dmax, ΔTx, and cost.

Main Methods:

• Utilized genetic operators: competition, recombination, and mutation on a population of trial alloy compositions.
• Employed an ensemble neural-network model to predict alloy glass-forming ability.
• Focused optimization on maximum casting diameter (Dmax), supercooled region width (ΔTx), and price.

Main Results:

• Successfully applied genetic algorithms to navigate complex alloy composition spaces.
• Identified promising novel glass-forming alloys, including specific aluminum-based and copper-zirconium-based candidates.
• Demonstrated the algorithm's utility in optimizing existing alloy systems, such as zirconium-based alloys.

Conclusions:

• Genetic algorithms offer a powerful and practical alternative to brute-force searches in alloy discovery.
• This approach enables the identification of novel, high-performance, and cost-effective glass-forming alloys.
• The methodology facilitates targeted alloy design for specific applications and material systems.