Deep Learning-Based Framework for Efficient Design of Multicomponent High Hardness High Entropy Alloys.

Area of Science:

• Materials Science
• Computational Materials Science
• Alloy Design

Background:

• Machine learning (ML) offers a new paradigm for high-entropy alloy (HEA) research.
• Traditional alloy design methods are limited, while purely data-driven approaches may not guarantee performance.
• Optimizing the balance between exploring new HEA systems and ensuring reliable performance is a key challenge.

Purpose of the Study:

• To develop a deep learning framework integrating materials domain knowledge and data-driven techniques.
• To optimize the design process for multicomponent, high-hardness HEAs.
• To address the limitations of traditional and purely data-driven alloy design approaches.

Main Methods:

• A material concatenation embedding (MCE) module coupled with a BiLSTM-CRF network was used to analyze 2698 research papers, extracting 8067 data points.
• Materials domain knowledge was incorporated to identify high-potential elements and processing conditions for a curated hardness dataset.
• A two-stage design strategy combining genetic algorithm (GA) and particle swarm optimization (PSO) was employed for alloy system exploration and composition refinement.

Main Results:

• Three novel HEAs were successfully designed: Cr$_{20.6}$Fe$_{22.5}$Mo$_{20.6}$Ti$_{18.3}$V$_{18}$, Al$_{9.32}$Cr$_{20.62}$Fe$_{21.71}$Mo$_{27.09}$Ti$_{21.26}$, and Al$_{6}$Cr$_{20.3}$Fe$_{19.5}$Mo$_{20.1}$Nb$_{18.8}$Ti$_{15.3}$.
• The predicted average relative error in hardness was less than 5%.
• The optimal designed alloy's hardness was only 38 HV below the historical record.

Conclusions:

• The proposed deep learning framework effectively integrates domain knowledge and data-driven methods for HEA design.
• This approach facilitates the discovery of novel high-hardness HEAs with predictable performance.
• The framework offers a promising strategy for advancing multicomponent alloy development.