Machine learning-enabled high-entropy alloy discovery (2202.13753v1)

Abstract: High-entropy alloys are solid solutions of multiple principal elements, capable of reaching composition and feature regimes inaccessible for dilute materials. Discovering those with valuable properties, however, relies on serendipity, as thermodynamic alloy design rules alone often fail in high-dimensional composition spaces. Here, we propose an active-learning strategy to accelerate the design of novel high-entropy Invar alloys in a practically infinite compositional space, based on very sparse data. Our approach works as a closed-loop, integrating machine learning with density-functional theory, thermodynamic calculations, and experiments. After processing and characterizing 17 new alloys (out of millions of possible compositions), we identified 2 high-entropy Invar alloys with extremely low thermal expansion coefficients around 2*10-6 K-1 at 300 K. Our study thus opens a new pathway for the fast and automated discovery of high-entropy alloys with optimal thermal, magnetic and electrical properties.

Insights into Machine Learning-Enabled High-Entropy Alloy Discovery

The paper "Machine Learning-Enabled High-Entropy Alloy Discovery" presents a novel methodology for accelerating the discovery of high-entropy alloys (HEAs) with desirable properties, addressing the immense challenge posed by their complex compositional spaces. The research introduces a closed-loop framework that integrates active learning with ML models, density-functional theory (DFT), thermodynamic calculations, and experimental feedback to efficiently explore the composition space of HEAs.

Key Contributions

The paper's primary contribution is the development of an active learning framework, which is meticulously designed to circumvent the limitations of traditional alloy design methods in high-dimensional composition spaces. The proposed method leverages:

• HEA COmposition Generating Scheme (HEA-COGS): A deep generative model optimized for small-to-medium experimental datasets, tailored to sample potential new functional alloys, particularly focusing on achieving low thermal expansion coefficients (TEC).
• Two-stage Ensemble Regression Model (TERM): This model includes multiple ML techniques—specifically multilayer perceptron (MLP) and gradient boosting decision trees (GBDT)—for robust prediction of the TEC of proposed alloys. The multi-stage approach integrates computational techniques with thermodynamics and experimental verification.
• Iterative Learning Process: The framework's iterative nature, guided by experimental results, enables a continual refinement of the predictive models, thus enhancing the accuracy and efficiency of the discovery process.

Numerical Results and Findings

The framework results in substantial strides in discovering compositionally complex alloys with low TECs. Noteworthy achievements reported include:

• Identification of two high-entropy Invar alloys with TECs as low as 2×10^-6 K^-1 at 300 K, recorded across only a handful of experimental iterations.
• Successful demonstration of a framework capable of more than fifty-fold increased efficiency compared to conventional trial-and-error approaches.
• Discovery of 7 new Invar alloys showcasing TECs below 5×10^-6 K^-1, benchmarked against both historical datasets and comparative conventional methods.

Theoretical and Practical Implications

The research elucidates pivotal insights into the inherent correlations between compositional structure and thermomechanical properties within a high-entropy context, bridging theoretical understanding and practical alloy design. This ML-enabled approach not only streamlines material discovery but also expounds upon the utility of active learning in managing the sparse data challenge pervasive in materials science.

Speculations on Future Developments

The methodological framework proposed in this paper bears promise for further applications in multi-faceted property optimization of HEAs. It projects a significant paradigm shift in addressing multifactorial material properties such as magnetism and corrosion resistance alongside mechanical characteristics. The model's adaptability invites potential refinement and broader applicability across diverse alloy systems, opening avenues for impactful innovations within industrial applications.

Conclusion

This paper exemplifies a concerted effort to harmonize computational advancements with experimental pragmatism, steering high-entropy alloy design into a more predictive, less serendipitous domain. The integration of ML with well-established theoretical models marks an informed step towards efficient exploration of vast, convoluted composition spaces, aspiring to satisfy both scientific inquiry and industrial pragmatism.