Papers › Interpretable Machine Learning for High-Strength High-Entropy Alloy Design
Interpretable Machine Learning for High-Strength High-Entropy Alloy Design
Anurag Bajpai, Ziyuan Rao, Abhinav Dixit, Krishanu Biswas, Dierk Raabe
The archive published only this paper's code-link row. Authors, date and abstract are from arXiv's metadata (CC0), read from the Kaggle arXiv metadata snapshot of 2026-09-12 where its title matched the archive's; the title is the archive's.
High-entropy alloys (HEAs) are metallic materials with solid solutions stabilized by high mixing entropy. Some exhibit excellent strength, often accompanied by additional properties such as magnetic, invar, corrosion, or cryogenic response. This has spurred efforts to discover new HEAs, but the vast compositional search space has made these efforts challenging. Here we present a framework to predict and optimize the yield strength of face-centered cubic (FCC) HEAs, using CoCrFeMnNi-based alloys as a case study due to abundant available data. Our novel Residual Hybrid Learning Model (RELM) integrates Random Forest and Gradient Boosting, enhanced by material attribute data, to handle sparse, skewed datasets for real-world alloys. A hybrid Generative Adversarial Network-Variational Autoencoder model explores new alloy compositions beyond existing datasets. By incorporating processing parameters, which determine the microstructure and thus strength, RELM achieves an R² score of 0.915, surpassing traditional models. SHapley Additive Explanations (SHAP) and Partial Dependencies enhance interpretability, revealing composition-processing-property relationships, as validated by experiments, including X-ray diffraction, SEM analysis, and tensile testing. The model discovered two novel Co₂₀Cr₁₆Fe₂₀Mn₁₆Ni₂₄Al₄ and Co₂₄Cr₁₂Fe₁₂Mn₁₆Ni₂₈Al₄Si₄ HEAs with a maximum possible yield strength of 842 and 937 MPa, significantly exceeding previously reported values for these alloy systems. This study pioneers interpretable machine learning in alloy design, providing a rigorous, data-driven approach to discovering, processing, and optimizing real-world materials. The findings highlight the critical role of both compositional and post-fabrication processing parameters in advancing the understanding of composition-processing-property relationships.
Code
Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.
Code Syntology ran Syntology
Not run by Syntology. Nothing on this page verifies that the listed code works.
Results from the paper archive 2025-07-28
No leaderboard rows for this paper in the archive.
Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections