Machine-Learning-Assisted Multi-objective Screening of Hardness and Oxidation Resistance in Refractory High-Entropy Alloys

Abstract Refractory high-entropy alloys (RHEAs) exhibit excellent mechanical strength and thermal stability, but their practical application is frequently limited by insufficient oxidation resistance. This study develops a machine-learning-assisted workflow for multi-objective screening of hardness and isothermal oxidation resistance in RHEAs. The selected baseline hardness XGBoost model and baseline oxidation random forest model achieve held-out R 2 values of 0.79 and 0.76, respectively. The current fitted models screen 109,571 quinary compositions and are compared with six alloys fabricated following an initial ML-assisted multi-objective selection targeting contrasting hardness–oxidation trade-offs. Among these alloys, HHHO_2 exhibits the highest measured hardness of 780 ± 8 HV, whereas LHHO_1 exhibits the lowest specific mass gain of 2.75 ± 0.04 mg/cm 2 , corresponding to an oxidation-resistance index I ox of − 0.44 after oxidation at 1000 °C for 20 h. The comparatively large oxidation-prediction errors motivate development of a time-resolved oxidation model incorporating oxidation duration and a composition-weighted Pilling–Bedworth ratio. Within the complete feature set, gradient boosting achieves a five-fold cross-validation R 2 of 0.92 ± 0.02 and a held-out R 2 of 0.94. For three further alloys selected using the initial advanced-model predictions, the advanced model reduces the mean absolute I ox prediction error from 0.46 to 0.21, representing an approximately 53% reduction relative to the baseline oxidation model. The refined HHHO candidate exhibits a hardness of 661 ± 11 HV and a specific mass gain of 6.20 ± 0.01 mg/cm 2 . Repeated out-of-fold SHAP and cross-model ranking analyses are used to assess the robustness of the learned associations, while multivariate applicability-domain analysis identifies candidates requiring extrapolative interpretation. Overall, the workflow provides an experimentally informed approach for prioritizing RHEA compositions with different hardness–oxidation trade-offs while explicitly accounting for model variability and compositional applicability.

Authors

Institutions

Publication Details

Journal
High Entropy Alloys & Materials
Published
2026-11-16
DOI
https://doi.org/10.1007/s44210-026-00087-5
Primary Topic
High Entropy Alloys Studies
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Machine-Learning-Assisted Multi-objective Screening of Hardness and Oxidation Resistance in Refractory High-Entropy Alloys

Stéphane Gorsse, An‐Chou Yeh, Wei-Chih Lin
5 links
High Entropy Alloys & Materials
High Entropy Alloys Studies
article

Machine-Learning-Assisted Multi-objective Screening of Hardness and Oxidation Resistance in Refractory High-Entropy Alloys

Stéphane Gorsse, An‐Chou Yeh, Wei-Chih Lin
article en

Abstract

Abstract Refractory high-entropy alloys (RHEAs) exhibit excellent mechanical strength and thermal stability, but their practical application is frequently limited by insufficient oxidation resistance. This study develops a machine-learning-assisted workflow for multi-objective screening of hardness and isothermal oxidation resistance in RHEAs. The selected baseline hardness XGBoost model and baseline oxidation random forest model achieve held-out R 2 values of 0.79 and 0.76, respectively. The current fitted models screen 109,571 quinary compositions and are compared with six alloys fabricated following an initial ML-assisted multi-objective selection targeting contrasting hardness–oxidation trade-offs. Among these alloys, HHHO_2 exhibits the highest measured hardness of 780 ± 8 HV, whereas LHHO_1 exhibits the lowest specific mass gain of 2.75 ± 0.04 mg/cm 2 , corresponding to an oxidation-resistance index I ox of − 0.44 after oxidation at 1000 °C for 20 h. The comparatively large oxidation-prediction errors motivate development of a time-resolved oxidation model incorporating oxidation duration and a composition-weighted Pilling–Bedworth ratio. Within the complete feature set, gradient boosting achieves a five-fold cross-validation R 2 of 0.92 ± 0.02 and a held-out R 2 of 0.94. For three further alloys selected using the initial advanced-model predictions, the advanced model reduces the mean absolute I ox prediction error from 0.46 to 0.21, representing an approximately 53% reduction relative to the baseline oxidation model. The refined HHHO candidate exhibits a hardness of 661 ± 11 HV and a specific mass gain of 6.20 ± 0.01 mg/cm 2 . Repeated out-of-fold SHAP and cross-model ranking analyses are used to assess the robustness of the learned associations, while multivariate applicability-domain analysis identifies candidates requiring extrapolative interpretation. Overall, the workflow provides an experimentally informed approach for prioritizing RHEA compositions with different hardness–oxidation trade-offs while explicitly accounting for model variability and compositional applicability.

High Entropy Alloys & Materials
Centre National de la Recherche Scientifique (FR), Nanyang Technological University (SG), National Tsing Hua University (TW), Institut de Chimie de la Matière Condensée de Bordeaux (FR), Institut Polytechnique de Bordeaux (FR), CNRS-International-NTU-Thales Research Alliance (SG)
National Tsing Hua University
High Entropy Alloys Studies
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon