Machine learning method to investigate the influence of alloy elements on non-metal interstitials in HCP metals

Guodong Lu, Xiuzhi Qiu, Zhixiao Liu, Dong Wang, Tianguo Wei, Yi Zhao, Wangyu Hu, Huiqiu Deng,
Machine learning method to investigate the influence of alloy elements on non-metal interstitials in HCP metals,
Materials & Design,
Volume 259,
2025,
114747,
ISSN 0264-1275,
https://doi.org/10.1016/j.matdes.2025.114747.
(https://www.sciencedirect.com/science/article/pii/S0264127525011670)
Abstract: Interactions between non-metal interstitial atoms and substitutional elements in HCP metals critically affect material properties, yet existing predictive models lack systematic integration of these effects. This study develops a data-driven framework to accurately predict formation energy and stability while elucidating their underlying mechanisms. We integrated density functional theory (DFT) calculations with machine learning (ML) techniques to investigate non-metal interstitials in HCP metals containing substitutional elements. We constructed an automatically procedure to label the stability of interstice which is very important for constructing dataset. The dataset comprised DFT-derived formation energies (for prediction), labeled stability of non-metal interstitial atoms (for classification), and VoronoiNN-extracted geometric and elemental descriptors. Various ML algorithms were trained, with SHAP and SISSO analyses identifying key features and yielding a predictive equation, subsequently validated on an independent dataset. The linear equation demonstrated strong generalization capability (R2 = 0.951 on validation). Atomic radius differences and electronegativity emerged as critical descriptors, highlighting distinct interstitial behaviors in HCP structures. This research delivers an efficient predictive tool that reduces reliance on computationally expensive DFT calculations and accelerates alloy defect investigation. By addressing the gap in systematic interstitial-substitutional interaction models in HCP metals, this work enhances understanding of defect energetics and establishes a foundation for applications across broader crystal systems.
Keywords: Machine learning; Non-metal interstice; HCP metal; SISSO