Machine learning driven rationally design of amorphous alloy with improved elastic models

• Published in: Materials & design 2022-08, Vol.220, p.110881, Article 110881
• Main Authors: Li, Zhuang, Long, Zhilin, Lei, Shan, Tang, Yulin
• Format: Article
• Language: English
• Subjects: Amorphous alloy; Elasticity; GFA; Machine learning; Plasticity
• ISSN: 0264-1275; 1873-4197
• DOI: 10.1016/j.matdes.2022.110881

[Display omitted] •Composition and structure descriptors were distinguished as the input of machine learning.•The ML-based elastic models exhibit higher accuracy, wider applicability and better interpretability than that of other reported models.•Molar volume and electronegativity were identified as key descriptors determining G and K of amorphous alloys.•Performances of our proposed elastic models were well verified via GFA and plasticity prediction in two ternary systems.•A general framework for rational design of amorphous alloys with desired properties was proposed. Rational design of amorphous alloys from the viewpoint of elasticity can be helpful as it offers close correlations with glass forming ability (GFA), thermal stability, mechanical properties and so on. Here, by separately employing composition and structure descriptors as input, we successfully optimized, generated and interpreted the elastic predictive models via various machine learning (ML) approaches, which exhibit distinct advantages of high accuracy, simple operation, wide applicability and good interpretability relative to that of previously reported elastic models. Meanwhile, the performances of our developed elastic models were well verified via GFA and plasticity prediction in two ternary amorphous alloy systems. Finally, based on the above improved elastic models, we proposed a general framework for rational design of amorphous alloys using four steps strategy. Our results demonstrate the great potential to accelerate the composition screening and property optimization of amorphous alloys.