Efficient alloy design strategy for fast searching for high-entropy alloys with desired mechanical properties

Abstract

The exponentially large compositional space of high entropy alloys (HEAs) offers more possibilities for designing alloys with desired properties. However, it also poses challenges to using the traditional “trial and error” approach in alloy design. In this work, an XGBoost model for predicting the elastic properties of the NbTiVZr family across the entire compositional space was established by combining density-functional theory (DFT) calculation results as the dataset with machine learning (ML) algorithms. Furthermore, considerations of charge transfer were incorporated into the solid solution hardening (SSH) model, and the model was further modified. Through comparing plasticity evaluation indices, the parameter D (γSurf/γGSFE) was determined to be suitable for predicting the plasticity of NbTiVZr alloys. A full compositional space model for yield strength and plasticity has been constructed based on the modified SSH model and the parameter D, respectively. Ultimately, an alloy design system combining the full compositional space models for yield strength and plasticity was established, achieving good consistency with experimental results. And a non-equiatomic alloy with a yield strength exceeding that of equiatomic alloys by 32.2 % (1409 MPa), while maintaining 29.27 % compressive strain was discovered. In conclusion, this work provides an efficient design strategy for alloys with desired properties.

Keywords

• Density-functional theory (DFT)
• Machine learning
• High entropy alloys
• Mechanical properties
• Alloy design