Machine learning assisted design of high-entropy alloys with ultra-high microhardness and unexpected low density

Abstract

High-entropy alloys (HEAs) have attracted considerable attention for their exceptional microstructures and properties. Discovering new HEAs with desirable properties is crucial, but traditional design methods are laborious and time-consuming. Fortunately, the emerging Machine Learning (ML) offers an efficient solution. In this study, composition-microhardness data pairs from various alloy systems were collected and expanded using a Generative Adversarial Network (GAN). These data pairs were converted into empirical parameter-microhardness pairs. Then Active Learning (AL) was employed to screen the Al-Co-Cr-Cu-Fe-Ni system and identify the eXtreme Gradient Boosting (XGBoost) as the optimal ML master model. Millions of data training iterations employing the XGBoost sub-model and accuracy evaluations using the Expected Improvement (EI) algorithm establish the relationship between HEA compositions and microhardness. The proposed sub-model aligns well with experimental data, wherein four Al-rich compositions exhibit ultra-high microhardness (>740 HV, with a maximum of ∼780.3 HV) and low density (<5.9 g/cm3) in the as-cast bulk state. The hardening increment originates from the precipitation of disordered BCC nanoparticles in the ordered AlCo-rich B2 matrix compared to the dilute B2 AlCo intermetallics. This lightweight, high-performance alloy shows potential for engineering applications as thin films or coatings.

Keywords

• High-entropy alloys
• Machine learning
• eXtreme Gradient Boosting
• Microhardness
• Microstructure