Machine learning assisted design of new ductile high-entropy alloys: Application to Al-Cr-Nb-Ti-V-Zr system

Abstract

The search for new high-entropy alloys (HEAs) with desired properties is an urgent problem that is hardly solvable experimentally due to the extremely large number of possible alloy compositions. Thus, methods for theoretical prediction of HEA's properties play a key role. Currently, effective predictive models are based on machine learning methods and modern data analysis algorithms. Here we address developing data-driven machine learning models (DDML) to predict the ductility of HEAs. We have built several DDMLs and found that the best approach is based on the Support Vector Classifier, which significantly outperforms phenomenological models (balanced accuracy of 0.784 and F-score of 0.824). By combining this model with a previously developed yield strength prediction model, we have predicted and fabricated novel HEAs of the Al-Cr-Nb-Ti-V-Zr system with good mechanical properties. An obtained Al₁Cr₉Nb₃₅Ti₅V₄₀Zr₁₀ alloy demonstrates a combination of high strength at room and elevated temperature, combined with good ductility at room temperature.