Machine-learning model for predicting phase formations of high-entropy alloys

The phase formations of high-entropy alloys (HEAs) are essential to their properties, but efficient prediction of them remains a challenge. In this work, we propose a support vector machine model that is capable of distinguishing stable body-centered cubic (BCC), face-centered cubic (FCC) HEA phases, and the remaining phases out of the 322 as-cast samples with a cross validation accuracy over 90% after training and test. With the model, we predicted 369 FCC and 267 BCC equiatomic HEAs from the composition space of 16 metallic elements, one order larger than the number of available experimental data. Furthermore, dozens of refractory HEAs with high ratios of melting temperature to density have been screened out. Eleven of them agree with recent experiments and the 20 quinary ones with highest melting temperatures are validated through first-principles calculations. The proposed model is complementary to the calculation of phase diagrams and ab initio methods and could serve as an effective guide for designing new HEAs.