Accelerated design of low-activation high entropy alloys with desired phase and property by machine learning

Low-activation high-entropy alloys (HEAs) have been regarded as novel candidate structural materials for fusion reactors due to their excellent mechanical and radiation resistant properties. Nevertheless, the potential vast composition space brings a prominent challenge in the design of low-activation HEAs. Herein, a new strategy based on machine learning (ML) was proposed to accelerate the exploitation of low-activation HEAs with desired phase and property. Two optimized classification models with accuracy of > 85% were developed to identify single-phase body-centered-cubic (BCC) solid-solution (SS) HEAs. One regression model with correlation coefficient (R) of > 0.9 was constructed to predict the hardness of HEAs. The phase and hardness prediction models were combined in accordance with an integrated design strategy to identify the low-activation HEAs with desired phase and property from 284,634 candidates. With this design strategy, a new desired single-phase BCC Fe35Cr30V20Mn10Ti5 low-activation HEA with 555.9 ± 15.3 HV was designed and fabricated via only two experimental iterations. Besides, two new phase selection rules with accuracy of > 95% were given to efficiently screen SS and BCC HEAs, respectively. Our research work not only provides new ideas to search for low-activation HEAs, but also can be extended to the integrated design of structure and performance of other advanced materials.