Machine-learning and high-throughput studies for high-entropy materials

The combination of multiple-principal element materials, known as high-entropy materials (HEMs), expands the multi-dimensional compositional space to gigantic stoichiometry. It is impossible to afford a holistic approach to explore each possibility. With the advance of the materials genome initiative and characterization technology, a high-throughput (HT) approach is more reasonable, especially to identify the specified functions for the new HEMs development. There are three major components for the HT approach, which are the computational tools, experimental tools, and digital data. This article reviews both the materials informatics and experimental approaches for the HT methods. Applications of these tools on composition-varying samples can be used to obtain stoichiometry effectively and phase-structure-property relationships efficiently for the materials-property database establishment. They can also be used in conjunction with machine learning (ML) to improve the predictability of models. These ML tools will be an essential part of HT approaches to develop the new HEMs. The ML-developed HEMs together with ML-created other materials are positioned in this manuscript for future HEMs advancement. Comparing all the reviewed properties, the hierarchical microstructures together with the heterogeneous grain sizes show the highest potential to apply ML for new HEMs, which needs HT validations to accelerate the development. The promising potential and the database from the HEMs exploration would shed light on the future of humanity building from the scratch of Mars regolith.

Automated summary

The development of high-entropy materials (HEMs), with their vast compositional space, requires a high-throughput approach to identify functionalities and establish a materials-property database. This article reviews the use of materials informatics and experimental methods in high-throughput approaches. Machine learning is identified as a crucial tool for accelerating the development of new HEMs. Hierarchical microstructures and heterogeneous grain sizes show the highest potential for applying machine learning in new HEMs.