Machine Learning Derived Periodic Table for High Entropy Alloys

This work explores new high entropy alloys (HEAs) which address the challenges in trade-offs between strength, ductility and various environmental effects. To address the limitations in traditional regression approaches which can generate large amounts of data rapidly but do not account for the complex interplay in correlating properties, we instead apply an unsupervised learning approach. By mapping the high dimensional nature of the systematics of elemental data embedded in the periodic table, the influence of specific combinations of elements on engineering properties of HEAs are captured. This approach is designed to capture the interplay between chemistry, microstructure and phase stability, which allows us to identify chemical design rules for improving mechanical properties with minimal trade-offs. This work uses a graph representation approach to capture the thermodynamic and structural complexity of high entropy alloys (HEAs). This approach has<br/>been used for materials discovery based on first principles, but now we are using it to design engineering alloys. We identify the potential existence of new combinations of phases not previously identified by tracking the connections in the network, which are analogous to tie lines in a traditional phase diagram representation. In this way, mechanical properties are rationally designed through proposed chemical design rules across the entire HEA search space, resulting in a machine learning based representation of a periodic table based on HEA properties.