Breaking data barriers: advancing phase prediction in high entropy alloys through a new machine learning framework

High entropy alloys (HEAs) represent a promising ADVANCEMENT in the context of Industry 4.0, embodying the principles of interconnectedness, automation and real-time data. The formation of phases in HEAs is predominantly influenced by the composition of the alloy and its corresponding thermodynamic properties. With an expansive composition space, HEAs exhibit diverse phase formations. While previous research has primarily focused on solid solution features, establishing a direct link between compositions and phase formations is crucial for accelerating the development of novel HEAs. Machine learning algorithms have been extensively employed to predict phases in HEAs; however, their effectiveness relies heavily on the availability of large datasets. Unfortunately, acquiring precise data remains challenging due to the nascent nature of HEA research. In this research article, we propose a novel machine learning algorithm to predict alloy phases, even in scenarios with limited data. The primary objective is establishing a robust relationship between elemental properties and phases, enabling phase predictions based on alloy compositions. Our study achieves an accuracy of 92% using a dataset comprising only 118 samples. The proposed work is compared with state-of-the-art approaches, addressing data scarcity challenges in HEAs and advancing predictive modeling for phase predictions in these alloys.