Abstract
Alloy design has long been a knowledge-guided approach coupled with experimental trial and error. However, facing high-entropy alloys (HEAs) with complex compositions and large design space, traditional experimental methods often fail. In this paper, an active learning (AL) framework is proposed to achieve the discovery of ultra-high saturation magnetization strength (MS) alloys based on limited data. The framework is based on the AL efficiency to filter the best combination of surrogate model and acquisition function, and we obtained that the combination of Gaussian Process Regression (GPR) + Expected Improvement (EI) has the highest AL efficiency. Through 3 iterations and 9 sets of experimental recommendations, 3 new alloys with better performance than Fe65Co35 (218 emu/g), which is known as the largest MS alloy in the as-cast state, are discovered in the search space containing 110,198 compositions.****Among these discoveries, the Fe70Co25Ni4Si1 alloy achieved a remarkable MS of 225.76 emu/g. This demonstrates the effectiveness of our framework in rapidly designing soft-magnetic HEAs with elevated MS.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call