Machine learning correlated with phenomenological mode unlocks the vast compositional space of eutectics of multi-principal element alloys

• Published in: Materials & design 2022-07, Vol.219, p.110795, Article 110795
• Main Authors: Chen, Kaixuan, Xiong, Zhiping, An, Miaolan, Xie, Tongbin, Zou, Weidong, Xue, Yunfei, Cheng, Xingwang
• Format: Article
• Language: English
• Subjects: Eutectic; Machine learning; Mechanical performance; Multi-principal element alloys
• ISSN: 0264-1275; 1873-4197
• DOI: 10.1016/j.matdes.2022.110795

[Display omitted] •Knowledge assisted machine learning achieves prediction accuracy higher than 90%.•Eutectics variation significantly alters the microstructures.•Mechanical performance is widely tunable for eutectic CoCrFeNiHf alloys. Eutectic multi-principal element alloys (MPEAs) present a vast compositional space of eutectics, providing a great potential to tailor mechanical performance. However, only limited eutectics have been determined since MPEAs were brought to light in 2014. It still remains a huge challenge to efficiently identify the eutectics. Here, we propose a novel strategy to determine eutectic compositions via phenomenological mode and machine learning, which is validated with Co-Cr-Fe-Ni-Hf/Al MPEAs. Phenomenologically, approximate eutectics can be calculated via the addition of binary eutectics when ignoring the effect of Co-Cr-Fe-Ni interaction. Then, these eutectics are quantitatively corrected by only adjusting Hf content through machine learning. A prediction accuracy higher than 90% is achieved. Noticeably, the variation of eutectic compositions significantly alters the microstructures, leading to great changes in mechanical performances. These findings can potentially pave the pathway to explore the vast compositional space of eutectics and dramatically accelerate the development of eutectic MPEAs.