Predicting New Single/Multiphase Structure High Entropy Alloys by Pattern Recognition Network

The phase structures of a dataset of 634 set high entropy alloys (HEAs) were predicted by machine learning methods and classified into four categories: bcc, fcc, bcc+fcc and others (containing intermetallic compounds and other structural alloys). The algorithm was a Pattern Recognition Network (PRN) using the cross-entropy as loss function, which could predict the phase formation probability of HEAs. The PRN algorithm was proved to be over 87% accurate on the test data. With the increase of Al content in AlxCoCu6Ni6Fe6 and AlxCoCrCuNiFe HEAs, the transformation from fcc to fcc+bcc and then to bcc structure was successfully predicted by the PRN algorithm. Meanwhile, the AlxCoCu6Ni6Fe6 (x=1,3,6,9) HEAs were also prepared and characterized, the experimental results showed that the phase structures were highly consistent with the prediction.