Machine learning guided BCC or FCC phase prediction in high entropy alloys

Abstract

High entropy alloys （HEAs） have excellent properties because they can form simple solid solution (SS) phases, including body-centered cubic (BCC) phase, face-centered cubic (FCC) phase, or FCC + BCC phase, so phase prediction is the first step in alloy design. In current research, machine learning (ML) approach had been widely used to guide the discovery and design of materials. The prediction of HEAs phase structure based on machine learning (ML) is a hot topic. In this work, five ML algorithms were utilized to predict HEAs for SS and amorphous (AM) phases based on 399 collected data sets, including 120 BCC alloys, 87 FCC alloys, 82 BCC + FCC alloys and 110 a.m. alloys. To enhance the model's accuracy, grid search and K-fold cross validation were used to optimize performance. Valence electron concentration (VEC) and ΔHmix exhibit high importance in prediction in compared to other parameters. The results show that the random forest can effectively distinguish BCC phase, FCC phase, mixed solid solution phase (FCC + BCC) and AM, with an accuracy is 0.87. After that, the CoCrFeNiAlx (x = 0, 0.5, 1) system alloys were characterized by XRD and SEM-EDS. The experimental results validated that the phase structure of CoCrFeNiAlx alloys changed from FCC to BCC + FCC and BCC with the increase of Al content, which is consistent with the ML prediction.

Keywords

• Machine learning
• High entropy alloys
• Phase structure
• Prediction