Development of single-phase BCC refractory high entropy alloys using machine learning techniques

[Display omitted] •Development of BCC RHEAs with high liquidus temperature by integrated approach using computation and experiment.•Latin hyper-cube sampling technique is used to extract RHEAs dataset consisting single-phase and dual phases.•Single-phase BCC phase prediction is done using multiple machine learning (ML) algorithms.•Phase prediction is done effectively by random forest (RF) algorithm with high F1 score of 0.93.•Predicted three single-phase BCC RHEAs by ML methods are validated by thermodynamic simulation and experimental techniques. The current study presents the application of both computational and experimental techniques in the quest for novel single-phase BCC refractory high entropy alloys (RHEAs) with high liquidus temperature and phase stability. The phases of RHEAs are predicted using different machine learning (ML) algorithms such as Logistic Regression (LR), Gradient Boosting (GB), Support Vector Machine (SVM), Decision Tree (DT), K- Nearest Neighbor (KNN), Random Forest (RF), and Artificial Neural Network (ANN). Latin hyper-cube technique is used to extract 489 datasets consisting of 243 single-phase BCC solid solution (SS) and 246 non-single-phase RHEAs & then multiple machine learning methods are used to train datasets. With high F1 score of 0.93, training accuracy of 99.4% and a test accuracy of 93.88%, the phase prediction is done effectively by RF algorithm which distinguishes between single-phase BCC solid solution phase and non-single-phases (SS+Intermetallics) RHEAs. Subsequently the three predicted RHEAs with BCC structure such as Mo-Nb-Ti-V-W (Tm = 2916 K), Mo-Nb-Ti-Ta-W (Tm = 2909 K), Mo-Nb-Ti-V-Ta-W (Tm = 2780 K) are compared with thermodynamic simulation method. Finally, the designed three RHEAs are synthesized experimentally, and the formation of BCC structure is confirmed.