Machine-learning-guided phase identification and hardness prediction of Al-Co-Cr-Fe-Mn-Nb-Ni-V containing high entropy alloys

A machine learning technique based on artificial intelligence (AI) has emerged as a potential tool for accelerating the search and design of new high entropy alloys (HEAs) and predicting their mechanical properties. This study demonstrates the implementation of a cutting-edge machine learning framework integrated with optimization strategies to predict the phase formation and hardness of several HEAs, including Fe25−xCo25Ni25Cr20V5Nbx (x = 2.5, 5, 7.5, 10 at. %) HEAs. In this investigation, five machine learning (ML) models, namely Decision Trees, Random Forest (RF), Bagging, AdaBoost, and Extra Trees, were used to identify phases. Additionally, six ML models, including XGBoost, Gradient Boost, Bagging, Extra Trees, RF, and artificial neural network (ANN), were employed to predict hardness. After evaluating the performance and optimization of each model, an Extra Trees classifier (with accuracy = 0.893) and an ANN (with R2 = 0.95, and MAE = 34.91) model showed the best predictive capabilities for phase and hardness prediction, respectively. Finally, we utilized the ML-based model to predict the phase (for 16 HEA compositions) and hardness (for 12 HEA compositions) of various HEAs, after which we validated them with experiments. The Extra Trees model successfully identifies the phase of both previously reported HEAs and currently studied Fe25−xCo25Ni25Cr20V5Nbx (x = 2.5, 5, 7.5, 10 at. %) HEAs. The ANN model predicted hardness matched the experimentally measured hardness with an average error of 13.25 %. The results of our experiments are tracked with our predictions, suggesting that ML-based approaches could be helpful to design HEAs in the future.