Unraveling phase prediction in high-entropy alloys: A synergy of machine learning, deep learning, and ThermoCalc, validation by experimental analysis

The phase formation in high entropy alloys (HEAs) presents a significant challenge due to the complexity of their composition and the intricate interactions between multiple elements. The machine learning (ML) and deep learning (ANN) models play a crucial role in phase prediction for HEAs owing to their capability to handle intricate, multi-dimensional datasets and capture nuanced relationships between composition and phase formation. This article seeks to enhance the understanding of phase prediction in HEAs by utilizing ML, ANN, ThermoCalc, and experimental validation techniques. Parameters such as δ, VEC, and Tm, influential in predicting phases, were discerned using the Pearson correlation method. Various ML models, including kneighbors, bagging, adaboost, decision tree, extra trees, and ANN, were employed for predicting phase formation in HEAs. The ANN model exhibited an impressive accuracy of 90.62 %, while the extra trees model achieved an accuracy of 89.73 %. These ML and ANN models adeptly predicted the observed phases in experimental results, correctly identifying both HEAs Co10Cr19Fe30Mn23Ni9Ti8 and Co7Cr22Fe29Mn24Ni14Ti4 as having a face-centered cubic (FCC) + intermetallic (IM) structure. However, it is noteworthy that ThermoCalc and other ML models almost misclassified these HEAs. Both alloys primarily consist of an intermetallic phase enriched in titanium (Ti) and manganese (Mn) while exhibiting a noticeable depletion of iron (Fe) content. A comparison of these approaches underscores the significance of experimental validation in assessing the accuracy and reliability of phase predictions in HEAs, revealing the strengths and limitations of each method.