Composition based machine learning to predict phases & strength of refractory high entropy alloys

Abstract

Refractory high-entropy alloys can function at temperatures exceeding those of nickel-based superalloys. Aluminum, as an alloying element, contributes multiple advantageous characteristics to various high-temperature alloys. The Aluminum containing RHEAs have the potential of being the best high temperature materials. In the present study we use the machine learning(ML) technique to determine the phase and yield strength of aluminum containing RHEAs. In this regard, we created the Al-RHEA dataset from the published [1] compilation of RHEA data. We applied multiple ML algorithms to the training set and determined that the CatBoost algorithm gave the best performance. We optimized the hyperparameters of this algorithm and tested it for robustness using cross-validation methods. The CatBoost model predicts the yield strength of test data accurately (R2=0.98). The algorithm was applied to estimate the yield strength for alloy compositions absent from our current dataset, achieving accurate predictions for these unrecorded alloys indicating that the model has learnt the underlying rules to predict the yield strength sufficiently. We then predict the effect of varying aluminum content on yield strength of RHEA. The model predictions were rationalized in view of published data on Al-RHEAs. We also developed the CatBoost classifier model that predicts the phases formed in the alloy of a given composition accurately. The cause for errors in phase prediction is discussed.