Prediction of Ti-Zr-Nb-Ta high-entropy alloys with desirable hardness by combining machine learning and experimental data

Traditional alloy design depends heavily on trial and error experiments, which are neither cost-effective nor efficient, particularly for the development of high-entropy alloys (HEAs) using a broad composition space. Herein, we combine a machine learning (ML) model with phase diagram calculations (CALPHAD) to design Ti-Zr-Nb-Ta refractory HEAs with a desirable hardness. The extreme gradient boosting (XGBoost) algorithm is used to train the ML model based on the Ti-Zr-Nb-Ta HEA hardness dataset from CALPHAD-assisted experiments. As a result, the most important features (i.e., the Ta content, melting point, and entropy of mixing) are determined via feature selection and model optimization. Moreover, the high performance of the ML model is validated experimentally, and the prediction accuracy reaches 97.8%. This work provides not only an interpretable ML model that can be used to predict the hardness of Ti-Zr-Nb-Ta HEAs but also feasible guidance for the development of HEAs with desirable hardness.

Automated summary

Traditional alloy design methods have limitations, prompting the use of machine learning and phase diagram calculations to successfully design refractory high-entropy alloys like Ti-Zr-Nb-Ta HEAs with desirable hardness. The XGBoost algorithm was utilized for training the model, leading to the identification of crucial features and achieving a high prediction accuracy of 97.8% through experimental validation.