Machine-learning-assisted design of high-hardness high-entropy alloys based on component space screening

Previous machine-learning (ML)-based studies on high-entropy alloys typically depend on alloy composition and empirical parameters to construct and select features, a process that is not only cumbersome but also prone to reducing the computational accuracy due to feature redundancy. In the present study, the process of the feature construction and screening relying on empirical parameters in previous alloy hardness predictions has been simplified, and a direct mapping relationship between the composition and hardness has been established. Inputting the alloy composition, based on five ML models — random forest (RF), extreme gradient boosting (XGBoost), support vector machine (SVM), decision tree (DT), and multilayer perceptron (MLP) — to predict the hardness of the alloy. Grid search and K-fold cross-validation are utilized to optimize the model and improve prediction accuracy. Among them, the XGBoost model demonstrates outstanding predictive performance (R2 = 0.94, RMSE = 54.9, and MAE = 35.4). According to the prediction results, we selected the alloy with the highest predicted hardness for experimental fabrication and validation. The measured hardness of the alloy reaches 841 HV, with a relative error of only 3.3% compared to the predicted result, which is significantly higher than the hardness of the Al-Co-Cr-Fe-Ni-V alloys in the dataset. The introduction of Shapley additive explanation (SHAP) revealed that the Al and V elements have a significant impact on hardness improvement, which aligns with the analysis results of the fabricated alloy. This study establishes a rapid prediction framework from the composition to hardness, providing new insights for the design of novel high-entropy alloys.