Molecular dynamics simulation and machine learning of wettability in the Ag-Cu/ FeCoNiCrMn high-entropy alloy system

Brazing is one of the commonly used joining methods for dissimilar materials, where the wettability of the metal filler on the base metal directly determines the quality of brazing performance. Currently, research on wettability involves capturing and reproducing the spreading process of metal fillers using cameras, which is inefficient and consumes substantial resources. With the continuous advancement of computer computing power, large-scale material simulation calculations have become feasible. In this study, FeCoNiCrMn-based high-entropy alloys and non-active Ag-Cu brazing fillers are selected as the research objects, and a universal wettability judgment method based on data analysis is proposed. MD simulation combined with machine learning (ML) was employed to investigate the wettability of Ag-Cu on non-equiatomic FeCoNiCrMn HEAs. A database describing the relationship between material composition and wettability was established based on high-temperature simulation experiments. Four ML models for wettability learning tasks were studied and compared, including deep neural network (DNN), kernel-based extreme learning machine (KELM), support vector machine (SVM), and extreme gradient boosting (XGBoost). It was found that the DNN and XGBoost model outperformed the others in solving the regression problem of wettability. The prediction accuracy (R²) values are 68.8% and 60.2%, respectively, while the prediction errors (RMSE) are 3.7° and 4.4°, respectively. Through SHAP (Shapley Additive Explanations) analysis, it was revealed that Mn and the melting point of HEAs are factors influencing the wettability of this interface system. Finally, the accuracy of the XGBoost model was verified by designing high-temperature wetting experiments of Ag-Cu with HEAs of different compositions. The results indicate that the combination of computational research and machine learning can realize accurate prediction of wettability for complex HEA systems and enable reasonable analysis of material properties through data mining.