Predicting the grain boundary segregation energy of solute atoms in aluminum by first-principles calculation and machine learning

Grain boundary (GB) segregation energy is an important factor affecting the segregation behavior of solute atoms and the mechanical properties of alloys. In this study, first-principles calculation combined with machine learning (ML) algorithms were used to calculate and predict the GB segregation energies of solute atoms in Al alloys. Five GB structures and 44 common solute atoms in aluminum were selected for the calculations, and a database of 924 groups describing the relationship between GB characteristics and GB segregation energy of solute atoms was constructed. Calculation results and feature importance analysis show that the atomic radius and Voronoi volume of solute atoms play significant roles in determining segregation energies. Nine ML algorithms, including three linear regression models, four decision tree models, a support vector regression model, and an artificial neural networks model, were employed to predict the GB segregation energy. The results indicate that increasing model complexity leads to an overall improved prediction accuracy. The performance of decision tree models is generally better than that of linear regression models. The artificial neural network model exhibits the highest performance, demonstrating a promising combination of accuracy and efficiency, and ten cross-validations confirmed the robustness and generalization ability of the model on the prediction task of GB segregation energy.