Abstract: Dynamic recrystallization (DRX) behavior is influenced by multiple interacting factors, including deformation temperature, strain rate, and solid solution of elements. Traditional empirical formulas and physically based models struggle to capture the complex evolution mechanisms of DRX under such multivariate interactions, making the prediction of DRX behavior in low-alloy steels challenging. This study developed a machine learning (ML) model based on elemental coupling effects and recrystallization features to predict DRX behavior by constructing a comprehensive information base from experimental and literature data. Principal component analysis (PCA) was applied to extract the principal components that characterized the elemental coupling effects. The mathematical expressions of these coupled principal components were constructed using an eigenvalue decomposition algorithm, systematically revealing the synergistic mechanisms of multiple alloying elements on DRX behavior. Based on these findings, an XGBoost model was developed to predict DRX behavior. Feature importance analysis further demonstrated that the influence of principal components on DRX exhibited a dynamic evolutionary law. The model achieved excellent predictive performance (R² > 96 %) across a range of strain rates (0.01, 0.1, 1, and 10 s^-1), significantly improving the accuracy of DRX volume fraction predictions and offering a valuable theoretical guidance for the design of high-performance low-alloy steels.

Key words: Dynamic recrystallization, Machine learning, Elemental coupling effects, Principal component analysis, Feature importance analysis