Low-cycle fatigue life prediction method for stud connectors based on interpretable machine learning

Low-cycle fatigue is a common failure mode of stud connectors in bridges. Accurate prediction of their life is crucial for material design and engineering applications. However, traditional theoretical formulas and experimental methods suffer from limitations such as low accuracy and individual variability. This study aims to develop a high-precision prediction model for low-cycle fatigue life using machine learning methods, providing a new approach for material performance evaluation. Firstly, through literature analysis and correlation analysis, key feature variables were identified: f_u, ln(τ_max), ln(Δτ). Secondly, the predictive performance of nine machine learning models was compared by combining cross-validation and hyperparameter optimization. Based on the principle of complementary advantages, an ensemble model was established using Random Forest (RF) and Extreme Gradient Boosting Tree (XGBoost) as the basis. Finally, the SHAP tool was introduced to explain the model’s decision-making process. The results showed that compared to individual models, the integrated model reduced the MAPE by 8.91% and the RMSE by 14.83%, while increasing the R² by 7.32%. The main factor affecting the low-cycle fatigue life of studs is ln (τ _max). The interaction between ln (τ _max) and ln (Δ τ) has the greatest impact on the low-cycle fatigue life of the stud. These findings not only enhance the understanding of fatigue mechanisms in stud connectors but also provide a robust framework for optimizing material selection and design in steel–concrete composite structures.