Prediction and analysis of bearing capacity and failure mode of grouted sleeve connections under high temperature based on machine learning

Abstract Grouted sleeve connections are a key technology for ensuring the quality of reinforcing‐bar connections in precast structures, and their high‐temperature and fire resistance directly affect the safety of prefabricated buildings under extreme fire conditions. Previous studies have shown that elevated temperatures reduce the ultimate capacity of grouted sleeve connections and gradually shift the failure mode from reinforcing‐bar fracture to pull‐out failure. This study proposes a machine‐learning‐based and interpretable framework for predicting the ultimate capacity and failure mode of grouted sleeve connections under elevated temperatures. A database comprising 549 uniaxial tensile tests conducted under post‐high‐temperature and elevated‐temperature conditions was established, and multiple individual, ensemble, and combined machine‐learning models were systematically compared. For capacity prediction, Bayesian optimization combined with cross‐validation was used to determine the optimal model hyperparameters. XGBoost exhibited superior predictive performance and higher accuracy than conventional calculation methods, while CatBoost and LGBM showed good performance in failure‐mode classification. Model robustness was further evaluated using one‐hot encoding and grouped cross‐validation. Although model performance decreased under strict source‐wise data separation, satisfactory generalization ability was retained. SHAP and partial dependence plot (PDP) analyses were subsequently employed to reveal the effects of key parameters on model predictions. Temperature, reinforcing‐bar anchorage length, and grout strength were identified as important factors governing the failure mode of grouted sleeve connections. Finally, a graphical user interface was developed based on the best‐performing models, providing a practical tool for rapid performance assessment of grouted sleeve connections under elevated‐temperature conditions.