A Hybrid Artificial Intelligence Framework for Predicting and Optimizing Post-Fire Bond Performance in Reinforced Concrete

The degradation of the bond between steel rebar and concrete during fire exposure significantly endangers structural safety. Existing empirical models often fail to capture this complex behavior. This study presents an integrated framework that couples a predictive model with a genetic algorithm (GA) to estimate post-fire bond performance and optimize fire-resistant designs. A comprehensive database was assembled containing over 1000 experimental data points, and four machine learning models were evaluated. Among these, XGBoost achieved the highest accuracy in predicting bond strength ( R² = 0.949), normalized bond strength ( R² = 0.907), and failure mode (Accuracy = 0.983). SHAP analysis confirmed the model’s physical validity. The analysis identified temperature as the most dominant negative factor, followed by bar diameter and concrete strength. The validated XGBoost model was then used as the fitness function in a GA-based optimization to address the inverse design problem and identify optimal material and geometric parameters. The optimal configuration shifted from high-strength materials at temperatures below 600 °C to more ductile materials at higher temperatures. Stirrups also became crucial for confinement at temperatures above 600 °C. This research provides an interpretable, high-accuracy AI framework: https://concrete-bond-strength-predictor-mvo3gwvjts6eredemqq9xc.streamlit.app/ .