Enhancing bond performance in SRC structures: a computational approach using ensemble learning techniques and sequential analysis

Composite structures using Steel Reinforced Concrete (SRC) are increasingly being praised for outperforming standard steel or reinforced concrete structures in terms of strength, stiffness, corrosion resistance, and cost-effectiveness. The interfacial bond strength of steel and concrete is critical for establishing composite action in SRC structures, but correct computation remains difficult. This study addresses the above limitations by employing Ensemble Learning Techniques (ELTs), a subset of Machine Learning, focusing on the application of Decision Tree, AdaBoost, Random Forest, and Extreme Gradient Boost algorithms. The study improves engineering knowledge of bond performance by introducing a basic set of input parameters for SRC columns. Furthermore, the Shapley Additive exPlanations (SHAP) approach is included to enhance the interpretability of the model by providing insights into parameter impacts. The analysis of EL models shows that XGBoost outperforms other models in both training and validation, exhibiting strong correlation coefficients and perfect alignment with benchmarks. Sequential analysis, a novel approach for establishing ELT’s predictive power, further highlights XGBoost’s world-class performance. SHAP analysis identifies “relative concrete coverage” and “relative side coverage” as factors that affect the prediction of bond strength. Overall, this study significantly contributes to engineering knowledge by demonstrating the effectiveness of Ensemble Learning Techniques in predicting bond strength for SRC structures, offering a valuable and practical design tool.