Blast Performance Enhancement of RC Columns Using Machine Learning

Reinforced concrete buildings may experience complete or partial progressive collapse due to column failure under blast loading. Therefore, predicting the structural response of columns under blast loading can support preventive measures that reduce fatalities and financial losses. Using thirteen features related to column and blast properties, this study presents machine learning models to predict the maximum lateral displacement of reinforced concrete (RC) columns strengthened with CFRP sheets under blast loading. A total of 1263 CFRP-strengthened numerical configurations were generated from 421 unstrengthened baseline specimens, including 121 collected from the literature and 300 developed using ANSYS AUTODYN, by applying two CFRP wrapping schemes and three CFRP thicknesses. The strengthening schemes consisted of full-height and middle-third CFRP wrapping. The resulting blast-performance enhancement was assessed based on the reduction in maximum lateral displacement. XGBoost, CatBoost, LightGBM, and Gradient Boosting algorithms were evaluated. The XGBoost model was interpreted locally and globally using the SHAP algorithm to determine the significance and contribution of each input parameter. The accuracy of the XGBoost model was evaluated by comparison with results from a previous study and ANSYS AUTODYN finite element simulations. The XGBoost model achieved high predictive performance, with R2, RMSE, MAE, and MAPE values of 0.97, 16.237 mm, 7.722 mm, and 5.1%, respectively, demonstrating its ability to predict blast response and identify relationships between the key parameters and maximum lateral displacement.