Data-driven prediction and evaluation model for explosion damage in reinforced concrete columns with densified stirrups

Reinforced concrete (RC) columns serve as core load-bearing members in engineering structures. Under explosive loads, they may sustain damage and potentially trigger the collapse of the entire structure. Therefore, rapidly and accurately assessing the damage state of reinforced concrete columns subjected to explosive forces is crucial for ensuring structural safety and implementing protective measures. This study proposes a data-driven model based on machine learning (ML) to predict damage indices of RC columns under explosive loading. The study constructed a comprehensive database comprising 3,133 samples, integrating 259 literature-derived data samples and 2,874 numerically simulated supplementary data samples. This dataset encompasses both uniformly reinforced RC columns and densely reinforced RC columns meeting seismic design requirements. Eleven key parameters were selected as input features, with RC column damage indices serving as output features. Six ML models were employed to predict damage indices under explosive loading, and their predictive accuracy was comparatively analyzed using four regression evaluation metrics. The results indicate that the Tabular Prior-data Fitted Network (TabPFN) demonstrated optimal prediction accuracy and generalization capability, achieving an R² value of 0.989 on the test set with MAE and RMSE values as low as 0.018 and 0.03, respectively. Further interpretability analysis of the TabPFN model was conducted using the SHAP (SHapley Additive exPlanations) method. Analysis indicates that explosive weight, detonation distance, and column cross-section depth are the key features governing RC column damage severity. Notably, the contribution of volumetric stirrup reinforcement ratio to enhancing RC column blast resistance significantly outperforms that of longitudinal reinforcement ratio. Finally, through comparative analysis with finite element simulation results, the TabPFN model was validated to exhibit excellent generalization performance under unseen operating conditions. In terms of computational efficiency, the proposed model achieves a single-prediction inference time of approximately 0.5 s, representing a significant efficiency gain of nearly four orders of magnitude compared to conventional numerical simulation methodsThe established model enables rapid and precise prediction of damage to reinforced concrete columns under explosive loading, providing guidance for structural blast-resistant optimization design and post-disaster rapid assessment.