Modeling and Prediction of Blast-Induced Response in RC Columns Using Graph Neural Networks

Efficient and accurate prediction of the structural response of reinforced concrete (RC) components under explosive impact is critical for emergency repair decisions, structural strengthening, and protective design. Existing fast calculation approaches, such as analytical models and lightweight data-driven methods, offer high computational efficiency but are limited in accuracy when it comes to calculating three-dimensional structural response fields. This paper proposes a rapid damage prediction model for RC columns based on Graph Neural Networks (GNN). By leveraging the neighborhood aggregation mechanism of the GNN to efficiently transmit mechanical interaction information within the structure, the model establishes an end-to-end mapping between explosive load inputs and the three-dimensional structural responses of the column, enabling fast and accurate damage prediction. A multi-scenario feature coupling training strategy is further introduced, enabling the GNN model to adapt to variations in reinforcement ratios, explosive charges, and detonation locations, thereby significantly improving its generalization performance across different blast conditions. Results show that the model completes a single prediction in just 55 milliseconds, achieving a speed improvement of more than 4 orders of magnitude compared to traditional methods, with a prediction error below 3.33% and high-precision damage prediction under various explosive conditions. The model maintains high prediction accuracy across multiple blast scenarios, demonstrating excellent robustness. This study highlights the potential of GNN-based approaches in blast-induced damage prediction and provides an innovative, data-driven solution for rapid structural assessment and protective design in blast engineering.