Machine learning‐based damage classification of ultra‐high performance concrete slabs under contact explosions

Abstract The prediction of failure modes in ultra‐high‐performance concrete (UHPC) slabs is critical for guiding post‐blast operational planning and repair strategies for protective structures following blast events. This study addresses the requirement by investigating the application of machine learning (ML) methods to failure‐mode classification. Three representative failure modes, that is, crater, spalling, and breach, were considered, with a dataset of 638 samples compiled from field experiments and numerical simulations. The prediction performance of four ML models, including Support Vector Machine (SVM), Random Forest (RF), XGBoost and Backpropagation (BP) neural network, was evaluated using a randomly selected test set. Among the evaluated models, RF achieved the highest accuracy of 93.8%, while BP obtained the highest mean test accuracy of 91.7%. A comparison with existing methods highlighted the advantages of ML‐based prediction in enhancing classification accuracy at about 20.7 percentage points. Furthermore, the influence of key input parameters on model performance was analyzed, providing interpretability of the results and improving the reliability of the prediction framework.