Efficient damage prediction and sensitivity analysis in rectangular welded plates subjected to repeated blast loads utilizing deep learning networks

Abstract

The uncertainty in constitutive parameters significantly affects structural responses. This study examines the impact of these parameters on the damage to rectangular welded plates under multiple impacts using deep learning methods. A validated finite element model was used to generate a dataset by varying the constitutive parameters. Several surrogate models based on the Johnson–Cook models were compared for prediction accuracy. An attention-based neural network was applied for global sensitivity analysis of multiple-impact damage. The results indicate that models with attention mechanisms provide superior accuracy and efficiency for the damage of plate under repeated blast loading. Moreover, material parameters like density and yield strength are more influential under single impacts, while damage parameters become critical under repeated impacts. These findings offer insights for optimizing the safety of rectangular welded plates under varying impact conditions.