Predictions of the responses of stiffened plates subjected to underwater explosion based on machine learning

Underwater explosions can cause significant damage to ship structures, and quickly assessing the extent of the damage is crucial for improving warship combat capability. This paper proposes the use of machine learning algorithms to rapidly assess the damage of stiffened plates subjected to underwater explosions. The algorithms use structural responses of the plates obtained by numerical simulations, which are benchmarked by experimental results, as a database. Fractures and plastic deformations are both taken into consideration. The support vector machine algorithm is used to determine the criterion for fractures or plastic deformations, while a back propagation neural network model and a support vector regression model are both used to predict the plastic deformation and fracture area of the plates. The support vector machine model accurately classified different cases of fractures or plastic deformation with a training accuracy of 99.4%. The back propagation neural network model has regression values of 0.99 for predicting fractures and 0.97 for predicting plastic deformation, both of which are higher than those predicted by the support vector regression model (0.96 for the prediction of fracture and 0.90 for the prediction of plastic deformation). Therefore, the back propagation neural network model provides a more accurate assessment of damage to stiffened plates subjected to underwater explosions and can be used for rapid assessment.