Structural Deformation Prediction of Underwater Exploded Ship Hull Girder Using Machine Learning

Abstract Near-field underwater explosions can cause substantial damage to offshore ship structures, presenting considerable risks to their integrity. This study focused on rapidly predicting girder structure deformation in ship hulls subjected to near-field explosions from small equivalent-weight spherical charges underwater. The Runge–Kutta discontinuous Galerkin method (RKDG) was employed to calculate the explosive load generated by the spherical charge. This load was then applied to the nonlinear finite element solver software, ABAQUS, to determine the maximum deformation of the ship hull girder structure under the impulse load. By comparing the results with experimental data, the accuracy of the proposed model was validated, confirming that the RKDG finite element coupling calculation effectively simulates the response characteristics of spherical charges in near-field explosion scenarios. Subsequently, two machine learning algorithms driven by data, namely extreme gradient boosting (XGBoost) and random forest (RF), were employed to dynamically predict the maximum girder structure deformation in ship hulls. The analysis demonstrated that both models successfully predicted the maximum deformation. The root mean square error for the XGBoost model (27.67) was lower than that of the RF model (50.31). The XGBoost model also fitted 96% of the training data, compared to 94% for the RF model. Moreover, the relative error of the XGBoost model (6.25%) was lower than that of the RF model (10.38%). Overall, XGBoost is highly suitable for predicting girder structure deformation in ship hulls subjected to underwater explosions.