A Machine Learning-based Tool to Correlate Coupled and Uncoupled Numerical Simulations for Submerged Plates Subjected to Underwater Explosions

In naval engineering, understanding underwater explosions is crucial for structural integrity and safety, particularly for combat ships. Coupled numerical analyses, which account for fluid-structure interaction (FSI), are accurate but computationally expensive and impractical for real-time applications. In contrast, uncoupled methods are efficient but overlook FSI effects. This study introduces a data-driven approach using a feedforward Deep Neural Network (DNN) to estimate FSI-induced displacements from uncoupled simulations. Trained on numerical datasets of blast-loaded plates with varying characteristics, the DNN predicts the coupled displacement field based on structural parameters of uncoupled simulations. Results demonstrate that this framework provides a fast and reliable alternative to coupled simulations, offering a practical engineering tool for underwater blast scenarios. This work serves as proof of concept that deep-learning-enhanced uncoupled simulations can replace coupled ones, with validity beyond the specific structure in the case study.