Prediction of Water-Entry Trajectories and Hydrodynamic Characteristics of Projectiles Based on Machine Learning

Conventional approaches for predicting water-entry dynamics—including physical models and numerical simulations—suffer from computational inefficiency, limited adaptability to complex scenarios, and inadequate real-time capabilities. This study overcomes these limitations through a machine learning framework employing fully connected optimized neural networks (FCNNs). Leveraging CFD-Fluent simulations coupled with a 6-DOF model, we established a refined dataset of 162 multicondition trajectories enhanced by standardized preprocessing. Our dual-FCNN architecture achieved high-precision predictions, demonstrating maximum trajectory errors of 0.344% (training) and 0.525% (testing), while specialized networks for hydrodynamic forces in x/y directions improved gener-alizability. Compared to traditional methods, the framework effectively resolves nonlinear fluid interactions involving turbulence and cavitation while reducing computational costs by 82% and enabling real-time analysis. These advancements provide a transformative methodology for marine ballistics with scalability for operational deployment.