Analysis and rapid prediction of underwater vehicle hydrodynamic loads during water-exit integrating CFD with GA-LSTM

Abstract

The water-exit process of underwater vehicles, characterized by trans-media dynamics, strong nonlinearities, and fluid-structure coupling, is significantly influenced by ocean environmental conditions. The efficient prediction of associated hydrodynamic loads has emerged as a critical focus in both engineering practice and theoretical research. This paper aims to achieve a high-precision, rapid prediction of water-exit loads. Therefore, an efficient computational methodology based on Computational Fluid Dynamics (CFD) and a GA-LSTM strategy was proposed to predict the macroscopic loads during water-exit under coupled wind-wave-current fields. Optimal Latin Hypercube Design (OLHD) has been used to systematically execute batch numerical simulations of various combinations of sea state levels, initial speeds and marine environmental parameters. This approach generated a comprehensive sample set that spanned the whole parametric space of dynamic load responses. The integration of the Genetic Algorithm's (GA) global search capability with the time-series analysis of Long Short-Term Memory (LSTM) networks has enabled the development of a highly effective predictive model for unsteady trans-media loads. A comparative analysis between model predictions and CFD simulation results demonstrates that this neural network approach enables rapid and accurate prediction of the variation patterns of external loads acting on the underwater vehicle under wind-wave-current coupled field, trained on high-fidelity CFD data. This methodology provides a robust theoretical and methodological underpinning for the simulation of trajectories and attitude dynamic for underwater vehicles during trans-medium process.