A Novel Artificial Intelligence Approach for the Rapid Hydrodynamic Prediction of Manta Ray-like Underwater Vehicles

The manta ray-like underwater vehicle exhibits excellent maneuverability and bio-affinity. To enhance the development efficiency of this vehicle and reduce the costs of numerical simulations and experiments, we propose a novel artificial intelligence-based hydrodynamic prediction method called Hydro-DDPM. This model can rapidly and accurately generate time-series curves of drag, lift, and pitch moment variations within a single swimming cycle by inputting motion control parameters such as flapping frequency and amplitude. Hydrodynamic data are generated through a combination of Gaussian noise diffusion processes and neural network-based reverse denoising processes, with motion parameters integrated into the neural network model using a self-attention mechanism. Utilizing the IB-SGKS algorithm, 180 sets of hydrodynamic parameters were initially obtained and subsequently augmented to create a dataset comprising 3600 samples. The results indicate that the model performs well in multi-variable time-series predictions, with maximum prediction errors not exceeding 5% and overall errors within 3%. Compared to traditional CFD methods, our approach generates a set of hydrodynamic data in just a few seconds, achieving a 10³-fold increase in efficiency. Furthermore, this method is versatile and can be applied to predict the performance of various underwater vehicles, including those with pump-jet propulsion, rotorcraft, and gliders, by training the model with different datasets.