NEURAL NETWORK-BASED PREDICTION MODEL FOR WALL PRESSURE SPECTRUM OF UNDERWATER VEHICLE

The primary source of hydrodynamic noise of the underwater vehicle is the pulsating pressure on the surfaces exerted by the turbulent boundary layer. During the early design stage, it is essential to evaluate the flow and acoustic performance of the underwater vehicle. Regulating the energy of wall pulsations can significantly reduce the self-noise and hydrodynamic load of underwater vehicles. The traditional water tunnel experiments are costly and complex. To reduce the prediction time, this study proposes a neural network-based method to predict the pulsating pressure spectrum on the surfaces of the underwater vehicle. The proposed neural network-based method relies on limited test data and is based on the SUBOFF, using steady flow field results as input parameters and experimental data as output parameters. The boundary layer information was extracted from steady CFD results and the neural network-based prediction model is trained using a total of 168 pieces of water tunnel test data. The comparison between the predicted results and experimental measurements demonstrates a high level of accuracy within the range of incoming flow velocity from 0 to 6.8m/s and Reynolds number ranging from 5.68 × 10⁵ to 3.86 × 10⁶. This neural network-based prediction model can be effectively utilized for designing low-noise underwater vehicle.