A MULTI-SCALE NETWORK FOR THE PREDICTION OF HYDRODYNAMICS IN UNDERWATER LAUNCH

The prediction of cavitation and the hydrodynamic characteristics plays a significant role in the design of the underwater launched vehicle. In recent years, the artificial intelligence technology has become an important prediction method for these parameters. In order to quickly predict the dramatic changes of the bottom pressure in the underwater launching process, a multi-scale deep learning network is developed. This neural network model is based on a one-dimensional convolutional network (1DCNN) and established with an encoding-decoding network structure. The input data set is decomposed into a smooth part and fluctuating part through different sampling frequencies. A large-scale low-fidelity network and a small-scale high-fidelity network are trained separately to achieve the response and capture of different physical processes. Firstly, the bottom pressure under different launch conditions are obtained through numerical simulation, and the mechanism of bubble dynamics is constructed as a physical input data. Secondly, the data set is decomposed into two parts to train deep learning networks with two different scales respectively. Finally, two sets of output data based on two networks are integrated to establish a full prediction model. Testing and verification indicate that this newly developed multi-scale network can realize the fast and accurate prediction of the hydrodynamics of the underwater vehicle under various usual launch condition. The predicted bottom pressure curve during any stage, including the smooth stage, the transitional stage, as well as the frequency and magnitude of oscillation are consistent with the numerical simulation results. As a result, this method can provide a basis for the prediction of motion and trajectory of the underwater vehicle.