Investigation of wake patterns of moving disturbances using convolutional neural networks

Kelvin waves result from the interaction between ship and surface waves and represent a distinctive phenomenon. Understanding their patterns is of utmost importance for coastal protection, as these waves can persist for extended periods. The characteristics of Kelvin waves are influenced by the speed and lengthscales of the ship, which have been extensively studied through hydrodynamic simulations. However, recent advancements in machine learning offer an alternative approach to explore their underlying physics. In this presentation, I will share a preliminary investigation into the wave patterns generated by moving disturbances, utilizing convolutional neural networks (CNNs). To simulate this physical system, we employ a high order spectral method that solves the Zakharov equations. To create a comprehensive dataset, we consider a wide range of values for the length scales, distributions, and propagating velocities of the moving pressure. Subsequently, we train a residual neural network (ResNet) using this dataset, and the ResNet is employed to predict the speed of the pressure disturbance. The results demonstrate the ResNet's prediction ability with impressive accuracy and robustness against random noise, highlighting the significant potential of CNNs in advancing our understanding of Kelvin waves and their dynamics.