Abstract
In marine engineering applications, a crucial demand exists for the accurate and dependable prediction of far-field noise emanating from marine vessels. Traditional full-order models relying on the Navier-Stokes equations prove impractical, and advanced model reduction techniques can be inefficient for reliable far-field noise prediction. Recent advancements in deep learning-based reduced-order models have demonstrated effectiveness, achieving speeds several orders of magnitude faster than full-order simulations through the utilization of convolutional neural network architectures. Despite their effectiveness, existing models encounter considerable difficulties when forecasting wave propagation over extended time horizons and making predictions for distant locations. This research endeavors to enhance the predictive capacity of underwater radiated noise in far-field scenarios by refining the network architecture and integrating bathymetry information into the neural network input. Reduced-order models utilizing deep learning often rely on auto-regressive prediction, lacking information about far-field bathymetry. To overcome this limitation, we introduce a new range-conditional convolutional autoencoder network, which incorporates ocean bathymetry data into the input. To showcase the efficacy of our range-conditional convolutional autoencoder network, we examine a benchmark scenario involving far-field prediction over Dickin’s seamount. Our proposed architecture adeptly captures the transmission loss over a range-dependent, varying bathymetric profile. The architecture can be integrated into an adaptive management system for underwater radiated noise while providing real-time end-to-end mapping between near-field ship noise sources and marine mammals.
