Session: 08-04-01 Deep Learning based ROM for Wave Dynamics and Prediction in Marine Engineering

Paper Number: 128032

128032 - Predicting Wave Propagation for Varying Bathymetry Using Conditional Convolutional Autoencoder Network 

The propagation of underwater noise in the ocean is a complex and dynamic phenomenon that is heavily influenced by the intricate geometry of ocean bathymetry. It has long been a challenging endeavour to develop generalised physical models capable of accurately predicting acoustic transmission loss in a wide range of oceanic conditions. In this context, we present a novel solution - the Range-Dependent Conditional Convolutional Autoencoder Network (RC-CAN), a deep learning based model designed for the real-time prediction of far-field underwater acoustic noise for varying bathymetry.

The RC-CAN model, being data-driven, provides a distinct advantage by relying solely on data, regardless of data acquisition methods. By learning a latent representation of the input ocean geometry, RC-CAN effectively reconstructs far-field acoustic signal transmission loss distributions across the provided ocean mesh. We have tested the RC-CAN model on a two-dimensional ocean environment with varying bathymetry, our findings highlight RC-CAN's exceptional ability to learn fundamental factors affecting acoustic signal transmission loss, including the spreading of sound waves, refraction, and reflection from ocean surfaces and floors. 

The implications of this research are enormous. RC-CAN's ability to capture the complexities of ocean acoustics holds great promise for real-time far-field underwater noise prediction. This innovation has the potential to revolutionise marine vessel decision-making and online control strategies, allowing for informed and timely interventions in the ever-changing underwater auditory landscape.

Presenting Author: Indu Kant Deo UBC

Presenting Author Biography: I am a Ph.D. student at the University of British Columbia (UBC) in Vancouver, Canada. My research interests lie at the intersection of physics-based machine learning, deep learning, and scientific computing. Presently, my research is focused on developing deep-learning based solutions for underwater acoustic sound propagation phenomenon.

Authors:

Indu Kant Deo UBC
Akash Vankateshwaran UBC
Rajeev Jaiman UBC