Data-Driven Machine Learning Approaches for Compressible and Incompressible Fluid Dynamics Modelling

Machine learning has been extensively applied in various domains, including natural language processing, computer vision, speech recognition, and robotics. Recently, it has been increasingly applied to scientific and engineering fields, such as material science, medical science, and fluid mechanics. While the graph neural network architecture has been successful in several areas, its inability to account for long-range forces and heterogeneous dynamic data with physics constraints has made certain scientific fields impenetrable.

This project seeks to build on these advancements by developing a machine learning framework that can accurately predict fluid dynamics over a wide range of scenarios, including multi-physics behavior and fluid-structure interactions. The framework will be validated on small- and large-scale compressible and incompressible fluid flow simulations of the circulation on the Faeroe Shelf using the FVCOM model.