ShipNet: A Geometric Deep Learning Surrogate for Real-Time Ship Hydrodynamics
Accurate prediction of hydrodynamic performance is essential for ship design, yet high-fidelity simulations remain too computationally expensive for large-scale design exploration and optimization. In this talk, we present ShipNet, a geometric deep learning surrogate that predicts hull pressure distributions and free-surface wave patterns directly from three-dimensional hull geometry and operating conditions. By combining point-cloud representations with graph neural networks, ShipNet learns physics-relevant geometric features and enables near real-time hydrodynamic predictions. We further investigate how transfer learning can improve generalization across different ship families, reducing the need for extensive simulation data when extending to new vessel types. Finally, we present ongoing work towards the hull optimization using an active learning framework, based on a gradient signed-distance-function autoencoder. Through examples involving both yacht and dredger hulls, we demonstrate how surrogate modelling, transfer learning, and active learning can be combined to support the development of data-efficient hydrodynamic prediction tools for a wider range of ship designs.
Topic: CWI-SC seminar DAMEN/MARIN
Time: Dec 3, 2026 11:00 AM Amsterdam
Join Zoom Meeting
https://cwi-nl-zoom.zoom.us/j/88401859098?pwd=VtF0eyBfDqxZckwHrjKqNPrCtcz610.1
Join and take notes
https://cwi-nl-zoom.zoom.us/j/88401859098?pwd=VtF0eyBfDqxZckwHrjKqNPrCtcz610.1&mynotes=on
Meeting chat link
https://cwi-nl-zoom.zoom.us/launch/jc/88401859098
Meeting ID: 884 0185 9098
Passcode: 154578