Session: 08-05-01 Deep Learning for Predictive Modeling in Marine and Offshore Systems

Submission Number: 157155

Hypergraph Neural Network Surrogate Modeling for Prediction of Vortex-and
Wake-Induced Vibrations 

Vortex-induced vibrations (VIVs) and wake-induced vibrations (WIVs) pose significant challenges in marine and offshore engineering, contributing to risks of structural fatigue and potential failure in marine infrastructure. Accurate prediction and mitigation of these complex fluid-structure interactions (FSI) are therefore critical for maintaining long-term structural integrity. However, conventional analytical and numerical methods often struggle to efficiently explore large parameter spaces and develop effective strategies for mitigating flow-induced vibrations. This study employs a finite-element inspired hypergraph neural network (GNN) surrogate model to investigate the coupled dynamics of freely oscillating cylinders, a well-established benchmark problem in VIV and WIV research, with the aim of achieving accurate predictions of FSIs at reduced computational costs. The proposed methodology integrates high-fidelity numerical simulations based on variational finite-element methods with GNN-based surrogate modeling to forecast the spatio-temporal dynamics of VIVs and WIVs. The surrogate model utilizes graph-based representations to capture intricate fluid-structure interactions, incorporating a multi-layer perceptron for mesh displacement updates and a hypergraph neural network for fluid state predictions. Quantitative analyses demonstrate that the GNN model effectively learns from high-fidelity spatio-temporal simulation data, providing stable and accurate predictions of VIVs and WIVs while reducing computational overhead. These results highlight the potential of GNN-based approaches for advancing rapid FSI prediction and facilitating efficient parameter space exploration. This research supports the development of physics-based digital twins within the context of marine and offshore engineering.

Presenting Author: Shayan Heydari The University of British Columbia

Presenting Author Biography: PhD Candidate and Graduate Research Assistant at The University of British Columbia
Research areas: Computational mechanics, Fluid-structure interaction, Deep learning, Graph neural networks