Reduced-Order Modeling of Ship Airwakes with Atmospheric Turbulence Effects using Dynamic Graph Networks

The interaction between the ship airwake and helicopter aerodynamics poses a challenge for the pilot workload and limits the operation range of the rotorcraft. The accurate characterization of this helicopter/ship dynamic interface problem requires the two-way coupling between high-fidelity aerodynamic simulation and rotorcraft comprehensive analysis. However, this approach becomes computationally intractable when realistic flow conditions, such as turbulent atmospheric boundary layer, are involved. This study proposes a novel reduced-order modeling (ROM) approach based on the message passing architecture of the graph neural networks (GNN). Instead of treating the entire flow field as combinations of modes as in conventional ROM’s, the GNN learns the interaction between the vortices in the flow field. This unique feature gives the GNN an unprecedented flexibility in the modeling of flow fields. The ROM is tested using vortex particle data synthesized from 2D snapshots of 3D CFD-based flow solutions. The results demonstrate the capability of the ROM to predict the time evolution of a vortical flow field and generalize to flow field patterns that are not present in the training dataset. The results establish initial capability and potential of the proposed ROM approach to be applied to the full-scale ship airwake problem and high-fidelity two-way coupled simulation of helicopter/ship dynamic interface modeling.