78th Annual Meeting of the APS Division of Fluid Dynamics (Nov 23 — 25, 2025)

V019: ExaWind: Revealing the dynamics of wind turbines with high fidelity simulations

This video examines the flow dynamics in a four-turbine wind farm. The data presented is from a large-scale geometry-resolved simulation of four five-megawatt wind turbines. These simulations were conducted with the ExaWind software stack. ExaWind integrates two computational-fluid-dynamics (CFD) solvers through an overset-mesh method: AMR-Wind, a structured-grid, adaptive-mesh-refinement CFD solver, solves the large-scale atmospheric boundary layer background flow, and Nalu-Wind, an unstructured-grid CFD solver, solves the flow around the wind turbine blades and tower. TIOGA is used for overset grid connectivity and solution communication at mesh overlaps. OpenFAST solves the nonlinear structural dynamics of the turbine blades and towers and provides forces and displacements for structural response that informs the fluid-structure interaction. The simulation visualized in this video includes four blade-resolved wind turbines operating in a turbulent atmospheric boundary layer. There are 500 million cells in the AMR-Wind solver, which is using 256 AMD GPUs of the Oakridge Leadership Computing Facility Frontier supercomputer. One Nalu-Wind solver instance is assigned to each wind turbine. Each turbine is discretized with over 13 million elements and solved using 448 CPU cores, for a total of 1792 CPU cores running four Nalu-Wind solvers. The cores on each Frontier node are allocated as 56 cores for Nalu-Wind and 8 cores for AMR-Wind operations on the GPUs. The ExaWind simulation is therefore using all available CPUs and GPUs concurrently. The visualization data results from the full flow field data of the simulation. To enable real-time flow visualization, the data is written to disk every 16 time-steps and lossy-compressed to a specific accuracy using ZFP. Custom tooling was developed to translate the compressed AMR-Wind data to a multi-resolution voxel grid and Nalu-Wind artifacts to blade and tower geometry. Blender is used to render the visualization, with the iso-contours and particle advection being performed with Blender's geometry node feature. NREL's Insight Center visualization cluster was used to compute the resulting content.