GPU-accelerated CFD analysis of wind loads on box-type billboard structures using CUDA

Resumo

Billboards are a common feature in urban and highway environments, but their structural reliability under wind loads remains a concern. Many structural failures of billboard structures are observed every year, usually during weather events with high-intensity winds. Research indicates that recommendations found in standards around the world lead to underestimated wind load values for billboard structures, raising interest in further investigations in this field. A numerical investigation of steel billboards with a closed two-sided plate on flat terrain has been conducted with a numerical formulation based on finite elements and the explicit two-step Taylor-Galerkin model, used to solve the equations of fluid dynamics, considering incompressible and isothermal flow. The finite element meshes are made using linear hexahedral elements with reduced integration and hourglass control. Turbulence is handled with Large Eddy Simulation (LES) and a dynamic Smagorinsky subgrid model. The algorithm was implemented in CUDA-Fortran, utilizing the processing capabilities of the graphical processing unit (GPU) to speed up the simulation speed. The billboard model used corresponds to a full-scale billboard mounted at the Texas Tech University. Several meshes were constructed for different wind directions. The results are compared to full-scale measurements by other authors to validate the simulation. Mean and instantaneous velocity, and pressure fields, have been analyzed, alongside load and moment history. The GPU-enabled algorithm is compared with an equivalent CPU-only parallel algorithm, highlighting a speedup of up to 8.7 times compared to the CPU implementation.