GPU-Accelerated Full-Wheel Large-Eddy Simulations of a Transonic Fan Stage

Abstract

A single-stage transonic axial-flow fan is simulated using graphic processing unit (GPU)-accelerated wall-modeled large-eddy simulations. The computational meshes are generated using Voronoi diagram-based approach. Two types of mesh are employed for grid sensitivity study. The first type uses isotropic hexagonal close-packed (HCP) seeding whereas the second combines the background HCP mesh with anisotropic boundary-layer type of mesh around the blade surfaces. The effect of the rotor tip clearance is examined using a rotor-only configuration. The sweep of the performance curve at full rotation speed is carried out for both the rotor and the stage. The computational results are compared against the experimental measurement and excellent agreement is observed when adequate near-wall resolution is employed. The code performance of the GPU-accelerated solver is compared to the CPU-based version. The high throughput of the GPU solver significantly reduces the computational cost. On a mesh with 84 million control volumes, 10 rotor revolutions can be computed within 1 day using 20 GPUs.