A GPU-Accelerated Full-Wheel Multi-Stage Turbomachinery Flow Solver

Cascade will validate the CPU-based moving mesh solver using NASA stage 37, high-pressure-ratio stage of an axial core compressor developed by NASA in late 1970rsquo;s. Cascade will port its moving version ofnbsp;the large-eddy simulation (LES) flownbsp;solver charLES to GPU-accelerated architectures. The moving mesh solver uses the same Voronoi diagram-based meshing strategy as the static mesh solver, however the meshing is now integrated with the solver to allow local regeneration of the Voronoi diagram when points are in relative motion. The current implementation for traditional architectures uses a conservative space-time formulation that allows for complex motions including collisions and full contact. Since complex motions and thus complex solver treatments are not required for the relatively simple solid-body rotational motion of turbomachinery, the development will be staged by first porting the Voronoi point search and cutting algorithms to the GPU, and simply re-cutting the interface cells in each time step, and updating the communication pattern. The entire algorithm can remain fully explicit, utilizing essentially the same solver as the static charLES for accelerated architectures. In regards to specific architectures, the static accelerated charLES is written in both CUDA and HIP, allowing us to leverage both NVIDIA and AMD accelerated architectures. Verification will be performed by comparing the GPU and CPU implementations in two stages. First, a comparison of the geometric data and operators (e.g. geometric conservation, volumetric fluxes and gradient operators) will be conducted on a mesh containing a moving disk part. Second, a comparison of the flow variables will be made on canonical flows (e.g. Euler vortex, 1D acoustic wave and solid body rotation) with and without a moving disk part. The moving solver calculations will be validated by comparing results against a direct numerical simulation of a rotating sphere.