Line-Based Unstructured/Structured Heterogenous CPU/GPU Framework for Complex Aerodynamic Flows

A heterogeneous solution infrastructure using both CPU and Graphics Processing Units (GPU) is presented for threedimensional aerodynamic flows on overset meshes. A line based solver on unstructured grids on CPU (HAMSTR) is coupled to a structured RANS solver on GPUs (GARFIELD) through a lightweight Python-based framework. In HAMSTR, "linelets" through the meshes are found in a robust manner and the solver uses line-implicit schemes and high-order reconstruction schemes along these linelets, similar to a structured solver. This methodology is extended to pure unstructured volume grids, to enable solution in complex geometries. Mesh deformation techniques are implemented using a spring analogy to solve for deforming meshes. The GPU solver GARFIELD can operate in an overset environment and shows competitive speedups when compared to well-established structured solvers. The solvers are integrated through a mesh connectivity tool (TIOGA) and are individually demonstrated for cases such as hovering rotor and a rotor in forward flight. Aerodynamic flow past a rotor hub was simulated using the solution framework and the predictions in unsteady drag and spectral components of the drag agreed reasonably well with experiments and other CFD simulations.