Accelerating Convergence of RANS CFD Simulation of External Flows Using Surface-Vorticity Panel Solver

Reynolds Averaged Navier-Stokes (RANS) based computational fluid dynamics (CFD) simulations are widely used to measure aerodynamic and hydrodynamic forces on airplanes, cars, ships, and more. However, these simulations demand heavy computational power, often slowing down design and optimization processes. Past efforts to speed them up have ranged from tweaking pressure solvers to using machine-learning techniques. This study explores whether advanced potential-flow solutions can kickstart RANS simulations to make them faster. We proposed that initializing velocity and pressure fields with predictions from a surface-vorticity solver, FlightStream, could speed up RANS simulations for steady, incompressible flows. To test this, we developed tools to transfer FlightStream's flow data into a popular commercial RANS solver. For a 2D NACA0015 airfoil at a Reynolds number of 3.6x105 and angles of attack from 0 to 10 degrees, initialized RANS simulations converged 2.8X–5.3X faster than those starting from scratch, proving the idea works. In 3D tests with the NACA0012 airfoil under the same conditions, the speedup was even more impressive due to the larger scale. Depending on how we measured convergence, speedups ranged from 7X–17X for lift and drag with strict criteria, up to 43X–794X when we relaxed the standards. Surprisingly, not all initial conditions worked as expected. We tried two setups in 3D: one with both velocity and pressure from FlightStream, and another with a simple uniform velocity matching the inlet. The uniform velocity unexpectedly outperformed the more complex setup, and adding the pressure field didn't help much. We think this is because the RANS solver quickly figures out the pressure on its own, and FlightStream's data had some gaps from manual processing. For the Streamlined NACA0012, the uniform flow was already a close match to reality, giving it an edge. This uniform velocity boost was a pleasant surprise, but it might not hold up in trickier flows with sharp changes or turbulence, where it could stumble. Meanwhile, FlightStream's predictions, though imperfect, could shine in complex scenarios. These results show our method has potential, but there's more to explore, especially for tougher cases.