Rotor Wake Simulation via Vortex Lattice Methods on a Workstation Using GPGPU Accelerators

In contrast to an aircraft’s propeller the wake of a helicopter’s main rotor interacts very strongly with the rotor itself, giving rise to a number of problems that can only be addressed if the geometry and intensity of the wake is known with a high degree of accuracy. There are three largely different approaches to this issue: CFD-methods appear to be the most wholesome approach, however sufficient vortex preservation is today only achievable with ultra-fine meshes, resulting in very high computational cost that prohibit the everyday use as a design tool. Prescribed wake models situated on the other side of the cost spectrum use global inflow parameters to deform the wake. The third pillar is a vortex-lattice-method called Freewake. Here the shed vortices of the rotor blade are explicitly tracked and their induction on each other and the blade itself is computed using Biot-Savart’s law. They do not suffer from numerical dissipation of vortices, while still being very fast compared to full-fledged CFD methods. They are typically operated on small to medium sized clusters and are therefore subject to the associated administrative overhead. The Institute of Flight Mechanics of the German Aerospace Center (DLR) operates an advanced prescribed wake model which is capable of simulating most standard steady-state flight-conditions with a good degree of accuracy. This cheap and fast method is the reason why the institute’s Freewake code has been used sparingly in the past. That being said, many cases like transient maneuvers, flight in the turbulent atmosphere and dissimilar blades cannot be represented by prescribed wake codes and Freewake solutions are needed. The Freewake code of the DLR Institute of Flight Systems was developed in the 1990s for massively parallel computations on distributed memory systems. Research of rotor dynamics would greatly benefit from regular, cheap and easy use of Freewake methods. This would be possible if one could run the Freewake on a workstation and so do away with all the external costs and the cumbersome administration. This talk focusses on DLR’s effort to create a Freewake compliant to these restraints. Since the original development of the code computational power has increased exponentially, but not enough that the original code could be directly used with satisfactory running time. To reduce the computational cost sufficiently to bring the DLR Freewake calculation to a single workstation, the existing code was extensively revised. Apart from various optimizations, the existing MPI-parallelization was enhanced by an openACC branch to use GPGPU accelerators to further reduce the time to solution. The talk starts with general issues of rotor-aerodynamics and the wake description. Then the Freewake approach used and the performance gain achieved thus far are discussed. The lessons learned with bringing the problem to GPGPU are presented. Closing an overview over several future enhancements is given