Analysis of Aerodynamic Indices for Racing Sailing Yachts: a Computational Study and Benchmark on up to 128 CPUs.

Abstract

This work presents a feasibility study for trustable and affordable CFD analysis of aerodynamic indices of racing sailing yachts. A detailed reconstructed model of a recent America’s Cup class mainsail and asymmetrical spinnaker under light wind conditions has been studied using massive parallel RANS modeling on 128 CPUs. A detailed comparison between computational and experimental data has been performed and discussed, thanks to wind tunnel tests performed with the same geometry under the same wind conditions. The computational grid used was of about 37 millions of tetrahedra and the parallel job has been performed on up to 128 CPUs of a distributed memory Linux cluster using a commercial CFD code. An in deep analysis of the CPU usage has been performed during the computation by means of Ganglia and a complete benchmark of the studied case has been done for 64, 48, 32, 16, 8 and 4 CPUs analyzing the advantages offered by two kind of available interconnection technologies: Ethernet and Infiniband. Besides to this computational benchmark, a sensitivity analysis of the global aerodynamic force components, the lift and the drag, to different grid resolution size has been performed. In particular, mesh size across three orders of magnitude have been investigated: from 0.06 million up to 37 million cells. The computational results obtained here are in great agreement with the experimental data. In particular, the fully tetrahedral meshes allow appreciating the beneficial effect of the increasing of the grid resolution without changing grid topology: a converging trend to the experimental value is observed. In conclusion, the present results confirm the validity of RANS modeling as a design tool and show the advantages and costs of a large tetrahedral mesh for downwind sail design purposes.