A Comparison of RANS and LES for Upwind Sailing Aerodynamics

Abstract

Yacht sails experience complex aerodynamics that are challenging to reproduce with numerical methods. The experiments of Fluck on an idealized upwind sail plan [1] [2] showed areas of flow separation and vortex structures that subsequent Reynolds Averaged Navier-Stokes (RANS) calculations have struggled to correctly simulate [3] [4]. Therefore it is of interest to see if other methods, such as Large Eddy Simulation (LES), are able to more accurately predict this flow. To this end the experiments of Fluck have been reproduced in Fluent, using RANS and LES. Both RANS and LES accurately model the attached flow on the lower region of both sails, but RANS fails in capturing the separation bubble occurring at the top section of the mainsail, predicting it as either too short, non-existent, or extended for the entire chord length. However, LES correctly models the leading edge separation bubble with its reattachment point located approximately halfway along the sail chord. The LES calculations are sensitive to the free stream turbulence intensity with an increase in the turbulence intensity from 3% to 12% shortening the separation bubble by half. Accurately reproducing the experimental geometry also improves the numerical results, with modelling the computational domain as a simple box or as the real wind tunnel results in different flow, with the angle of attack varying by 3°. In conclusion, LES, even if it is very sensitive to the mesh and the choice of the solver, is more accurate than a steady RANS calculation, although the computational time increases by a factor of 50.