Assessment of turbulence models for the transonic flow around the DLR-F4 wing/body configuration

Abstract

The transonic flow around the DLR-F4 Wing/Body configuration is simulated using the Baldwin-Lomax and Granville algebraic turbulence models, the Wolfshtein oneequation model and the Chen & Patel two-layer model, combining the Wolfshtein model near walls with the standard k — e model away from walls. Regarding the shock location, the best results are obtained with the Granville model, the models with transport equations being intermediate between the two algebraic models. Introduction One of the most important demands of industrial aerodynamics is the prediction of a 3D flow around a complete aircraft configuration or at least an assembly of its major components. Although the traditional algebraic models, including Baldwin-Lomax, are cheap and robust, their underlying physical assumptions severely limit the complexity of the flows which can be predicted with an acceptable level of accuracy. Reynolds-stress models, which do not involve the eddy viscosity hypothesis, have a high potential to provide simulations of complex 3D flows, due to their ability to capture anisotropy and the disproportionately sensitive response of the stresses to curvature-related secondary strains. However, they are penalized by a high cost, induced by the large number of additional equations, and poor numerical properties, which may affect the robustness of the code. Recently, very promising results have been obtained with the so-called non linear eddy viscosity models, in which the Boussinesq relation is generalized to a coordinate invariant quadratic relationship between stresses and strains to be coherent with the observation that any Reynolds stress is linked non linearly to all other stresses and strains. The big advantage with this class of models is that there is no longer a need for a six-component Reynolds stress closure. The first step towards those very attractive models is the implementation of a two-equation model based on the Boussinesq assumption for application to complex 3D geometries. This is already an advance on algebraic models since twoequation models are transport in nature and therefore can take account of history effects. The configuration investigated in this paper is the DLRF4 wing/body configuration, that has been studied experimentally in three different european wind-tunnels, with very high overall consistency of the experiments. The complete model (1.17 m span and 1.19 m body length) is representative of a realistic transport-type aircraft with a high aspect ratio (A = 9.5) transonic wing and an Airbus-type fuselage. The test case is defined as follows: = 0.75, a = Engineer, Aerodynamics Department Copyright ©1996 by the American Institute of Aeronautics and Astronautics, Inc. All rights reserved. 0.93°, Re = 3 10 (based on aerodynamic mean chord), transition being set at x/c = 0.25 on the lower wing and x/c = 0.15 on the upper wing (the fuselage is assumed to be fully turbulent). This parameter set corresponds to experimental lift and drag values (wing plus body) of CL = 0.602 and CD = 0.0352 and seems to be rather close to shock induced separation. The measurements include force and moment measurements of the whole configuration and chord-wise pressure distributions along 7 spanwise sections on the wing, ranging from TI = 0.185 to TI = 0.844, with each of them carrying 36 pressure holes. On the fuselage, there are another 44 pressure tubes. Since the experiments had not been performed with respect to code validation purposes, there are no boundary layer measurements available. Nevertheless, this experiment is of great interest for industrial code validation as it is close to reality and therefore is seen to be a basic test case to improve today's CFD codes. Four turbulence models have been considered: BaldwinLomax and Granville algebraic turbulence models, Wolfshtein* one-equation model and Chen & Patel two-layer k — e model. A special implementation allows their application to arbitrarily complex geometries and structured multiblock meshes. The calculations have been performed on a CRAY C98 computer. The majority of the loops have been written so to be both vectorized and parallelized (vector-concurrent mode).