The printable version is no longer supported and may have rendering errors. Please update your browser bookmarks and please use the default browser print function instead.

Ahmed body

Application Challenge 1-05 © copyright ERCOFTAC 2004

Comparison of Test data and CFD

Experiments provide very detailed data that offer a particularly difficult challenge to CFD. They showed that the drag crisis experienced by the body around 25°-30° is related to a dramatic change of the structure of the wake. The low-drag configuration (35°) consists in a massively separated wake, which is quasi-2D, while the high-drag configuration (25°) consists in a very complex, 3D wake structure, with a reattachment of the flow on the slant part and a strong interaction of the bubble with intense corner vortices, which are very energy-consuming.

EXP1 shows that fixing a splitter plate in the wake of the body, in the symmetry plane, forces the flow to turn back to the low drag configuration (massively separated wake). The mechanism underlying these phenomena is not clear, but it could be due to the fact that the splitter plate counteracts a flapping of the wake in the span-wise direction. Therefore, there are some evidences that large-scale unsteadiness of the wake could play a crucial role in the wake structure transition. It could also explain high levels of turbulent stresses above the slant part that are very difficult to predict with steady-state RANS calculations.

It appears from all the CFD results that the wake structure of the low drag configurations (35°) is correctly reproduced by all the turbulence models tested. The correct trend of the drag coefficient with the slant angle is correctly reproduced (CFD1), but the correct level is not found. In general, since the wake structure is correct, the pressure levels on the slant part are realistic, but the exact pressure repartition on the slant part and the vertical base are hardly reproduced.

Concerning the high-drag configuration (25°), the great majority of the CFD computations were not able to reproduce the complex, 3D structure of the wake: a massively separated wake is obtained, which shows that the wake structure transition is missed. The number of computation and the variety of numerical schemes and meshes give many indications that the main issue is not numerical, but linked to the physical modeling: turbulence model and steady-state strategy. It appears that only two types of modeling are able to reproduce the structure of the wake: LES (CFD9) and low-Reynolds number Reynolds stress model (CFD13). It should indicate that the large-scale unsteadiness of the wake must be resolved (the potential of URANS has not been investigated extensively yet) or, alternatively, the absence of large-scale unsteadiness resolution must be compensated by a very refined turbulence modeling (Reynolds stress transport equations and integration down to the wall). However, these partial conclusions are only based on one LES and one low-Re RSM computation. Additional studies are necessary to confirm these favorable conclusions.

The paper of Florian Menter extracted from the Proceedings of the 10th ERCOFTAC IAHR Workshop (https://hal.science/hal-03037095) with permission, gives a further comparison of experimental and CFD results, including various figures. This paper can be obtained by clicking here.

Update added in 2024 by F.R. Menter

Since the ERCOFTAC workshops, simulations have progressed with an increased focus on Scale-Resolving Simulations. The simulations fall in two categories: Hybrid RANS-LES model and pure LES model simulations. Hybrid RANS-LES methods seem well suited for the Ahmed 25° car. They avoid the high cost of LES near the wall of the attached boundary layers. The ability of such models to predict the complex flow topology for the 25° case depends however on the ability of the underlying RANS model to predict separation from the slant onset. For a discussion of hybrid methods with application to this current test case see e.g. Guilminesau et al. (2020), Ekman et al. (2020).

The application of LES to the 25° case proved surprisingly difficult. Up to the publication by Menter et al (2024) no LES with acceptable accuracy for the exp. Reynolds number has been achieved. For a review of LES studies see Menter et al (2024). Of special interest is that simulations at artificially reduced Reynolds number were able to predict the correct flow topology with separation and reattachment on the slant, even on coarse grids. However, at the exp. Reynolds number, the simulations showed a similar behavior to RANS models. In one set of simulations, the flow stayed attached (like with k- type models) and in another set, the flow stalled (like with SST type models). The authors in Menter et al (2024) confirmed this observation, even for much finer meshes than used previously (e.g. a 560 million block-structured hexahedral mesh resulted in fully attached flow). Only after turning to Octree meshes, which allow a three-dimensional refinement towards the wall, could a sufficient resolution of the boundary layer be achieved to allow a reliable prediction of separation and reattachment on the slant. The following pictures are taken from Menter et al (2024).

Figure 2: Zoom of Octree meshes O1 and O2 near the roof-slant intersection.

Figure 3: Flow structure on roof-center plane for WALE O1(top) and WALE O2 (bottom) meshes showing contours of vorticity.

Figure 4: Wall shear stress on the roof-center plane for WALE- O2 and WALE- O1 in comparison with SBES/RANS solution.

Fig. 2 shows two Octree meshes near the roof-slant onset of the Ahmed car. The coarser mesh has 230 million and the refined mesh has 320 million cells. Both grids are formally of sufficient near-wall resolution for a wall-resolved LES (with ∆x^+=∆z^+≈35,∆y^+=1 in streamwise, spanwise and wall-normal direction respectively). However, the 320 million cell mesh (O1) has an overall finer mesh in the central part of the boundary layer. This results in a resolution of finer turbulence structures in the roof boundary layer as seen in Fig.3. The improved resolution brings the LES closer to the wall shear stress distribution (Cf) of the SST/SBES model which can serve as a reliable reference for the zero-pressure gradient flow in that region (Fig.4).

Figure 5: Velocity profiles in center plane for WRLES on O1 and O2 grids, compared to experimental data.

Figure 6: Stress profiles for streamwise coordinate in center plane for WRLES on O1 and O2 grids, compared to experimental data.

Both meshes produce highly accurate representations of the separation bubble on the slant as seen from the velocity profiles in Fig. 5. Included in the figure is also a simulation on the O1 mesh where the WALE model was deactivated in the entire domain, which resulted in an even slightly better agreement with the experimental data. Fig. 6 shows the corresponding profiles for the streamwise stress component, which are again in good agreement with the experimental data, in contrast to RANS models which strongly underpredicted the stress level. While there are acceptable velocity profile results available for hybrid models e.g. Guilmineau et al. (2020), also none of these simulations captures the correct stress-level, especially just downstream of the slant onset. This points to a high consistency of the depicted LES simulations.

Figure 7: : Flow topology on slant of Ahmed car. Left: experimental oil flow (from Ahmed et al 1984). Right: Octree O1 – no model simulation.

Figure 7 shows the flow topology for the O1 (no model) simulation compared to the experimental oil flow. As expected from the close agreement in the velocity profiles, the agreement in the flow pattern is also very close.

Figure 8: Q-criterion plots for Octree O1 no model solution. Large picture has

Q=5\cdot 10^{6}s^{-2}

and smaller picture has

Q=1\cdot 10^{8}s^{-2}

.

Finally, Figure 8 shows the resolved turbulence structures using the Q-criterion with a zoom to the slant onset region for the O1 mesh. As seen, this mesh allows for a very fine resolution of the turbulence which is necessary to accurately capture flow reattachment.

References

Ekman , P., Wieser, D., Virdung, T., Karlsson, M., Assessment of hybrid RANS-LES methods for accurate aerodynamic simulations. J. of wind Engg. And Industrial Aerodynamics, 206, (2020), 104301.

Guilmineau E., Deng G.B., Leroyer A., Queutey P. Visonneau M., Wackers J., Assessment of hybrid RANS-LES formulations for flow simulation around the Ahmed body. Comput. Fluids 176, 302-319 (2018)

Menter,F.R., Hüppe A., Flad D., Garburak, A. V., Matyushenko A.A., Stabnikov A.S., Large eddy simulations for the Ahmed car at 25° slant angle at different Teynolds numbers. Flow, Turbulence and Combustion, 112, 321-343, (2024).

Contributors: Remi Manceau; Jean-Paul Bonnet - Université de Poitiers. — Update (2024) F.R.Menter, ANSYS Germany ,