The ONERA M6 wing is a pretty tough case for flow solver validation - especially regarding the fact that the experiment was not setup to later deal as a validation case for CFD.

My own experience is that doing best practice CFD may not help. In our institute we figured out that changing the trubulence model can help (look out for: Eisfeld, Bernhard (2010) Reynolds Stress Modeling for Complex Aerodynamic Flows. In: Proceedings. 5th European Conference on Computational Fluid Dynamics, ECCOMAS CFD, 14-17 Jun 2010, Lisbon. ISBN 978-989-96778-1-4.). Another colleague (unfortunately unpublished) succeeded with this case by adding more than 100 millino grid points to the mesh. Me myself got better results by changing the mesh strategy (see: Wild, Jochen (2011) Smooth Mixed Meshes for Acceleration of RANS CFD in Aircraft Analysis and Design. 49th AIAA Aerospace Sciences Meeting, 4.-7. Jan. 2011, Orlando, FL, USA.).

Basically it is not the question if or if not a double shock is resolved but more where the two shocks merge into a single one at the wing tip. A slight variation in the spanwise position has a pronounced visibility especially at the station y/b=0.8. My experience is that structured type sub-meshes are better able to reproduce than unstructured (as long as no special anisotropic triangulations are used). Additionally wake resolution may play a significant role to get the spanwise lift distribution (and therefore induced flow velocities) into a better agreement as a prerequisite to match shock locations.

For me one of the big uncertainties is the real shape of the rounded wing tip. In the original geometric description of the test data this is not included. There are several geometries around for the wing end, but all are retro-fitted to my knowledge. I would expect that the shape has some influence on the strength of the outboard shock and therefore also on the position of shock merging. Also, several geometries are around with a closed (sharp) trailing edge instead of the measured blunt one.

By the way, for me the 6° AoA case seems evenmore interesting, since it includes the risk of a wrongly predicted shock induced separation.     

If you want to have more information about trouble on the M6 wing you may also take a look into the discussion forums on http://www.cfd-online.com/. 

In the attachment you can see the results of some sample computations, performed by my colleagues with different numerical godunov-type schemes. As you can see, some flow elements can be resolved with TVD schemes, while with other schemes they cannot. May be you need to use more advanced numerical scheme in your computations or increase a number of cells significantly.

Article Mass transfer in close binary stars

Pavel,

The Grid was very fine... To be exact the the elements crossed 29 Million... So I guess more advanced numerical scheme is to be deployed.

The ONERA M6 wing is a pretty tough case for flow solver validation - especially regarding the fact that the experiment was not setup to later deal as a validation case for CFD.

My own experience is that doing best practice CFD may not help. In our institute we figured out that changing the trubulence model can help (look out for: Eisfeld, Bernhard (2010) Reynolds Stress Modeling for Complex Aerodynamic Flows. In: Proceedings. 5th European Conference on Computational Fluid Dynamics, ECCOMAS CFD, 14-17 Jun 2010, Lisbon. ISBN 978-989-96778-1-4.). Another colleague (unfortunately unpublished) succeeded with this case by adding more than 100 millino grid points to the mesh. Me myself got better results by changing the mesh strategy (see: Wild, Jochen (2011) Smooth Mixed Meshes for Acceleration of RANS CFD in Aircraft Analysis and Design. 49th AIAA Aerospace Sciences Meeting, 4.-7. Jan. 2011, Orlando, FL, USA.).

Basically it is not the question if or if not a double shock is resolved but more where the two shocks merge into a single one at the wing tip. A slight variation in the spanwise position has a pronounced visibility especially at the station y/b=0.8. My experience is that structured type sub-meshes are better able to reproduce than unstructured (as long as no special anisotropic triangulations are used). Additionally wake resolution may play a significant role to get the spanwise lift distribution (and therefore induced flow velocities) into a better agreement as a prerequisite to match shock locations.

For me one of the big uncertainties is the real shape of the rounded wing tip. In the original geometric description of the test data this is not included. There are several geometries around for the wing end, but all are retro-fitted to my knowledge. I would expect that the shape has some influence on the strength of the outboard shock and therefore also on the position of shock merging. Also, several geometries are around with a closed (sharp) trailing edge instead of the measured blunt one.

By the way, for me the 6° AoA case seems evenmore interesting, since it includes the risk of a wrongly predicted shock induced separation.     

If you want to have more information about trouble on the M6 wing you may also take a look into the discussion forums on http://www.cfd-online.com/. 

Prediction of shocks in numerical simulations is actually a pretty tough subject, and generally the best approach is to generate a refined mesh where you expect to find the shocks and then let an algorithm do further refinement for you. Also, depending on your spatial discretization method, discontinuities may be severely smoothed, so a weaker shock won't appear at all. For the numerical issues related to this topic, I'd suggest the book by E. Toro, but as a rule, Riemann solvers tend to do better when combined with some form of shock capturing method. A 3rd order MUSCL scheme will give you excellent results, but may be extremely tough to converge. If you're feeling bold, try developing a higher order Finite Elements code to see the results. Of course, correct representation of the geometry plays a huge role as well, so maybe scanning a 3D replica might generate a better model. It's just far too sensitive a case to pinpoint the exact reasons of your question. Last but not least, if turbulence is involved, then I'd suggest using a non-Boussinesq approximation to model it, especially in the 3D case. Either that or, if you have access to a supercomputer, try going for DES or any Scale-adaptive model. Interactions between shocks and boundary layers are painfully difficult to resolve, so you want a high-fidellity model. Hope it helps a bit...