The NACA 0012 develops a LE separation bubble in this Re/alpha range (as noted in the attached paper) which might confound matters even more.

Sorry have forgot Reynolds No 360000

Interested, I shall follow the answers.

Check your boundary conditions .

the boundary condition is critical issue and has a lot of data to change for try and error kindly i sent the attached files to help you see the problem and advice

for area at reference what data should be :-

1- is Area = cord x span or what , remember cord =0.1524 span .91 ?

2- is length = cord or span ?

3- is the depth = cord or span ?

at the configuration of the problem shall i assign the upper and lower edges as symmetry or inlet velocity?

The problem with all the turbulence models in FLUENT is that they are not based on physics, including k-omega which is in its SST format maybe the best choice but still semi-empirically based. k-eps is, in a way, complete nonsense, at least for a physicist.

I include a theory which starts with Kolmogorov 1942, completing it to give Karman constant as 1/SQRT(2*Pi) ~ 0.399 as only ONE of the results.

It is applicable for moderate curvatures of the r.m.s. flow streamlines. There are challenges which I would like to solve with my theory like Princeton superpipe. I already have a sketch "how to" but need and a contract ...

Hi Derbaz, with your Re = 36,0000 you are close to \infty and shoult really try my theory, which is for exactly this case. For finite Re we elaborated a generalization wich I attach.

But sorry, all this is nothing for engineers to quickly apply it. The application of these advanced turbulence theories deserves realt dedicated work and only THEN the engineer may join the party.

Best regards,

Helmut

5 to 10 % variation between experimental data and computational data is allowed because experiment conditions are different then computation

thanks to all of you , the result from Ansys is Cd=0.2 as in text file attached but from the paper of sand80-2114 experimental result at wend tunnel shows Cd=0.02

but for Cl both experimental and Ansys the same Cl = 0.9 which is great

but right now i am trying to correct my meshing by point wise software , i have just discovered a great way to adjust my meshing annd i am doing such a equation to make enhancement of meshing according to the length so i will return to you again . thanks for of of you

I think that at this angle of incidence separation starts. Remember that the majority of the RAS turbulence models are calibrated in boundary layer flow cases for a developed flow without adverse pressure gradients (or mildly separated flow). Then, you should check several points. One of them if the mesh quality for the boundary layer flow. If you use wall functions and you have separation they don't work anymore. I also think that for "industrial flows", i.e., cases of complete turbomachinery or a complete 3D wing airplane CFD modeling, RAS modeling is still a good option for design engineering compared to more sophisticated ones (DES, LES or similar). E.g., for LES modeling you need (Re)^9/4 mesh points and to define a time step in accordance with the "characteristic velocity" of the cascade energy eddies' in order to define an accurate mesh and time step size for "filtering". For these requirements massive CPU computations are necessary, then RAS still is a good option, but it is necessary an accurate case-by-case calibration of they.

kindly find out the validation case at this paper

1- NACA 0012 at 10 degree rey. 360000 all data found for experiment

Hi Derbaz,

Before considering the details about how each turbulence model treats near-wall behaviour I would revisit the basics again and again. Boundary conditions have already been mentioned above, however in FLUENT you also have to set your reference values (e.g. freestream velocity etc...). Check those because the coefficients are worked out from them. Also ensure your coordiante axis are the ones you are analysing afterwards; for example you could be reading c_D in the spanwise direction instead of along the stream (I know it's trivial but sometimes these differences are due to small misses!). You might also want to double check how the Mach numbers compare; depending on your altitude/pressure/density the aerofoil might be experiencing some compressibility effects. If the previous is all fine I would have a look at friction coefficient distribution along the chord. As it's been said you could have separation at 10 degree AoA and different turbulence models as well as what numerical scheme is used to solve the differential equations will provide different answers (which doesn't inherently mean right or wrong by the way!). If your simulation is underpredicting separation there you'd have a factor for the disagreement. If all the above is still fine then perhaps you'd like to look at what turbulence intensity are you feeding at the inlet of your domain versus the experimental data or reference that you are using. In the latest versions of FLUENT you can choose the k-w model with SST and enabling the inttermitency transition option; it really isn't costlier in terms of computing time than standard k-epsilon, it has the advantages that k-w-SST brings and it really does help with regards to laminar-turbulent matters (it is recommended in the guide to set intermittency to 1 at the domain inlet; in this case ensure it is far enough upstram of the leading edge). Good luck!

It looks like the NACA0012 stalls at about 12deg, which means the separation point is moving further from the trailing edge as the angle of attack increases. If you don't match this in the CFD model, it will have an impact on the drag calculation. In addition, as others have noted, the correct turbulence mode is essential to correctly capture the skin friction and the behavior of the separation. Another check you can make is to compare your results against pressure coefficient data. There is a sharp suction peak at this AoA that you will need to capture in the model, and you can also use the Cp profile to help verify the location of the separation point.

Dear Amit, "5 to 10 variation between experiment and simulation ... is allowed because ....

I do not fully agree. Your statement might hold for all the empirical turbulence models and their applications. But my new turbulence theory (2005, 2013 ...) does not use a single observation nor other empirical data and nevertheless predicts (among other fundamental constants and features of high Re turbulence) the von-Karman constant as 1/SQRT(2*Pi) ~ 0.399. Observations give 0.40 p/m 0.05. I.e. the range accepted by observationalists is up to 13 %, yes, but my prediction is more or less exact for Re = \infty.

If you have questions regarding the attached topic, please dont hesitate.

The NACA 0012 develops a LE separation bubble in this Re/alpha range (as noted in the attached paper) which might confound matters even more.

Check the boundary conditions first, then solution method. Some time very very small mistake in the numerical solution causes large difference in the results.

Initially, I suggest:

- Check which area you are using to generate the coefficient - in the tab "reference values".

- Also check which speed component is being used.

this is an important factors correct me if I am wrong :

1- 2 D problem

2- depth = 1

3- length = cord

1- Area = cord x span

Derbaz Khalil raises a very good point. Most experimental models are designed with end plates at either end of the airfoil to represent an infinitely log span so that the section lift, drag, and moment coefficients may be calculated. This is done by having a small gap between the end of the airfoil and the end plate. If your CFD model is of finite span, you will have the tip vortex effects which could skew your results. I don't know the details of how the turbulence models handle the 2D situation, but I seem to remember that they typically require a 3D model. If you run a 3D model, you will need to minimize the end effects if a section lift/drag coefficient is to be calculated.