[gregorv]

I am trying to model 2D wing profiles with simpleFoam without using wall functions as I am mostly interested in laminar wing profiles. I am currently trying to compare the NACA 23015 profile (not a laminar profile in itself, of course) computations at 0 AOA with experimental data at the Reynolds number of 260000. I also tried with other profiles with very similar results.

I am using the Spalart-Allmaras turbulence model, with the value of nuTilda at both the inlet and the wing itself set to a small number (10^(-8)) in dimensionless units. I played about with combinations of inlets and outlets and the calculations are very insensitive to that.

The problem I have is that (using the liftDrag tool as recompiled for 1.4.1) drag is overpredicted by a factor of more than 2. The pressure and viscous contributions to drag are roughly the same in size. I understand that the liftDrag tool can be used as-is for my problem seeing that the turbulent viscosity should drop to 0 on the wing itself?

To create the mesh, I am using an own made tool to create scripts for gmsh, and the meshes I use are regular (hexahedral) for the distances well outside the boundary layer then unstructured outside. I checked with paraFoam that the velocity takes a number cells to change from the value of 0 to the freestram velocity and that the velocity in the first cell is practically 0 (well, blue in paraFoam ;) ). Could there nevertheless still exist artificial viscosity due to the mesh that bumps the drag in my case to a larger than expected value? Also, the aspect ratio of the cells just next to the wing is naturally quite large (at the order of 20), is that a problem for the boundary layer calculations and the cause of my error? The results do change somewhat with further refinement of the boundary layer discretization, but this is to the order of 10% and nowhere near the factor of 2 needed. The solutions always converge nicely and give very plausible looking flow fields.

I would be glad to hear from anyone who has experience obtaining successful drag predicitions for laminar profiles.

[nuovodna]

i think that liftDrag tool doesn't work very well. can anyone post a corrected liftDrag tool version ?

[msrinath80]

Hi Emanuele,

I am now very curious. What exactlydo you mean by saying that the liftDrag tool does not work very well? Isn't the utility merely calculating the total force (pressure + viscous) along the direction of interest. Just how many ways are there to calculate this force? My point is that if your velocity and pressure fields are time and space accurate (in comparison with whatever metric [say experiments]), then the forces calculated from these primitive variables should be accurate as well?

[gregorv]

Nah, I believe liftDrag is doing its job correctly, it's really just a very simple summation code.

I did some further calculations and noticed that the pressure drag results are quite sensitive to the accuracy of the div(phi,U) scheme. Using

div(phi,U) Gauss linearUpwind Gauss cubic

setting gives much better results, and drops the pressure drag to below the viscous drag in my example. Now I am about 50% off in my drag estimates still. With the higher order schemes, however, the discretization does not affect the results much (less than 5% with a double refinement at the boundary in both directions). I will keep on digging and post my findings.

[msrinath80]

I agree with Gregor. I've done some preliminary validation for steady laminar flows (search the main forum). I know that the utility is calculating the forces accurately. I am still waiting for my other unsteady validation to finish. Once it is done I will confirm that the liftDrag utility is indeed correctly written. Please note that all my comparisons are being done for laminar flows only. I do this to ensure that I do not bring in the uncertainties associated with modeling (e.g. when dealing with turbulence etc.).