Hi guys
one of my colleagues took a look and gave me an important clue.
He hinted that the domain size plays an increasing role at high angles of attack. This is due to the fact that the wake region becomes larger, and imposing a constant pressure boundary condition right through a wake flow can introduce problems (because it is physically incorrect - the pressure isn’t constant). NASA has also been aware of this problem.
NASA made the domain so big that the airfoil cannot be seen at this zoom level. There’s another interesting page which mentions the effect of the farfield boundary. According to NASA the farfield boundary “can particularly influence drag and lift levels at high lift conditions”. Furthermore “with smaller farfield boundary extents such as 30 c, the boundary influence on the forces is greater (especially for drag)” (emphasis mine).
So we ran the 11.15 degrees (which was badly off) again with a bigger flow domain
This is still not up to the standard of NASA but it gives the wake enough space to swim away freely without interacting with the outlet BC.
We used comparable mesh settings (fineness 4) and to ensure we get roughly the same fineness as in the project with the incorrect results, we imposed a region refinement. We used the standard mesher with the same layer refinements, and the same boundary conditions. We tried both hex and pure tet meshing.
The hex mesh gave the following values:
forces forces1 output:
sum of forces:
pressure : (0.76146998137 0.000159678961654 110.118009483)
viscous : (0.203707401587 -0.00077078339203 -0.0322107856268)
porous : (0 0 0)
sum of moments:
pressure : (0.880791955197 -0.0873637693236 -0.0063370913922)
viscous : (-0.000224440531852 -0.00256230327931 -0.00155160496964)
porous : (0 0 0)
forceCoeffs forceCoeffs2 output:
Cm = -0.00103618864389
Cd = 0.0111214224598
Cl = 1.26848255642
Cl(f) = 0.633205089568
Cl(r) = 0.635277466856
faceSource faceSource30 output:
areaAverage(B1_TE1065) for U = (84.1164776596 -1.32189786544e-07 -1.18014982077)
areaAverage(B1_TE1065) for p = 0
areaAverage(B1_TE1065) for nut = 0.00114687930404
areaAverage(B1_TE1065) for k = 0.00382498200213
areaAverage(B1_TE1065) for omega = 3.32811613533
C_d = 0.01112 \\
C_l/C_d = 114.05
The tet mesh gives
Cm = -0.000659972381821
Cd = 0.0109396031182
Cl = 1.27931757233
Cl(f) = 0.638998813784
Cl(r) = 0.640318758548
These values are in good agreement with the values according to NASA.
Hannes
Hi @jprobst: your last post is a very good contribution to defining a ‘good practice’ for airfoils simulation we tried to assemble! From above presented results both HEX and TET meshes are performing is similar manner and CL/CD is making sense.
You did make whole domain bigger, but moving Outlet BC far enough (we need to discover what is enough) would be possibly good enough. Moreover, you did use the same BC and same BL (1 layer covering y+ > 80, patched airfoil face with denser mesh on that path), which makes me believe that proposed geometry tweak + mesh + BL + BC + minimal Numerics tweaking can be used to check more experimental results and be valid for TET and HEX meshes.
NASA document you cited ( NASA has also been aware of this problem.) shows relatively small variations of CD between 30c and 100c and hints about ‘farfield Point Vortex (PV)’ correction. Extrapolating that graph to 5c would not bring that 70% more CD, but in any case, relation is clear.
Remaining question:
NACA0012 simulations for AoA 3.95 - 9.99 (moderate wake) with small domain and TET 2019 mesh were already correct. TET 2020 mesh would invalidate them (CD will be much off). Do you have explanation , please?
I was personally starting to run sensitivity tests on the distance of the outlet from the airfoil last October. This is when I found out that the new TET algo is limited to a max BMB domain size of 1km.
That, aside, are you saying that the significant difference between Results of 2019 TET meshes and 2020 TET meshes may ‘disappear’ if we did the following :
Do a sim run on a 2019 TET mesh, that had a large enough distance to the outlet face.
Re-mesh that geometry with the same meshing parameters to obtain a 2020 TET mesh.
Do a sim run with the same BCs etc on the new 2020 TET mesh.
I can see that if (and that is a BIG if) the Results of the sim runs in Steps 1 and 3 are ‘close’ to each other, then there MAY be no reason to further question whether the 2019 TET meshes or the 2020 TET meshes give more accurate results…
If we agree that it would be worth it, I think I can do Steps 1 to 3 with a fair bit of effort…
.
.
.
The above aside, could you provide a link to each of the HEX and TET sim runs that you say seem to be in good agreement with NASA
Hi guys
sorry for the late reply. It’s been a busy week.
This is the link to the project: NACA0012 Airfoil study by Edoardo | SimScale
My opinion is that we can get good results for the airfoil with the current standard mesher, considering everything we found out in here. In summary
Keeping the boundary conditions, the boundary layer settings, numerics the same
Using a larger domain (in all directions) at least for the high angles of attack
I believe it is necessary to extend the domain in all directions because
In upstream direction we need to avoid interference with the velocity inlet BC
In downstream direction we need to avoid interference with the pressure outlet BC
And in the direction normal to the flow (“top and bottom”) because of the blockage effect of the airfoil
All of the effects mentioned above increase with higher AoA.
I agree with the procedure proposed by @DaleKramer.
Hex and tet should both give good results, although the hex will likely perform better.
Remaining question:
NACA0012 simulations for AoA 3.95 - 9.99 (moderate wake) with small domain and TET 2019 mesh were already correct. TET 2020 mesh would invalidate them (CD will be much off). Do you have explanation , please?
Just to help me understand, @Retsam are the results for low AoA with the current 2020 meshes incorrect? I can see in this post that the results for 11.15 AoA (relatively high) has been off (and I think we understand this now). Is it also the case for low AoA?
@jprobst: I’ve just re-run NACA0012 simulation for low AoA (3.97) with mesh 2020 and produced a table with results. I added also the run with Outlet set to ‘Mean Value’, which, for small domain, has real impact.
You can see that mesh 2020, with Fixed value Outlet gives CL/CD -6.73% compared to NASA experiment. Mesh 2019 was 1.81% off.
Using Outlet Mean value is providing much better value, but both CD and CL are bigger than in experiment.
Mesh 2020 is ~ 10% bigger and simulation takes twice the time (too many processors involved?).
In preliminary tests of 2020 mesh, I can confirm that making domain longer brings nothing. As you said, domain should be scaled up in two dimensions. Making it thicker is counterproductive as well.
Making it thinner was calling for a disaster, as strong mesh irregularities emerge again and cell counts go to the sky (making it only .8 cm thick instead of 2 cm was just a mistake from OnShape, but I learned the ‘no-go’ for that quasi 2D mesh).
Higher angles (> 12 AoA) still do suffer (CD too high), but I did not finish my research of best, commun conditions for NACA0012 simulation.
Hi @Retsam
Thanks for sharing those results! Hmm, that’s interesting. So the fact that the outlet boundary condition has an impact on the results should in principle be an indication that the outlet is too close. However you also state that “making the domain longer brings nothing” - this surprises me. I can also not give a stringent explanation why the “2019” mesh gave a better C_l/C_d than the “2020” one. It may have been a coincidence, but I feel that it’s quite unlikely.
I can understand why making the domain thinner will increase the element count drastically. That’s because the mesher will refine elements in order to keep their aspect ratio below a certain limit.
If a better explanation for the unexpected behavior of the lift/drag values comes to mind I’ll let you know.
Johannes
Here are results of moving outlet wall from 3.5 m to 9 m distance, still keeping Fixed Value BC. It was a surprise indeed, but there is no clear relation of CD or CL with that distance.
:
