|
|
UComparison.png (81,070 bytes) |
|
|
kComparison.png (72,268 bytes) |
|
|
|
|
|
Why do you think it is necessary to set the normal stresses? These should be calculated by the turbulence model rather than the wall-function? Also if all the components of the Rw are to be set to tauw couldn't we just use Rw[facei] = tauw; The reason the individual components are set at the moment is to specifically avoid resetting the normal stresses. |
|
|
If you look at the attached test case results, the present implementation results are clearly wrong. Resetting also the normal components on the wall produces results that are plausible and very close to the kOmegaSST results. Neglecting the correctWallShearStress function altogether also produces axisymmetric results but the results are of course wrong. Thus, the non-axisymmetric behavior appears to be caused by the modification in the shear components in correctWallShearStress and it appears that to get rid of it the normal components have to modified also. Either in the turbulence model, correctWallShearStress or somewhere else. |
|
|
Rw[facei].xx() = tauw.xx(); Rw[facei].yy() = tauw.yy(); Rw[facei].zz() = tauw.zz(); are definitely not correct; tauw is the shear stress and does not include the the contribution from k. If we assume that the anisotropy may be obtained from dev(symm(gradUw)) then it would be possible to estimate the normal stresses consistent with those from the k-epsilon model by including the spherical contribution (2/3)Ik. |
|
|
Try this: // Calculate near-wall Reynolds-stress tensor Rw[facei] = ((1.0/3.0)*tr(Rw[facei]))*I - nutw[facei]*2*dev(symm(gradUw)); |
|
|
I added the spherical contribution as you suggested and reran the test case. I'll attach comparison images. The first cell velocity is now slightly higher than with kOmegaSST while with the previous modified version it was somewhat lower than with kOmegaSST. However, with the spherical contribution there is also a strange kink in the velocity and turbulence properties in third (!) cell from the wall. UComparison2.png (98,815 bytes) |
|
|
kComparison2.png (92,815 bytes) |
|
|
kComparison2-2.png (92,815 bytes) |
|
|
U_profiles.png (20,684 bytes) |
|
|
nut_profiles.png (24,104 bytes) |
|
|
k_profiles.png (23,885 bytes) |
|
|
It would be better if anisotropy of the normal stresses were set to correspond to the expected value for the log-region or better still the value the model would produce if the near-wall region were adequately resolved rather than inferring it from dev(symm(gradUw)) which is not consistent with either the models or measurement. |
|
|
Increasing the couplingFactor eliminates the third cell kink with the spherical contribution. First cell velocity returns to the same or lower as it was without the spherical contribution. k in the second cell increases significantly as the couplingFactor increases. |
|
|
According to the 1978 Launder-Gibson paper for near-wall turbulence u1^2/k = 1.1 u2^2/k = 0.65 u3^2/k = 0.25 with a bit of tinkering to get the normal, flow and lateral directions this could be used to set the normal stresses. |
|
|
I am running the test case and noticed that the laminar viscosity is VERY low: mu 8.9e-5; which with rho = 1017.1 gives a kinematic viscosity of 8.75e-8. What is the fluid? |
|
|
Dichlorodifluoromethane near boiling point at 26.15 bar. In noticed that you pushed a fix with just the tauw at the wall. |
|
|
Yes but it is not entirely consistent with the formulation of divDevRhoReff which should also use dev(R_) if the entire spherical part is incorporated in the pressure. I am running an extensive set of tests of the various formulations and so far the form you initially proposed is proving the most reliable at least for your case. |
|
|
Resolved by commit 6f4908ed000d223b797825b308a5a7ae6bbc8a63 |
- Anonymous
