[lhcamilo]

Hello everyone,

there is this itch I have been trying to scratch and hopefully it is just me doing something dumb.

in any case, as a background I have been performing LES simulations on this channel and I have been calculating the production and dissipation. And upon calculating the dissipation and production terms of the Reynolds stress equation in the spanwise, vertical and streamwise directions and compering with a DNS database, I realised that something was wrong. The streamwise and vertical components seemed switched and the spanwise component did not match well with the reference.

After mulling over it for some time, I decided to take a look at each term individually. And this weekend I decided to take a look at how the Mean velocity gradients matched with the reference.

The Mean velocity I used,,UMean, was averaged through the flow after a rather long development period. So in order to calculate the components of the mean velocity gradient I computed the Tensor with

fvc::grad(UMean)

For this simulation I used a fixed pressure gradient of 1 and a viscosity of nu = 1/Re_tau for Re_tau = 180. The plot data is being produced with a modified version of the postChannel utility. The margenta color is the reference.

gradx.jpg

grady.jpg

grady_ref.jpg

gradz.jpg

The black crosses represent the data from extrating the components XX,XY and XZ from fvc::grad(UMean), the green line represents the data obtained by calculating each scalar element directly and the magenta line is the DNS reference.

Those gradients have been computed as a post-process, but the results are consistent with a on the fly computation.

Additionally fvSchemes is set as:

gradSchemes
{
    default         Gauss linear;
}

So, am I doing something really dumb or there seems to be something off with fvc::grad()?

Note. This was tested in openFoam 2.0.x,2.2.x and 2.3.x

I should also point out that the Mean velocity profile seems to be in very good agreement with the reference, so for all intent and purposes, the issue does not seem to be the UMean field.

[hjasak]

try

default cellLimited Gauss linear 1;
If this does not help, AND YOUR MESH IS REASONABLE, there is garbage in your velocity field.

Hrv