problem with DNS turbulent channel flow
 

Hi, all
I'm doing a research of turbulent channel flow with DNS. I referred to this thread " Everything you need to compute DNS in channel vs OF 2.1.0" by adding a source term in the UEqn representing pressure gradient.
I've uploaded my solver based on icoFoam. The problem I'm facing is the unsteady solving process. That is, as time goes on, the velocity at the centerline will decrease from my desired value, and this exist consistently along the solving process, I don't know how to deal with it.

Oh, my mesh is 64*128*64,the fluid is water, the channel is 12.56*2*6.28m, Ret=194, Re=2900, according to this , the ut=1.95358e-4m/s and the dp/(dx*rho)=3.82e-8m/s^2

The result is big so I can't upload, all the things except it are uploaded here, hope anyone can give me some suggestions.

ThanksAttachment 29088

Attachment 29091

Attachment 29092

Attachment 29093

I'm also a newbie in OF but I know that for a DNS simulation you need a finer mesh otherwise you'll not compute all the turbulent scale, hence the profile near the wall will be far from the real one. Did you try to refine the mesh and see if the results would improve or not?

From what i understood about OF on channel he will maintains the mass flow constant so the velocity profile is quite important, and depending on how the velocity develops near the wall the value at the center-line will change accordingly.

Unfortunately my knowledge on OF is quite scarce, if you found the culprit of the error please write it.

Thank you for your reply. I just adopt the same mesh as Kristofferson in his paper about rotating channel which is 128*128*128, while in order to make a quick watch at the calculation ,I chose a coarse mesh in the 2 periodic directions. So I think in the spanwise direction, the mesh is enough. While I will still try to refine the mesh in the spanwise to see if there is any improvement.

Also, I tried to use RAS to do a faster calculation, the residuals are also very large along the calculation as DNS. That's quite annoying~~

I'm having problem with the residuals too when I try simulating the backward facing step, in my case they will decrease like normal then suddenly they will increase and will have an oscillatory profile. I couldn't find yet the culprit.

Another thing that can influence your case is the schemes adopted try increasing the order and see if the computed solution change or not. In my case thank to changing schemes and the solvers in fvsolution i could increase the rate of convergence and eliminate the oscillatory behavior at the start of the simulation.

yes, the schemes using is important, there are so many schemes to chose. Now I'm installing the openfoam-1.6-ext in order to use the channelfoam. I want to see if it can do a different solution, hoping a improvement then :)