[camoesas]

Hey Felix,

Rereading my post you´re right its not clear enough. I extended the domain with no difference in the solution and convergence. And then I have changed PCG > GAMG and PCiCG > smooth solver. And I have got a better solution. But I could not reproduce this result. Must have been something else.

Now I ´ve build Up my case completely new (I have lost some files rm -rf * )
And now its working... Unfortunately I dont know what I have changed.

As you suggested I changed from 1. to second Order. In an first step just for U and then for k and omega. Of course convergence is getting worse and it takes longer to converge. The solution stops telling:

SIMPLE solution converged in 1176 iterations

Do you think I should run it a little bit longer. The residuals are still quite big.
Do I need second order for k and omega? The tutorial just uses Upwind.

I have tried a setup with wallFunctions because I wanted gather some experience. In my final geometry I wont be able to fully resolve boundary Layer.

I have also added my latest setup with that I achieve something like convergence.

Thanks for your help

Camoesas

[FelixL]

Good morning, camoesas,


well, well - the mysteries of computational fluid dynamics I'm glad it works now.

Yeah it's quite common that the residuals convergence rate decreases when increasing the accuracy of the interpolation schemes. Sometimes they even seem to stall. This is not that big an issue, though, because the residuals are just a hint for convergence - they don't neccessarily have to vanish, it's a question of their definition.
I'd monitor other integral parameters to check for convergence - in your case e.g. the pressure drop along the computational domain. If this is constant you have your convergence.

Regarding upwinding k and omega: "The chain is only as strong as its weakest link". So if you have a first order scheme somewhere, your overall accuracy degrades. Nevertheless, if you are unsure you can do a grid convergence check and see how it influences your solution.


Greetings
Felix

[camoesas]

HI Felix,

How could I monitor pressure drop along the domain? Do I have to write an expression or is there already a usable function.

Grid convergence is one of my next topics. Thanks for the reminder

camoesas

[FelixL]

Using the drag coefficient along the channel's walls should do the trick This should be constant in the end, too (for a steady state problem!).

Greetings,
Felix