hi, i am trying to simulate flow in curved duct. Flow is 0.7 mach, so i am using compressible, gradient based, solver and k-epsilon-realizable viscous model.

my problem is that k and epsilon residuals are not converging.

i tried to reduce the URF for these two from 0.8 to 0.4 and further to 0.4 but no improvement.

i also tried to use second order discretization for these two, but no improvement.

i am used unstructured grid with standard wall function and wall y+ is more than 300.

I am interested in the pressure profile at the exit and flow pattern inside the duct.

How to go about. thanks in advance.

can multigrid or increasing the mesh density will help?

Hi NISHIKANT TAMRAKAR,

Sure, increasing your grid resolution may help you in better converging your problem; you said as that your y+ surpasses the limit of 300, perhaps not only the walls, but also the rest of your domain needs more resolution to caption the general features of the flow.

If it does not work, perhaps you should investigate if your fluid flow is experiencing unsteady features, like vortex shedding or others: you can do solve this obviously by switching the unsteady solver.

And as you said, switching multigrid FAS scheme may also help: Try VCycles of k and e, with BCGSTAB smoother, and 2 pre-sweeps. But keep in mind that this will slow your machine and depending on grid resolution and computational resources may take quite a long time to solve.

Hope this helps! Good luck,

Freeman

can i try adaptation of mesh in fluent or i need to mesh it in gambit only? This is unstructured grid as i was not able to make structured grid with good boundary layer due to complex 3D geometry. Thanks

Hi again,

Yes, you can adapt directly in Fluent and, what is more, I think is the smartest decision, due to your unstructured grid. As your case study seems to be compressible flow, I would bet for a grid adaption by gradients of pressure.

Good luck with your sim,

Freeman

hi, thanks for your reply. i did the volume adaptation. initial cells = 439473 final cells = 584310 running realizable k-epsilon model with node base gradient and default relaxation and discretization.

got this error Error: 1000000: Failed to allocate 22.29MB memory (..\..\src\amginc.c:127)

Error Object: ()

my machine has 3GB RAM with Vista. what is the way out. Thanks and regards nishikant

Mmmm, that's quite disturbing... you should not have any problems with that machine you have and the cell size you're managing. I have run simulations up to 700.000 tetra elements in a centrino laptop 1.8GHz and 1Gb RAM: from 700.000 and more, the RAM got satured and needed to work from de Hard disk instead of the RAM, taking ages every iteration and having to stop it because it became unpractical to work like this.

But you see, although your machine could not manage that size of mesh, you wouldn't get any error message...

Sorry for not having solved your problem: I hope any other user will come here in your help.

Good luck and cheers,

Freeman

how to go about multi grid. please tell me some simple steps. Thanks

I am not very mastered at Multigrid method, but what I have tried sometimes after reading quite a lot here in CFD-online, is that switching to VCycles in k and e, with BCGSTAB smoothing (and, if required, some pre-sweeps -2 for instance-) is fine.

For this, go to solve->Multigrid. In this new window that appears, you see k and e multigrid discretization method: change it from "flexible" to "VCycle", and just in the same row, to your right, there are the AMG Smoothers: switch the BCGSTAB. Down the k and e, you can see another zone of the Multigrid window, where you can change the pre-sweeps, as I said above.

Hope it was clear enough. Good luck,

Freeman

[satyendra]

hi Freeman,

hi all,

i am doing analysis of a furnace in fluent 6.3(geometry modeled in gambit 2.24). i have made structured mesh in most of the volume except at the corners of the mould.

initially when the hot gases' velocity were low it was converging, but i was not getting required temperature inside mould. hence i increased the flow rate without changing the model.

But now except energy, all other residuals i.e continuity, k, x, y, z velocity and epsilon are fluctuating. following are the specs:
1. unsteady flow.
2. turbulent k-epsilon and RNG model
3. time step = 1

satyendra