Incompressible low Re computations

einsfliegen · June 4, 2016, 04:04

Hi guys,

I'm currently trying to compute an diagonal pump using Fine/Turbo with incompressible fluid under low Re number. Due to it's incompressibility and low Re number, I'm having a hard time procuring convergence. The global performance is very unstable, with a pattern of constant vibration and sometimes it diverges. How can I improve the stability and convergence of my computation under these circumstances? I need some tips on smoothing residuals and procuring convergence under low Re and incompressible.

Here are some of my parameters used:
Reynolds number: 147 ~ 884
Ref. Velocity: 0.03275 ~ 0.1965 m/s²
Hakimi Preconditioning Beta: 6000
Sweep scheme definition: Increase on higher level
CFL: 3

Thanks in advance for any help! Would really appreciate it.

DarylMusashi · June 4, 2016, 07:18

Dear Einsfliegen,

It has been observed in many cases that the convergence rate may be significantly improved by performing a higher number of sweeps on the coarsest levels. Instead of the the standard 1-2-16 scheme you can try a higher number instead of 16.

Firstly, convergence problems may occur if your grid has a poor quality. Can you preclude this? Which y+ values do you achieve when inspecting one of your solutions?

Secondly, which initialization method do you use? When using constant values you should think about if the chosen values are as close as to the final solution as possible. You could try to use as initialization a fully converged solution of a 111-computation (leaving all boundary conditions unchanged) instead. Often convergence is better on coarser grids.

Some words to beta*:
Such a high value increases the dissipation and therefore alters the precision of the solution. I just want to make sure you are aware of that. How did you estimate the value of 6000?

Kind regards
Holger

DarylMusashi · June 4, 2016, 07:44

Dear Einsfliegen,
The naming of your files is SA_Blut_* - which means you use Spalart-Allmaras as turbulence model? If your estimation of the Reynolds-number is right the flow is certainly laminar, which means you don't need a turbulence model.
Do you yet know something about the flow field or do you expect turbulent regions in your model, so a turbulence model is necessary?
At least you could give it a try and do a computation without a turbulence model for comparison.

Kind regards
Holger

einsfliegen · June 4, 2016, 08:25

Thanks Daryl for the tips. Will definitely try it out. The beta I used are based on trial and error. I tried to use the table provided by the tutorial in Numeca, but it still results on divergence. So I increased it. As for grid quality, the grid I used results in convergence for the computations I made with higher Re-Number. Difficulties appears only on low Re Number.
I used a turbulence model because I assumed that there will be turbulence due to the high speed rotation (RPM: 8000 1/min) of the impeller of my pump, but I will definitely give laminar model a try!

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Physics of an incompressible fluid	Jonas Holdeman	Main CFD Forum	87	April 22, 2019 10:40
Incompressible turbulence models: strange implementations?	AleDR	OpenFOAM	3	November 18, 2014 11:23
Incompressible simulation	brugiere_olivier	SU2	2	April 15, 2014 10:12
Incompressible, Low Re RSM	ivan_cozza	OpenFOAM Running, Solving & CFD	0	June 9, 2009 05:08
Multicomponent fluid	Andrea	CFX	2	October 11, 2004 05:12

June 4, 2016, 07:18		#2
DarylMusashi Senior Member Holger Dietrich Join Date: Apr 2011 Location: Germany Posts: 174 Rep Power: 15	Dear Einsfliegen, It has been observed in many cases that the convergence rate may be significantly improved by performing a higher number of sweeps on the coarsest levels. Instead of the the standard 1-2-16 scheme you can try a higher number instead of 16. Firstly, convergence problems may occur if your grid has a poor quality. Can you preclude this? Which y+ values do you achieve when inspecting one of your solutions? Secondly, which initialization method do you use? When using constant values you should think about if the chosen values are as close as to the final solution as possible. You could try to use as initialization a fully converged solution of a 111-computation (leaving all boundary conditions unchanged) instead. Often convergence is better on coarser grids. Some words to beta*: Such a high value increases the dissipation and therefore alters the precision of the solution. I just want to make sure you are aware of that. How did you estimate the value of 6000? Kind regards Holger

June 4, 2016, 07:44		#3
DarylMusashi Senior Member Holger Dietrich Join Date: Apr 2011 Location: Germany Posts: 174 Rep Power: 15	Dear Einsfliegen, The naming of your files is SA_Blut_* - which means you use Spalart-Allmaras as turbulence model? If your estimation of the Reynolds-number is right the flow is certainly laminar, which means you don't need a turbulence model. Do you yet know something about the flow field or do you expect turbulent regions in your model, so a turbulence model is necessary? At least you could give it a try and do a computation without a turbulence model for comparison. Kind regards Holger

June 4, 2016, 08:25		#4
einsfliegen New Member Yi Fei Lim Join Date: Apr 2016 Posts: 2 Rep Power: 0	Thanks Daryl for the tips. Will definitely try it out. The beta I used are based on trial and error. I tried to use the table provided by the tutorial in Numeca, but it still results on divergence. So I increased it. As for grid quality, the grid I used results in convergence for the computations I made with higher Re-Number. Difficulties appears only on low Re Number. I used a turbulence model because I assumed that there will be turbulence due to the high speed rotation (RPM: 8000 1/min) of the impeller of my pump, but I will definitely give laminar model a try!