[thijs1909]

Dear all,

For a while I am struggeling to obtain a mesh independent solution using translational periodic boundary conditions for a straight 3d pipe, both laminar and turbulent. The goal is to invesigate by specifying a certain mass flow rate for a pipe with a helix structure (spiral offset) and study the pressure gradient for different helix geometries. For that I want to use periodic boundaries and only simulate a few pitches. But before that, I am struggeling to obtain an mesh independent solutions for a laminar and also a turbulent pipe flow.

My diameter is D=5,3mm and the periodic length is L=10mm. I have an incompressible fluid (water) and use the SIMPLEC algorithm. I notice that for a constant mass flow rate and mesh refinement (1,5x in every direction) I do not obtain a mesh independent solution for the pressure gradient. My mesh is structured (O grid butterfly topology) and low Reynolds. For a laminar pipe flow I do obtain a hagen-poiseuille parabolic velocity profile, but the pressure gradient is not converging. All residuals are nicely converging to 1e-5.

For momentum I use second order upwind, second order for pressure, second order upwind for turbulence, Least squares cell based for gradient.

Laminar periodic pipe flow:


Turbulent periodic pipe flow using RNGkepsilon:


I can not seem to find out what is the problem for the pressure gradient. I also checked for different partition setup, first I used metis decomposition and next I use decomposition along z-axis, so each processor only has two neighbors. It seem to influence the result, more osscilation, but still no assymptotic value.



My definition of a mesh independent solution is that with increasing number of cells, the value of interest converge to an asymptotic value.

Kind regards,

Thijs

[LuckyTran]

First, fits also my experience with turbulent flow in straight ribbed pipes.

However, I noticed that I was not achieving iterative convergence. That is, letting the simulation run for 100k or 1M iterations, the pressure gradient had very low frequency oscillations and the amplitude of these oscillations was more than the deviation in the pressure gradient vs mesh count.

I think you have to give up the mindset that you are looking for mesh independence & instead call it by its proper name, mesh dependence. That is, you want monotonically asymptotic behavior where it doesn't exist. You should recalibrate your thinking in terms of asymptotic +/-.

I see people running into these situations more often nowadays, even at the AIAA CFD benchmarking workshops.

[thijs1909]

Originally Posted by LuckyTran

First, fits also my experience with turbulent flow in straight ribbed pipes.

However, I noticed that I was not achieving iterative convergence. That is, letting the simulation run for 100k or 1M iterations, the pressure gradient had very low frequency oscillations and the amplitude of these oscillations was more than the deviation in the pressure gradient vs mesh count.

I think you have to give up the mindset that you are looking for mesh independence & instead call it by its proper name, mesh dependence. That is, you want monotonically asymptotic behavior where it doesn't exist. You should recalibrate your thinking in terms of asymptotic +/-.

I see people running into these situations more often nowadays, even at the AIAA CFD benchmarking workshops.

I contacted ANSYS again while it is a very strange observation for a Hagen-poiseuille flow. If this very famous problem cannot be solved, I think it is very alarming.

[Kushal Puri]

Originally Posted by thijs1909

Dear all,

For a while I am struggeling to obtain a mesh independent solution using translational periodic boundary conditions for a straight 3d pipe, both laminar and turbulent. The goal is to invesigate by specifying a certain mass flow rate for a pipe with a helix structure (spiral offset) and study the pressure gradient for different helix geometries. For that I want to use periodic boundaries and only simulate a few pitches. But before that, I am struggeling to obtain an mesh independent solutions for a laminar and also a turbulent pipe flow.

My diameter is D=5,3mm and the periodic length is L=10mm. I have an incompressible fluid (water) and use the SIMPLEC algorithm. I notice that for a constant mass flow rate and mesh refinement (1,5x in every direction) I do not obtain a mesh independent solution for the pressure gradient. My mesh is structured (O grid butterfly topology) and low Reynolds. For a laminar pipe flow I do obtain a hagen-poiseuille parabolic velocity profile, but the pressure gradient is not converging. All residuals are nicely converging to 1e-5.

For momentum I use second order upwind, second order for pressure, second order upwind for turbulence, Least squares cell based for gradient.

Laminar periodic pipe flow:


Turbulent periodic pipe flow using RNGkepsilon:


I can not seem to find out what is the problem for the pressure gradient. I also checked for different partition setup, first I used metis decomposition and next I use decomposition along z-axis, so each processor only has two neighbors. It seem to influence the result, more osscilation, but still no assymptotic value.



My definition of a mesh independent solution is that with increasing number of cells, the value of interest converge to an asymptotic value.

Kind regards,

Thijs

Sometimes with the refinement, specially in the structured mesh, quality of the mesh start degrading, that is also sometimes cause change in results with refinement. hope you have checked all the quality parameters for all the meshes and maintained same kind of quality in all meshes.

[thijs1909]

Originally Posted by Kushal Puri

Sometimes with the refinement, specially in the structured mesh, quality of the mesh start degrading, that is also sometimes cause change in results with refinement. hope you have checked all the quality parameters for all the meshes and maintained same kind of quality in all meshes.

Hi,

I want to reply that I fixed the issue. The problem was in the SIMPLE algorithm to simulate a translational periodic flow. Mesh refinement did not result in a converging solution for the variable of interest. In my case the pressure gradient for simple pipe flow.

I switched to a coupled solver, with the pseudo transient option and now I do get a converged solution for the pressure gradient in translational periodic flow.

It is not clear to why the SIMPLE algorithm is not working properly and I do not have an clear answer for that.

Notice that this result is just my observation and the used version of Fluent was v19.0 . With the coupled solver I also ran in a memory error:P But luckily running on multiple nodes fixed the issue.

Kind regards,

Thijs

[LuckyTran]

You pulled a Millikan and you got lucky. If I change enough settings and collect enough eventually I find a bunch of data points that on a curve I've already drawn.