[aerospain]

Hi everyone,

I am running a grid convergence study on a 2D square cylinder by refining four times an initial coarse grid (by refineMesh). All grids have been run by 5000 iterations and each of them, besides the first one, start from a mapped solution of the final iteration (5000) of the previous coarser grid (done by mapFields).

My surprise came when opening in paraView the last two solutions (pictures attached) and the finest grid gave an "unsteady" looking solution.

Mesh04 the first picture and Mesh05 the other one.

Any hints, please?

Thanks!

[linnemann]

We probably need a little more info such as schemes for example.

Depending on the Re the flow around a square will rarely be steady-state and you could simply have refined the mesh to such a degree that the numerics start behaving more transient since the fluxes wont be "damped" as much.

Its just an idea

[niaz]

Dear aerospain
you solved a case not steady state. simplefoam does not have problem. your test case physically is unsteady.

[aerospain]

Originally Posted by linnemann

We probably need a little more info such as schemes for example.

Depending on the Re the flow around a square will rarely be steady-state and you could simply have refined the mesh to such a degree that the numerics start behaving more transient since the fluxes wont be "damped" as much.

Its just an idea

Dear linnemann,

Thank your for your help. I was aware from the beginning that my "physical" problem is unsteady and the "computational" one as not capturing the whole extent of the physics.

But, as it is usually done in grid convergence studies, you first run a steady solution to define your mesh size, and the you run unsteady solutions to define your time step. Besides, I will need to compare steady solutions and time-averaged (unsteady) solutions in a couple of months.

What you mention about the fluxes goes along the lines of what I've been talking to some colleagues during lunch break.

I've decided to run two meshes in unsteady mode to check how the turbulence behaves and if I can observe anything "going wild".

cheers!

[ternik]

Originally Posted by aerospain

Dear linnemann,

Thank your for your help. I was aware from the beginning that my "physical" problem is unsteady and the "computational" one as not capturing the whole extent of the physics.

But, as it is usually done in grid convergence studies, you first run a steady solution to define your mesh size, and the you run unsteady solutions to define your time step. Besides, I will need to compare steady solutions and time-averaged (unsteady) solutions in a couple of months.

What you mention about the fluxes goes along the lines of what I've been talking to some colleagues during lunch break.

I've decided to run two meshes in unsteady mode to check how the turbulence behaves and if I can observe anything "going wild".

cheers!

Aerospain,
defining the mesh size based on a steady state solution results is O.K., but as you said, you do not achieve steady state solution (results) because your case is time-dependent! From that point of view, I think your approach is (might be) not O.K.
In your case, I would:
1. perform the numerical modelling of time-dependent flow;
2. do some "premature" modelling to get the impression on the time-step and mesh-size scale;
3. choose fine enough time-step, fix it and do the grid dependence study (for a given/fine enough time-step) using three consistently refined meshes;
4. once the "optimal" mesh (giving the mesh-size independent reults) is determined, use three (or more, if needed) consistently refined time-steps to show (prove) that the solution obtained on a chosen mesh size is time independent.

In addition, I suggest to go through the following paper

A.S.R. Duarte, A.I.P. Miranda, P.J. Oliveira. Numerical and analytical modeling of unsteady viscoelastic flows: The start-up and pulsating test case problems, Journal of Non-Newtonian Fluid Mechanics, 154 (2008), pp. 153-169.


where the similar "issue" is resolved.

Cheers,
Primoz

[aerospain]

Thank you Ternik,

Your step-by-step explanation asserts me on what I had in mind.

Just one comment, I've been able to produce (numerical) steady results by starting from zero time all the grids. Before, I was using the results of each coarser mesh as a restart point for the following finer mesh. This allowed for the disturbances present in the solution to build up and show the physical unsteadiness after the third finer grid.

Have a great summer!

C.