[fede0122]

Hi everyone!

I'm studying the compressible effects of a coolant (Novec 649) when passing through a simple grid. I was wondering why when I use a single cross (see Photo 1) I can get convergence with velocity inlets close to 25m/s (before the sim diverges). The moment I use the scaled-up version, I can only achieve 10 m/s. The dimensions of the first test are 4x4x215mm, the grid being 1.5mm thick. For the second one, the dimensions are 50x25x215mm, the grid is also 1.5mm thick. (Both have a 15mm buffer at the inlet Z=0 to Z=15mm before the grid.)

On paper, the difference in result shouldn't exist. Both should have relatively similar behaviors meaning reaching Mach 1 approximately with the same inlet velocity.

Thanks a lot for any guidance!

I´m using these settings:
Mesh Type: Trimmed Cells
# of Cells 30k+ for Full domain and
Space: 3D
Material: Gas -> NOVEC 649
Flow: Coupled
Equation of State: Real Gas
Real Gas Equation of State: Peng-Robinson
Time: Steady
Viscous Regime: Turbulent
Reynolds Averaged turbulence: (K-Omega or Spalart Allmaras)

For Boundary conditions I have:
Static Temperature 375K
Pressure: 281MPA

[sbaffini]

Well, the problems might be everywhere (are meshes identical except for a scale factor?), but in general you clearly aren't solving two identical problems, and the fact that you only refer about dimensional quantities suggests that you haven't considered the non dimensional numbers of the problem, which would clearly suggest the two problems are not the same.

Originally Posted by fede0122

On paper, the difference in result shouldn't exist. Both should have relatively similar behaviors meaning reaching Mach 1 approximately with the same inlet velocity.

This sounds aneddotical to me.

But let get things clear: does the first case actually converges, like to 1e-6 or better? Because, from the way you wrote your post, it seems that both cases diverged, just the smaller one took more time/iterations

[fede0122]

Originally Posted by sbaffini

Well, the problems might be everywhere (are meshes identical except for a scale factor?), but in general you clearly aren't solving two identical problems, and the fact that you only refer about dimensional quantities suggests that you haven't considered the non dimensional numbers of the problem, which would clearly suggest the two problems are not the same.

Okay, so the meshes... I say are "identical" considering that I've used the same mesh settings, the same refinement mechanisms (volumetric control, the 2 blocks have the same location). I'm a beginner in CFD, I don't know what non-dimensional issues my simulations might be causing. Boundary conditions, initial conditions... don't know what else?

Originally Posted by sbaffini

But let get things clear: does the first case actually converges, like to 1e-6 or better? Because, from the way you wrote your post, it seems that both cases diverged, just the smaller one took more time/iterations

Yes! The residuals converge to 1e-5, others at 1e-6... (except TKE which does at 0.001). However, they only converge when the fluid stays below Mach 1... meaning I get convergence @Mach=0.9xxxxx.

What I can't wrap my head around is why the fluid reaches Mach=1 @26m/s with the individual cross (walls of the tunnel = symmetry plane) ... but the full-sized grid (50x25) reaches Mach 1 at much lower speeds... only 14m/s

Thanks for the reply, I really appreciate it

[LuckyTran]

Let's say you repeated the mesh exactly in whichever direction to truly make both cases geometrically identical and also assuming you have the exact same initial condition and what not and it truly is identical in every sense.

When you increase the extent of the computational zone, you also increase the available number of length scales in these directions. A multigrid solver that aggressively accelerates convergence will use these larger length scales, otherwise it would take you ~80x more iterations to converge to the same tolerance. The errors and their associated length scales are not the same so you can not expect the evolution of the solution to be the same iteration for iteration across the two grids. It can certainly lead to situations where you converge for a single grid cross but don't converge for an array, or vice versa.

All of this is just to say, it happens. Don't worry about it. Just do the things you normally would do to get a case to converge.