Calling experienced Elmer users ...

I am struggling some to learn to use the CFD capabilities of Elmer. Once I get there I will explore convection in a heatexchanger.

I started with a base case - flow over a step - as in the examples section, but in 3d instead of 2 and with air from the builtin material library.

The mesh I used - at least the one I started with - was used to run the test case in code-saturne which reasonably quickly converges to reasonable and steady solutions for different inlet velocities, using the k-e turbulence model.

I cant get the system to converge to a stable and plausible solution. The only modification to the mesh that had a certain stabilizing effect was when I eliminated the boundary layer mesh.

I calculate incompressible. I also tried with different fluid (water) with no real benefit. Leading me to suspect solver settings is my challenge. I use relaxation down to 0.5 and tried to see the effect of a time based approach.

I still dont feel confident. Is there a good reference for learning to use Elmer for fluid jobs, ideally a case comparable to the one I do (the step) here but 3d?

Another thing that puzzles me is that if I want the pressure to be absolute, I expect to do so by applying an absolute external pressure (e.g. 101300 Pa as atmospheric pressure) on the outlet and velocity at the inlet, and let the solution determine the pressure at the inlet. When I do this Paraview plot the pressure distribution as negative values. This leads me to suspecting that application of boundary definitions (and application of turbulence models for that matter) is fundamentally different than code_saturne. On the other hand the descriptions in the documentation does not indicate that, so I think that the results are completely out of the physical world, so to say.

Why not just stick with code_saturne you may ask. This is because in the end I will need a coupled analysis way beyond what can conventiently be done with code_saturne.

## Learning to use the Navier-Stokes solver

### Re: Learning to use the Navier-Stokes solver

One experiment is a transient simulation to see what happens.

Length of volume is 0,5m and velocity is 1 m/s The video linked to below shows what happens with velocity. Individual frames are spaced 0,2s.

https://u.pcloud.link/publink/show?code ... dfzHo3z5hV

Apparently something goes out of control due to a reflection at the bottom which is a No-slip wall BC or maybe originating at the outlet which (in this run) was a kind of collimator (only velocity in direction 1 was unconstrained).

In the second plot below (same run) the pressure is color coded and scaled velocity vectors shown, all in the middle plane only:

https://u.pcloud.link/publink/show?code ... FXVm4WjYzX

Flow seems to develop in a reasonable way until unstability.

Length of volume is 0,5m and velocity is 1 m/s The video linked to below shows what happens with velocity. Individual frames are spaced 0,2s.

https://u.pcloud.link/publink/show?code ... dfzHo3z5hV

Apparently something goes out of control due to a reflection at the bottom which is a No-slip wall BC or maybe originating at the outlet which (in this run) was a kind of collimator (only velocity in direction 1 was unconstrained).

In the second plot below (same run) the pressure is color coded and scaled velocity vectors shown, all in the middle plane only:

https://u.pcloud.link/publink/show?code ... FXVm4WjYzX

Flow seems to develop in a reasonable way until unstability.

### Re: Learning to use the Navier-Stokes solver

OK, so the instability in the transient run was (of course) courant number related, i.e. shorter timesteps required.

Here is the history with 10m/s at inlet, air, timestep 0.5ms and the 51 video frames are 2.5ms apart: https://u.pcloud.link/publink/show?code ... fWLfsI1rHk

Looks OK convincing to me. Now I need to find a setting for the solvers of Elmer do its magic and reach a similar solution preferable a bit quicker than direct timestepping.

If possible.

Here is the history with 10m/s at inlet, air, timestep 0.5ms and the 51 video frames are 2.5ms apart: https://u.pcloud.link/publink/show?code ... fWLfsI1rHk

Looks OK convincing to me. Now I need to find a setting for the solvers of Elmer do its magic and reach a similar solution preferable a bit quicker than direct timestepping.

If possible.