[wyldckat]

Hi Matt,

Many thanks for sharing the case! I think I found the reason why you have the outlet re-injecting flow back into the tank

Remember when we talked about mesh resolution? Well, there are two more details you should have checked:

1. Time resolution, but only in the sense of inspecting what was happening around the outlet. And what happens during the first second is extremely important, so you should have saved more time snapshots for the first second.
2. Resulting from the previous point, the conclusion is that the extension of the high resolution zone around the outlet is not nearly enough for the flow to properly stabilize!

I honestly have to say that this is another one of those crazy CFD cases I really like! Check the sequence of attached images!

So, what I did for examining this case, was:

1. Change the run time settings in "controlDict" to simulate only 1 second and to save every 0.01s.
2. Used the "Slice" filter to view the flow near the outlet.
3. Changed the representation to Uz and the scale limits to around +-0.095 m/s.
4. Applied the "Glyph" filter to the "Slice" entry and chose to display the glyphs as 2D arrows, with the "Scale Mode" off, the respective factor at 0.02 and unchecked the "Mask Points" option.

The explanation of what happened is as follows:

1. After 0.01s, the flow was already well oriented, after some pressure adjustment had already occurred.
2. At around 0.06s, a circulation centre appeared (aka a vortex) near the centre of the outlet boundary.
3. At around 0.15s, there is a clear shape inside the refined zone, which looks like a ball-reshaped vortex of fluid (imagine a conic vortex, constrained a bit at the top of the cone).
4. This shape is moving upwards, away from the outlet.
5. At around 0.25s, the centre of this ball-reshaped vortex hits the boundary between the high refinement zone and the lower refined zone of the mesh, near the outlet.
6. After that, at around 0.28s, the ball-reshaped vortex is changing shape, due to the loss of mesh resolution, leading the Finite Volume algorithm to do it's best to preserve the mass flow inside the cells. This makes the centre of the vortex completely change its dynamics and it's as if the refinement boundary became a flow bouncing wall.
7. At around 0.33s, the centre ball-reshaped vortex completely outside of the high refinement zone and has got a centre of upward flow of around 0.04 m/s!
8. At around 0.36s, the centre of the vortex has already managed to create a massive upward suction effect (>0.09 m/s) near the boundary between refinement levels, which has already lead the FV algorithm to propagate the suction effect from the centre of the high refinement zone!
9. At around 0.60s, the high suction effect that was felt inside the high refinement zone, has reached the boundary between the levels. And the awesome detail is that the centre inside the high refinement zone has already reached an upward speed of 0.985 m/s!
10. At 1.0s, the end of my simulation, shows that the maximum upward speed reduced to 0.935 m/s and is located above in the lower refinement zone.

And so, the reason for the problem is now explained! It's not the boundary conditions themselves to be blamed here, nor a problem with a high Courant number, nor even the schemes! The problem is the mesh refinement that is not respecting the requirements of the flow itself.


In a few minutes I'll upload the modified case to my Dropbox account, along with some images of the flow.

edit: The links to the files on my Dropbox:

• This is the slightly modified case I used for the diagnosis (6MB): https://www.dropbox.com/s/6wyycur7qa...ank_3D.7z?dl=0
• This is the full image sequence, with snapshots every 0.01s from 0.0s to 1.0s (25MB): https://www.dropbox.com/s/j8groc8skb...quence.7z?dl=0
• This is the Summary sequence, with 10 evenly spaced snapshots from 0.01s to 1.0s (4MB): https://www.dropbox.com/s/i1hgy31f7g...quence.7z?dl=0

Best regards,
Bruno