It would appear that the problem is inherent to the manner in which the fluid simulation works.
Similiar to problems you would encounter if the domain were to be completely filled with fluid,
it seems that positioning the inflow object in the scene, such that it becomes enveloped by
fluid also causes problems.
I re-created your scene in the Mantaflow fluid simulator (blender code implemetation by sebbas,
and based on the framework created by Nils Thuerey and associates- the same person who created the
original elbeem fluid simulator for blender), and it had the exact same issue.
I don’t want to mangle the explanation by sebbas as to why this happens, so I’ll just post the
conversation I had with him (and lucaRood) in #blendercoders:
gimble: sebbas: I was trying to re-create the issue that DavePSB was having with the elbeem fluid simulator, and it seems that mantaflow reacts the same way -> https://blenderartists.org/forum/showthread.php?410338-Fluid-Inflow-Fails-To-Fill-Simple-Tank-Depending-On-Position-In-Tank
gimble: sebbas: The mantaflow version is here: http://pasteall.org/blend/index.php?id=44369
gimble: sebbas: I was wondering if I set something wrong, or is this just an issue to be avoided?
sebbas: gimble: ah yes, I see what you mean
gimble: sebbas: The inflow rate noticeably changes once the inflow object becomes “submerged”
sebbas: gimble: i am not sure why elbeem behaves like that but for manta i know the reason:
sebbas: the thing is that the inflow object only refills cells with new particles
sebbas: so once the liquid level matches the inflow objects level, the inflow object has no cells to refill
sebbas: and hence it stops adding new particles
sebbas: so yes - I’d say this is a case you should avoid
gimble: sebbas: Ah, ok. hmmm, so “just the way it is” sort of situation I suppose? Not sure about why he set it up like that, and I can’t really think of any solutions to overcome that situation…
gimble: sebbas: thanks for the details
LucaRood: sebbas: Are you talking about particle sims or voxel sims?
LucaRood: sebbas: cuz you talked about “cells”, and later said it stops adding “particles”.
sebbas: LucaRood: I just meant the FLIP particles that manta creates (you dont see them but manta uses them to create the mesh)
LucaRood: sebbas: Ah FLIP… that explains it
gimble: sebbas: elbeem uses something similiar (or similiar explanation why it happens there)?
sebbas: gimble: i’d say elbeem does the same (just a guess …)
LucaRood: sebbas: What integrator is manta using for FLIP?
sebbas: LucaRood: uff not sure …
LucaRood: sebbas: Ok, np.
LucaRood: sebbas: I asked about the integrator, because depending on what it uses, it would be trivial to fix the submerged inflow issue.
sebbas: LucaRood: ah ok - to fix the problem I think it should also be possible to explicitly add new FLIP particles at each step (so not just refilling empty cells)
sebbas: I think there is an option for that in manta
LucaRood: sebbas: Yeah, but the issue is that if you add new particles within an existing fluid mass (filled cell), the simulation tends to explode when using most explicit integrators. Hence the integrator question
sebbas: LucaRood: its using IntRK4
LucaRood: sebbas: Ah yeah, that is a explicit, though it is a fourth-order integrator, so should not be too “explosive”
sebbas: LucaRood: ah okay - so something like second order would be much more “explosive”?
LucaRood: yes
LucaRood: with rk4 you might be able to get away with just artificially adding particles, if the inflow isn’t too high.
So, if I interpret what sebbas said properly, the problem stems from the manner in which the inflow object
adds fluid to the domain: if fluid hasn’t moved away from the inflow object’s volume due to gravity,
initial velocity, or some other external force, it sees the fluid voxel as being full, and no additional
particles/fluid is added.
It appears that there is a possible solution, but it might lead to “exploding particles/fluid” if the
inflow rate is set too high. Then it becomes an issue of if this is implemented in the Mantaflow fs,
and if it would also be ported to the elbeem fs…
In the mean time; what can be done?..
It seems that the scene setup is somewhat “artificial”, in that you normally don’t see fluid materializing
in the middle of a tank out of thin air. Usually there would be a pipe into the tank either from above or
from below the tank. From above is no problem, but from below is. In that case you would have to resort to
some trickery or perhaps route the pipe from above the tank to enter from below the tank; keeping the
inflow object situated above the tank.
Another option would be to use SPH fluid particles and then use the CubeSurfer addon to mesh the result.
I tried a quick mockup, and it seemed to work, at least on a coarse level, though it is also subject to the
“exploding particle” isssue if you’re not careful.