Stuck? 
support@mecway.com
support@mecway.com
Mecway
FEA
Self-supporting paddling pool
Trying to model a self-supporting pool, both empty (should collapse) and full (should stand). I have tried a segment, but frictionless supports to impose circular symmetry cause failure. Tried axisymmetric, says line2 elements not allowed. Tried full revolve, have problems with central elements. There must be a simple way to approach this? Attempts attached, material is a fibre-reinforced polymer sheet, properties I am kind of guessing at for now. Can't help thinking non-linear is better, but I don't think I can use hydrostatic pressure. This is only for interest/learning, not vital, so don't spend ages on it. Some simple suggestions welcome. Thanks.
Howdy, Stranger!
It looks like you're new here. If you want to get involved, click one of these buttons!
Comments
Revolving creates badly shaped elements at the axis of rotation. Correct them by merging the duplicated center nodes and using Mesh tools -> Correct collapsed elements.
The constraint issues are because CCX has a limitation that requires all constraints on the same node to be parallel or orthogonal to each other. Frictionless support makes constraints in arbitrary directions so when they're combined with other constraints, they're sometimes incompatible. Here, it should be possible in principle to rotate them to make that work but that special case isn't implemented in Mecway's .inp export. I got around all that by using a full quadrant so they're naturally orthogonal and specifying the constraints explicitly so they're exactly orthogonal.
Configurations only contain different suppressed states and solutions. So you can't have different meshes or analysis settings in them.
The real thing has an inflatable sausage that provides additional rigidity to the top lip plus some support from bouyancy. I think we could have pressurized chamber but bouyancy is probably beyond us, other than specifying upward forces on some nodes.
Victor, in your reply you said it was non-linear, but the analysis was actually linear. Non-linear gives me all sorts of red flags, including the non-following of the hydrostatic that you pointed out. I'm guessing a slip of the fingers from you.
Anyway, we got some sort of resemblance to real life, and I learned a little bit. I'll try some other shapes, vertical sides and sloping out, but I think they'll prove inferior.
I'd be skeptical of any results from linear analysis since the displacements when it's collapsed will be much larger than the rule of thumb of half the wall thickness. Non-linear will also allow you to include contact, though it'll be a bit of trial and error because you have to anticipate which pairs of surfaces might come into contact. You're right that Mecway doesn't allow you to ramp components of the hydrostatic pressure load in a quasi-static analysis and CCX has no "native" hydrostatic pressure so it's not really practical to do by hand.
Buoyancy comes for free with hydrostatic pressure so that should be easy.
Yep, I left it set to linear, sorry. But you can switch to non-linear and solve without making any other changes. Just keep it set to the CCX solver which can do shells, unlike the internal non-linear solver.
I attach two files because they solved slightly differently for the full config. The only thing that I knowingly did differently was that for non-lin2 I solved all configs simultaneously. There was such a delay before any progress showed, I opened a second window and did it one config at a time. The simultaneous solve eventually got there, and - unlike the other one - there is no error reported in the solution (hydrostatic not being a follower). Displacements slightly different too. Poisson still at zero in these, that will be the next thing to try.
I notice the material's 100 MPa Young's modulus is in the circumferential direction (V) rather than thickness (W). Is that a mistake?
I can't reproduce the different solution you got in Pool non-lin2.liml (58.42 mm displacement magnitude on "full" configuration). Regardless of whether I used Solve all configurations or Solve, it still gives 42.79 mm displacement magnitude.
There's a misleading issue I just found with the time step size. If you have a value set, like you do here, even with Quasi-static turned off, then it uses that as the initial time step size. That slows down the solver by making it start searching with the very small value. So it's better to set it to zero (automatic), but doesn't affect the solution.
I put 0.3 Poisson's ratio in and the Full configuration was very similar but the Empty one changed to a symmetric deformation shape with very small displacement. It looks the same as your 2nd screenshot but it hasn't actually folded up because the display scale factor is high. I suspect it needs to buckle to fall down so you may need to add an imperfection in the shape or load to initiate that.
The Young's modulus thing is definitely my error, I'll try to understand how that happened. The non-reproducable thing may have been me too.
With Poisson at zero, the empty solution was very unsymmetrical. I tried Poisson 0.45, and the empty solution was similar to the static 3D, collapsed and passing through itself. I tried CONTACT (as you said, trying to anticipate which surfaces would meet), and it seemed to do nothing. I cranked up the slope value quite high, and tried putting in values in the friction fields, which I believe are not needed. I tried also swapping slave/master. I read in another discussion that midside nodes in slave elements can be an issue, but not sure if that's relevant. I was using faces, I will experiment to see if nodes or whatever makes a difference.