malbrecht
Moin,

I assume you created your scene file in 16.5 - it threw some errors for me (I don't have 16.5). But anyway, if you want to use FEM for this, the most likely problem is that your geometry isn't divided into small enough tets. When you embed the tets geometry (“solidembed” node), the default values are good for starting off, but you often have to change the min/max scale values to match the simulation requirements. Try basing your tets on 1 and then set max to 1.2 and min to 0.2 or something like that - might already solve the problem. You can also try to enlarge the collision geo, but given the slow solving of FEM simulations, my personal preference is to work with original sizes as far as possible.

Also, collisions often appear due to motion that is just too fast, so you may have to increase your substeps to catch all penetrations (but with your image given above I assume you have to tweak the tets size first).

If you do not want to use FEM for this specific simulation (for performance reason) I am sure you could get good results from Houdini's cloth solver (which, as I understand, isn't exactly a FEM solver internally) OR even a PBD (position based dynamics) solver like “grains”.

Marc

P.S. / Edit - attached is a frame from a quick “grain” (particle based in this case) solver setup. No intersections. But (as to be expected with a particle based solution) quite hard to control in terms of “haptics”. Dramatically faster than FEM, but still a lot slower than specialized solutions (like syFlex etc).

Traceback (most recent call last):
  File "dynamics_popgrainswet", line 4, in <module>
  File "C:/PROGRA~1/SIDEEF~1/HOUDIN~1.268/houdini/python2.7libs\doppoptoolutils.py", line 650, in createSandSource
    landinnode = createSandSourceFromObject(origobj, style)
  File "C:/PROGRA~1/SIDEEF~1/HOUDIN~1.268/houdini/python2.7libs\doppoptoolutils.py", line 437, in createSandSourceFromObject
    0, 0)
  File "C:/PROGRA~1/SIDEEF~1/HOUDIN~1.268/houdini/python2.7libs\toolutils.py", line 357, in connectInputsAndOutputs
    connectMultiInputsAndOutputs(newnode, branch, inputs, outputs)
  File "C:/PROGRA~1/SIDEEF~1/HOUDIN~1.268/houdini/python2.7libs\toolutils.py", line 338, in connectMultiInputsAndOutputs
    outputs[i][2])
IndexError: tuple index out of range

malbrecht
Moin,

I assume you created your scene file in 16.5 - it threw some errors for me (I don't have 16.5). But anyway, if you want to use FEM for this, the most likely problem is that your geometry isn't divided into small enough tets. When you embed the tets geometry (“solidembed” node), the default values are good for starting off, but you often have to change the min/max scale values to match the simulation requirements. Try basing your tets on 1 and then set max to 1.2 and min to 0.2 or something like that - might already solve the problem. You can also try to enlarge the collision geo, but given the slow solving of FEM simulations, my personal preference is to work with original sizes as far as possible.

Also, collisions often appear due to motion that is just too fast, so you may have to increase your substeps to catch all penetrations (but with your image given above I assume you have to tweak the tets size first).

If you do not want to use FEM for this specific simulation (for performance reason) I am sure you could get good results from Houdini's cloth solver (which, as I understand, isn't exactly a FEM solver internally) OR even a PBD (position based dynamics) solver like “grains”.

Marc

P.S. / Edit - attached is a frame from a quick “grain” (particle based in this case) solver setup. No intersections. But (as to be expected with a particle based solution) quite hard to control in terms of “haptics”. Dramatically faster than FEM, but still a lot slower than specialized solutions (like syFlex etc).