Add FEniCS as a solid participant for elastic-tube-3d

[IshaanDesai]

This PR adds FEniCS as a solid participant to the elastic-tube-3d case.

[IshaanDesai]

THIS ERROR IS NO LONGER SEEN
Currently the results in FEniCS show an error. The edge of the tube at z=0.005 appears to not be constrained with zero displacement and actually moves as the neighboring nodes are displaced. This can be seen by placing a Point at (0, 0, 0.005) and observing the tube edge.

The fixed boundaries are defined in the following way:

tutorials/elastic-tube-3d/solid-fenics/solid.py

Lines 13 to 18 in cddd741

	def clamped_boundary(x, on_boundary):
	"""
	Filter nodes at both ends of tube as they are fixed
	"""
	tol = 1E-14
	return on_boundary and ((x[2] - 0.0) < tol) and ((x[2] - L) < tol)

and the clamped boundary is defined in FEniCS as:

tutorials/elastic-tube-3d/solid-fenics/solid.py

Line 65 in cddd741

fixed_boundary = AutoSubDomain(clamped_boundary)

and the constraint of zero displacement is applied in the following way:

tutorials/elastic-tube-3d/solid-fenics/solid.py

Line 77 in cddd741

bc = DirichletBC(V, Constant((0, 0, 0)), fixed_boundary)

Currently there is no reasonable explanation why the tube end nodes displace. @BenjaminRodenberg do you have an idea what is going wrong?
Resolved in a later commit.

[precice-bot]

This pull request has been mentioned on preCICE Forum on Discourse. There might be relevant details there:

https://precice.discourse.group/t/does-the-fenics-adapter-support-3d-case/696/2

[BenjaminRodenberg]

I think the case is overall working. I'm a bit concerned about the results compared to the CalculiX solver, but not sure if this is in scope of this PR.

[BenjaminRodenberg]

Do you want to generally keep this watchpoint or did you just introduce it for verification of the results? Please add an image of the results to the README.md of this case similar to https://github.com/precice/tutorials/tree/master/perpendicular-flap#post-processing. This helps a lot when releasing and checking the cases. Please also provide a script (you can probably just use a modified version of https://github.com/precice/tutorials/blob/master/perpendicular-flap/plot-all-displacements.sh).

[BenjaminRodenberg]

I already added a script to this PR. Here is the plot that I created:

The circumferential displacement is (for both solid solvers) negligible as expected. The radial and axial displacement also looks good (as far as I can judge here...). What worries me a bit is the large difference between the CalculiX and the FEniCS case. Do you also observe this? It might also be a problem with my system, because I did not update CalculiX or the adapter for quite a while.

[IshaanDesai]

What worries me a bit is the large difference between the CalculiX and the FEniCS case. Do you also observe this? It might also be a problem with my system, because I did not update CalculiX or the adapter for quite a while.

I am aware that there is a large difference in the results of CalculiX and FEniCS. I think this is the case because of the choice of the elements and also the solvers in CalculiX and FEniCS. The CalculiX participant has C3D10 elements. For the FEniCS participant we use the Cylinder function to construct the geometry. The FEniCS version we use offers very little control over the type of elements we can have, especially when we use in-built geometry constructing functions. In FEniCS-X there is greater control over the choice of elements.

I think we should not worry too much about having the same results with CalculiX and FEniCS for this case as it is a complex case which also takes a long time to run. We can present this tutorial merely as a case to showcase the 3D functionality.

[BenjaminRodenberg]

I agree. It's already good to have an example for 3D the disagreement between FEniCS and CalculiX is a different story. Can you document it in an issue and also add the possible reasons you mentioned above?

[BenjaminRodenberg]

The disagreement between CalculiX and FEniCS should be documented in an issue. Rest is good to go 👍