Thermal stress simulation: TET ELEMENT XXX HAS SINGULAR GEOMETRY

[roseR2YZL]

I'm working on a thermal stress simulation with a relatively simple 3 piece assembly. I've run a straight up static stress simulation and everything seems happy, but when I clone the static stress simulation and try and add a thermal stress, everything gets really unhappy. Here's a sample warning I'm getting:

Warning: TET ELEMENT 1681882 HAS SINGULAR GEOMETRY
Cause : The specified element is not valid or is highly irregular. The following will result :
- The element stiffness matrix will not be valid.
- The global stiffness matrix may be singular.
- The solution may not be valid.
Action : Check the element connectivity and grid point coordinates.If the element is a bar or beam
with pin flags specified, check that the pin flags are valid.
Warning: Exceeded the limit of 100 warnings
Error: STIFFNESS MATRIX SINGULAR OR NON-POSITIVE DEFINITE
Cause : A singularity or non-positive definite has been detected in the stiffness matrix during the
preconditioning phase of the iterative solver.
Action : Investigate the model for a lack of constraint. If using shell elements, either set the
K6ROT model parameter to a value between 1.0 and 100.0, or set the SHELLRNODE model
parameter to ON. Also, check for elements with bending stiffness (line and shell
elements) improperly connected to elements without (solids). If you are unable to
locate the source of the singularity, consider using the sparse direct solver.
Job has finished with errors. Receiving results.
Error: Solver Error
Error: An error occurred while solving the model.

I've tried making all my contacts bonded and I've introduced fillets on all my edges, tried extremely fine meshes and regular sized ones, but I must be missing something because nothing seems to be working. I've attached the assembly in question if anyone wants to take a stab at it.

Questions I have: what is a singular geometry and how do I beat it?

[John_Holtz]

Hi @roseR2YZL

That's strange that the stress analysis runs but the thermal stress does not. Maybe there is some logic that "changes the model" for the thermal analysis? I am trying the analysis with the "Create Curved Mesh Elements" turned off ("Manage > Settings > Mesh > Advanced Setting").

To answer your question, a singularity indicates that the matrix of equations cannot be solved. As indicated in the error message, this is probably due to a distorted element. (Other cases of "singularity" are caused by missing constraints which causes the model to be unstable, or sometimes a combination of strong material and weak material can make the solver have numerical singularity because of roundoff.)

[mcmillr · Accepted solution]

I created a Modal study and placed a single constraint and I am seeing the same warnings.

There is something wrong with the part '10 pin rotating female'. It appears to generate a good mesh, but when the mesh is sent to the solver it has badly distorted elements.

If I simplify this part so that it is just a half cylinder, it still has problems.

If the model is Exported to STEP and then re-imported the problems appear to be gone.

Perhaps that could be a short term workaround until we can figure out a more surgical fix.

[roseR2YZL]

The re-imported .step worked, thanks for that.

If I may, why did you make a modal study? Just to see if the error wasn't limited to thermal stress?

[roseR2YZL]

I see, so what would cause a distorted element?

[mcmillr]

I am glad the export/import allowed you to move forward.

Creating a modal study is sort of a 'sanity check'. It requires the least amount of setup and the results are pretty predictable. I was immediately able to see some odd looking displacements and the mesh geometry warnings were still being generated. There appears to be some issue related to the geometry of the part. Since the part was created in Fusion its possible there is some sort of bug in the modeler which is producing geometry the mesher does not like. If I create a similar shape from scratch I don't see any problems.

Even though the mesh looks fine in Fusion, the mesh being sent to the Nastran solver is clearly not right. (We have an internal way of viewing the nastran input file.) I will engage with our mesher and solver team to understand why the mesh becomes corrupted, but only for this particular part.

I will follow up once we have a better understanding of the problem

Thanks for your patience and understanding!

[mcmillr]

We found the root cause of this issue. Some unusual (but valid) numeric values in the model exposed a bug in the generation of the solver input file.

We are working on a fix.