Pardiso solver returns no error, but all results are 0

General Questions
1

Hello,

I’ve been studying an issue (stress due to uniform pressure) on a mechanical component with several variations in it’s design.
It’s a mildly complex containter-type object, which needs to resist pressure from the inside, and we’re changing around just the “reinforcements” on the outside of the shell.
BCs, Loads and their surfaces don’t vary at all between tests, just the outside geometry.

I’ve been using PrePoMax with Pardiso (any other solver wouldn’t get the job done).

In the latest variation, which is the most complex, the solver doesn’t return any exception or errors, but all answers are “0” (no displacement, no stress etc), I tried this with different steps and inputs but nothing changed.

Any idea where to get started to figure what’s wrong?
Is there a way to get a higher level of logging from Pardiso?
Could you reccomend a procedure to simplyfi the meshing? (I’m fairily new to this world)

Thanks a lot

2

I was solving a medium size model days ago, and have the same problem, the solver ends with no error code, but not result at all. The problem was that the model was too big and was unable to save the results, the solution was to decrease the size of the mesh to have results.

Is a thing that would like to be solved, it happened several times but there is no warning that the fault was the size of the probelm.

1 Like
3

And for editing the mesh, you must start in the CAD:

  1. Apply simmetry, magically your mesh will be half or quarter zise! => Cut the part in half/quarter…or even a radial segment if is ciclic
  2. Remove small details as small holes, fillets, chamfers or features => Model thinking in FEA, first structural shapes/features, then small features, and finalize with finishing features as fillets, chamfers and so on, so then is easy to recover a simplified model going back in the part history tree.
  3. Is a part with constant thickness walls? Try to use shell elements, be carefull with simetries, sometimes is better to use complete models with shells on CCX. Also with the element size and order (always second order!).

In any case don´t take your first results as an absolute true, keep in mind the limitations of the solver and bc, and make several runs with different element size until your find that the results converge.

2 Likes
4

What mesh sizes are we talking about? Number of nodes?

Is it the problem of saving the file or a memory problem? If it is a memory problem, using a different solver might help (Iterative).

It it is a problem of saving a file, using a binary output might help. However, it is not yet supported in PrePoMax.

1 Like
5

it’s a bit less than 2 * 10^6 nodes, where the previous iteration which worked was 1.4 * 10^6.

I don’t know where the issue is as the error isn’t handled, but using symmetry to simplify the problem helped

6

Thanks a lot! these reccomendations were really helpful, I’m working now with a simplified model suing symmetry and it’s made life a whole lot easier

not sure yet where the problem was, but it was definitly related to model size

7

CalculiX may still struggle with large models (with millions of nodes). Not always and not with all solvers so if the model can’t be simplified then it’s best to try with another solver. Fortunately, most models can be simplified in some way or at least you can start with a coarser mesh and then keep refining it as needed, based on the convergence of the results.

8

I was informed that there is a CalculiX version that supports larger models: ccx_static_i8.exe (and ccx_dynamic_i8.exe) solvers are versions which have full-length pointers. It was published a few years ago and has a larger memory capability than the standard version. You can still find it in package:

http://www.dhondt.de/calculix_2.19_4win.zip

It is not fully stable, but anyway, one could solve about double-size models with it.

Maybe someone could ask for the new version with the increased capabilities.

2 Likes