FEM / FEA with highly heterogenous modell
22/02/23
Hi there,
i came across some robustness issues with the radioss solver and it is very hard to determine the true nature of the cause. I want start with a short description of the model and what elements are involved and what we found so far, in order to get insight / help to narrow the error causation down to a single element.
The idea of the modell is to simulate deformation in bone tissue. Imagine a box, like a crop from a Computer tomography (CT) imgae volume. This box has different materials inside depending on the brightness of the pixels, and each element is represented as a tetrahedral. The assigned material property is PLAS_JOHN, and eMod value as well as the density are generated using the metadata from the CT image. So now you have a box with the mapped materials, whereas the variety of stiffness occurs around 5 MPA ot 4000 MPA and is highly heterogenuous distributed. Now for stress simulation i want to compress this box and i am wondering about the vonMises stresses and the displacements of each and every node. Therefor a single Force is applied in the magnitude order of 400-600 N.
Now a force on the top surface of this box is imposed and the lower surface is set to be rigid as if the box stood on the floor. As the solver begins, at some point there is a divergence of the error value and with further computation its starts to point out there are negative / zero volume elements, untill the solver breaks down due to energy error limit reached.
For my model it is critical that the error diverges, since this causes high stresses in elements and alters my solution. So the target would be to have a robust computation, which never diverges in error or energy terms and does not fail the computation itself.
Now what i experienced is, if i set a lower threshold and artificially lift the stiffness of all elements lower than 100 MPA to 100 MPA, the computation gets more stable. But obviously the results are getting a little higher in terms of stresses.
So since there are numerous settings in the radioss solver and flags and so on - where i feel it is not a good point to start looking at, i wondered if this is a conceptual error, that heterogenous material with a high stiffness range is anyways hard to handle for the solver and more precisely if we could just delete elements with too low stiffness and run the computation with sorts of "holes"?
I feel the error must be inbetween the geometrical aspect of the elements or with the material set up. Any otherway i'm thrilled for help 
lets open this discussion, thanks
G.
23/02/23
Follow up on the problem. I looked at the internal energy / external force work, and it seemed to diverge very very quickly. As i looked closer into my engine settings, i saw that i used the DT/NODA/CST with a default value of 0.9 in [ms] units. In general i think i wanted to optimize the computation before it was even stable - so this made the computation a lot more stable without crashing. Now i am asking myself about the error rate between the cycles. If there is are some values rusing up to -30% but later on converge to lets say -4%, does that result in a probably bad simulation? Is it best if the simulation has a steady error?