Hybrid mesh feasibility: where large deflection without Mesh aggregation not working for large model - 3D mesh

[nitin_hadpe]

Hello Developer team,
Recently we came across an error: Error :301150 with Autodesk support and we realized (Based on Autodesk support team) that 

The investigation has exposed an issue in the code:

• The incident is triggered for large 3D models running large deflection with mesh aggregation disabled and upgrade tetras set to either automatic or all.
• While the model size and type of geometry (ratio of thicker vs thin layer areas) is the driving factor, we cannot estimate beforehand if the incident will occur or not.
• Until a solution is released, the recommended solution for this type of models is run large deflection with mesh aggregation enabled.
• The provided ****ysis had gaps in the node numbering. It is important to make sure the node labels had been squeezed before running any ****ysis --> this we already tried.

With this scenario there was a thought (Knowing it's very much tricky but may doable), Can we have a hybrid mesh (DD+3D) where we can have 3D in complex area where Midesh is not suitable so we have to go with 3D and other area we can keep DD (Thin area / uniform thickness area) to keep element count in control?

I guess other product simulation software has such facility, may be tricky for process but want to check.

Case number to know more: CaseNo:22274199.

Looking forward to having feedback. 

[nitin_devkar3HFV4]

This is a much-needed enhancement in the Autodesk solver. We always struggle when we have a heavy model with 3D mesh with an objective to capture critical areas like grills or finer rib network on plastic components.

With a possibility to have hybrid mesh (2D shell in major areas and 3D in the critical area of interest, it would result in a comparatively lighter model which is faster to solve with a possible enhancement in Autodesk solver to manage models with hybrid mesh. 

Looking forward for positive steps in this direction by Autodesk developer team.

[raalteh]

Moldflow had a Hybrid development for a good number of years (early 2000s) but eventually abandoned this. One major hurdle was to have the solution in the different domains behave similarly. The solutions at the different domains are connected, and if the domains behave not ‘very close to identical’, the over all solution (in for instance warpage) can be very heavily influenced by the mesh domains themselves. For example if a 3D element domain would shrink more than a Flat area domain just because of the element type, and they are connected, then the combine structure would not likely build in a warp due to differential shrink page of the mesh domains. This is very hard to avoid and there would not be a way to distinguish this from the ‘real’ differential shrinkage effect. This situation would give a different solution depending on how the mesh is constructed. At the time the Moldflow development codes had a few different schemes to create the Hybrid mesh and all could give very different answers (and we hear this holds true with competitive solutions).

Moldflow got close to releasing the Hybrid mesh technology commercially but ended up abandoning the project very late in the game because it had become clear that the solutions were extremely mesh dependent and there was no clear path to solve this. The advantages are speed (and model size) but the solutions were not reliable.

[nitin_hadpe]

Thanks for update, I wasn't aware that.
In fact, I am following Moldflow from MPI 4.1 (~ Y2002/3) and till that time there was nothing such like hybrid mesh, maybe it was before than that.  If I understood well, hybrid mesh tried but could not show close results so abandoned the further development. 

We are exploring what different simulation software are doing at the moment on very high model 3D warpage but thought to check with Autodesk as well on possibility of Hybrid solver.

One thing and you also stated same that mesh type should not show variability in results so solver should behave similarly, and we have seen even single mesh type we (mid vs 3D) we can see some differences, so Hybrid won't work with current set up and need investigation at solver level (If it's in interest of global community). 

[raalteh]

A lot of 'under the hood' solver work is to reduce the mesh dependency of the analyses, as well as to reduce the differences between Midpland/Dual Domain and 3D. This is not straight forward and is an area of continued work. 

That said, making the results mesh independent will probably not happen any time soon; its more likely that the the results will not change much beyond a certain level of refinement, but I would not expect we'll be able to get the results for 1 layer of tets to be the same as the current default 10 layers of tets (for example)