Properties and Performance issues - large models

Hi,

I'm recently using Catia FMS to generate large models, up to 200 / 300 000 elements.
But I have a huge issue with computer performance.
My first question concerns properties mapping. Is it possible to generate different properties (one for each region) in a single mesh ?  Example : A wing skin has many different thicknesses. If a different mesh is generated for each panel, the model may get up to 80 Gbytes in size (?!) - which is unacceptable. But if the whole skin is treated as a single panel(surface), which solves the memory issue, the whole skin needs to have the same thickness, which is unreal. 
How to solve this case ?
Thanks in advance and cheers,

JP Lirani

So the question really is : meshing the whole skin as one surface solves the performace issue but, is it possible to associate this one mesh with different thicknesses (for different "sub-regions" or faces) ?

Hi
i think i understand your problem except for size. Are you saving the analysis result file? See "Save Managment" to save only the analysis model!

also what kind of mesh are you using? i think you can stick to shell elements !

can you upload some screen shot !

thanks
robbie

Hi Robbie, thanks for your post !

The matter of size is memory usage, not exactly disk usage. The final model occupies about 1 Gbyte in disk, but to create the mesh the processor uses more than 80 Gbytes (or crashes before that in smaller machines ..).
I've noticed that bar elements work well and are lighter (as expected), but the use of several (more than 1000) shell elements ("advanced surface mesh" shell to be more precise) really consumes a lot of RAM memory to create. That's why I'm considering joining everything and creating one mesh.. but then I have these different properties to handle !

The matter of size is memory usage, not exactly disk usage

you need a 80GB RAM machine ? wow that's huge memory requirement. my understanding of FEM is fairly elementry. but i would not expect that you can replace shell with a bar/beam elements. (again i am not an expert in this matter).

you need not join the surface before meshing!

my argument is for making shell elements using surfaces !

-make many surfaces (but do not join)
-mesh them independelty! switch on automatic mesh capture..then you can give different 2D properties to each domain!
-use constraining elements on each domain so that your meshes are constrined ! (you can do this by using "project external elements").
-use quality checker to check the quality of mesh

you can increase the size of the mesh to bring down the memory requirement till you have a working model ! (also model only one side of the plane, unless you are looking at buckling characteristics)

what kind of work do you do?

Robbie

Thanks again, Robbie !
I agree - huge memory. And no, I cannot replace the shell elements by beams - they do behave as shell.
I proceeded exactly as described in your steps : generate all surfaces, mesh independly, etc , but that ended up in consuming this huge memory ! I cannot increase mesh size as it must respect mesh quality requirements.
This work is in the context of strucural analysis for aerospace industry.
I hope I could map different properties for s single entity, for instance, a property for each face of my surface, but I do not know if this will work. If not, how should I proceed to work with 1000 different properties ?

Hi JP,

I understand your issue as: you trying to use the property mapping for 300,000 elements and it's taking VERY long to solve and/or the amount of disk space used is huge. I know of this limitation with property mapping and it's really a function of the algorithm in the API being used to assign each element a different value. For each spatial location you specify a thickness value, the software has to sample that location and then locate the nearest element and assign the value (or interpolate). Yes, I agree this is not the most efficient method. I suggest you file an enhancement request to have that API updated.

Now, that said maybe we can find an alternate way of doing this. How many thickness mappings do you have (1/surf or 1/element)? Namely, if there are only a few, you could use groups. By creating a couple groups near the meshed surface, you can assign a different thickness value to each group. Thus, accomplishing a similar results (..assuming each element is not different).

Another technique would be to use the composite workbench (if you have access). The algorithm is different with composite workbench and how it assigns thickness values.

Cheers,
Greg

Hi
* meshing involving 200 000 to 300 000 (i can imagine) will always be time consuming and frustrating.

Maybe, You can consider creatig a meta model for different parts of the wing which will give you the kind of properties that you are looking for (stiffness, stress ). these meta models should be much less memory intensive. which can be assembled to compelte an analysis run. Such optimized meta modeling (changing individual element thickness) is probably difficult and time consuming in CATIA.

what did you mean by property? if you just want to quickly change the thickness then connect field to a design table and this should enable you to change the thickness fast. Maybe you might have rely on scripting if you want to change the material also.

I dont have CATIA right now and so i cannot checkit out.
But try meshig the whole surface and when you want to give 2D properties, select the geometry in the tree.

(Surface C=A+B)

mesh surface C

give 2D properties to Surface A and B (do not select the mesh as support for it). I dont expect you to get performace enhancement. (if it works)

Good Luck!

Thank you for your posts, I wil try to answer them and clarify some points.

Greg, the main issue I find in dealing with many properties is not disk size or solver time, it is really the time and RAM memory usage to generate the mesh. I agree that the property mapping method is the main issue here, and since it will not change in the short-term, let's discuss workarounds..

Robbie, meshing 200.000 elemnts does not have to be so time consuming. Another commercial packages mesh that model in a matter of seconds (yes, seconds). And even in Catia, when I mesh the whole wing surface at once, memory drops to 1Gb, instead of 40 Gb !

I'll try to explain more about the context.

A wing skin (surface) is made of many wing panels, as seen in figure 1(wing_panels.jpg). In this picture, each panel has a different color. Each wing panel has a its own thickness. So, in Finite Element nomenclature, each wing panel is represented by a group of shell elements (mesh) and must have a different shell property. This shell property is exactly its thickness. This may be seen in figure 2(wing_mesh.jpg).
Note that this is a sample wing, with a couple of wing panels. A real wing may have more than 1000 panels and thus, 1000 meshes and properties..
Using this approach, each wing panel has its mesh and its property. But when generating this mesh for large models, the memory needed to complete the process is too big.
So what I considered to be one solution is to mesh the whole wing surface (all panels at once) and try to create properties for each panel. But it did not work.
Greg, answering your second question, I use one property for each wing panel, so it may get to 1000 properties. How do you use groups ? Is it possible to define more than 1 group for a single mesh ?
I hope this helps to understand the problem.

Thanks again for all contributions and cheers,

JP 
 

 

JP,

It may be best for me to just show you. I'll send you a PM with my contact info.

Cheers,
Greg

i think Grega is taliking about "surface grouping" in the meshing work bench!

good luck!