Unexpected difference between Mecway and CCX solvers

ehoule

Good morning,

I tried to make a quick analysis of a handrail using shell elements and mecway solver, nothing very special there, except maybe a shell offset.

I had some doubts about the results, so I decided to simply switch to the CCX solver and got quite a big difference between the results (by a factor over 2 for VM stress and displacement)

Modelized again using internal beam elements and results are much more closer to the CCX results. Did I made something wrong?

Victor

Hello ehoule

I think you might be running into a problem that the internal solver's shell elements have with some curved geometries. You can see that the solution is wrong because it isn't converging with mesh refinement:

6342 elements. Max. displacement = 0.6273 inch
25368 elements. Max. displacement = 1.276 inch
101472 elements. Max. displacement = 2.091 inch

In contrast, the CCX solution doesn't change much with refinement.

disla

Hi,

I have taken a look to it and, to me, it seems a problem of the internal solver with pinned supports.
Mecway is constraining rotations when it shouldn't. The simplified version of the handrail .liml file attached shows what I'm talking about.

Run with Internal has a solution (Should not as it is underconstrained).
Run with cxx the problem is unconstrained (Right solution)

Victor

It looks like some of the shell elements with certain curved shapes are leaking moment to ground. It improves slowly with mesh refinement. I don't understand what's causing it and maybe I need to replace the entire shell formulation to solve this and other problems.

Here's a simpler 3-element example that shows only a small rotation when it should be unconstrained.

disla

Hi Victor,

I do not know if it may help but I have noticed that the thickness expansion in ccx and Mecway is not the same.
There is a small difference. (See Pictures)

Another curiosity is that the moment leaked to the ground depends on the model distance to the origin. The difference can sometimes be some orders of magnitude.(see file)

Another One is the Y rotation that change sign in one of the models.

Victor

You had me even more worried there, but I think your example is actually OK because the displacements are huge (~100000 x mesh size) and would be swamped by any actual constraints. So I wouldn't worry about the position dependence or inconsistent rotation direction either. I don't see those effects with my example that has much smaller displacements (more spurious stiffness).

You're right that the internal solver expands shells differently to CCX. They are expanded in a single direction, whereas CCX shells are expanded according to the local normal at each node. Usually, the difference doesn't matter, but it can do if the surface area for loads is important, or the size of offset shells. I know this is not ideal but legacy.