Hard to understand HyperSizer calculation in sandwich optimization

[iankim]

Hi,

I've been working on sizing a structure with unstiffened panel/sandwich panel concept.
As I've been accumulating results after several iterations,
I found that HyperSizer seemed to distribute the load too unevenly,
and it led the panel with too much load to be too thick, which is unreasonable.
Attachments I put to this message shows that evenly distributed Nx originally
(1st run) gets to be changed such that bottom face carries too much compressive load
(3rd run).
This causes the total weight to increase considerably(top face has smallest thickness
and cannot become 0 in the real world) and hinder speedy convergence.

I'd like your any advice or comment on this topic, thanks.

Ian.

[iankim]

If you check the trend of variables change as shown in the figures attached
(result of 1st run, 3rd run and 5th run respectively),
you'll find the thickness of bottom panel and honeycomb are going to extreme,
and it leads the weight to increase continuously.

The weight of 'Skin_Up_A' group changes from

40.3 -> 42.1 -> 47.4 -> 51.1 -> 54.2 kg

as iteration progresses, and it doesn't seem to me that it converges to any value.

What's the problem supposed to be???

[Phil]

Hi,

What kind of loading do you have on this panel?  Are there pressure loads?  Are you superimposing pressure.  One thing I am thinking it could possibly be is that if you are superimposing pressure and also have moments from the FEM then you could be double-bookkeeping the moment load, which would cause one facesheet to see a higher load than the other.   If this is the case you might try "zeroing" the moments from the FEA using the checkbox on the FBD tab.  If you do not have the "Superimpose Pressure" button turned on, then do not zero out the FEA moments.

One thing you could try is to link the top and bottom facesheet material and thickness (these options are found on the Concept tab).    Doing this will cause the optimized result to have identical facesheets.

[iankim]

Thanks, Phil.

As shown in the attachments of this post,
there is no superimposed pressure.
All load are nodal loads in FEM, so your first comment doesn't apply to this case.

Secondly, I tried to optimize the problem with top and bottom facesheets
linked together in the first place as found in the second attachment,
only to witness the weight becomes even greater.
That's because the thickness of top facesheet follows that of bottom facesheet,
while bottom sheet has to be thick but top sheet doesn't need to be so, though.
This means I think that load distribution to each part doesn't change
regardless of whether two sheets are linked or not.

What do you think of this...?

[Phil]

It seems like you are getting a very large bending moment introduced into your panel in addition to an axial load.  This would cause the panel to load up heavily in one face and not the other.   For example, if you have pure moment, then you would get a tension in one face and compression in the other.  Then if you apply a membrane tension to the panel in addition to the moment, then it would make the tension face worse while relieving the force in the compression face.

On the Options tab of the sizing form, do you have sandwich midplane (the second option) chosen as the reference plane?  If not, then you might be loading your panel in the FEM in one of the facesheets which might cause this behavior.

Can you help me understand the loading that you are applying to your panel?  Do you have free rotation edges on your model or are they fixed edges?

[iankim]

On the option tab, sandwich midplane(second one) is chosen.
So this is not the issue.

There wasn't big bending moment originally.
But I suppose moment gets to increase as iteration
progresses, to balance the uneven load between
top and bottom facesheet,
plus the effect of the increase of honeycomb thickness.

I attached the load distribution of the panel(A_3) I posted earlier
to comply your request.
They are external load, internal load and internal moment
calculated from Nastran freebody repectively.
All loads are just vertical nodal loads originally as shown in *ExtLoad.jpg.

The last one indicates the boundary condition on the whole wing
and A_3 is the part of yellow area.
You may think it is influenced by DOF 135 fixed nodes but it's not true,
because A_3 is not the only problematic panel.
A_7 and A_8 which are away from root BC are showing similar tendency
for example.

Do you think I have to ignore moment during iteration?
I tried to optimize the model with unstiffened and stiffened concept earlier
but there wasn't any problem in convergence.
(Total weight decreased gradually and rapidly converged to some value
after a couple of iterations)
Only this sandwich concept optimization produces this phenomenon
even with the same FE model.
What do you think I have to do to solve it?

[Phil]

Its really hard to tell what could be causing this behavior.  I can see where a bending moment is being introduced from your forces but I'm not sure why it would not be converging.  If you send the FEM, I might be able to look at it.  I would only need the bulk data file and any include files, but would not need the *.F06 file.

[iankim]

I sent you the bulk data file by email.
Please check your email box.
Thanks.

[iankim]

While you're investigating the bulk data file,
I'm posting the convergence graph from HyperSizer analysis result of several design concepts.
While unstiffened and stiffened concepts show fast and downward convergence,
sandwich concept show slow and upward convergence.

I'm wondering if this is general phenomenon caused from HyperSizer calculation algorithm
or it's just specific to this model.
Any idea of yours?

[Phil]

I think this behavior is specific to this particular model.  The loads on this particular panel are such that one facesheet wants to load up while the other face does not.  I could see from the forces you sent earlier that a substantial bending moment is introduced to the panels.  If you also have in-plane loads, this would mean that the loads on one face would be alleviated and the loads on the other face enhanced, which could cause this behavior.  I can see from your plots that the solution is essentially converged at about 7 iterations.