It depends on the particular case but in general, the following techniques are used for large models:
- simplifying the geometry (including fewer details)
- using shell and beam elements wherever possible
- taking advantage of symmetry
- treating selected parts as rigid
- using bonded connections instead of nonlinear contact conditions
- using submodeling and substructuring (superelements)

last time I checked a real structure doesn't have singularities. The solid should have "proper" fillet rads, don't suppress them. Mesh the crap out of it. I just finished a model of a 1,000,000 TET10s ... ok, it took an hour to run, and I could only run one load case at a time, but so what. If I Had to I'd've learned re-starts to speed up the multiple cases (but I have only 10 so "rinse, repeat").

"Hoffen wir mal, dass alles gut geht !"
General Paulus, Nov 1942, outside Stalingrad after the launch of Operation Uranus.

User "FEA way" already gave many useful tips. The structure in question seems to be some sort of ship or offshore structure, so I would suggest to investigate the marine FEA literature, classification society recommendations and maybe course material from some university (many of the prestigious unversities offer course material online these days).

Hourglassing should be avoidable by using proper plate/shell elements (e.g., MITC4 elements or whatever the state-of-the-art shear locking free element is these days) which are not integrated in a single point.

Singularities can be avoided by using common sense. At sharp corners with small element sizes or skewed elements (aspect ratio of triangles or rectangles too small to provide well-conditioned Jacobian for integration in reference element), there may be a large peak of stress. If it seems un-physical, it probably is, and if your material is steel, local plastification and stress redistribution often ensure that such stress peaks do not occur in the actual structure.

Singularities can also be caused by load application (RM plate elements, for example, do not handle point loads well in theory) or supports (supporting solids on a line or on a point is nonsensical). Surface loads and line or surface supports are often a safe bet for plates and shells.

Good technical comments have been offered already. I will offer a strategic one.

An FEA book isn't going to address this need. It's a mix of isolating the key question/quantity from "noise", your engineering intuition, and FEA skills that will have to come at play here. By noise, I mean questions that are hurled towards a modeler from non-FEA experts; some questions are feasible, most are not. In rare cases, however, you end up having to use a brute force approach by throwing a gazillion elements with the trade-off that your turnaround time takes a hit.

*********************************************************
Are you new to this forum? If so, please read these FAQs:

http://www.eng-tips.com/faqs.cfm?fid=376
http://www.eng-tips.com/faqs.cfm?fid=1083

Thank you all for the responses.

I didnt explain myself correctly, sorry.

My question is not related with this particular example, but for a procedure for a common aproach to redundant structures, where the elastic limit is far away from analysis results.

The example, has no hourglassing, no corner singularities, by definition of classification, no stress beyond elastic limit is allowed, and above all, there is a condition of not "changing too much the geometry", this problem requires a good aproach of how and where apply the thickness, and what i am asking is a condensed procedure.

For me, in this moment, the answer is kill the normal stresses as they are the bigger, giving inertia, increasing thickness on flanges, after that i would go to shear, with thickness on webs, but maybe, some of you could give another insight, from a previous experience.

Thank you all.

Regards

So what you seem to be asking is “how do I design this type of structure?”. That is not a FEA issue. FEA is a tool in the design process. You need to talk to some peers / supervisors who have experience designing whatever type of structure this is.

yes, How do I learn how to design airplanes ?

1) work in the industry for 10 to 20 years,
2) design a bunch of bad ones first ...

"Hoffen wir mal, dass alles gut geht !"
General Paulus, Nov 1942, outside Stalingrad after the launch of Operation Uranus.

Unless I have missed something obvious, I agree with the previous comments.

However, the images make me curious: You are concerned that the random spectrum analysis is showing you very high stresses. How confident are you about the PSD that was fed as input? Is the PSD an accepted standard? Are those stresses 1/2/3-sigma stresses?

*********************************************************
Are you new to this forum? If so, please read these FAQs:

http://www.eng-tips.com/faqs.cfm?fid=376
http://www.eng-tips.com/faqs.cfm?fid=1083

we definitely have created a trend of producing more and more numbers (or pictures of numbers) with less and less real meaning, until ...

"Hoffen wir mal, dass alles gut geht !"
General Paulus, Nov 1942, outside Stalingrad after the launch of Operation Uranus.