Stress concentration in solid modeled with TET10

[Markeng512]

Hi everybody,
I received a very big complex to model. For one of the attachmetn I removed a very big fillet leaving an 90 deg angle. In the simulation I have a peak of stress in that area. Would it be possible to show through an analytical calculation the stress value in that area if there was the radius modeled?

 

[whitwas]

I can't see an attachment, but for simple geometries, there are stress concentration factors in "Roark's Formulas for Stress and Strain".

There wouldn't be a way to convert the peak stress from FEA to an actual stress with radius because the peak stress in FEA is a massive error. There's software that does something clever with shells to obtain maximum stress from stresses a small distance away from the corner, without having to model the radius. But I think that's a secret sauce they've invented themselves.

 

[Mustaine3]

In Abaqus is a functionality called Surface-based Submodeling, that allows to have a closer look at that region by running only the interesting portion of that part.

 

[Markeng512]

Thanks Mustaine3,
but in reality I am using Femap that I think doesn't have this functionality.

 

[rb1957]

why'd you remove the fillet rad ? that's an olde school solution to the problem of model size.

can you make a free body of the part with the fillet, and run it in isolation ?

another day in paradise, or is paradise one day closer ?

 

[corus]

If it's a very big fillet, ie. not just a fillet weld for example, and you've removed it then you'll have altered the structural stiffness and so all your results wouldn't be correct. If it is a weld fillet radius then you wouldn't normally include it as I've yet to find a welder who can make a perfect radius at a weld.

 

[NRP99]

I think of "linearization of stresses in near and/or along the peak stress location" as an option if available directly in FEMAP. From this you can find out peak stress component among other average stress. Check out Stress linearization calculation online if not available directly.

Another more visual and crude option is to check the stresses away from the discontinuity where they are very much continuous and compare it with your corner stress. The ratio of the two will give you the SCF.

EDIT:
Tets are more stiffer than bricks. Use of Tets at corners will give you unrealistic rise of stresses that is more due to tet element's stiff behaviour than corner effect. use bricks and check it yourself.

 

[whitwas]

Stress linearization might be a good idea - not so much so you can find the peak stress (impossible here because of lack of radius) but so you can identify it and ignore it because a bit of plastic deformation there wouldn't cause failure.

The SCF for a 90 degree internal corner is already known to be infinite so you can't/don't need to estimate it from the ratio of stresses. That would give a wrong finite value.

While it's true that tets perform worse than hexahedra with the same number of nodes, you should have already made sure the results are independent of the mesh so both element shapes should be giving the same results (or both showing no convergence as in this case). If they're not, then the model is already wrong even without the stress concentration. Tets give results that are just as correct as bricks but you may need more of them. Bricks are an optimization to reduce the number of nodes and aren't necessary for correct stresses.

 

[NRP99]

whitwas

You said correctly about SCF but I proposed this way of checking for general feeling of stress concentration. I am just wondering about your following statements.

Tets give results that are just as correct as bricks but you may need more of them. Bricks are an optimization to reduce the number of nodes and aren't necessary for correct stresses.

My understanding is bricks give more correct results than tets for same analysis provided all things are correct(in the sense of analysis setup and physics) and of course with less no of nodes.

 

[whitwas]

As you refine the mesh, both tet and hex converge to the same solution, so they're both correct. If you use fewer nodes with the hex elements, then yea, you're right, hex can be more accurate with fewer nodes.

 

[NRP99]

Here is the reference of past threads which discusses the same issue but with different and deep angles.

http://www.eng-tips.com/viewthread.cfm?qid=2912

http://www.eng-tips.com/viewthread.cfm?qid=141134

http://www.eng-tips.com/viewthread.cfm?qid=376766

I guess we need to look into past threads before posting answers(me)/questions here.