reliability of FEA stresses with out testing. 4

[kamal11]

Hi all,

So far, in my analysis carrier, I was redesigning components to a bench mark design using deflection or frequency as a critieria, But not for the stress values. Now, I need to find the highest torque a particular gear box can with stand. I understand , "Entire carriers are based on gear designs..."

I understand stresses due to FEA are dependent upon the loads, constriants, materail properties, geometry and their variations, singularties etc,...
With the analysis we did, I many times saw, stresses on FEA analysis above the yeild strength by 2~5 times.
I may need to fine tune my analysis to get better results by looking more closely at the application.

My understanding is people tweak their FEA models, by corelating them to the test results and then with experience and comparision with test data, they start comparing their designs.

This question keeps on bugging me all the time, what ever components I analyse. How close are my FEA results to reality...

With out testing,
how do we determine, which analysis will predict the failure stresses correctly. linear, non linear material or contact analysis etc..

Assume that we did the best FEA analysis with out testing the product and send a report saying the componet will with stand this much of loading.....

What will be the realiability of such statment based upon FEA results. how far can we depend on FEA.
What should be our approach when we need to design for failure using FEA. .
Regards
Kamalakar.

 

[johnhors]

"If one is not sure that his bounday conditions and loads may be in-appropriate, one can list down on papers serially
diiferent boundary conditions and loads which one may feel that may be applicable.
Work out on anlysis with all different bounday conditions and loads and write down all the results under different conditions."

This statement is very worrying! A load is a boundary condition. For static components the analyst must be able to sketch a free body diagram, label ALL the applied loads and check for a balance of forces and moments. Otherwise the analyst should not be running FEA !

 

[cbrn]

Hi,
Johnhors, I agree. But what Salilkrishnarao wanted to say is most probably another thing.
Say, you have a scenario where it is unclear, at the first sight, which FE boundary conditions pattern (restraints + loads + ...) would be the most APPROPRIATE (I would like to use this term) in order to figure out the "real" behaviour of the analyzed part. Let's say, in addition, that all these patterns are globally equilibrated and compatible. But let's say you still can't forecast which one is "the best". Said like that, it may sound ridiculous. In fact, it partially is. But sometimes in order to appreciate how much you are approaching the reality, you might have to run analyses with different BC patterns, because the system response itself could guide you discerning which abstraction model is most appropriate. You will say: "and how would you know which response is absurd and which is not?"; perfectly right: here comes experience and knowledge of the "real" parts. That means also: here comes previous testing of homologous components in similar situations.
I think that's what Salil wanted to say, not that the "analyst" should or could run randomly a bunch of simulations with casual BC in order to see what happens. This last method would rather be applied by a well-instructed monkey...

Regards

 

[johnhors]

cbrn, thanks for your input, my approach is that if the loading is not determinate or cannot be rationalised into a determinate load system (the free body diagram with balanced determinate loading) , then a knowledge of the supporting structure is essential for a correct analysis. Sometimes if you're modelling a component that is bolted onto an aircraft, boat or car, you may be able to obtain a stiffness matrix for this purpose, otherwise the supporting secondary structure should be included in the overall model.

At risk of getting on my soapbox again! How many software demos/tutorials/novice users have you seen where loading is applied at one end of part and the other end is fully fixed? Thus transforming the part into a cantilever (a mathematical entity that doesn't exist) and a wholly inappropriate behaivour results. Ideally supports/restraints should be kept to a minimum, as an over restrained model produces fictitious results.

 

[cbrn]

Hi,
Johnhors, you hit good points in my opinion. I agree that an important part of the success of a simulation model is to place the "bounding limits" where they are appropriate. My working field is very far from yours and rarely we encounter situations where the matrix of a "superelement" must be known in order to get valuable results from the components interested by the analysis. But, we encounter something similar when a part is supported entirely by others (which, in their turn, can be considered "fixed" to the ground, i.e. fully restrained in some points/edges/faces) and we don't want or can't afford including them in the analysis altogether: it's a case where idealized restraints would be extremely inappropriate: we can't leave the part unrestrained or it will experience FBM; we can't fix-restrain it because it would be unrealistic. Sometimes, calculating the equivalent stiffnesses of these supporting parts, for the DOFs of interest, can be very long and difficult (sometimes we have chains of components involving three or four "levels" of this kind of abstraction, and the parts themselves are too complicated to be analyzed altogether, or sometimes the interaction between the parts forms a "loop" so you have to guess and iterate...), but it is also a part of the work that I personally love, I find it extremely interesting. Maybe I'm mad... ;-)

Regards

 

[GregLocock]

In my world uncorrelated FEA models with assumed load cases will kill people, and then we'd go to court.

So, I've got say that this thread gives me the creeps.

Cheers

Greg Locock

Please see FAQ731-376 for tips on how to make the best use of Eng-Tips.

 

[cbrn]

Hi,
GregLocock, I don't understand your grief: your situation is the same for everybody (have you ever seen what a penstock breakdown can do? I had the opportunity to see...), and it seems to me that no one here said that it is suitable to run uncorrelated FEA models (b.t.w., what do you mean by "uncorrelated"?). OK, I write this because I feel I could have been misunderstood.
You mean that you've never come across the question "and now, what BCs do I put to capture AT BEST the real phenomenon?" ? If no, there are only two cases:
1- you're lucky
2- you're very experienced;
having read other posts of yours, I am sure of the second hypothesis, but then it's the same thing that several persons herein have said in other terms... Where the point open by the Original Poster was: can FEA really capture the "reality"?
IMHO, the vast majority of the posters have said "Yes it can, but it depends upon you to properly set up your analysis: you're here for that." ALways IMHO, we could also stop this thread, as you say.

As a last thing: where I can easily have been misunderstood: when I said "...you have to guess and iterate", I meant: when you can't forecast the complete behaviour of a very complex structure (some of those I deal with would require a Cray to be analyzed "in one pass", and I think you also experience this situation, due to your workfield), formed by parts whose behaviour under a WELL-DEFINED loadcase depends upon the behaviour of other parts under THE SAME WELL-DEFINED loadcase, then I think you can operate just like the iterative solvers do: 1st hypothetical BCs on the "seed" part -> calculate response of the "seed" part -> calculate interaction with other parts -> calculate response of these parts -> update BCs on the "seed" part -> calculate response -> calculate interaction -> and so on, until two successive responses differ less than a WELL-DEFINED criterium. Or you also can use sub-modeling, or compute superelements, or... etc... Do you disagree? I would be interested to know your opinion.

Best regards

 

[GBor]

cbrn,

There is a software package that you may want to look at. It balances the loads on a model automatically so that a few minor boundary conditions well away from any point of interest is sufficient restraint.

http://www.roshaz.com

It has its limitations, but for the price, it's powerful. Not sure if maybe I misunderstood your modeling methods of progressive local models, but this package sounds like it could be useful.

Garland E. Borowski, PE
Borowski Engineering & Analytical Services, Inc.
Lower Alabama SolidWorks Users Group

 

[GregLocock]

cbrn - what is the title of this thread?



Cheers

Greg Locock

Please see FAQ731-376 for tips on how to make the best use of Eng-Tips.

 

[cbrn]

Hi,

Gbor - thanks, I'll give a look as soon as I have time. Maybe it could be useful for some of the "progressive modeling" situations.

GregLocock - title is "reliability of FEA without testing", if I put on my glasses well ;-) I understood the OP was worried that the FE simulations could deviate from reality (assumed that testing is the reality, which is not always true...) whenever not supported by a physical test. I also understood the whole thread tried to find out at which conditions a FE analysis can live its own life without a corresponding test. At the end, it seemed the majority if not all agreed that FE on its own IS reliable, when you have knowledge of what you are inputting (the famous "garbage in, garbage out"). Maybe it was not necessary to come to 27 replies to state this obvious thing, maybe I misunderstood something from the start, maybe the necessity of doing tests at any design step depends on the application field, maybe yours require it whether mine doesn't... I seem to remember that Boeing 767 was the first civil airplane to come to final prototype assembling without any intermediate simulacra... So the "total abstract" design is perhaps not so impossible... How many Boeing 767 have crashed down ?

Anyway, I'm a bit disappointed: even if I may have expressed idiot statements, I was waiting for a constructive comment and not for an ironic question...

 

[rb1957]

cbrn,
"I seem to remember that Boeing 767 was the first civil airplane to come to final prototype assembling without any intermediate simulacra... So the "total abstract" design is perhaps not so impossible..." ... i think that has more to do with 3D CAD design than with stress analysis by FE.

also the B767 was rigorous tested as a structure before entering service.

my 2c worth ... it's the engineer behind the FEA who adds the intelligence to the result. with knowledge and experience, a competent engineer can model a structure and reach conclusions about its strength. otherwise, you can replace your expensive FEA package with a much cheaper random number generator, or a drawing package that allows you to colour in the spaces !

 

[cbrn]

rb1957 - I 100% agree.

As regards Boeing, well, "maybe" I exaggerated a bit... I would rather have spoken of examples taken from my own field (yes, though "injury-critical", some components are NEVER tested before they come into service, simply because... it's impossible !), but refering to Boeing was more impressive ;-) !

Regards