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.

 

[corus]

The reliability of your statement will depend on the competence of the person doing the analysis, the verification of the results, and the methods by which you assess your results.

Competence I'd define as the amount of experience that the person has. Verification as the analytical means by which you checked the results. The methods by which you assess your results is normally defined by the design by analysis rules you have followed in the appropriate design standard.

It would also be wise to validate your results by measurements but this is not always necessary (or possible) if the verification process has been rigorous enough to give confidence in your results.

corus

 

[kamal11]

So are you agreeing with me in saying, with out experience on the specific product line, and with out testing, One should not release a report saying, the component will with stand this much amount of loading

 

[GregLocock]

Using real live data to predict stresses in important areas we get errors of around 30%.

The test to test repeatability of those loads is around 20%.



Cheers

Greg Locock

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

 

[corus]

Testing some times cannot be carried out on a component or a structure for the simple reason that the component hasn't been made yet. I said that testing by measurements is wise (if it can be carried out) but otherwise it is not necessary. It is a matter of confidence in your results, and the confidence you have in the person providing the results. It's also advisable to have someone else check/verify/approve the work who is also competent.

corus

 

[cbrn]

Hi,
I also many times wondered "how much" a calculation can approach reality, be it done by FEA or not. In fact, many times we speak about "analytical formulations" as the best way, when possible, to perform calculations, as opposed to "numerical approaches" such as FEM, FDM, CFD, etc...
But we forget (or want to forget?) that even the best analytical formulations are nothing but based upon abstraction models which can approach reality more or less, it depends.
The same is for FEM: the more complex your analysis, the more adherent is the abstraction model with the reality, the more close are the results to the test ones. When possible. In fact, why would you perform a time- and resource-consuming ultra-detailed FE analysis, if you'd better directly test the specimen or, worse, if you doubt that the numerical model would be in any case inadequate? Design time-saving, of course, but if you are totally in doubt about the principle, then forget about FE, forget about analytics, forget about formulas, forget everything, and convert from engineer to test-specialist!!! Make prototypes of everything at every design step, and you're done!
Well, I "extremized" the terms of the question, of course, but when I see simulation results which are not "physical", then it's up to me to filter them out (i.e. decide reasonably if and to what : why these results are there where they are? Which assumption was I undertaking when building and solving my model? Which "real-life" effect was I neglecting? Which "collateral" effects were instead introduced into the simulation model, due to the way it was built up? Am I completely aware of them, i.e. have I completely understood the theory of FE which applies to the simulation I am doing? And so on...
It may not be completely surprising that Greg Locock gets simulation results which are 30% in error w.r.t. "reality": this can happen, if his application field involves phenomena which are hardly covered by FE algorithms. On the opposite, in my own field, numerical prediction have many times proved to give errors far within 10% error (and, sometimes, even around 1%...) w.r.t. test data (when it was possible to do testing!).
But, when you get high uncertainties: why would you completely disregard FEM, if it allows you to discern between design variants even with the given uncertainty?

Regards

 

[GBor]

Corus' answer is the reason consultants exist. We have a lot of confidence in our abilities to produce the right answer with FEA because we have tested our modeling techniques against many live tests. A new object is generally just another means of combining the analytical formulations with which we are comfortable.

There are some techniques, however, to boost your confidence. We do mesh convergence studies to make sure that a more refined mesh doesn't change our answers by more than about 10%. We look for areas where the gradient is greater than 10% of the overall stress scale in a single element and determine if 1) it is an area of concern, and 2) if the stress is a geometric issue, or an actual, realistic stress problem. This is where the engineer comes out in us...

We also rely on our "gut feel" as to whether the software is producing reasonable answers and we perform simple models that we can hand calculate the answers, gain confidence in the more simplistic models and use that information to make certain we are using the right element formulations, load applications, boundary conditions, etc.

Garland

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

 

[cbrn]

Garland, could not agree more with what you said. A star!

Regards

 

[051174]

To make a good correlation predicting test results by FEM forecasting don't forget mohr theory and how stress are related to strain. i suggest to you before checking your model to see strain theory because gauges give apparent stresses and fem gives real stress.

 

[GBor]

Thanks, cbrn!

 

[prost]

051174: could you please explain the phrase "apparent stress"? As I understand it, gauges (I assume you mean 'strain gauges') measure local extension. If the local extension is small enough and if the strain field is well behaved (no sharp gradients, for example), the conversion from extension to strain is easy. From the strain gauge measurements, you can use Hooke's law to calculate the stress. Is that the stress you mean when you said the gauges measure "apparent stress"?

 

[RvL04]

I too could not agree more with GBor. I've worked with a wide range of material test data, some good, some bad. I was in the fortunate position of being able to use both material test data to build my model as well as component test data for model verification purposes. A good analyst / consultant can make this work, provided that his test data is good to begin with, of course.

 

[vonlueke]

prost: I think "apparent stress" would typically refer to a portion of strain gauge output, unrelated to strain, caused by internal heating of the strain gauge (a resistor) due to its electrical current. I think heat increases the resistance of a resistor, as does positive strain, causing the resistor to output a larger voltage drop. So I assume strain gauge output needs to be corrected for self heating, or thermal output. But like you, I don't think studying strain theory would address this issue of strain gauge circuit heat dissipation. I wonder what order of magnitude or percent error this could cause in the strain readings. Or is it usually not a problem?

 

[prost]

Googled 'apparent stress'--most people are using it to state the value of 'force applied divided by cross sectional area'. Normally they mean the 'initial, undeformed cross sectional area', which is normally different than the deformed area, which can be very hard to measure accurately. If the displacement and strain are 'small', then the undeformed and deformed areas are pretty close in magnitude.

If the displacement and strain are 'small' and the strain gage is in a uniform strain field, shouldn't you be able to calibrate the resistance effect you are citing, vonlueke?

 

[btrueblood]

How do you calibrate a strain gage? I would assume that gages are generally certified on a lot-by-lot basis by the manufacturer, who tests some sample of the lot via a standard coupon and pull test. But calibrating a single gage, i.e. the particular gage you are about to glue to the surface of a load test prototype, is done ... how?

As far as thermal effects, generally strain gages must be either thermally compensated (by adding a chunk of wire or other resistance device with the same tempco and approx. resistance as the gage into the circuit, and then assuming the temperature diff. between wire and gage is negligible); or by keeping the gage and substrate (test article) at a constant, uniform temperature, and using the manufacturer's published temp-vs.-resistance charts to correct the reading. The calibration chain in both sequences is pretty convoluted.

The precision of a typical strain gage measurement can approach one percent or less, but that requires a lot of control of the variables mentioned above. Even the best designed strain gage installations (pressure transducers), the ones reporting 0.25% accuracy, are not accurate to better than +/-1% "out of the box", they always have a bit of zero shift - we always calibrate the transducers using a deadweight tester before installing them, and periodically thereafter. For "in-the-field" measurements using bonded gages, or measurements taken from operating machinery, I would be happy to be within 5-10% of an analytical prediction, and usually that is "good enough" to prove or disprove a failure theory.

 

[prost]

Calibrating a strain gage

http://www.dataq.com/applicat/articles/an23.htm

http://bits.me.berkeley.edu/beam/sg_7.html

to name a couple

 

[btrueblood]

No, prost, the first article is mislabelled. It tells you how to calibrate a pressure transducer, which uses a strain gage. It doesn't tell you how to calibrate a strain gage. The second article discusses using a load simulator (known resistance value in the resistance chain) to simulate a strain-based resistance change in the gage, for the purpose of calibrating one's instrumentation.

What I am trying to point out is that there is no good way to calibrate a strain gage used in load testing: i.e. I am going to bond a gage onto a structure, whose strain vs. load characteristic is not known, or not known very well. I want to know that the gage I am bonding is going to give me an output (e.g. a resistance change that is proportional to strain as in/in X 10-6, or microstrain) that is accurate to within some percentage. The only way to know what the proportionality factor truly is for a given gage is to bond it to a structure, and load the structure. This usually results in the inability to re-use the gage, and in no way could the gage be re-used without recalibration. So: for measuring unknown strains on a not-well-modelled structure undergoing load testing, your measured strain accuracy is only as good as the manufacturer's lot testing and quality control, and your results will vary from ideal by a value that (hopefully) the gage manufacturer can tell you.

Read here:

http://www.omega.com/techref/strain-gage.html

where it states: "In a stress analysis application, the entire gage installation cannot be calibrated as can some pressure transducers. Therefore, it is important to examine potential error sources prior to taking data." It then goes on to list (some) of those error sources. Notably absent from their list is what the "nonlinearity" effect due to large strains is caused by, and what hysteresis effects can occur if large + strain is followed by large - strain on a repetetive basis over many cycles (e.g. a vibration test).

 

[salilkrishnarao]

Hi,
I do feel that for predicting failure/ life of product using FEA CAN be 100 % accurate, provided the Analyst know what he is doing and why he is doing and what he expects out of FEA.
Among many, I will recommend new starters to go through
Dermotmonaghan.com, chalice-Enginnering web pages to learn about Practical FE Analysis.

I feel personally that, if after getting results on FEA, if company wants to make prototpe and test or any kind of measurement to verify FEA Results, Its Ridiculous and pitty on the part of the company.

After purchasing Analysis software worth sufficient odd amount, if one need to validate its results with again huge amount of money. It won't sound good.

I feel personally that nowadays Analysis have forgotten that the basic and pre-requirement for FE Analysis is sound basic understanding of Physics, Mechanics and Strength of Materials along with the sufficient knowledge of Finite Element Method basic theory.
As a rule of thumb, its always better to start the project with question, Do I aspect out Analysis. On what basis I will accept whether results are correct or wrong?

My Own Physophy, FEA always gives correct results for the data/(input) you supplied to software.
Thus it i always necessary to understand physical significance of Boundary conditions and Forces we are inputting.
Hope it sounds ok.

 

[vonlueke]

prost: I think I overstated the resistance effect. (Yes, it could be properly calibrated). After further scanning, it seems the majority of thermal output (apparent strain) is instead probably due to differential thermal expansion of the strain gauge, adhesive, and test part materials. But this would only be an issue if the test temperature greatly differs from the installation temperature. You would also of course get a large apparent strain output, even at room temperature, if one uses the wrong gain or gauge factor settings, etc.

salilkrishnarao: You made some good points, but I don't feel testing is a pity after getting FEA results, due especially to the uncertainty of modeling accurate boundary conditions in certain assemblies in certain projects.

 

[salilkrishnarao]

Hello, Mr VonLueke,
During my starting carreer in FEA, I too use to feel the same that , To validate FE Results we need Testing and some other way. But things have changed lot, I had interactions with testing majors in country and worldwide.
It was pitty to feel on them when they replied that to check load flow path/gradient flowpath they have apply strain gauges everywhere and then confirm the gradient flow path of deflection and the stresses. I personally felt that testing peoples don't have that good understanding of physics of structure and its response to various loading condition. Ok Then FE Analyst can good testing expert.
But then Question come! Thus Budget Justified?

Cost of Analysis software is awesome and then the really high end FE Analyst salaries are generally on sky.
With that can company Justify the cost of testing along with FE Analysis. Can the Design Hours and cost justified?

I don't feel that I became God Father in the field of Analysis. But I use to do the validation in Analysis itself.
General Sources of errors:
1. Meshing ( But considering that one can do that with good Homework, It can be achived 100% good mesh. Also choce of Elements with respect to loading patern and Geometry need to accounted)
2. Boundary Conditions & Loading: You are true. Mojor sorce of erros comes here.
I will try to take my Office time to cover this,
there are 1000 ways you can apply bounday conditions, but what is correct depends upon your judgement, which again comes from sound understanding of Components behaviour under loading which comes from sound basic knowledge of physics, Mechanics and mainly Strength of Materials.

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.
Then comes which one is correct?
So Finally answer is DESIGN INTEND. It means how your part is design. For what your part is designed. How it is supposed to behave and which boundary conditions do you feel giving results matching with the DESIGN INTEND.
I feel that that's the way I do Validation of FEA results.

Again Global major organisation, generally asks FEA vendors what results do they feel that, they are going to give solution to them , before assigning project to consultants.
Thats the test of consultant's capablities.
I can write thiusands words, but other wise my company will thrown out me.
salil