How well does FEA predict real world

[McDesign25]

Has anyone validated their predicted FEA stresses or displacements with simulated testing? If for example, you wanted to validate the deflection of a simple rectangular cantilevered beam, how close could you predict displacement? Within 20%, assuming of course you accounted for infinite stiffness, material tolerances, modulus changes, load cell deflection, misalignment? Is it possible to mimic within 10% using a linear FEA solver and a good test setup, assuming you stay in the small deflection region. I have read articles in trade magazines on this manner, but have not heard what the majority of FEA user experiences has been. I too realize this is a luxury to be able to test your FEA results, due to time crunches, etc.

 

[johnhors]

Assuming linear FEA, small deflection (i.e. no plasticity and no non-linear geometry) then on highly complex structures with equally complex loading conditions, I consider any differences between rosette strain gauges and FEA that are lower than 20% to be a success. When differences are greater than this, it is usually possible to find discrepancies between what you have modelled and what was actually tested. On simpler structures (e.g. tube bending) I have seen and would expect agreement to be within 5%. A major cause of error in FEA models are poor boundary conditions applied to be the model, also bear in mind that in most cases it is practically impossible to place a rosette strain guage exactly at a position of peak stress as predicted by FEA. Lastly only consider using rosette guages, a strip guage tells you strain in one direction and nothing more than that!

 

[fkmeyers]

We do a considerable amount of testing and we have found good agreement between test deflections and those predicted by our Nastran solver. As part of our design process we build a test verification FEA model which is representative of the actual test boundary conditions, the test specimen actual dimensions and test loading. This model almost always is within 10% of actual test results and often as good as 5%. The stresses are a different story. Getting these to agree requires a very small strain gage with a precise location that is consistent with the FEA model. Often strain gages span an area of changing strain or are located in areas where the peak stresses are not occurring (due to the inability to locate a gage where ever you really want to). Generally we get strain results that are within 20% of tested values. A lot depends on how the gage is mounted, where, what quality of gage is used, the strain gage size, the adhesive used, the number of cycles, the strain level, etc. If we are close on displacements we know we have the stiffness right.

 

[fkmeyers]

To further clarify we are usually within 10% or better on displacements and 20% or better on strains.

 

[McDesign25]

We test a variety of our plastic molded parts for our product, and I am usually within 20% on displacement, we have not tried to find the stress with strain gauges. Are any of your experiences with molded plastic parts, filled with glass or no glass fibers? If so, did you predict the deflections of polymers with the glass fibers better than polyomers without glass fibers? Thanks for you replies. They make me feel better about what is possible in practice.

 

[johnhors]

McDesign25 - only with high tensile steel alloys, aluminium and titanium, no plastics or polymers.

 

[fkmeyers]

We do have experience with composites and I can tell you they are a real problem to get any decent strain readings from. The adhesive that bonds the gage is an issue as well. If you grind down any coatings and have a smooth surface you can get within 30-40% most of the time but the technician mounting the gages needs to know what they are doing. I recommend coupon testing first to establish techniques and correlation to known values before testing any real structures.

 

[McDesign25]

fkmeyers,

Could you explain coupon testing? Are you referring to standard tensile bars?

Thanks.

 

[fkmeyers]

Yes. These are typically flat dog bone shaped coupons with a strain gage located at the center of the coupon. The purpose of the testing is to establish bonding techniques. The testing is very predictable with known strain states based on extentiometer readings that can be compared to the readings from the mounted strain gage.

 

[GregLocock]

I'm designing a welded box section chassis for an old car. We've measured the deflection of the old chassis along its length for a couple of load cases and are generally much better than 10% error (obviously you can get huge %age errors if the displacements are small). So on a simple cantilever you should be able to get much better than 10%.



Cheers

Greg Locock

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

 

[fkmeyers]

I agree. We normally see less than 5% error on displacements for metallic structures especially when you are out of the noise of the instrumentation. If the displacements are very small there are other issues that garbage the data and very precise instrumentation is required which is typically not used. In our testing displacement gages are usually bonded to the structure using dental cement and small metallic tabs and then attached to wires which are fixed to the gage and then to a "rigid" structure. Proper calibration eliminates most of the error in the system itself.

 

[IRstuff]

FEA is also only as good as the input. For certain applications, every nut and bolt and every joint can make a difference. In our applications, we're interested in angular deflections on the order 10s to 100s of microradians. FEA has generally failed miserably for us in predicting ultra-small angle deflections.

TTFN

 

[McDesign25]

IR Stuff,

What inputs are you referring too? Loads or Restraints or Mesh or Material Properties? Could you define failing miserably? What methods do you use to predict deflections then, experimental such as strain gages, etc?

 

[IRstuff]

Input, as in the modeled unit. The level of versimilitude is a direct function of the degree of dependence of the behavior on the relative contributions of the components in the unit.

Failing miserably as in a predicted deflection of 20 microradians and getting milliradians instead. Measured with theodolites.

TTFN

 

[rkn40]

It is difficult to come up with an absolute percentage. The corrrelation depends on the geometery, material properties and boundary conditions. Also, there is always some variability due to test setup. Considering all this, less than 20% variation for linear static analysis and 10% variation for modal analysis is not uncommon. Please read this article on correlating linear static analysis

http://www.feadomain.com/content.php?article.48

 

[Baheej]

Two years ago I was working as a tech support engineer for a CAD and FEA software company. Most of the customers seemed satisfied to be within 10-20% in their predictions (both stress and displacement) for simple models/cases. Others seemed to expect a better agreement.
Often it is easy to blame the model, because the solid data does not lie (although it can be deceiving or not reporting what you are looking for), but as IRstuff pointed out, the output has a lot to do with the input.

 

[GregLocock]

As an ex development engineer I think there is some validity in questioning the test data - in fact measuring static stiffness is quite difficult because it is very hard to get good terminations. For my projects I often end up modelling the test rig as well as well.

The best test data (the most difficult to mess up) is free-free modal data. However that is also the sort of test that bears the least resemblance to the actual operating conditions. The checks that can be done on a modal survey are quite thorough, so the test engineer can be quietly confident of his results long before the FEA people get involved.

I'd have to say >75% of the time the FEA model is mostly to blame for significant differences between test and analytical results. And it is very rarely a coding problem.







Cheers

Greg Locock

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