Test results vs FEA results. 5

[prost]

johnhors: because strain gages seem to work well only in areas of small to no strain gradients, and work not so well in the really interesting areas, the high strain gradient areas, then they would appear to be of limited utility for correlation with numerical analysis. I have only limited experience with correlation of strain gage readings with FEA, so perhaps others could enlighten the rest of us ignorants.

Photoelastic coatings work very well in many many cases. There is also a lot of research in thermoelastic measurement techniques that use infrared to detect tiny changes in temperature which sometimes can be correlated very well with strains and stresses.

Every measurement technique, as well as every numerical method, has problems, so I wouldn't discount any technique just because it has limitations, I would only make sure those limitations weren't distorting my results.

 

[gfbotha]

Prost, you may also enlighten me about photoelastics - I'm under the impression it is only useful in the elastic range of the photoelastic material? So, in context of this thread which is about substantial plastic yielding of metal, will the photoelastic material typically have sufficient range to correlate in?

Thanks

 

[johnsmith2]

photoelastic material need to be transparent? So. light can pass through.

 

[prost]

subject: photoelastic measurements of plasticity in metals

Don't have this article, otherwise I'd share it.

http://www.springerlink.com/content/t8l13p625u675843/

 

[xerf]

http://www.doitpoms.ac.uk/tlplib/photoelasticity/index.php

 

[GregLocock]

A technique I have heard of, but never used, is SPATE.

http://cat.inist.fr/?aModele=afficheN&cpsidt=1128136

GenRad or SDRC were promoting it in the late eighties.

Cheers

Greg Locock

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

 

[FEAHappy]

Typically, I don't have time for runs to check for convergence. I check the "energy norm error." (the difference between nodal and elemental stresses) My usual practice is to refine the mesh, with mesh controls, finer and finer until the meshing and run time slows.

"Finer the better" with H elements draws negative comments, However, I have not heard a refutation of this practice. In meshing finer and finer, I have seen a realistic appearing stress gradient change to a stress field containing spots of high stress that make no sense. This occured with a very fine mesh; (Global mesh/100)

In one reply, and in reality, inexpensive FEA tools are criticized as simplistic or useless. I never hear why, just that these are "toys" . Without naming names, I have seen excellent correlation between a lower cost and a higher cost package in a complex linear stress analysis of an assembly.

In the world that I have worked in for 30 years, engineers do not have master's degrees in engineering mechanics, and companies will not invest in the high end FEA packages.

Often, checking with hand calculations is not useful due to simplifying assumptions. Checking with a simple FEA model is one technique. Perhaps a more useful question than checking with hand calculation might be
"Did you do a free body diagram?"

There is a value in using FEA in that the problem must be defined. The user must think about the boundary conditions.

 

[xerf]

Without naming names, I have seen excellent correlation between a lower cost and a higher cost package in a complex linear stress analysis of an assembly.

It is not difficult to program the code for doing a linear stress analysis without meshing capabilities. Actually, there exist free FEA tools which are far of being simplistic.

 

[GregLocock]

"In one reply, and in reality, inexpensive FEA tools are criticized as simplistic or useless. I never hear why, just that these are "toys" . Without naming names, I have seen excellent correlation between a lower cost and a higher cost package in a complex linear stress analysis of an assembly.
"

No, I said of limited use. Don't misquote me. I did not say they were toys.

Tell me, if they were as useful as full powered products, why would people pay for the full powered products? All you have to do is compare the feature-list, typically, to see what the differences are. At least one of the bundled products cannot handle any complexity in boundary conditions, and makes dangerous assumptions about them, and I doubt any of them can deal with assemblies.

So far as 'toys' go FWIW at home I do the vast majority of my FEA in the 200 node demo version of nastran for windows. And get very good correlation- typically within 5% of the correct (measured) static stiffness. But, I have full control of the model.

Cheers

Greg Locock

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

 

[GBor]

Tell me, if they were as useful as full powered products, why would people pay for the full powered products?

I think it is generally support. If you are only doing linear, static analysis, there are a variety of GREAT product ranging from free to $20k that will all do a great job, but if you have any questions about how to run them, the ones that cost generally come with better support.

All you have to do is compare the feature-list, typically, to see what the differences are.

Glad you added the word "typically".

At least one of the bundled products cannot handle any complexity in boundary conditions, and makes dangerous assumptions about them, and I doubt any of them can deal with assemblies.

There are a few that handle assemblies depending on what you mean by "bundled", what you are trying to do with the assembly, and how you create the CAD model.

I think the "moral" to this story is, "Prepare yourself with the knowledge of your own limitations and the limitations of your tools or those for whom you design your products may pay for it with their lives"

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

 

[cbrn]

Hi,
Garland and Greg, I personally think that you both hit good points.
While it's true that in the linear elastic, static, domain (and not only, I'd say...) the FE solvers behave all almost the same way (hopefully!!!), be them hi-end ultra-expensive systems like Ansys Multiphysics (but then, why get the Multiphysics license level if it's only for linear static?) or 100%-free packages, when it comes to capabilities in handling "exotic" boundary conditions then the music changes: for single example (and only to name systems I directly know or knew), up to v.2004 (and perhaps v.2005) CosmosWorks couldn't handle non-holonomic boundary conditions, while this capability has been existing in ANSYS at least since v.5.3 with COMBIN37 element. And the list could be VERY longer... Of course, the model you have to analyze can be as simple as you want, but as Greg says if you need a particular boundary condition and your package doesn't handle it, you'll have no chance to get any realistic result, I think.

 

[johnhors]

If you supply analysis work for certain (large) companies, you have no choice about what analysis software you use to do the job.