Remark: In verification of FE analysis one of the  important issue is to show the convergence wrt mesh density. In fact you show that the result is stabilizing /converging/ if you are computing with a set of FE models with refined meshes for the same mechanical model.

Dear Dr. Tarkes Dora Thanks for sharing your knowledge

Regards

Prabhakar P

Remark: In verification of FE analysis one of the  important issue is to show the convergence wrt mesh density. In fact you show that the result is stabilizing /converging/ if you are computing with a set of FE models with refined meshes for the same mechanical model.

Thank you Alexander Konyukhov , As you explained above the answer was  very useful stating it is must to do convergence study to identify the error difference in finite element result and to ensure the solution is converged

I agree with all answers.

Typically, the percentage error levels for the case studies  are within  0-20 %.

The lower the error, the more reliable is FE model.

Dear Daniela Siano

Thanks for sharing your knowledge and stating more accurate the FE model the more accurate the result we get. by considering the complexity of the problem the fine mesh has to be done in the area of focus which helps to get most accurate result and thanks again for clarifying 10-20 % as ideal error level

Regards

Prabhakar P

I think accepted error level depends on the problems complexity and materials too.

If the problem is simple and the material is noble error must be less then 5%

when it involves some complexities (interraction, contact, complex rehology) it could be greater then 10%.

In some cases when datas are not available (geotech) the error could be G.T 20%.

Dear Sir,

Thanks for sharing your knowledge and thanks again for sharing the information that at any cost the error should not exceed 20 %

Regards

Prabhakar

One more point that you must consider is how you will cite this information about reasonable errors in your publications. You should do a research in reliable journals in your specific area, looking for papers where the authors have made comments or drawn conclusions based on this error margin for FE problems. This way, you will show more consistence when stating that 10% or 20% is accepted.

Maybe some of these researchers who have answered your question may suggest their own papers. In my case, my FEA results are compared with theoretical ones (analytical solution) or other author's results, and the error is about 3% to 5%. But these are not experimental errors, which are usually bigger indeed.

Thank you

Dr. Wederley Miranda

For sharing the information, that it is must to know how we compare and to which source we compare the FE analysis result before justification of error

Could be relevant to you

Dr. Krishnan Umachandran

Thank you for the article

I saw this question very late and You have already got so many good answers….still I would like to add my views:

There is a famous quote of Richard Feynman

“It doesn't matter how beautiful your theory is, it doesn't matter how smart you are. If it doesn't agree with experiment, it's wrong”

and as per my understanding this is in the background of validation and verification, simulation is useless if it is not matching with actual test data. Now your question is with how much accuracy?

Depends on measurement…..e.g. if you are subjecting a structure for some mechanical loads and you are measuring deformations through LVDTs and strain through strain gauges….first check deformations. As displacements are indicator of global behavior of structure they must be in very good agreement with your predictions… if not some serious problem in your simulation/understanding….pattern as well as magnitude……if pattern is out… you are totally wrong… for variation in magnitude there must be some good reason…..

Strains give local behavior….if good number of strain gauges (say 80% of total number) are matching with simulation result with a good accuracy….and remaining 20% are out….you need some local measurement at gauge location and need to correct your simulation…..

It depends on geometry and element used in simulation (assuming all convergence studies are carried out and you are simulating correctly)…..I have personal experience of testing thin pressure vessels for internal pressure that were simulated with axisymmetric elements in ANSYS match was >95% by magnitude almost for all gauges and DTs…..for thermal problems with RTD measurement it comes within 1deg C accuracy (again depends on problem)…..Both the above mentioned tests were relatively simple so accuracy of prediction was very high….but I have experienced in complex structures with complex loadings at some places there were very large differences and we were not able to find the reason till last…so as your experience with one particular test will increase and if you keep on updating your simulation after every test…..it will move towards perfection…..but don’t try to find out a number to relax your work...as many people have already pointed out for first time 10-20% will be a good number....but in my personal opinion parameters that give global picture should match with in 10% even in first test….try for the best match every time….

FEM does not solve for a single element, it is assembled and solved as a whole (FDM, on the other hand, sweeps through a mesh and solves implicitly, element by element)

http://www.eos.ubc.ca/courses/eosc433/lecture-material/L9-DeformationAnalysis.pdf

Dear All,

Is there a reference one can site as to the acceptable percent error?

Good question.. Please share me the best answer might you trust...

Regards…