ERRE_NOEU_ELEM and ERRE_ELEM_SIGMA

ldc · 2008-12-08 00:54:56

Hi to all forum,

Can somebody explain me how to use the ERRE_NOEU_ELEM and ERRE_ELEM_SIGMA error estimators?
How to use their option? Does it exist reference documentation? I looked for it but could not find anything.

I attachment a simple 3D model with .comm file.

Thanks in advance to anybody would help me.

Luigi

Attachments:
ErrorTest.zip, Size: 573.72 KiB, Downloads: 419

delmas · 2008-12-08 11:26:15

Hello Luigi,

In Code_Aster exists few types of error estimators : estimators in energy norm and estimator in quantities of interest.

For the first type, there are two methods available in Code_Aster
- Only for mechanical analysis and 2D. Recovery based estimators (Zienkiewicz and Zhu like) : you have to compute the recovered stress field with SIGM_NOZ1_ELGA or SIGM_NOZ2_ELGA and the error estimated field with ERZ1_ELEM_SIGM or ERZ2_ELEM_SIGM. A reference : http://www.code-aster.org/DOCASTER/EDF/ … 41001a.pdf
- For mechanical and thermal analysis in 2D and 3D. Explicit residual based estimators (Babuska and Rheinboldt like) : SIGM_ELNO_DEPL and ERRE_ELEM_SIGM or ERRE_ELEM_TEMP. References : http://www.code-aster.org/DOCASTER/EDF/ … 41002b.pdf http://www.code-aster.org/DOCASTER/EDF/ … 41003a.pdf

For the second type, it is a bit more complicated. You have to compute a primal problem and a dual problem. A reference : http://tel.archives-ouvertes.fr/tel-00311947/en/ (there is an exemple of Code_Aster command file at the end)

About the question how to use : http://www.code-aster.org/DOCASTER/EDF/ … 20801b.pdf
These options provide an error field. The highest values show you where you have to refine your mesh to improve the accuracy of your results.

In Code_Aster, you can refine your mesh with the command MACR_ADAP_MAIL.

The difference between ERRE_NOEU_ELEM and ERRE_ELEM_SIGMA is the localisation of the values (nodes for the first one and element for the second one).

I hope that it will help you.

ldc · 2008-12-08 16:45:41

delmas wrote:

A reference :
http://www.code-aster.org/DOCASTER/EDF/ … 41001a.pdf
http://www.code-aster.org/DOCASTER/EDF/ … 41002b.pdf
http://www.code-aster.org/DOCASTER/EDF/ … 41003a.pdf
http://www.code-aster.org/DOCASTER/EDF/ … 20801b.pdf

Hello delmas, thanks a lot for your reply. Unfortunely, I have no access to documents you indicated, they seems to be reserved to authorized people only (maybe EDF developers?).

Is there some way to have them?

Thanks again for your help, best regards.

Luigi

delmas · 2008-12-09 07:02:24

The links available from internet:

http://www.code-aster.org/DOCASTER/Man_R/R4/R41001a.pdf
http://www.code-aster.org/DOCASTER/Man_R/R4/R41002b.pdf
http://www.code-aster.org/DOCASTER/Man_R/R4/R41003a.pdf
http://www.code-aster.org/DOCASTER/Man_U/U2/U20801b.pdf

ldc · 2008-12-09 17:12:20

Hello delmas,

thanks for your hints, I will study everything.

I have another problem now on my model. I run the model and asked for error estimation, as in the follow. Aster gets the following alarm messages and does not calculate the error estimation. It says champs are inexistant, but I asked for the solution of these fields, you can check from the .mess file.

Where is the error? I made a lot of trials, without any results. By the way, the model is a mixed beam - solid one. Whit another model, built only with bricks, it works perfectly and I can get the error estimation.

Thanks in advance for your help.

Luigi

Solution=CALC_ELEM(reuse = Solution,
INFO=1,
TOUT_ORDRE='OUI',
RESULTAT=Solution,
OPTION=('SIGM_ELNO_SIEF','ERRE_ELEM_SIGM','EQUI_ELNO_SIGM','SIGM_ELNO_DEPL','SIEF_ELNO_ELGA','ERRE_ELNO_ELEM','EFGE_ELNO_DEPL',),
TYPE_OPTION='TOUTES',
CRITERE='RELATIF',
PRECISION=1.E-3,
MODELE=FEMQuad,
CHAM_MATER=Mat,
);

!----------------------------------------------------------------!
! <A> <UTILITAI8_14> !
! !
! !
! Champ inexistant SIEF_ELNO_ELGA !
! !
! ou EFGE_ELNO_DEPL !
! !
! NUME_ORDRE 1 on ne calcule pas l'option SIGM_ELNO_SIEF !
! !
! !
! !
! Ceci est une alarme. Si vous ne comprenez pas le sens de cette !
! alarme, vous pouvez obtenir des résultats inattendus ! !
!----------------------------------------------------------------!

!---------------------------------------------------------------------------!
! <A> <CALCULEL3_5> !
! !
! !
! pour calculer SIGM_ELNO_SIEF il faut SIEF_ELNO_ELGA ou EFGE_ELNO_DEPL !
! !
! !
! !
! Ceci est une alarme. Si vous ne comprenez pas le sens de cette !
! alarme, vous pouvez obtenir des résultats inattendus ! !
!---------------------------------------------------------------------------!

!----------------------------------------------------------------!
! <A> <UTILITAI8_14> !
! !
! !
! Champ inexistant SIGM_ELNO_DEPL !
! !
! ou SIEF_ELNO_ELGA !
! !
! ou SIRE_ELNO_DEPL !
! !
! NUME_ORDRE 1 on ne calcule pas l'option ERRE_ELEM_SIGM !
! !
! !
! !
! Ceci est une alarme. Si vous ne comprenez pas le sens de cette !
! alarme, vous pouvez obtenir des résultats inattendus ! !
!----------------------------------------------------------------!

!----------------------------------------------------------------!
! <A> <UTILITAI8_14> !
! !
! !
! Champ inexistant ERRE_ELEM_SIGM !
! !
! NUME_ORDRE 1 on ne calcule pas l'option ERRE_ELNO_ELEM !
! !
! !
! !
! Ceci est une alarme. Si vous ne comprenez pas le sens de cette !
! alarme, vous pouvez obtenir des résultats inattendus ! !
!----------------------------------------------------------------!

Attachments:
Urban_40_v9.mess, Size: 34.15 KiB, Downloads: 421

delmas · 2008-12-10 19:02:04

About the meaning of your error messages, the order of the names is important. You need the last fields to compute the first one....

('SIGM_ELNO_DEPL',
 'SIEF_ELNO_ELGA',
 'SIGM_ELNO_SIEF',
 'EFGE_ELNO_DEPL',
 'EQUI_ELNO_SIGM',
 'ERRE_ELEM_SIGM',
 'ERRE_ELNO_ELEM',),

But the error estimator does not work with beam models, only with C_PLAN, D_PLAN, AXIS or 3D models...

Post your files, I'll try to think about error estimation in such situation.

ldc · 2008-12-12 19:33:18

Hi delmas,

I followed your advice, not it works!!!

If you want to check my model, I have to send to your personal address, it is too big to post on forum (more than 10 Mb).
I am trying to estimate stress near weldings on a minibus frame.

I wold like to know your opinion about quality of solution near holes (see attachment). Elements are quadratic.

Let me know if you want me to post the model.

Many many thanks for the moment, ciao.

Luigi

Attachments:
ErrorImages.zip, Size: 986.92 KiB, Downloads: 470

ldc · 2008-12-14 16:46:29

Hi again,

I read the documentation you advised me. It is very useful but.. now I have more questions.

For sample, If I use error estimator TERMS2 on element solution: this is relative error of the stress jump between elements. What "relative" does it mean? What's the total"? Total strain energy? Total jump error? In that case what could be an orientative value to estimate if the solution is good or not on the elements of interest?
For sample, in ansys there is an error that compares the energy norm on elements with the total deformation energy. I know, this value should not be more than 30-40%. I now is a rough estimation but is a starting point....
Is there in aster something similar?

ERREST seems to be different from that type of estimator, so what's could be an orientative value?

At the end we know that error estimation depends also on the sensibility of the engineer. So maybe, we could change name to this topic and try to involve more people on the forum to share their experience on error estimation. It could be very interesting to have numeric values as a reference (error near boundary, error near holes, between contact surfaces, near fillett etc.).

I looked similar arguments on documentation but I did not find anything. Same thing on forum topic.

Thanks in advance for your reply, ciao.

Luigi

delmas · 2009-01-29 12:02:05

Dear Luigi,

Sorry for my late answer.

About your little questions : in Code_Aster, the field "ERREUR" has many components. But you have to consider only the two first (ERREST and NUEST). ERREST is an absolut value and NUEST is relative value (absolut value normalized with the stress tensor norm).

The question of error estimation is wide.

Error estimation depends on what you want to know and how much you want to pay for it. If you want a very accurate result, you need a very fine mesh. But if you don't want to have a very fine mesh you need a very good error estimator (very expensive too). Often, we try to be in the middle of both solution...

In Code_Aster, error estimators are error indicators : they show were the error is important but you can't use the value of the error. (my result is 2, my error is 0.5, the true result is 2.5). That's why you have to use this information in an adaptive process. Very accurate estimators exist but they are very expensive (more than the analysis?!) and then usless in industrial structural analysis.

The huge mistake is to think that we can share numerical values because the error depends on the geometry, the way you respect the geometry, the mesh refinment, the elements, the loads. And when you solve the same problem, what is acceptable does not depend on the engineer but the final goal of the computation.

But i agree that it will be very interesting to compare experience on error estimation.

JD

#1 2008-12-08 00:54:56

ERRE_NOEU_ELEM and ERRE_ELEM_SIGMA

#2 2008-12-08 11:26:15

Re: ERRE_NOEU_ELEM and ERRE_ELEM_SIGMA

#3 2008-12-08 16:45:41

Re: ERRE_NOEU_ELEM and ERRE_ELEM_SIGMA

#4 2008-12-09 07:02:24

Re: ERRE_NOEU_ELEM and ERRE_ELEM_SIGMA

#5 2008-12-09 17:12:20

Re: ERRE_NOEU_ELEM and ERRE_ELEM_SIGMA

#6 2008-12-10 19:02:04

Re: ERRE_NOEU_ELEM and ERRE_ELEM_SIGMA

#7 2008-12-12 19:33:18

Re: ERRE_NOEU_ELEM and ERRE_ELEM_SIGMA

#8 2008-12-14 16:46:29

Re: ERRE_NOEU_ELEM and ERRE_ELEM_SIGMA

#9 2009-01-29 12:02:05

Re: ERRE_NOEU_ELEM and ERRE_ELEM_SIGMA

Board footer