Welcome to the forums. Please post in English or French.

You are not logged in.

#1 2017-05-19 08:27:58

seasoulte
Member
Registered: 2017-05-05
Posts: 7

Friction makes the solution no-convergence and symmetry treatment prob

I am simulating an impact problem using OpenSource code. I found code_aster should be the most powerful package. After two weeks struggling, and asking in the forum, I successfully made case with contact but not friction. It runs well but when I add in the friction, it reports no solution after a while. I am using CONTINUE formulation, COULOMB is used because I want to define the friction coefficient. To reduce the computation load, I also did one case by cutting the impactor and the target to half. To ensure symmetry, I am using AFFE_CHAR_MECA to restrict the movement in y direction.

 AFFE_CHAR_MECA(MODELE=XXX,
                                            DDL_IMPO=(_F(GROUP_MA='ballSym',),
                                                                     DY=0,),),); 

This will cause a single matrice error. If I remove this restriction, the solve can continue but still ends with a non-convergence error. Firstly, I thought it is because duplicated nodes in the mesh, but after examination in Salome, it seems there is no duplicated nodes.

I know it is very troublesome to get the friction on and speedup the simulation. However, it should be doable in code_aster. I also attached an animation done by LS-DYNA to show how the problem is. It is straightforward in DYNA.

Thank you All.
                                                     
Full model first.


Attachments:
60-45fric.zip, Size: 1.1 MiB, Downloads: 34

Offline

#2 2017-05-19 08:30:01

seasoulte
Member
Registered: 2017-05-05
Posts: 7

Re: Friction makes the solution no-convergence and symmetry treatment prob

Half Model.


Attachments:
60-45halffric.zip, Size: 397.61 KiB, Downloads: 30

Offline

#3 2017-05-19 13:55:20

Johannes_ACKVA
Member
From: Ingenieurbüro für Mechanik, DE
Registered: 2009-11-04
Posts: 615
Website

Re: Friction makes the solution no-convergence and symmetry treatment prob

hello,

I downloaded your half-model (60-45halffric.zip) , but there is no symmetry condition in the .comm-file!

If you at any time have applied a symmetry condition which cuts a slave surface with friction then a typical problem appears: certain DOFs are made dependent more than once: The degree of freedom perpendicular to the plane of sym is fixed by the symmetry condition AND it is linked by the friction in that direction. The solution is to exclude the friction linkage of these nodes in the direction perpendic to the sym plane (results remain exactly the same because those nodes do never move in that direction because of the sym fixing)

So this kind of exclusion is done with SANS_GROUP_NO_FR + DIRE_EXCL_FROT. Example:

contact = DEFI_CONTACT(
    ZONE = (_F(
            GROUP_MA_MAIT = 'master',
            GROUP_MA_ESCL =  'sla',
            COULOMB=0.1, 
            SANS_GROUP_NO_FR=('listofgroupsofnodes',),
            DIRE_EXCL_FROT=(0,0,1), # direction perpend to plane of sy
            etc  etc  
           ),),
    FROTTEMENT = 'COULOMB',
    FORMULATION = 'CONTINUE',
   ..... 
   );

After having solved this kind of problem you can benefit from the very shorter CPU-time of your half-model.

Another way to make the analysis much faster (if the movement in the contact are not to big) is to use

contact = DEFI_CONTACT(
    ZONE = (_F(
            ),),
      REAC_GEOM = 'SANS'   # no loop for geometry update
   );

Probably you can find some interesting aspects here

code-aster.org          /forum2/viewtopic.php?id=21300

and - of course - in U2.04.04

Regards
Johannes_ACKVA

______________________________________________________________________
CODE-ASTER-courses at Ingenieurbüro für Mechanik, Germany

*** CODE-ASTER INTRO + CONTACT + MATERIAL
     09-13 October 2017

*** CODE-ASTER  DYNAMIC ANALYSIS
     30 Nov - 01 Dec 2017


Ingenieurbüro für Mechanik
D 91717 Wassertrüdingen / Germany

www.code-aster.de                                                Training & Support for NASTRAN and CODE-ASTER

Offline

#4 2017-05-23 03:08:05

seasoulte
Member
Registered: 2017-05-05
Posts: 7

Re: Friction makes the solution no-convergence and symmetry treatment prob

Thank you, Johannes,

Yes, the one I uploaded is with symmetry condition removed to make the simulation run.

Now I am using SANS_GROUP_NO_FR to exclude the nodes of slave surface at the symmetry plane (those nodes of the master surface are not excluded). It runs and convergences at each time step only when I use an automatic division of time-steps.

DEFLIST1=DEFI_LIST_INST(DEFI_LIST=_F(LIST_INST=timelst,
                                     METHODE='AUTO',
                                     PAS_MINI=1.0E-12),
                        ECHEC=_F(ACTION        = 'DECOUPE',
                                 SUBD_METHODE  = 'MANUEL',
                                 SUBD_PAS      = 4,
                                 SUBD_PAS_MINI = 1e-12),
                        ADAPTATION=_F(EVENEMENT='SEUIL'),
                        );

But it is tremendously slow. Without friction the simulation time is about 3 hours but according to current progress it will take 3 days if friction included.

I also tried to use penalty for continue and frottment, which is really fast, but in that case the impact ball will not rebounce  but stick to the target within the given simulation time (it will continue to deform to the target surface if simulation continues, I guess).

Any suggestion to make the simulation faster, but still can reflect the behavior of rebounce and shape change of the ball and target, because energy loss during this process is one of my concern.

I attached the newest case with symmetry condition and automatic time steps.

Thanks.


Johannes_ACKVA wrote:

hello,

I downloaded your half-model (60-45halffric.zip) , but there is no symmetry condition in the .comm-file!

...
     30 Nov - 01 Dec 2017


Attachments:
halffricautostep.zip, Size: 263.46 KiB, Downloads: 31

Offline

#5 2017-05-23 14:59:19

Johannes_ACKVA
Member
From: Ingenieurbüro für Mechanik, DE
Registered: 2009-11-04
Posts: 615
Website

Re: Friction makes the solution no-convergence and symmetry treatment prob

your .comm file looks good. Sadly, including friction in a contact analysis augments very considerably the CPU-time, so don't be surprised.

What you could try additionally:

(1)
REAC_GEOM = 'CONTROLE', NB_ITER_GEOM=1 instead of REAC_GEOM = 'AUTOMATIQUE'
(see here: http:// code-aster . org / forum2/viewtopic.php?id=21219

(2)
PENALTY contact: I m really not shure if the penalisation of friction causes additional amortissation (damping) . Could you try with a longer simulation time and then confirm you result? Or report that the rebounce happens at later simulation time? Perhaps your penalty stiffness are very low so that you have a deep penetration before the rebounce.

(3)
The HTT-scheme (DYNA_NON_LINE(...SCHEMA_TEMPS=_F(SCHEMA='HHT'...) is known to cause damping (so perhaps it is not the fault of the penalty contact). A non-dissipative scheme is  DYNA_NON_LINE(...SCHEMA_TEMPS=_F(NEWMARK). Since energy loss is an important concern for you you should use NEWMARK

(4)
Since your CPU-time will alway be "long" I would make effort to reduce the number of degrees of freedom (DOF): the same fineness in the region of contact, but strongly biasing to rather large elements away from the contact. Also a coarser mesh of the upper half of the ball.

(5)
a smaller slave surface (fewer slave nodes) will further reduce the number of  DOF (but it is easier achieved with the coarser mesh, see above).
The slave group (for a smaller slave surface) can be created with DEFI_GROUP, so no return to work and split the geometry!

(6)
if you can do with a really very low number of slave nodes (<100 or <50) you could try if the DISCRETE contact with LAGRANGIAN is in this case faster than the  CONTINUE contact (which has it's big advantage for large number of slave nodes)

about (2) and (6) I m really not shure and would be very interested to get you feedback

Regards
Johannes_ACKVA

______________________________________________________________________
CODE-ASTER-courses at Ingenieurbüro für Mechanik, Germany

*** CODE-ASTER INTRO + CONTACT + MATERIAL
     09-13 October 2017

*** CODE-ASTER  DYNAMIC ANALYSIS
     30 Nov - 01 Dec 2017


Ingenieurbüro für Mechanik
D 91717 Wassertrüdingen / Germany

www.code-aster.de                                                Training & Support for NASTRAN and CODE-ASTER

Offline

#6 2017-05-29 07:49:25

seasoulte
Member
Registered: 2017-05-05
Posts: 7

Re: Friction makes the solution no-convergence and symmetry treatment prob

Hi,  Johannes and others,

I tried for one week and here are some findings. In summary, until now the model is still not considered successful.

1) Firstly, of course I coarsen the mesh both in the impactor part and the target part (the target thickness was also reduced). It runs much faster, so I did not change  REAC_GEOM. I tried CONTINUE formulation and STANDARD ALGO_CONT  without friction and it runs very fast, the result shows a reduction of the normal velocity but not the tangential velocity. That is explainable as we do not include frictions.
2) I changed to PENALISATION ALGO_CONT and adjust the COEF_PENA_CONT value. Different value results in different node velocities but we can see it is reasonable and  the value is not too far from the STANDARD ALGO_CONT method.
3) After that I added in the friction and also used the PENALISATION method. Because I checked some documents and found that ABAQUS, DYNA are all using penalty methods. The results are weired. The maximum normal and tangential velocity are all higher than the initial velocity, which I think it is not possible, and it is not seen for frictionless case. The node velocity difference between nodes are much higher, whereas in the DYNA results they values difference is small. I have double checked the elastoplastic material properties in both ASTER and Dyna and I am sure they are the same. Example figures are attached (tangential velocity before and after the rebound).

4) I also tried NEWMARK but the results show very discrete value for different nodes, i.e., some nodes in the "node clouds" appear higher or lower values than the adjacent nodes.


I will continue to learn code_aster and to see if the model can be improved to get similar results as that in Dyna.

Thank you.

Johannes_ACKVA wrote:

your .comm file looks good. Sadly, including friction in a contact analysis augments very considerably the CPU-time, so don't be surprised.

What you could try additionally:

(1)
REAC_GEOM = 'CONTROLE', NB_ITER_GEOM=1 instead of REAC_GEOM = 'AUTOMATIQUE'
(see here: http:// code-aster . org / forum2/viewtopic.php?id=21219

(2)
PENALTY contact: I m really not shure if the penalisation of friction causes additional amortissation (damping) . Could you try with a longer simulation time and then confirm you result? Or report that the rebounce happens at later simulation time? Perhaps your penalty stiffness are very low so that you have a deep penetration before the rebounce.


*** CODE-ASTER INTRO + CONTACT + MATERIAL
     09-13 October 2017

*** CODE-ASTER  DYNAMIC ANALYSIS
     30 Nov - 01 Dec 2017


Attachments:
vx.zip, Size: 37.68 KiB, Downloads: 29

Offline

#7 2017-06-16 10:44:04

Johannes_ACKVA
Member
From: Ingenieurbüro für Mechanik, DE
Registered: 2009-11-04
Posts: 615
Website

Re: Friction makes the solution no-convergence and symmetry treatment prob

seasoulte wrote:

The maximum normal and tangential velocity are all higher than the initial velocity, which I think it is not possible

yes, it is possible: look to your figure after the rebound. The large X-velocity happens because the ball has got a rotation.

Did you create with LS Dyna the same kind of pictures showing something really different? If there are still discrepancies I m rather confident that you can clear them by playing with the integration scheme (NEWMARK seems me the best) and using very very small time steps.

It is also more clear to compare not only  coloured pictures, but also calculating the energies and to sum it up for the whole ball, something like:

CALC_CHAMP(.. ENERGIE...(ECIN_ELEM, ENEL..

Sum=POST_RELEVE_T(ACTION=_F(OPERATION='EXTRACTION',
                                 INTITULE='SumOfEnerg',
                                 RESULTAT=Result,
                                 NOM_CHAM='ECIN_ELEM',
                                 GROUP_NO='Ball' or GROUP_MA='Ball'? i m not sure?
                                 RESULTANTE=('TOTALE',),)

If you have problems with the calculation time I would simplify the whole to an axisymmetric model, just for that comparision with LS.Dyna and for  parameter studies


seasoulte wrote:

..... very discrete value for different nodes, i.e., some nodes in the "node clouds" appear higher or lower values than the adjacent nodes.

I do not understand what you mean

Regards
Johannes_ACKVA

______________________________________________________________________
CODE-ASTER-courses at Ingenieurbüro für Mechanik, Germany

*** CODE-ASTER INTRO + CONTACT + MATERIAL
     09-13 October 2017

*** CODE-ASTER  DYNAMIC ANALYSIS
     30 Nov - 01 Dec 2017


Ingenieurbüro für Mechanik
D 91717 Wassertrüdingen / Germany

www.code-aster.de                                                Training & Support for NASTRAN and CODE-ASTER

Offline

Board footer