Convergence issue (general tips)

[bjorn_fallqvist]

Hello,

I am having difficulty getting a model to converge. It is a model which involves very large deformations (but no contact). It runs very stable up to a certain point where it basically "runs into a wall". I am guessing that up to this point the behaviour is smooth, but then when the material at this point starts to stretch more and more, it gets much harder to find a solution. The problem can be generally described as pressurizing a bag.

This is what the convergence history looks like:

Time        Percent     DT       L Iter. Residual
   0.000000   0.00     0.010000  1   0   0.0000E+00
   0.010000   0.33     0.010000  1   1   7.9477E-01
   0.010000   0.33     0.010000  1   2   1.0709E-02
   0.010000   0.33     0.010000  1   3   8.5464E-04c
   0.020000   0.67     0.010000  1   1   6.6037E-01
   0.020000   0.67     0.010000  1   2   7.5254E-03c
   0.030000   1.00     0.010000  1   1   4.7514E-01
   0.030000   1.00     0.010000  1   2   5.8254E-03c
   0.040000   1.33     0.010000  1   1   3.4853E-01
   0.040000   1.33     0.010000  1   2   4.7869E-03c
   0.050000   1.67     0.010000  1   1   2.7449E-01
   0.050000   1.67     0.010000  1   2   4.3138E-03c
   0.060000   2.00     0.010000  1   1   2.2872E-01
   0.060000   2.00     0.010000  1   2   4.1690E-03c
   0.070000   2.33     0.010000  1   1   1.9860E-01
   0.070000   2.33     0.010000  1   2   4.1572E-03c
   0.080000   2.67     0.010000  1   1   1.7709E-01
   0.080000   2.67     0.010000  1   2   4.1921E-03c
   0.090000   3.00     0.010000  1   1   1.6029E-01
   0.090000   3.00     0.010000  1   2   4.2438E-03c
   0.100000   3.33     0.010000  1   1   1.4614E-01
   0.100000   3.33     0.010000  1   2   4.3069E-03c
   0.110000   3.67     0.010000  1   1   1.3364E-01
   0.110000   3.67     0.010000  1   2   4.3929E-03c
   0.120000   4.00     0.010000  1   1   1.2231E-01
   0.120000   4.00     0.010000  1   2   4.5402E-03c
   0.130000   4.33     0.010000  1   1   1.1191E-01
   0.130000   4.33     0.010000  1   2   4.8458E-03c
   0.140000   4.67     0.010000  1   1   1.0232E-01
   0.140000   4.67     0.010000  1   2   5.5194E-03c
   0.150000   5.00     0.010000  1   1   9.3600E-02
   0.150000   5.00     0.010000  1   2   6.9283E-03c
   0.160000   5.33     0.010000  1   1   8.6519E-02
   0.160000   5.33     0.010000  1   2   9.4035E-03c
   0.170000   5.67     0.010000  1   1   8.5322E-02
   0.170000   5.67     0.010000  1   2   1.1362E-02
   0.170000   5.67     0.010000  1   3   7.8842E-03c
   0.180000   6.00     0.010000  1   1   1.1110E-01
   0.180000   6.00     0.010000  1   2   1.1999E-02
   0.180000   6.00     0.010000  1   3   5.7486E-03c
   0.190000   6.33     0.010000  1   1   1.8934E-01
   0.190000   6.33     0.010000  1   2   1.9690E-01
   0.190000   6.33     0.010000  1   3   1.0105E-02
   0.190000   6.33     0.010000  1   4   7.1759E-03c
   0.200000   6.67     0.010000  1   1   2.7777E-01
   0.200000   6.67     0.010000  1   2   2.5949E-01
   0.200000   6.67     0.010000  1   3   9.0875E-03c
   0.210000   7.00     0.010000  1   1   3.0643E-01
   0.210000   7.00     0.010000  1   2   2.4690E-01
   0.210000   7.00     0.010000  1   3   7.2738E-03c
   0.220000   7.33     0.010000  1   1   2.5610E-01
   0.220000   7.33     0.010000  1   2   1.4178E-01
   0.220000   7.33     0.010000  1   3   6.5146E-03c
   0.230000   7.67     0.010000  1   1   1.8876E-01
   0.230000   7.67     0.010000  1   2   5.5400E-02
   0.230000   7.67     0.010000  1   3   4.3195E-03c
   0.240000   8.00     0.010000  1   1   1.4351E-01
   0.240000   8.00     0.010000  1   2   1.9818E-02
   0.240000   8.00     0.010000  1   3   3.0770E-03c
   0.250000   8.33     0.010000  1   1   1.1547E-01
   0.250000   8.33     0.010000  1   2   7.4187E-03c
   0.260000   8.67     0.010000  1   1   1.0545E-01
   0.260000   8.67     0.010000  1   2   4.9805E-03c
   0.270000   9.00     0.010000  1   1   9.0345E-02
   0.270000   9.00     0.010000  1   2   3.8216E-03c
   0.280000   9.33     0.010000  1   1   7.9443E-02
   0.280000   9.33     0.010000  1   2   3.7428E-03c
   0.290000   9.67     0.010000  1   1   7.1928E-02
   0.290000   9.67     0.010000  1   2   3.6637E-03c
   0.300000  10.00     0.010000  1   1   6.6100E-02
   0.300000  10.00     0.010000  1   2   3.5460E-03c
   0.310000  10.33     0.010000  1   1   6.1110E-02
   0.310000  10.33     0.010000  1   2   3.4560E-03c
   0.320000  10.67     0.010000  1   1   5.6546E-02
   0.320000  10.67     0.010000  1   2   3.4840E-03c
   0.330000  11.00     0.010000  1   1   5.2184E-02
   0.330000  11.00     0.010000  1   2   3.5791E-03c
   0.340000  11.33     0.010000  1   1   4.7930E-02
   0.340000  11.33     0.010000  1   2   3.7271E-03c
   0.350000  11.67     0.010000  1   1   4.3783E-02
   0.350000  11.67     0.010000  1   2   3.8994E-03c
   0.360000  12.00     0.010000  1   1   3.9827E-02
   0.360000  12.00     0.010000  1   2   4.0531E-03c
   0.370000  12.33     0.010000  1   1   3.6279E-02
   0.370000  12.33     0.010000  1   2   4.4362E-03c
   0.380000  12.67     0.010000  1   1   3.3807E-02
   0.380000  12.67     0.010000  1   2   8.1883E-03c
   0.390000  13.00     0.010000  1   1   3.1914E-02
   0.390000  13.00     0.010000  1   2   9.8297E-03c
   0.400000  13.33     0.010000  1   1   2.8432E-02
   0.400000  13.33     0.010000  1   2   7.0273E-03c
   0.410000  13.67     0.010000  1   1   2.4682E-02
   0.410000  13.67     0.010000  1   2   5.3665E-03c
   0.420000  14.00     0.010000  1   1   2.1676E-02
   0.420000  14.00     0.010000  1   2   4.9013E-03c
   0.430000  14.33     0.010000  1   1   1.9327E-02
   0.430000  14.33     0.010000  1   2   4.8288E-03c
   0.440000  14.67     0.010000  1   1   1.7582E-02
   0.440000  14.67     0.010000  1   2   4.8893E-03c
   0.450000  15.00     0.010000  1   1   1.6444E-02
   0.450000  15.00     0.010000  1   2   5.0106E-03c
   0.460000  15.33     0.010000  1   1   1.5903E-02
   0.460000  15.33     0.010000  1   2   5.1718E-03c
   0.470000  15.67     0.010000  1   1   1.6018E-02
   0.470000  15.67     0.010000  1   2   5.3588E-03c
   0.480000  16.00     0.010000  1   1   1.6664E-02
   0.480000  16.00     0.010000  1   2   5.5736E-03c
   0.490000  16.33     0.010000  1   1   1.7780E-02
   0.490000  16.33     0.010000  1   2   5.8154E-03c
   0.500000  16.67     0.010000  1   1   1.9282E-02
   0.500000  16.67     0.010000  1   2   6.0829E-03c
   0.510000  17.00     0.010000  1   1   2.1075E-02
   0.510000  17.00     0.010000  1   2   6.3763E-03c
   0.520000  17.33     0.010000  1   1   2.3076E-02
   0.520000  17.33     0.010000  1   2   6.7005E-03c
   0.530000  17.67     0.010000  1   1   2.5264E-02
   0.530000  17.67     0.010000  1   2   7.0504E-03c
   0.540000  18.00     0.010000  1   1   2.7584E-02
   0.540000  18.00     0.010000  1   2   7.4320E-03c
   0.550000  18.33     0.010000  1   1   2.9998E-02
   0.550000  18.33     0.010000  1   2   7.8440E-03c
   0.560000  18.67     0.010000  1   1   3.2505E-02
   0.560000  18.67     0.010000  1   2   8.2869E-03c
   0.570000  19.00     0.010000  1   1   3.5086E-02
   0.570000  19.00     0.010000  1   2   8.7623E-03c
   0.580000  19.33     0.010000  1   1   3.7757E-02
   0.580000  19.33     0.010000  1   2   9.2859E-03c
   0.590000  19.67     0.010000  1   1   4.1369E-02
   0.585000  19.50     0.005000  2   1   6.6893E-02
   0.582500  19.42     0.002500  3   1   8.1757E-02
   0.582500  19.42     0.002500  3   2   1.6046E-02
   0.581250  19.38     0.001250  4   1   9.2190E-02
   0.581250  19.38     0.001250  4   2   1.2774E-02
   0.581250  19.38     0.001250  4   3   8.9851E-03c
   0.582500  19.42     0.001250  4   1   1.9329E-01
   0.581875  19.40     0.000625  5   1   1.7487E-01
   0.581563  19.39     0.000313  6   1   1.6868E-01
   0.581406  19.38     0.000156  7   1   1.6703E-01
   0.581328  19.38     0.000078  8   1   1.6668E-01
   0.581289  19.38     0.000039  9   1   1.6664E-01
   0.581270  19.38     0.000020 10   1   1.6666E-01
   0.581260  19.38     0.000010 11   1   1.6668E-01
   0.581255  19.38     0.000005 12   1   1.6669E-01
   0.581252  19.38     0.000002 13   1   1.6670E-01
   0.581251  19.38     0.000001 14   1   1.6670E-01
 5.8125E-01  19.38   6.1035E-07 15   1   1.6670E-01
 5.8125E-01  19.38   3.0518E-07 16   1   1.6670E-01
 5.8125E-01  19.38   1.5259E-07 17   1   1.6670E-01
 5.8125E-01  19.38   7.6294E-08 18   1   1.6670E-01
 5.8125E-01  19.38   3.8147E-08 19   1   1.6670E-01

So as you can see something drastic happens at about 19%, but I'm unsure exactly what it is. Neither the stresses, nor the displacements seems to make a jump" when I plot it against time. Maybe it has something to do with the elements starting to experience an abrupt change in stiffness (previously it was more or less pure bending, but as the bag tightens, it becomes almost pure membrane stress)? Could it be some kind of locking?

[S.LI]

As you mentioned here, the problem is because of material/structure softening.

The converged solution is very hard to be located, especially for force controls.

Try to see if you can simulate your problem with Riks method, though there are still limitations in Algor Riks.

[shakeel.mirza]

"I am guessing that up to this point the behaviour is smooth, but then when the material at this point starts to stretch more and more, it gets much harder to find a solution."

Hi Bjorn,

As a general rule, in such case, you might have to start with a fine mesh in that area of deformation, so that during the analysis, the mesh quality is not affecting your analysis.

Do you see highly deformed/distorted mesh in your analysis at that point convergence problem ?

Regards,

Shakeel

[bjorn_fallqvist]

Hi, Shakeel

It doesn't look very distorted to me, and I haven't gotten any warnings. I already have a quite refined mesh in the areas experiencing large deformations, and I making them even finer would reduce the running time too much.

I'm currently trying to run it as a MES Riks analysis...will there be a possibility to run this with pressure and surface loads in the future?

Best regards

Björn

[shakeel.mirza]

Hi Bjorn,

All depends on the particular problem that you are having with this model.

Will it be possible to share the model (of course that will be used internally for support purpose) ?

Please feel free to write me directly to my email  shakeel.mirza[at]autodesk.com.

Seems that you are in European time zone as well.

Regards,

Shakeel

[bjorn_fallqvist]

Thank you. I will write you an email.

Best regards

Björn

[xli]

According to your description, you might be experiencing snap-through points that Riks method as S.Li suggested might help depending on how difficult your model is and how good the Riks method implemented. But it seems that is a correct way to go. Refining mesh may help you getting closer to the point but software might still have difficulty to find the solution branch after that.

-xli

[bjorn_fallqvist]

Hi,

I have converted the pressure to nodal forces for use in Riks analysis, but this is taking a very long time. In an hour and a half, it has only calculated four iterations. Is this normal?

Also, will pressure be included as a possible load in future releases of Algor?

Thanks

Björn

[xli]

To your 1st question: the Riks method actually uses a different solver due to matrix may be indefinite (caused by sigularity at few DOF). That solver may take longer time to solve.

-xli

[bjorn_fallqvist]

Hi,

Does the Riks method differ between different packages? I used it throughout my thesis in Abaqus, and it was very quick there. Do you think that you will have support for surface-based loads with this solver in the future?

Best regards

Björn

[xli]

Hi Björn,

Riks method actually was recent years implemented in MES, you might send improvement request of what you are asking for pressure load support. As a developer new to Autodesk system, I am actually not familiar to the official process though, sorry about that.

As I have told, Riks method needs to use indefinite solver that is general slower than regular positive definite solver we use, in fact, about orderly more expansive versus model size. I think ABAQUS should be using similar kind slow solver but they might be more smart to know when to switch to this solver and then to switch back to fast ones, or doing something like that. Well, this is just my guessing, I have no direct experience to ABAQUS Riks method. OK, I do have an idea: you might try regular solver to solve to right about snap-through (before seeing those level reductions happens), then stop this MES job, then create another Riks method job with same model and restart from the unfinished one. Only if you have suffered the slowness with a big size model. It is just an idea, I never tried. Good luck!

-xli 

[bjorn_fallqvist]

Thanks, Xli

I will try this tomorrow. Just to ensure that I have understood how restarting with Riks' method works, the load that it takes and scales (with the load proportionality factor) is actually the nodal forces, no matter the type of load applied in the MES (the restarted) anlysis. The problem I would have thought I had is that in Riks, since no pressures are allowed, I could only have applied nodal forces, and these cannot be made to follow the surface (very important, since there are so large deflections here).

But I can make a MES analysis with the pressure ramped to, say, 0.01 bar. Then I take the results from this analysis, and restart it as a Riks analysis and then use load factors up to 0.25 bar to see the behaviour without actually applying any loads in the Riks analysis? Is this correctly understood?

Thanks

Björn

[bjorn_fallqvist]

It didn't work very well at all. First off, I got this warning:

Warning: Surface and body loads are treated as non-proportional thus held constant throughout the Riks analysis. If they are critical to the process you wish to model, consider converting them to equivalent point-type loads.

Then the program quit with this error:

** Error : reading restart file, r9. This model has been updated to use new enhancements in elements, The analysis must be performed from the beginning in order to perform a restart analysis in the future. ** Error (1)

I restarted from a model that I had not modified in any way at all.

How can I convert a pressure to equivalent point-type loads? I can't think of any way to make them follow the normal of the surface.

Edit: It seems that it doesn't give me any warning or error when I use a surface variable load. Is this a viable option, do you think? What makes this so fundamentally different from a surface force or pressure, that it can be used?

[bjorn_fallqvist]

I'm trying to run a restart analysis with Riks, but I keep getting this error message

 ** Error : reading restart file, r9.
    This model has been updated to use new enhancements in
    elements, The analysis must be performed from the beginning
    in order to perform a restart analysis in the future.

What's wrong with the model? I haven't updated anything at all.

[S.LI]

Hi Björn,

The error is because the "restarted" analysis needs some information that doesn't exist in the previous analysis.

For example, in old analysis, traction/pressure has been used, and it is removed in the "restarted" analysis. Then you will see the error.

The only way is to rerun the model.

[bjorn_fallqvist]

But I haven't removed anything. After the first analysis is finished, I immediately save it as a new file, and restart from the old one without making any modifications at all.

[S.LI]

Did you restart Riks from a Riks or MES or other analysis types?

I'm not sure if you can restart Riks from MES or others.

[bjorn_fallqvist]

According to the helpfile you can:

"This load then creates a displacement, and this displacement is incremented at each step of the analysis. (If the analysis were linear, the applied loads would be linearly interpolated between a multiplier of 0 at time 0 to a multiplier of Total number of steps x Initial load factor increment at the end of the analysis.) For any other loads to be considered (pressures, gravity and so on.) you should run a standard MES/nonlinear structural analysis up to the point when the loads reach the appropriate magnitudes. Then you should save the analysis as a new filename and restart the analysis using the Riks method. Prescribed displacements should not be used in a Riks analysis."

I restarted it from a MES.

[xli]

Björn,

It probably belongs to an area of developing in MES's Riks method. It at least may help speeding up in Riks analysis. We should have tested it before suggesting you, I was low confident about this idea developed in this thread. I will try it a little bit if there any work around we will let you know. Sorry about it.

-xli

[bjorn_fallqvist]

Hi,

No problem. Anything that might be of help. Right now I'm trying to see what happens if I use an original (displacement) convergence requirement of 0.01, and when the time step is much reduced (I'm guessing it corresponds to a more or less "flat" region in a force-displacement diagram) I activate convergence relaxation with a factor 100. What happens as the bag unfolds is really of no interest to us, so this should help speed things up until the elements start experiencing memebrane stresses (instead of just following the rest of the structure, so to say).

Thanks,

Björn