Static Non-linear FEA simulation errors - Newton convergence could not be reached

Using SimScale Solid Mechanics / FEA
21

Hi @DaleKramer,
you can change the global physical contact settings in the parent tree node for the physical contacts:

physical_contact_settings
physical_contact_settings623×549 50.7 KB

At “Solution method” you can change between “Penalty Method” and “Augmented Lagrangian method”.

The comment you cited means that this setting can only be changed globally for all contact pairs and not individually per contact - so all contacts are either defined as Penalty or Lagrangian.

Good news: we will actually remove this restriction in the next 2 weeks. Then you can select this setting for each contact individually.

Best,
Richard

2 Likes
22

Theoretically, we should see this coin diameter at 1.0s (from CAD program):
FullSizeCoin

23

And the horses are coming around the bend, with Lagrange at 0.8s just a hair ahead of Penalty1e14 at 0.8s.

But wait all the horses have stopped and it looks like none of them can get past that critical 0.8s to the 1.0s finish :frowning:

Here is as far as Lagrange got:

Lagrange-8s
Lagrange-8s1632×975 289 KB

And here is how far Penalty1e14 got:

Penalty1e14-8s
Penalty1e14-8s1662×974 249 KB

Hey, race fans, any idea why the last 4 horses have slowed to a snails pace at about 0.8s after a quick ~15 core hours to get there and then sim had to be cancelled after ~60 core hours after no further progress and no errors in solver logs :question:

24

I made a copy of the project and was able to simulate the whole displacement.

Here is the link: SimScale

Some changes I made:

  • Applied the elastic constraint to the cylindrical face of the coin, with an isotropic value of 1000 (N/m) this is arbitrary and works, but is generating 1 N of force en each node for each 1 mm of displacement, what should be taken into account when analyzing results.
  • Turned some numerical knobs, the most important being lowering the tolerance to around 0.0001, lowering initial time step to 0.05, number of processors to 16.

I think the following step is to implement a hyperelastic material model, or whatever is closes to the material being used. Please evaluate if results are satisfactory.

2 Likes
25

Great advice on the numerics :slight_smile:

EDIT: But I think see that the Numerics changes you made to my 'Run 2 - penalty … ’ were:

  1. ‘Tolerance’ from its 0.000001 default to 0.0004 (you say this may be the most significant change that allowed completion)
  2. ‘Max Newton iterations’ from its default of 15 to 20

And changes to the Simulation control were:

  1. ‘Initial time step’ from default 0.1 to 0.05
  2. ‘Number of subdivisions’ from default 2 to 4
  3. ‘Number of processors’ from 32 to 16 (which forces ‘Number of parallel processes’ to change from 16 to 8)

Am I correct on this, if so perhaps edit your post for those that follow.

The two leaders so far had elastic on coin cylindrical face, but it is good to know you had to increase the elastic stiffness to 1000 N/m for your success…

AND, ggiraldoRun1, using modified parametesr of penalty1e14 , appears out of nowhere, with a horse that does not die in the diameter race, to take a slight lead and the pack is just coming around the bend :joy:

ggiraldoRun1
ggiraldoRun11687×974 216 KB

The race to the finish seems to be a matter of changing the state of matter in the lead horse :rofl:

I am on it :wink:

At this point I am not too concerned that I match my exact material properties until we get something running to full diameter, so anyone can still join the race here … :wink:

1 Like
26

This horse race sprouted up on this topic and was finished there … :wink: