Fully Coupled Thermal-Stress Converagnce Problems 1

[grnlan03]

I am attempting to run a fully coupled thermal-stress analysis on two aluminum plates. I am using hex elements (C3D8T) with about 3000 elements. I can run the model when the mesh is very course but when I try to run the model using a fine mesh to the model will not converge. I do not know why this is happening if anything I would expect it to not converge using the course mesh.

Also is there a limit on the aspect ratio i.e. the length, width, and height of a hex element?

I can post the model if needed.

Thanks

 

[xerf]

I believe there is a restriction on the mimimum time increment (against spourious oscillations) if you are using quadratic elements. Something like:
delta_t>function(density, conductivity, element size ^2)

However, documentations states that this is not an issue with first order elements.

Any relevant warning messages, about why the solution diverges?

Material softening at high temperatures may lead to local instabilities (which can be mesh dependent).

 

[corus]

I have seen this in the past where elements appear to go inside out with a finer mesh and it was a matter of trial and error to get the mesh size just right but not too coarse. As xerf says though, it's better to describe the problem better, ie. is it a time increment problem, contact or excessive yielding, or what?

corus

 

[grnlan03]

I beileve I got it to converage after I decreased the Time/LPF Inc or Increment size from the default settings of Initial = 1 Min = 0.00001 to Initial = 0.0005 Min = 0.0000005. The model converaged around 0.0013 which seems odd since this is above the default Min value. Also the Time/LPF Inc seems start at 0.0005 decrease, increase and decrease again before it converages.

I really don't why this worked. Do you all know why this worked when I changed the increment size and what that does in the model. I'm sure this is in the documentation but sometimes it is better to here it from a actual person.

Thanks

 

[corus]

Default time value is 1. If this doesn't converge then the time step becomes 0.25 and it tries again, before decreasing it again for another attempt. After about 12 attempts (I forget the actual number) of decreasing the time step it'll give up and go home. By setting the initial to 0.0005 you've allowed it to decrease the time step to a value where it will converge. Increasing and decreasing the time step to converge isn't so unusual. It depends what hiccup caused it to slow down a while before continuing on. Possibly a contact problem where a node didn't quite make contact at a particular time?

corus

 

[DanStro]

I apologize up front if this is obvious.

When I see that the time increment keeps going below what I have set for the minimum I go to the mesh verification and see what the minimum stable time increment is and then set the minimum to be below that. That usually works, but I can see that it might not do the trick if the problem is due to a discontinuity from changing contact conditions during a run.

HTH,
Dan

 

[grnlan03]

When I change the time step what exactly am I changing? I guess really don't know what the time step does and why making it smalller allows it to converge.