SOL 106- not converging

[spicynoodles]

There was a previous thread similar to this which I read, but in didnt prove helpful. So here I go posting this thread, hopeful that someone has run into a similar situation.

I am running SOL 106 for a beam-column type of analysis, nonlinearity for the displacements. The constraint is simply supported on one end, and roller on the other end, free to translate along the column's own axis. The actual loading case is an axial compressive load and distributed load along the column, which yielded a non convergent solution. The column is modeled by shells, and is loaded eccentrically by the compressive load.

Now, I did do a few tests by I varying the magnitude of the loads, and running only compressive and only distributed load. What i found is that I can find a convergent solution if i decrease the load enough. So there is a specific magnitude for which any runs with loads higher than that wont converge. If the model runs just fine with smaller loads, does that mean its an actual physical instability of the structure? which would mean that I need to re-design my part? Or is it only an algorithm occurence? Im leaning towards the first possibility. Please give me any kind of input- it will be appreciated. Thank you.

 

[zerodof]

In sol 106 ,make sure you dont have any rbe2 or bar elements as they are linear elements.

The other thing to try out is to make more than one load cases and increase the number of increments on NLPARM card until you see any convergence. Note that convergence level ,and add another loadcase and rerun with higher number of increments. Try this several time , and you will see it converging unless you have very flexible part or mechanism.

 

[spicynoodles]

I dont have any rbe2 or bars. I tried increasing the NLPARM card several times, but my model still seems to converge at about 90% of load, and then diverges. Maybe my part is too flexible like you mentioned, which means i should stiffen it.

 

[ThomasH]

No convergence can also mean some kind of structural collaps. Do you have anything like yield limits in the model?

 

[EricZhao]

It looks like your beam is buckled under compressive. If it is true, just find the load step that your solution begins to diverge. That is the buckling load.

Usually, it need a small structural imperfection to have buckling. But sometimes, numerical roundoff can also introduce buckling.

 

[spicynoodles]

Thanks for all the great comments. In response to the last posts:

ThomasH,
I dont have any yield limits in the model. I have linear isotropic & orthrotropic materials in the model, so it shouldnt recognize yields.

EricZhao,
Ive already thought about that. I did a test where i imposed 2 of the same load cases, with different magnitudes, but I know both exceed the number that allows convergence. Both did not converge, and diverged at a percentage of the full load. One was at about 82% load, the other at 90% load. When i calculated the failure loads for these 2 cases, they werent the same magnitude. So it doesnt consistently give me the same buckling load. But, I still tend to believe that this might be hinting a physical geometric instability that actually exists. But i dont know this for sure.

 

[EricZhao]

I would suggest that you do simple eigen buckling analysis. Usually, the result from eigen buckling is larger than the actual buckling limit. So, if your currnet load is larger than the eigen buckling limit, you are in trouble already.

 

[kellnerp]

To continue on Eric's thought and to confirm it run your Sol 106 analysis at 50% load. If it converges you probably have collapse. If it will not converge you probably have other issues.

Another thought. If the compressive load causes the beam to arch up into the distributed load then is it possible you are seeing some kind of snap through?

 

[spicynoodles]

The buckling analysis was a good idea. However, I ran that, and found my current load is NOT bigger than the eigen buckling limit. I was hoping that would be the case.

I did do runs with less load and found that my solution converges. kellnerp, what do you mean by "collapse."? Is that something that would happen physically, or something in the algorithm?