Model doesn't converge

[flyforever85]

I have a 2D model that doesn't converge.
It's a model of an helicopter with specific loads, the solution should be found through inertia relief, but it doesn't converge even with different nodes completely locked.
I have several hinges and mpcs. Checked them all and it seems to me there is no problem

I ran a modal analysis, and it seems right: 6 rigid modes and then several modes coming from different part of the structure and different lump masses attached to the structure.

I'm removing one by one the hinges and the couplings and in some cases it converges in some other it doesn't. But there is no rule

Any idea on which checks I should do to understand why my model doesn't converge?

 

[FEA way]

But there are no unconnected regions now, right ? Which type of step do you use ? Your approach with checking connections one by one is very good, even such inconsistence gives valuable information that convergence issues may be caused by these connections. I would try automatic stabilization in the step settings. Also take a look at the results right before the analysis stops (unless it happens immediately after the simulation is submitted). It may show some spurious deformations.

 

[flyforever85]

I don't think there are unconnected because a) I don't see them in the modal analysis and b) the solution takes up to 50 iteration before to stop. When I see the results it seems they are converging to a solution though abaqus aborts because the last time increment is larger than what it needs.

The step is static analysis with the default paramters.

I tired automatic stabilization but it's not allowed since it's an inertia relief analysis.

 

[FEA way]

It's hard to say more without seeing the model but you may have to apply some boundary conditions anyway (even though inertia relief is used). That's because springs, dashpots, MPCs and couplings (from your previous post I remember that you have at least the latter) may generate internal forces and moments that can't be balanced by inertia relief. You may also consider removing inertia relief and using BCs instead. There are ways to constrain the model in such manner that no false stresses appear in results.

 

[flyforever85]

I think I narrowed down the problem to the floor attachment. I'll try to explain at the best of my possibilities.

The floor is composed by 2 flat surfaces: the top is a honeycomb and the bottom is a flat aluminum sheet. In between there is a tube/pipe, that my longeron. The attachment is "U" shaped: the longeron sits in the concave part of the U, the bottom of the U is attached to the bottom sheet and the two vertex fo the U are attached to the top honeycomb.
I simulate this attachment as a hinge that goes from the top floor to the longeron and another hing from the longeron to the bottom sheet. I think abaqus does not really like it and I don't know how to make a better simulation of this configuration

 

[FEA way]

Could you attach a picture of a very simple hand drawing with these main features of your model ? It would be much easier to understand the way it looks like for someone who doesn’t know much about helicopters, like me.

 

[flyforever85]

Thank you again for your kind availability but apparently I solved the problem. And still I'm not sure how.

I noticed that every step took several iterations, I thought it was weird since it's a linear analysis with beams. None of the steps converged but the first. So modified the second step and made it equal to the first and didn't converge as well. I decided to delete them all and recreate them again and guess what, all the steps (9) converged with just one iteration and the analysis finally works. I've never had a problem in the steps, I guess what happened will be a mistery.

 

[FEA way]

Actually recreating the analysis from scratch surprisingly often solves the problem. That's because in case of some more advanced simulations we make small mistakes, accidentaly click something or select/unselect, make mistake in units and so on. It all can make the analysis fail. Checking each option is the first thing to do but in those complicated cases we may still miss an important detail. That's why, when everything else fails, it's good to start from the beginning.