CCX Error "Too Many Cut backs, best solution and residuals are in the frd file"

Aftab

Hi
I am using contact for the attached file, however trying so many options , i am getting above error with no convergence what so ever. Can someone kindly look into this for me.
Best Regards
Aftab

rwhirley

Hello Aftab,
I had a quick look and I believe some of the normal vectors in your contact surface definitions may be inconsistent. I got a partial solution by deleting contact surface #2, changing to nonlinear dynamic analysis (so inertia will smooth the solution) and using a very small time step for debugging (1E-4). You can see the troublesome area in the attached file.

All the normals on the tank side should point outward and on the strap side should point inward. You might give this a look.

I'll look further when I get a few more minutes but thought this might be helpful.

Regards,
-Robert

rwhirley

The attached file clearly shows the problem after just one step. For simplicity I deleted all the ring stiffeners, but the problem is still present. My suspicions (in addition to a contact surface normal vector issue, as discussed above):
1. It is close to the triangular transition between quadrilateral faces - this may be causing the contact search algorithm in ccx to fail.
2. Sometimes contact algorithms don't do well with higher order elements. Try switching to all 4-node elements and see if the problem goes away.

If you can't find the problem, I suggest setting up a simple model problem with a short cylinder and one sling so you can iterate rapidly until you zero in on the problem. I think you are close to a solution.

Maybe someone else has some suggestions also.

Regards,
-Robert

Aftab

Many Thanks Robert for your time and advise. I will look apply these suggestions and see how it goes.
Best Regards
Aftab

Aftab

Hi Robert
I have changed the quad8 to quad4 and have checked al the normals. Removed the triangular mesh from the contact region. seems from partial solution the contact is happening but the final solution does not converge. I will try a much smaller model.

rwhirley

I have played with your model a bit and I think there are some problems beyond the contact. I deleted all the contact and fully fixed the ends of the tank cylinders. I changed the gravity load to ramp from 0 to 9.8 m/s2 over 1 s. Still, the model would not solve as a nonlinear quasistatic model in ccx, even with a very small time step. I changed it to a linear analysis with full gravity load and it solved easily, and the results appeared plausible. I tried making all the shells the same thickness (changed the 50mm to 20mm) - that helped a lot in the nonlinear solution, but it still won't fully solve.

I have no experience using shell elements in ccx; I do know they are not "real" shell elements and that ccx expands them into 3d elements somehow. I don't know enough about this to anticipate problems, but I am suspicious that something is amiss there. I pass it along for what it's worth.

Aftab

Many Thanks Robert for your input, I really appreciate it.

rwhirley

I had a few more minutes to play with this model over lunch today. This was a bit tricky, but at least the contact is sorted out.
1. Tank density increased 10X to account for contents weight
2. Deleted forward & aft tanks; something about their merge to center section was problematic. I didn’t dig into it. I added x & y constraints to tank ends.
3. Added a soft spring element at the top of the tank to provide some z-direction stiffness for the quasistatic solution
4. Corrected gravity load application to linear increase to 1g at t=1 (keeps increasing for t>1.0).
5. Run to 5X gravity, just to challenge contact (normal gravity at t=1)
6. Model now converges easily and results look plausible. I did not check them carefully.

I'll have to defer to someone more knowledgeable in ccx shells for nonlinear analysis to help with the problem in merging the forward and aft tank sections into the center tank.

Hope this is helpful.

-Robert

Aftab

Much appreciated Robert. Looks quite reasonable and results are comparable with the linear! I will check in detail. Thanks again, that was some help.