Non-linear analysis of a buckling cylinder 2

[Hurricanes]

I am currently trying to solve a localised buckling problem in a cylindrical piece of equipment modelled using plate elements. The buckling comes about due to a combination of thermal effects and eccentric loading on the cylinder.

When I run a linear static and a linear buckling analysis I get some nice answers out that reasonably replicate what is observed out on site.

When I try to run the non-linear geometry analysis I cannot get it to converge.

Does anyone have some tips, tricks or techniques to get around this problem? I have simplified the geometry and the mesh, fiddled endlessly with load increments... as the cylinder is buckling does this mean the non-linear analysis is unstable and won't converge?

 

[corus]

If it has buckled then you won't get a converged solution as in theory the displacements have gone off to infinity. In some FE programs you can see the solution prior to failure, ie. if the load is ramped then you might see the solution at, say, 90% of the load. Of course, if the problem is highly non-linear, with contact, and non-linear material properties, then you may be better using an explicit analysis, though this can take a long time to run.

Tara

http://tinyurl.com/4ydjg7m

 

[rb1957]

linear buckling means elastic material and predicts failure at something else than the applied loads, or the FEM is predicting stresses higher than yield with the applied loads (hence failure).

if the latter, then apply less load for your non-linear run. maybe apply the load without the temp (just to see what happens).

 

[flash3780]

It depends on the software you're using, but there are usually switches to turn on for this sort of analysis. ANSYS uses "stabilization" which is basically applying dampers to the body to help walk it through the instability. ABAQUS uses a RIKS solver to converge (been a while, but that's what I remember anyways).

 

[Hurricanes]

Thanks to all of you for your responses.

I am using Strand7 to run the analysis, which I don't believe allows me to perform an explicit analysis. Nor does it do anything fancy to force convergence.

I am planning to proceed as follows;

1. Modify the geometry slightly so the buckling mode I am interested in is the 1st mode in the linear buckling analysis (stiffeners in areas I am not concerned with buckle at lower load factors)
2. Run the non-linear geometry analysis and find the fraction of the load which brings about the onset of buckling (point at which the stiffness matrix becomes singular)

Then it is a case of determining how to prevent buckling occurring via some sort of stiffening arrangement.

My concern now is that the thing out on site has already buckled so in devising a solution to prevent further buckling, should I really be modelling the permanent deformation in the cylinder and applying some sort of stiffening solution to that?

 

[corus]

If your simple linear buckling analysis has given reasonable results that agree with observations, then I'd stick to those results. They give some confidence in the model to others for when you look at the post buckled shape, which is the real point of the exercise. So the answer to your last question is yes. I'd use the same loads/factors on the stiffened/post buckled shape to prove that it doesn't buckle as the original/unbuckled structure did.

Tara

http://tinyurl.com/4ydjg7m

 

[rb1957]

how have you defined your material curve ? i imagine that perfectly elastic-plastic material def'n would be a problem.

have you asked Strand7 helpdesk ?

from your linear run, what's the highest stress ? this'll tell you the %age of load that the structure should be able to carry to yield.