Reduced integration = convergence difficulties?

[Anonymous]

Hi,

One of my analyses yielded a far too high collapse load (with MES), so I tried running an eigenvalue analysis, and this is much lower; actually approximately correct when compared with reality. I am thinking that the problem is due to some kind of shear locking. The structure is basically a strut supporting a wall, on which a hydrostatic pressure is acting. From this, the load causes a normal compressive stress in the struct, causing it to buckle. I'm using shell elements for this, but as I said, the structure behaves too stiffly (there's a factor of difference of about 20 between reality and MES).

I tried to activate reduced integration for transverse shear terms, but now the analysis is having difficulty converging. I'm guessing maybe the elements become too flexible with fewer integration points?

Any suggestions?

Thanks

Björn

[zhuangs]

Which element formulation and which material model of the shell element type are you using? 

You can check

"Element Definition" -> "Advanced" -> "Element Formulation";

"Element Definition" -> "General" -> "Material model".

-Shoubing

[Anonymous]

Hi

I'm using the general shell formulation (large displacement, 3rd order integration, total lagrangian) and an elastic-plastic von-Mises material model (hardening modulus zero, though).

[zhuangs]

I am wondering whether thin shell can help.

-Shoubing

[Anonymous]

I'll try this.

Thanks.

[Anonymous]

That helped with the convergence problem, but I'm still getting a too high collapse load.

[zhuangs]

It's great that thin shell can help convergence proble.  Now let's focus on the result of collapse load.

1. I would like to make sure that what the "Reality" means, expreiment?
2. What kind of material are you using? Is it selected from "Autodesk Simulation Material Library"?  If yes, are all the material properties (such as Mass density, Modulus of Elasticity, Poissom's Ratio, Yield Strength, Strain Hardening Modulus) exactly the same as the expreiment data?  If not, whether are all the user-defined material properties using the material properties from the experiment?
3. I prefer the following setting are used: (1)  "Element Definition" -> "General" -> "Analysis Type" = "Large Dispacement"; (2) "Analysis Parameters" -> "Advanced" -> "Equilibrium" -> "Nonlinear iterative soulution method" -> "Full Newton w/ line search".
4. If the time step is too large, try a smaller one, which can help the results.

-Shoubing

---

@Anonymous wrote:

That helped with the convergence problem, but I'm still getting a too high collapse load.

---

[Anonymous]

Hi,

1. Reality is the result of a real-life prototype built, which collapsed.

2. It's an elastic-plastic material model of a regular steel. I'm not sure exactly what kind of steel they are using, but using a different kind of steel I doubt would give that much discrepancy. As I said, using linear buckling results in the approximately correct collapse load.

[S.LI]

Hi, Björn,

The convergence issue is because of the strong nonlinear in your model, which appears as

1.) shell element with large deformation;

2.)buckling

3.) elastoplastic material model with zero hardening.

I have to say this is tough.

My suggestions on your original model (with general shell with reduced integration) are:

give an appropriate nonzero hardening parameter

or

converting pressure to nodal force and using Riks method

or

replacing pressure with prescribed displacement, and checking the reaction force when it buckles. I know this is a little bit unreasonable on your hydrostatic case.

[Anonymous]

Hi,

Yes, I can imagine it is. The hardening parameter shouldn't really matter; buckling occurs far below the yield stress.

Converting to nodal forces may be a good idea, but would be a bit cumbersome. I think I'd use a variable surface load instead. When we're on that topic, how come surface loads are not allowed in a Riks' analysis, but variable surface loads are? They're both converted to equivalent nodal forces, so why can't e.g. pressures be used also?

Thanks

Björn

[S.LI]

Do you mean " variable surface loads" works with Riks?

I never notice this.

It's nice if yes. At least there are more options.

Sorry for not allowing general surface loads on Riks.

[Anonymous]

Yup, it works for some reason. Not sure why surface loads don't works.