Solution Performance in Contact Problem

[Chebeba]

Hi all! So I'm new to Abaqus, but I already like it a lot, having much better performance than packages I used before. Especially the mesher...

Anyway, I am now trying to run a contact problem. I have done the hinge tutorial, and expanded upon it some, and also successfully analysed a small simplification of my real problem. All this using A/Standard, with contacts being between a rigid body and a deformable, like in the tutorial.

Now when I try to increase my mesh density to something a little more realistic than the tutorial or my simplified testcase, the solution times go through the roof with amazing speed, even though the model is identical in all other respects. Much more so than a problem not using contacts would do with the same density increase.

Are there any pointers for how to improve performance when using contacts? My simulation times are on the order of many hours for a two part problem, 1 rigid shell and one deformable solid and on the order of 150000 mesh elements on a dual Opteron/2.6GHz rig. Does not seem realistic! Setup identical to Hinge Tutorial from Getting Started manual.

(I get the same slow solution for the Hinge Tutorial if I just increase mesh density there)

/c

 

[corus]

I'm not sure how the contact works in Abaqus but I believe that additional degrees of freedom are created. So increasing the mesh density will have a more severe increase in the number of unknowns and hence run time. Also, solving simulataneous equations involves numerical operations in the order of the square of the nomber of unknowns. Doubling the number of nodes will make the run time increase by a factor of 4, etc.

corus

 

[Chebeba]

Hmm... Maybe so. I was just not used to simulation times increasing according to this law from other FEA packages...

The good thing is I have now realized that my problem was totally not suitable for Abaqus/Standard solution, but Abaqus/Explicit using Quasi-Static analysis with Mass Scaling handles it like a charm. Contacts in /Standard seems to have limited usefullness.

Cheers,
/C

 

[W00dman]

The difference in CPU time could also come from the contact formulation that is used...

Abaqus/Standard uses a lagrangian multiplier formulation by default which doubles the amount of unknowns in the problem.
You can try with an Augmented Langragian formulation (all versions), or a penalty formulation (for 6.6 only) with the *SURFACE BEHAVIOR map. These formulations keep the same number of unknowns.

Abaqus/Explicit uses exclusively a penalty formulation.

Cheers

 

[Engelien01]

I mean 150000 are a lot of elements...

but do you change the default information in the job panel?
- Ram space
- go to maximum