Using Symmetry To Reduce FEA Runtime 3

[JGard1985]

Hi all

A recent project spawned a water-cooler discussion regarding symmetry theory and FEA. I developed a worked example using ANSYS (LSDYNA) to show underlying symmetry concepts, manual application, and advantages to symmetric analysis.

I think its a great primer, to help new FEA engineers understand underlying principles of symmetry in FEA in any software package.

https://www.xceed-eng.com/reducing-fea-runtime-with-symmetry/

Might be worth a read, comments very much appreciated

Thanks!

Jeff
Pipe Stress Analysis Engineer

http://www.xceed-eng.com

 

[corus]

You've applied rotational restraints to a solid model when there are no rotational degrees of freedom. The example you chose would have been more apt if it had been a shell model.

 

[JGard1985]

Corus

Thanks for giving it a look over. I applied the "no rotation" DOF to the edge of the tank liner which is shell elements. Looking at my figure with "thick shell" option turned on this isn't apparent. I will update wording to describe which elements are shells and which are solids

Thanks again for your comments!

Jeff
Pipe Stress Analysis Engineer

http://www.xceed-eng.com

 

[IDS]

A few points:

At least in my area (civil structures), I'd say that many structures can be approximated with a symmetrical structure, rather than most are symmetrical.
Asymmetrical loading can be modelled with a combination of symmetrical + anti-symmetrical loads, with appropriate boundary conditions. Perhaps worth adding something on that.
That approach of course requires linear elastic behaviour.
Perhaps worth adding some words about checking that the boundary conditions are correct.

By the way, I posted a link to your engineering-apps article in the Structural engineering forum:

http://www.eng-tips.com/viewthread.cfm?qid=426581

Doug Jenkins
Interactive Design Services

http://newtonexcelbach.wordpress.com/

 

[ESPcomposites]

I assume the example should represent best practice FEM use as well. That said, some comments:

- A "potato" like part is the classical way to demonstrate how to set up the symmetric constraints when solid elements are used (also an appropriate use of solid elements). It is much easier to show the BC's on the cut planes.
- For thin structures (such as this problem), where shells are usually a better choice, a different set of BC would be used (including the rotational DOF) to set up the symmetry conditions.
- For this problem, shell elements would increase the computational efficiency far more than using symmetry with solid elements (making symmetry a moot point). Again, is this really the best example?
- Is a single solid element through-the-thickness sufficient? If not, then you may not want to imply this is a best practice example. Solid elements are very inefficient for thin structures (may require multiple elements through-the-thickness and aspect ratio requirements may force you to use many elements).

Also, in my experience (aerospace), I have rarely been able to use symmetry (usually only for study models). Most complex structures (that would actually benefit from reduced run times) usually have some sort of asymmetry due to the small design features, loads, eccentricity, etc. Therefore, if required to use the complete model, one would want to make it as efficient as possible (leading back to the use of shell elements for the example problem).


Brian

http://www.espcomposites.com

 

[rb1957]

asymmetry in structure or loading is common.

another day in paradise, or is paradise one day closer ?

 

[GregLocock]

Although symmetry reduces run times, it increases the likelihood of a jellyware/fingertrouble mistake. It seems to me that in simple cases it reduces run times, but in complex cases it is either not applicable, due to actual assymetry, and increases the probability of an incorrect model. The time spent debugging an error in b/c s is vastly more expensive than the extra run time for a full model.



Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376

http://eng-tips.com/market.cfm?

 

[swertel]

I used to create symmetric models all of the time, now I don't bother.

The mathematics behind FEM hasn't changed. The solvers are still performing the same calculations on the stiffness matrices. But, my CPU and GPU performance has increased dramatically. Getting the FEM to look more like the actual end-item makes it much easier for me to explain the results to non-technical management than it was to try to explain why a partial model with symmetry rules applied yielded accurate results.

In other words, cranking out a full model doesn't take much longer than cranking out a symmetric model just a few years ago. I might as well analyze the entire geometry.

--Scott
www.aerornd.com

 

[Mustaine3]

For experienced analysts symmetry is a nice method to save computation time. The chance to make a mistake with the symmetry BC is not so high (again, for an experienced user).

In Abaqus/CAE I can mirror (or pattern) my result in postprocessing to show the full model with results. That prevents stupid questions.

 

[ESPcomposites]

Also consider that you could spend a lot of time setting up a symmetric model; the next day you may be asked to run a new set of loads (not a symmetric load case). You would have wasted a lot of time by not "future proofing" your model and your boss may not be happy about this. The labor to set up a new model does not usually make up for the saved run time.

With today's computers, solve times are rarely the issue, provided you construct the model efficiently (i.e. for the example problem, shells would be far more efficient than solids). There are certainly exceptions, but symmetric models are not the norm (based on previously stated disadvantages).

Symmetry has its place (academic study and certain problems), but it may end up costing you in the long run. The other thing to consider is that if your run times are very long (for linear solutions), symmetry may not address the root cause of this; there is usually a bigger problem with the approach at that point.

 

[scheah]

For a full model consisting of 80k nodes as in the example, the trade-off consideration lean towards solving the full model instead of a symmetry model with less than 5 minutes 'wasted'. That said, I used symmetry in a recent large problem dealing with nonlinear materials and frictional contacts.

While born out of necessity of limited computing power, I believe symmetry still makes sense for certain problems today.


Best regards,
Jason

 

[ESPcomposites]

Symmetry has benefits for certain problems (contact usually being one of the most time consuming types of nonlinearites). But the statement that "most structures and their loads are symmetric" seems questionable for most problems where the computational savings would pay off with modern computers (i.e. more complex and computationally taxing models tend to be asymmetric).

Perhaps a better example of the use of symmetry with solid elements would be a lug/clevis joint (half symmetry) where nonlinear material and contact is used. That would not only be an appropriate use of solid elements, but also an case where computational savings might pay off. The downside is that from a practical perspective, we often allow the load to be slightly off-axis, which causes a bending load (a scenario where symmetry is not appropriate).