Random Vibration - Very Long Solution Times 2

[Maxham1]

I've been working on a random vibration model that has the following:

#Nodes = 37K, #Elements = 17K, DOF=158K

It's a model of a power supply and the problem is that the solution takes DAYS to solve. There are 116 resonant frequences. If I just try to solve the random vibration with just using the first 40 resonant frequencies, I can get a solution in about 4 hours. But anything longer the computer (Quad Core, 3 MB RAM) takes very, very long.

Any ideas? Being pointed in the right direction would help. Are there any resources that would help in how to handle large problems like this? (Actually I didn't think it was a large problem considering the stats above).

 

[GregLocock]

I agree it isn't a large problem, by itself, but suspect that it is trying to store 116*158k results in memory. Which still doesn't seem much.

try solving with 10 20 30 resonances and you've already done 40, to see how long they take.

Cheers

Greg Locock

SIGlease see FAQ731-376 for tips on how to make the best use of Eng-Tips.

 

[Maxham1]

I was also thinking that maybe it's trying to write out to memory. Most of my computational background is in CFD, but I've done some FEA mostly static analysis, which runs fine. I'm wondering if it's got to do with contact definitions or the lack thereof.

I'm also looking for any resources on running random vibration, more than just the basics like I have with Schroff (good book nonetheless).

 

[GregLocock]

Ah, you can't do random FEA with contacts in the frequency domain

Cheers

Greg Locock

SIGlease see FAQ731-376 for tips on how to make the best use of Eng-Tips.

 

[Maxham1]

Wow, that's an eye opener - thanks! How does one do a vibration analysis of an assembly where many of the parts are in contact - just turn off all contact sets?

 

[GregLocock]

Yes. Or use time domain software.

Cheers

Greg Locock

SIGlease see FAQ731-376 for tips on how to make the best use of Eng-Tips.

 

[Maxham1]

Greg, I just wanted to make sure we're talking about the same thing when it comes to contacts. I'm talking about contact sets, that is, defining that two surfaces are bonded. I'm having a hard time understanding why Solidworks Simulation wouldn't allow for the analysis of two (or more) pieces bonded together.

Were you talking about contact analysis above? Thanks for the help.

Dave

 

[kellnerp]

Random vibration is still a linear type analysis. One of the ways you can linearize a non-linear structure is to put springs in the areas of contact. You then have to tune the stiffness and damping of the springs to realistically behave like the joint.

Sounds like you are doing an out of core solution. Cosmos doesn't always use all the RAM so the fact that you are going out of core so soon doesn't surprise me. Task manager can tell you if this is in fact happening if you turn on the VM, page fault and IO data columns. Cosmos also doesn't use multiple cpu cores.



TOP
CSWP, BSSE

http://www.engtran.com

http://www.niswug.org

"Node news is good news."

 

[GregLocock]

re contacts -yes I was assuming you meant contact bcs, not glue between components.

In that case you are on the right track.



Cheers

Greg Locock

SIGlease see FAQ731-376 for tips on how to make the best use of Eng-Tips.