Random response analysis with Nastran 1

[cavva78]

Hello,
I'm looking for someone expert in random response analysis for help, since Nastran
and Femap guides are not very explicative. I try to list some doubts:

-I apply an acceleration load on 4 nodes with a PSD input (in g^2/Hz) with a model
in SI units. When I ask as output acceleration I expect to get again g^2/Hz, right?
And for displacements, forces and stresses which units do I get?

-Is it possible in some way to have a Von Mises stress as output from a random analysis?

-I was asked to calculate differential displacement output between 2 nodes as PSD
by random analysis: is there any way to do that? Is it possible to simply subtract
2 PSD displacement curves (taking absolute value)?

Thanks a lot, and any reference to tutorials or guides is helpful.

Marco

 

[SamSSX]

I have done random vibration response analysis with Patran/Nastran. I would think Femap/Nastran should be similar.

Patran/Nastran is unit neutral. All they ask for is a consistent set of units. I always convert accel PSD into my consistent unit system (for me, it is in^2/sec^4/Hz) for input in Patran. The results I get will have units of in/sec^2 for accel, in for disp and lbf for force.

>-Is it possible in some way to have a Von Mises stress as
>output from a random analysis?

You can calculate RMS values for all stresses (including von Mises) after specifying input excitation.

>-I was asked to calculate differential displacement
>output between 2 nodes as PSD by random analysis: is
>there any way to do that? Is it possible to simply
>subtract 2 PSD displacement curves (taking absolute
>value)?
No, that is not the way to calculate the PSD (auto spectral density, to be exact) of relative disps of two nodes.

First, remember that auto spectral density is the Fourier transform of auto-correlation of a random variable x(t):

Auto-correlation[x] = E[ x(t)*x(t+tao) ]

And cross spectral density is the Fourier transform of cross correlation of two random variables x1(t) & x2(t):

Cross-correlation[x1, x2] = E[ x1(t)*x2(t+tau) ]
Cross-correlation[x2, x1] = E[ x2(t)*x1(t+tau) ]

Suppose x1(t) and x2(t) are the disps at node 1 and 2, respectively, then

Auto-correlation[x2-x1]
= E{ [x2(t)-x1(t)] *[x2(t+tau)-x1(t+tau)] }
= Auto-correlation[x2] + Auto-correlation[x1]
- Cross-correlation[x2, x1] - Cross-correlation[x1, x2]

Take the Fourier transforms of both sides of above equation, we get:

Auto-spectral-density[x2-x1]
= Auto-spectral-density[x2] + Auto-spectral-density[x1]
- Cross-spectral-density[x2,x1] - Cross-spectral-density[x1,x2]


>Thanks a lot, and any reference to tutorials or guides is helpful.
You should get it from your tour tech support.

Sam SSX

 

[cavva78]

Thanks Sam,

-ok for the units.

-For Von Mises, in Femap I request stresses as output but Von Mises doesn't appear in the results, only single components (sigma_x, sigma_y, etc). Is there a particular Nastran command to have Von Mises in output with random response?

-I'm missing some theory of random vibration I know... anyway if I want in output auto spectral density of some disps I just ask for random output, and I know how to do it in Femap. But for cross spectral density how can I do it? Do you know the commands in Nastran? Note that I have only one acc input on 4 nodes of my model. There are no cards in Femap asking for cross spectral density or something similar

 

[SamSSX]

I can only use Nastran through Patran. So I am not familiar with Nastran commands. In Patran, "Random Analysis Tool" can get RMS of von Misese stress response and (at least according to its documentation) cross spectral density.

 

[cavva78]

Could you please send me a Nastran input file taken from a random analysis set up by Patran asking for Von Mises output (or differential displacement output)? Maybe there is some tutorial already done, and you can just export the .dat file..

 

[eaolsen]

Hi All-

I am looking for basic instructions on how to perform PSD analysis in Patran/Nastran. Does anyone have a document they could share?

Thanks,

Erik

 

[mechengdude]

When I used PATRAN/NASTRAN for dynamic random analysis, I found the help line one of the best I've ever used. Suggest you call MSC software directly. My understanding at the time was a gentleman by the name of Mohan Barbela was responsible for writing the code for the dynamic random functionality. He was based out of Texas and was able to explain things in far more detail than I ever needed. I don't think it's good practice to put up an individuals contact information on-line without their consent so I'll just list the main office address for MSC Solutions where he was based. Bear in mind this information is about 6 years old.

MSC.Mechanical Solutions
1000 Main Street Suite 190
Grapevine TX 76051
817.481.4812

 

[Spirit]

cavva78,


at this link

http://femci.gsfc.nasa.gov/random/index.html

you can find a brief yet complete tutorial on what you should know to perform Random Analysis with NASTRAN.

There is a lot of other interesting stuff on other subjects as well.

I hope it helps.

Regards


Spirit



'Ability is 10% inspiration and 90% perspiration.'

 

[tomirvine]

I perform random vibration finite element analyses via a modal analysis which yields frequency response functions(FRFs) for the nodes of interest.

Then I develop my own power transmissbility functions from the FRF functions, outside of the FEA software.

Then I apply a base input PSD (or applied force PSD) to the power transmissibility function to calculate the response PSD at nodes of interest. This is also a post-processing step outside of the FEA software.

The idea that we must use various auto and cross-correlation functions to calculate relative displacements is technically correct but may be overwrought.

We can simply use the complex absolute displacement FRFs to develop relative displacement FRFs, from which we can calculate relative displacement power transmissibility functions. This method is easy.

Tom Irvine

http://www.vibrationdata.com

 

[40818]

Sorry to but in, but having just read the comment from Sam SSX:

"I can only use Nastran through Patran. So I am not familiar with Nastran commands"

Please don't feel agrieved about my rambling, but how can you use a pre/post processor but have no clue about what it does and how it does it? Its a big bug bear of mine that people learn to become cad monkeys/ FE jockeys, but have no idea about what the hell the program does. You should LEARN Nastran and USE patran. So many times programmes like Patran manipulate the data incorrectly (with bugs or features), or just give you garbage answers because it is able to.
I dont mean that you have to understand the code, but you must at the least be familiar with the limitations and abilities of it.
For example, would you know why linear buckling in Nastran is acceptable for some situations and not other. And why, or how the choice of elements (especially between lower/higher order or solids/shells) can make a marked difference between results.

Really gets me this does.

Arrghh.