Frequency analysis software for structure borne noise

[frizzit]

I am looking for some data analysis software for calculating the structural mobility/impedance based on modal hammer and random vibration exciter measurements. Meaning I am going to collect a bunch of high frequency data with a bunch of high bandwidth accelerometers and I need some way to analyze impedance of the structure between 6 and 20 khz. I'm still working on the method of testing and I'm trying to determine how to analyze the data. I'm not looking for resonances, though I am looking for how well those frequencies are transmitted through the structure. If anyone has any suggestions on what software I should be using and how best to analyze the data I would appreciate it.

 

[GregLocock]

There's two parts to this. To create FFTs you can use anything that comes to hand from Excel upwards, depending on how much you want to hand code/spend. Off the shelf 4+ channel FFT analysers will have no problems with your requirements.

However what you are trying to do thereafter is trickier and until you decide wat you are actually trying to do then it is hard to say, B&K and LMS probably have modules to do what you want, but again rolling your own might be worth it, for which I'd use Octave or Matlab.

The general field is called statistical energy analysis, and you are probably interested in structural intensity, which is directly analagous to acoustic intensity. The marvelous thing about intensity is that it tracks the flow of something human beings can't detect and don't care about.

Cheers

Greg Locock


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

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

 

[frizzit]

I'll do some digging on the structural intensity topic. I've also been barking up the LMS and B&K trees already. My application requires 1 input (modal hammer) with ~14 accelerometers on the structure. I believe my setup is somewhat involved, though I've reviewed some setups with hundreds of points. My current processing software is a bit lacking in frequency analysis (DIAdem). I can do multiple 1 in 1 out FRF, though I don't know how to calculate mobility/impedance, I'm still digging around some B&K/Endevco/PCB technotes. I would prefer to be able to analyze all 14 responses based on a given input.

Most of my experience is in injury analysis and dynamic/quasi static events. This whole vibration analysis is a bit new to me, thanks for the input. Another question kind of on topic, I've been exploring the FFT function averaging and overlap. I have no idea what is standard for frequency analysis at such high frequencies. Increasing the block size (points per fft) decreases the frequency step(bin size), though it "clutters" the plot. Is there a good average, standard resolution that is commonly used? Or do you go with what looks good.

 

[GregLocock]

Mobility (velocity/force) and impedance (you name it) are trivial frequency domain maths, assuming you start with accelerance (acceleration/force) FFTs like any normal human being.

14 response channels and 1 input is well within the capabilities of any serious laptop based system, but the front end will obviously cost more than a 4 channel unit, I'd GUESS $2000 per channel.

"Another question kind of on topic, I've been exploring the FFT function averaging and overlap. I have no idea what is standard for frequency analysis at such high frequencies. "

There is no standard. Your frequencies aren't especially high, in the next office they work at 20 kHz all day. Many people will run 50% or 75% overlap, there is no penalty (but little advantage) in going to 99% except for calculation time.

"Increasing the block size (points per fft) decreases the frequency step(bin size), though it "clutters" the plot. Is there a good average, standard resolution that is commonly used? Or do you go with what looks good."

Typically 400 lines usable (ie 1024 samples/frame), or 800. But that very much depends on what you are trying to do, sometimes I use 16, other times 10240.

Cheers

Greg Locock


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

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

 

[frizzit]

So digging round the forum I'm guessing if I divide the fft acceleration X(w) by 2*pi*frequency*i it will net me V(w), then I can divide that by F(w).

X(w)*2*pi*f*i V(w)
--------------- = -----
F(w) F(w)


Can I integrate the acceleration channel to velocity first and then use the fft of that to calculate mobility with a spectrum H0. H1 is standard FRF for noise at the output which is what I was using for my impact testing.

and if I wanted impedance I could choose F(w)/V(w) = Z(w)

Spectrum H0 DIAdem calculates the transfer frequency response by dividing the FFT of the output signal (A) by the input signal (E): FFT(A)/FFT(E). DIAdem averages the amplitudes of the individual transfer functions.

Spectrum H1 DIAdem specifies the cross spectrum and the auto spectrum for each signal pair. DIAdem calculates the transfer frequency response by dividing the averaged spectra: Middle(cross(A,E))/middle(auto(E)). DIAdem does not average phases, because phases can delete each other.

Spectrum H2 DIAdem specifies the cross spectrum and the auto spectrum for each signal pair. DIAdem calculates the transfer frequency response by dividing the averaged spectra: Middle(auto(A))/middle(cross(E,A))

 

[frizzit]

except my acceleration is in g's so:

(X(w) * 9.807)/(2*Pi*f) gets me m/s

http://www.ctconline.com/pdf/pubTechPapers/01-Beginning Vibration Analysis.pdf

page 57

 

[GregLocock]

If I were you I'd just modify the final FRFs, don't bother modifying the inputs to the FFT.

A->V is just a 1/(j*w) weighting, once you are in consistent units.

Cheers

Greg Locock


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

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

 

[SomptingGuy]

Be very careful about peak vs RMS - know what you have. Also watch out for window correction factors.

- Steve

 

[Strong]

Good luck going to 20 hkHz with a modal hammer. The ones that I have (old but still functional) do not go beyond 5 to 10 kHz with a metal tip. I am interested in what hammer you are using that can apply an impact with flat frequency response to 20 kHz.

Walt

 

[frizzit]

Walt,

it's not quit a hammer at all

http://www.pcb.com/spec_sheet.asp?model=086E80

I would never use it to hammer anything, it's so tiny.

Steve,

I have not gotten that far, and I hope to find a paper/primer/technote/website on conversions and correction factors. It seems like my math software can do it... it's just not in a black box function.

 

[GregLocock]

Yes that is the right hammer.

If you haven't done a hammer survey before then you have a whole new skill to learn, and if you haven't done a high frequecy survey before then you will need to practice a lot.

Cheers

Greg Locock


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

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

 

[Strong]

I don't own that "hammer" or impulse application device, but have used it several times for testing small compressor blades. I had to be really innovative to get clean hits (good rebound without multiple impacts). The blades (structure) was so light that I used a microphone to avoid mass-loading from the 2-gram accelerometer. Take a look at the force spectrum and Coherence to see if you are really getting all the way to 20-kHz with that "hammer"! Impacting a large structure with a bigger hammer can be easier than working on tiny lightweight structures. Good luck with that!

Walt

 

[GregLocock]

Sure. I've seen that hammer get good data off a 1/5 scale plastic model of a car body at 30 Hz, but I've also used it to get good data at 7 kHz off a disk brake. The stiffer and heavier the test item the better for high frequency work.


Cheers

Greg Locock


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

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

 

[Strong]

"I've also used it to get good data at 7 kHz off a disk brake." I agree Greg that this frequency range is possible, but the original posting said testing to 20-kHz! Perhaps more clarification of the application would help.

Walt

 

[frizzit]

I'm exploring the use of glyphXe for frequency analysis currently. One of our other locations has a license for this software.

We are not looking for frequency modes of resonance. We need to induce some energy in the 6-20 kHz range and find the best path for those frequencies through the test article.

http://www.testing-expo.com/europe/08txeu_conf/pdf/day_1/01-01-lauerer.pdf

 

[GregLocock]

Resonances or not, rubbish data won't help you. Performing an impact test and getting good coherence at high frequencies may require a great deal of practice, many hits, or may even be impossible, depending on the test item, the hammer and accelerometers and the operator. Coherence is not the be all and end all of good data but it is a necessary minimum.

Once you have good coherence your next check is reciprocity, which helps to define if your system is linear. If you do not get good reciprocity then all may not be lost but you are outside my field of experience when it comes to analysis. There can be several reasons for poor reciprocity.

Cheers

Greg Locock


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

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