LabVIEW

cancel
Showing results for 
Search instead for 
Did you mean: 

sound vibration 1/3 octave

I am trying to reconcile the results of a 1/3 octave analysis done on an HP spec analyzer and that processed through the Sound & Vib tool box. There are some big differences and I would like some help understanding why. The input signal was split so the same transducer was used for both measurements. The HP took samples for 16 seconds and the recorder feeding the LV routine ran for 48 seconds at 50kHz rate.
 
Let me know if you have some insight.
 
Frank
0 Kudos
Message 1 of 14
(3,575 Views)
We need more information.

Since the two devices are averaging for different lengths of time (by a factor of 3!), the signals must be stationary in order to yield the same values for the 1/3 octave spectrum. That is the first issue to address.

The next issue is what is the nature of the differences. Are the sprectra different in amplitude, frequency, or both.

I do not have the Sound & Vibration Toolbox but including your code would be helpful in general, especially if the problem is one of improper amplitude or frequency scaling, or incorrect input parameters in the functions.

Some instrument model numbers may be helpful as well.
jc
Mac 10.4
LV7.1
CLD
0 Kudos
Message 2 of 14
(3,557 Views)

I appreciate the response but I'm really looking for somebody who knows what goes into the toolbox and how that would compare to a more traditional spec analyzer. The manual talks about meeting 'ANSI S1.11-1986, Order 3, Type 1-D, optional range' for the filter specs. I'm confident that the routine is scaled correctly etc because I can run a cal signal through it (159 Hz at 1g) and get the correct 120 dB ref 1 micro g.

On the issue of averaging, I can't argue. The spec analyzer only takes 16 s but doesn't talk about averaging over this time period. It definately averages when we perform narrow band measurements because you can set that up with the user interface screens.

Maybe I'l just try taking the 1st 16 sec of data and compare those two data sets.

If I don't get any other response, I'll get back to you.

Thanks,

Frank

0 Kudos
Message 3 of 14
(3,556 Views)
You are probably going to be asked about the nature of the differences, i.e. what concerns you. I am afraid that you won't get as much response from a vague general description of the problem as you will with a detailed description, e.g. "The spectrum has the right shape but is different in amplitude." or "There are the same number of peaks but they are in different locations." or "The two prominent peaks have different relative amplitudes."

I don't think you will get a useful answer in the absence of probing questions unless you provide more detail. I am going off line. I'll check in tomorrow and see how it's going. Good luck.
jc
Mac 10.4
LV7.1
CLD
0 Kudos
Message 4 of 14
(3,551 Views)
Frank,

Thanks for posting to the NI forums.  I will be happy to discuss any descrepancies you see if you provide a few more details.  What exactly is the difference you are seeing?  Are you using one of the SVT examples to take the measurement or is it a program that you wrote yourself?  The first thing I would confirm is how the HP device is taking the measurements and try to set the parameters of the 1/3 Octave vis the same.  For example is the HP machine applying any weighting (A Weighting, B Weighting, etc)?  Is it performing any windowing or averaging?  If it is performing some sort of averaging what type is it using?  Is it doing exponential or linear averaging, for example?  I think using the same period (16s) of data is an excellent idea.  Let me know if you have any further questions about the sound and vibration toolkit and good luck with your application.

Regards,

Neil S.
Applications Engineer
National Instruments
0 Kudos
Message 5 of 14
(3,539 Views)

Neil,

Here are the things I know -

I'm using one of the SVT examples,

the HP is using the ANSI S1.11 - 1986, Order 3, Type 1-D filters (same as LV), now weighting for these tests (we use it for airborne noise), don't know about averaging but the manual talks about 'a 2 second stable average'

my acquisition rate of 50kHz (2.5 x 20kHz) was not fast enough with a true center frequency of 20158.74 Hz desired,

I'm concerned about anti-aliasing - HP has it, I hadn't worried about it yet,

16 s worth of my 48 s didn't matter (see attached Excel file) - there are two sets of data in the workbook, noise data (Chart 1) and calibration data (Cal) the file shows that by about 800 Hz the two solutions diverge, converge, and diverge again,

I've also attached a couple of screen shots from the routine showing the 1/3 octave plot and the scaled time series data. The primary noise is coming in at around 7 Hz - showing up in the 10 Hz bin. The magnitude agrees reasonably well with the HP but I'm concerned about the linear rise in the reported levels at the ferquency goes up.

I've got some more homework with the HP to figure out what it is and isn't doing...

Any ideas?

Frank

Download All
0 Kudos
Message 6 of 14
(3,519 Views)
Frank,

I assume that you are using a 9233 for you measurements with LabVIEW.  Is that correct?  If you are the 9233 has a variable anti-aliasing filter that changes its cutoff based on sampling frequency.  Despite this antialiasing, I believe the data at the higher frequencies is being effected by the lower sampling rate of the 9233.  I don't believe that the differences you are seeing have to do with the Sound and Vibration toolkit but with the differences in hardware and sampling rates.  Could you confirm that you are using the 9233?  Have you attempted the test with another one of our DSA boards such as the 4472 or 4461?

Regards,

Neil S.
Applications Engineer
National Instruments

Message Edited by Neil S. on 12-18-2006 09:42 AM

0 Kudos
Message 7 of 14
(3,514 Views)
I think you are right to suspect aliasing. If the digitizer you are using with the LabVIEW app doesn't have a filter on the front end, you will deifinitely see noise issues. A good test to perform is to put a filter on the front end of the digitizer (30 KHz or lower corner) and observe the effect. In the cal data, the most drastic differences show up below about 60dB down from the main lobe of the calibration signal. This may be due to differences in noise levels and dynamic range between the two digitizing devices. Some instrument model numbers would be helpful.

Some comments:

1) If you are using accelerometers, and the signals that you are analyzing are background noise signals of your vibrating system, then the linear rise is probably real. If you have an unknown background vibration level it could have any shape and still be correct. The 3dB per octave rise suggests that your noise is velocity flat is the high frequency region. Is this a reasonable expectation for your system?

2) Why does the scren capture not look like the data in the spreadsheet? Are they two different events?
jc
Mac 10.4
LV7.1
CLD
0 Kudos
Message 8 of 14
(3,509 Views)
jc,
 
I'm capturing data on a TEAC GX-1 recorder at 50kHz - no anti-aliasing that I'm aware of.
 
The system is mechanical with accelerometers. The screen captures don't go with the Excel file, they are from a background measurement. The measurement was conducted simultaneously with an HP 35670A spec analyzer from the same accelerometer. The two instruments agree on the basic ~10 Hz 70 dB but differ in the higher frequencies. HP does not show the same rise at the high end. Unfortunately I don't have that data but saw it on the display.
 
I'm actually okay with the cal data, as you note, it is 60 dB down from the main cal. My concern is the high end where in the 'real' mesurement (Excel file) where I'm 5-17 dB high using the toolbox routine.
 
I'll be looking into the noise floor and aliasing issues. I'll also be capturing more side-by-side data sets.
 
Frank
0 Kudos
Message 9 of 14
(3,504 Views)
Thanks for the info. I am leaning toward aliasing. One important piece of information will help a lot. What is the resonance frequency of your accelerometer. Generally, accelerometers have very high resonance frequencies. The response is generally enormous at resonance, but the resonance is usually well above the frequency range of interest. If your input driving force is broad band, e.g. impulsive or random, this resonance will be excited. If you are not filtering out the resonance, then there will be a strong high frequency signal which will be aliased. The difference in the two spectra may, in fact be due to the different filtering on the front ends of the digitizers. I suggest that this is what you are seeing. The rise with frequency, then, is due to the filter (or lack thereof) on the TEAC.
jc
Mac 10.4
LV7.1
CLD
0 Kudos
Message 10 of 14
(3,501 Views)