Rule of thumb for the number of averages in an FFT for vibration analysis?

[Wayne_B.]

I'm doing vibration analysis on rotating machinery and would like to average the FFT to minimize variations of the data (rms averaging).  How does one choose the optimum number of averages? 

Also, why is it necessary to specify the number of averages (for example, in the Spectral Measurements express VI) as well as telling the VI to restart averaging.  Would not the latter would be sufficient when the Spectral Measurements VI is enclosed in a loop?  Thanks.

[preston_johnson]

Thanks for your post. 

In my experience 20 is a good number.  It really depends however on a number of factors. 

1) does the frequency of the rotation of the machine (speed) change?  If so, averaging from an FFT would not work well.  I would use the Order Spectrum in the Sound and Vibration Measurement Suite if speed changes.

2) are loads changing (and speed is constant or you are using order spectrum)?  If so you will want a smaller number of averages so you do not average out the effect of fast load changes.

3) try a few numbers and note when the noise floor appears stable.  This may be the best method to choose the number. 

4) how often do you want a result?  A smaller number of averages will produce an "averae complete" result more often. 

Use the averages complete boolean to gate passing on the results (to the user interface, file, further analysis, etc.).  With restart averaging on, it will restart another averaging process and toggle the boolean (gate).

Hope this helps. 

[Wayne_B.]

Thanks, Preston.  Your reply is very helpful regarding choosing the number of averages.

Though I understand your reasoning (I think) about using the "Averages Complete" boolean to gate out information, I'm still a bit foggy on why I would, for example, specify twenty averages within the Spectral Measurements VI and place the VI inside a FOR loop set to run twenty times.  If the VI continues to average until it gets the Restart boolean true input, is there a redundancy here or do I misunderstand the way the Spectral Measurements VI is set up?

Thanks,

Wayne

[preston_johnson]

Check out this simple example.  I generate a noisy signal and "gate" the output of the completed average signal using a case structure.  I use 20 averages.  For rotating machinery, the sound and vibration measurement suite also offers order spectrum with averaging.  Further, the sound and vibration measurement suite tools have a function for auto restart of averaging.  I had to use a shift register to make that happen here.

[Wayne_B.]

Thanks, Preston.  Very helpful example VI.

Wayne