Vibrations Analyser Uncertainty Evaluation

[200500]

Hello,

I built a vibrations analyser using Labview and a National PCI Board. Now I want to evaluate it's behavior according to the recomendations of the Guide to the Expression of Uncertainty in Measurement (GUM), published by ISO or ANSI.

I have taken sensors, conditioning units and DAQ board's uncertainty data to compute the combined uncertainty of the measurement chain (composed by: PCB Accelerometer -> PCB Signal Conditioner -> National DAQ -> FIR Band Pass -> Hanning Windowing). However, I have some trouble in dealing with the uncertainty of the FIR bandpass filter and the hanning window error. The filter has a amplitude ripple in the passband of 1 db higher or 1 dB lower the input value. In addition, the hanning window adds 1.42 dB of "worst case error". I don't want to calibrate the system in the hole range I intend to use it. Instead of doing this, I would like to determine a combined uncertainty that includes these uncertainty of the FIR and Hanning algoritms.

For the FIR filter, it seems to be easy, and I assumed 1 dB higher or lower of uncertainty, but in hanning I don't know how to put it in my counts.

What is this 1.42 dB "worst case error" and how does it apear in the measurements results? Is this an error that can assume higher or lower values inside this range (1.42 dB)? Or is this an error that has an absolute deviation of 1.42 dB (I mean, 0.71 higher and 0.71 lower the real value)?

Thankyou a lot,

Marco

 

[electricpete]

I believe the nature of any finite windowing error is that it will take a discrete single-frequency peak and "spread" it across a range of frequencies.

(Multiplication by window function corresponds to convolution in the frequency domain).

The error lies in the fact that the energy can be spread across multiple bins, making the spectral peak appear lower than it actually is.

So, in my opinion this error will always be in a direction to make indicated vib less than or equal to actual.

Note this error only applies if you are concerned with the amplitude of a given peak. Different story if you are talking about an overall.

=====================================
Eng-tips forums: The best place on the web for engineering discussions.

 

[electricpete]

http://www.dataphysics.com/support/library/downloads/articles/DP-Windowing.pdf

Look near the bottom of the 2nd column on first page.

If the actual frequency falls midway between bins the worst case error of 15% (1.4db) occurs. The "reported" (indicated) is less than the actual.

=====================================
Eng-tips forums: The best place on the web for engineering discussions.