FFT Integration of Accel Signals 1

[AbuNader]

Using a 4-channel OR24 analyzer with high sensitivity accels. Application is building vibration.

My analyzer will digitally integrate - no problem. But the frequencies of interest are low (<10Hz).

When I integrate to the required (by the client) velocity, I see what looks like a ski slope. That does not show up in the raw, unintegrated signal. My accels' are supposed to be accurate to less than a Hz. And the ICP cut-off frequency on my analyzer is 0.35Hz.

Is the integrated-to-velocity spectrum accurate at the very low frequencies? And, if so, is there a simple way to explain the low frequency slope that a non-FFT user can understand?

 

[GregLocock]

Do you see the ski slope in the time domain, or the frequency domain? If you could post the values for the acceleration spectrum 0-10 Hz, and the velocity for the same frequency range, it would be much easier to guess what is going on.

Incidentally I am a little bit suspicious of the frequency response of your accelerometers.

Bear in mind that digital integration is just a case of dividing by the frequency, so any low frequency noise will suddenly become apparent.





Cheers

Greg Locock

 

[electricpete]

Some discussion of causes of ski-slope effect

http://www.wilcoxon.com/knowdesk/ACF217.pdf

Integration increases the effect of any broad-band low-frequency noise. If noise is present due to sensor overload from heavy impacting or mounting problems, ski slope will be there.

Even without these problems the normal noise can be a problem if you are particularly interested in low frequency applications. Some discussion on the subject of overcoming ski slope monitoring low-speed equipment using SKF monitoring equipment:

http://www.skfcm.com/service/support2/New Library/CM3051 CMVA60 ULS.pdf

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

 

[Tmoose]

OROS chucked me and my registration attempt out thru the swinging doors and into the street. So I could not see their published specs. If I understand it right, you collect TWF data, then process it via software. I assume you could pick a starting point other than the beginning of the data set for analysis.

On our various general purpose machinery monitoring data collectors/analyzers To make the best of the situation we have to supply power to the accelerometer for at least 30 seconds before acquiring data. This was very apparent monitoring a double integrated overall signal. Even with continuous power The overall value was HUGE for several seconds. On later collector versions we had a double-key-strike option allowing us to initiate the data collection only when we were ready, after the overall had settled out, which reduced the ski-slope nicely. Still not at the low frequencies you are attempting to monitor.

When I used an earlier collector with a neck-strap I found my normal fidgeting could induce a huge ski-slope. I was bending the cable at ~ 1 Hz.

More than one accelerometer vendor has mentioned both mechanical and thermal settling-out times are required to avoid spurious ski-slope. Maybe one of these would help?

http://www.foxunlimited.com/Fur-Girls.jpg

 

[tomirvine]

I recommend using a B&K 2635 (or equivalent) charge amp with a B&K charge mode accelerometer. (Disclaimer: I am not a B&K salesman).

This charge amp model performs integration on the analog acceleration signal in the time domain after passing the signal through an analog high-pass filter. The high-pass filter frequency has several settings. A typical setting might be 1 Hz. The filter removes the DC offset which would otherwise cause the slope effect in the velocity signal.

Tom Irvine

http://www.vibrationdata.com

 

[GregLocock]

Read the manual for the 2635. Wonderful though it is there are limitations to the integration circuit. I'm not sure if they are described in the user manual or the service manual.



Cheers

Greg Locock