LabVIEW

cancel
Showing results for 
Search instead for 
Did you mean: 

assumed erronous amplitude in low end of frequency spectrum

Solved!
Go to solution

I use PSD function to get the amplitude spectrum of accelerometers on a pump, thru a DAQ device.

im sampling at 25600hz, using 512 samples once every second for the spectrum.

 

i dont do any averaging, as i had some difficulties making it work as i expected.

 

even when the pump is at standstill im picking up high amplitudes in the lowest or second lowest frequency bin. 

 

see attached pic: frequency spectrum mms2

 

when i do frequency domain integration, these amplitudes become more prominent. note that i set the 0hz bin to zero after the integration, as i didnt figure out how to do integration at 0hz.

 

see attached pic: frequency spectrum mms

see attached pic: frequency spectrum mms zoomed

 

If i turn off windowing; hanning -> rectangular, then these amplitudes almost disappear. maybe because spectral leakage smooths it out? blackman window made them larger than hanning.

 

Is there some known reason this could happen, is there some simple way to correct? my plan forward is to increase frequency resolution (use 8000 samples maybe) enough that i can drop the lowest bins from measurements, hopefully the pc will handle it.

 

i dont have VI available to post as im writing topic.

0 Kudos
Message 1 of 9
(2,262 Views)

Hi Jan,

 

why do people think it's ok to attach some blurry and distorted images to explain their problems?

 

Well, is the pump connected to some AC supply? Or is there an AC supply nearby? (That peak seems to be at 50Hz…)

Best regards,
GerdW


using LV2016/2019/2021 on Win10/11+cRIO, TestStand2016/2019
0 Kudos
Message 2 of 9
(2,228 Views)

Hi Gerd.

 

Because thats what they had available when writing the topic past midnight 🙂

 

i increased number of samples to get frequency resolution down to 3,125 hz and the problematic amplitudes are still in the lowest and second lowest bin (so im not picking up the mains power), so now i can omit these two bins from my overall calculation and still have an acceptable freq range. we'll do measurements with a handheld device also to see if results in mm/s match.

 

still it would be nice to know the cause. im using 9234 cards with ac-coupling enabled. edit: and pseudodifferential wiring enabled. must admit i dont 100% understand these terminal wiring choices. the accelerometers seem to be connected to ground thru a resistor.

 

Best regards Jan Egil Glæstad

0 Kudos
Message 3 of 9
(2,220 Views)

@GerdW wrote:

why do people think it's ok to attach some blurry and distorted images to explain their problems?


@JanEgilGlaestad wrote:

Because thats what they had available when writing the topic past midnight 🙂



Please use the quote button...

 

The code used to produce the spectra might be useful.

 

Often frequency spectra a 'symmetric', so the highest frequency is in the middle, the lowest frequencies on the left and right. That would reduce your two peaks to one problem.

 

Could those low frequencies simply physically be there?

 

Could it be a DC offset in the signal?

0 Kudos
Message 4 of 9
(2,197 Views)

wiebe@CARYA wrote:

Could it be a DC offset in the signal?


I assume the Original Poster has never had a course on Signals, Spectra, FFTs, etc.

 

Bob Schor

0 Kudos
Message 5 of 9
(2,189 Views)

wiebe@CARYA wrote:

Could those low frequencies simply physically be there?

starting to think so. now the whole testrig has been torn down, but ill test again in a couple of weeks with other sensors. and try with function generator or just shorting the input. i also had a brief conversation with a vibration technician today and if i understood him correctly, it is common with some "garbage" in the very low end of spectrum with accelerometers.

 

 


wiebe@CARYA wrote:

Could it be a DC offset in the signal?


there is a DC offset in the signal, but i assumed this would be filtered out with the AC coupling enabled? when i test again, ill try with a 1uf cap in series with the signal.

0 Kudos
Message 6 of 9
(2,184 Views)

@JanEgilGlaestad wrote:

wiebe@CARYA wrote:

Could it be a DC offset in the signal?


there is a DC offset in the signal, but i assumed this would be filtered out with the AC coupling enabled? when i test again, ill try with a 1uf cap in series with the signal.


Apparently. At least theoretically.

0 Kudos
Message 7 of 9
(2,148 Views)
Solution
Accepted by JanEgilGlaestad

@JanEgilGlaestad wrote:

there is a DC offset in the signal, but i assumed this would be filtered out with the AC coupling enabled? when i test again, ill try with a 1uf cap in series with the signal.

NO!  If there is a DC component of your signal, remove it.  This means (I really shouldn't have to tell you this!) that you take the mean of your signal (I believe there is a function for this) and subtract it (this process is known as "Zero-meaning the signal").

 

If there is no DC component, then no DC component will appear in the Spectrum, and it won't "leak" into adjoining spectral components.

 

Bob Schor

0 Kudos
Message 8 of 9
(2,135 Views)

thanks bob.

 

ill make sure to zero-mean the signal. lack of zero-meaning is probably the cause. i guess even a tiny bit of dccould cause this. of course i cant verify for sure that there wasnt any low level freq components since the testrig is taken down and equipment shipped.

 

anyway i believe my results should be correct after i omitted freq components below 10hz.

 

Best regards Jan Egil Glæstad

0 Kudos
Message 9 of 9
(2,127 Views)