ok, dumb question from someone who doesn't use autocorrelation. What the heck is this waveform telling us? Particularly the side-band looking things (I know that's not the right term).
The waveform is saying "Hey, when I compare the "left" half of the stored waveform to the "right" half, all I see repeating itself is the inner race frequency."
The "sidebands" are not really SB, they are all, or appear to be, inner race frequency.
What I thout I was seeing was a series of smaller time peaks and every 5th peak is larger than the rest.
After studying the frequencies listed on the side of the graph, it looks like the spacing between the smaller peaks is BPFI and spacing between larger peaks is around 1x (BPFI ~ 5 orders). So BPFI impacting modulated by 1x makes sense. It just took me awhile to figure that out. In an indirect way, each of those "peaks" (for lack of a better word) in the autocorrelation TWF represents the magnitude of a repeating impact. In a time waveform it is easy to recognize when a peak represents an impact because we have a ringdown. It's not quite as easy in an autocorrelation waveform where there is no ringdown (a repeating peak in the autocorrelation TWF could also represent a sinusoid instead of an impact... right?). But I can see there might be an advantages in case when time waveform has other stuff in there to complicate things.
This message has been edited. Last edited by: electricpete,
Originally posted by electricpete: But I can see there might be an advantages in case when time waveform has other stuff in there to complicate things.
That's exactly why I use it--to uncomplicate things. If there is a high correlation factor, I might dig a little deeper. If the correlation plot is relatively flat, ie. not much to correlate, then I'll usually move along. It's another tool.
Patrick
Posts: 381 | Location: NJ | Registered: 19 April 2004
The correlation factor calculation is applied in this case to a PeakVue TWF where impacting will not show up in a same way as it does in a regular TWF.
Each peak in the correlation factor calculation represents by design the degree of periodicity (repeatability) but its magnitude is also affected by the ratio of this periodic signal to noise. Therefore even a perfectly repeatable signal in presence of high noise will be low in magnitude in the correlation factor plot.
IMO this should not affect problem severity assessment. As long as there is a repeatable pattern, such as in Danny's attachment, even of small amplitude, the problem is confirmed.
As I said previously same periodic signal with modulation can be probably clearly seen in the spectrum.
Could you attach it, Danny?
David
This message has been edited. Last edited by: David_G,
Posts: 980 | Location: Texas | Registered: 22 February 2005
What information is provided in the autocorrelation plot that is not seen in the peakview spectrum. I presume the peakview is showing harmonics of the of 8534 with motor speed sidebands.
Also-- while we are on the subject--are any of you aware of a good description of the use of autocorrelation. Although it is not a part of CSI's software, I have not found useful literature on the plot on the CSI website.
Posts: 45 | Location: US | Registered: 26 May 2005
Typo---Although it is now part of CSI's software....I see some of you have edited your posts. How do you do that for those of us that kant spull? Thanks lawrencep.
This message has been edited. Last edited by: Testtech,
Posts: 45 | Location: US | Registered: 26 May 2005
Wow Bill, That was pretty intense. It looks like it's made for peakvue more then anything. Why don't they show any 360 deg plots as they give the option to do in RBM?
Posts: 11 | Location: West Point Va | Registered: 19 November 2005
I have attached plots of the spectrum and the expanded twf with the fault frequency line shown. I had to expand the twf from its original 48 shaft revs so that the fault frequency lines would show up.
It's pretty easy to see in the spectrum, not so easy in the twf then clear as a bell in the autoccorelated twf.
I use the spectrum for preliminary diagnosis, the autocorrelated twf for confirmation and the twf for severity. I almost never make a call based solely on PeakVue on a motor bearing. This one is a Level 3 of 5 for me which means check the lube (they don't) and keep on running for probably at least another year. Maybe less because it is an inner race. I'll try to remember to update the posts until they let it lock up.
"Each peak in the correlation factor calculation represents by design the degree of periodicity (repeatability) but its magnitude is also affected by the ratio of this periodic signal to noise. Therefore even a perfectly repeatable signal in presence of high noise will be low in magnitude in the correlation factor plot."
Can you elaborate on the last statement and maybe give an example of where that would occur?
Danny
Posts: 1595 | Location: Midlothian, VA, US | Registered: 22 February 2005
It follows from the equation (2) on p.3 and property 2 on p.4 in the referred paper.
In your example, although there are clearly standing out peaks in the PeakVue spectrum, the value of AF=0.25 and less. This is a result of significant noise present in the signal. Noise can also be seen by eyeballing in the expanded PeakVue TWF.
BTW, one can clearly also see in the AF modulation at 1x. It is also present in the spectrum.
David
Posts: 980 | Location: Texas | Registered: 22 February 2005
Since we have been discussing autocorrelation, I thought I'd throw this in. This fan bearing exhibited elevated high frequency vibration. The TWF shows a once per revolution impact. The Peakview data shows a peak at shaft speed (1041) with another peak at 15562. This is about 15x shaft speed. The autocorrelation plot shows a very strong shaft speed frequency with an additional repetitive frequency at 15360. I believe what all this means is that we have a once per revolution impact in the bearing that is exciting a resonance at at 15562. In this case, the autocorrelation seems to provide the simplest insight into this problem. The shaft speed vibration is very low. There is no indication of looseness. I expect there is a defect in the bearing that is being struck once per revolution.
