Kurtosis miss indication of impacting

[Sofistioelevib]

Hello Guys,
i have 2 data acquisition with 2 waveform one with impulse and one without.
I made some calculation with my instruments and it is very easy to evaluate impulsiveness of the waveform.
Why, instead, Kurtosis seems have missed this impulsiveness of the waveform?

Let me know about your opinion.

regards

 

[IRstuff]

Why do you think it should? Kurtosis is about the characteristics of a tail on a statistic distribution, which doesn't appear to be what you are measuring.

TTFN (ta ta for now)
I can do absolutely anything. I'm an expert!

https://www.youtube.com/watch?v=BKorP55Aqvg

faq731-376 forum1529 Entire Forum list

http://www.eng-tips.com/forumlist.cfm

 

[GregLocock]

I think you'd get better results by looking at the crest factor, peak over RMS, or the amplitude distribution.

Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376

http://eng-tips.com/market.cfm?

 

[Sofistioelevib]

IRstuff
i'm starting to use kurtosis as "new frontier to find bearing problem" as some salesman use to sell his products.
Some other speak about technique to find impacting or impulsive signal....
I wanted to try personally and i found it is not really as aspected.
Anyway i know some technician use it with good results.
Or i have misconception about how Kurtosis work and the waveform in attach it is a rare case where kurtosis doesn't works or kurtosis is useful for other type of investigation.

I'm not really familiar with statistics and i didn't find yet a simple explanation about how Kurtosis work.

GregLocock
I used crest factor several time but also in this case the crest factor value is not really big. In this case is simple to say crest factor is better indication because you have a comparison signal. In any case i could not to assert impulsiveness only from crest factor because it is very low value.

 

[IRstuff]

Given that your "signal" rides on another signal of the exact same frequency, it seems dubious to extract the difference in any unfiltered means, unless you compare measurements wit, and without, the spikes. It just seems to me that something like an unsharp mask approach to extract only the high frequency spikes would be required.

TTFN (ta ta for now)
I can do absolutely anything. I'm an expert!

https://www.youtube.com/watch?v=BKorP55Aqvg

faq731-376 forum1529 Entire Forum list

http://www.eng-tips.com/forumlist.cfm

 

[IRstuff]

I think this article says something similar:

http://downloads.hindawi.com/journals/sv/2002/647652.pdf

TTFN (ta ta for now)
I can do absolutely anything. I'm an expert!

https://www.youtube.com/watch?v=BKorP55Aqvg

faq731-376 forum1529 Entire Forum list

http://www.eng-tips.com/forumlist.cfm

 

[GregLocock]

Fundamentally the issue is that the there is relatively little energy in the spike compared with the base sinusoid. One approach I've seen for a signal like this is to gate it, that is, average the time history for each cycle starting from the same shaft position. On the other hand, as IRStuff implies, a high pass filter may be all that is needed.

Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376

http://eng-tips.com/market.cfm?