Sign up Calendar Latest Topics Donate
 
 
 


Reply
  Author   Comment   Page 2 of 2      Prev   1   2
Curran919

Avatar / Picture

Sr. Member / Supporter
Registered:
Posts: 599
Reply with quote  #16 
Ah... I think we are talking about different things. When I said peak, I meant the group of 0-pk and pk-pk, not the spectral frequency peak. [rofl] Whoops! Sorry John.

I meant, that if the y-axis of a spectrum is labeled IPS 0-pk, the measure must be pseudo 0-pk and cannot be true 0-pk.

Though, if you want to start a talk about scalloping, leakage and the beauty of flattop, we can do that too!
John from PA

Sr. Member
Registered:
Posts: 963
Reply with quote  #17 
Quote:
Originally Posted by Curran919
I meant, that if the y-axis of a spectrum is labeled IPS 0-pk, the measure must be pseudo 0-pk and cannot be true 0-pk.


I actually think that a peak on the y-axis that is labeled IPS 0-pk, is very close to true peak however it may be obtained.  The analyzer is using a monolithic filter to establish a voltage amplitude.  It is a sine wave and hence the scaling can work.

Curran919

Avatar / Picture

Sr. Member / Supporter
Registered:
Posts: 599
Reply with quote  #18 
We're still not on the same frequency here. Here's a real velocity waveform. The time is about equal to the spectral frame length. Here, 0-pk ~ 4.5 mm/s, right? Now the spectrum on the right is RMS. If we scale that up 1.4x, the pseudo peak of specific components would be max 1.13mm/s. What do you compare this value to? Its not a question of accuracy, its about comparing apples to orchards.

pseudopeak.png 


John from PA

Sr. Member
Registered:
Posts: 963
Reply with quote  #19 
With all the spectral peaks it isn't the best data to work with but I gave it a shot.

Step 1.  I broke your spectrum into the (11) largest peaks.  Visually, and going from left to right, they appear to be in amplitude as follows:

0.45 mm/s rms
0.80 mm/s rms
0.32 mm/s rms
0.28 mm/s rms
0.07 mm/s rms
0.08 mm/s rms
0.06 mm/s rms
0.26 mm/s rms
0.58 mm/s rms
0.20 mm/s rms
0.13 mm/s rms

Step 2.  Each of these amplitudes is obtained by the action of a monolithic filter so they represent the rms amplitude of a near perfect sine wave.  Thus scaling up by a factor of √2 to reach the peak value of each component applies with fairly good accuracy.  On that basis the values below in red are the 0-pk values of each component

0.45 * √2 = 0.64
0.80 * √2 = 1.13
0.32 * √2 = 0.45
0.28 * √2 = 0.40
0.07 * √2 = 0.10
0.08 * √2 = 0.11
0.06 * √2 = 0.08
0.26 * √2 = 0.37
0.58 * √2 = 0.82
0.19 * √2 = 0.28
0.13 * √2 = 0.18

Step 3.  Sum the calculated 0-pk values and compare that to your visual "shot" of the 0-pk waveform that is estimated to be 4.5 mm/sec.

∑ = 4.57

Very close!  

My Excel tabulation is shown below.

tabulation.jpg 



John from PA

Sr. Member
Registered:
Posts: 963
Reply with quote  #20 
The spectral component breakdown

spectral breakdown.jpg 

Curran919

Avatar / Picture

Sr. Member / Supporter
Registered:
Posts: 599
Reply with quote  #21 
Quote:
Originally Posted by John from PA
Sum the calculated 0-pk values and compare that to your visual "shot" of the 0-pk waveform that is estimated to be 4.5 mm/sec. ∑ = 4.57 Very close!  


That's the step I couldn't infer. Honestly, I've never seen someone linearly sum peaks from a spectrum before, and I can't imagine it a physical basis for it. Is this a recognized method? If you sum the highest 20 peaks, you get 5.8 0-pk. Hell, if you take the linear sum of the entire spectrum, you will get 30 0-pk! You've got to 'fudge it' for it to make sense.

The only situation in which your method would work as a workaround, would be with a discrete number of frequencies, all non-multiple of eachother (e.g. all prime numbers, so all phasing is possible), and you have a very large (infinite) time waveform to allow for all of the frequencies' peaks to line up perfectly at some point in time. ONLY THEN will the linear sum of the peaks (assuming 0 scalloping and 0 leakage, which would be impossible with 'prime number' frequencies) equal the 0-pk value derived from the waveform (without fudging).

I can definitely imagine waveforms where the pseudo 0-pk overall (RMS summation) will have much higher error than your method (quasi linear summation), but from my perspective, its still strictly an empirical quantity, and calling it true 0-pk is just not... true. [frown]
John from PA

Sr. Member
Registered:
Posts: 963
Reply with quote  #22 
Quote:
Originally Posted by Curran919


That's the step I couldn't infer. Honestly, I've never seen someone linearly sum peaks from a spectrum before, and I can't imagine it a physical basis for it. Is this a recognized method?

I can definitely imagine waveforms where the pseudo 0-pk overall (RMS summation) will have much higher error than your method (quasi linear summation), but from my perspective, its still strictly an empirical quantity, and calling it true 0-pk is just not... true. [frown]


We are just doing what Fourier said we could do; any complex waveform can be broken down into a series of sine waves.  Go to https://en.wikipedia.org/wiki/Fourier_series and watch the video develop (takes about 10 seconds) along the right edge.  It goes from a complex waveform created from six harmonic functions to the generation of a spectrum.

Key here is to remember that the spectrum analyzer is a monolithic device.  Monolithic means a single block and in this instance the "block" is mostly defined by the span and the lines of resolution.  The amplitude of that block is a sine wave and for that we can apply √2 to an rms ampitude and yield a peak value.

In my example of the 50% duty cycle bipolar square wave (2 EU peak to peak) I stated that the pseudo peak value was 41.4% high.  In contrast a 2% duty cycle pulse going from 0 to +2EU will have a pseudo peak that will be 60% low.

You might want to set up a function generator and do some simulations yourself.  Just use something like an o'scope or true peak meter to adjust the overall amplitude so it is constant.  

So the summary is right along with the point you were trying to make; pseudo peak is of little value. 

fourier.jpg 



fburgos

Sr. Member
Registered:
Posts: 670
Reply with quote  #23 
according to TA of charlotte the digital overall from spetrum is


root(sum(Amp1^2+amp2^2+amp.....))/root(1.5)

but also this is an rms i guess you need to correct each amplitud by root(2)
OLi

Sr. Member
Registered:
Posts: 1,913
Reply with quote  #24 

Is the TWF avg 1 and FFT avg 479 or another avg, just curious? 
Practically, what would you like the pk to indicate?
A value that can be related to the RMS or derived from it or
a value that give you the peak including the off chance of a bearing fault
transient being captured on top of the signal? Or the similar value that you would
get from a old time analog peak detector of those variations of design that would give?
So I think in my mind I would like that, the same as a analog peak detector with some decay setting.......
I can't see the use for it but that's me, trying to find a limit close to bearing clearing limit in a eddy probe setup is
interesting. 2 weeks ago I was called as the data was obviously wrong, indicating 130% of the clearance set in the bearing
and when I told them it was RMS it was even more upsetting and when it was eventually found to have that large clearance also IRL
it was totally confusion..... So it is not only peak detection that give problem here. So my conclusion is that I don't like pseudo peak it is useless
just call it RMS if that is what it is. How much pseudo I am willing to accept would be difficult to say. Having all computer power we have now even in the phones it
would not be any problem to do the peak value evaluation on a correct Inverse FFT with phase and all and get the peak from the processed TWF if needed. If you would do it
mid 1980's on a 8 bit Z80 CPU things would be different but today I think you should try to get the peak as good as you can get it if you need it but that's only me.

 


__________________
Good Vibrations since early 1950's, first patented vibrometer 1956 in the US.
http://www.vtab.se
John from PA

Sr. Member
Registered:
Posts: 963
Reply with quote  #25 
Quote:
Originally Posted by fburgos
according to TA of charlotte the digital overall from spetrum is


root(sum(Amp1^2+amp2^2+amp.....))/root(1.5)

but also this is an rms i guess you need to correct each amplitud by root(2)


That is a correct statement.  But if the amplitudes (Amp1, amp2, etc.) are RMS, then the final result is RMS.  RMS = root mean square, or the square root of the sum of the squares.  But the point of the discussion is there are many devices out there that then take the RMS value and scale or electronically gain the RMS up by a factor of √2 (1.414) to reach peak.  That only gets you a "pseudo" peak value that can have huge errors.

The operation of using √2 *rms = 0-peak only holds true for a sine wave.
fburgos

Sr. Member
Registered:
Posts: 670
Reply with quote  #26 
Agree, pseudo peak is really just an rms, I´ve used rms value for trending all my carrer, my country lacks of "normatives" on my previus company someone was first instructed by schenck (German company) with ISO2372, then we all follow ISO path.

Never put my mind in to the real difference between peak and rms, I now se a clear difference between real peak, rms and crest factor, thank you all.
RustyCas

Avatar / Picture

Admin
Registered:
Posts: 1,810
Reply with quote  #27 
I've never really gotten into the intricacies of signal processing, so I can't speak to this with the intelligence of John, or Curran, or Pete.  But unless you are commissioning a machine, or evaluating it relative to a "standard", I don't see that it matters as long as you are consistent, within the environment you are working in.  I've always used "peak" velocity, which if I understand correctly, in the CSI world is √2 * RMS value.  This is partly because of the customers I work with.  Displacement, velocity, and acceleration mean very little to them.  They pretty much have no idea what I'm talking about.  They just want to know how soon it's going to break. 

Over time I calibrate them by making reference in my reports as "vibration is fairly high at 0.5 in/sec" (0.35 ips RMS is 'fairly high', right?) or "vibration is extremely high at 1.1 in/sec" (correct?) or "vibration has increased from 0.3 to 0.6 in/sec this month" (I think they intuitively grasp that vibration 'doubling' in a month is probably bad).  'Peak' is higher than RMS, so I think it's more conservative to use that.  After all, I'm not asking them to decide if it's "bad" or if the machine needs to be shutdown or rebuilt -- that's what they are paying me for.  Since I seldom call for any action by the customer on early stage faults, when I write something up I'm usually trying to convince them it's time to take action.  I'm trying to draw them a picture that is self-explanatory.  I always say if I can't use the data to draw a picture that is clear and convincing, then maybe there's not really a problem. There are exceptions of course, and I'll say "the data is not convincing, but this needs to be addressed ASAP!" 

But this is the rough-and-tumble world of mostly run-of-the-mill equipment where the "change" in vibration is much more important than accurate amplitudes.  It's where I live and work.  But I understand that this is just a subset of what is out there, that there are many instances where accuracy absolutely matters.

Sorry if I've stated the obvious, but I think it's important for those who are perhaps new to this world to understand that much of what is written, taught, and stated has to be evaluated in light of what type work you are doing, and who your audience is.  It's important to "know" the terms, rules, and conventions, but what you do with that knowledge... well, it just depends.
 

__________________
"The trend is your friend"
Curran919

Avatar / Picture

Sr. Member / Supporter
Registered:
Posts: 599
Reply with quote  #28 
Quote:
Originally Posted by John from PA
We are just doing what Fourier said we could do; any complex waveform can be broken down into a series of sine waves.


Okay, so I'll say now that I am painfully familiar with the fourier transform as a superposition of perfect sine waves. I had just never heard the term monolithic filter, but that is neither here nor there.

You are forgetting a critical piece of the fourier transform though, and one that we normally ignore because it is of little interest. A fourier transform gives the complex vector of each frequency, so it also gives phase of the components, relative to the beginning of the frame. The waveform cannot be reconstructed with the component magnitudes alone!

Lets look at the example you gave, the fourier series that makes a square wave. This works because the component n+1 has a minimum where component n has a maximum. It provides deconstructive interference that keeps the superposition amplitude low. However, see how all of the components start at the same phase to the beginning of the frame. If we take all of the even components (2nd, 4th, 6th) and shift them 180 degrees, the peaks of all of the components will line up perfectly once per period, and as a result, constructive interference will greatly increase the pk-pk amplitude of the superposition waveform with no change to the RMS of the superposition.

In fact, with this shifted form of the square wave signature, where 'all the planets align' in every period, your linear sum technique of calculating the overall peak will work perfectly. However, it explicitly works terribly for the normal square waveform approximation in your figure.

Quote:
Originally Posted by Olov
Practically, what would you like the pk to indicate?


I've read quite a few proponents of peak claim that severity for some vibration failure modes is proportional to peak more than RMS. However, I imagine those same failure modes may be even better correlated to crest factor. Therefore, measuring true peak may be useful, but every reporting vibration levels in peak may not be.

Quote:
But this is the rough-and-tumble world of mostly run-of-the-mill equipment where the "change" in vibration is much more important than accurate amplitudes.


Well said. You can measure your vibration in yards per minute pseudo peak-peak for all I care if all you are doing is trending it, you will get the same outcome. The only problem is when you need to handover your tasks to someone else, or as Oli has said, when you need to change out self-integrating accels with new ones that may report something different and your trend suddenly takes a 41% bump.
electricpete

Sr. Member
Registered:
Posts: 647
Reply with quote  #29 

I'm late to the party.  Lots of good discussion and I’m pretty sure everyone understands the underlying relationships. All that’s left is opinion  or spin. Here’s mine:

If I have a bunch of sinusoids, I can compute the rms of the sum as the SRSS of the rms magnitudes of the individual sinusoids.  It is well defined in math textbooks worldwide.    There is very little potential for confusion.

If you choose for some perverted ;-) reason to apply a scale factor to the individual rms magnitudes first, and then combine those scaled sinusoids using SRSS,  then you will come up with something vibration people call an “overall” that is scaled by the same factor as the individual rms sinusoids were.  That’s what we do for our pseudo peak overalls: instead of considering the rms of the individual sinusoids, we consider the individual sinusoid by their peak value and the resulting pseudo peak overall of the sum is sqrt(2) times higher than the rms of the sum.  It is an unfortunate choice, but it’s what has been done a lot of places… taught in technical associates training… built in option for the software we use… pretty common in the US including at my plant.   It introduces vast potential for confusion due to new terms like “overall” that are not defined anywhere outside the vibration industry (and still needs to be further clarified by pk/0, pk/pk or rms within vib industry).  Likewise if you use the term "peak" there is now a question of true or pseudo.  

You may ask which is more natural way to consider the individual sinusoid: by it’s peak or its rms.  I look to the electrical power world. If someone tells you you have 5 amps ac current at 120 volts ac, they are invariably talking rms. It's what's built into the meters. It is also built into familiar equations like resistive power P = I*V, where P in watts, V in volts, I in amps (That equation and unit system would give incorrect results if you plugged in peak values of voltage and current).   And that is when we are talking about a single pure frequency like 60hz!!!.... When we need to combine multiple frequencies the inclination toward rms should be even stronger imo. 

The sole benefit I can see for abandoning the rest of the world and characterizing the individual sinusoids by their peak (rather than rms) is that we can easier look at sinusoidal TWFs and judge the peak than the rms.  Is it worth it?  My opinion, no way.  But that's just an opinion, everyone's got one....

Previous Topic | Next Topic
Print
Reply

Quick Navigation:

Easily create a Forum Website with Website Toolbox.