Dear Guys,
Do you know how to identify "glitch" due to electrical problem on running shaft surface (on proximity measurement) by monitoring in spectrum, time waveform ??
I believe that glitch will be displayed as an impulsive and asymmetric time waveform..but how to separate it to mis-alignment or rubbing event?
If you have some example of the spectrum & TWF, please submit in this forum.
Thanks so much..

Rahardi
Chevron Indonesia

I hope this link helps,

http://www.bently.com/articles/apnotes/an002.asp

regards

Glitch can take on many forms, ranging from the short duration spike to an almost sinusoidal waveform from residual magnetism. Hence there isn't anything typical you can look for in the spectral domain.

What is commonly done is to observe the signal during a coast down of the machine and see if the amplitude tends to hang up to some higher than desired value.

Good software can then quantify the signal in various ways and can actually remove the glitch from the full speed signal, assuming you have a keyphasor or once-per-turn reference.

John from PA

This message has been edited. Last edited by: John from PA,

As mentioned, slow roll is the best way to identify scratches but this can also be done at speed. First, you will have to collect both X & Y waveforms simultaneous. Second, look for the event to repeat 90 degrees in direction of precession. Precession is determined by standing at the back of the driver and looking down the train. Counter clockwise rotation is "X to Y precession" & clockwise rotation is "Y to X precession". For example, if precession is X to Y then the X probe will see the scratch first, then 90 degrees later the Y probe.

Bad scratches will show up in the spectrum as harmonics. It's like taking a spectrum of a keyphaser signal.
I will try to post some pictures later.

Performing a cross-correlation on the x and y probes at low speed is interesting (Usually, this is not too difficult to determine by other means, but I have used this in reports - it shows the great value an analyst can bring ). This yields high + correlation at the phase lag of the probes.

If the glitch, noise, has some degree of independence (perhaps further from specific locations), then one can see this well.

Thanks for all your response,
I try to summarize from Erik & Bill:
If I illustrate your explanation in orbit, the display will be like in the attachment. Doesn't it?
Please your all comments..
And Erik, I really appreciate if you post your glitch TWF/Orbit.

For John,
How come "glitch" will display almost sinusoidal? Could you describe me its mechanism & illustration?

In my case, I can't do much experiment in the equipment. Operation must be kept running and I only can take orbit by 2120-2CH from Bently panel without keypahsor or photo-tach.
Please, give suggestion how to prove the glitch problem.

Thanks..

Regards,

Rahardi
Chevron Indonesia

This message has been edited. Last edited by: Rahardi,

Orbit with scratches

Spectrum of orbit above

Another Orbit with scratches

Erik,
Thanks for all your post. They are very useful.
Now, however without phase measurement, I'm sure that my problem is glitch.

Best Regards,

Rahardi
Chevron Indonesia
rahardi@gmail.com
Mobius Institute - Cat III
Vibration Institute - Cat II

I have posted this before, but it might help you out. Especially in seperating rub from glitch...

orbit_reference.ppt (264 Kb, 28 downloads)

Rahardi, a shaft that hasn't been properly degaussed will commonly give a very clean almost sinusoidal waveform. If you have X-Y transducers at the exact same axial location they will also be displaced by 90 deg at slow roll.

There is also a situation with chrome plating on a shaft that can also cause a waveform that is almost sinusoidal at slow roll. Some people consider this a "specialized" case of glitch.

John from PA

Eric, the last time I saw something similar to your post of 17 January 2008 08:44 PM it was traced down to an air operated four-jaw lathe chuck that had about 200 psi instead of the design 90 psi due to a bad regulator. Careful analysis of the time domain plot establised that the "scratches" were at exactly 90 deg intervals. The lathe operator was chucking over brass shim but there was so much pressure being applied it caused localized "glitch."

John from PA

John,
The frequency at which events occure always have to be varified in the time domain and related to possible external sources. The examples posted are for 5000 and 10000 HP machines with no possible external vibration sources at 0 and 90 degrees and not 180 & 270 for the same probe and then 90 degrees later the same events occure for the other probe.

Rahardi,
could be a scratch but please collect both waveforms simutaneous if possible on your 2120. This does not require a keyphaser but does require you to set up your analyzer to collect two reading simultaneous.

Steve mentioned rubs. Rubs occur at a point or points in time/ perhaps periodic. Runout is somewhat of a time shift phenomenon. Runout may have a 'noise' characteristic as seen in an autocorrelation - The signal may not be completely statistically independent (noise) with time shifts, but weakly (significant time shifts have no dependence upon the un-shifted runout. Periodic components in the vibration will come through on an operating machine.)

Cross correlations is still relevant at running speed conditions. The noise component will show the time shift in the cross-correlation.

The best way is still at low speed. Runout resembling a rub, though will show the time shifted nature instead of peaking at 0 time shift (and of course multiples of the running speed period.).

Bill,
can you say that in English . I'm sure some of us read that and said what the heck did he just say, including myself. Thanks

One can apply two channel signal analysis to look for delays in one signal from the other using correlation functions.