Fractinal harmonics, such as 1/2x or 1/3x orders don't make feel anyone good in general. How do you assess the situation in the attached file? It is a direct driven overhung fan with tapered sleeve mounted spherical rolling elements bearings. Speed is accurate down to
1 RPM and as you can see it has a few exact 1/3x and many 1/2x orders consistently present in the data.

Should it be taken seriously? What do you think?

David

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

One_third_order.doc (40 Kb, 77 downloads)

Dave,

It may not represent an immediate shutdown condition, but internal bearing looseness should be corrected to to minimize premature bearing/shaft/housing wear and failure. Has the 1xSS and HFD acceleration levels gone up over time?

It is always smart to use early warning information to mitigate epensive repairs or an uncontrolled outage. If someone else wants to continue running with a fault condition, then he/she should accept the consequences.

Walt

I think the best severity assessment is done from the waveform. Why would you look at integrated, windowed, averaged spectrums to determine severity? Spectrum analysis has its place, but most people rely on it way too much. Post acceleration TW and Peakvue if possible.

Good comments from Walt and Bill.

I have never heard of 1/2x and 1/3x harmonics present in same spectrum.

It appears to me there are no odd harmonics of the 1/2x (no 3/2x no 5/2x, no 7/2x etc). Are you positive the 1/2x is exactly 1/2x and not perhaps vibration coulpled from another machine running half as fast?

If it is real, my guess would be looseness at more than one location (I would not think a single instance of loosenss would cause this).

Perhaps by external measurments you can check for loosess at the machine base.

If I'm pretty sure it's "looseness" I'm seeing, I interpret fractional harmonics as "gross looseness." How serious is it? How serious is any internal bearing clearance problem. It's one of the things I don't worry a lot about because if a bearing has adequate lubrication and clearance it will run until it falls apart which is usually quite a while. My response is usually "repair when convenient".

A rub can also present fractional harmonics. So can binding in a metal-to-metal coupling at slower speeds. Just remember that fractional harmonics in the spectrum probably don't represent real, physical "events" that are occurring every 2nd or 3rd revolution, but are more likely just the FFT's way of representing random or quasi-random events in the waveform (as Bill alludes to).

I'm with Bill on using the waveform as my basic severity indicator. Until I start to see lots of impacting, or high amplitude spikes, there's probably not a lot going on.

I collect acceleration waveforms for ALL rolling element bearings and trend the "max peak waveform" (CSI's term) value.... not the peak-peak acceleration.... I am concerned with "impacts" which may be peak-peak, but are often only peak values.

IMO, the velocity waveform is only good for low order mechanical problems... looseness, imbalance, misalignment. The velocity waveform is a "derived" representation and does not accurately reflect real-world events. The acceleration waveform is simply the raw output from the accelerometer (digitized, of course), so it always gets the most weight with me. ALL spectra are derived representations, so I don't think it matters as much which units you use. The spectrum is not "real" anyway.

I agree with Rusty that fractional harmonics represent gross looseness. This comes rather from experience then from understanding of physics or from math modeling of looseness. Yes, there is no events occurring every 2 or 3 revs, but instead there is modulation of 1x every 2 or 3 revs and that is what produces fractionals. In this sense FFT is real, one just has to interpret it properly. As a side note the fractionals are rather 1/3 sideband harmonics, then harmonics of 1/3x being fundamental, IMO.

Having agreed on fractionals as indication of "gross looseness" I believe that looseness is present in the bearing fit on the shaft rather then due to excessive internal clearance and this is obviously dangerous.

As Pete noticed, having both 1/3x and 1/2x is strange and I agree with it. Is it an indication of 2 independent sources of looseness?...
Regarding transmitted vibration possibility, although there are plenty of other machines running in close proximity and all of them including the problem one are mounted on a mezzanine, there is nothing running at 1/3x or 1/5x.

I looked previously at the TWF and PeakVue, nothing attracted my attention. But I'll take a look again and will post it.

Regarding Bill's comment on overly reliance on spectra. I agree that TWF will better show impacting but to fish out 1/3x harmonics from a TWF? Not always easy. IMO, the fact that for integration errors. What is important is the mere ability to detect fractional harmonics using a spectrum, IMO.

David

quote:

Yes, there is no events occurring every 2 or 3 revs, but instead there is modulation of 1x every 2 or 3 revs and that is what produces fractionals.

On a number of shaft rubs with 1/2X vibration the shaft hits the stator every other rotation in all likelyhood. It works in theory and simple experiment too.

One can more easily detect an exact sub-harmonic using a triggered time waveform than a spectraa.

Fratcional harmonic..
not only rubs
The principal cause is non-linearity maybe of movement or stiffness variations for example craked shafts.
When 1/2x is caused by looseness there are 3/2, 5/2 etc.

Ricardo Góz from Brazil

Here is some more food for thoughts, although not much, IMO. (Please read comments under the plots). It turns out that spectrum is the best indicator in this case, IMO. One can add raised floor as another symptom but with the fan sitting on a mezzanine this might be expected and has nothing to do with a loose bearing on the shaft.

This fan will be inspected and I'll post results.

Again, 1/2x along with 1/3x pattern is probabaly not very common. Is it a result of two sources of looseness?

David

Sys_fan4_fractionals.doc (150 Kb, 56 downloads)

One thing to consider may possibly be Rotating stall of the fan. Based on what I am seeing in your data, this is not an occrurence that just happened but is present in all the spectra. Rotating stall should show a dominant 2/3 order (which I believe that I see in your data) as well as possible harmonics @ 4/3, 6/3 etc. If this fan has an inlet damper, it may be worth it to check and ensure that it is installed in the correct direction.

Viberscott,

Talking of fractional harmonics I have to note that only 1/3x (very small magnitude) and 2/3x are present, no any higher orders. On another hand there are plenty of half harmonics which may point to a bearing loose fit.

But thanks regarding rotating stall idea. This fan has an outlet damper and dust collector with filter bags on inlet. I guess stall is feasable in this case.

David

The numerous and random 5 G impacts and low correlation in the PV TW are very informative. It shows that the impacting is very random, and confirms a looseness condition.

Interesting observation regarding the PV autocorrelation. This looks like a 'random' signal - The low correlation relates to the signal be un-correlated with time delays of itself, in this sense random. Does this just go hand in hand with the advise not to average PV?

Note that the autocorrelation from the time signals does not have this random characteristic, which is more appealing to me physically. This does not show a random signal, but the PV signal has the random character.

Bill,
It is random by definition, even if it doesn't look like it is. The PV correlation basically correlates 1/2 the time to the other 1/2. Low values = high randomness or noise. I don't think one average has anything to do with it, we dont average time waveforms, we average spectrums and keep the last time block or special time block if needed. So the lack of repetition means somethimg; looseness and non-linearities or speed changes most likely.

quote:

Originally posted by Bill Kilbey Mobius Institute:
The numerous and random 5 G impacts and low correlation in the PV TW are very informative. It shows that the impacting is very random, and confirms a looseness condition.

Yes, it is informative in a sense that it is random, but does it lead specifically to a conclusion that bearing is loose on the shaft? Could be bunch of other reasons.

I added Autocorrelation plot (Fig. 3-1) for the regular TWF (why not?) and it also shows randommness.

David

Sys_fan4_fractionals.doc (158 Kb, 21 downloads)

When I had a CSI system and an acceleration TWF that was noisy, I would change units to velocity. Since integration is a smoothing process, in some cases, a pattern was easier to see. I would also at times use a higher Fmax and a lower number of lines to maybe get a better pattern in the accleration TWF when zooming in on the TWF. At times, I would also take a low Fmax, high no. of lines spectrum to drop the noise floor and and see the peaks better.

I'm not sure what Rusty means by saying that velocity TWF is derived. I would appreciate some additional explanation on that. Do you mean that it is calculated by integrating the acceleration signal? If so, that does not make it useless in my opinion, but it might not be as good as a raw velocity TWF. These days though, who has a true velocity transducer handy?

I'm not implying that you should use a velocity waveform in place of an acceleration TWF, but I would not discount the usefulness of looking at the velocity waveform as a piece of the puzzle. I'm not trying to turn this thread into a acceleration vs. velocity vs. PV discussion either. I'm just saying that I found it useful to me to look at velocity TWF's along with all the other data representations to attempt to better understand the issue.

I also agree that you have some looseness somewhere. I would take is seriously, but it seems to be trending pretty flat since July.

Michael Titone

quote:

Yes, it is informative in a sense that it is random, but does it lead specifically to a conclusion that bearing is loose on the shaft? Could be bunch of other reasons.

Read my second post: "looseness and non-linearities or speed changes most likely." I assume you have done a good visual inspection with eyes, fingers, ears and strobe? Rules out a few things. Flow noise? Belt rate? Have you calculated all forcing functions? High res. below 100 hz just for giggles?

Any plans to work on it? Always good to fix things.

This message has been edited. Last edited by: Bill Kilbey,

quote:

that does not make it useless in my opinion

Michael, I get into enough trouble on my own... please don't help me out.

What I said was:

quote:

IMO, the velocity waveform is only good for low order mechanical problems... looseness, imbalance, misalignment.

The velocity TWF is certainly not "useless".... I just don't think that for higher order problems (bearing problems, gearmesh, lubrication) it's as usefull as an acceleration TWF. (And I am primarily speaking of rolling-element-bearings). If I can only have one TWF, I want acceleration. That's an opinion based mostly on experience... I've not done a detailed study of velocity vs acceleration TWF's.

quote:

It is random by definition, even if it doesn't look like it is.

I don't know to what the "it" refers, and in general this statement doesn't make much sense to me. Random by definition? How do you define random?

Random often denotes that the signal has its amplitudes distributed by some probability distribution (sometimes Gaussian). Also, time translations of the signal (like the autocorrelation does) have no correlation with the signal - the amplitude at a future time is not predicted from the present value.

This latter property produces an autocorrelation with a peak at zero (time delay) and zero elsewhere. The PV signal has this characteristic, but the time waveform doesn't.

This makes me wonder about the value of PV. Is it just a random signal?

quote:

This makes me wonder about the value of PV. Is it just a random signal?

I have used PeakVue and the autocorrelation function in the 2120/30 and Machinery Health Manager software many, many times on real signals from real equipment. Faults that generate random amplitude and frequency (boundary lubrication, cavitation, looseness) usually have very low to zero correlation. Defects that produce repetitive signals (brg defects, rubs, RBPF, GMF etc. ) show high correlation. It is a very good tool for separating random/non-random events.

I might suggest you have a conversation with Jim Robinson at Emerson for your answers, or read the attached paper. You may also want to read and pay attention to this discussion board. Many people will tell you the "value" of PeakVue in their real life case histories and experiences. If you actually use it, you may be able to make a more sound judgement.

Autocorrelation_as_a_dignostic_tool.pdf (1,970 Kb, 27 downloads)