Gentlemen,

The attached data is from a SKF 1315 bearing that I have been monitoring since it was new in 2000 and it has been in 24/7 operation since then. After the installation, the bearing exhibited a frequency at 6.73X which matches the BPFO data. Over the last few years, the BPFO amplitude has ranged between 0.015 in/sec to 0.030 in/sec. But last data set indicates an amplitude of 0.035 in/sec.

The bearing is in a SKF SAF 615 pillowbock and is used in a fan application. The data presented was recorded vertically (on the top) with a magnet-mount accel (everytime).

The curious thing is that this frequency in not present on horizontal or axial directions on this bearing. Nor is it present on the other bearing nor on the motor ... nor on the bearing support base. Which brings me to the question: "When is a bearing really bad?"

Do you have envelope spectra?

Making a call on a velcity spectrum out to 24,000cpm is 'difficult'

Its better to look at acceleration data, preferably spectra out to at least 5kHz - that's where you see the damage developing and growing. Generally you see the spectral content of the damage getting lower in frequency as the damage grows.

In general I have found that overalls of less than 6G (true peak) are nothing to really worry about. Getting up to 15G (true peak) is a real concern.

You should examine time waveforms as well to see the impacting effects of damage.

However, trends are always the best indicator of whether or not more action is justified.

By the time you pick up damage in the velocity spectrum the bearing hasn't usually got a lot of life left in it.

Based on your data, showing BFPO which does not appear to be deteriorating, I suspect that you might be seeing 'false brinelling' (or standstill damage) of the outer race. If so, then you have a bearing which is dmaged but not failing - we have plenty of these which we have added to a 2 weekly route instead of the usual monthly - they can run for many years like this.

In short, you should be looking at the higher frequency spectra (or some sort of demod measure) and seeing an ongoing worsening trend before making a call.

Additional information - no matter what the diagnosis, the customer will NOT change bearing until October (unless it fails).

My intent on this post is mostly to get some debate on criteria that others are using for diagnosis. Primarily I am curious at what point you guys would start getting the "change it now jitters". After all, changing too early means lost bearing life and changing too late means additional components may be damaged. It is a fine line, so where do most stand?

Duncan,
This machine is readily accessible and I can take any additional data with any parameters that you feel might be relavant. What parameters would you suggest?

Buzz,
FMax of 24,000 CPM is 20X rotor speed and this is not enough? Considering that a magnet-mount accel has a resonance of 2kHz, wouldn't a 5kHz signal include data above the "maximum useable frequency range" that sensor manufacturers specify?

Does this 15G(pk) fault level come from a standard or from experience? The reason I ask is that my primary reference is the GM Vibration Specification, which is one of the few freely available Specs that addresses something other than 1X amplitudes. I've had machines with 30G that are "perfectly normal" and ones with 1G that scare the heck out of me.

The trending for this machine is roughly in six month increments, the last overall data actually decreased 2% from the previous. So far the trend is no climbing.

Regarding the false brinelling, you nailed the original cause. This bearing was sitting on the shelf for four years prior to installation. After the installation, I immediately inspected six other bearings on the same shelf. All six were found to have false brinnelling on the races. I wanted to put them in the circular file, but the customer decided to keep them "just in case". (I think he meant "just in case I want to annoy the vibration guy some more" )

In short, you want to see a 0-5kHz FFT and time data in acceleration (G). You also want some demod data, what parameters would you suggest?. Anything else? Anyone? I am planning on being there this afternoon, so posting now will get you data today.

IMHO 20X rotor speed is not enough. I use 0-30X in velocity (@ 1600 linesto check for unbalance msalignment and other relatively low frequency faults)and 0-5kHz in Acceleration (@ 3200 lines specifically for reb faults).

I am not really interested in the maximum usable range of the acclerometer (although I need to be aware of it) since I am not making absoute reference measurements that will stand up in court - I am looking for changes to alert me to a developing problem NOT checking accel specs (although I need to be aware of them).
How do demod measurments get made with magnetically mounted transducers?
Purists will argue that a magnetically mounted accel might well have a resonance of 2kHZ but, using a strong pull two leg magnet I do get repeatability (which is the important thing) and if there is nothing to excite the resonance then nothing is in the spectrum.

Usually (with the exception of cage faults) a REB fault starts with low amplitude high frequency vibration components and then moves down the spectrum.

The limits I suggested are based on my experience - a REB with less than 6g (true peak NOT pseudo)is generally OK - a newly installed, properly fitted REB will exhibit less than 1G peak - with alarms being set accordingly. Gears, gear pumps and small lobed lube oil pumps are inherently noisy and can go wel above 15G, somtimes as high as 50G with hi-amplitude vibration in normal operation - here I use frequency band alarms.

Standards are far too general - OK if you have nothing better (e.g. just starting a program) but nothing beats looking at the trend of each parameter, especially if the alarms are correctly selected to cut down on the drudgery - once the trends start to go up its time to take a closer look.

Also, a lot can happen in 6 months - our max monitoring interval is 3 months - we vary things depending on the level of concern - typically monthly monitoring picks up most developing faults at an early stage, but we will go daily if neccessary, until we get a good understanding of what is happening. Its easier to take a measurement to make sure than issue a work order and be wrong.

The accel spectrum and waveform would be useful but I still wouldn't make a call until I had seen a trend of each, say maybe a week apart.

I think Buzz is right on the money on all those points. I wouldn't dream of trying to diagnose a bearing severity by looking at the fundamental fault frequency and not the higher harmonics and TWF.

You need a higher Fmax. Mounted bearing resonant frequencies are somewhere in the range 1khz - 2khz (60kcpm-120kcpm) and that is usually where most of the spectrum action is if you have sharp impacting. I usually put on a log scale and try to see the patterns. If only one fault shows, outer race is least severe, inner race and ball moderate severity and cage definitely the worst although I haven't seen a cage fault.

Generally, the more complex the pattern, the worse. Adding 1x sidebands is a little worse. Multiple types of faults are more severe. Changing pattern toward higher magnitudes or lower frequencies might be an indication of increased severity (depends who you talk to).

You could in your data have multiple types of faults besides outer race and you aren't seeing them because you're not looking high enough.

If you don't want to spend a lot of time interpretting those complex families, demod simplifies it a litte. Shifts everything down to low frequencies so you can see at a glance what are the actors. Sometimes if I look a demod first it helps me go back to the regular spectrum and recognize the patterns.

TWF is another important parameter. Peak TWF acceleration is a good thing to look at if you're looking for a simple quantitative measure (although I'm sure others will point out it's not perfect... often goes down). CSI on their website has some guidelines for judging severity based on peak TWF (they are talking about peak in the peakvue TWF but it's pretty darned close to the same thing as peak in the acceleration TWF as long as your Fmax is high enough). I can find that link if you want. They have fairly detailed criteria on acceration TWF true peak/peak that vary based on the speed (lower speed has lower limit) and location of the fault (inner race has lower limit since it is attenutated more).

Hi Chris,

Since you are "on site" this day and have the VB commtest meter, why not try some

normal data with an Fmax of 1K Hz and 2K Hz and 3K Hz,and 3200 lines

And some Demodulation from say,with a band pass of 125 Hz to 1250 Hz with a 1K Hz Fmax and 1600 lines plot

and maybe a Demodulation with a band pass of 500 Hz to 5K Hz with 3200 lines and Fmax of 2K Hz.

Be interesting to see what shows.

Here is a link to the CSI vib papers.

http://www.compsys.com/enews/knewspro.nsf?OpenDatabase&...=1&Count=50&Expand=7

I can't find the exact one I was looking for. It was an article by Robinson and Berry on using Peakvue to judge bearing severity.

Attached is a part of that paper I excerpted (I have the paper on my harddisk...downloaded it for free back in the days when I could find it). I don't think there is any copyright problem with posting this here, but if there is, please let me know.

Again these limits apply to Peakvue TWF true peak/peak. If you don't have peakvue TWF, use acceleration TWF true peak/peak with the understanding that it might read a little low if your Fmax is not high enough.

CSI_PeakVue_TWF_LIMITS_Only.PDF (108 Kb, 60 downloads)

The other comments are good ones. You can't really judge using data only from a normal spectrum but what you see could well be the result of false brinelling. If that's the case and there's no significant damage on the other bearing surfaces, then there's probably plenty of life left in the bearing - bearings that only have false brinelling of the outer race typically last a long time. The bearing flaws that are the easiest to detect have the least effect on bearing life and outer race flaws are the easiest of all to detect.

To get a real sense of the severity, you need to make envelope spectrum measurements in a controlled bandwidth with enough averages so that you get a smooth noise floor in the envelope spectrum - at least 12 and more is better. As I think you know, my company, Vibrotek, has a pair of products called DREAM that have the option of expert system diagnostics. Commtest licenses our diagnostic modules as an add-on for Ascent Level 2 and refers to it as Ascent Level 2 plus Diagnostics. Furthermore, the Commtest data collectors that have envelope detection are actually quite similar to our DC-12M except that the Commtest instrument has a higher resolution A/D converter for better signal to noise ratio plus more memory.

I have a copy of Ascent Level 2 plus Diagnostics but I don't have a Commtest data collector and right now, I don't have any way to import your data unless Commtest authorizes it and you send me a copy of your database. I have created a diagnostic measurment using only the bearing info in my copy of Ascent. It recommends using an envelope spectum taken radially(I usually use horizontal if possible) with a 500 Hz maximum frequency and an input bandwidth to the demodulator of 4 kHz to 5 kHz; lower frequency input bands that include the bearing resonances are apt to give data that is difficult to make sense of. If you post the resultant spectrum or send it to me, I'll look at it and try to apply DREAM's rules; some are rather easy but some are more complex and really could use the software diagnostics. You can also do this to an extent yourself by applying the technical articles on vibrotek.com.

As I'm sure you are aware, I am a VENDOR, both for our own systems and as a supplier to Commtest.

Sometimes soon in these edits, the "this post has been editied by;" message will appear. I'm setting a record for typos.

This message has been edited. Last edited by: Duncan Carter,

Duncan,

Apologies for an over developed sense of cynicism but, what does your system bring to a situation such as this?

I once asked a question of some avid PeakVue users and whilst they aggressively defended their use of this tool, none actually answered the question.

Whislt I have enormous difficulties in predicting when fan blades are about to come off or detecting failures of bearings fitted with phenolic cages (which is why we banned the damned things and moved to bearing fitted with brass cages), I do find that using the following measurement schedule works very well for me.

Velocity Trends:
Overall unfiltered amplitude
Velocity Amplitude: filtered band pass from 20X to 2kHz
Velocity Spectrum 0 to 30X

Acceleration Trends:
Overall unfiltered amplitude
Acceleration amplitude: band pass from 20X to 5kHZ
Acceleration spectrum: 0 to 5kHz
Acceleration waveform: to about 8 shaft rotations

I find that this schedule identifies at least 90% of problems in rolling element bearings (especially that bandpassed accleration reading - with every thing else giving providng supporting evidence of an ongoing failure), giving plenty of time to allow more investigation and planning of how to deal with the situation.

The remaining 10% of problems are somewhat complicated by the presence of high frequency components such as rotor bar passing frequencies, gear mesh or cavitation - these are easily dealt with by setting frequency band alarms so that changes in gearmesh are covered, as is the appearance of components that are not part of normal operation e.g. bearing damage components.

So the same question to you: what would your system bring to my way of doing things?

Buzz,
Chris was open to do additional measurements and the rectification included in the demod procedure do in my opinion and experience add some features that is worth considering. I have used Duncan´s and his companions way to process the data for some time and the only problem so far is the severity. It does trigger for the most minor bearing problem at a very early stage that may take years to develope to the stage where replacement is strongly adviced. I can not in all situations pass that info to the end customer in a direct report as he in some cases may pull the bearing at next stop and come back chasing me with at shotgun as he see nothing with his bare eye.... So I may indicate and advice on lube procedure and wait for accel 2-20KHz get above 3g and or "our" L method above 600Hz in velocity is above 2-3 mm/s and or the demod looks like a garden fence and or the SOB sounds like a concrete mixer in my stethoscope (yellow handle analyzer also accepted)......... I have a case where only the last case was apparent due to what was found a bearing full of condensed water... All other (at least 3+ measured parameters for bearing) but sound indicated nothing so I don´t rely on a single indication when doing a calling from a single measurement. I will be back on a case of "spectral" healing tomorrow for you to look at and I hope comment. Olov

Buzz - I really haven't used the Commtest version of the diagnostic system very much I don't have a way to get data as I don't have a Commtest data collector) so I'll have to make comments based on our version. Unlike many other systems, ours has an enourmous amount of testing and refinement that has gone on for the last 20 years. What it does that PeakView and most other systems don't does is handle the measurement issues in a way that allows you to get numerically meaningful diagnostic data for the expert system to work with. PeakVue, on the other hand, gets bearing data in what, in my opinion based on 35 years of bearing test experience both as a consultant and as a test system designer, in the worst possible way.

I first built a hardware demodulator as a internal accessory to an IRD 350 in 1971, and except for being done with hardware, it was a direct analogy to what PeakVue does. I started making digital and mixed analog/digital measurement systems for my own use as a consultant in 1986. For digital demodulaltion, I did the same thing that CSI did with PeakVue. Like everyone else, I didn't have a good way of anwering the questions, "How bad is it? How long will it last?".

Beginning a year or so later than I started, my Russian colleagues and their colleagues and predecessors at the naval laboratories in St. Petersburg started a long and well funded investigation of the bearing diagnostic problem. Like everyone else in the 1970s, they hit the same problems (No, CSI did not invent envelope detection as they announced in their November, 1971 Sound and Vibration; patents expired.) In the early to middle 1980s, the Russians concluded that there was no way to get numerically significant data using a broadband envelope detection measurement and no way to get a meaningful measurement if the bearing resonances were included in the measurement.

One researcher, our scientific director, Dr. Natalia Barkova, realized that there was a way to solve this by applying the following concepts:

1. As stated earlier, do not use bearing resonant frequencies. Instead, use frequency bands away from the resonances where the frequency response of the bearing is more uniform and predictable.

2. There are two major sources of bearing signals, impacts sources that are normal to the contact surfaces and friction sources which are in the direction of the contact surfaces. Because these wources have different bounday condictions, their signals have different frequency responses. By selecting the input filter band to the demodulator, it's possible to control the relative sensitivities of the friction and impact levels, thus giving you the beginning of establishing a meaninful way to establish severity.

3. For demodulated signals, the relationship of severity to the height of a spectral line is a function of the input filter bandwidth to the demodulator and the resolution of the spectum of the demodulated data. This relationship is called the modulation index.

4. The signal attenuation within a bearing affects the level of signal measured outside the bearing. Assuming that a transducer is placed closest to the outer race which is the most common case, the path to the outer race has the least attenuation, the path to the rolling elements has greater attenuation, and the path the inner race has the greatest attenuation. These attenuation factors are a function of bearing surface contact speeds, frequency, and bearing dimesnions. To make numerically meaningful diagnostics, these factors must also be weighed.

Our software can do something that, as far as I know, is unique. It can calculate a meaningful and statically relevant estimated minimum safe operaing period for machine components including bearings. In practice, it can calculate minimum safe operating periods up to 200 days or 20% MTBF for no faults down to 1 day for impending failures, it's designed to do/approach this in at least a dozen steps in most cases. It cannot calculate a realistic time to failure; neither can anyone elses system. For its original shiboard purpose, it could answer the question of "Is there significant risk of failure during this cruise?" or "Should it last until the next turnaround? The not only considers each fault but also considers all the combinations of failure probabilities of each fault.

I've given you a very short overview of my answer to your question. Our web site has an extensive library of technical papers that go into much greater detail at:

http://vibrotek.com/articles.php

I especially recommend that you read the first and last articles in General/Overviews and all the articles in Bearings, Data Acquisiton and Bearings, Diagnostics and, especially the S&V articles. Once you've done this I'd be happy to discuss this further but it really would save a lot of typing if you would read and digest these articles first.

Duncan

All good stuff but, in the manner of a typical politician, you seem to be answering another question to the one I posed so, how about a simple answer to the question which was

"What would your system bring to my way of doing things?"

I am a great believer in the KISS principle and once the band passed acceleration starts to trend upwards then there is a huge likelihood that the bearing is failing. This is then reinforced by confirmation of fault frequencies in the spectra. Is this wrong or is there a better way of doing things?

How could I justify to my management that spending money on your products/technique would bring an improvement over what we already do?

I have absolutely no doubt that your method works, as does PeakVue, Spike Energy and all other demod techniques but I would really really like to know what I would achieve by adopting any of them.

Sorry - I have tried reading the papers and I get lost early on, they go far too deep - I don't need to know how to design an automobile from first principles in order to achieve my primary objective of getting from a to b by driving one.

My real objective in this case is to spot failing bearings at an early stage.

I am not trying to be overly antagonistic, anything that makes my job easier is worth a go - for once I would just like to see an explanation in simple terms, rather the the 'BS baffles brains' method that so many vendors seem to take when challenged.

Most modern systems can spot failing bearings at early stages. The question is, "How bad are they in terms of when they might or might not fail. Several good, simple systems to detect flawed bearings were introduced in the early 1970s. It didn't take much time before users realized that this really didn't help very much. Most bearings are rugged and can operate with flaws, especially like false brinelling flaws. If you change every flawed bearing that you find, you'll change a lot of bearings and most, long before they really need to be changed. Determining how bad in terms of useful life is a complex process. Our expert systems were originally designed for use by Russian sailors. That's about as far down the sophistication food chain as you can get.

Read OLi's post again.

Because we use information about bearing friction sources, we can also detect and diagnose installation problems which are often correctable without major damage.

Gentlemen,

Here is the data and some details about the data parameters:

click here for the files

velocity FFT with 24kCPM Fmax, 400 lines, 4 averages
400Hz_vel_FFT.jpg
side note: this parameter was specified by customer for his reports

acceleration time, 2048 samples for 5 revs (266ms) (VB default parameter)
5rev_acc_time.jpg

acceleration time, 8192 samples for 800ms
800ms_hi-res_acc_time.jpg

acceleration FFT, 1kHz Fmax, 3200 lines, 4 averages
1kHz_acc_FFT.jpg

acceleration FFT, 2kHz Fmax, 3200 lines, 4 averages
2kHz_acc_FFT.jpg
(side note the above data shows a very clear resonance peak near 2kHz)

acceleration FFT, 15kHz Fmax, 400 lines, 4 averages
15kHz_acc_FFT.jpg

2-10kHz demod, 20X Fmax, 800 lines, 4 averages (VB default parameter)
2-10kHz_demod_20X_FFT.jpg

2-10kHz demod, 20X Fmax, 800 lines, 4 averages (VB default parameter)
2-10kHz_demod_40X_FFT.jpg

500Hz-5kHz demod, 1kHz Fmax, 3200 lines, 4 averages
500-5kHz_demod_FFT.jpg

Ralph - perhaps your unit is newer than mine, as the Fmax you specify is not supported with those bandpass sets. The last data on the list was as close as I could get with my system.

Pete - Thank you for the paper, will read it when I have a chance.

Duncan - Perhaps when I have the money I will upgrade my system to include your module. Having the ability to drop in a parameter that would be able to tell me something like "2 months until self destruction" would be a valuable tool.

I respect you and your company, but due to an incident with another vendor (software company), I am NEVER going to release my databases to anyone (other than customers). It was a very painful situation.

Buzz - You made an excellent point about "nothing to excite the resonance then nothing in the spectrum". One of my mentors used to say, "bells don't ring themselves". I have see many times that someone is trying to reengineer a system with elaborate mass and/or stiffness changes, without attempting to address the forcing function.

Regarding your "over developed sense of cynicism", I think that all the really good analytical minds have this quality to some extent. I would agree that standard units of measure are most widely used. And I have yet to find a competent vibration guy that will call a fault based SOLELY on PeakVue, Spike Energy, or any of the other "miracle measurements" on the market. They will ALWAYS fall back to one or more of the measurement paramters you listed. However, I would like to keep an open mind, new developments come with new technology.

OLI - We walk a fine line being in the consultant seat. We want the customer to react on somethings, but not as much on others. This case would be one point, when I arrived at the customer today to take additional data, their maintenance guys swarmed me. After I explained that nothing was wrong, I just needed additional data they backed off and watched me from a distance. As I was heading for the door, I looked back to see one of them whipping out a grease gun. So I turned around and went back to take a second set of measurements (nothing changed or I would have posted them).

Your opinion of needing more than one source before advising a bearing replacement is very valid. But I don't have a "yellow handled analyzer", I bought the black handled one, perhaps should I upgrade?

To All - I am still looking for the answer to my original question ...

"When is a bearing really bad?"

What is the point, between the initial fault detection and the glowing meltdown, when one would absolutely have to say "CHANGE THIS BEARING"?

I have seen vibration charts that indicate multiple stages of bearings defects. But I have yet to see a proportional failure rate when a defect is advancing thru these stages. If a bearing goes from stage 3 to stage 4 in 60 days, how long should it take to hit stage 5? Months? Days? Hours? I realize that there are multiple factors that would affect this, but does anyone have any solid RTF data? Something that would be able to reference a time frame, perhaps?

Take Care,

This message has been edited. Last edited by: Chris Olsen,

Chris - check your links as theys seem incomplete.

Duncan - Thanks! I was on it, but you caught me before I got it fixed. The link opens a directory that has the files in it.

Take Care,

I figured that out. As a SWAG, I say outer race wear/brinelling with perhaps a bit of eccentricity. I'm guessing at the effective bandwidth of the measurement but I'd say it's not very bad at this point.

The resonance near 2 kHz is typical and to be avoided for the envelope measurements if possible.

Chris, mine "yellow handle" is black and contain silicon, still works for me.
Buzz, I don´t want to intimidate, only inform on what I found as being valid, basically using any demod function your hardware support, try it, there is no black magic just some more info to be found and to make your (or at least mine) head to twist some more once in awhile. You got the levels where I shout wolfe, not that there are cases where you find surviving bearings at 3 and on occasion 10 times those levels. On the other hand if the client spend money on doing a scan I would like to try my best to make sure a bearing is called at a "proper" level where it is used as long as it can and the odds for survival significantly longer are low. I hate if I call a bearing and it is not replaced and then fails as that cancel the work I do completely to being of no use at all and I hate it. So I try to keep some margin, sometimes slim. I realy try to make sure that any source problems that lurk behind the bearing failure is taken care of. I have a couple of cases as most of you I guess have where design, cost and whatever makes a good fix of a problem do not happen and it always irks me. So the major problem is not to call the bearing at a "proper or optimal level" but to make sure it wouldn´t happen again.... Olov