Some of you tuned out of the other thread on crawdad piles etc before we got to the practical stuff. For this thread, no physics or math problems allowed.

How do you judge when it is time to call a bearing?
How much change. What levels. Etc.

For my own purposes I am primarily interested in knowing at what point it is safe to make a call and be assured there will be visible damage on the bearing when removed, assuming it is a machine that can be taken out of service for maintenance without much trouble.

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

Calling based on level changes generally doesn't work well. The easiest flaws to detect have the least effect on bearing life. Flaws on multiple surfaces are much more significant than flaws on single surfaces in that multiple surface flaws allow momentary lock-up of the bearing and greatly increase the forces operating on the bearing. For a simple graphical example, look at Figure 14 in:

http://vibrotek.com/articles/sv95/part2/index.htm

quote:

For this thread, no physics or math problems allowed.

I'm hoping for plenty of responses to this one. This is the type of information that I can use now while I am still just cutting teeth.

I agree, if I find 2 or 3 faults eg. inner and, or outer and balls, FFT and, or Demod looks like a garden fence and all other bearing HFD values are sky high and it sounds like stone crusher using anything from the "yellow handel analyzer" to the stethoscope you can be sure to have damage you can see. If you then have to or dare to gently inject some grease and all values increase 200% I usually say the final is (very) near. If it goes down and comes back within hours, you still have severe problems but bearings can be kept operating surprisingly long times with small and often regreasing. Olov

I would not use a specific number as a treshhold dividing bearing with a visible defect from the one where defect is not as visible.

It is common knowledge that if there is a symptom then the cause is there. The question is old as human race: how severe?

If historical data is available, then a climbing trend (ampitude, # of harmonics, sidebands, etc.) of 3 consequitive readings will warrant bearing removal.

Trend.doc (62 Kb, 67 downloads)

quote:

Originally posted by vibbase:
It is safe to assume there is visible damage when it is audible. If you are unable to listen to the bearing directly, you will have to depend on the amplitude of the defect peaks in your acceleration data whether HFD'd or not. This is from my own experience and I have seen some very, very, very small defects show up as 10 g's in the spectrum. If the defect is in the load zone the amplitudes will be higher.

Here in the paper mill, on critical equipments like a winder roll for example, the call for a bearing change is made even when nobody can hear a suspect sound on the bearing housing. If the peak acc. overall level has increased by a minimum of 10Db. and one or many bearing defect frequencies appears in the demodulation spectrum don’t matter the amplitude level, the call is made to replace it for the next shutdown when possible. With hundreds of machine to be monitored, i don't have the time to measure the same equipment on a daily basis to follow the evolution of the defects. Everybody knows (at least maintenance people) that a small scratch on a bearing race means the beginning of the end of the bearing's life anyway.

Thanks for all the great comments. Keep them coming.

In the other thread I made a comment that some people had claimed success using peak twf acceleration, even without trending. I think there are people on reliability magazine board that used that approach but I can't find it on Oli's search (thanks anyway Oli). I did find this excerpt in an article presented at the Enteract 2000 conference: "There’s Still Value in Overall Vibration Measurements" By John C. Johnson of Balance Plus, Wichita, Kansas (Enteract articles available for free btw)

"I started using true-peak acceleration measurements after attending an advanced vibration analysis class in 1993 taught by Nelson Baxter. Talking to Nelson during a break, I learned one of the most valuable methods of detecting defective bearings that I had ever learned before or since. He told me to look at the acceleration time waveform and identify the highest amplitude peak, either negative or positive (this is truepeak acceleration). Amplitudes over 7 g’s for ball bearings and 12 g’s for roller bearings are a strong indicator the bearing is defective. I applied this over the next few years and found it to be very accurate on
machines that do not produce a lot of high frequency energy during operation......Earlier in this paper, alarms for true-peak acceleration amplitudes for rolling element bearings were stated as 7 g’s for ball bearings and 12 g’s for roller bearings. These numbers work very well on most machines if the signal path to the accelerometer is relatively short and direct and in the load zone."

Note he talks about alarm levels but not trending. I don't know if this was an inadvertant omission but it looks like he works straight off levels.

CSI's famous 100 page article talks about Peakvue TWF peak/peak alarm levels. [Yes, Peakvue is different and better than acceleratio TWF peak/peak but I lump them together] They give a more elaborate method to determine alert and alarm levels. Outer race fault has higher alert level than inner race since it is easier transmitted. Low speed machinery has lower alarm levels. They also make it very clear that the trend is more important than absolute value. The same idea as David and Alec discussed above. I like the idea there can be a number to trend with automated alarms to help us track the condition over time.

It also makes sense to me that just like with normal overall, in addition to looking at the number we should look at all of the characteristics available. Particularly those that Martin pointed out in the other thread and Oli mentioned above. As the defect progresses we expect to go from one fault of one type to multiple faults of multiple types. The TWF gets increasing complexity including modulation. The spectrum gets more families of harmonics and sidebands. I wonder about the cause of those 1x sidebands but it's a subject for another thread.

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

I was aware of a tone of anger in my posts so I have deleted them. Just so you know, I quit smoking on May 12th after 30 years. Cold turkey, just a little tense at times. I'm sure I'll mellow in time. Thanks and sorry about the harsh words, I did not intend to offend anyone.

No problem. We all get that way once in awhile. Keep those great comments coming.

Duncan - your email doesn't seem to work. Is it the right address?

quote:

Originally posted by electricpete:
"... look at the acceleration time waveform and identify the highest amplitude peak, either negative or positive (this is truepeak acceleration). Amplitudes over 7 g’s for ball bearings and 12 g’s for roller bearings are a strong indicator the bearing is defective".

The problem with this approach, as I see it, is that measured "true" peak acceleration will greately depend on selected sampling rate and therefore is not truely the peak value. The actual true peak is provided by PeakVue which samples vibration at a fixed rate (40kHz)and stores the peak value for each predetermined period of time. Being that high, this frequency also encompasses the frequency range of stress waves emission.

Using peak time waveform levels is a technique that has been around for quite a while, at least since the middle 1960s. In the analog hardware days, this was mostly done using oscilloscopes, either as add-on devices for instruments such as the IRD 330/340 that also had optional accelerometer accessories or in instruments such as the Metrix 5115 that used accelerometers as the standard transducer and that also had an oscilloscope as the display for the operator. By the middle 1970s, there were also more specialized instruments that could measure peak levels and also calculate parameters such as crest factor. Generally, if the crest factor was above 3.5 to 4, there would be a visible defect that might or might not trend to a worse defect. If your only criteria for a successful program is finding a visible flaw, this is adequate. Many users of such equipment in the 1970s reached the conclusion that it wasn't so good as flaws could be detected long before the bearing needed to be replaced. The other problem with this method is that problems that can cause severe damage can be missed but if your criteria is that only finding visible flaws to avoid being second guessed, peak detection and crest factor works.

Broadband envelope detection methods like PeakVue can be considered a next step up from viewing acceleration peaks in the time domain. Applying a high pass filter can and usually does improve the ability to see peaks from flaws earlier. There are two main reasons for using envelope detection compared with simple time domian signals. One is the compression of the time domain signal that allows significatly longer blocks of time domain data to be stored. The second and more useful feature is that envelope detection allows the calculation of envelope spectra which usually is much more useful for diagnostics. The default PeakVue bandwidths give users familiar with time domain analysis a lot of comfort in that levels tend to be close to what you see in normal time domain data; unfortunately this takes away much of the ability to judge severity in terms of life but that's another story.

pete - I just successfuly sent myself a message. Maybe my spam filter got it. Try again.

I seem to remember from the old board that the US Navy used a system where they viewed either acceleration or demodulated spectra on a dB scale and if the peaks were X dB above the noise floor they called the bearing. Does anyone remember the specifics?

Vibbase--I feel your pain. After a pack and a half for almost 35 years I am cigarette free since last Oct. so it can be done. Have actually cracked two teeth from aggressive gum chewing, gained 15 lbs and lost all my friends but they say it is worth it. On the positive side, as a reward I have learned to enjoy a good cigar at 5:00pm every Saturday afternoon.

dj

dj - the Russian Navy developed a method for bearing diagnostics in the middle 1980s that evaluated the severity of envelope spectrum peaks relative to the noise floor in a controlled bandwidth ( necessary because the bandwidth and resolution of the spectrum affect the size of the spectrum peaks). This is the basis for my company's bearing diagnostics and there's a lot of information on this in the articles on our web site at vibrotek.com.

VENDOR WARNING

Duncan

I don't rely much on trending for analyses unless it is something approaching real. I'm looking at CSI trending data and Pos AH going skyward at a terrific rate; Pos AV going down in severity and similar throughout.

Where is failure? That's what we want to know and if your trend is only history and your trend is up and down and down and up and ??? then it's useless but if it is mathematically correct and can tell you where you will be 3 months from now then go with it.

I have a machine that always consistantly fails along the same path/curve. And from 7 g's to 14 g's is predictable and basically always the same and you know basically how long it will run and you have a few weeks to plan.

I'm getting interupted every 15 secs - bye for now. D D D that's right!

I read the paper Duncan posted on the other thread, along with other papers on the website.

There was a phrase in the paper (backed up by some statistics) about Long MTBF for single defect bearings, but severely shortened MTBF once multiple defects are detected.

For beginners like myself, I feel a little better about several single defect systems I currently have running.

But I would like to hear opinions from this group about this concept... MTBF for single defect VS multiple defect bearings.

The concept seems to make sense to me; and it validates the general concept of "orderly" waveforms vs "chaotic" waveforms.

As any weatherman can attest... Weather is difficult to predict because it is choatic.

What is your monitoring frequency: quarterly?
Then you have to know if your machine will run throught that monitoring period or not - if not, then when will it fail?

Often I get a bugger; quarterly and defects that lend to the thinking that failure will likely occur in three months - great! Before or after next monitoring interval. This is a very tough shot to call unless you are planning on reducing your monitoring frequency to monthly. In-plant program, you can opt to monitor monthly through failure. Outside consultant - call the shot. On a number of occasions I have called the shot as, "I believe it will safely make it thought next intrval and we will plan a schedule from there?" Sometimes at the next point, it is at failure and you are getting close to having egg on your face.

Most machines fall into quarterly monitoring IMHO. Unless cirtical (super critical and circumstances??) and the manager calls or dictates a frequency. 1200 RPM or less is basically always quarterly. 1800; could be monthly, bi-monthly or quarterly; 3600 generally always monthly. If criticallity is more than this - full time monitoring is probably the way to go.

I'm not a bearing expert but I know there are many types of them. How about use FMEA to identify all potential failures for bearings and evaluate their likehood and consequence based on past occurences? Or is any FMEA available from bearing manufacturers? From a maintenance strategy point of view, I think we all agree to replace a bearing once the consequence is no longer acceptable to those bearing experts. However I appreciate their contributions better if I can see all the details on paper ie based on proper/comprehensive analysis (qualitative rather than quantitative). TQ

Great comments all around.

Sam - you're right, monitoring frequency may certainly play into it, in terms of projecting how much worse it might get between now and next monitoring period. We monitor from quarterly to yearly depending on equipment. If there were a problem on a critical piece of equipment we put it on increased frequency monitoring. But we wouldn't put it on increased monitoring unless we have also called the bearing and written a work order for correction. That's just the way things work for us - nothing can go on increased monitoring unless we have a plan for getting it off (generally correct the condition within one or two regularly scheduled availability periods).

Josh - I see you are focused heavily on the true equipment reliability impact associated with bearing defects as rightly we should be. I don't have any answers. I have not seen many bearings take down the machine and most of the ones I read about were cage failures. Maybe someone else has more comments.

All
Again in my view there are two constraints for calling a bearing:
#1 - "Repair after visible defect" - as Duncan said you don't want to be embarassed in front of the mechanics by calling a bearing and then they can't see anything wrong with it. So to avoid embarassment we should only call a bearing when we are sure they will be able to visibly see something wrong with it. (David - my cutoff for "visible" is something you don't need to squint to see)

2 - "Repair before failure" - You usually have to call it so it early enough so it can be repaired before failure (with the exception of run-to-failure equipment).

I think most of the focus is usually on #2 (repair before failure). That may in some circumstances be the best call from an overall plant economic perspective (just-in-time repair?). But at least in my environment, it is usually considered a success as long as you meet item #1 (repair after visible). As some people mentioned in this thread, it can only get worse. Why spend all the energy trending it for years instead of just fixing it. (don't answer why... it was a rhetorical question... really depends on your plant environment).

How do we judge #1 and #2
For #2 (repair before failure) - I think that the appearance of multiple defects, and sidebands and increased complexitiy (from the multiple families of patterns) in TWF and spectrum, is most useful. Also the nasty noise that Ron mentioned.

For #1 (repair after visible defect) - I get the feeling from Duncan's comments (and I have heard it somewhere else before but I'm not sure where) that peak acceleration (and possibly crest factor) is most useful for judging when we can expect to see visible damage.

The existence of a worsening trend in either the level or the pattern clearly play into #2 (repair before failure). I'm not sure to what extent it plays into #1.

Interested in comments on the above and I still have some questions geared toward making better calls per criterion #1 (call after visible defect):

Questions

A- As far as crest factor goes, it seems like you are ratioing to another number in the denominator (rms acceleration) that brings into play a bunch of additional factors.... like ringdown characteristics after impact, and possibly other non-defect vibration. I think maybe the justification for crest factor is that any attentuation of the signal affects both the peak and the rms so we have a ratio independent of attenuation.... but we know the high frequency components are attenuated much higher anyway so I don't think that really buys us anything. Therefore to me it sounds like peak acceleration is a better number than crest factor. What do you guys think? (especially Duncan)

B - To what extend do you'all agree peak TWF is a good parameter for judging #1 (repair after visible)

C - Do you think TREND of peak acceleration is an important component of judging a bearing by criterion 1 (call after visible defect) or can we really call based on magnitude alone (possibly considering lower alarm for inner race vs outer race as per CSI article).

D - Any suggestions on what would be a high enough peak acceleration level (assumed you have a high enough samplingto get pretty darned close to the true peak and long enough samplte to get a few representative impacts), to come close to guaranteeing a visible defect. Ron's comments suggest maybe that number needs to be above 10g's at least for outer race defect. I'm thinking the alarm numbers from CSI might be a good number - 12 g's pk/pk for outer race fault. And if your TWF peak isn't as true as Peakvue, then you are even more "conservative" from the standpoint of being safely past the visible stage.

All my comments in this thread assume that the pattern you are looking at is a bearing defect pattern. Random pattern or impacting at 1x etc is different subject.

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