Martin
Those are great practical points to keep the discussion grounded.

Other ways you mention to judge severity - trending, and apperance of increasing number of distinct faults which at some point become chaotic, and increased modulation in TWF / sidebanding in spectrum. I agree.

Regarding trending - we are just beginning to take acceleration TWF pk/pk values on some of our machines as part of the routine. Never have taken before except during special investigatory measurements,so limited opportunities for trending at this time on most of our machines. Even if you trend you still would like an idea what is good and what is bad. It gives us the same comfort levels when looking at accleration true peak as we have looking at overall velocity levels and comparign to ISO standards. (I know that I don't care about a jump from 0.02 ips to 0.04 ips and I care about a jump from 0.2ips to 0.4ips but I don't have the same feel for acceleration TWF peak.)

Although TWF peak or pk/pk acceleration (including peakvue TWF etc) is not discussed in ISO standards, I believe it is pretty widely used to judge rolling element bearing defect severity and I have heard a lot of people claim great success even without trending... of course only applies when the pattern is recognized as rolling element defect. Anyone want to chime in on your success with this method?

Can the rolling bearing math problem be solved exactly? No way. Too many variables. And we haven't even talked about attenuation of the signal. But aren't you curious why that chart from CSI looks the way it does?

Pete:
I don't know if the CSI chart is purely mathematically derived, or whether it was derived from actual readings and failures.

I know I have fan inboard pillow block bearings on belt driven air handlers, for example, that run far above these levels for long periods of time. I've come to understand that there is something about the inboard bearing that elevates pkvue above normal, (belt tension effecting the load zone?) Lubrication greatly effects Pkvue also. I see clear drops at shutdowns, then back up in 2-3 months.

I use peakvue only to direct me to look closer at regular spectra. Generally speaking, I begin by using the scale in the chart, but adjust to known conditions. It really comes down to the analysis of each machine on its own.

Is there anyone here willing to make calls on Pkvue data in the absence of indications in regular spectra?

Pete:
In regards to using new Acceleration data with no trending:

If you haven't noticed something in Velocity Spectra(bearing defects) , and you record High amplitude acceleration in new TWF, I would switch to Acceleration in the spectra to see from where the signal is coming. Mark two of the twf peaks and get the freq.

Chances are you have something like pump cavitation, or some other less critical high frq phenom.

I've got liquid ring pumps that hit 70g in TWF, but it is all flow and frition on the ring. And it depends on how much moisture is coming through the pump at the time of the reading.

Normally one would say that 60 pkpk indicates failure approaching, but not in this case.

If I matched it to a bearing defect, that would be another story.

Martin - I think I agree with what you're suggesting.

A few things to clarify my opinion:
- When I talked about Peakvue I was talking about the Peakvue TWF true peak, not Peakvue spectrum or Peakvue overall. I don't use CSI equipment, so my terminology may be incorrect.
- As I said in my post directly above, "TWF peak or pk/pk acceleration (including peakvue TWF etc)... of course only applies when the pattern is recognized as rolling element defect"
- I don't recommend anyone to make a call without checking the TWF and the spectra are consistent with the call (If it sounded different, then I apologize). Certainly I hope anyone uses all info available and the ones you highlighted are among the most important (trending and change in pattern indicating increased defects).

Ok:
For a confirmed bearing defect signal, and an "equal" defect (in terms of structure) on two "equal" machines, 1775 rpm vs 3550 rpm, obviously the 3550 rpm machine will degrade quicker. ie Speed kills as was said before.

The acceleration amplitude will also be higher all else being equal. But if all else is truly equal (same bearing, same load) the ability of the bearing to endure the impacts is comprimised faster.

In terms of absolute alarms: if both use the same alarm level, obviously the 3550 will trigger the alarm first. But you may need that extra time because your time to failure is shorter!

So maybe setting higher alarms for higher RPM is not such a good idea??!!

What do you think?

Woops.. One other thing

For this example the "load" complicates the matter. Load relative to full load of the machine or an absolute static load for both?

If we choose a static load, the low relative load for the 3550 machine may equalize the expected acceleration impacts.

But load can have different effects.. increasing or decreasing impacting. I don't think using a higher load to anticipate higher acceleration is a good guideline.

quote:

My rump-meter tells me the acceleration and force is roughly proportional to speed^2. (maybe someone can explain that with a math model?) It's the same

.
Theorem of work and energy tell :
Work applyed on a particle is equal to change in cinetic energy (K)
W=F*d=(1/2*m*V*V)- (1/2*m*Vo*Vo)
W= K - Ko
.
where:
W:force , F:force , V:final speed
m: mass , Vo:initial speed
.
Work is proportional to force and is proportional to V*V

.
Regards

Martin wrote :

quote:

If we identify a series of impacts in twf that match a bearing frequency, and that series of impacts modulates in amplitude, the impacts represent time when the bearing defect is in and out of phase with other components.

By extension, a series of identified bearing impacts which does not modulate (much) in amplitude indicates that a greater percentage of a given impact is in fact the bearing defect component of the impact.

Low amplitude in the envelope is when several impacts are "almost" synchronized.

Just an observation (and I admit I haven't read everything posted in this thread)... based on my experience, I would surmise (gut feeling) that -- at most -- 20% of the bad bearings I detect show distinct bearing defect frequencies. So I don't spend a lot of time looking for distinct bearing defects. I would also surmise that > 95% of the bad bearings I detect show rapidly increasing levels of peak acceleration in the TWF. So that is what I look for (mostly). If a defect frequency presents itself, then I analyze it, of course.

The detection and analysis of bearing defect frequencies is way over-sold, in my opinion.

Martin - I'm glad you joined the discussion. Those are some good questions and comments.

I agree with your logic on the conclusion that the crawdad pile effect has two different implications for alarm limits that tend to cancel each other. That is in fact the exact position that I took during the original post of this thread. But then after I got finished arguing that force was proportional to speed squared, Rusty went and agreed with me and that really scared me ;-) so I had to quickly change my mind and argue the other side. (It wasn't really Rusty - just wanted to play devil's advocate). The other side is that F ~rpm^2 is in some ways is inconsistent with bearing life theory as well as peak acceleration TWF limits. And when I discussed the peak acceleration TWF limit you'll notice I used the terms "if we assume the goal is to call a bearing when it has visible defect" (not if the goal is to avoid failure). With that assumption you would tend to change your alarm with speed. Maybe that seems a little sneaky to change the assumption, but after all the approach for calling a bearing should satisfy both calling the bearing AFTER appearance of visible defect and calling the bearing BEFORE failure. And the goal of my discussion was to see what other answers people come up with. And along the way there has been some great comments.

So, I agree, the possible answers to the "apparent conflict" on TWF limit being constant with speed:
1 - The underlying basis for the limit is to avoid failure rather than to identify when visible damage begins.
2 - The limit is empirical based on experience.
3 - less likely - bearing impact force doesn't increase with speed (for a fixed load and size of bearing).

If #1 is the answer and the assumption that impact force changes with speed is true, it is still an interesting result that we would expect to see less damage on a fast speed machine that we call at X g's. than on a slow speed machine.

Since no one has jumped in on the other "apparent conflict" between the crawdad effect (Fimpact~RPM^2) and the classical life prediction of the bearings, I'll tell you my view. The standard says that the ratings are based on "fatigue life" and I think I remember hearing somewhere that this is intended to be the time to first appearance of a spall. And of course the crawdad pile effect doesn't come into effect until the first defect shows up or until contamination is introduced. In the real world where we have a useful life beyond first spall and we have contamination, it seems like the crawdad effect would tend to make the standard overestimate the life of fast-speed bearings as compared to low-speed bearings. But I hope no-one takes my statements too seriously in this thread which I identified as philosophical. It is not intended to be a proclamation of a new principle or recommendation on how to monitor machines. Just trying to make sense out of things.

Rusty

I didn't understand what you said about often not seeing defect frequencies. Are you saying:
1 - you often don't bother to try to match defect frequencies up with the specific bearing (neither do we)
or
2 - you don't see any non-syncrounous stuff at all in the spectrum
or
3 - that you see a lot of non-syncrounous stuff and raised noise floor but it is not worth the time to figure out?

We always see non-sync stuff when we call a bearing but then again we historically focused very heavily on the velocity/acceleraiton spectra and not much at all on TWF or demod spectra, even for bearing calls. In the future we plan to pay more attention to TWF.

By the way, I am shocked that anyone hasn't read this whole thread ;-)

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

Rusty:
As I have said before, I am still developing my sense of severity for bearing defects. What you wrote is the direction I seem to be going:

" I would also surmise that > 95% of the bad bearings I detect show rapidly increasing levels of peak acceleration in the TWF."

But I was suprized about this:

"based on my experience, I would surmise (gut feeling) that -- at most -- 20% of the bad bearings I detect show distinct bearing defect frequencies. So I don't spend a lot of time looking for distinct bearing defects."

Do you mean in the spectra? This seems right to me because I've seen defect frequencies in the spectra remain stable for long periods.

For example, I have many machines where I have no idea which bearings are in the system. They also happen to be belt driven, which can sometimes put a lot of peaks into the lower orders.

So if I have a stable Asynch peak between 2-3 orders on a belt driven system, and the TWF is only 2gs pkpk I don't get excited.

But if my TWF is 15g pk-pk, I don't expect those kind of impacts to come from a belt.

In that sense the TWF can be a deciding factor.

But we had been discussing a "Confirmed Bearing defect" in the spectra, and at what pk-pk point do we call it for 1800 vs 3600 rpm machines.

Rusty are you saying you primarily look for Asynchronous impacting in TWF?

When you have a confirmed bearing defect in spectra (say BSF), but the TWF has low pk-pk values do you let it ride until the TWF values begin elevating?

And if that is the case, do you set up your alarms primarily on pk-pk twf, then dig deeper when the alarm trips

This is the thing that gives me fits.

If you consider a defect out of the load zone, it may not degrade that quickly.. in fact it could go for years.

The reason I am so interested in this thread is because our program is growing.. up to now I have been able to review each an every spectra and TWF. But soon I will really need to rely on alarms to trigger further investigation, and have the confidence not to look into machines which are within limits.

Pete,

I think the reason Dr. Robinson's first paper said pk-pk waveform was due to RBMWare showing a negative value in the waveform. The waveform is rectified so it is really a 0-peak value. The newer versions do not show a nagative value in the waveform.
Crawdad piles can be hard on lawnmowers and lakes. They tunneled through my dam and I kept losing water. I would dig down through their burrows and refill them but they would just tunnel back through in a few days. I read somewhere to pour a quarter cup of ammonia down each hole and the problem would end, it worked. You might try it in your ditch and yard.

Interesting discussion.
If you were mowing over the crawdad piles, AND your lawn had a roof 1 inch above the top of your head, your would notice a different response too!

Stan - It's amazing how bumpy those piles of dirt can be. Good reason to skip mowing the lawn every other weekend... don't want fatigue failure on my lawnmower. Wife doesn't seem to buy it though.

Jim - now I am really curious. How does CSI recommend you compare your peak/0 value to pk/pk alarm limit? As discussed above there can be up to a factor of 2 difference.

Pete,
Dr. Robinson's paper you refered to was only on PeakVue data. The PeakVue waveform is always a 0-peak value. They have a pk-pk and a max-peak waveform trend parameter, I think either will give you the same value on a PeakVue waveform. When using pk-pk the bottom value will just be 0. The regular acceleration waveform has a pk-pk value.

Thanks Jim. That is still a little confusing to me (so what else is new
;-) ). But since I don't have Peakvue, I'm going to ignore the idiosyncracies of CSI terminology and focus on its applicability to me:

I have an acceleration TWF. I plan to compare the pk/pk value of my time waveform directly to the CSI chart labeled pk/pk. Does that sound correct? (Assuming my sample rate is high enough and sample duration is long enough)

quote:

Originally posted by electricpete:
Another philosphical question/comment.

A crawdad is also called a crawfish... it looks a little bit like a lobster. People in Louisiana might even tell you it tastes like a lobster. Myself, I don't eat stuff that grows in my ditches ;-) (Just kidding you Louisiana guys)

Hey EPETE. I bet those crawfish are doing tricks with all this rain. By the way, lobsters don't even come close to tasting like a big pot of boiled spicy crawfish. Down here we don't eat them out of the ditches, they are grown on "Crawfish Farms", no joke, it's a big business. What we do down here with the piles we don't want in our yard......drop 1 moth ball down the hole........they are history. The dirt really dulls your mower blade fast too, huh?

Hey Don!

I've heard from friends that crawfish are pretty good. I just haven't got the courage myself.

I generally try to ride over everyting in my path when mowing the lawn (crawdad piles, fire-ant-piles, spanish moss, fallen branches, small critter carcasses etc), so I have not gotten very good life on my blades. And being stupid enough to buy a lawnmower from S*ars, I pay twice as much for a blade that lasts half as long. Don't get me started...

Plenty of rain for my day off today. Let's hope Emily doesn't take a northward turn or else we won't be laughing anymore here along the Texas coast!