Hi guys, I found this write up by our former one year experience "Expert" For Analysis, most of the statements are fair, but... for Routine Data Collection? Please, give me some good comments about it. Thanks Manny K
----------------------------------------------
----------------------------------------------
"Methodology for collecting routine route data to catch all bearing failures in the earliest detectable stage

From what I can gather, there are several ways that a bearing degradation can occur. These are as follows:

1. Random high frequency raised noise floor begins to appear in the 50-70x running speed range. This is indicative of very minor defects in the load zone, and is hard to distinguish between normal wear and a bearing problem in some cases. This is the ideal method for bearing failure, the so called L10 life failure, whereby material fatigue begins to gradually spall the bearing surfaces.
2. Spalling of the bearing surface without warning. It is possible for a bearing to flake or spall between one monitoring period and the next. No ultra high frequency data will appear at all, making ultimate reliance on “seeing it coming” from the upper frequency range a risky method.

To accurately analyze conditions, the most important tool is the time waveform, which is difficult or impossible to analyze if more than one range is not used. For instance, too high a resolution requires a long waveform, where high frequency events would be masked. Too low a resolution produces a short waveform where lower frequency events may not even be contained. The FFT can only interpret what is in the time waveform. There is no magic software that can find something that an analyst cannot find by looking at the proper range and resolution time waveform. I am not suggesting we go to all long, 4 second time waveforms where we cannot see higher frequency data. However, I also know that you cannot find lower frequency data in a 100 ms waveform if the actual event takes 90 ms to occur. The fact is, the analyst or the computer requires 3-5 events to be able to find something. Too many, and it is hidden, too few, and there is no pattern. This kind of makes it hard to do routine readings with one setting.

Do not forget that all bearings are not kind enough to fail the same, predictable way. If that were the case, no analysis would be required, and software could easily be written to detect problems and spit out a work request according to some magic formula.

VCI does not say only look at this range, or only look at that range. They take thousands of readings for clients on a routine basis where they always take 3-4 different waveforms and corresponding FFT’s. They may have more time than we do, but we should be heading in that direction to know what we really have out there, and what may come up and surprise us".

OK, I'll bite. I use 8-10 occurences of the highest expected defect frequency of the inner race. This varies with machine speed of course. I will add 1 or sometimes 2 additional ranges where required for watching those low or high frequency generating items but it is very specific to the the type of machine. I don't do it as a general thing. I have been doing this work for about 10 years now and have had pretty reasonable success.
To comment on the author of the speil you quoted, it sounds like a vendor trying to posture himself as some sort of expert in order to boost sales. Anyone can sound like a techno geek in this field if you know the right words..........

It seems obvious this is a portion of a longstanding back-and-forth discussion so it's tough to pick the real context. From the portion you have excerpted, it does look like the author has dismissed the value of looking for bearing defects using a high-Fmax spectrum or demod spectrum. He mentions only elevated noise floor but no recognition of the fact that specific patterns of non-syncronous frequencies with harmonics and possible sidebands that provide unambiguous indication of bearing defect (although severity is never straightforward).

As far as time waveform I agree it is useful in conjunction with the spectra. In particular acceleration TWF peak seems the most widely accepted means for assessing severity of a rolling element defect.

At the end it seems he is suggesting several TWF's of different resolution on every point on every machine. Wow- that would be a tremendous unnecessary burden in data collection time and data storage. IMO, if there is to be included TWF on every point, certainly one TWF would do the job and more could be obtained on an as-needed basis.

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

Pete, that was an old complete write up sent to our supervision, just that. Around that time is when AP sets were changed "drastically", adding many extra points to the route with very low Fmax and very high resolution (to trend each peak), and reducing our Fmax of the other points (60-70X) to very much lower values. We are still in the proccess of correcting all that. After some relexions, I tend to think that maybe (???) we were to become data collectors for the outside outfit to do the analysis. Insane. . . . .

I'm not sure what he is talking about when he talks about resolution in the waveform. The only way to detect high frequency energy in a regular accel waveform is to choose a higher f-max, the length of the waveform will then be determined by what I call the resolution (eg.1600 lines).

I quess the quote defines resolution in the TWF as time difference between two consequitive data samples. The quote seems reasonable to me for a detailed data analysis because of the following reason.

When trying to catch high frequency events, high Sampling Rate=2.56*Fmax and, therefore, high Fmax is required. At given #Lines, it leads to a short span of the TWF data acquisition time, which potentially may cause insufficient number of lower frequency events or their complete loss in the TWF since T=#Lines/Fmax. So, different combinations of #Lines and Fmax are required if one wants to maximize TWF analysis.

Of course, this is impractical in a route set up.

It seems like it's all about compromises--taking enough data to warn of and/or identify impending problems in enough time to do something about it--balanced vs. the time/opportunity to collect & analyze that data.

It seems like anyone touting ONLY one method/facet to collect & analyze hasn't put enough time in the field yet... Shorter/longer time waveforms, lower/higher Fmax's at different resolution, HFE, etc. are all tools at our disposal. Use of just one seems to risk missing what others will make clear; use of all of them all the time is overkill--and unrealistic! I've had to cut as many corners as possible to keep up on my monthly routes at my plant--you can bet that what is left is a mix of technologies, and every point has a purpose! Just my $.02--for all that's worth!

Tony