I would appreciate your thoughts on the attached comparison. At a recent trade conference, the subject came up about the best way to monitor for bearing defects. We all know about 10X inner race as a good fmax but one guy suggested going way out to where the bearing natural frequency would occur and he finds that this is the place to see those initial impacts. Since I have a bearing showing a defect in the HFD spectrum, I took a reading at 7kHz fmax acceleration with just a small high pass filter and was able to clearly see the defect "way out there". Now, my question is this; Would it be better to do both? I'm sure there are those of you out there who never use HFD for bearing monitoring and some who depend on it. What I'm finding is that the HFD can be variable depending on the condition of the mounting and the state of the actual defect. Perhaps a plain old high fmax/lines spectrum is just as good and more reliable. Your thoughts?

Accelfltrcompare.doc (74 Kb, 79 downloads)

Maybe it depends on what a person wishes to know and if you are reporting bearing conditions or reporting equipment that needs to be worked on in order to prevent unscheduled downtime.

In the spectrums you showed, under my current setup with most equipment my waveform pk-pk acceleration trend would have picked up and shown an increase.

I could then go out and look for the cause by acquiring a larger fmax spectrum, If I wanted to see it.

Or I could have chose peakvue and selected 1000hz fmax and 1000hz filter for most 1800rpm equipment and saw what the cause was.

In the end it is my job to report bearing conditions as a ( 1 ) If they are likely to cause downtime if not worked on during the upcoming shutdown.

I do go on and list as ( 2 ) and ( 3 ) bearing conditions, 2's being bearing they could consider changing if they have time and manpower and 3's being really only for me to keep watch on and monitor regularly and upgrade as needed.

Your familiarity with your equipment should give you an opinion on the stage of failure
and how urgent the need for repair is.

"just my 2 cents"

Mike

Hi Ron,

quote:

I took a reading at 7kHz fmax acceleration with just a small high pass filter

I am not clear on the "small HP filter" setting you used.

Is the top spectrum an enveloped spectrum or just a normal spectrum taken in Gs with some type of HP Filter that doesn't make it an enveloped data?

I see, in the upper left corner of the top spectrum, a note that the HP is 2.1075 Hz, but see no note in the lower spectrum, except the words "high frequency" in the center at the top and the top spectrum says "none" in this same area. Are both enveloped or what? Is this 2.1075 Hz sort of like a low end cut off?

I am not familair with the meter and system you use, could you explain more of what and how the data is taken? What is a "small HP Filter" and how does it differ from the lower spectrum, Etc.

I agree with what Mike said, but he uses, as I do, CSI and is a totally different setup from what you have. His system of collecting data has been proven over years with a great success rate.

Thanks, Ron.

[QUOTE]Is the top spectrum an enveloped spectrum or just a normal spectrum taken in Gs with some type of HP Filter that doesn't make it an enveloped data?

I see, in the upper left corner of the top spectrum, a note that the HP is 2.1075 Hz, but see no note in the lower spectrum, except the words "high frequency" in the center at the top and the top spectrum says "none" in this same area. Are both enveloped or what? Is this 2.1075 Hz sort of like a low end cut off?

I am not familair with the meter and system you use, could you explain more of what and how the data is taken? What is a "small HP Filter" and how does it differ from the lower spectrum, Etc.

I agree with what Mike said, but he uses, as I do, CSI and is a totally different setup from what you have. His system of collecting data has been proven over years with a great success rate.

QUOTE]

Top is normal with a "low cutoff" which I call High pass meaning anything below is not shown. This is to filter out low end noise but you know that already.

The lower spectrum is High Frequency filter which is a 5KHz gSE (HFD) filter meaning anything below this filter is not shown. The fmax is 500Hz but the box will still use the full dynamic range which is 40KHz when performing this measurement thereby capturing the peaks that exist above the fmax.

The box I use is a dataPAC 1500 taking data from a stud mounted 100mv/g accelerometer. My usual method is to collect HFD filtered data on each bearing plus a velocity spectrum with at least 10x the highest frequency of interest which may be bearing defect or gearmesh. My usual filter is called overall on normal non HFD data which defaults to 2x fmax/lor.

My thought here is to add the 7KHz acceleration spectrum with an overall filter to each bearing. Just another view of the data. The reason is to see the "real" peaks as they exist above 5KHz rather than depend on the HFD filter to place them in the lower range. I assume you understand how this works.

Also, do you think this post is worth the time? I may be out in left field here. My goal is to simplify the data I collect but still ensure that I capture the defects as they occur. It's not a question of success rate. I'm also not wanting to reinvent the wheel either. I appreciate your input.

If you had an AP SET that looked out to the high frequency area you have a concern about and were able to view this AP's overall value on your meter before moving on to another point in the route, this would be of some help.

Example:
Say your normal Fmax of a point is 1000 Hz and your #2 AP Set's covered area was from 1KHz to 10KHz (or whatever range you choose) and you had a value set in your mind as to where the overall amplitude should be before acquiring additional data out to this area, then collecting high frequency data on every point would not be required.
But if your meter does not allow you to view the AP Set's values quickly while in the field, then this is out.

Is this as clear as mud?

That's pretty clear. I can take acceleration overall and see the amplitude as I collect with the previous amplitude showing.
So, this would be a way to trigger collecting off route high frequency data.
The other way is to set the storage parameter of the point at "collect on alarm" but that would mean loading the alarm values with the route data which I have found causes problems with memory capacity in the collector.
I think I'll give it some more thought and do some experiments. I have been using the same set up for the last 10 years with success but I'm interested in trying some different approaches. Mainly to be sure I'm getting the best data for each point with the most efficient amount of time spent. Thanks for the input. If you have more, bring it on.

Something to consider: I've always setup a machine by monitoring it running - look at the spectra over different base-band frequencies...... it will be very obvious where the interest are and what frequencies are of interest and should be monitored. I think you can hear the rocks cry out if you listen close enough............Just a thought.

I wonder why vendors still have HFD in their instruments if they also have demodulation methods? Isn't it true that the latter are superior?

Isn't HFD some kind of atavism left in the instruments for those who like old good times?

When doing consultancy work or measurements at long intervals for customers that prefere to do that I feel a need to cover all bases. Bearing faults when new start generating frequencies hi up. When getting older they decrease in frequency. So if you don´t know the age of the fault I use a couple of wideband HFD type measurements and demod to be sure to capture whatever there is to see. I also find the HFD type methods easier to trend and to have quick say on the level of damage from the readings up front but that may be from what I am used to only. Special cases are VFD induced bearing faults that I have found very quick produce hi HFD readings for pretty long time and then just as fast fails and where demod in some cases look different until the final stage probably due to the different way these bearings are destroyed compared to the "usual" methods. I also find HFD good indicators for lube status and I don´t find that so easy using demod. So maybe it´s out of habit and long measurement intervals or one shot measurements I find HFD methods useful. So I use at least 3 bearing methods at the same time to be sure. Normally they indicate the same but in some cases they differ. Olov

Olov,

Altough HFD vs demodulation issue deserves a separate thread, let me just say here that demodulation in my view is advantangeous.

First, (in case of PeakVue) raw signal is sampled, after passing through HP, at a rate of 100,000 samples/sec and records peak value in a predefined time interval. HFD, on the other hand, just comes up with an overall RMS value in a predefined frequency range and is derived by using analog circuitry.

Secondly, a demodulated TWF also can be trended indicating possible lube problem, but it produces a spectrum that may or may not indicate a bearing defect as well.

Thirdly, demodulation will also give you a sufficiently advanced warning.

David

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

So, demodulation, HFD, PeakVue etc. are, as we know, filters that can highlight high frequency peaks that occur over 5KHz for example and can then show them on a lower range spectrum. The process is looking "way out there" for you. But, if you take an unfiltered acceleration spectrum with a 10KHz+ fmax and sufficient number of LOR, you can see them yourself anyway. It may boil down to a question of which is faster and which one do you trust more, the digital/analog process or your eyes. And don't get me wrong, I use the gSE filter in my dataPAC (5KHz) with a lot of success. So, I'm not advocating to stop using these tools, I want to understand whether it's the best approach for detecting these failures. I'm basically looking at all the angles.