Vibration Analysis?

[JMBacker]

Hello Everyone,

I'm building a small machine to test and qualify bearings using vibration analysis.  I'm fairly new to the vibration world, but I've read several articles and documents on the subject.  I'm writing a program to take the vibration readings (See attachment), and I have a few questions.  I've read about the sound and vibration analysis toolkit from NI, but this is not in the budget right now.  If I can prove this testing will work, then we'll have to re-evaluate the need for the add-on toolkit.

My current program seems to take readings accurately and repeatably, but the information I'm getting isn't painting a very good picture for me.  I can test a brand-new bearing and one that you can feel and hear has defects, and see what appears to be the same readings.  You can't tell the difference between good and bad bearings.  I've read about defect frequencies for bearings etc., but this doesn't seem to show anything either. 

I've seen some discussion on here about converting from g's to in/sec^2 etc. but I don't really understand this and I'm not sure of the importance of doing this conversion?

Basically, I need to know if there's anything I can do better in the analysis and manipulation of the data, without the sound and vibration toolkit, to get a better reading from the bearing.

Is the spectrum tool right for this?  How should I set it up?

Does anyone have any good links to places I can learn more about vibration analysis?

Is anyone doing this type of testing?  Do you already have code, or are you using a different tool than LABView?

Anything else?  We have a real potential for saving money here, I'm looking for any help.

Thank You,

Jeremy Backer

[johnsold]

Jeremy,

Can you post some typical data?  Ideally one set with a new bearing and one with a bad bearing.

I have done some bearing testing (with analog circuits, before LV).  My concern is that if you cannot see any differences in the raw data, you probably will have difficulty extracting results in software.  What kind of transducer are you using?  How is it attached to the bearing or other part of the system?  Are you testing bearings in machine where they are under load and the machine is doing productive work or is this a test as the bearings are being manufactured?

The inner while loops in your program can be removed.  Dataflow assures that the digital write occurs before and the digital read after the for loop.

Is data acquisition synchronized in any way with the bearing speed?  If not, the averaging may be hiding the results you want.

Lynn 

[JMBacker]

Lynn,

Thanks for the reply.

I'm attaching some pictures of the test rig.  As you can see, we have a small servo motor sitting on top of the bearing, which is a small needle/thrust bearing, used in the planetary gears of automatic transmissions.  The readings are being taken by a general purpose 100mV/g accelerometer, which you can see is mounted on the bottom of a fairly thick steel plate.  I understand there can be some transmissibility problems when going through too much material?  We seem to be getting a reading though.  The only weight on the bearing is the motor and mounting plate, 5-10lbs.  I considered adding weight, and that will probably be my next move.

I'm also attaching some runs I've made.  One set is of a brand new bearing, the other set is of one I stuck in a sandblaster.  It is now very obviously damaged, rough and noisy when you turn it.  You should be able to open these files with the .vi attached to my previous post.  The first tab will be the waveform, raw data.  It is not accurately calibrated to g's, but I think it should be in the ballpark.  The runs are comparable to each other anyway.

I'm not currently alligning the readings with the motor revolution.  Maybe this is something I should look into?  I have a very capable motor/drive setup for this...

I was previously using a legacy daq device, with which I couldn't wire the error lines into the digital line write tool.  You are correct that I can now remove the while loops.

Whatever else I can provide to help you assist me here, let me know.

Thanks for the feedback.

[Jeremy_B]

Hi Jeremy,

You're data looks pretty noisy, it is hard to see any difference between the signal from the good bearing and the bad.  What does the signal look like without any bearing at all in the rig?  You might have a lot of vibration in the rig itself, and it looks like it could have a high enough amplitude to obscure the real signal you are looking for. 

[johnsold]

Jeremy,

I agree with Jeremy_B that you do not have any bearing signal.  The accelerometer location is the most likely suspect, although the signal conditioning electronics may be a problem also.

What I would expect to see is a signal with several components:

1.  The rotational speed of the motor and its harmonics.

2.  The bearing frequency which is the shaft speed multiplied by the number of moving elements in the bearing.  This is the number of balls or rollers in ball/roller bearings.

3.  Brad band noise, possibly modulating the other components. 

4.  (possibly) Some power line frequency interference.  If the servo amplifier is the PWM type, something at the pulse frequency.

All of your data sets have a very weak periodic component (-60 dB).  I did not attempt to calculate the frequency, but my guess would be power line frequency.  The motor speed probably varies a bit, especially with the rough bearings.  This would broaden the peaks.

Is your accelerometer aligned so that its sensitive axis lines up with the direction of the expected vibration?  The pictures do not show the mounting details.

Lynn 

[johnsold]

Jeremy,

I had additional thoughts last night.

Is the transducer attached outside the bolt circle of the bearing support plate?  The vibrations outside that clamping node may be much reduced.

Most accelerometers have a finite bandwidth and many have a resonance just above the highest usable frequency.  Your data seems to show neither of these effects.  Something is not working in your sensing hardware.

Lynn 

[JMBacker]

Hey guys, I've realized the same things.  For some reason, when I switched from the old board I was using to the new board, I lost a lot of the reading.  Not sure why yet?  Sorry for the bum info.  I'm going to put the old board back on and re-take some readings.

The first attachment is basically the same vi as before, modified to open the readings taken with the old board.

The second attachment is readings.  New bearing, Sandblasted bearing, No bearing motor spinning only, and no motor.  These definitely look better and have better information, but they're still not telling me anything.

I looked into the motor setup to make sure I was turning the RPM I thought I was.  I wasn't, it was set up for the wrong encoder.  These readings are taken at 2000 RPM.

The bearing I'm using has 45 rollers.

Nothing is bolted down here.  The bearing just sits in a recess in the base plate.  The motor sits on the bearing, and is held from rotating by 4 guide pins/rods.  Everything is held down by gravity only.  There is considerable mis-alignment/movement sometimes, but I figured we could look through this.  I'm obviously not worried about mis-alignment issues, as this isn't where the bearing will be used.

The accelerometer is bolted to the bottom of the plate, and while we didn't do a precision alignment, it is in the correct axis and lined up fairly closely.  It is mounted outside the bearing.  Should it maybe be mounted directly under the center of the bearing?

Thanks again for all the help!

Jeremy Backer

[johnsold]

Jeremy,

Now you clearly have some real data.

The motor speed lines show up in all the spectra except the No Motor of course. 

The first third of the spectrum (to the large peak) seems to be dominated by the bearings, but the correlation for good and bad is not obvious.  The bearing peaks are slightly broadened and the baseline between peaks is higher with the bad bearings, but I doubt it would be easy to detect in software.

Is it possible that the whole bearing is rotating in the fixture for both the good and bad ones?  Can you hear or feel a difference between good and bad bearings?

I am still thinking that either the transducer location or the fixture is the problem.  When I worked on a similar problem 30 years ago, our vibration technician spent quite a bit of time moving the transducer around on the housing to find a location where he could reliably pick up the signals we wanted.

Lynn 

[JMBacker]

So, doing some math here:

I'm new at this, so let me know if I miss the boat...

Obviously, the motor speed should be 2000RPM/60=  33.33 Hz.

There are definite peaks at 33 Hz.

My roller cage will rotate at half speed of my motor.

A defect on either inner or outer race should show up at:

22.5 Roller impacts / second

* 2000 (RPM)=

45,000 impacts / minute

/60 (seconds)=

750 impacts / second (Hz)

750 Hz, or every .00133 sec. 

Correct?

The info shows a peak at 750-800 Hz, but it doesn't show any higher on the bad bearing vs. the good bearing.

I measured the roller diameter at .077".  Circumference of the roller = 2*3.14*.5*.077 (2pi*r) = .242"  The I.D. of the bearing path is 1.58", the O.D. is roughly 1.93", so the average bearing path diameter should be 1.75", circumference of 5.5".

5.5/.242 = 22.7 roller revolutions per roller cage revolution, 11.35 roller revolutions per motor revolution.

Follow all that? 

11.35*2000(RPM) = 22,700 roller revs / minute.

22,700/60 = 378.33 roller revs / second = 378.33 Hz.

Roller defect frequency should be 378.33 Hz?

Not sure what I'm seeing here?

What are you guys thinking?

[JMBacker]

The bearing housing has tabs which I've shaped the base plate for, so it can't rotate.  However, the motor to bearing mating could be slipping.

I think the next thing I'll pursue, because it's the easiest, is adding some weight.  Then, I'll experiment with the transducer placement.

Thanks for all the advice Lynn.  I'll re-post if I find anything out.