Our plant has eight circulating water pumps (vertical) that run at approximately 295 rpm. Vibration data is collected monthly at the motor outboard position. Vibration data is collected with an accelerometer integrated to both velocity and displacement. I have noticed that the displacement spectral data shows low frequency noise below running speed, somewhat like ski-sloping, but with a number of discrete peaks. I did some testing today, collecting data with both a standard 100 mV/g accelerometer (which is normally used) and a 500 mV/g Low Frequency accelerometer. I also tried some different high-pass filters (2 Hz, 2.3 Hz, 2.67 Hz, and 4 Hz).

The attached plots compare data collected from the same machine, position, data collector, and cable using a 100 mV/g accelerometer and a 500 mV/g accelerometer.

The data from the low-frequency accelerometer is "cleaner", but I'm curious why the data from the 100 mV/g accelerometer shows the distinct peaks below running speed.

The specification sheet for the 100 mV/g accelerometer indicates a frequency response of +/-5% from 1.5 - 5000 Hz, while the 500 mV/g indicates +/-5% from 0.5 - 2000 Hz.

Any thoughts/comments?

Circ_H2O_Spectrum_Plots.pdf (53 KB, 72 downloads)

Steve, perhaps all 100mv accels are not "created equal".... the $69 "industrial" accelerometers are cheap for a reason. To start with they have a ceramic element. I'm sure there are other differences.

Of course, you may be using a top quality accel, but if not it might be interesting to do some quality comparisons.

At 295 rpm, I assume these are large, high flow pumps. You're bound to have a lot of "flow noise", much of it sub-synchronous.

The 100 mV/g accelerometer is a Wilcoxon 793, not a "cheapy" as far as I know.

No, that's a very good accel (as far as I know) and the one I use for most all routine work.

Now I'm curious. I have probably 10 different types/brands of 100 mv accels... I guess I need to compare them at low frequencies to see if I see the same phenomena that you are seeing.

looks like noise on both transducers. However if it is some kind of electrical interferrence at a fixed voltage level the 500 mv/g accelerometer will interpret it as vibration but at 1/5 the amplitude as the 100 mv/g transducer. Looks like that is what you have on the 2 plots.
Regards,
Rick

I don’t have a lot of experience on this. But fwiw, I’m inclined to agree with Rick.

Outputting in displacement involves double-integration, which will enhance low frequency content. To the extent there is any broadband white-like noise present in acceleration signal from any source, the low-frequency part will be amplified.

That broadband “noise” in the acceleraiton signal can be either real (from flow, rub etc) or artificial noise inherent in the accelerometer electronics.

I think that trying accelerometers with different sensitivites as you did is a good tool to separate those two different types of broadband signals.

The electronic noise will tend to be constant voltage magnitude, regardless of high or low sensitivity accelerometer. This same voltage will look like a higher vibration on the low-sensitivity accelerometer, lower vibration on the higher-sensitivity accelerometer.

The “noise” from flow gives higher voltage for the more sensitive 500mv/g accelerometer and lower voltage for the less sensitive 100 mv/g accelerometer. Since the voltage changes in proportion to the sensitivity, the vibration from this source looks the same on both accelerometers.

Since the stuff below 1x is higher on the low-sensitivty acceleromter, it is probably electronics noise.

Since the stuff above 1x looks the same on both accelerometers, it is probably real flow or machine "noise".

At least that seems to make sense from thinking it through. I haven’t paid a lot of attention to that aspect of our CW pump/motor data collection, so don’t have much experience.

Also I assume settling time was sufficient.

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

Steve,
I used to demo multi channel modal analysis systems. We were supposed to use an aluminum weldment. I always demo'd on whatever the customer would eventually want to test. Probably why I sold a lot of systems.
Anyway, the demo system I had used a 10 mV/g triax. I was asked to do a simple modal on a 2000 hp motor. It was there that the lesson of the 'brain dead' accelerometer was learned the hard way. The data looked sort of OK. There were some peaks, but they weren't that clean and the curve fit results were ghastly!
Moved over to 1 v/g accels and a 3 lb impact hammer with a 10 lb extender weight on it's butt and the softest tip in the box.
I agree with RL and EP. Your data collector is amplifying whatever it gets. In the case of the 100 v/g accel, that is insufficient signal, so it amplifies noise.
Regards,
Ron Brook

I would like to recommend that any newbies reading these boards that have all kind of questions about tranducers and signals and why they get what they get look into a one day course on digital signal processing. I believe PCB/IMI is running one currently.

Steve,

Electronic noise is a plausable idea, but I see in the 100 mv accel plot what looks like defined peaks below the 1x. I would rather expect the electronic noise to be random.

What are the peaks telling us if any? It also strange that if this is truly electronic noise, why it ubruptly ends right before 1x?

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

Dave,

My initial post referenced the "discrete peaks" below 1X. I don't believe that it is real, because I see it on all 8 motors. These are large ( 2500 hp / ~200,000 gpm) vertical motors/pumps. There are four per unit, and the two sets of four are physically separated, so I can't see a physical phenomenon that would affect them all. The spacing between the peaks is approximately 30 cpm. Any physical phenomenon that would create these peaks from a kingsbury thrust or a radial (sleeve) guide bearing?

Waterfall_List.pdf (56 KB, 20 downloads)

Steve,

Do the peaks produced with the 100 mv accel also exist if you utilize same settings but in a different location? ( I mean whether or not they are due to internal noise ?).

Also consider in general that signal is integrated twice and in the low frequency range same g's will be of is the fact that these peaks are comparable with 1x ). I bet, in velocity the issue does not look that dramatic.

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

I dont know if this will help but once on a troubleshooting call we had a customer using a CSI 2130, a 100 mV/g accel and a 500 mV/g accel. The 500 mV/g was producing ski slope and the 100 was not. The time waveform showed a bias voltage shift in the 500 mV/g but not the 100 mV/g. In our test we had point 1 setup for 100 mV/g and point 2 for 500 mV/g. When the user had built point two he just used all the same settings he had for point one except he changed the sensitivity to 500 mV/g.

So, to test a theory, I made a measurement with the 500 mV/g on point 1 and got good data. It was 5x too high, but it was good data. Point 2 didn't produce good data even with the 100 mV/g sensor.

Later I helped the customer get "high level" CSI tech support on the phone and listened in. I'm not an expert on CSI but basically the tech support person told our customer to "let the software build the 500 mV/g low-frequency data point rather than copying the 100 mV/g and changing the sensitivity."

This solved the problem.

Hope that story helps. Do you have the time waveform data? You can easily see if its bias shift if there is a very low frequency sine wave superimposed on the vibration data (i.e. its drifting above and below 0).

-Mike Scott
-IMI Sensors

I think I am going to have to go down the path that David was going down. This does not appear to be electrical noise as I have seen it.

I will offer this test.

Set up a point with the highest resolution that you have available. Set the FMAX fairly low - something like 24000 CPM. Set the start freq to zero but the low frequency cut off a bit higher - use 24000/6400*8= 30 CPM. Do not average. Collect one sample. If the data still has discreet peaks, it is a fair bet they exist due to something other than sensor issues.

Your integration slope will be about 4 bins of resolution - which is exactly what you have in the 3000 CPM 800 line reading.

Also- if you have a velocity or displacement sensor, give that a try.