There seems to be enough side comments going on about this subject to make it worthy of a new thread.

Personally, if I want a velocity TWF or spectrum, I see no problem using the integrated output of an accelerometer (other than a small increase in the low-frequency noise content).

I know there are people out there who disagree. I would like to understand what objections they have to integration.

Well, I'm no expert on the details of integration but I have used accelerometers since I started 10 years ago and I have used velocity as my primary analysis measurement with (in my humble opinion)great success so, I know it works just fine. There are purists out there (like Sam) and they are entitiled to do or say as they wish. I think it boils down to keeping it simple for overall equipment montoring and getting into more detail by using acceleration if needed. I actually collect both on most of my points anyway. I just use different filters on the g's reading such as gSE. I also only look at acceleration waveforms, rarely velocity waveforms.

BTW Pete; I think the title of this post should read Integrated Velocity vs. Acceleration.

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

I think all will agree that velocity is the universal language. It is readilly understood as a severity that all relate to.

Velocity as a measure is good and has its place and a velocity measurement may be best suited for a given application.

Velocity integrated from acceleration is a different topic in my opinion. I've been using acceleration since 1979; have an acceleraton based vibration system. In my own sytsem there's an acel spectrum displayed and also a list of frequencies and amplitudes in g's, IPS, and mils. All analyses are done basically in g's but language in a formal report is generally velocity for ease of understanding. There isn't a pure integration from acel to vel; it is relative and not absolute - so the difference is little! boo-hoo! After 25 years of acel, working with acel, in acel, it's second nature to me. Also it doesn't produce the cumberson problems of going back and forth. You already have the RR so why jump into the volkswagon? A picute is worth a thousand words and you can't analyze what you can't see. Too simple? The acelerometer produces a voltage at a given frequency, integrating it to velocity doesn't magically make it a velocity signal. It is as good as it gets in its raw form. Now, there are many functions; hilbert transform, cesptrum, etc........ many analyses techniques. I'm not limited to any one nor am I recommending any one! Use all tools available. It's junk science that gets to me. I'm just back in from the field collecting data with a CSI, RTA and ME-42 --- the CSI is off in left field show values that can't be produced on anything else. Back in the lab and downloading into the PC and wham-o the 7 g's that was displayed on the CSI screen in the field can't be found in the data anywhere on the PC?????? You can manipulate to make a number of different readings that look like junk science to me that do not conform to anything or any law of physics and maybe collect false data. Anyway they never agree with other data from other analyzers. So how do you CSI users function?

Since my family have used velocity probes since they made their first vibrometer w. velocity probe around 1950 I agree to be somewhat biased but not by a bias voltage . Anyway a true moving coil velocity probe do have some advantages, first signal transmission since it may use balanced transfer on 2 twisted shielded cables and any induced electrical interference may be reduced at the receiving end by at least 70-80dB. So in a rolling steel mill it´s the only thing I know that give a signal that only comes from vibration at the end of the cable. I have only seen something similar in airborne hi cost accels w/o builtin charge amp´s. Any induced interference using a accel w. builtin or outside charge amp. will produce the sum of the vibration and interference signals at the end of the cable At lo freq. (less than 5Hz) velocity probes may give some more signal and certainly a way better signal/noise ratio. At high freq (above 6-8KHz) the output and linearity is limited but how linear is a accel so close to the resonance, vel. probes don´t have that resonance problem and I have seen 2xGM from a gbx at 18KHz from a vel. probe that gave accels major problem as it excited the resonance.
Long time stability, I have seen velocity probes in use for 40+ years, since they are built from copper and other pure materials they loose output much slower than a semiconductor device. Drawback as I see it is cost due to production in small numbers and that sometimes I/you like to have raw acceleration from a accelerometer so I am not that stubborn so maybe you would like to buy a velocity probe w. builtin accel? I know, it will be to expensive. From strict measurement point of view, if you want velocity, use a probe that give that direct, if you want acceleration, use a probe that give that. The less conversion the less problems. So for everyday use I have (analog)integrated velocity from an accel and our latest vibrometer do use a accel with builtin amp. so I have to reluctantly bend for the cost effectiveness and that is how cost influence what I use compared to what I belive. (I still carry around a simple vibrometer w. vel. probe just to be sure) For difficult places like steel mills and hi quality demand places like nukes and slow speed places w. electric interference and long service periods and long cables like hydro power plants I still can promote the use of moving coil velocity probes as the extra cost can be justified. You can also remote test velocity probes for function in operation at the end of the cable w/o access to the probe, handy in nukes. Sorry for ranting I couldn´t resist. Olov

The fact is that the integrated data is useful; so use it. They are not mutually exclusive.

I like Sams comment about the RR and volkswagon.

I use CSI equipment, and have done just fine. In terms of conformity to physics, well, thats another thread.
I look at the data in all forms, as I am still developing my sense of severity.

quote:

A picute is worth a thousand words and you can't analyze what you can't see. Too simple?

Sam,

Can you post us an example or two of what one would miss in a spectrum, by storing intergrated data in velocity rather than in acceleration and then analyzing it in intergrated velocity and (derivative) acceleration, and assumming everyone stores their waveform in real acceleration.

Also can you post the comparison of the data from the CSI meter to the other two you mentioned? This seems strange that there would be such a dramatic difference. The only thing I really don't trust is the CSI conversion of a stored acceleration waveform to velocity or displacement. I have never collected each one and stored them seperately for comparison, but the conversion doesn't look real, especially the displacement waveform.

Comparing acceleration measurements to integrated velocity measurements from an accelerometer, I don't see any problems except as follows:

1. The high frequency noise caused by bearing deterioration is filtered out by the integration process. No surprise here. Look at integration as a 6 dB/octave low pass filter.

2. Analog integration requires infinite gain as the frequency approaches zero. This isn't possible or usually usefull, so integration circuits typically stop integrating at 5 - 10 Hz. Anything below this cutoff frequency will be lower than the expected level by a little or a lot depending on the design of the integrator circuit. The practical effect of this is that integrated velocity isn't the correct choice for very low speed machines when you might expect it to be the best choice.

Comparing analog integration of an accelerometer to a velocity transducer.... A velocity transducer operates ABOVE its natural frequency, with little output below this frequency. A typical velocity transducer has a resonance of about 5 Hz, making the usefull range about the same as an integrated accelerometer. Olov's velocity sensors are the exception to the rule, with a resonance of 1 Hz or so I believe? This extends their lower limit to be better than an integrated accelerometer signal. Of course, yoou won't get the high frequency response of an accelerometer to detect bearing deterioration.

Regarding Sam's comment that integrated acceleration provides "relative" velocity, I really don't see the difference. Sure, an impact-type event might show some differences in characteristics, vibration levels from the usual culprits such as balance and misalignment should show no difference at all.

Jon
Spintelligent Labs

Jon, I have to admit, I cheat with electronics so it has a damped mechanical resonance at 35-40Hz and is in fact from that down to 1Hz an accel, not a piezo one but anyway so it´s integrated conventionally to velocity in that range but it give very good hi freq. data for such a device and can and is used for bearing condition. I have seen a report from a 3 letter bearing mfg. that in short say, ´If you like early detection and a instable trend use acceleration and peak rectifier and if you want a some later detection but stable trend use velocity and RMS rectifier´, so I use both....
By the way the analog integrators in the old CSI boxes go down to 0.1 and 0.15Hz if I remember correct. I have not tried them knowingly to that but the components used indicate that. I can add that using hi output accels like 500mV/g do improve signal levels at lo freq. so you have some more signal to integrate but will have a lower hi freq. performance so either way you turn, you have the rear side of your pants pointing backwards.... It is always a compromise of many things. Olov

quote:

Originally posted by electricpete:

Personally, if I want a velocity TWF or spectrum, I see no problem using the integrated output of an accelerometer (other than a small increase in the low-frequency noise content).

Digital integration of an acceleration TWF will produce noise (due to integration errors) just in the first several bins of a spectrum and this could be easily cut out from the spectrum. So, I agree, this is not a problem.

With the velocity TWF integrated from acceleration TWF the story is different. Integration will add low frequency content, so the velocity TWF will be drifting in time.

It is not easy to see it in a velocity TWF, but if one wants to see the drift, use Y and X eccelerometer and plot the orbit in velocity. You are going to see as each revolution produces a velocity orbit which is slightly shifted from the previous one.

OEMs could filter out the low frequency using analog filters but, I quess, it is hard to eliminate it completely.

It's quite practical to have low noise digitally integrated and double-integrated velocity and displacement time domain data from accelerometers if you do more than use the simplest of methods such as are found in the typical data collector. To get the kind of signals that you used to be able to get on the oscilloscope outputs of old, analog, IRD, Metrix, Bruel & Kjaer or others, you also need to filter the low frequencies appropriately like these old analyzers did, either with the transducer or with hardware filters. There are comparable ISO specs for how the low frequencies should be filtered for overall levels.

GOSH!!!!

This thread died in one days's time, in fact in 9 hours and 13 minutes.
What happened?
Didn't get too deep I am sure.

Seems like a lot of unanswered questions were left hanging.

I don't think the thread died: Jon did a very good job answering. Sometimes integration will drop the levels into the noise floor or below a preset threshold and out of the analyses picture.