Vibration analysis

[joost]

Hi Radu,

I haven't tried to run the app note you refer to, however, now that you are talking spindles...

1. I take it you measure during the spindle's idle time? If not, the grinding vibration will over shadow any of the bearing vibration, misalignment etc you're trying to pick up.

2. I am familiar with high speed motorized spindles up to 20-30KRPM Take the high end 30KRPM this means a speed of 500Hz. Most applicable bearings would give you an inner/outer race defect frequency of about 10 times the speed and you want to pick up at least 5 (ideally 10-20) harmonics to get some resolution. Soooh that means a sample speed of 500 * 10 * 5 = 25kHz on the low side.

3. Bearing signals are impactive in nature. It is best to apply a non-linear method to emphasise these impacts. Common in the vibration industry is the so called envelop method which is nothing more than:
signal --> band pass filter --> rectifier --> smoothing filter. You'd do your FFT on the enveloped signal.

Hope this helps.

[RPJ]

Thank you Joost.

I actually need to record the grinding vibrations.
I was initially thinking to use a 30 kHz sampling rate but I believe it is OK to use 10 kHz for grinding chatter or forced vibrations. I'm not really looking to see the behavior of bearings but more like capture process vibrations through the bearings. It is probably not the best solution but this is not the point here... The low sampling rate is also explained by the fact that I am simultaneously recording force and power signals and I am forced to do it at the same sampling rate as for vibrations.

Any way, your suggestion might work for my case. You were saying: "signal --> band pass filter --> rectifier --> smoothing filter. You'd do your FFT on the enveloped signal." Should I understand that you would just use the FFT advanced vi to get some of the data that interests you? Do you have an example that you could share?

I would like to see the displacement and the dominant frequencies. Would a power spectrum be given by the advanced vis? (I'm using LabView 7.1).

Thank you.

[joost]

Hi Cobus,

For A to V, your unity gain should be at 61.4Hz (g to ips)
For V to D, your unity gain should be at 318.3Hz (ips to mil pp)

Not sure what you're filters are doing but the relations above must be observed for a true conversion.

Btw keep in mind that a derived displacement value compared to an actually value (e.g. from an eddy current probe) may show quite a bit of error particularly in the low frequencies.

Joost

[joost]

Hi Radu,

I actually don't use the labview software (at all) but matlab. My background is in vibration monitoring, hence the recommendations I gave you. Yes you would want to do FFT analysis on the enveloped signal. However, the enveloper emphasizes "impact" like signals (such as bearing defects, shocks etc). It will in essence deminish any signal that is of singular sinusoidal behavior or non-complex repetitive.

Set the band pass around your region of interest and set your smoothing filter corner frequency to twice that of your high pass corner of the band pass. Start with a 3 or 4 pole filter set, nothing too steep. Don't have very small band pass filters; you want to preserve the group delay within the pass band.

You may be able to infer displacement from double integration but I would expect some 10-20% error in your readings (I don't have very good experiences with double integration). What dominant frequencies are you expecting? (i.e. sources)

Good luck,

Joost

[RPJ]

Joost,

It was a surprise to see that you are not actually using labView.
Since you have background in monitoring vibrations, could you tell me what are the usual parameters of a vibration (acceleration signal) that you would use to characterize that vibration and compare it with other? For example: If I measure forces I would like to see the average or maximum force such that I can later tell 'this process required a higher force and it was this much'. In case of vibrations what would I consider? I am a little confused by the fact that everybody talks about acceleration, displacement, dominant frequencies, power spectrum and power spectral (or spectrum) density, but nobody seems to be able to focus on something specific.
What parameters/magnitudes would you usually compare when you look at different vibrations? For example: you get a chart of acceleration from sensors, then what can you do to this signal - you could integrate (twice) and find out the displacement graph that would show you the trend in that vibration (if it increases or not); you could apply an Fast Fourier Transform that wiould allow you to get a chart having frequency on X axis (?), you could determine the power spectrum that is also displayed in a chart... A lot of charts but what would be a value that you would search for and actually display, where do you get the dominant frequencies and their values?

Thank you

[joost]

Hi Radu,

I normally use specialized software for vibration analysis. I use NI data acquisition for ad hoc prototyping and use matlab for back end data analysis during prototyping. I've used Labview in the past but found it fairly cumbersome compared to a programmatic language.

First of all, the difference between Acc, Vel and Disp: for your purposes there is absolutely no need to integrate all the way to D. What will happen if you do is that frequencies say above 100Hz or so are suppressed to the extend that FFT analysis yields nothing useful. Particularly if you sample at 10kHz, most of your spectrum isn't going to show anything useful. For this application we always use 100mV/G accel sensors and in most cases I simply diagnose in Accel though many of our customers add Vel to it as well (Vel measurements are such a standard thing to do in the industry, there is no way escaping them). What you should keep in mind is that between Acc and Vel the unity gain is at 61.4Hz so an accel signal at 6.14Hz is 10 times higher in Vel. Consequently a signal at 614Hz is 10 times smaller in Vel.

We never diagnose during grinding, always during idle. This because we're interested in the condition of the bearing and misalignment or unbalance. One cannot perform this measurement while grinding as the grinding process adds a tremendous amount of vibration. Since you mentioned to be interested in measuring during grinding I cannot tell you much here. Measurements taking during idle can be as much as 0.25Gp as I have seen but in general you need to relate the actual overall level to the size and type of the spindle. Bearing failures have a very characteristic behavior and actual signal amplitude of the bearing defect signals is actually not the first parameter I look at before calling it a failure. It is the harmonic pattern I am interested in.

As far as a spectrum calculation, I use a simple FFT (not a power spectrum) as this is standard in the industry and provides the best visibility on higher frequencies (power spectra lower a frequency's amplitude by the root of the actual frequency).

Sooh, what to look for in your spectra? First determine which failures you want to capture, than determine what that failure looks like in time domain (e.g. unbalance looks like a sine, a bearing defect looks like an impact/pulse) than determine what that time domain signal looks like in a spectrum.

In general I would recommend a customer to baseline a series of known good spindles and use that baseline to set simple overall value alarm levels. In addition, I would make sure that from all spindles the bearing frequencies are known (determined by the bearing type) and watch for frequencies to pop up that correspond to the bearing frequencies. These are best monitored by a straight simple accel FFT as well as a enveloped accel FFT. In order to help automate all this data diagnosing we have specialized routines that sort of crunch the spectra down to a single number that indicates the likelihood of a bearing defect being present.

There are several bearing databases that could tell you which defect frequencies they produce. Some of the databases are free or cost $300-500 bucks or so. If you know your bearings I can see if I can find the frequencies in the databases I have here.

Hope this helps.

Joost

[RPJ]

Joost,

Thank you for your reply. It is very helpful.
I talked to some people meanwhile and I came to similar conclusions.
As for the bearings, I am not sure yet. If I can find out I'll let you know; maybe you can find something. However, I would have hard time using your data without a paper or company reference; on the other hand it would make a great reference point.

Radu

[RPJ]

If anybody out there has more comments or could share a simple vi for vibration analysis I would appreciate your help. I hate reinventing the wheel...

[Cobus]

Joost,
So what you mean is that I should multiply my signal by 10 after integration at a freq. of approx 6Hz ??

Another few questions:

** Is this relation true for metric convertion - no difference when using inches or mm ??
** I employ a high pass Butterworth filter (std VI - 2nd order) before each integration. How do you calculate the effect of filtering on the amplitude of the result ?? For example if I have a signal with fundemental freq of 4 Hz @ 7mm p-p. How do I determine a good low cut-off filter setting and how would you have to correct the amplitude after filtering & integration to still obtain an accurate result ??

As you may notice I do not have much knowledge in this field but would appreciate any help I could find.

Kind Regards

[RPJ]

Joost,

I found out about the bearings: they are angular contact bearings type 7210 TYSULP4.
Do you have anything shearable on these type of bearings?

What (good) books would you recommend for vibration or modal analysis. I need to read more about this subject and I am not sure what to choose. I'm interested mostly in practical aspect (what to acquire, what tools/analysis should I apply, what should I look for in the output, and how should I descipher or interpret this output). Theory is good but I do not want to get lost in 'the forest' of formulas and not reach the target...

Thank you for your time and expertise.

Radu