FFT of bearing vibration

[RKN]

Hi,

I was given with bearing vibrations in time waveform, for analysis I need to convert that to frequency domain, in my VI I have used the specral analysis tool to convert time domain to frquency domain. I just need someone to confirm that I'm proceeding in the correct way. from the time domain waveform I can see the bearing has roller faults as well as inner race faults. please confirm whether my investigations is correct. Find the attached VI.

RKN.

[RavensFan]

What do you want us to check?

Your VI consists of a missing subVI called bearing_2.vi, and an Express VI that is set up to do spectral measurements to give FFT in magnitude and phase.  On the surface, that looks like it should work.

[Remmix]

Hi RKN,

Could you pls help. I have acquired a series of more or less similar data that i need to investigate or evaluate possible causes. How did you evaluate that the time series waveform has roller and inner race faults?

Thanks

[johnsold]

Since you did not include any data, it is hard to tell what charactersitics are best used to identify various faults.

Often you can obtain the shaft speed from a fundamental frequency component of the signal or from a separate tachometer signal. From that and the number of rollers you can determine the frequency of components which will be generated by the bearing. 

One way to do this is to establish a range of signal amplitudes (or ratios of amplitudes of harmonic to fundamental) for "good" devices and a range for "bad" devices. With a little luck the ranges will not overlap. In that case a simple amplitude comparison will work.

Another factor is that bad bearings may have considerable random noise associated. It may modulate the bearing frequency signal or simply be superimposed on the total signal. The spectral characteristic of noise is wide bandwidth and may be useful.

It is also possible to get a "click" or "bump" once per revolution which means that it will be at the shaft speed. This type of signal usually has a harmonic signature which is different from that of a good device, but the differences may be small. A time domain detection of the click might work better than a spectral analysis.

Lynn

[Remmix]

Lynn thanks for your reply. I will attach the data that i have. I have wired the data files into a Spectral Express (FFT) VI and i can detect the peaks in frequency domain. I was not sure how I would be able to distinguish 'good' bearings from 'bad' ( or the faults in them) or if the signals are burried in random noise. I have read the white paper on vibration analysis on NI website which mentioned that you can use Cepstrum (Quefrency) to get more accurate bearing fault analysis by obvious harmonics from the spectrum however my graph did not produce any readings so far. I don't have the Professional version which comes with the Cepstrum VI at this point though, I used an equivalent logic of a Cepstrum VI which was posted on the forum previously.

I guess the shaft speed will be the 1st harmonic in the spectrum. I only know the number of balls in each bearing. Any pointers appreciated

Regards,

Raz

[johnsold]

Raz,

I did something similar to this before NI developed LabVIEW. We took about 100 devices and made measurements of the sound spectra. Using other methods the devices were categorized as good or bad. We manually compared the spectra to determine which characteristics were associated with good and which with bad. We found frequency ranges where there was no overlap between the good and bad levels and used a threshold between the levels as the decision criterion.  We actually had both gear and bearing problems to identify so we ended up with three different tests for the separate types of problems.

The key point I am making here is that no one can tell you where the threshold will be or which frequency bands or which harmonics will be the best indicators of the problems in your devices. You will need to evaluate your equipment and make those decisions.

Lynn