time frequency views of machinery vibration

[vgarzani]

Used to be PdM person at NucPowerStation, now am pump Eng. but still dabble. I was intrigued by Gaberson's recent article in S&V on Wavelets. Found an excellent time freq. toolbox for Matlab at:

http://crttsn.univ-nantes.fr/~auger/tftb.html

- skip to the chase, the Choi-Williams distribution is calculated pretty efficiently by the function tfrcw.m (requires Matlab and I think the signal processing toolbox)
Current process is a kludge, but is getting better - export waveform data from MasterTrend / RBMware database - then import to Matlab and mess with. Just finished automating some of this to speed it up -> using ODBC access and some VB to efficiently write waveform files which can be easily read by Matlab. Then put together tf views with scaling of Time in shaft revolutions for X and Freq in orders for Y. This has almost made the process reasonable. But the views seem to be very interresting to show "what is happening when".
Before doing a tfrcw(sig) you might want to decimate and then make it analytic with hilbert(sig). Having fun with these novel views - anyone else doing something similar?

 

[electricpete]

Well, I work at a nuc station on electric motors, somewhat involved in pdm. I do some signal processing with matlab. But I never got interested in wavelets or alternative views of my data.

What can it do for us? either practical-wise or just giving us some insight.

 

[GregLocock]

We've used wavelet data in the past for human response to sounds - without a great deal of success I might add.

The ear/brain behaves in some respects like a constant %age bandwidth device, whilst at other times it behaves more like an FFT analyser. In particular the tone discrimination performance of people when listening to short bursts of tones often seems to exceed the limit set by the Heisenberg uncertainity principle (B*T>=1). Wavelets offer a visual analogy to that process. However, I am told by people who have studied hearing far more deeply than I this is not a particularly meaningful model of the hearing process, what we should be doing is looking at the modulus of response of lots of little band pass filters. Then you can get into the great fight of whether we can hear phase. If anyone is interested my answer is yes... depending on how you define phase.

Cheers

Greg Locock

 

[vgarzani]

Have to admit - no new insights or catches yet. Some interesting structure apparent in geardrive data - but you guessed it, we had it called anyway. I do have a situation that seems to be pretty well depicted with time-freq view, I think involving unsteady flow in a single stage vertical. Approximately every 8 shaft revolutions (not fixed) there is excitation of some apparent mode between 2 and 3 times running speed. And vane pass seems to go along in a counterpoint pattern.

This used to be the kind of feel for situations you used to get by watching a real time analyzer in the field or with tape recorded data.

Electricpete - are you doing any partial discharge monitoring? If so possibly doing more than the canned IRIS processing?

 

[electricpete]

Yes, we have 30 13.2KV motors and 2 25KV generators with Iris partial discharge equipment. I have taken their course and studied some of the literature and articles. I guess I don't find much need to go beyond their simple rules for looking at things like +/- predominance, line-to-ground or phase-to-phase activity, temperature dependence, load dependence (special test). I enter my summary statistics into excel and automatically import winding tempearture, ambient temperature, and ambient humidity. I let excel compute correlation with those parameters, but I haven't seen any useful patterns other than using the correlation to identify temperature dependence. What kind of stuff have you seen done in that area.

I didn't mean to put down your idea by asking for the practical side. I was just trying to figure out if it is worth investing my time in. It sounds like for more complex signals that are not striclty periodic it may have some good uses.

Since you're a matlab guy you might be interested in this page that I have posted a few times on this forum

http://www.geocities.com/pschimpf/impacts/electricpete.htm

I have done quite a few other similar things using Maple. I have a few models that evaluate unbalanced magnetic pull on a motor in the presence of static, dynamic, eccentricity, and the behavior of RBPF. Nothing too earthshaking. Also working on a model for vibrations in the presence of broken rotor bar but not done with that one yet.

Your name rings a bell as if perhaps I have met you at a seminar or something. Does V stand for Vincent?