Vibration analysis using wavelet transform

[AX3L]

Hi everyone!

I'm making a system that will detect all different kinds of fault on different machinery like bearing, misalignment etc. I understand the concept on wavelet transforms and its advantages over the regular FFT for these kind of signals but I can't see how to compare and analyse the values of the wavelet transform to in fact detect any faults (automatically, of course). If anyone have any knowledge or know where I can find some information in this area it would be deeply appreciated!

Sincerely
Axel

 

[electricpete]

google wavelet machinery condition monitoring

=====================================
(2B)+(2B)' ?

 

[AX3L]

Electricpete:

Thanks for the hint, but I've done some serious research for quite some time now without any real success so far. The thing is, that there's a lot of papers of this topic showing that wavelet is able to detect faults and showing different kinds of wavelet plots for different faults. And while I obviously can see that there's a difference, I cannot find any good reliable relationship between faults and the plot (like in the fft analysis, where bearing faults shows at x times the rotational frequency).

 

[GregLocock]

I do use wavelets but only as an aid to problem identification while listening to the signal. I imagine setting up an automated analysis based on wavelets would be rather tricky, or at best would be the same as using STFT.

Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376

http://eng-tips.com/market.cfm?

 

[AX3L]

Thanks for the answer Greg, informative as usual!
Tricky it is for sure, and it doesn't get easier since you need different wavelets for different applications. But maybe identifying the fault isn't the right approach here, maybe it's better to find just that something is wrong and then use something like fft to find what's actually wrong (or use manual inspection).

But Greg, is there anything special you're looking for in your wavelet plot or is it just that something has changed? And what wavelets algorithms are you using the most?

 

[amanuensis]

Hi,
It's important to have a smart approach when using wavelets.

The choice of the kind of wavelet to use is very very important.

To do this, you have to look at your time-history signal and see precisely the shape of the curve where appears the default.
Then, the idea is to select the wavelet that matches the best the shape of the default.

Then the wavelet spectrogram gonna amplify the default, even though it can scarcely be seen on the time history signal.

 

[AX3L]

amanuensis, I'm not entirely sure what you mean by looking at the time-history signal and what means by default and to amplify that. If you could please explain the steps a bit more in detail it would be very helpful!

Or are wavelet uneccesarry complicated? The problem I have are large and fast variations on the RPM as well as load of the machines so I assume Wavelet and STFT is a good approach to this. But maybe there are better methods?

 

[GregLocock]

I'll try and find out the wavelet function, but it is 12 years since I used them last for noise signals.



Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376

http://eng-tips.com/market.cfm?

 

[AX3L]

Thanks Greg, it would be very appreciated!