2
$\begingroup$

Last week I've been trying to implement the Order Analysis in MATLAB in vain. I've read a lot of docs about it like this one , but I still can't figure it out. This will be my last shot.

Say I have recorded 10sec audio file of a rotating machine, the sampling frequency is 96Khz. In this 10sec interval, the speed of the machine will continuously increase, say from 800rpm to 1500rpm. Is there any way to calculate the Order Analysis for this machine? I'll be grateful for any hint!

$\endgroup$
  • $\begingroup$ Do anyone did plot the order tracked spectrum in python? $\endgroup$ – mouli ravindran May 23 at 5:56
1
$\begingroup$

Well my friend, if you read the following B&K tutorial about the Order Tracking, then you should encounter so called Vold-Kalman Order Filtering. Personally I don't like it (rather prefer Chirp-Fourier based methods for extracting the response of the system).

You can find plenty implementations in the internet. When you look for MATLAB, then please refer to:

Vold-Kalman order tracking code

Second generation Vold-Kalman Order Filtering.

And most certainly I suggest you to search the bottom of the prof. Jiri Tuma website: click!

For more theory please take a look at - I'm sure you will like it:

Characteristics of the Vold-Kalman Order Tracking Filter

Vold-Kalman Order Tracking Filtration

Main Principles and Limitations of the Current Order Tracking Methods

Realization of the Vold-Kalman Tracking Filter - A Least Squares problem

Advanced Vold-Kalman Filtering Order Tracking - do not be scared, it's not in chineese ;)

$\endgroup$
1
$\begingroup$

Well, I am doing the same kind of research myself, and I did meet the same problems you've met.

For my case, the fluctuation in speed is approx. 1%, not as large as yours, but enough for disturbing fault diagnosis. Thus I presume what you really want is not actually 'fix' the speed variation, so that what have been masked by smear effect due to this variation in frequency domain can be revealed---but simply plot the frequency components in angular domain instead of frequency domain. In latter case I believe what have been kindly answered above is quite useful. But if you were doing fault analysis and try to fix small speed fluctuation, I think refer to TSA method or just simply resample the data in angular domain to reach uniform sampling points between each shaft spin will do.

Hope it will help you 2 years earlier and I hope if anyone see this please comments.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.