0
$\begingroup$

I have vibration data from WT (gearbox) sampled at 25.6 kHz. From a previous post I received a lot of good recommendation.

What I am trying to do:

Find early stage bearing faults by using an envelope analysis aka. an high frequency resonance technique. This technique can:

Magnify the low amplitude, high frequency signal by filtering out low frequency components. .. fault indication frequencies can be determined in the envelope spectrum through a spectrum analyzer

-Fault Detection Of Rolling Element Bearings by C. Liu

Current method:

  1. Apply bandpass filter (low: 5000, High: 6000Hz)
  2. Rectify resulting signal (absolute value AND squaring to remove aliasing)
  3. Apply lowpass at 6000 Hz

-> Fault detection: Do this for many other 10 second samples over time to look for changes in kurtosis, FFT plot etc.

enter image description here

Zoomed in images of the signals processed:

enter image description here

Questions:

  • Should I upsample before or after rectifying?
  • Does this process look correct?
  • Should I do another LP filter to "clean" up the signal?
  • Should I do another LP filter at a lower cutoff frequency to "clean" up the signal?

Thanks in advance.

My approach resembles this one, from -Fault Detection Of Rolling Element Bearings by C. Liu enter image description here

| improve this question | | | | |
$\endgroup$
  • $\begingroup$ Your BPF has a bandwidth of about 1kHz. Squaring will about double the bandwidth. So you could drop the cutoff of the final LPF to about 2kHz to get a ‘better’ envelope. It also looks like you are using an absolute value for your rectification rather than squaring, which will result in more smearing. Maybe try squaring, LPF, then square root? I haven’t done envelope detection this way so not sure. $\endgroup$ – Dan Szabo Mar 12 at 16:25
0
$\begingroup$

I think the question as is, is unanswerable, but here I am making an attempt of correct it with these comments as answer.

Should I upsample before or after rectifying?

For the algorithm, if you are finally assessing against a threshold in magnitude, you should not care about down|upsampling. Do not add that.

Does this process look correct?

Strictly speaking, you are taking the algorithm from your document. Since your document has 200 pages, this is someway hardly to say if you are taking correctly the approach taken by the author.

Should I do another LP filter to "clean" up the signal?

This depend on the specific fault you want to capture for the bearing mechanics. From there, you can decide which frequencies to focus, and how to assess them.

Should I do another LP filter at a lower cutoff frequency to "clean" up the signal?

Again, please check specifically how your algorithm is built. Once you have it implemented, we could study modifications or alternatives.

| improve this answer | | | | |
$\endgroup$

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.