my task is to develop a condition monitoring system of a machine module. For that I am using the vibration sensors placed on the machine housing. I want to focus on only one module of the machine, a rack and pinion system, in which the rack carries out one reciprocating movement (back and forward) per machine cycle. I have carried out some test measurements to see if I see the vibrations corresponding to the cycle of the module. I could not see any clear vibrations that I could trace to rack and pinion module. Here is the qualitative representation (The signal above is the vibration, the curve below is the rotational speed of the rack): Qualitative vibration signal and speed representation The rotational speed curve is just an assumption, since I do not have any speed sensor in place to measure it.

There are many rotating gears in the transmission of the machine and I need to extract only the vibration signal of my module. What I want to do is to install a proximity sensor next to the pinion teeth to be able to measure its rotation. I would then know when exactly does my rotation start and end, and I would be able to differentiate the signal flanks to get the instantaneous rotational speed of the pinion. Once I have this information I would be able to extract the parts of the signal when the cycle occurs and analyse them using STFT or Order Tracking (using the calculated instantaneous speed). enter image description here My questions are:

  1. Is this approach correct? Or should I just divide the signal into sections corresponding to the cycle of the whole machine and analyse them?
  2. Are there some ready methods to deal with this task, that I could read on?
  3. Can I just cut the signal exactly where the movement of the pinion starts and use it just like that or should I use some windowing function?
  • $\begingroup$ No, we should immediately be thinking in anything involving obtaining the spectra (FFT, spectrogram) of your signals, and observing the evolution and discovering of any peaks... $\endgroup$
    – Brethlosze
    Apr 15 '17 at 20:35
  • $\begingroup$ Thanks for your answer! The FFT i did with my data is very "smeared", and the spectrogram image also does not yield any useful data. The peaks I see might come from different modules of the machine. I have read that in case of non-stationary movement as a source of vibration the right thing to do is resampling the data with the rotational frequency and not using the usual fft and stft. What makes you think such approach is not correct in my case? $\endgroup$ Apr 18 '17 at 9:49
  • $\begingroup$ I insist, the proper technique for your cyclic machine keeps being taking the spectrograms. Nobody is forcing you to follow the de-facto, but you must do that analysis, and double checking you are measuring the proper thing. Remember to measure the solid parts of the machine, and not just some housing. The sensor will measure the element in which you install it, hence you must ensure the element has a good mechanical impedance with your source (solid steel/concrete). You must have the FFT analysis done before trying to invent anything new. And besides, there is a physical reason for this. $\endgroup$
    – Brethlosze
    Apr 19 '17 at 14:02

Your Answer

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

Browse other questions tagged or ask your own question.