Chatter detection

[Marci_2023]

Hi, my name is Marci, and this is my first post here,

I would like to achieve the online (real-time) chatter detection of turning processes by power spectral analysis. I use producer-consumer architecture, in the producer loop I read (10000 samples at a time) the signal of the TDMS and send it to a que and in the consumer I read from the que and use the Spectral Measurements Express VI's Power Spectrum with Hanning Window with linear results. So as far as I know this is an STFT method. I  can see the power spectrum of all of the signals, but my problem is that I don't really know how to filter these signals, and I can't really reproduce the mathematical formulas of chatter identification by Power Spectrum Analysis in LabVIEW and thus i cant do the feature extraction from the spectrum. I would like to detect aperiodic frequency components, because the periodic ones are probably due to spindle rotation. So in this case I figured that I would have to use some kind of a Mask to sort the periodic frequencies and aperiodic ones. So if anybody has a better understanding of this, or can recommend me ways to do this, that would help a lot. I cannot attach the TDMS files, cause they are way too big, even in a zip, but I attached how the signal looks in time domain. 
I use Labview 2020, advanced signal processing toolkit 2021 and DAQmx 2021. 
Thanks in advance!
Marci

[GerdW]

Hi Marci,

---

@Marci_2023 wrote:

I cannot attach the TDMS files, cause they are way too big, even in a zip, but I attached how the signal looks in time domain.

---

We cannot analyze a signal just from an image of a graph…

What about attaching zipped TDMS files containing just the parts of interest? (Sampling at 10kS/s results in files of <80kB/s/ch…)

---

@Marci_2023 wrote:

I  can see the power spectrum of all of the signals, but my problem is that I don't really know how to filter these signals, and I can't really reproduce the mathematical formulas of chatter identification by Power Spectrum Analysis in LabVIEW and thus i cant do the feature extraction from the spectrum. I would like to detect aperiodic frequency components, because the periodic ones are probably due to spindle rotation. So in this case I figured that I would have to use some kind of a Mask to sort the periodic frequencies and aperiodic ones.

---

You probably know the spindle speed, so you also know its base frequency (rpm/60).

You may filter that base frequency and its multiples (±range) from your FFT result…

(When there are gears and bearings then they also produce their own base frequencies you need to take into account.)