Frequency content of a tachometer signal

I have a tachometer signal from a rotating shaft. I am interested in the frequency content of the signal, since I know that the rotational frequency of the shaft is not exactly constant.

I attached a picture of the signal in its raw form. I tried to do a simple fft of the signal, but that did not give me the answer I was hoping for since the FFT only shows the fundamental rotating frequency (and its multiples) and not much else.

How should I proceed?

Thank you for any help!

Answer 1

[Mikko]

I will answer myself, in case someone is having the same kind of problem: I got what I wanted from the signal using a different software, which converted the pulse signal to angular speed (deg/s) signal. By taking the fft of this angular speed signal I got a nice graph showing the frequency content which I was looking for.

I do not know exactly what kind of operation the software did, but apparently it is about frequency to voltage conversion. All I had to tell the program was the number of teeth on the gearwheel. I can find plenty of information about this conversion in Google, but almost all of it is about electronics, I guess it is a pretty popular thing to do in that domain.

I will maybe try to do the conversion sometime on my own using Matlab, I can post the solution here if I manage to come up with a solution.

Answer 2

[Wayne King]

Just looking at your signal, why would you expect anything more from the Fourier transform than energy at the fundamental frequency and its harmonics? I can't tell how "wide" these pulses are from your plot (what would they look like zooming in) but you essentially have a pulse train with very,very little (if any) additive noise. So what you are reporting from the Fourier transform is exactly what I would expect.

[Mikko]

Hi Wayne,

The fft was just something that I did as a first try since I did not know how exactly I should approach this problem.

The pulses are 2...3 samples "wide", if that helps.

[Wayne King]

But Mikko you say "..since the FFT only shows the fundamental rotating frequency (and its multiples) and not much else."

My question is what do you expect? What you say FFT shows is exactly what I would expect for the frequency content of your signal. That signal should have a discrete spectrum.

[Mikko]

Ahh now I see, I did not understand at first. What I expect is some content on the frequency spectrum on frequencies lower than the fundamental rotating frequency.

I forgot to mention: The measurement itself is made at the flywheel, which has teeth. Each pulse/peak corresponds to one tooth passing the sensor. The pulse signal is not triggered at once per shaft revolution, but let's say a 100 times per revolution (100 being the number of teeth per shaft revolution). Maybe this helps.

[Wayne King]

I don't think you should expect frequency content lower unless there is a DC shift in the signal. From what you describe, I think you're seeing in the spectrum pretty much what I would expect. If you attach the signal as a .mat file and give the sampling frequency, I can take a look.

[Mikko]

I believe that the low frequency content should come from the small time differences of when the pulse occurs. In other words, the pulses should not come at exactly even intervals all the time.

There is content in there, it has been recorded with other methods (but using the same sensors) before. Now I just got the raw signal and I should be able to get it visible. I don't know if I am doing things the right way in simply using FFT, is there some more advanced signal processing involved in a case like this?