What it is about
The time signal from an accelerometer mixes all the machine’s vibration sources in a single trace, and it is difficult to interpret. The Fourier transform breaks that signal down into a sum of sine waves of different frequencies. The spectrum is the plot of the amplitude of each one against its frequency. The FFT is the efficient algorithm that calculates it from sampled data.
Why it is useful
Each mechanical phenomenon leaves its mark at characteristic frequencies. Unbalance appears at the rotating speed, gears at the gear mesh frequency, and defective bearings at frequencies associated with their geometry. Seeing the frequencies makes it possible to know which component is producing the vibration.
Parameters that determine the result
- Maximum frequency: the sampling rate must be more than twice the highest frequency of interest (Nyquist criterion). An anti-aliasing filter prevents high frequencies from appearing as false low frequencies.
- Resolution: it is inversely proportional to the record time. Finer resolution is obtained by recording for longer or with more lines, which makes it possible to separate closely spaced peaks.
- Window: because the record is finite, the calculation assumes that it repeats, and this causes energy to leak between frequencies. A window (for example, Hanning) reduces the effect at the cost of widening the peaks.
- Averaging: reduces random noise and stabilizes the result.
How to read a spectrum
- Identify the rotating speed and mark its multiples (1X, 2X, 3X…).
- See which peaks are harmonics of the speed and which are not.
- Look for sidebands around major peaks.
- Check the background noise level at high frequencies.
- Compare with previous spectra from the same point.
Limits
The spectrum assumes a machine under stable conditions during the record. With variable speed or load, the peaks widen. It also does not show transient events in time; for those, the waveform must be examined as well.