An interpolated FFT spectrum benefits from amplitude (and possibly frequency) recalculations based on the leakage to either side of the FFT line (or bin) containing a peak. A peak located left of center causes the neighboring line to have more “stray” or leaked amplitude than the right-hand side line would, and the proportions of leakage will change as a peak’s frequency moves within a line (the latter should hopefully be recognized as the result of a variation in speed).
For very small RPM changes, the principal measurement comparative problem will arise from the change in position of the peaks relative to the FFT lines. On non-interpolated systems using a Hanning window, this may mean a rise or drop in amplitude of as much as 16% on some peaks, leading the analyst astray in overanalyzing an otherwise perfectly stable machine behavior.
In and of itself, the Hanning window only causes artificial reductions of amplitude, never an artificial increase. Since an FFT spectrum’s specific peak may already be affected by a drop, displacing it (when the RPM changes minutely) can provoke a rise as significant as the stated 16%. Interpolation resolves many of these issues and saves considerable time in restraining artificial variations. Averaging will also deliver better stability, albeit in a different context than what this article considers.
The reader will draw his own conclusions as to the value of interpolated versus non-interpolated monitoring data, whilst keeping in mind that there can occasionally be a need to refer to the unaltered FFT.
© 2007 by François Gagnon