KutupASP DijitalSpectral leakage is the spreading of a peak's energy into neighboring frequency bins when the record does not hold a whole number of periods. A window function tapers the record edges, which reduces that spreading. Hann suits general vibration, flat-top suits amplitude accuracy, and the rectangular window suits whole-period or transient records.
The FFT treats the record as if it repeats. With a whole number of periods, the start and end match and the peak stays in one bin. Otherwise a discontinuity forms between the ends, and energy spreads into neighboring bins. This is spectral leakage.
A window function multiplies the record by weights that taper toward zero at the edges, which reduces the discontinuity. It has a cost: the peak widens and close components become harder to separate. The Hann window is a balanced choice for general vibration. The flat-top window gives accurate peak amplitudes but makes peaks wide. The rectangular window, meaning no window, suits only whole-period or transient records.
Consider a hypothetical record with fs = 1,024 Hz and a duration of 1 second, so N = 1,024 and the bin width is 1 Hz.
A real machine has several frequencies and changing speed, so no single record length suits every component; that is why windows are used. The Welch method averages overlapping segments. For an 8,192-sample record with 1,024-sample segments and 50 percent overlap, the segment count is (8,192 − 1,024) / 512 + 1 = 15.
Window choice is often done badly:
For locating peaks and separating close components, Hann is usually the balanced choice. If you must report the amplitude of one peak accurately, use flat-top. In either case, record which window was used in the report.
Yes. A window changes the weight of each sample, so amplitude scaling is affected. Check whether your software applies the window correction factor. The simplest test is a sinusoid of known amplitude.
Example numbers are hypothetical, not measured field results.
After reading, you can compute frequency resolution for a vibration record, justify your window choice, and read spectral peaks as clues rather than diagnoses, using a worked example.
After reading, you can set a sampling rate from your analysis band, recognise a false low-frequency peak caused by aliasing, and see why slow PLC or SCADA logs fall short for spectra.
