← Back to the chapter · Next detail →

Technical detail · Edge Data: Sampling and Statistical Compression

Motor harmonics, aliases, and payload

A 1,800 RPM shaft turns at 30 Hz. Suppose a hypothetical source produces eight impulses per revolution: 8×30=2408 \times 30 = 240 Hz. This is an illustrative impulse train, not a bearing-defect prediction. Its second and third harmonics are 480 Hz and 720 Hz, so the highest illustrated frequency gives:

fsample≥2×720=1440 Hzf_{sample} \geq 2 \times 720 = 1440\text{ Hz}

A margin of 2.5 times this minimum gives 3,600 Hz, rounded to about 4 kHz. At only 100 Hz, the Nyquist frequency is 50 Hz and the 240 Hz component aliases to:

∣240−round⁡(240/100)×100∣=∣240−200∣=40 Hz|240-\operatorname{round}(240/100)\times100|=|240-200|=40\text{ Hz}

That false 40 Hz pattern can hide fault evidence and allow a developing failure to go undetected. At 4 kHz and two bytes per sample, raw traffic is 8 kB/s, or 8 kB per one-second FFT analysis window. Ten frequency peaks at 120 B/s give a 66.7-fold reduction, rounded to 67-fold, before metadata and transport overhead. This retains the input’s 16-bit sampling resolution before reduction, but does not preserve its complete waveform. FFT reduction is a candidate for battery-powered monitoring when the retained peaks preserve the required faults.

← Back to the chapter · Next detail →