Reading a vibration log

What the FFT plot is telling you

A fast-Fourier transform of a vibration signal converts it from a time-domain trace into a frequency-domain plot: amplitude on the y-axis, frequency on the x-axis. Every peak is a repeating force inside the machine — a gear mesh, a shaft rotation, a vane pass, a bearing race frequency. The shape of the plot tells you which forces are dominant, which are merely present, and which are running away from the baseline.

Cross-reference each peak against the running speed (1× RPM) and the operating frequencies of the components you know are inside the asset. A spike at exactly 1× with no harmonics usually means unbalance. A spike at 2× with phase-stable 1× often points to misalignment. Sidebands spaced at 1× around a carrier ring frequency are the classic outer-race bearing signature.

Common failure signatures

Imbalance shows energy almost exclusively at 1× RPM with the harmonic series decaying quickly; severity rises linearly with the imbalance moment. Misalignment puts comparable energy at 2× (parallel) or 1× and 2× combined with a third peak (angular). Looseness gives a family of harmonics rich enough to look noisy, with sub-harmonics at 0.5× and 0.33× showing footstep-style crosses on the waveform.

Bearing defects start small — a ringing inner race tone, an outer-race impact train that breaks up into sidebands as the defect grows. Don’t wait for a smeared broadband rise: by then the inner race is spalling and the corrective work has already shifted from cleaning to replacement.

What to escalate, what to trend

Escalate now: any peak that crosses the alarm threshold, any sideband structure that didn’t exist in the last reading, any broadband rise accompanied by a temperature gain. These are not trending problems — they are read-the-gauge-right-now moments.

Trend and log: peaks that grow less than 10–15% over a quarter, harmonic creep on a known-loose footbolt, sub-harmonics on a coupling that has been loose for six months. Capture the date, the amplitude delta, the operating condition, and the suspected cause. The value of the log is that six months from now you can show the curve instead of arguing from memory.

1. Plot the alarm thresholds on the same axis as the trending data, then read the gap, not just the number.
2. Note the load and the speed at every reading. Two peaks at the same RPM at different loads are different stories.
3. Write one sentence on probable cause into the work order the moment you see the spike, not after the corrective work is half-done.
4. If a corrective action does not bring the vibration back inside the alert band within the agreed interval, escalate — don’t re-log the same point.