Machinery health
From vibration data to a useful signal
Sampling, sensor mounting, windowing and operating conditions must be considered together before interpreting an FFT plot.
The measurement chain starts at the sensor
An accelerometer’s bandwidth and range matter, but so does its mounting. Surface condition, orientation and cable movement can change the signal. Record sensor position and axis to make measurements repeatable. Store machine speed, load and temperature alongside the trace.
Choose sampling for the question
Select sampling rate from the highest frequency of interest. The Nyquist condition is only a starting point; the anti-alias filter transition band also matters. Uncontrolled high-frequency input can fold into misleading lower-frequency content that a later software filter cannot recover.
Record length sets frequency resolution
FFT bin spacing is sampling frequency divided by sample count. A longer record gives finer separation for stationary signals, but changing operating conditions may mix different machine states in the same record. Document window choice, amplitude correction and units.
One peak is not a diagnosis
A spectral peak alone does not prove imbalance, misalignment or bearing damage. Review speed-related components, harmonics, sidebands, the time waveform and changing loads together. Baseline comparisons are useful only when measurement conditions are sufficiently comparable.
Keep the analysis repeatable
Record the sensor and mounting, preserve raw data, verify units and retain filter and sampling parameters. Include the processing version with the output. These records make software analysis a traceable part of machinery-health decisions.
Illustrative plots are educational. Decisions about real equipment require appropriate field measurements and expert assessment.
