Vibration Analysis

FFT Analysis: Reading a Vibration Spectrum

How to interpret the fast Fourier transform output in vibration analysis, identifying fault frequencies, harmonics, sidebands, and noise floor changes that indicate specific machine faults.

Key Concept

A vibration spectrum is a map. Frequency is the horizontal axis, amplitude is the vertical axis, and each peak is a landmark that corresponds to a specific physical event in the machine. Reading the spectrum means knowing what landmark to expect at what frequency, and what it means when a landmark grows.

Frequency → Amplitude noise 1× Imbalance 2× Misalignment 3× BPFO 1× sidebands Noise floor Synchronous (1×, 2×, 3×) Non-synchronous (bearing freq.) 1× 2× 3× BPFO FFT spectrum anatomy: synchronous peaks at 1× and harmonics indicate imbalance or misalignment. Non-synchronous peaks at bearing defect frequencies indicate bearing damage.

The Frequency Axis: Orders and Hertz

The X axis of a vibration spectrum can be labeled two ways. Absolute frequency uses Hz (or CPM, where CPM = Hz x 60). Orders express frequency as multiples of shaft speed: 1x is shaft speed, 2x is twice shaft speed, and so on.

Orders are preferred for fault diagnosis because they remain consistent across different shaft speeds. If you collect a spectrum at 1,500 RPM and another at 1,800 RPM, the order scale shows the same fault patterns in the same positions. The Hz scale does not.

Bearing fault frequencies are fractional orders, calculated from the bearing geometry. BPFO (ball pass frequency, outer race) typically falls between 3x and 7x depending on the bearing. Expressing it in orders means the calculation holds regardless of the operating speed at collection time.

Scale Unit Best for Limitation
Absolute frequency Hz or CPM Electrical faults (line frequency), resonance Fault positions shift with speed changes
Orders Multiples of 1x Mechanical faults, bearing damage, imbalance Requires accurate shaft speed reference

The Amplitude Axis

Three units appear in vibration spectra, each suited to a different frequency range.

A spectrum plot shows one unit. Confirm which unit the plot uses before interpreting any amplitude level. An amplitude of 0.5 means something very different in g than in mm/s.

Identifying the Shaft Speed (1x) Component

The 1x component appears in every rotating machine spectrum. It represents any force that repeats once per shaft revolution: residual imbalance, slight eccentricity, small amounts of shaft bow.

A clean 1x with low amplitude and no significant harmonics is the baseline signature of a well-balanced machine. Amplitude thresholds depend on machine type and speed, but the absence of harmonics is the more diagnostic indicator.

A dominant 1x with no other elevated peaks is the classic pattern for imbalance. The amplitude typically increases with the square of shaft speed, which helps distinguish imbalance from other 1x sources such as misalignment, where axial amplitude also rises.

Harmonic Series

When a mechanical force repeats at a fundamental frequency, energy appears at integer multiples of that frequency. These are harmonics: 2x, 3x, 4x of the fundamental.

The harmonic pattern, not just the fundamental, carries the diagnostic information.

Harmonic pattern Likely fault Confirming evidence
Dominant 1x, elevated 2x (especially axial) Angular misalignment High axial vibration, phase difference across coupling
Strong 1x, strong 2x (radial) Parallel offset misalignment Phase checks at coupling, 180-degree radial relationship
Series at gear mesh frequency (GMF, 2xGMF, 3xGMF) Gear mesh forces GMF = shaft speed x tooth count
Harmonics at bearing fault frequency (BPFO, 2xBPFO) Advancing bearing damage Confirmed by envelope spectrum
Multiple harmonics of 1x, high amplitudes Rub or looseness Half-order peaks sometimes present with rub

Sidebands

A sideband is a peak that appears at a fixed frequency spacing above and below a carrier frequency. Sidebands form when one frequency modulates another in amplitude or frequency.

The sideband spacing identifies the modulating source. The sideband amplitude relative to the carrier indicates fault severity.

Bearing inner race (BPFI) sidebands

Inner race defects rotate through the load zone with each shaft revolution. This produces amplitude modulation at shaft speed. The result: sidebands at BPFI plus and minus 1x shaft speed (BPFI +/- 1x). Multiple sidebands indicate more severe damage.

Gear mesh sidebands

A damaged gear produces sidebands around gear mesh frequency at the shaft speed of the defective gear. Sidebands at GMF +/- shaft speed of the input gear indicate input shaft gear damage; sidebands at GMF +/- shaft speed of the output gear indicate output shaft damage. A single damaged tooth generates a once-per-revolution modulation; a worn gear generates sidebands that are more numerous and more symmetric.

Motor rotor bar sidebands

Broken or cracked rotor bars in induction motors generate sidebands at twice line frequency (2xLF, typically 120 Hz) plus and minus shaft speed. The sideband pair at 2xLF +/- 1x is the rotor bar fault signature. Amplitude of the sidebands relative to the 2xLF peak indicates severity.

BPFI sidebands: BPFI ± 1x, BPFI ± 2x, BPFI ± 3x GMF sidebands: GMF ± n x (shaft speed of defective gear) Rotor bar: 2xLF ± 1x, 2xLF ± 2x

The Noise Floor

The noise floor is the broadband background energy distributed across the spectrum between the discrete peaks. In a healthy machine, it is low and relatively flat.

A rising noise floor indicates distributed energy: many small impulses rather than one localized defect. This is a distinct diagnostic signature from a single elevated peak.

A clean noise floor with a single discrete peak at a known fault frequency is the pattern for an early-stage, localized defect. This is the favorable detection scenario: the fault is specific and the machine is otherwise healthy.

The Envelope Spectrum (High-Frequency Envelope Analysis)

The standard velocity spectrum is effective for fault detection above a certain severity threshold. For early-stage bearing defects, a different technique is required.

Envelope analysis (also called demodulation or HFD) processes the high-frequency portion of the vibration signal, typically 1 kHz to 20 kHz. In this range, bearing defect impacts excite structural resonances in the sensor and housing, producing a high-frequency burst with each defect contact. The technique extracts the amplitude modulation pattern from this high-frequency carrier, producing an envelope spectrum that shows the modulating frequency.

The envelope spectrum shows BPFO, BPFI, BSF, and FTF even when their amplitudes in the standard spectrum are below the detection threshold. This extends detection to Stage 1 and Stage 2 bearing damage, when intervention prevents catastrophic failure rather than responding to it.

Bearing damage stage Detectable in standard spectrum? Detectable in envelope spectrum?
Stage 1 (subsurface fatigue, no surface damage) No Yes (subtle)
Stage 2 (surface pitting, early spalling) Marginally Yes (clear)
Stage 3 (spalling spreading, noise floor rising) Yes Yes (strong)
Stage 4 (advanced damage, heat generation) Yes (dominant) Yes (may be distorted by noise)

Practical Spectrum Reading Sequence

A consistent reading sequence prevents missed findings and confirmation bias (looking only for what you expect to find).

  1. Establish shaft speed: obtain from tachometer, nameplate, or by identifying the 1x peak. Record in RPM and convert to Hz for the absolute scale.
  2. Mark expected fault frequencies: calculate BPFO, BPFI, BSF, and FTF for the bearing at this measurement point using the bearing geometry and confirmed shaft speed. Mark these positions on the spectrum before examining amplitude.
  3. Check overall vibration level: compare the RMS velocity to the applicable ISO 10816 zone boundary for this machine type and power rating. This establishes whether the machine is in a normal or abnormal overall condition.
  4. Examine each expected peak location: starting with 1x, check amplitude against the baseline for this specific measurement point. A 25 to 50 percent increase above baseline at any peak location is a trend worth flagging; a doubling is an alert.
  5. Compare against baseline at each fault frequency: BPFO, BPFI, and any harmonic positions. Absence of a peak where one is expected is not a problem. Presence of a peak where none was in the baseline is a finding.
  6. Examine the noise floor: compare the background energy level to the baseline spectrum for this point. A rising floor with no new discrete peaks suggests distributed damage or fluid-related issues.
  7. Examine sideband patterns: for any elevated peak, check for symmetric sidebands. Identify the sideband spacing and match it to a known modulating source (shaft speed, gear shaft speed, 2x line frequency).

Document the findings, not conclusions. Record: measurement point, date, amplitude at each fault frequency, overall level, noise floor observation, any sideband spacing detected. Conclusions follow from trend data across multiple collection intervals.

Continuous Vibration Monitoring with Tractian

Tractian sensors collect vibration spectra continuously, detect fault frequency changes automatically, and alert your team before damage reaches Stage 3.

See Mechanical Monitoring
mechanicalmonitoring.com is an educational resource produced by Tractian. Content is provided for informational purposes only and does not substitute for site-specific engineering assessment. Always consult qualified personnel and applicable standards (ISO 10816, ISO 13373, ISO 20958) before making maintenance decisions.