An engineer monitoring a speed increaser gearbox recorded 8g overall acceleration with 20-40g impact peaks occurring once per motor revolution. Time waveform analysis revealed the impacts coincided with a specific tooth engagement point. The diagnosis: a chipped gear tooth on the bull gear. Without vibration analysis, that gearbox would have run to catastrophic failure.
Vibration monitoring catches gearbox problems early – but only when you understand what the data means. This guide covers the complete diagnostic process: setting up measurements, identifying frequency signatures, diagnosing faults, and determining when action is required.
Setting Up Vibration Measurements
Reliable diagnosis starts with reliable data. Poor sensor placement or incorrect analyzer settings will mask the very frequencies you need to see.
Sensor Placement on Gearbox Housings
Mount sensors as close to bearings as possible. For gearbox housings, this typically means:
- Input bearing housing – captures motor-side frequencies and input shaft bearing condition
- Output bearing housing – captures load-side frequencies and output bearing condition
- Intermediate shaft bearings – on multi-stage units, each stage needs coverage
Stud mounting with grease coupling provides the best high-frequency response – essential for bearing analysis where defect frequencies often exceed 2 kHz. Magnetic mounts work for routine gear mesh frequency (GMF) monitoring but lose accuracy above 2 kHz due to coupling resonances.
My practice: stud mount for bearing baseline readings, magnetic acceptable for trending GMF on walkthrough routes.
FFT Resolution and Frequency Range Settings
Gearbox spectra contain closely spaced frequencies that require adequate resolution to separate. For most industrial gearboxes:
Frequency range (Fmax): Set at 3.25x the highest gear mesh frequency. This captures GMF, its first three harmonics, and surrounding sidebands without wasting resolution on empty high-frequency space.
Resolution: Use at least 1600 lines if Fmax is below 2000 Hz. For higher Fmax settings, use 3200 lines. Insufficient resolution causes adjacent peaks to merge, hiding the sideband patterns that reveal fault type.
Example: A gearbox with GMF at 600 Hz needs Fmax of at least 1950 Hz (600 x 3.25). At 1600 lines resolution, each line represents 1.22 Hz – enough to separate sidebands spaced at shaft speed.
Understanding Gearbox Frequency Signatures
Every gearbox component generates specific vibration frequencies. Calculating expected frequencies before analyzing spectra tells you exactly where to look – and prevents misidentification.
Gear Mesh Frequency and Harmonics
GMF is the fundamental signature of gear operation:
GMF = Number of Teeth x Shaft RPM
For a 24-tooth gear running at 1492 RPM: GMF = 24 x 1492 = 35,808 cpm (597 Hz)
Healthy gearboxes show GMF and typically 2x GMF in the spectrum. The presence of higher harmonics (3x, 4x GMF) or pronounced sidebands indicates developing problems.
Sidebands appear at intervals equal to shaft speed. Their spacing reveals which shaft has the problem:
- Sidebands at input shaft speed = input gear fault
- Sidebands at output shaft speed = output gear fault
For a gearbox with input at 1492 cpm and output at 942 cpm, sideband spacing immediately identifies the defective gear location.
Bearing Defect Frequencies
Bearing frequencies depend on geometry. Exact values require bearing part numbers, but these approximations work for planning analysis:
| Frequency | Formula | Typical Range |
|---|---|---|
| BPFO (outer race) | 0.4 x N x RPM | 40% of balls x speed |
| BPFI (inner race) | 0.6 x N x RPM | 60% of balls x speed |
| FTF (cage) | 0.4 x RPM | Less than half shaft speed |
Where N = number of rolling elements.
For a bearing with 10 balls running at 1800 RPM: BPFO approximates 7200 cpm (0.4 x 10 x 1800). The actual value depends on contact angle and ball diameter, so use manufacturer data for precise analysis.
Diagnosing Gear Faults
Gear problems show up as changes in GMF amplitude, harmonic content, and sideband patterns. Each pattern points to a specific fault type.
| Symptom | Probable Cause | Verification |
|---|---|---|
| High 1x GMF only | Normal wear, minor misalignment | Check trend – stable is acceptable |
| High 2x GMF | Gear misalignment, worn tooth profile | Verify shaft alignment records |
| Evenly spaced sidebands | Localized damage (chip, crack) | Sideband spacing = defective shaft speed |
| Random noise floor rise | Advanced wear across multiple teeth | Compare to baseline spectrum |
| Very low frequency sidebands | Hunting tooth frequency – severe localized damage | Indicates tooth pair has been compromised |
A practitioner forum case illustrates a common challenge: elevated 2x GMF with low sidebands. The analyst suspected wheel/pinion axis misalignment. The manufacturer disputed this, blaming sensor application (magnetic mount on rough surface). Both possibilities were valid – which reinforces a critical principle.
When a significant change in gear vibration amplitude occurs, check for other mechanical problems before condemning the gears themselves. Coupling misalignment, loose foundation bolts, and resonance can all produce gear-like signatures. Time waveform analysis showing once-per-revolution impacts (as in the introduction case) provides stronger evidence of actual gear damage than spectral amplitude alone.
Detecting Bearing Deterioration
Gearbox bearings fail progressively through four distinct stages. Each stage has characteristic frequency signatures and remaining life estimates that guide maintenance decisions.
The Four Stages of Bearing Failure
Stage 1 – Subsurface Fatigue (>10-20% L10 life remaining)
Microscopic cracks form below the raceway surface, generating ultrasonic frequencies in the 20-60 kHz range. Standard vibration analysis won’t detect this stage – it requires ultrasonic monitoring or specialized high-frequency demodulation.
Stage 2 – Surface Spalling (5-10% L10 life remaining)
Cracks propagate to the surface. Natural frequencies of bearing components (typically 2-6 kHz) become excited. Sidebands appear around these resonances but discrete defect frequencies haven’t emerged yet. Envelope analysis becomes effective here.
Stage 3 – Discrete Defect Frequencies (<1-5% L10 life remaining)
BPFO, BPFI, BSF, and FTF appear in the standard spectrum. When you can identify specific bearing defect frequencies, the bearing has consumed over 95% of its fatigue life. Plan replacement at the earliest opportunity.
Stage 4 – Widespread Damage (~1% L10 life remaining)
Discrete frequencies disappear into a rising broadband noise floor. Ironically, overall vibration levels may decrease as defect edges round off. This is pre-failure territory – the bearing can seize without additional warning.
The practical implication: by the time BPFO shows up clearly in your spectrum, you’ve already lost most of the bearing’s life. Earlier detection requires envelope analysis.
Using Envelope Analysis for Early Detection
Envelope analysis (demodulation) extracts low-energy bearing defect signals from within high-frequency resonance bands. The process:
- Band-pass filter the signal around a bearing resonance (typically 1-20 kHz)
- Rectify the filtered signal to extract the modulation envelope
- Apply FFT to the envelope signal
The resulting spectrum shows bearing defect frequencies that would be invisible in standard analysis. Detection range extends to 3-10x shaft speed, covering the bearing frequency range where standard spectra show only noise.
Limitation to know: envelope analysis loses reliability in Stage 4 when damage becomes widespread and modulation patterns break down. It’s an early detection tool, not a failure confirmation tool.
Setting Severity Thresholds and Action Levels
Knowing something is wrong matters less than knowing when to act. Two threshold systems exist – and both have value.
ISO 10816-3 General Machinery Zones
The ISO standard defines four severity zones based on velocity (mm/s RMS):
| Zone | Velocity (mm/s RMS) | Condition |
|---|---|---|
| A | 0-1.4 | Newly commissioned |
| B | 1.4-2.8 | Acceptable for long-term operation |
| C | 2.8-4.5 | Acceptable only for limited periods |
| D | >4.5 | Damage occurring |
These thresholds were developed for general rotating machinery – pumps, motors, fans. Gearboxes are different.
Gearbox-Specific Limits
According to the Vibration Monitoring Handbook, gearbox-appropriate limits are:
| Level | Velocity (mm/s RMS) | Acceleration (m/s2 RMS) |
|---|---|---|
| Satisfactory | <10 | <70 |
| Maximum Allowable | 15 | 150 |
The “satisfactory” gearbox limit of 10 mm/s RMS is 3.5x higher than ISO 10816’s Zone B boundary of 2.8 mm/s. Gearboxes inherently generate more vibration than simple rotating equipment due to mesh forces and multiple shaft interactions.
Using generic ISO thresholds for gearboxes leads to excessive alarms and unnecessary shutdowns. Using gearbox-specific limits without trending leads to missed deterioration.
Combining Absolute and Baseline Approaches
The NREL Wind Turbine Gearbox Condition Monitoring Round Robin demonstrated that baseline-driven pattern detection outperforms absolute threshold monitoring for complex gearboxes. Multiple analysis partners achieved better diagnostic accuracy when comparing against baseline spectra than when using only absolute values.
My recommendation: use absolute thresholds for overall health screening and immediate danger assessment, baseline trending for specific fault detection and progression tracking. A 25% increase from baseline warrants investigation even if absolute levels remain “acceptable.”
Key Takeaways
Gearbox vibration analysis follows a logical sequence: collect quality data, identify frequency signatures, diagnose specific faults, then assess severity for action decisions.
Critical numbers to remember:
- Gearbox satisfactory limit: 10 mm/s RMS velocity (not ISO’s 2.8 mm/s)
- When BPFO/BPFI appear in spectrum: <5% L10 life remains – plan replacement
- FFT setup: Fmax at 3.25x GMF, 1600+ lines resolution
Before condemning any gearbox component, verify the diagnosis with time waveform analysis and rule out external causes – coupling alignment, foundation, resonance. The vibration pattern indicates the problem, but the source isn’t always where the symptom appears.
The whole point of vibration monitoring is catching problems early. Stage 3 bearing detection still gives you time to plan. Waiting for obvious symptoms means accepting unplanned downtime.
