Gearbox failures don't happen without warning — they happen without detection. Gear tooth pitting begins as microscopic surface fatigue, progresses through macro-pitting and spalling, and terminates in tooth fracture over timescales ranging from weeks to months depending on load, lubrication, and metallurgical quality. AI analyzing gear mesh frequency sidebands and harmonics compresses that detection window from weeks before failure to months before it. Get iFactory Support and connect your gearbox vibration data to AI diagnostics today.
Identify Pitting, Scoring and Root Cracking Before Tooth Fracture
iFactory AI tracks gear mesh frequency sidebands and harmonics continuously — trending degradation severity from minor surface wear to imminent tooth failure with mathematically grounded severity scoring.
The Gear Failure Cascade: Surface Fatigue to Tooth Fracture
Every gear failure mode follows a predictable physical progression — and each stage in that progression produces measurable changes in the vibration spectrum that AI models are trained to identify and classify. Understanding the progression informs both the detection strategy and the urgency of intervention at each severity level. Contact iFactory to map your gearbox failure modes to the iFactory diagnostic model.
Stage 1
Micro-Pitting (Frosting)
Sub-micron surface fatigue cracks develop in the pitch line contact zone under high contact stress and thin elastohydrodynamic film conditions. Vibration signature: slight increase in gear mesh frequency (GMF) harmonics 3rd through 5th. Detectable by AI 8–16 weeks before progression to macro-pitting.
Stage 2
Macro-Pitting (Spalling)
Surface cracks propagate and intersect, removing material in discrete pits. Each pit creates an impulse at tooth mesh that sidebands the GMF at shaft rotational frequency intervals. AI sideband amplitude tracking provides quantitative severity assessment distinguishing Stage 2 from both Stage 1 and Stage 3.
Stage 3
Scoring (Adhesive Wear)
Lubrication film collapse under extreme pressure or temperature causes metal-to-metal contact and adhesive material transfer between meshing tooth flanks. Vibration signature shifts: broadband noise floor elevation plus distinct GMF harmonic amplitude increases differentiating scoring from pitting.
Stage 4
Root Cracking (Bending Fatigue)
Fatigue cracks initiating at the tooth root — the highest stress point in bending — propagate across the tooth cross-section. Root cracks produce a distinctive phase modulation of the GMF plus ghost frequency components that AI separates from surface-damage signatures using cepstrum analysis.
Stage 5
Tooth Fracture (Catastrophic)
Complete tooth separation produces an impact impulse at every revolution of the gear, detectable as a strong periodic impulse at shaft rotation frequency with GMF amplitude collapse. At this stage, continued operation causes rapid secondary damage — replacement rather than repair is typically required.
Stage 6
Abrasive Wear (Contamination)
Lubricant contamination by metallic debris or external particulates creates abrasive wear on tooth flanks. Distinct from fatigue-initiated pitting: AI oil debris monitoring correlation with vibration data differentiates contamination-driven wear from contact fatigue, directing the correct root cause intervention.
Gear Mesh Frequency Analysis: The Core Detection Method
Gear mesh frequency (GMF = shaft speed × number of teeth) and its harmonics are the primary carriers of gear health information. The spectral structure around GMF harmonics encodes the nature and severity of gear tooth damage in mathematically predictable ways. iFactory AI continuously monitors this spectral structure against baseline models for every gearbox in your facility.
Scroll for more →
| Failure Mode | Primary Spectral Indicator | Secondary Indicators | AI Analysis Method |
|---|---|---|---|
| Distributed Pitting | GMF harmonics 1×–5× amplitude increase | Broadband noise floor elevation, kurtosis increase | Harmonic amplitude trending + statistical kurtosis |
| Localized Pitting (1 tooth) | GMF sidebands at ±1× shaft speed and harmonics | Time-domain impulses at shaft rotational period | Sideband amplitude + time-synchronous averaging |
| Root Cracking | GMF phase modulation + ghost harmonics | Cepstrum quefrency components at shaft period | Cepstrum analysis + phase demodulation |
| Scoring | GMF 2× and 3× elevated + broadband rise | Oil temperature correlation, debris sensor spike | Multi-sensor fusion: vibration + oil + temperature |
| Tooth Fracture | GMF amplitude collapse + shaft 1× impact impulse | Highly elevated kurtosis, impulsive time waveform | Envelope detection + kurtogram analysis |
| Abrasive Wear | Gradual GMF harmonic amplitude growth across all harmonics | Oil debris particle count correlation | Wear rate modeling + debris sensor fusion |
Degradation Severity Trending: From Minor Wear to Imminent Failure
Gear Mesh Sideband Growth Rate
Key Progression Metric
The rate of sideband amplitude growth is more prognostically meaningful than absolute amplitude. AI models fit growth curves to sideband history and project time-to-threshold crossing — the point at which continued operation exceeds acceptable risk for your production context.
Kurtosis Factor (Impulsiveness)
Fault Severity Indicator
Kurtosis measures the impulsive character of the vibration signal. Healthy gearboxes show kurtosis values of 3.0 (Gaussian distribution). Localized damage such as a single pitted tooth produces kurtosis values of 6–10. Imminent fracture with tooth debris present commonly shows kurtosis exceeding 30.
Detection Lead Time vs Failure Mode
Average: 6–20 Weeks
AI detection lead time varies by failure mode. Distributed pitting — the most common gear failure mode — is detectable 12–20 weeks before functional failure. Root cracking provides 4–8 weeks. Sudden-onset scoring from lubrication failure may provide only 1–3 weeks, making continuous monitoring non-negotiable for this mode.
Diagnostic Accuracy by Method
AI Fusion: 94% Accuracy
Combining gear mesh frequency analysis with time-synchronous averaging, cepstrum analysis, and oil debris monitoring achieves 94% diagnostic accuracy in controlled validation studies — correctly classifying failure mode, severity level, and affected component for intervention planning.
Time-Synchronous Averaging: Isolating Individual Gear Contributions
In multi-stage gearboxes, vibration signals from all gear pairs superimpose on a single accelerometer output — making it difficult to isolate which gear pair is producing a fault signature. Time-synchronous averaging (TSA) synchronizes vibration data to individual shaft rotation periods and averages hundreds of revolutions, suppressing non-synchronous components and revealing the contribution of each gear pair independently. iFactory implements TSA automatically for all monitored gearbox stages.
Gear Condition Indicators (GCIs) Primary Metrics
GCIs are statistical features extracted from TSA-processed signals that quantify specific aspects of gear health. The FM4 (fourth statistical moment of the difference signal) and NA4 (normalized fourth moment) are the most diagnostically sensitive to tooth damage — FM4 tracks developing faults, NA4 is sensitive to both developing and distributed damage across all teeth.
Cepstrum Analysis for Root Crack Identification
Cepstrum analysis converts the log-magnitude spectrum to the quefrency domain, where periodic families of sidebands appear as distinct quefrency peaks at shaft rotation intervals. Root cracks, which modulate GMF amplitude and phase at the shaft rotation period, produce cepstrum peaks that are diagnostically unambiguous — not confounded by other failure modes the way GMF sidebands can be.
Envelope Analysis for Impulsive Damage
High-frequency resonance excitation by impulsive events (pitting, tooth fracture) is extracted by bandpass filtering around structural resonances (typically 5–20kHz), rectifying and low-pass filtering to extract the envelope signal, then spectral analysis of the envelope. Localized damage producing impulses at shaft rotation frequency appears clearly in the envelope spectrum.
Oil Debris Monitoring Correlation
Inline oil debris sensors counting metallic particle size distributions provide independent confirmation of wear severity. AI correlating particle count rate with vibration severity scores reduces false positive rates and improves confidence in severity classification — particles confirm material loss, vibration indicates mechanism and location.
iFactory Gearbox Monitoring Capabilities
Multi-Stage Support
Simultaneous TSA analysis for every shaft stage in a multi-stage gearbox from a single sensor set
Real-Time GMF Tracking
Gear mesh frequency tracked dynamically as speed varies — no fixed-speed assumption required for analysis
Failure Mode Classification
AI classifies detected faults by failure mode — pitting, scoring, root crack, or fracture — not just severity level
RUL Projection
Remaining useful life projections with confidence bounds updated at each measurement cycle
Gearbox Monitoring Implementation: 6 Steps
01
Gearbox Documentation Audit
Collect nameplate data and gear drawings for every monitored gearbox: number of teeth per stage, shaft speeds at rated load, and gear ratio. This data is used to calculate theoretical GMF values and fault indicator frequencies before any sensor data is collected.
02
Sensor Placement Optimization
Accelerometer placement on the gearbox housing closest to the high-speed shaft bearing provides the best signal-to-noise ratio for mesh frequency analysis. iFactory provides placement guidance for each gearbox geometry to maximize detection sensitivity across all gear stages.
03
Tachometer Signal Integration
Accurate shaft speed measurement via once-per-revolution tachometer pulses is essential for time-synchronous averaging and slip-compensated frequency analysis. iFactory supports optical, magnetic, and encoder-based tachometer inputs with automatic pulse validity checking.
04
Baseline Spectrum Establishment
Capture 2–4 weeks of baseline vibration data at representative load conditions with the gearbox confirmed healthy by inspection. The baseline GMF spectrum, kurtosis values, and GCI measurements establish normal operating bounds for each specific gearbox under its typical operating conditions.
05
Alert Threshold Configuration
Configure alert thresholds relative to each gearbox's own baseline — not industry generic values. A lightly loaded gearbox running with wide safety margins warrants different thresholds than a heavily loaded unit already near its design capacity. iFactory supports per-asset threshold configuration.
06
Maintenance Integration and RUL Planning
iFactory integrates remaining useful life projections with your planned maintenance schedule — automatically surfacing gearboxes whose projected failure dates conflict with planned production campaigns and recommending intervention windows. Book a demo to see RUL planning in action.
Frequently Asked Questions
Can AI detect gear pitting before it is visible during an inspection?
Yes — and this is precisely the diagnostic value. Micro-pitting and early-stage macro-pitting produce vibration signature changes detectable by AI well before pit sizes become visible during a borescope inspection. Waiting for visual confirmation means waiting until Stage 2 or Stage 3 damage is already well established.
How does AI distinguish gear pitting from bearing defects in the same gearbox?
Gear pitting produces fault indicators at gear mesh frequency and its sidebands (multiples of GMF ± shaft rotation frequency). Bearing defects produce indicators at bearing defect frequencies (BPFO, BPFI, BSF, FTF) which are mathematically distinct from GMF-related frequencies. AI trained on both fault families classifies the frequency structure to isolate the fault source.
What causes gear tooth scoring and is it preventable?
Gear scoring is caused by elastohydrodynamic lubrication film collapse — typically from lubricant starvation, excessive operating temperature, incompatible lubricant viscosity, or contaminated lubricant. It is entirely preventable through correct lubricant selection, oil condition monitoring, and temperature management. AI detecting early scoring signatures triggers investigation of the lubrication system, not just the gearbox.
Does AI gearbox monitoring work for planetary gearboxes?
Yes, with additional complexity. Planetary gearboxes produce more complex spectral structures because planet gears rotate and orbit simultaneously. iFactory's planetary gear analysis calculates planet pass frequency, ring gear mesh frequency, and carrier rotation frequency separately, applying the appropriate TSA synchronization reference for each component's analysis.
Turn Gearbox Vibration Data Into Failure Predictions
iFactory AI analyzes gear mesh frequency sidebands continuously — giving reliability teams weeks or months to plan targeted interventions before tooth fracture forces unplanned shutdowns.
