Main-bearing condition monitoring follows the slow rotating shaft of a wind turbine under complex, changing loads. Temperature, vibration and acoustic measurements can raise concerns, but do not themselves establish the physical damage or remaining life of a bearing.
The gold standard combines calibrated, speed-aware signals with independent fault evidence, appropriate operating comparisons and explicit uncertainty.
eduKateSG · Wind Turbine Main Bearing Condition Monitoring
Find your next learning step
Calculate slow-shaft and bearing frequencies, evaluate signals and test whether a model’s output is a credible diagnosis.
Full chapter index · Practice and transfer · Education root · How X Works
Full chapter index
Identify the component · Chapters 1–2
Calculate the indicators · Chapters 3–4
Select the observations · Chapters 5–7
Evaluate a fault claim · Chapters 8–10
Follow scientific sources · Chapters 11–12
CHAPTER 1 OF 12 · Identify the component
1. The main bearing supports the slow turbine rotor shaft
The main bearing is a mechanical interface between the rotating rotor and the supporting turbine structure. Its job includes allowing the shaft to turn while passing substantial non-rotational forces into the machine frame. This differs from a high-speed generator bearing and from bearings located inside a gearbox.
Main-bearing architectures and access arrangements differ across turbine designs. Their sensors may be installed on housings, shafts or nearby components and consequently measure different responses to the same physical event.
The 2026 Wind Energy Science drivetrain-monitoring review describes main-bearing condition monitoring as difficult because loading is unsteady, rotation is comparatively slow and repair consequences can be large.
An elevated vibration or temperature reading is an observation, not automatically a damaged bearing. The gold standard starts with component identity, loading regime, valid instruments, measurement method and independent evidence.
This article offers educational analysis only. Decisions about safe continued operation, inspection or bearing replacement require qualified engineering and the machine’s approved procedures.
CHAPTER 2 OF 12 · Identify the component
2. Multiple load and damage mechanisms can overlap
Wind loads on a main shaft are variable. Rotor torque, thrust, aerodynamic imbalance, wind shear, turbulence, gravitational effects and structural vibration may influence what its supporting bearings experience.
Possible damage mechanisms involve rolling-contact fatigue, contamination, lubrication breakdown, assembly or fit problems and abnormal loading. Their symptoms can overlap with changes caused by operating conditions or sensors.
The WES review distinguishes acoustic, vibration, strain, thermal and SCADA measurements and discusses how different fault mechanisms may be difficult to identify from one indicator alone.
A variation in temperature could reflect warmer ambient air or cooling, while a vibration burst could be transmitted from an unrelated drivetrain stage. Two sensors responding to the same gust are not automatically independent evidence of damage.
A diagnostic plan should predict which observables would change under each competing explanation and seek a suitably independent means of checking the proposed cause.
Maintain the turbine, bearing, sensor and intervention histories so a signal before a maintenance event is not treated as directly comparable to a changed configuration afterward.
CHAPTER 3 OF 12 · Calculate the indicators
3. Worked example: rpm and the slow rotation signal
Take an imaginary main shaft operating at 12 revolutions per minute. Its rotation frequency is f_rot = 12/60 = 0.20 Hz, meaning 0.20 cycles per second.
One revolution takes 1/0.20 = 5 seconds. A simple once-per-revolution signal would repeat every five seconds under this steady hypothetical operating condition.
If the shaft increased to 18 rpm, f_rot = 18/60 = 0.30 Hz and the corresponding period becomes approximately 3.33 seconds.
These changes arise from normal speed differences and do not prove any new mechanical fault. Bearing-related features, rotor imbalance and other structural responses can occur near rotation-related frequencies.
Coarse, ten-minute SCADA averages can describe broad speed and temperature context but may not retain individual transient vibration cycles, making sensor sampling a separate design question.
The calculation shows why a frequency must be reported with a shaft-speed reference and operating state before any fault label is considered.
CHAPTER 4 OF 12 · Calculate the indicators
4. Worked example: outer-race frequency from bearing geometry
For a simplified rolling-element bearing, the outer-race ball-pass frequency is often represented as f_BPFO = (N/2)f_rot[1−(d/D)cosθ], where N is number of rolling elements, d their diameter, D pitch diameter and θ contact angle.
Take a fictional bearing with N = 12, d/D = 0.15, θ = 0° and main-shaft f_rot = 0.20 Hz. The geometry factor in brackets is 1−0.15 = 0.85.
Calculate f_BPFO = 6×0.20×0.85 = 1.02 Hz. Relative to 0.20-Hz shaft rotation, the theoretical order is 1.02/0.20 = 5.1.
That result is a predicted characteristic periodicity, not measured evidence of an outer-race fault. Load, sliding, geometry and speed variation can shift real observed components.
A signal near 1.02 Hz might also be produced by another machine interaction. Appropriate detection requires the documented bearing, valid sampling and corroborating physical evidence.
Never copy the fictional bearing geometry into another drivetrain and assume its predicted fault frequencies remain applicable.
Suppose a fictional calibrated accelerometer collected four evenly weighted acceleration samples [0.4, −0.4, 0.8, −0.8] in m/s² during a defined observation window.
RMS acceleration is the square root of mean squared signal: sqrt[(0.16+0.16+0.64+0.64)/4] = sqrt(0.40) ≈ 0.632 m/s².
The arithmetic mean is zero, but the oscillating signal has nonzero magnitude. RMS is a convenient summary but does not by itself specify the source of the oscillations.
If shaft load or wind turbulence changed, broadband RMS could increase without a bearing defect. Even at stable load, sensor attachment or filter settings may alter its value.
Record sampling frequency, bandwidth, sensor orientation, calibration and machine operating state when making comparisons.
A fault-specific analysis may need frequency bands or modulation indicators in addition to RMS. The fictional result is not a safety acceptance limit.
CHAPTER 6 OF 12 · Select the observations
6. Thermal response can mask timing and confound causes
Main bearing temperature responds to heat from mechanical friction, surrounding components, lubrication and the environment. Large structures have thermal inertia, so the signal may lag a sudden change in physical condition.
Imagine an invented reading of 54°C with ambient 19°C, a 35°C difference. On a warmer day the bearing reads 59°C with ambient 24°C, still a 35°C difference.
The example shows why a five-degree increase in absolute bearing temperature does not necessarily mean increased deterioration. It does not prove the bearing is healthy because loads, lubrication and thermal paths could also have changed.
A 2025 study integrating control charts and machine learning reported limitations of fixed temperature alarms for early main-bearing degradation because thermal changes may develop slowly.
Baseline models should use suitable operating context rather than comparing startup measurements with long periods of steady operation indiscriminately.
Temperature evidence is one component of monitoring, not a direct measure of the shape or size of a bearing crack.
CHAPTER 7 OF 12 · Select the observations
7. Vibration and acoustic emission capture different physical responses
Accelerometers detect mechanical vibration transmitted to their mounting points; acoustic-emission sensors may detect short transient elastic signals associated with cracking or frictional processes. The two methods use different bandwidths and have distinct sources of noise.
Envelope analysis can help expose repeated impact-type modulation in a chosen vibration band. But a main bearing is near a larger mechanical assembly that may transmit gearbox, generator and structural vibration.
The 2026 WES review discusses time-domain RMS and kurtosis, spectral and envelope methods, and time-frequency techniques for nonstationary bearing faults.
Rotational-speed variation can smear fixed-frequency components, making order tracking or other speed-aware techniques potentially useful when the measurement conditions support them.
A peaked signal or high kurtosis can also arise from noise, mounting changes or a non-bearing transient. A candidate periodicity becomes a diagnosis only when credible physical reference evidence supports it.
Prefer complementary measurements selected for their physical sensitivity, not just the greatest number of algorithm-generated scores.
CHAPTER 8 OF 12 · Evaluate a fault claim
8. SCADA and fast sensors need a shared operating timeline
SCADA frequently records average power, speed, temperature and status on a coarser time scale than high-frequency vibration or acoustic emission acquisitions. Each has value, but their sampling rates must not be confused.
A large signal-processing model cannot reconstruct bearing impacts never recorded by its acquisition bandwidth. Sensor placement, aliasing, timestamps and calibration are therefore necessary parts of the result.
The 2024 Fraunhofer wind turbine dataset provides a research example combining accelerometers, tachometer and weather records under known fault scenarios. Its small-machine observations do not certify performance on a large offshore bearing.
When aligning records, check exact acquisition intervals and operating states. A ten-minute temperature statistic and a vibration window recorded during a different load event may not describe the same condition.
Label shutdown, servicing, sensor replacement, communication gaps and abnormal environmental states. An apparent trend crossing a change in instrument configuration may not reflect physical deterioration.
Keep raw observations alongside derived features so that another analyst can reproduce both the baseline and the anomaly.
CHAPTER 9 OF 12 · Evaluate a fault claim
9. Lubrication evidence and physical inspection can help attribution
Lubrication and contamination can influence rolling-contact conditions and bearing reliability. Depending on the bearing design, qualified lubricant analysis or inspection can provide more direct information about particular failure mechanisms than a temperature trend.
However, an oil or grease sample represents its sampling location and method. A detected particle does not automatically identify which exact rolling contact generated it.
Maintenance events can also change temperature and vibration baselines. Record repairs, lubrication interventions and sensor calibrations to prevent mistaking an administrative change for sudden bearing recovery.
The drivetrain review describes many potential root causes of premature main-bearing failures and notes that their relative importance is difficult to establish in real conditions.
Combine independent observations only when they genuinely provide separate evidence; correlated measurements of the same wind or load can create the illusion of corroboration.
The scientific conclusion should separate a proposed cause, measured data and what independent inspection or analysis would be necessary to confirm it.
CHAPTER 10 OF 12 · Evaluate a fault claim
10. Validation and remaining life are separate inference problems
Detecting a difference from a healthy baseline does not prove a mechanical failure. Even a confirmed fault does not automatically establish when the bearing might become unusable, since progression depends on mechanism, loading and intervention history.
Suppose a fictional 120-case independent evaluation contains 20 fault cases and 100 nonfault cases. A monitor correctly flags 16 faults, misses four, raises ten false alerts and correctly rejects ninety nonfault cases.
Recall is 16/20 = 80%, precision is 16/(16+10) ≈ 61.5% and false-positive rate is 10/100 = 10%. The values are invented and cannot certify a diagnostic system.
Consecutive recordings from one degrading machine are highly related. Splitting them randomly between training and test sets can exaggerate generalization. Independent fault events, machines or time blocks provide a stronger challenge.
Remaining useful life models additionally need trustworthy event times and treatment of censored cases where failure never occurred during observation. Presenting a precise countdown without these limits misrepresents uncertainty.
The gold standard names each stage—observation, anomaly, proposed defect, independently confirmed damage and life estimate—rather than collapsing them into a single health percentage.
CHAPTER 11 OF 12 · Follow scientific sources
11. Practice, FAQs and linked drivetrain reading
Practice A: For 12 rpm, compute 0.20 Hz and five seconds per revolution. Why can a one-revolution signal have a cause outside the bearing?
Practice B: For 12 elements, d/D = 0.15, θ = 0 and shaft 0.20 Hz, confirm BPFO = 1.02 Hz and order 5.1.
Practice C: The four fictional samples [0.4, −0.4, 0.8, −0.8] m/s² give RMS ≈ 0.632 m/s². State why a bigger RMS is not a unique fault label.
Practice D: With 16 hits, 4 misses and 10 false alerts, calculate 80% recall and approximately 61.5% precision.
FAQ — Is main-bearing monitoring interchangeable with generator-bearing temperature diagnostics? No. Component loads and measurement physics differ.
FAQ — Is normal temperature evidence of no mechanical fault? No. Thermal response may lag and some faults have weak thermal effects.
FAQ — Can one predicted bearing frequency authorize replacement? No. Qualified inspection and appropriate corroboration are needed.
Transfer the analysis to a larger turbine with different bearings and supports. Redefine expected frequencies, sensor placement, load distributions and independent fault evidence.
Verified connected eduKateSG roots for distinct mechanical and statistical questions:
- The Gold Standard of Wind Energy — main drivetrain and loading
- The Gold Standard of Wind Turbine Gearbox Vibration Monitoring — separate gear-and-bearing signatures
- The Gold Standard of Wind Turbine Generator Bearing Temperature Diagnostics — different thermal sensor and component
- The Gold Standard of Wind Turbine Blade Condition Monitoring — aerodynamic and composite structure
- The Gold Standard of Wind Turbine Power Curves — operating-state context
- The Gold Standard of Physics — mechanics and oscillations
- The Gold Standard of Statistics — model testing and uncertain fault labels
CHAPTER 12 OF 12 · Follow scientific sources
12. Primary evidence and next study plan
The 2026 Wind Energy Science drivetrain review surveys main-bearing difficulties, sensors and uncertain remaining life. Research from Fraunhofer and other laboratories illustrates multisensor studies and the limits of scaling one test context to another.
All rotations, accelerations, temperatures and fault scores in this guide are fabricated classroom cases. They are not diagnostic conclusions, component acceptance criteria or permission for maintenance.
An audit-ready project should archive bearing identity, speed, loading, vibration and thermal data, instrument calibrations, feature definitions, event labels, sensor changes, independent inspections and model uncertainties.
Primary peer-reviewed and institutional sources for monitoring and validation:
- Wind Energy Science — Drivetrain Condition Monitoring (2026) — main-bearing sensing and reliability context
- Springer — Control Chart Integrated Fault Detection (2025) — temperature alarms and models
- Scientific Data — Fraunhofer Turbine Monitoring Dataset (2024) — independently labelled experimental cases
- Mechanical Systems and Signal Processing — Main-Bearing Test Rig (2025) — scaled measurement limitations
- TU Delft — Physics-Based SCADA Main Bearing Modelling — thermal-baseline method
- Renewable and Sustainable Energy Reviews — Condition-Based Maintenance — system-level diagnostic and maintenance scope
The immutable Clementi Mathematics instructional reference shows the teaching approach; it is not an engineering inspection standard.