Blade damage is a ladder. At the bottom are surface conditions: coating loss, leading edge erosion, lightning receptor damage and gelcoat cracks. In the middle are structural conditions invisible from outside: bond line disbonds, shear web cracks, laminate wrinkles, core failures and root insert movement. At the top is a blade that ruptures or leaves the rotor. An insurer's estimate from 2014 put the world fleet at around 700,000 blades with roughly 3,800 blade failures a year, and the cost of resolving an incident in the order of one million dollars (GCube, reported by Modern Power Systems, 2014). A blade condition monitoring system therefore has to answer four questions: which rungs of the ladder it sees, what it misses, how often it looks, and what that costs in money and turbine downtime.

What does a blade inspection drone find, and what does it miss?

A blade inspection drone photographs the outer surface of all three blades, usually with the rotor stopped in one or more fixed positions so that all four faces can be imaged.

What it detects: surface damage. Leading edge erosion and its stage, coating and gelcoat cracks, lightning strike marks and receptor damage, open trailing edges, and delamination that has reached the surface. Automated blade damage detection on drone images is well studied; one open study reported damage location and type suggestions at almost human-level precision (Shihavuddin et al., 2019). What it misses: everything beneath the surface. A disbonded bond line, a spar cap wrinkle, a cracked shear web or a moving root insert gives no external sign until it cracks the shell.

Cadence and cost: the cheapest way to image the whole blade surface at inspection resolution, priced per turbine, with the rotor stopped only for the flight. A review of leading edge erosion notes that the cost of scheduling full turbine inspections leads many operators to inspect only every two to three years, so failed repairs go unnoticed until the next round (Herring et al., 2019). In our judgement annual external imaging is the minimum where erosion, lightning or a known defect is in play, and the blade inspection frequency should follow the blade type's history, not the budget.

What does an internal blade inspection robot see that a drone cannot?

An internal blade inspection robot, a crawler or camera carrier with its own lighting, enters at the root and travels along the cavity, imaging the spar caps, shear webs, bond lines and inner skin.

What it detects: the structural rungs. Starved, cracked or disbonded bond lines; shear web cracks; wrinkles where they show on the inner surface; core damage; lightning tracking inside the shell; moisture; and at the root the laminate around the inserts, where cracking or whitening shows; the inserts themselves are buried and out of sight. This is where serial manufacturing defects live.

What it misses: the outer blade, which narrows until nothing passes, and anything below the inner surface. Damage inside the laminate needs non-destructive testing: a national laboratory built a prototype external crawler with phased-array ultrasound to find subsurface damage that is not visually evident, still a prototype in 2020 (US Department of Energy, 2020). A wrinkle buried in a spar cap may show as a ripple on the inner skin, or not at all; ultrasonic testing is the usual way to confirm the ply geometry.

Cadence and cost: rotor locked, blade pitched, a technician at the root and a tether managed, so it costs more per blade than a drone flight. It is not an annual item; the checklist below sets out when it is due.

Blade sensors: strain, acceleration, acoustic emission and fibre-optic monitoring

Blade sensors turn inspection into blade structural health monitoring: they measure continuously instead of looking occasionally. The families are set out in reviews of blade damage detection (Du et al., 2020).

  • Strain sensors at the root, electrical gauges or fibre Bragg gratings read at many points along an optical fibre, measure flapwise and edgewise bending and show changes in stiffness and asymmetry between blades.
  • Accelerometers measure natural frequencies, which shift when damage changes stiffness. In our judgement the shift from early damage is small next to that from temperature or ice, so the method mainly sees large damage near the root.
  • Acoustic emission sensors listen for the ultrasonic bursts released as fibres break or a bond line advances. They detect damage as it grows, which no camera does, but the sensor must be near the source, so coverage of a long blade is partial.
  • Microphones inside the blade cavity listen to airborne sound: the change in a blade's own noise when a crack opens or a trailing edge splits, and the impact of loose material on the shell. In our experience this is the most common commercial form of blade acoustic monitoring; one microphone at the root hears further along the blade than a contact sensor, but locates the source less precisely.

Blade acoustic monitoring also has an external form: a ground or nacelle microphone listening for the change in aerodynamic noise when a crack or open trailing edge whistles. A 2025 study with a ground microphone reported test accuracy between 0.87 and 0.93 for damaged versus normal blades, and its authors state that more data from damaged blades is needed before the method generalises (Yang et al., 2025).

What sensors miss: anything slow and quiet. Erosion, coating loss and a wrinkle that has not begun to delaminate produce no signal, and a retrofit usually covers only the root.

Root and pitch monitoring

Root and pitch monitoring uses the pitch drive's torque or current, root bending moments where measured, and the relative signals between the three blades. In our judgement a loosening root or a moving insert changes one blade's behaviour relative to its neighbours before it produces a visible crack, so blade-to-blade asymmetry is the signal worth trending. Sensors are a capital item per turbine plus a data service, justified only where the blade type has a known structural weakness.

SCADA blade anomaly detection: what ten-minute data can and cannot tell you

Every turbine already records SCADA data: power, wind speed, rotor speed, pitch angle, temperatures and alarms, typically as ten-minute averages. Reviews group the approaches into trending, clustering, normal behaviour modelling, damage modelling, and assessment of alarms and expert systems (Tautz-Weinert and Watson, 2017). Normal behaviour modelling, the most used, flags drift from a model of healthy operation; a recent study on ten-minute data from five onshore wind farms found generator bearing, generator fan and rotor brush failures (Chesterman et al., 2023).

SCADA is good at the drivetrain, where a bearing warms over weeks. Blades leave fainter traces: a power curve lowered by erosion or a pitch fault, vibration made asymmetric by rotor imbalance, a pitch drive working harder on one blade. In our judgement SCADA usually says that something about the rotor has changed, rarely what or where. It decides where to send the drone or the crawler.

Lightning detection

Blade lightning protection is governed by IEC 61400-24. Lightning detection on a wind turbine registers that a strike occurred, sometimes its magnitude, and which blade took it; it does not say whether the down conductor carried the current or the laminate did. A study of 304 lightning damage cases from wind farms in the United States found the damage concentrated at the blade tip (Garolera et al., 2016). A strike registration should trigger an external inspection of that blade; the count on its own is not monitoring.

Satellite wind turbine monitoring: what Sentinel-2 and Sentinel-1 can and cannot see

Satellite wind turbine monitoring, as most often offered, uses free, open data from the Copernicus programme. Sentinel-2 is optical, with four bands at 10 m resolution and a five-day revisit for the two-satellite constellation, cloud permitting (Copernicus SentiWiki). Sentinel-1 is C-band radar, which sees through cloud and at night, at 5 by 20 m resolution in its standard mode with a six-day repeat from two satellites (ESA).

A dense Sentinel-1 time series has followed deployment, vessel interactions and operational events at 15,606 offshore wind locations (Hoeser et al., 2026).

What satellites can see, given a time series: whether a rotor is present or a turbine has become a bare tower; whether a jack-up vessel or large crane is beside it, which signals a major component exchange; long outages; and construction or decommissioning progress.

What the free data cannot see: blade damage of any kind, whether the turbine is generating, and not reliably rotation or pitch angle. At 10 m per pixel a 60 metre blade is six pixels long and a crack a centimetre wide is a thousandth of a pixel. Commercial optical satellites image far more finely, at a price per scene, and in our judgement finely enough to show a missing blade, a blade on the ground or a rotor parked for weeks, but never erosion, cracks or disbonds. Satellite blade monitoring is therefore site monitoring, not blade condition monitoring.

Cadence and cost: the Copernicus imagery is free and the analysis is the cost; its place is portfolio surveillance and a dated public record of when a rotor stopped or a crane arrived.

How should an owner combine wind turbine blade monitoring systems?

  • Continuous and free: SCADA screening for power curve, imbalance and pitch anomalies; lightning strike registration; a satellite time series for rotor presence on remote sites.
  • Annual, or set by risk: drone imaging of every blade, flown and classified the same way each time, with the blade inspection frequency written into the operations plan per blade type.
  • Triggered: external inspection of any blade after a lightning registration, a SCADA rotor alarm, or an unexpected stoppage seen from orbit.
  • Event-driven and internal: crawler inspection at end of warranty, before purchase or refinancing, after a failure of the same blade type anywhere, and on a defined sample when a serial defect is suspected.
  • Continuous and structural: blade sensors where the blade type has a known root, bond line or web weakness.

The two mistakes we see most often are inspecting on the budget's schedule rather than the blade's, and treating a report's damage category as fact rather than one inspector's judgement. Both are cured by teaching the people who read the reports enough to challenge them.

Frequently asked questions

How often should wind turbine blades be inspected?

Many operators inspect only every two to three years. Our view is that annual external imaging is the minimum where erosion, lightning or a known defect is present, and that internal inspection belongs at end of warranty, at a transaction, and after a failure of the same blade type elsewhere.

Can SCADA data detect blade damage?

Sometimes, and late. SCADA blade anomaly detection can show a lowered power curve, a rotor imbalance or a pitch drive working harder on one blade. It will not identify a bond line disbond or a wrinkle before the blade's behaviour has changed.

Can satellites monitor wind turbine blades?

Not their condition. At the 10 m resolution of the free Copernicus data a turbine is a few pixels and a blade defect far smaller than one; commercial higher-resolution imagery can show a missing blade or a rotor parked for weeks, but not erosion, cracks or disbonds. A time series can show whether a rotor is present, whether a vessel or crane is at the turbine, and whether a site has been idle. That is site monitoring, not wind turbine blade monitoring.

Do blade sensors replace inspection?

No. Sensors catch fast structural failure modes near where they are fitted, usually the root. They do not see erosion, coating loss, most surface cracks or the unmonitored outer blade, so a fleet with a known structural weakness needs both.

What does an internal blade inspection robot detect?

The structural interior: bond lines, shear webs, the core, lightning tracking inside the shell, and the root laminate around the inserts. It cannot reach the narrow outer blade or see inside the laminate, so a suspected wrinkle or subsurface disbond still needs ultrasonic testing.

How Apex Wind can help

We are an independent blade engineering consultancy: Apex Wind sells no inspection services, sensors, software or monitoring products and takes no referral fees. For disclosure, our founder separately co-founded a company that builds crawler robots for internal blade inspection. In a transaction, our blade technical due diligence reads the inspection and monitoring record for which methods were used and what they could not have seen. For teams that commission inspections and read the reports, our blade training covers how blades are built, how they fail and how to challenge a classification. If you are deciding what to monitor on a specific fleet, contact us.