Measuring blade aerodynamic flow state: what attached and detached flow cost in power

Why is the blade aerodynamic flow state invisible to SCADA?

Below rated wind speed, power depends on the power coefficient, which the controller sets through tip speed ratio and pitch. The standard strategy holds pitch fixed and varies rotor speed to keep the tip speed ratio at its optimum, yet the paper notes that the ideal settings can differ from the manufacturer's because of local aerodynamic effects, and drift with blade wear and weather (Voisin et al., 2026).

What the controller cannot see is the boundary layer. In attached flow the air follows the suction side and lift grows with angle of attack. As the angle rises, the flow separation wind turbine blade sections undergo starts at the trailing edge and moves forward; past the stall angle lift collapses, drag rises and the load fluctuates. A power reading shows the consequence, not the cause, and the paper's review states that nacelle anemometry cannot tell whether the flow on the blade is attached or detached.

SCADA also hides aerodynamic imbalance. The paper cites a laser-optical survey that found pitch misalignment on 38% of 195 turbines, and summarises the cost: lower production, more vibration, higher loads and shorter life, with even a small misalignment on one blade raising main shaft loading. None of this raises an alarm.

What does a blade surface flow sensor measure?

The tell-tale sensor wind turbine engineers may know from sailing is a strip of yarn on a sail: it streams in attached flow and flutters when the flow separates. The instrument in the paper, which the authors call the eTellTale or eTT sensor, is an electronic version: a silicone strip glued to the suction side, its movement read magnetically through a Hall-effect sensor at 200 Hz. The statistic used is the standard deviation of that signal, low when the strip lies quietly in attached flow and high when separated flow agitates it.

The paper's review is candid about an earlier strain-gauge version of the sensor: mechanically fragile, drifting over time and with temperature, and unable to give an absolute measurement. The method here relies on the signal's statistics, with thresholds set in a wind tunnel.

How was it calibrated for angle of attack and lift?

Calibration took place in the high-speed wind tunnel at CSTB Nantes on a full-scale aerofoil section scanned from a 2 MW turbine blade at 80% span, with pressure taps measuring the aerodynamic force while four sensors at 15% chord recorded their signals. From that data the authors set the attached and detached thresholds and trained a machine learning model linking the sensor statistics and wind speed to lift coefficient (the normal force coefficient, CN) and angle of attack, over -5° to 22° in a reduced model and to 30° in a larger one, at 20 to 40 m/s (Voisin et al., 2026). That turns a stall indicator into angle of attack measurement blade by blade in the field, though the estimates remain model outputs, trained on a two-dimensional section in low-turbulence tunnel flow and applied to a rotating blade in turbulent wind.

How was the field test on two V27 turbines set up?

The campaign ran at the SWiFT facility of Sandia National Laboratories in Texas, on two Vestas V27 turbines of 225 kW named A1 and A2, with support from TotalEnergies' R&D department as the paper acknowledges. Five sensors were glued to the suction side of each blade at radii of 6 to 10 m, alternating between 15% of chord from the trailing edge and a station further forward, and are named by radius, R6 or R9; a mast supplied wind speed at 31 m and pressure, temperature and humidity at 27 m. About 20 days of testing between August 2024 and January 2025 gave 1,545 hours of data on A1 and 610 on A2. The reference pitch is 1°, equivalent to the 0° reference of the standard controller; the pitch offset wind turbine trials covered -3° to +5° on A1 and only 0° and +1° on A2.

Weather correction and the power curve flow state split

Generator power was corrected to a standard air density of 1.225 kg/m³ from the mast data. Then, within each wind speed bin, the samples were separated by whether the sensor at a chosen radius reported attached or detached flow, giving three curves: all data, attached only and detached only. A power loss criterion expresses the gap between the last two as a share of operating time and as potential production.

What did the power curve split show about attached and detached flow?

The headline figures are those of the abstract: production with attached flow was 15% above average, production with detached flow was 30% below, and 15% of potential power was lost during 33% of operating time in low and medium wind (Voisin et al., 2026).

The attached flow detached flow power figures for A1 at the +1° reference pitch, split by the state at the 9 m sensor, are tabulated in the paper as:

For A2 the gains at the same wind speeds were 32%, 18% and 3%. Above about 7.5 m/s the curves converged and there was no further gain at fixed pitch; the paper describes R9 as entering a chaotic regime it calls "hard stall", at the wind speed where the V27's power coefficient peaks.

The wind turbine power loss flow separation was causing showed most clearly between the two machines. Nominally identical, A2 out-produced A1 by 41% at 4.5 m/s, 52% at 5.5 m/s and 29% at 6 m/s; restricted to attached flow at R9, A1's curve reached A2's, because A1's R9 was mostly detached below 6.5 m/s while A2's stayed attached. The best production came where R6 was stalled and R9 attached, short of hard stall. The paper also records that A1's pitch wandered between 0.5° and 1.5° while A2's stayed within 0.97° to 1.03°, which could explain part of the gap.

What did the pitch offset trials and the trim recommendation show?

Of the offsets tested on A1, -3° gave the highest production: 42.1 kW against 20.8 kW at +1° at 5.4 m/s, 62.3 kW against 46.2 kW at 6.8 m/s, and 7% and 12% more at 8.1 and 8.6 m/s. At -3° the rotor reached its rated 43 rpm at about 6 m/s, against about 10 m/s at +1°, which the authors read as consistent with the earlier V27 literature they cite.

The trim method is the part with the most direct application. Using the tunnel-trained model, the authors estimated angle of attack at the 6 m radius on each blade in 1 m/s wind speed bins and each blade's deviation from the three-blade average. The recommendation the paper states is a pitch offset per blade, as a function of wind speed, that brings every blade to the average angle of attack in that bin and so removes aerodynamic imbalance. The spreads were not small: on A2 between 4 and 5 m/s, blade A sat 2.47° above the average and blade C 2.35° below; on A1 between 5 and 6 m/s the extremes were +1.39° and -1.32°. The conclusions list what the method can address: pitch misalignment identification through angle of attack analysis, the static pitch that captures the optimum angle, and, as future work, dynamic pitch control driven by the sensor statistics at the 15% and 45% chord stations (Voisin et al., 2026).

What does this imply for pitch calibration, blade differences, erosion and control?

The following are our reading of the results, not findings of the paper.

Pitch calibration. Misalignment is normally checked geometrically; the paper's approach checks the aerodynamic outcome, the angle of attack each blade actually experiences at a given wind speed. A geometrically aligned blade can still differ from its neighbours through twist tolerance, surface condition or a pitch system that hunts around the set point, as A1's did, and a trim that varies with wind speed catches all three.

Blade and turbine differences. Two turbines of one type on one site differed by 29 to 52% in low wind, and the flow state explained most of it. A power curve deficit has many possible causes; blade aerodynamic performance monitoring on a sample of turbines can separate an aerodynamic cause from a control or sensor cause before a dispute or a transaction.

Erosion and soiling. The paper does not test a damaged blade, but it states that the best settings change with blade wear. In our experience, leading edge roughness brings separation forward and lowers the stall angle, so an eroded blade at the same pitch spends more time in detached flow. A flow-state measurement would show that shift directly, and how much of the production lost to leading edge erosion a pitch trim could recover.

Real-time control. The paper proposes tracking the optimum angle of attack rather than holding a fixed pitch. That is a manufacturer's control problem and remains untested; what an owner can use sooner is the diagnostic side, the split power curve and the per-blade angle of attack.

What are the limits of the study?

The results rest on two 225 kW research turbines at one site, about 20 days of testing, one sensor type, and far fewer hours and only two pitch settings on A2. The relative gains at 5 m/s are large because the absolute power there is small, around 20 to 40 kW, and the headline percentages refer to average production in low and medium wind during the test, not to annual energy. The -3° result is specific to these turbines and is not a recommendation for any other machine. The angle of attack figures come from a tunnel-trained model, the code is confidential and the data are under a non-disclosure agreement. The competing-interests statement discloses that two authors are board members of the company behind the sensor and one is the inventor of the related patent. Mohammed Fajar, as the author contributions statement records, assisted with the analysis and, like all co-authors, reviewed and edited the paper.

Frequently asked questions

What is the blade aerodynamic flow state?

It is the condition of the boundary layer on the blade surface, chiefly the suction side: attached, where the air follows the surface and lift rises with angle; separating, where flow detaches from the trailing edge forwards; and stalled, where lift collapses. The blade aerodynamic flow state sets the power a section produces at a given wind speed and pitch, and SCADA does not record it.

How much power does flow separation cost a wind turbine?

On the two V27 turbines in the Wind Energy Science paper, production with attached flow was 15% above average and with detached flow 30% below, and 15% of potential power was lost during 33% of operating time in low and medium wind. Those figures come from two small turbines over about 20 days, not from a fleet.

Can the method detect pitch misalignment?

The paper reports that it can, by estimating angle of attack on each blade at a fixed radius across wind speed bins, which yields a per-blade pitch offset as a function of wind speed. The estimate depends on a tunnel-calibrated model, so it complements a geometric pitch check rather than replacing it.

Does Apex Wind sell or recommend flow sensors?

No. Apex Wind sells no sensors, hardware, repairs or monitoring products and takes no referral fees. Our founder is a co-author of the paper; our interest is in what the method tells an owner about a rotor, not in any particular instrument.

How Apex Wind can help

We are an independent blade engineering consultancy, and our reading of a performance question starts from the blade rather than the SCADA screen. In a transaction or a performance dispute, our blade technical due diligence examines whether a power deficit has an aerodynamic cause, whether pitch calibration and blade condition have been checked, and what evidence a figure needs before it enters the financial model. For engineering, operations and asset management teams, our blade training covers how a blade section produces lift, why it stalls, and how erosion, soiling and pitch offsets change the answer. To explain an underperforming rotor, or to decide what to measure on it, contact us.