The experiment consisted of flipping the rotor of the 1.5 MW wind turbine from upwind to downwind. This was achieved by pitching the blades 180° and yawing the nacelle 180°. To an observer located upwind of the turbine, the rotor continued to spin clockwise and the pitch rate of the three blades was in the same direction. However, in the downwind configuration, the blades flew behind the tower and experienced a strong wake deficit induced by the presence of the tower. The experiment was motivated by the need to quantify the increase in fatigue loads and acoustic emissions of the downwind rotors compared to the upwind rotors, since the downwind rotor configuration is proposed for very large and very flexible wind turbines .

The downwind orientation of the rotor represents an additional challenge for the pressure belts, which were used to characterize the effect of higher-frequency step changes in loading during blade passage. Many of the findings discussed in this work also apply to conventional wind turbines with upwind rotors.

The aeroelastic model of the turbine is available in NREL's aero-servo-elastic solver OpenFAST. The simulations presented in this study were run with OpenFAST v3.3.0 , which was coupled to the industrial controller provided by GE Vernova in the form of a compiled dynamic link library. Both polars – in the form of lift, drag, and moment coefficients – and coordinates of the airfoils adopted along the span of the blades are known. The aeroelastic model uses a simple Euler–Bernoulli beam formulation for blades and towers and lifting lines for blades that include the effects of unsteady airfoil aerodynamics. Unsteady airfoil aerodynamics are simulated with the Beddoes–Leishman formulation including the Minnema–Pierce modifications . The tower wake is modeled with a standard Powles shadow model , which is implemented in OpenFAST as described in .

The coordinates of the airfoils were fed to the EllipSys2D two-dimensional (2D) fluid dynamics solver , which generated lift and drag coefficients as a function of angle of attack for both clean and fully turbulent conditions. To capture transitional boundary layers, the eN transition model is used . A satisfactory match was found between these predictions and the airfoil coefficients shared by GE Vernova within the aeroelastic model. These predictions were later compared with the experimental measurements (see Sect. ).

The extruded belt, with cross section shown in Fig. , is made of 32 channels, each with a diameter of 0.8 mm. The belt is 100 mm wide and has a maximum thickness of 1.8 mm. A pressure tap hole is drilled into each channel at a specified chord position, which is chosen to maximize the resolution of the pressure distribution and to minimize the disturbance among the pressure taps. The positioning of the taps is discussed in Sect.  and shown in the Figs.  and . Each blade belt is connected to two 16-channel miniature pressure scanners via a printed fixture. The pressure scanners, with a size of 37 mm × 32.4 mm × 9.2 mm, are operated in “calculated differential mode”, where one of the ports is used as reference and connected to the static port of the five hole probe at this radial position. The full-scale range is -400 to 200 mbar with a resolution of 0.03 mbar. The scanners are attached to the blade surface close to the trailing edge and connected with a signal cable to the data acquisition system in the flyboard. Further details of the instrumentation can be found in .

The two blade belts were designed to be installed at distances from the blade root of 18.90 and 31.45 m, where r/R = 0.51 and 0.84, the chord 1.91 and 0.96 m, and the relative thickness 25 % and 18 %. The two locations respectively correspond to 18.35 and 5.80 m from blade tip. The locations, at mid-span and toward the tip of the blade, were chosen to focus on a simpler case where flow is assumed to be close to 2D. Future experiments will be able to focus on more challenging blade regions, such as closer to the root and tip, where flow is known to be more three-dimensional (3D) . At the two locations, the chords are known and the airfoils correspond to two master airfoils of known coordinates and performance. The length of each belt was set to ensure that the ends of the belt reached the trailing edge on both the suction and pressure sides.

The belts on the tower were installed at a height of 46.7 m from the base of the tower. The height was chosen because it corresponds to the tower path of the outer belt on the blade in an undeflected turbine configuration. To cover the entire circumference of the tower, two belts were wrapped around the tower, each made of two 32 channel segments of a length of 9.7 m. The two segments were vertically shifted by one belt width (see the photo on the right in Fig. ) to allow the connection of a 64-channel pressure scanner, which was installed in an almost airtight box attached to the tower opposite the prevailing wind direction at a location approximately 1 m above the pressure belts. The scanner is the Scannivalve MPS4264 electronic pressure scanning module with a full range of 4 inH2O (0.145 Psi) and with the static port open inside the box with the scanner. Unlike the belts on the blades, the positions of the taps on the tower were uniformly distributed, except for a few faulty taps not shown (Fig. ).

The pressure tap (also known as port or orifice) layout used at both belt locations was based on a cosine spacing law. This arrangement produced higher port density at the leading edge and sparser port density at the trailing edge of both pressure and suction surfaces. The spanwise spacing of ports minimized upstream ports tripping the boundary layer for downstream ports, though some degree of disruption of the airfoil boundary layer might be expected in some conditions from the spanwise-running ramps at the edges of the pressure belt (cf. ). The port layout is shown in Figs.  and . Both the (a) chord positions (Figs. a and a) and (b) span positions (Figs. b and b) of the ports are presented. The span position plots show the pressure belt from above as wrapped around the leading edge at sarc≈ 2 m and sarc≈ 1 m for the two stations.

Note the spanwise layout axes are not to scale. The red vectors are 15° wedge angles from vertical to assess downstream transition spreading. Additionally, pressure belt tubing connectors interfered with optimal port locations near the trailing edge. These interference areas are depicted as shaded red trapezoids in Figs. b and b.

In the authors initially developed an optimization routine to minimize the number of pressure ports needed to accurately measure lift coefficient. This work showed that 32 pressure ports with cosine spacing already had a lift coefficient error below 0.01. Therefore, there was not a significant advantage to using an optimized layout.

Due to the long blade sections and large tower diameter of the 1.5 MW turbine, the pneumatic pressure belts used in this experiment included extended runs of small-diameter embedded tubing that connected surface taps to miniature pressure scanners. Such configurations are known to introduce dynamic distortions in unsteady pressure measurements, including signal attenuation, resonance amplification, and phase delays. These distortions result from viscous friction and acoustic wave propagation within the tubing and are a function of frequency . To ensure accurate pressure signals (both in magnitude and time), which are necessary for analyzing unsteady aerodynamic phenomena such as blade/tower interaction, it was essential to characterize and compensate for each port's dynamic response.

Following the full-scale measurement campaign, the pressure belts were removed from the turbine and transferred to a laboratory environment for post-test characterization of the channel connecting the taps to the transducers. Each pressure channel was individually characterized using an in situ tubing characterization device, which generated a rapid pressure pulse at the belt inlet (e.g., the surface pressure). The resulting pressure response at the scanner end of the tubing (e.g., the remote-raw pressure) was recorded simultaneously with the surface pressure, thereby enabling quantification of the system's frequency-domain transfer function. The tubing system geometry was then modeled, and theoretical transfer functions (e.g., the system response) were generated and compared with experimental data to accurately capture variations in the as-built geometry . In practice, reconstruction of the surface pressure requires an inverse process; the remote-raw pressure is used to estimate the surface pressure. A Wiener-filtered inverse model of the system response was developed based on the as-built tubing system geometry and compared with each port's measured reference data, following the methodology of . The filtering is necessary to avoid amplifying noise present in the remote-raw pressure when reconstructing the surface pressure. These models facilitated the derivation of optimal reconstruction filters capable of suppressing noise amplification while accurately restoring phase and amplitude across the relevant frequency range.

Figure  presents an example of the dynamic characterization of one representative pressure port. The left panel shows the recorded pressure signals in the time domain, clearly illustrating the distortions induced by the tubing system. The surface pressure exhibits a sharp high-frequency pressure pulse, whereas the remote-raw pressure is significantly delayed and attenuated, showing resonant behavior (overshoot) caused by acoustic and frictional effects within the tubing. The Wiener-filtered reconstruction produces the remote-reconstructed pressure that demonstrates a significantly improved comparison to the surface reference input, effectively correcting both amplitude attenuation and phase delay. The frequency-domain plot (right panel) confirms these observations: the pressure pulse applied to the surface has frequency content from low frequency out past 500 Hz, the remote-raw pressure is significantly attenuated at all frequencies greater than 10 Hz, and the remote-reconstructed pressure accurately restores the broadband frequency content up to 400 Hz. The ability to reconstruct pressure signals with frequency content up to 400 Hz ensures the ability to resolve unsteady aerodynamic phenomena relevant to blade/tower interactions.

It is important to note that the tubing characterization step was conducted at a significantly higher sampling frequency than that used during field testing to ensure increased fidelity of the modeling approach. However, the actual field measurements were acquired at a sampling frequency limited to 100 Hz; consequently, the maximum reconstruction frequency of the test data will be constrained by this sampling frequency bandwidth.

The reconstructed pressure signals were compared against the known input signals from the calibration procedure in both time and frequency domains. For each port, the reconstruction was evaluated based on its ability to recover the sharp pressure rise and broadband content of the calibration impulse. Overall, the reconstruction technique demonstrated effective recovery of high-frequency pressure content up to approximately 20 Hz across many channels, which is critical for resolving unsteady aerodynamic features in the current field test campaign.

The validated reconstruction models were then applied to the raw pressure data acquired during the two downwind test campaigns. The reconstruction was performed as a postprocessing step, channel by channel, using the individualized Wiener-filtered system response model developed for each port. The corrected pressure signals provide a better representation of the unsteady surface pressures on the blade and tower, with significantly reduced influence from tubing-induced distortions through the reconstruction process. These reconstructed datasets form the basis for the aerodynamic analysis and comparisons with numerical predictions discussed in subsequent sections.

The flyboards, one adjacent to each belt on the blade, consist of a metal frame with a plastic cover that aims to minimize impact on the surrounding airflow. The dimension of the flyboards is approximately 30 cm in spanwise and chordwise directions and about 10 cm in height; see Fig. . They contain the data acquisition unit, WiFi, and batteries (see the middle photo in Fig. ). Each flyboard also carries the five-hole probe, which has a length of 0.42 m. The data acquisition system samples the data from the pressure scanners on the blade, the five-hole probe, and a three-axis accelerometer in the flyboard at a rate of 100 Hz; the data are GPS time-stamped. The WiFi antenna transmits data in real time to the ground for visualization and postprocessing. Note that if the WiFi malfunctions, the data are also stored in the data acquisition system housed on the flyboard and can be retrieved physically by accessing the flyboard. To maximize battery charge, the data acquisition and WiFi system has a start/stop function that can be controlled remotely, so the system only operates when measurements are ongoing.

A third data acquisition system is installed at the bottom of the tower and samples the 64 channels from the pressure scanner connected to the pressure belts on the tower at 100 Hz. In postprocessing the data are aligned with the blade pressure data based on the GPS timestamp.

The five-hole probes, manufactured by Aeroprobe Inc., have a diameter of 6.35 mm and a spherical head. The pressure ports and the probe coordinate system are shown in Fig. . Two different probe coordinate systems were used in this investigation. The coordinate system defined by the angle of attack α and the angle of side slip β was used to characterize the aerodynamics at the wind turbine blade. The coordinate system defined by the yaw angle θ and the roll angle Φ exploits the symmetry of the theoretical shape of the five-hole probe. The conversion between the two coordinate systems is given by 1α=arctan⁡(sin⁡(ϕ)tan⁡(θ)),2β=arcsin⁡(cos⁡(ϕ)sin⁡(θ)), and 3ϕ=arctan⁡(sin⁡(α)/tan⁡(β)),4θ=arccos⁡(cos⁡(α)cos⁡(β)).

The five-hole probe was calibrated in the red wind tunnel at DTU . The red wind tunnel provides a maximum flow velocity of 40 ms-1 and a turbulence intensity of less than 0.1 %. The test section is 2000 mm long and has a cross section of 750 mm × 500 mm. The tip of the five-hole probe was placed on a turn table 750 mm downstream of the inlet (Fig. ). A reference Pitot tube was placed 60 mm downstream of the inlet.

The calibration was performed at two flow speeds, 20 and 40 ms-1. Five differential pressures were measured during calibration: 5Δp16=p1-p6,6Δp23=p2-p3,7Δp45=p4-p5,8Δpt6=pt-p6,9Δps6=ps-p6, where pn is the pressure at port n=1…6, pt is the total pressure from the reference Pitot tube, and ps is the static pressure from the reference Pitot tube. The reference dynamic pressure is calculated by 10q=Δpt6-Δps6.

The four coefficients, 11Cα=Δp45Δp16,12Cβ=Δp23Δp16,13Cq=Δp16q,14Cs=Δps6q, are calculated by using the calibration pressures from the reference Pitot tube. The coefficients Cα and Cβ are solely defined based on pressures measured with the five-hole probe. They are used to infer the angle of attack and angle of side slip according to the procedure outlined below. The static and dynamic pressure of the flow can only be inferred directly from the measured pressures at the five-hole probe if it is aligned with the flow. The coefficients Cq and Cs were introduced to compensate for the effects of probe misalignment by calibration when determining the static and dynamic pressure.

The results for a flow speed of 20 ms-1 are displayed in Fig. . The results of the calibration at a flow speed of 40 ms-1 are not displayed, because the result is very similar to the result at 20 ms-1. In addition, it was found that each calibration coefficient depended mainly on one flow angle and the influence of the other flow angle could be neglected. Hence, the approximations 15Cα=f(α),16Cβ=g(β),17Cq=h(θ),18Cs=k(θ), were made. The calibration coefficients do not depend on flow speed. This approximation holds if the flow Mach number is lower than 0.3. The calibration functions were assumed to have polynomial shapes and were displayed in red in Fig. .

The data reduction using the calibration coefficients was done in the following way.

Cα and Cβ were computed with the measured five-hole probe pressure differentials.

The deployment of the measurement system, which was planned for the last 2 weeks in April 2024, was challenged by rainy, windy, and cold weather. On Sunday, 21 April, the weather was dry and sunny, but the low temperature in the morning required the removal of ice from the top of the nacelle, the hub, and the blade surface in the area for the belt before the installation could start. Despite nonoptimal installation conditions, both pressure belt systems on blade 1 were installed on this day from a lift (left and middle photos in Fig. ). The already installed inner belt and flyboard are visible at the mid-span of blade 1.

The middle photo of Fig.  shows the belt and the flyboard installed at the outer blade span location. The belt is toward the blade root, and the flyboard is toward the blade tip. The installation included the correct positioning of the two five-hole probes, which were aligned with the mean angle of attack at the two spanwise stations. Angles of attack for alignment of the five-hole probes were estimated by running the turbine model in OpenFAST through six turbulent seeds from the normal turbulence model with an average wind speed at the hub height between 5 and 11 ms-1 in steps of 1 ms-1.

The installation of the measurement system without the tower belt was completed in the morning of the following day. Instead of taking time to mount the tower belt, a measurement campaign of several hours was conducted that day.

Unfortunately, rainy weather followed the next several days, until Monday, 29 April, when it was possible to install the tower belt. The right photo of Fig.  shows the installed pressure belt with the vertical displacement of the belts relative to each other. A short measurement campaign was conducted in the afternoon that day with all three pressure belts working.

Data were collected on two days: 22 April (D1) and 29 April (D2) 2024. During D1, only the two blade systems were available. During D2, all systems were available, but data on the inner station of the blade were partly corrupted by the presence of water in the inner blade belt.

The pressure data were filtered using conditions extracted from the SCADA data to find time periods of fully operational data. The following filters were applied:

active power larger than 150 kW,

The application of these filters reduced the dataset to 1.4 h for D1 and 0.28 h for D2.

Data from the five-hole probe were first converted to inflow velocity and inflow angle using the calibration obtained during wind tunnel testing. Next, these quantities were used to estimate the angle of attack and relative velocity based on EllipSys2D computational fluid dynamics (CFD) simulations. The trailing edge pressure measured from the belts was interpolated first, and then the pressures were integrated to determine aerodynamic forces and coefficients. The results were time-averaged (10 samples, resolving up to 10 Hz), filtered based on SCADA data, and binned in terms of angle of attack and azimuth. The pressure coefficient, Cp, for the measured data was calculated using the static and dynamic pressure from the five-hole probe.

In Fig. , the measured Cp distribution of the inboard section binned on various angles of attack is presented and compared to EllipSys2D CFD results. There is a clear discrepancy, especially on the suction side, resulting in an overall loading difference with the model results. This could be attributed to shape and roughness imperfections but is probably not least due to the impact of three dimensional flow effects in combination with a rather thick airfoil of 25 % in atmospheric turbulent flow. The missing suction peak could be due to this.

However, the conversion of the measured inflow angle by the five hole probe to AoA based on 2D CFD simulations is also an important source of uncertainty combined with the uncertainty in estimating the geometrical angle of the five hole probe relative to the local chord.

In Fig. , the measured Cp distribution of the outboard section binned on various angles of attack is presented and compared to EllipSys2D CFD results. The simulation results show a good match with the measurements, especially the fully turbulent ones, indicating good estimation of the overall loading. The improved correlation between measurements and simulations for this station compared with the inboard station mentioned above can probably also be attributed to a more 2D inflow at this station. We also notice here that the suction peak at 8° and, in particular, at 10° is well captured by the pressure measurements.

In Fig. , the measured binned lift coefficients of both sections are compared to the EllipSys2D CFD data. The measurements for the inboard section indicate a lower lift slope, which could be attributed to shape and roughness imperfections and the uncertainty mentioned in the conversion method from the input angle to the angle of attack, the former being visible in the pressure distributions. The deviations correspond to the pressure distribution differences shown in Fig. .

The measurements for the outboard section match quite well with the fully turbulent simulation data, which is also in line with the good correlation between the experimental and simulated pressure distributions in Fig. .