Laminar-turbulent transition characteristics of a 3-D wind turbine rotor blade based on experiments and computations

Laminar-turbulent transition behaviour of a wind turbine blade section is investigated in this study by means of field experiments and 3-D computational fluid dynamics (CFD) rotor simulations. The power spectral density (PSD) integrals of the pressure fluctuations obtained from the high frequency microphones mounted on a blade section are analyzed to detect laminar-turbulent transition locations from the experiments. The atmospheric boundary layer (ABL) velocities and the turbulence intensities (T.I.) measured from the field experiments are used to create several inflow scenarios for the CFD simulations. 5 Results from the natural and the bypass transition models of the in-house CFD EllipSys code are compared with the experiments. It is seen that the bypass transition model results fit well with experiments at the azimuthal positions where the turbine is under wake and high turbulence, while the results from other cases show agreement with the natural transition model. Furthermore, the influence of inflow turbulence, wake of an upstream turbine and angle of attack (AOA) on the transition behaviour is investigated through the field experiments. On the pressure side of the blade section, at high AOA values and wake conditions, 10 variation of the transition location covers up to 44% of the chord during one revolution, while for the no wake cases and lower AOA values, variation occurs along a region that covers only 5% of the chord. The effect of the inflow turbulence on the effective angle of attack as well as its direct effect on transition is observed. Transition locations for the wind tunnel conditions and field experiments are compared together with 2-D and 3-D CFD simulations. In contrast to the suction side, significant difference in the transition locations is observed between wind tunnel and field experiments on the pressure side for the same 15 airfoil geometry. It is seen that the natural and bypass transition models of EllipSys3D can be used for transition prediction of a wind turbine blade section for high Reynolds number flows by applying various inflow scenarios separately to cover the whole range of atmospheric occurrences.

industrial applications, the e N method together with empirical criteria for transition mechanisms that are not covered by this approach, such as bypass and attachment line transition, keeps its place as a practical method (Krumbein, 2009). The design process for the wings and airfoils still requires the use of laminar-turbulent transition modelling in Reynolds-averaged Navier-Stokes (RANS) solvers. The current analysis involves a coupling of the e N transition model with the RANS solver considering its accuracy for high Reynolds number flows in wind turbine applications (Sørensen et al., 2014). 5 The experimental studies on laminar-turbulent transition on aerospace applications goes back a long way compared to the research conducted on wind turbines. While the inflow turbulence intensity for an airplane wing in cruise is lower than the one experienced in a wind tunnel, it is higher for a rotating machinery or wind turbine rotors (Hernandez et al., 2012). Transition analysis performed for wind tunnel experiments in controlled conditions includes measurements on wind turbine airfoils equipped with pressure taps and sensors, balance system and a wake rake (Ceyhan et al., 2017); infrared thermography (Joseph 10 et al., 2016); rotating turbine blade equipped with pressure sensors, strain gauges, balance system and particle image velocimetry (Schepers and Snel, 2007), rotating wind turbine and wind turbine blade experiments by oil visualization, stethoscope and flush-mounted unsteady pressure sensors (Lobo et al., 2018), wind turbine airfoil with pressure sensors and high frequency microphones (Özçakmak et al., 2019).
In addition to the DAN-AERO experimental campaign   (Troldborg et al., 2013), of which the current 15 analysis is based on, there have been other field experiments on boundary layer transition on rotating wind turbine blades using microphones glued on the surface (Van Ingen and Schepers, 2012), hot film and pressure tubes (Schwab et al., 2014) (Schaffarczyk et al., 2017), microphones on the suction side in addition to the ground based thermographic cameras (Reichstein et al., 2019). All these experiments pointed to a fact that more field experiments are needed on the wind turbine blades in order to characterize the transition behaviour with inflow turbulence and rotational effects. Moreover, determining the relevant 20 frequency ranges for the atmospheric turbulence and the occurrence of the T-S waves under real atmospheric conditions is needed in order to investigate the combined effects of the turbulent wind and the blade rotation on transition. Modern wind turbines usually operate in wind farms where the inflow is affected by the wake of the upstream turbines. They are also exposed to high free-stream atmospheric turbulence and wind shear. The differences in the transition behaviour of an airfoil section tested in a controlled 2-D wind tunnel environment and a blade tested in 3-D field experiments at real operational 25 conditions is discussed in previous works and it is seen that full rotor blade section exhibits different transition characteristics than in the 2-D case (Madsen et al., 2019a)  . The difference between the design conditions for rotor and airfoils and the real operating conditions leads to inaccurate predictions of the loads and the performance as observed in previous studies from Sørensen (2009) Chaviaropoulos et al. (2015).
. The difference between the design conditions for rotor and airfoils and the real operating conditions leads to inaccurate 30 predictions of the loads and the performance as observed in previous studies from Sørensen (2009) Chaviaropoulos et al. (2015).
In this study, the transition characteristics of the LM 38.8 blade on the NM80 2.3MW wind turbine is analysed by field experiments (DAN-AERO project), and computations with the DTU in-house CFD EllipSys code (Sørensen, 1995) (Michelsen, 1992), (Michelsen, 1994). The present experimental analysis is based on high frequency microphone measurements that enables 35 acquiring data at higher sampling frequencies and allows a higher resolution (with the number of microphones placed on both upper and lower surfaces) than the previous studies. This paper is focused on the analysis of the DAN-AERO 3-D transition rotor measurements in a wind farm and the validation of the transition models in the EllipSys3D CFD in-house solver using this experimental data. The atmospheric turbulence, wind shear and wake effects on transition behaviour of a wind turbine blade section is discussed. The effect of these parameters on the effective angle of attack and velocity on the blade section as well as 5 their potential direct effect on transition is discussed. Comparison of the field experiments with the CFD simulations enlightens the transition behaviour of the wind turbine blades and enables improvement of the design and aerodynamic prediction tools.

Field Experiments : Set-up and Instrumentation
The main objective of the DAN-AERO project was to establish an experimental database for aerodynamic, aeroelastic and aeroacoustic issues that are significant for the design and operation of MW size wind turbines . The laminar-10 turbulent transition investigation of this campaign contains both 2-D wind tunnel tests  as later analysed by (Özçakmak et al., 2018) (Özçakmak et al., 2019), and 3-D field experiments (Troldborg et al., 2013).
In this study, field experiments are analysed in order to investigate the laminar-turbulent transition characteristics of a 3-D rotor blade. The tested turbine is placed at a wind farm in Tjaereborg, Denmark, which consists of 8 turbines in 2 rows. The test turbine is a 2 MW NM-80 wind turbine with LM-38.8 blade. The rotor diameter is 80 meters. The site and the test turbine 15 (denoted as 'NM80') is presented in Figure 1. The wake cases presented in this paper are from an upstream wind turbine that is located around 6 rotor diameters (6D) upstream of the test turbine.
The rotational speed, yaw, pitch and rotor azimuth angles are measured at the nacelle. In addition to the pressure taps placed at four different sections on the blade, 56 high-frequency microphones are installed about 1 mm below the blade surface at a section 36.9 meters from the hub (3.1 m from the tip of the blade). The same section is also equipped with a pitot tube for 20 measuring the relative velocity. The pressure taps are placed 36.8 m from the hub (3.2 m from the tip) next to the microphones.
The wind direction and wind speed information of the inflow is obtained from the anemometers and wind vanes placed at the meteorological mast (denoted as 'MM') 2.5 diameter far from the test turbine, see Figure 1. At some specific wind directions, the test turbine is in the wake of the upstream turbine, effect of which is also discussed in this study. Both wake and no-wake conditions are analyzed. 25 The angle of attack values of the field experiments presented in this article are derived from the measured normal force on the blade. The correlation between the normal force and the angle of attack is generated by the HAWC2 (Horizontal Axis Wind turbine simulation Code 2nd generation) (Larsen and Hansen, 2007) simulations which is based on the principle blade element momentum theory with an aero-elastic model of NM80 turbine using existing polars. A previous study of Boorsma et al. (2018) has shown a good correlation between measured and computed normal forces. The acquisition properties of the 30 instruments on the test blade and the meteorological mast are listed at Table 1.

Data processing
The pressure fluctuations in time domain (10 second series), obtained from high frequency microphones placed chordwise on the blade section, are analyzed in the frequency domain by fast Fourier transform (FFT) analysis. The sampling frequency of 10 the data is 50 kHz acquired over 10 seconds. The data is divided into smaller time segments of 0.0410 seconds. The window size of 4096 is used with a 50 % overlap. For each time segment, the power spectral density (PSD) is calculated by the short-time Fourier transformation analysis (Welch, 1967).
The PSD (P s,p ) of the pressure fluctuations obtained from the microphones are integrated in a frequency interval from f1=2 kHz to f2=7 kHz(see Equation 1). The integration within a certain frequency range gives the standard deviation (σ), which The transition location on the upper and lower surfaces is detected by the highest chordwise derivative of the pressure level as in Equation 2. The derivatives that are above a threshold level of 250 dB are selected for transition detection.
The transition detection method is illustrated in Figure 3. EllipSys3D is an in-house CFD solver for incompressible Navier-Stokes equations in general curvilinear coordinates by a multiblock finite volume discretization, here applied in RANS mode. Rhie-Chow (Rhie, 1982) interpolation is used in order to avoid odd/even pressure decoupling. The third order QUICK (Quadratic Upstream interpolation for Convective kinematics) upwind scheme is used for the convective terms. SIMPLE (Semi-Implicit Method for Pressure-Linked Equations) algorithm is 5 used to enforce the pressure/velocity coupling (Patankar and Spalding, 1983) (Patankar, 1980). The Message-Passing-Interface (MPI) is used to parallelize the code for executing on distributed memory machines with non-overlapping domain decomposition (Sørensen et al., 2011).
Mesh generation is done by the 3-D hyperbolic grid generation program HypGrid3D (Sørensen, 1998) The turbulence is modelled by the k − ω SST eddy viscosity model (Menter, 1993). The boundary layer transition prediction The angle of attack from the EllipSys3D simulations on the blade section is determined from the annular averaging method by determining the local induced velocities at the blade (Hansen et al., 1997) (Johansen and Sørensen, 2004). In order to calculate the induced axial velocity, V ind , at the rotor plane, annular averaging of the axial velocity at several upstream and 30 downstream locations (z i and z j ) at a given radial location (in this case, the same blade section as in the microphone ex-periments) is performed. Then, V ind is found by the interpolation of these averaged streamwise velocities to the rotor plane.
Lagrangian polynomial interpolation is used to determine the velocity at the rotor location (z = z 0 ) as in Equation 3.
Having calculated V ind , the effective local flow angle (α) is found from Equation 4, where ω is the angular velocity, R is the distance to the hub from the measured tested section on the blade, θ is the combined pitch twist angle.
The Fieldview (2017) software is used in order to postprocess the EllipSys3D simulation results, to extract the annular averages of the axial velocity and for the flow visualization.

EllipSys3D semi-empirical e N transition model
Transition to turbulence in EllipSys3D is governed by the semi-empirical e N model (Drela and Giles, 1987), which is based 10 on linear stability theory. The conventional e N method is a semi-empirical method not considering receptivity. In the semiempirical method, while linear stability analysis is done for the governing equations, the transition is assumed to take place when N reaches a previously correlated value from the experiments. Therefore, the empirical part comes from the N value at transition, which makes the model semi-empirical. In the e N method , the amplification of small disturbances are calculated for several frequencies, and the spectrum of the most amplified ones is identified. The critical N factor for each type of flow is 15 determined empirically, and the transition point is detected from this empirical value of the critical N factor.
In the EllipSys3D code, the boundary layer parameters (the displacement thickness δ * , momentum thickness θ and the shape factor H in Equation 5) are needed as an input to determine the occurrence of transition. U e is the velocity at the outer edge of the boundary layer which is determined from the Navier-Stokes computation, and y is the direction perpendicular to the wall/surface. The boundary layer thickness δ is defined as the location where u = 0.99U e . In EllipSys3D, the edge velocity 20 (U e ) is taken as the maximum tangential velocity.
Although, these boundary layer parameters can be calculated by integrating the velocity profile from the Navier-Stokes equations, it is shown by Stock and Haase (1999) that it requires an excessively fine computational grid. Therefore, the boundary layer parameters are found from the von Karman boundary layer equations (Von Kármán, 1921). 25 The momentum integral equation (Equation 6) and the combination of the kinetic energy equation with von Karman's momentum equation (Equation 7) are solved for H and θ. U e is determined from the Navier-Stokes computation. In each iteration, these equations are started from the stagnation line and integrated downstream on the surface until the transition point is found.
3 ) and D is the dissipation per unit area. The U n term, the 5 velocity normal to the wall, comes from the assumption of axial symmetry. These equations are presented in terms of boundary layer parameters and the skin friction coefficient C f , where x is the horizontal direction, τ w is the wall shear stress, ρ is density, µ is the dynamic viscosity, θ * is the kinetic energy thickness, and H * is the energy thickness ratio as shown in Equation 8.
Closures for C f , C d and θ * are calculated based on Falkner-Skan velocity profiles. The stagnation line is normally found as 10 the location where the pressure coefficient based on relative velocity is equal to 1. In order to ensure that the entire stagnation line is found at each time step, the pressure coefficient value is checked and the Navier-Stokes solution is analyzed. Along the stagnation line, Blasius flow and across the stagnation line Hiemenz flow are assumed. By this way, realistic initial values for H and θ are obtained and the stagnation line can be located between two computational cells.
The development of the imposed wave perturbations' amplitude is computed along the boundary layer based on spatial 15 analysis. A check is carried out during the integration of the boundary layer equations to determine if the disturbances are amplified or damped. As neutral stability is passed, amplifications are determined for a range of temporal frequencies.
The N factor is the natural logarithm of the ratio of the disturbance amplitude at a specific location to its amplitude at the neutral stability point. Integration of the N amplitude is carried out until it reaches a certain value for which transition is said to occur. This value is set according to the turbulence degree that is present in the experimental conditions for comparison 20 with the simulations. In order to build a relation between the turbulence level and the amplification factor, Mack's expression (Mack, 1977) is used as an estimate.
Transition to turbulence is handled by the intermittency factor (γ) when solving the Navier Stokes equations. The eddy viscosity (µ T ) obtained from the turbulence model is multiplied with the intermittency factor which controls the effective viscosity (µ ef f =µ + γµ T ), where µ is the molecular viscosity. Intermittency The intermittency factor is calculated from Equation 10. This equation is obtained by combining the statistical theory for 5 transitional flow by Emmons (1951), and the expression that represents the production rate with Gaussian distribution using the Dirac delta function by Dhawan and Narasimha (1958) and the Chen and Thyson (1971) formulation: where the subscript tr is the transition onset, ν is the kinematic viscosity, σ is here the spot propagation rate, and n is the non-dimensional spot formation rate, n = n · ν 2 /U e,T r 3 (Mayle, 1998).

10
The intermittency factor is calculated on the surface and then solved for the entire boundary layer and wake within the transport equation (Michelsen, 2002).
where the source term, S, is obtained by evaluating the transport terms for previously determined intermittency values.
3.1.1 Bypass transition model 15 When the amplitude of the disturbances are strong, such as for high free stream turbulence or large roughness elements, the the linear stages of the transition process is bypassed. In this case, transition happens in the absence of T-S waves and the disturbances are amplified by a non-linear process.
The approaches for modelling bypass transition in industry involves low Reynolds number turbulence models, and models using experimental correlations that relates free-stream turbulence intensity to transition Reynolds number based on momentum 20 thickness Re θt (Reza and Amir, 2009).
The e N method accurately predicts the transition for free stream turbulence levels, less than 2% for T-S dominated transition, but for higher levels it is bypassed (Biau et al., 2007). It should be noted that transition can also be dominated by other mechanisms; for instance by a streak breakdown for turbulence levels around 0.65% (Suder et al., 1988). The e N method in EllipSys3D can be used together with a bypass criteria. For the bypass transition model, Suzen and Huang (2000) empirical 25 model is used.
Abu- Ghannam and Shaw (1980) suggested that for the attached flows, transition onset can be obtained by correlating Re θ to the free stream turbulence intensity as in Equation 12. By maintaining the strong features of this correlation in adverse pressure gradient regions, more sensitive response to the favourable pressure gradients is obtained by Suzen et al. (2002) by re-correlating the transition criterion to the free stream turbulence intensity and acceleration parameter K t .
Re θtr = (120 + 150T u (−2/3) )coth 4(0.3 − 10 5 K t ) (12) 5 Here, K t is the minimum value of the acceleration parameter in the downstream direction (Michelsen, 2002), which can be expressed as ν/U t 2 (dU/dx) t where U t is the boundary layer velocity at onset of transition (Suzen et al., 2002). T u is the turbulence intensity at the transition onset. Under high turbulence intensity conditions, this correlation fits well with adverse pressure gradient regions. In EllipSys3D, for the bypass transition cases, where turbulence intensity levels are high, this correlation is used and the criteria for natural and bypass transitions are checked simultaneously in the code. The higher of the two 10 intermittency factor is used. Moreover, separation induced transition is also checked with a bubble model inside the boundary layer solver.

Numerical Set-Up
The computations are performed for three different grid levels for the 3-D simulations, and five different grid levels for the 2-D case. The grid independence is ensured and the results of the finest grid are presented.

15
The 3-D full rotor simulations are performed as transient calculations with 1200 steps per revolution for all grid levels.
The problem is approximately axisymmetric. The CFD Simulation input parameters are listed in Table 1    , deviating from the 2-D case (Figure 4-a).
In order to analyze the causes of this variation, inflow characterization is performed. The PSD is integrated in the frequency range from 2 kHz to 7 kHz for transition detection (presented in Figure 5-b). Several frequency intervals for PSD integration is also attempted in order to characterize the inflow turbulence from the microphone signals. It is seen in Figure 5-a that when 20 the PSD is integrated in the frequency interval from 100 Hz to 300 Hz, the microphones closer to the leading edge shows high pressure levels on this frequency range capturing the pressure response to the inflow turbulence. Therefore, this frequency range is selected for the inflow analysis. For the quantitative comparisons, a microphone located very close to the leading edge (at x/c = 2.2%) on the pressure side, in the laminar boundary layer, is selected to represent the inflow turbulence (L p,i ) that the blade section is exposed to.
The inflow turbulence levels (L p,i ) from the microphone analysis, the relative velocity obtained from the pitot tube on the blade section, and the angle of attack, which is derived from the forces, are presented together with the detected transition points as a function of the azimuthal angle in Figure 6. Two different cases from the measurements are shown in Figure 6.
5 Figure 6 -left shows a partial wake case (9% wake-affected rotor area) with a free stream velocity (U ∞ ) of 7.2 m/s. Figure 6 -right demonstrates a half wake case (48% of the rotor area covered with wake affected inflow) with U ∞ =8 m/s. Both cases have the same pitch setting and the same T.I.= 2.8% obtained from the meteorological mast. The wake shadow from the rotor view and the top view is also shown at the bottom of the figure for each case. The swept area that is influenced by the wake according to the wind direction was calculated by estimating the wake expansion depending on the velocity induction factor.

10
Each line at V rel , AOA, L p and x/c tr plots corresponds to a single revolution. For the 9% wake case, each revolution tends to have a similar behaviour, on the contrary, for the 48% wake case, there are some discrepancies observed between each rotation.
For both cases, it is noticeable that the decrease in relative velocity and angle of attack and the increase of the inflow turbulence leads the transition point to move closer to the leading edge at the pressure side. The tilt, yaw and wind shear effects on AOA and Vrel are analyzed by HAWC2 simulations. Yaw misalignment (which is the difference between the angle measured on the 15 nacelle and the wind direction measured from the meteorological mast at the same height) was checked for the cases presented in this paper. The mean absolute yaw error is found to be less than 5 degrees, and the effect of the maximum yaw on V rel and AOA change is found to be no more than 1% by HAWC2 analysis. Moreover, the wind turbine has a 5 • tilt angle that causes 1% change in V rel and 0.2% change in AOA according to this analysis. Considering that the cases presented here are not under a strong shear, and comparing those variations with the experiments, it can be concluded that the azimuthal behaviour of the relative velocity and the angle of attack is governed by the inflow turbulence, mainly from the wake of an upstream turbine. In order to distinguish the effects of the AOA and inflow turbulence, the data is divided into AOA bins. A previous study of Madsen et al. (2019a) with data from the DAN-AERO experiments shows that for several angle of attack bins, there is a correlation between inflow turbulence and the transition location. Increasing the turbulence content in the range of 100-300 Hz moves the transition process closer to the leading edge at the pressure side. At the suction side, transition points are detected within the first 13% of the chord, and no correlation could be established with the inflow turbulence.

Angle of attack effect on transition
The relative velocity, turbulence levels (L p,i ), AOA and the detected transition points on the pressure side for two different pitch cases are compared in Figure 7 as a function of time. These two cases belong to different measurement sets; a low pitch case (right) with 7 m/s free stream velocity with 2.6% T.I. and a high pitch case (left) with 7.2 m/s free stream velocity with 2.8% T.I. The high pitch case is under 8.9% wake from the upstream turbine while the case with low pitch angle is under no wake

Comparison between CFD and Experimental Results
Since the effective angle of attack in the experiments is derived from the force measurements, in order to have a direct comparison, the experimental forces are compared with the forces obtained from the EllipSys3D simulations. The F x and F z forces the experiments. The wake shadow falls in the azimuthal range from 200 to 340 degrees for this case from the measurements.
While large scale vortices contained in the wake of the upstream turbine might mean higher turbulence intensities, there is a also a velocity reduction due to the energy extracted from the wind by the upstream turbine. Therefore, in the wake region, the experimental force shows agreement with the forces obtained from the simulations with a lower inflow velocity (the fitted region is highlighted by the turquoise block). It is also observed from the data that high wake cases introduces bigger amount 5 of variation in the sectional force compared to low wake cases.
It is seen that the experimental force variation for 4 revolutions is comparable with the numerical results. For further validation, the transition points are also compared for the pressure and the suction sides in Figures 10 and 11 respectively. Figure 9, 10 and 11 show the results for the same measurement dataset. Three different turbulence intensity levels are shown in most of the cases to cover several ranges of turbulence intensity levels in the atmosphere.

10
It is seen from Figure 10  transition results covers most of the scenarios that the turbine is exposed to during rotation. 5 On the suction side, in Figure 11, the opposite behaviour of the transition point is noticeable with the angle variation. The regions where the transition point is closer to the leading edge corresponds to the high angle regions. The regions under wake at higher azimuthal angles correspond to the decreasing V rel and AOA (see Figure 6-right) which moved the transition point further downstream. In this region on the suction side, although the turbulence intensity is increasing, the individual effect of the increasing AOA is more prominent than the effect of the inflow turbulence itself. However, variations between each rotation 10 is also noticeable, which might be due to the inflow turbulence. It should be noted that although there are rotational changes in the transition point, x/c tr , on suction side of the rotor blade section, the detected transition locations are considerably close to the leading edge, so the relative movement is not as prominent as in the pressure side and it is harder to reach a reliable conclusion. The pressure coefficient results from the pressure taps and those from the simulations are also compared. In the experiments, 4 blade sections are equipped with pressure taps and the current analysis shows the most outward section (next to the microphones) that has the highest velocity change with the azimuth compared to the other sections. The pressure coefficient is calculated as follows: where V ∞ is the free stream velocity measured on the meteorological mast, r is the radial position of the section where the  The Cp values obtained from 3-D simulations and experimental Cp value at 90 • azimuthal angle is presented in Figure 13-b.
EllipSys3D simulations for various free stream velocities observed during the acquisition time of the pressure measurements fit well with the 3D experimental results. In Figure 13-c, EllipSys3D results for fully turbulent, natural and bypass transition are shown for 270 • azimuthal position. This is the region where there is wake affected inflow in this measurement set, therefore numerical results obtained for lower freestream velocities show agreement with the 3D experimental results. low and no wake cases, the transition location movement is within 13% along the chord in one revolution for the suction side.
The more downstream transition locations, seen at low AOA values, fit with the results from the simulations with the natural transition model. This indicates that the natural transition type is also present on the suction side. Moreover, the 2-D and 3-D experimental results show agreement in the high AOA range on the suction side and the transition locations are in very close proximity of the leading edge in most of the cases.

Conclusions
In this study, the analysis of the field experiments and results from the 3D CFD simulations are presented to characterize the laminar-turbulent transition behaviour of a wind turbine under real atmospheric conditions. The data from high frequency 5 microphones placed on a wind turbine blade section are analyzed in the time and frequency domains. The transition locations are detected from the standard deviation of the pressure fluctuations, which are integrated between 2kHz and 7kHz. The inflow turbulence behaviour is obtained from one of the microphones placed nearby the leading edge by integrating the spectra from 100 to 300 Hz. The inflow velocity is obtained from meteorological mast measurements and used as an input parameter in the CFD computations. The T.I. for the simulations is obtained from the relative velocity measurement from the pitot tube placed 10 on the blade section.
The field experiment results showed that the transition behaviour on the wind turbine blade in real operating conditions differs from the model in the wind tunnel, caused by the influence of the inflow turbulence and the wake from another turbine.
These factors change the relative velocity, so the effective AOA on the blade section, and, besides, inflow turbulence is observed to have some direct effects on transition. 15 The effect of the wake is visible from the variation of the detected transition points at each revolution. As the wake affected rotor area increases, bigger jumps of the transition position is observed during one revolution. At the low and non-wake cases, each revolution is almost identical and the transition behaviour is mainly governed by the angle of attack changes due to the inflow velocity. The angle of attack effect on transition is analyzed by comparing results from the two different pitch settings under similar inflow conditions. It is seen that for the pressure side, at low AOA cases, the transition position is not affected 20 by the variations during a revolution as much as in the high AOA cases. Changes in AOA is found to be highly correlated with transition locations during a revolution and the variations among different revolutions are due to the inflow turbulence.
The normal sectional forces from the experiments and simulations are compared in order to quantify the rotational changes of the force, and analyze the differences among several revolutions. Moreover, by binning the sectional forces from the experiments by the inflow velocity, the range that is covered by the simulation results obtained with various N numbers for the 25 natural transition and T.I. for the bypass transition is identified. It is seen that the field experiments and the 3D simulations are comparable. seen that for high AOA and wake cases, the movement of the transition point covers up to 44% of the chord on the pressure side in a single revolution, a value that drops to 5% at low AOA and for no wake cases. On the suction side, changes in the transition position is also observable, and the field and wind tunnel experiments agree in the high AOA range. It is seen that, on the suction side, the effect of AOA is more prominent than the direct effect of the turbulence intensity, though it is not 5 easy to reach a conclusion as the transition positions are in very close proximity to the leading edge (within x/c = 1 − 13%).
Therefore, at these physical conditions, the suction side is not suited to distinguish the type of the transition mechanism. It is visible from the pressure coefficient results that for azimuth angle 270 • , where there is a wake from an upstream turbine for the presented case, the experiments fit with the low velocity 3D simulation results for natural and bypass transition models.
On the other hand, 90 • azimuthal position corresponds to high AOA region in the field experiments, a suction peak increase It is seen that the e N semi-empirical transition model and bypass transition model in EllipSys3D can be used for high 15 Reynolds number flows (Re=5 million) in real atmospheric conditions. Using both models can cover the range of transition positions that is seen in the field experiments with a relevant choice of the amplification factors and T.I. values.
Several inflow scenarios are simulated separately in EllipSys3D as it is hard to control high turbulence in the wake region and handling varying N factor and T.I. in a single simulation. Simulations with more inflow characterization can be studied in the future in order to simulate the real inflow conditions from the experiments. Moreover, detailed characterization of the inflow 20 turbulence measured on the blade with high sampling frequency instruments in field experiments is needed to separate relevant frequencies that affect boundary layer transition. By more field experiments and high resolution simulations, laminar-turbulent transition predictions can evolve, and eventually contribute to the aerodynamic prediction and the design of the wind turbine blades.
Data availability. Data is available upon request to corresponding author.