Krogstad 2011

WIND ENERGY
Wind Energ. 2012; 15:443–457

Published online 13 June 2011 in Wiley Online Library (wileyonlinelibrary.com). DOI: 10.1002/we.482
RESEARCH ARTICLE
An experimental and numerical study of the

performance of a model turbine
P. -Å. Krogstad1 and J. A. Lund2
1 Department of Energy and Process Engineering, The Norwegian University of Science and Technology, Trondheim, Norway
2 Agder Energi, Kristiansand, Norway
ABSTRACT
The performance of a 0.9 m diameter model wind turbine using the National Renewable Energy Laboratory S826 airfoil
profile has been investigated both experimentally and numerically. The geometry was laid out using blade element
momentum (BEM) theory, and a detailed description of the geometry is given here. The design was tested experimentally
and gave a peak power coefficient of CP D 0:448 at the design tip speed ratio of D 6. After the model tests had been
undertaken, numerical calculations were performed by means of fully three-dimensional computational fluid dynamics
(CFD) simulations using a k ! turbulence model. It was found that the BEM correctly predicts the shape of the power
and thrust coefficient curves, with the efficiency coefficient virtually identical to the measurements at the design condi-
tions. At higher tip speed ratios, the performance is over-predicted. The estimated thrust was, however, consistently too
low by a shift of the order of CT 0:1 in the normal operating tip speed range. The high-resolution CFD predictions
(using about 3:5 106 grid points) reproduced the model thrust coefficient almost perfectly, and the predicted power coef-
ficients were also very close to the measurements, although the agreement with the measurements at high tip speed ratios
were only marginally better than those from the BEM method. At the design tip speed ratio, the CFD over-predicted the
power coefficient by merely 2%. The good agreement between the measured and computed performance at model scale
assures that accurate predictions of turbine performance at full-scale conditions are also possible with high-resolution CFD.
Copyright © 2011 John Wiley & Sons, Ltd.
KEYWORDS
wind turbine; model test; numerical simulation
Correspondence
P. -Å. Krogstad, Department of Energy and Process Engineering, The Norwegian University of Science and Technology, Trondheim,
Norway.
E-mail: per.a.krogstad@ntnu.no
1. INTRODUCTION
The wind turbine design phase is normally started by doing general performance studies using a blade element momentum
(BEM) method. When the optimum design has been sorted out, more refined flow studies may be performed using more
sophisticated software that solves the Navier–Stokes equations in a rotating frame of reference. In this way, possible trouble
areas may be found and the geometry may be modified to reduce these.
Because of the considerable costs involved in developing a wind turbine, it is absolutely essential that the predicted
behavior of the turbine may be trusted. Since only limited information is available from full-scale measurements, it may be
necessary in some cases to perform model tests to verify the performance of the turbine before a full-scale turbine is made.
The scale of the model is necessarily quite small, being typically two orders of magnitude less than the full-scale turbine.
Therefore, the data obtained from the model are likely to suffer from scale effects.
In order to reliably assess the performance of wind turbine prediction tools, well-documented experiments are required.
This is difficult to obtain on site at full scale, because a complete test case description will need full documentation of
the incoming flow in time and space. To account for this, a few reasonably large wind turbines have been tested in
wind tunnels at very high cost. The by far largest model tested in this way so far is the 10 m diameter NREL turbine
(National Renewable Energy Laboratory, Golden, CO), which was operated in the huge 24.4 m by 36.6 m closed test
Copyright © 2011 John Wiley & Sons, Ltd. 443

Study of the performance of a model turbine P. -Å. Krogstad and J. A. Lund
section of the NASA Ames wind tunnel. These tests were completed in 2000. Even a model at this scale will suffer from
scale effects since it is only about one-tenth of a modern-size turbine and it is tested in a closed environment.
However, for developers of prediction methods, the data generated are invaluable. Despite the high costs of these full-
scale experiments, more than 1700 test conditions were investigated. The data from the tests have been distributed widely
and have been used, e.g. in the ‘blind comparison’ tests undertaken in 2000. Here, a large number of experts were invited
to predict the turbine performance before the data were released (see Simms et al.1 or Sørensen et al.2 ). This exercise
demonstrated that considerable uncertainty in the estimates was to be expected from all types of prediction methods at
that time.
Later, the MEXICO project was undertaken in Europe by testing a 4.5 m diameter turbine in the 9:5m 9:5m open test
section of the DNW wind tunnel (see e.g. Snel et al.3 ). Schreck et al.4 made a comparative study of the data from both the
NREL and MEXICO tests and demonstrated that an important cause for the failure to predict the measurements correctly
is due to the rotational lift augmentation that may be hard to incorporate in most prediction methods that do not solve the
full set of equations of motion. They concluded that rotational effects cause the blade to stall at significantly higher angles
of attack than what follows from two-dimensional (2D) predictions, leading to much higher peak lift coefficients.
Large-scale tests have limited possibilities to give information about wake development and turbine interaction because
of the limited streamwise extents of these large-scale wind tunnels. Here, smaller models are needed. In the review paper
by Vermeer et al.5 on wind turbine wakes, a list of 19 wind tunnel investigations at conventional wind tunnel scales is given
and their results were discussed. Most of the tests were designed for studies of wake development behind a single turbine,
but a few investigations included multiple turbines, where wake interactions could also be studied by operating the turbines
in clusters. Adaramola and Krogstad7 investigated the interaction between two 0.9 m turbines, similar to the one used in the
present study, operating in-line. They showed that the interaction is sensitive not only to the distance between the turbines
but also to how they are operated. By varying the tip speed ratios and yaw angles, we could achieve a significant increase
in total power production.
Hansen et al.6 gave a review of turbine performance prediction methods of varying complexity, such as BEM, actuator
disc and Navier–Stokes based methods. They concluded that BEM methods may perform well provided that good airfoil
lift and drag data are available as function of angle of attack and Reynolds number. This will be discussed in some detail
later in this paper.
The present study was initiated when the well-established wind turbine technology was to be used in a very different
environment (tidal current power generation). It then became essential to investigate how well the performance of a model
turbine could be predicted using traditional prediction methods, one of which was a commercially available Reynolds
averaged Navier–Stokes solver (Fluent Inc., Lebanon, NH). Therefore, model tests and predictions from various methods
were initiated. It was found that the computational fluid dynamics (CFD) method reproduced the model scale data quite
well, giving confidence to the scaling of the model results to full-scale predictions provided that the numerical grid system
is sufficiently refined to account for the increase in Reynolds number.
The layout of the paper is as follows: in Section 2, the model and its test environment are given in sufficient detail to
allow a simulation to be set up and the accuracy of the experiment to be scrutinized. Section 3 gives the details of the
numerical simulations. Comparisons of the CFD computations with the design BEM predictions and the data from the
model experiment are presented in Section 4. Finally, Section 5 gives some conclusions from this study.
2. MODEL TESTS
2.1. Model
The model geometry was laid out using an in-house developed BEM method, incorporating the Prandtl correction for
tip losses and Glauert correction for the thrust force (see, e.g. Manwell et al.8 ). The design tip speed ratio was set to
D !R=U1 D 6 (! is the angular speed of rotation, R is the blade radius and U1 is the reference velocity), and the rotor
was designed with three blades using the NREL S826 profile shown in Figure 1 throughout the blade span (see Somers9
for profile definition). This is a 14% thick profile with a separation ramp at the back, designed for a Reynolds number of
Re D !Rc= D 2 106 to give high maximum lift, gentle stall and insensitivity to surface roughness (c is the chord
length, and is the kinematic viscosity). The insensitivity is achieved by promoting very early transition to turbulence on
the suction side. This ought to also reduce the Re dependence that was considered a bonus since the profile was going to
be used at a model scale that would typically reduce the Re by a factor 20.
In order to determine the flow characteristics at the model Re, we computed the expected airfoil performance by using
the freeware XFOIL package available from Massachusetts Institute of Technology.10 The blade was then laid out for a
nominal angle of attack of 7ı and CL 1:29 with an estimated lift to drag ratio of CL =CD 58. These numbers were
obtained using a Reynolds number of Re D c!R= D 105 at the blade tip. Because of the small scale of the wind turbine,
transition effects could be important. To simulate the effects of wind tunnel free-stream turbulence intensity and blade
Wind Energ. 2012; 15:443–457 © 2011 John Wiley & Sons, Ltd.
444
DOI: 10.1002/we
P. -Å. Krogstad and J. A. Lund Study of the performance of a model turbine
0.3
0.2
0.1
y /c
0
-0.1
-0.2
-0.3
0 0.2 0.4 0.6 0.8 1
x /c
Figure 1. The NREL S826 profile.
surface roughness, we used a transition amplification ratio of Ncrit D 3:0 (see Drela10 for details). This caused transition
from laminar to turbulent flow to occur closer to the leading edge than in the standard smooth surface, no free-stream
turbulence case. As the local blade Reynolds number is very dependent on the rotational speed of the blade, a wide range
of XFOIL calculations were made to take the Reynolds number sensitivity into account. The airfoil performance at these
low Reynolds numbers is shown in Figure 2. The profile appears to be quite sensitive to Re in the intended operational
range, so it is important that these effects are included in the performance calculations.
The twist and chord distributions obtained are shown in Figure 3. The chord length at the tip is c D25:8 mm. This gives
a stubbier blade than is normally observed on modern wind turbines. However, since the purpose of the blades was to
Figure 2. Airfoil data used for the BEM calculations. Ncrit has been set to 3:0.
0.1 40
0.08 30
0.06 20
c (m)
(o)
0.04 10
0.02 0
0 -10
0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5
r (m) r (m)
(a) (b)
Figure 3. (a) Chord length and (b) twist distributions for the model blade.
445
DOI: 10.1002/we
provide a test case for CFD codes, a large chord length was chosen in order to reduce the Reynolds number effects as much
as possible and to ensure a good geometrical accuracy of the machined blades. Since the aspect ratio of the blade would
be smaller than usual, it would also be a harder test case for prediction models that have problems to account for tip losses
correctly. For the same reason, the tip was therefore cut off squarely.
The blades were machined in aluminum and were mounted on a d D 0:09 m hub. To be able to pitch the blades to
any angle of attack, we made a short transition region to a circular cross section between r=R D 0:109 and r=R D 0:151.
The transition surface was made using a linear interpolation between the two shapes. The shape of the blade is shown in
Figure 4, and Table I gives the distribution of the chord length and twist angle as function of the radius. Together with the
airfoil geometry specification found in Somers,9 this gives a complete blade definition.
(a)
(b)
Figure 4. The model blade. (a) Azimuthal projection. (b) Streamwise projection.
Table I. Blade definition.

r =R c =R ˛
0.016667 0.030000 –
0.050000 0.030000 –
0.10889 0.030000 –
0.12222 0.11000 38.000
0.15000 0.18096 37.055
0.18333 0.17802 32.544
0.21667 0.17114 28.677
0.25000 0.16250 25.262
0.28333 0.15335 22.430
0.31667 0.14434 19.988
0.35000 0.13578 18.034
0.38333 0.12782 16.349
0.41667 0.12050 14.663
0.45000 0.11379 13.067
0.48333 0.10766 11.829
0.51667 0.10207 10.753
0.55000 0.09696 9.8177
0.58333 0.092286 8.8827
0.61667 0.088002 7.9877
0.65000 0.084068 7.2527
0.68333 0.080446 6.5650
0.71667 0.077104 5.9187
0.75000 0.074014 5.3045
0.78333 0.071149 4.7185
0.81667 0.068487 4.1316
0.85000 0.066009 3.5439
0.88333 0.063697 2.9433
0.91667 0.061536 2.2185
0.95000 0.059512 1.0970
0.98333 0.057613 0:71674
446
DOI: 10.1002/we
The experimental investigation was performed in a closed return wind tunnel with a test section that is 2.7 m wide,
2 m high and 12 m long. A picture of the model in the wind tunnel is shown in Figure 5. The model was mounted on a
six-component balance, which enables the three forces and moments to be measured. In this case, only the thrust force
will be reported, which was measured with an uncertainty of 0.5% at the design tip speed ratio. (This uncertainty increased
to almost 2% for the measurement at the lowest setting.) The tunnel speed was monitored by the pressure drop across
the inlet contraction to the test section. The conversion factor between the reference speed at the plane of the rotor and
across the contraction was obtained by a direct calibration. The flow uniformity across the test section was measured to be
constant within ˙0:27% of the mean velocity U1 over the plane covered by the rotor area, and the turbulence intensity
was Iu D 100% u0 =U1 0:3%.
The rotor diameter was chosen to be D D 0:9 m, which is close to the maximum that could be used without producing
significant interference effects from the wind tunnel walls on the thrust and power coefficients (see, e.g. Spera11 ). The
rotor swept area was 11.8% of the wind tunnel cross-section area. In the well-documented NREL/NASA Ames full-scale
wind turbine tests,1 the area ratio was 8.8%. The blockage effect on the reference velocity in these tests were thoroughly
investigated and specified to be ‘substantially less than 1%’. This was further investigated both by Sørensen et al.2 and
Gomez-Iradi et al.12 by means of CFD calculations. Both investigations concluded that ‘the effect of the tunnel walls on
the obtained CP values is, for all practical purposes, negligible’. As in the present case, these tests were performed in a
closed test section. With our model being somewhat larger, relatively speaking, the blockage effect in the present case is
somewhat higher, but is still expected to be small.
The model was designed as a fully operational wind turbine. The rotor was controlled by a 0.37 kW electrical asyn-
chronous motor by using a frequency converter where the speed of rotation could be set. The rotor, which had been
dynamically balanced, could be operated up to about 3000 rpm. The converter allowed power to be supplied to, or con-
sumed from, the motor, depending on the operating condition. When the turbine produced power, this was dissipated in a
300 W electrical heater.
The design of the nacelle is shown in Figure 6. Between the two front bearings of the main support shaft is a five-lead set
of Hottinger slip rings (Hottinger Baldwin Messtechnik GmbH, Darmstadt, Germany). These were used to transmit signals
from a set of strain gauges fitted to one of the blades to enable the root bending moment to be measured (not included
here). At the end of this shaft, an optical positioning device was fitted that was used to generate an electrical ramp signal
that gives the angular position of the rotor. The signal frequency from the position detector may also be used to measure
the speed of rotation. One of the blades was also fitted with a reflecting surface that could be used for remote optical speed
of rotation measurements.
A Hottinger 2 N m torque transducer was mounted between the main shaft and the drive shaft from the motor. It is essen-
tial that the torque transducer is mounted as close to the rotor as possible to reduce shaft frictional losses that may affect
the power coefficient data. During the power measurements, the brushes of the slip rings were therefore removed. Hence,
the only parasitic loss affecting the measured power coefficients are those of the two forward ball bearings. The sensitivity
of the torque transducer was calibrated prior to each test by using calibrated weights applied to the tip of one of the blades
when this was locked in a horizontal position. The torque transducer also had an electronic speed of rotation output. Hence,
Figure 5. Picture of the model turbine.
447
DOI: 10.1002/we
four independent systems could be used to accurately measure the angular speed. The motor was mounted under the tunnel
floor, and the power was transmitted to the rotor shaft by means of a belt drive.
After suitable amplification and filtering, the various signals were logged on a PC using LabView software (National
Instruments, Austin, TX) at 16 bit resolution until a stable mean was obtained.
2.2. Measurements
The blades were laid out so that the airfoil quarter chord was common for all radii and placed along a radius aligned in
the plane of rotation. The blades could be pitched around the quarter chord, and for the tests reported here, the model was
operated at a pitch setting of 0ı and the power was stall controlled. During the experiment, the turbine was operated at
constant wind speed, while the speed of rotation was varied to obtain results for different tip speed ratios (). Therefore,
the Re in the tests depends almost linearly on . The tip Re was thus changing by a factor of about 5, from about 3 104
for 2 to 1:8 105 at D 10.
A set of initial measurements were taken over a range of free-stream velocities to check the Re dependence on the
performance of the turbine. It was found that the power coefficient curves were virtually independent of velocity for
U1 9 m s1 (see Figure 7, which shows a small selection of the data obtained). The majority of the measurements were
therefore performed at a nominal free-stream velocity of U1 10 m s1 , which was also the design velocity.
The peak power coefficient at CP 0:45 was found to be surprisingly high. The model has previously been tested with
blades on the basis of the NACA 632 family, which only gave CP 0:28 at design conditions. These blades were also
found to be much more Re dependent and only gave CP 0:28 after the blades were fitted with tripping devices. This was
not found necessary for the new blades, which were only lightly sanded with fine emery paper after machining to produce
Figure 6. The design of the nacelle.
0.5 1.2
1
0.4
0.8
0.3
CP C T 0.6
0.2
0.4
U inf =7.2ms U inf =7.2ms
0.1 0.2
0 0
0 2 4 6 8 10 12 0 2 4 6 8 10 12
(a) (b)
Figure 7. (a) Experimental power and (b) thrust coefficients. Reynolds number dependence tests.
448
DOI: 10.1002/we
1.3 0.06
0.055
1.2
0.05
1.1 0.045
CL CD 0.04
1 0.035
0.03
0.9 XFOIL (turbulent)
Fluent SST Enhanced wall treatment XFOIL (turbulent)
Fluent RNG Enhanced wall functions 0.025 Fluent SST Enhanced wall treatment
Fluent RNG Standard wall functions Fluent RNG Enhanced wall functions
Fluent RNG Standard wall functions
0.8 0.02
0.01 0.1 1 10 100 1000 0.01 0.1 1 10 100 1000
+ +
y y
(a) (b)
Figure 8. Lift and drag calculations as function of maximum y C at the first grid point. ˛ D 7ı and Re D 105 .
a small amount of surface roughness. Since the experimental setup and all instrumentation were the same, it was concluded
that the new blades were a far better and less Re-dependent design than the previous one.
The thrust data show an almost perfect collapse for all but the lowest velocity. Since the data were to be used for compar-
ison with CFD simulations of the rotor, the drag of the model without the blades was measured. This gave a drag coefficient
of CD D 0:137, which has been subtracted from the data presented in this report.
Further information about the flow within the wake behind the turbine and the sensitivity of the turbine performance on
the wake from an upstream turbine has been reported in Adaramola and Krogstad.7
3. NUMERICAL SIMULATIONS
3.1. 2D simulations
Initially, a set of 2D computations of the flow around the NREL S826 airfoil were performed to see how well the Fluent
and XFOIL software packages would agree and to investigate the required grid distribution. The Fluent CFD simulations
were performed using the fully turbulent k ! shear stress transport (SST) and the k re-normalization group (RNG)
turbulence model options. At the inlet, the approximation D 2k was used, and k was adjusted so that the free-stream
turbulence intensity at the airfoil was about 1%.
The XFOIL results were obtained using a very low critical amplification ratio .Ncrit D 0:01/ in order to imitate the fully
turbulent conditions in the turbulence models used in Fluent.
A grid independency study was made, with a focus on how to reduce the number of grid points without losing accuracy.
A hybrid grid was chosen, with a high-resolution structured boundary layer grid around the airfoil and an unstructured grid
in the outer domain. A geometric growth rate of 1.4 could be used for the unstructured grid without much loss in accuracy.
In the boundary layer grid, the first grid point was always placed below y C D 5 from the airfoil surface, and for most of
the blade, it was significantly closer. This was performed because enhanced wall treatment proved to be necessary to obtain
good predictions of the surface friction. It was also found that the total height of the boundary layer grid should be at least
10% of the chord length, in order to capture the flow gradients in cases when the airfoil stalled. Results from one of the
grid independency studies is presented in Figure 8. Both the standard and enhanced wall functions were used in the k
RNG model calculations.
As observed by Baxevanou and Vlachos,13 it was found that the lift coefficient was not very sensitive to the grid geometry
or the number of grid cells used, whereas the predicted CD was, as expected, much more sensitive.
Grid-independent estimates for CD and CL were obtained, and the k ! SST turbulence model gave results that were
very close to those from XFOIL. The k RNG model with the enhanced wall function option always predicted higher
CD and CL values (Figure 9). It is clearly seen that there is a good match between the results using the SST model and the
XFOIL results for Ncrit D 0:01, which corresponds to the fully turbulent CFD calculations. The k RNG over-estimates
CD by almost 30% at the operating angle of attack (Figure 8(b)) and does not describe the stall behavior of the airfoil in an
adequate manner (Figure 9). Hence, the k ! SST model was used for the three-dimensional (3D) simulations, which is
also the method recommended by many others, e.g. in the review paper by Hansen et al.6
These 2D calculations gave confidence not only to the data that had been used in the BEM to design the rotor but also to
the choice of 3D grid system to be used in the fully 3D Fluent calculations.
449
DOI: 10.1002/we
As the Reynolds number is quite low, the effect of transition over the airfoil is expected to have an influence of the wind
turbine performance. The transition process depends on Reynolds number as well as the free-stream turbulence level and
surface roughness. The Reynolds number and turbulence level effects were studied extensively by Devinant et al.14 who
determined the characteristics of the NACA 654 -421 airfoil for chord-based Reynolds numbers ranging from 1 105 to
7 105 and turbulence levels from 0.5% to 16%. The study showed that these effects are most evident at angles of attack
corresponding to stalled operation. Significant increases in maximum lift coefficients and reduction in drag were found for
certain combinations of Reynolds numbers and turbulence intensities.
On the basis of the turbulence level in the wind tunnel, a critical amplification ratio of 3.0 was chosen in the XFOIL
calculations to imitate the expected transitional effects in the experiment. The effect of laminar flow over the forward part
of the airfoil is clearly seen in Figure 9 as a significant reduction in drag and a slight increase in the lift in the normal
operating range of the airfoil. This increase in L=D ratio will not be included in the CFD data, which do not incorporate a
transition model. Therefore, the CFD predictions could be expected to under-predict the wind turbine performance slightly
compared with the BEM design data.
3.2. 3D CFD calculations
By taking advantage of the periodic planes to reduce the computational domain to one-third of the rotor, we performed
the 3D simulations for the model turbine in a rotating frame of reference. The geometry of the tower was therefore not
included in the calculations. The computational domain extended 4:5D upstream and 7:8D downstream of the rotor plane,
which corresponds closely to the streamwise dimensions of the wind tunnel test section. Because the computations were
performed in cylindrical coordinates, the exact boundary conditions of the wind tunnel walls could not be used. Instead, the
radius of the computational domain was adjusted so that the flow area was the same as in the wind tunnel test section. This
is the same methodology that was used by Sørensen et al.2 for their NREL Phase VI predictions. The boundary condition
at the outer surface was specified as a wall. In this way, it was hoped that some of the effects of the finite flow domain that
could influence the measurements from the model might also be included in the computations.
The blade geometry was extracted from the CAD file generated by the company that machined the blades. This assured
that if there were any differences between the design and the actual model geometry, this would be included in the compu-
tations. It also allowed an accurate description of the geometry that the company had generated to taper the blade towards
the cylindrical section at the root used to fix the blade to the hub. The geometry, including the nacelle (which consists of a
hemispherical hub and a cylindrical main body, L D 6:7d D 60 cm long), was meshed using the Gambit software package.
A structured grid was used for the pressure and suction sides of the blade surface, with a total of 200 cells in the span-
wise direction and 300 cells around the blade in the chordwise direction. The tip and root section of the blade, as well as
the nacelle and hub, were meshed using triangular cell faces. About 100,000 faces were used to describe one blade and
one-third of the nacelle. Figure 10 shows the root section of the grid constructed.
A boundary layer grid was attached to the wall boundaries using the TGrid software (ANSYS, Inc., Canonsburg, PA),
and grid height and growth rates were adjusted in accordance with the experiences drawn from the grid sensitivity studies
made in 2D. The first calculation node was set 0.03 mm above the blade surface to keep y C below 5 everywhere along the
blade surface. Inspections of the results showed that y C stayed below 5 in the leading edge region and below 3 on the rest
of the blade for all simulations, giving confidence that the enhanced wall treatment was suitable for the grid.
2 2
1.5 1.5
1 1
CL CL
0.5 0.5
XFOIL N crit = 3.0
XFOIL N crit = 3.0
XFOIL N crit = 0.01 (turbulent)
0 XFOIL N crit = 0.01 (turbulent) 0
Fluent SST (turbulent)
Fluent SST (turbulent)
Fluent RNG (turbulent) Fluent RNG (turbulent)
-0.5 -0.5
-10 -5 0 5 10 15 20 25 30 35 0 0.05 0.1 0.15 0.2 0.25
CD
(a) (b)
Figure 9. CL and CD calculated using XFOIL and Fluent. Re D 105 .
450
DOI: 10.1002/we
Figure 10. Section of the grid used for 3D calculations.
The volume grid was meshed using an unstructured tetrahedral grid, and pyramidal cells were used to obtain a smooth
transition between the quadrilateral cells in the boundary layer grid and the rest of the volume grid. A total of 3:5 106
cells were used to mesh the computational volume.
The SIMPLER algorithm was used to couple the pressure and velocity fields, and the k ! SST model was used to
compute the turbulent eddy viscosity. The momentum equations were solved using a first-order upwind scheme for the
convective terms, and first-order pressure interpolation was used.
The calculation parameters were adjusted to match the experimental settings by keeping the inflow velocity at 10 m s1
while varying the rotational speed so that results for tip speed ratios of D 3; 4; 5; 6; 7; 9 and 11 were obtained. Air viscosity
and density were estimated before the experimental results were available, resulting in a Reynolds number on the basis of
rotor diameter and free-stream velocity, which was 3.4% lower than in the experiments. On the basis of the Reynolds
number sensitivity tests performed, this is not expected to add any significant bias to the results.
4. RESULTS AND DISCUSSIONS

An initial set of calculations of the performance for the model turbine was undertaken using the BEM method. These cal-
culations were performed in the same way as the model was to be tested, i.e., with a fixed free-stream velocity, which was
set to give the correct tip Re at design conditions. From a large library of lift and drag data as function of Re generated
using XFOIL, the Re dependence along the blade span was accounted for. In the same way, the -dependent Re variations
were included in the calculations. The computations were performed for tip speed ratios from D 2 up to where the power
coefficient would first become negative. This showed that the operational range of the NREL S826 profile used would be
from ˛ D 5 to 35ı angle of attack. The first indications of stall were predicted to occur near the root at D 4, and the
blade was found to be fully stalled at D 3. The XFOIL predictions showed that at the design Re D 105 , the profile will
have zero lift coefficient at ˛ 4ı . This point was passed when > 9, and as a consequence, the inner part of the turbine
then starts to feed energy into the flow while the outer section is still extracting energy.
Because of the small scale of the model turbine and the complicated shape of the blade, it is difficult to set the blade
angles with sufficient accuracy. To study the effects of blade pitch misalignment, we therefore performed calculations for
blade settings pitched at ˙1ı with respect to the design angle of attack. The results are shown in Figure 11 and demon-
strate that possible misalignment errors will affect the thrust coefficient data most severely. The power coefficient is most
sensitive to misalignment near the best point of operation, whereas the thrust coefficient appears to be very sensitive at high
tip speed ratios. It is therefore essential to align the blades properly. Therefore, a special alignment rig was made for the
blades. Using this tool, we estimated that the blades were aligned with an uncertainty of less than 0:25ı .
Figure 12 shows the measured power and thrust coefficients as function of the tip speed ratio (data given in Table II).
Included in the graphs are also the design predictions from the BEM method and the CFD calculations.
The power coefficient predicted by the BEM method and the CFD calculations follow the same distribution. As expected,
the differences are smallest close to the design tip speed ratio, where the turbine is operating under normal conditions
451
DOI: 10.1002/we
0.5 1.5
=0
=0
= +1
0.4 = +1
= 1
= 1
1
0.3
CP CT
0.2
0.5
0.1
0 0
0 2 4 6 8 10 12 0 2 4 6 8 10 12
(a) (b)
Figure 11. Performance predictions for three pitch angle settings using the BEM method. Ncrit D 3:0.
0.5 1.4
Measured Measured
BEM BEM
Fluent 1.2 Fluent
0.4
1
0.3 0.8
CP CT
0.6
0.2
0.4
0.1
0.2
0 0
0 2 4 6 8 10 12 0 2 4 6 8 10 12
(a) (b)
Figure 12. Comparisons of power and thrust coefficients from CFD and BEM with the measurements.
(5 < < 7). For lower tip speed ratios, significant stall effects are present close to the root of the blade that violates the
assumption of 2D flow that is the very basis of the BEM method. On the basis of previous studies of rotational augmen-
tation (e.g. Ronsten15 ), it is expected that the BEM model under-predicts the forces in the inner part of the blade for low
. Despite the fact that corrections for this effect has not been included in our BEM method, it appears to perform well in
predicting CP for low . This could be an indication that the airfoil data used are too optimistic for high angles of attack.
At higher tip speed ratios, the flow is expected to be largely 2D over most of the blade, so good correspondence between
the BEM and CFD prediction methods is expected. The small differences between the estimated CP by the BEM and CFD
methods found at high tip speed ratios ( > 7) could be related to the differences in the assumptions about the way the
boundary layer develops, e.g. whether the flow is turbulent from the leading edge or develops more naturally. For moderate
angles of attack, this affects primarily the drag coefficient and to a limited degree the lift, as seen in the XFOIL calculations
shown in Figure 9. (The data labeled Ncrit D 0:01 corresponds to transition right at the leading edge, whereas Ncrit D 3:0
predicts the expected natural transition.) The increase in drag in the fully turbulent CFD simulations will have an influence
that will be more dominant when the tip speed ratio increases, as the effective velocity gets more aligned with the plane of
rotation. This causes a reduction in the blade torque and hence in CP .
There are significant differences between the two model predictions for CT (Figure 12(b)). The CFD results are found to
be in close agreement with the measurements for most tip speed ratios, whereas the BEM method under-predicts the thrust
forces for all but the highest tip speed ratios. The low BEM predictions of CT cannot be explained by a possible misalign-
ment of the blades. If the blade angle was offset by a few degrees, this would not significantly influence the predictions
of CP , as shown in Figure 11(a). However, it will have a significant effect on the CT data, but mainly at high where the
BEM predictions are in closest agreement with the measurements (Figure 12(b)).
The under-prediction could therefore be an effect of finite domain size, since the measurements were preformed in a
wind tunnel where blockage effects are always present to some extent. This effect is included in the CFD calculations
because the computational domain cross-section area was similar to that of the wind tunnel but is not part of the BEM
predictions. If this obstruction of the flow by the turbine is significant, this will increase the thrust force. However, if this
452
DOI: 10.1002/we
Table II. Measured power and thrust coefficients.

CP CT
1.0000 0.014205 0.15800

1.5000 0.025568 0.20400
2.0000 0.046402 0.26400
2.5000 0.071023 0.32500
3.0000 0.12027 0.39300
3.5000 0.29735 0.52800
4.0000 0.37405 0.64500
4.6000 0.42235 0.74400
5.1000 0.43182 0.78900
5.6000 0.43655 0.84100
6.1000 0.43750 0.89100
6.6000 0.43182 0.93400
7.1000 0.41856 0.96700
7.6000 0.39489 1.0040
8.1000 0.36080 1.0310
8.6000 0.32008 1.0560
9.2000 0.26894 1.0820
9.6000 0.21212 1.1030
10.200 0.14867 1.1210
10.700 0.079545 1.1370
11.200 0.0075758 1.1400
11.300 0:013258 1.1440
was, the case it appears reasonable to assume that the blockage effect will be most significant at high , where CT is at its
highest. The figures show that the discrepancy is largest around D2, where the blade is fully stalled and the thrust force
is indeed very low. It is therefore unlikely that the under-prediction of CT can be attributed to blockage effects. It will be
shown later that the BEM method generally predicts too low forces in the streamwise direction and fails to account for the
significant increase in lift caused by 3D effects that are present, as the region of separation starts to develop near the root
of the blade when decreases.
There may also be other effects that limit the predicted lift coefficient of the blade and thus underestimate the total thrust
force on the blade at all but the highest . As pointed out in Section 3.1, one such possibility is that transition effects are
not captured correctly in the database used in the BEM model. This is an effect that will be Reynolds number dependent
and therefore most significant for low .
Figure 13 shows a pair of vector plots at D3, taken roughly at the mid-span of the blade and near the tip. It is apparent
that the whole blade operates in a deep stall mode over the entire suction surface at this tip speed ratio. As expected, the
vector plots for D 6, which is the design tip speed ratio, show that the flow is well behaved at all spanwise stations
(Figure 14). It is also interesting to observe that the CFD has captured the small separated region near the upper surface
trailing edge. This is the separation ramp that has been designed into the profile by NREL to increase the lift coefficient
at normal operating conditions by generating more upper surface suction. It also controls the movement of the separation
point at high angles of attack and has the additional benefit of causing a gentle stall characteristic, as shown in Figure 9.
Local force coefficients across the blade were extracted from the CFD predictions in order to obtain a better understand-
ing of the differences between the two models used to predict the wind turbine performance. Despite the two methods
producing very similar estimates for CP and CT , it is clear from Figure 15 that there are large differences in the com-
puted spanwise distribution of forces acting on the blade. The presence of rotational lift augmentation for D3 is evident,
increasing both the thrust and torque forces acting on the inner sections of the blade. This effect is highly 3D, caused by
spanwise flow on the blade surface, and is therefore not captured by the BEM method.
The sharp drop in the forces predicted close to the tip of the blade by the CFD method is caused by a region of sepa-
rated flow due to a combination of high angle of attack and the square geometry of the tip. This effect cannot of course
be reproduced in a BEM method. It is also obvious from the figure that at this operational condition, the Prandtl tip loss
correction in the BEM method produces far too strong effects. While the BEM fails to give the correct spanwise torque dis-
tribution (Figure 15(a)), it is at least roughly correct in the important outer part of the blade. For the thrust force, however,
it consistently under-predicts the forces (Figure 15(b)).
The agreement between the methods is much better near the design condition, where the assumptions inherent in the
BEM method are better justified. When D 7 (Figure 16), the agreement between the computed load distributions is
acceptable, mainly because of the absence of the severe 3D effects found at lower tip speed ratios. Figure 16(b) shows
that Prandtl tip loss factor now gives a good description of the loss of circulation close to the tip of the blade. However,
453
DOI: 10.1002/we
(a) (b)
Figure 13. Vector plots at D 3. (a) r =R D 0:44; (b) r =R D 0:89.
(a) (b)
Figure 14. Vector plots at D 6. (a) r =R D 0:44; (b) r =R D 0:89.
0.3 0.6
BEM =3 0.5
0.25 CFD =3
0.4
2dF / U R
0.2
2dF / U R
2
2
0.3
0.15
x
z
0.2
0.1
0.1
0.05 0 BEM =3
CFD =3
0 -0.1
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
r /R r /R
(a) (b)
Figure 15. Force distributions along the blade in (a) tangential and (b) streamwise direction. D 3.
the BEM method is again clearly under-predicting the thrust forces, as was demonstrated in Figure 12(b). There is better
correspondence between the tangential forces acting on the blade in this case, but the CFD method predicts a somewhat
steeper tip loss effect in the tangential forces than the BEM method.
The fact that the Fluent results have been obtained using a fully turbulent description of the boundary layer, while the
model was laid out from BEM results obtained using a transitional model, should give a lower CP value at this tip speed
ratio in the CFD, since CD is increased in the fully turbulent predictions (Figure 9(b)). This is clearly not the case, as the
CFD predicts a higher tangential force over most of the blade. The effect must therefore be due to the spanwise interaction,
which is not included in the BEM method. Three-dimensional flow structures were observed both near the root and tip of
the blade also for the design tip speed ratio. As both methods fail to predict the torque acting on the blade for high tip
speed ratios, it appears that the local forces acting on the blade are not well described by any of the methods. If an incorrect
454
DOI: 10.1002/we
0.2 2.5
2
0.15
2dF / U R
2dF / U R
1.5
2
2
0.1
x
z
0.05 BEM =7
CFD =7 0.5 BEM =7
CFD =7
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
r /R r /R
(a) (b) =
Figure 16. Force distributions on the blade in (a) tangential and (b) streamwise direction. D 7.
2 0.5
XFOIL XFOIL
XFOIL (turb) XFOIL (turb)
r =0.25 r =0.25
1.5 r =0.50 0.4 r =0.50
r =0.65 r =0.65
r =0.80 r =0.80
r =0.95 r =0.95
1 0.3
CL Cd
0.5 0.2
0 0.1
-0.5 0
-10 -5 0 5 10 15 20 25 -10 -5 0 5 10 15 20 25
(a) (b)
Figure 17. CL and CD derived from CFD results at various spanwise positions for D 3; 4; 5; 6; 7; 9 and 11 compared with XFOIL
predictions.
description of transitional effects is the main cause of the errors, the load distribution on the blade will be affected by this,
not only by a change in sectional lift and drag coefficients of the airfoil, but also by the observed flow three dimensionality
introduced. That transition has a significant effect on the flow three dimensionality, and therefore, the load distribution was
clearly demonstrated in the predictions of the NREL Phase VI performance undertaken by Sørensen.16
By inspecting the average axial velocity in the rotor plane, we can determine the axial induction factor by using the
reduced axial velocity method described by Johansen and Sørensen.17 In this case, the angular velocity was also investi-
gated, but found to be too low to have any significant effect on the results. With knowledge of the local rotational speed
and blade twist angle, the local angle of attack can be determined. Using the calculated blade forces, we can determine the
local lift and drag coefficients and compare their values with the 2D input of the BEM method.
The results (Figure 17) agree well with the earlier findings of Johansen and Sørensen.17 They compared predictions
from the ElipSys3D program package with the NREL Phase VI wind turbine data. There is a good agreement between the
fully turbulent XFOIL predictions and the lift coefficients from the CFD for all but the root and tip sections (r=R D 0:25
and 0.95 in the figure). The lower lift slope for these spanwise positions in the CFD data is due to finite span effects (see,
e.g. Bertin and Cummings18 ). The increased lifting force caused by rotational augmentation at D3 is clearly seen as the
very high local CL of almost 2:0 for the r=RD 0:25 section in the upper right-hand corner of Figure 17(a).
There are larger differences in the drag coefficients derived, and it is seen that CD is generally higher than the 2D
reference data, especially at low tip speed ratios (high angles of attack, ˛).
5. CONCLUSIONS
The investigation presented shows that the standard BEM method is capable of giving a good impression of the perfor-
mance of a wind turbine. In addition to the approximations built into the method, the performance of the method obviously
relies critically on how well the performance of the airfoil sections applied are known. In the present investigation, the
performance of the model turbine was calculated by assuming natural transition. Compared with the data obtained from
455
DOI: 10.1002/we
wind tunnel tests, it was found that the general shape of the power and thrust coefficient curves were correctly predicted,
but the thrust coefficient was consistently lower than the measurements over most of the operating tip speed range whereas
the power coefficient prediction was too high for high tip speed ratios. At tip speed ratios below the design value, the blade
element method predictions of the power coefficient agreed well with the measurements. The high CP estimated at high
indicates that XFOIL may have under-predicted the drag force on the airfoil, whereas the consistent under-prediction of the
thrust coefficient is likely to be due to a combination of under-prediction of the drag and a possible blockage effect from
the wind tunnel walls, which is not included in the BEM calculations. That XFOIL predicts too low drag in this case was
verified by the 2D CFD simulations performed using the Fluent software. Fluent consistently predicted about the same lift,
but higher drag.
The turbine performance was also estimated using high-resolution, fully 3D CFD simulations by means of the Fluent
software with the k ! SST turbulence model. To give a good representation of one-third of the flow domain, and by
taking advantage of the geometrical symmetry, we found about 3:5 106 grid points to be necessary. Although these cal-
culations did not produce perfect predictions of the performance, these calculations were in general much closer to the
measured values. This is especially true for the thrust coefficient, which matched the data of the model turbine throughout
the operational range.
While the blade element method computes the performance of a turbine operating in an infinite field, the computational
domain for the 3D CFD was closely matched to that of the wind tunnel test section, both with respect to width and length.
The good agreement with the measured thrust force and the consistently low values from the blade element method may
suggest that the experiment is somewhat affected by wind tunnel wall effects.
On the basis of the information obtained from this investigation, we have demonstrated that the Fluent CFD software
package is able to predict the wind turbine performance at model scale quite well. This is important information since
good agreement at model scale will give confidence to the calculations when the turbine performance is to be estimated
at full-scale conditions. Although the BEM method also predicted the behavior of the turbine well, it has, as expected, a
much higher degree of uncertainty. It was demonstrated that some of the shortcomings are caused by the 3D flow pattern,
which develops along the blade span when the turbine is operated far from its design tip speed ratio. The effect is partly
due to rotational lift augmentation and partly by 3D effects caused by transition. Neither effect is included in a standard
blade element prediction.
REFERENCES
1. Simms D, Schreck S, Hand M, Fingersh LJ. NREL unsteady aerodynamics experiment in the NASA-Ames wind tunnel:
comparison of predictions to measurements. NREL / TP-500-29494, NREL, 2001.
2. Sørensen NN, Michelsen JA, Schreck S. Navier–Stokes predictions of the NREL Phase VI rotor in the NASA Ames
80-by-120 wind tunnel AIAA, AIAA-2002-0031, 2002.
3. Snel H, Schepers JG, Montgomerie B. The MEXICO project (Model Experiments in Controlled Conditions): the
database and first results of data processing and interpretation. Journal of Physics: Conference Series 2007; 75: 012014.
4. Schreck S, Sant T, Micallef D. Rotational augmentation disparities in the MEXICO and UEA Phase VI Experiments
NREL/CP-500-47759, NREL, 2010.
5. Vermeer LJ, Sørensen JN, Crespo A. Wind turbine wake aerodynamics. Progress Aerospace Sciences 2003; 39:
467–510.
6. Hansen MOL, Sørensen JN, Voutsinas S, Sørensen NN, Madsen HA. State of the art in wind turbine aerodynamics and
aeroelasticity. Progress Aerospace Sciences 2006; 42: 285–330.
7. Adaramola MS, Krogstad P-Å. Wind tunnel simulation of wake effects on wind turbine performance. Proceedings of
European Wind Energy Conference, Warsaw, Poland, 2010; 64 – 67.
8. Manwell JF, McGowan JG, Rogers AL. Wind Energy Explained. John Wiley & Sons: New York, 2002.
9. Somers DM. The S825 and S826 Airfoils. National Renewable Energy Laboratory 2005. NREL/SR-500-36344.
10. Drela M. Xfoil v. 6.97. [Online]. Available: http://web.mit.edu/drela/Public/web/xfoil/ [accessed 2008].
11. Spera DA. Wind Turbine Technology: Fundamental Concepts of Wind Turbine Engineering. ASME Press: New York,
1994.
12. Gómez-Iradi S, Steijl R, Barakos GN. Development and validation of a CFD technique for the aerodynamic analysis
of HAWT. Journal of Solar Energy Engineering 2009; 131: 031009.
13. Baxevanou CA, Vlachos NS. A comparative study of numerical schemes and turbulence models for wind turbine
aerodynamics modelling. Wind Engineering 2004; 28: 275 – 290.
14. Devinant P, Laverne T, Hureau J. Experimental study of wind-turbine airfoil aerodynamics in high turbulence. Journal
of Wind Engineering and Industrial Aerodynamics 2002; 90: 689 –707.
456
DOI: 10.1002/we
15. Ronsten G. Static pressure measurements on a rotating and a non-rotating 2.375 m wind turbine blade—comparison
with 2D calculations. Journal of Wind Engineering and Industrial Aerodynamics 1992; 39: 105 –118.
16. Sørensen NN. CFD modelling of laminar–turbulent transition for airfoils and rotors using the Re model. Wind
Energy 2009; 12: 715 – 733.
17. Johansen J, Sørensen NN. Aerofoil characteristics from 3D CFD rotor computations. Wind Engineering 2004; 7:
283 – 294.
18. Bertin JJ, Cummings RM. Aerodynamics for Engineers, 5th ed. Prentice Hall: Englewood Cliffs, 2009.
457
DOI: 10.1002/we

Krogstad 2011

Uploaded by

Copyright:

Available Formats

Krogstad 2011

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Krogstad 2011

Uploaded by

Copyright:

Available Formats

WIND ENERGY

Wind Energ. 2012; 15:443–457

An experimental and numerical study of the

Copyright © 2011 John Wiley & Sons, Ltd. 443

Figure 1. The NREL S826 profile.

Table I. Blade definition.

Figure 5. Picture of the model turbine.

Figure 6. The design of the nacelle.

3.2. 3D CFD calculations

Figure 9. CL and CD calculated using XFOIL and Fluent. Re D 105 .

Figure 10. Section of the grid used for 3D calculations.

4. RESULTS AND DISCUSSIONS

Table II. Measured power and thrust coefficients.

1.0000 0.014205 0.15800

Figure 13. Vector plots at D 3. (a) r =R D 0:44; (b) r =R D 0:89.

Figure 14. Vector plots at D 6. (a) r =R D 0:44; (b) r =R D 0:89.

You might also like