Received: 3 July 2017 Revised: 5 September 2019 Accepted: 10 September 2019

DOI: 10.1002/we.2419

RESEARCH ARTICLE

A new tip correction for actuator line computations

Kaya Onur Dağ Jens Nørkær Sørensen

DTU Wind Energy, Technical University of

Denmark, Lyngby, Denmark Abstract
The actuator line method (ALM) is today widely used to represent wind turbine loadings in
Correspondence
Kaya Onur Dag, ̆ DTU Wind Energy, Technical computational fluid dynamics (CFD). As opposed to resolving the whole blade geometry, the
University of Denmark, 2800 Kongens Lyngby, methodology does not require geometry-fitted meshes, which makes it fast to apply. In ALM,
Denmark.
tabulated airfoil data are used to determine the local blade loadings, which subsequently are
Email: [email protected]
projected to the CFD grid using a Gaussian smearing function. To achieve accurate blade loadings
at the tip regions of the blades, the width of the projection function needs to be narrower than
the local chord lengths, requiring CFD grids that are much finer than what is actually needed in
order to resolve the energy containing turbulent structures of the atmospheric boundary layer
(ABL). On the other hand, employing large widths of the projection function may result in too
large tip loadings. Therefore, the number of grid points required to resolve the blade and the
width of the projection function have to be restricted to certain minimum values if unphysical
corrections are to be avoided. In this paper, we investigate the cause of the overestimated
tip loadings when using coarse CFD grids and, based on this, introduce a simple and physical
consistent correction technique to rectify the problem. To validate the new correction, it is first
applied on a planar wing where results are compared with the lifting-line technique. Next, the
NREL 5-MW and Phase VI turbines are employed to test the correction on rotors. Here, the
resulting blade loadings are compared with results from the blade-element momentum (BEM)
method. In both cases, it is found that the new correction greatly improves the results for both
normal and tangential loads and that it is possible to obtain accurate results even when using a
very coarse blade resolution.

KEYWORDS

actuator line, LES of wind turbines, tip correction, wind turbine blade loadings

1 INTRODUCTION

The actuator line method (ALM) was developed as a numerical technique to facilitate the representation of the rotor blades in computational
fluid dynamics (CFD) simulations of wind turbines.1 The ALM is based on a blade-element approach in which body forces are introduced in the
Navier-Stokes equations along lines representing the blades of the wind turbine. It was originally developed for simulating the wake behind
a single horizontal axis wind turbine,1-4 and has later been extended to simulate flows in wind farms,5-12 and applied to vertical axis wind
turbines,13,14 tidal turbines,15-17 and helicopters.18 For a review of the ALM and related methods, the reader is referred to Breton et al.19 In the
ALM, the body forces are obtained from tabulated airfoil data employing the local angle of attack along the blades as input. The local angle of
attack is defined from the local relative velocity, which is determined at each time step while running a simulation. After having determined the
local forces, they are projected from the lines representing the rotor blades to the CFD grid points. A main issue of the technique is that it is
required to smear out the body forces in order to avoid discontinuities in the resulting force distribution. This is typically carried out by using a 3D
Gaussian filtering function with a given projection width 𝜖 , which eventually defines the thickness of the volume on which the blade forces are
projected. It is known that very small 𝜖 values cause oscillations in the flow field, whereas large values result in inaccurate loading distributions. 20
There has been a fair amount of work on determining the optimum value of the 𝜖 parameter. Troldborg20 suggested that fixing 𝜖 to twice the
CFD grid resolution (Δ) is a good trade-off between having high accuracy and avoiding oscillations in the flow field. In the work by Jha et al,21 it

Wind Energy. 2019;1–13. wileyonlinelibrary.com/journal/we © 2019 John Wiley & Sons, Ltd. 1
2 DAĞ AND SØRENSEN

was shown that oscillations could be avoided by using a blended dissipative scheme with 𝜖 = 2Δ. It was also noted that a constant 𝜖 parameter
throughout the blade-span causes too large blade tip loadings. Additionally, Jha et al21 introduced a new methodology to vary the 𝜖 parameter
throughout the blade span in order to achieve a better loading distribution around the tip region. Their methodology improves the estimated
loading distributions, but, as usual for ALM, it requires case-specific tuning of the parameter. In the work by Shives and Crawford,22 the effect
of the 𝜖 parameter and the resolution of the CFD grid was investigated for a fixed wing. For this case, it was concluded that the 𝜖 parameter
should be in the order of 1∕4c, where c is the chord length of the local airfoil section. Further, they stated that, in order to be consistent with
the vortex line approach, the self-induction at the vortex center should be zero. Their work also showed that 𝜖 values smaller than 4Δ do not
satisfy this condition. In a recent work by Martínez-Tossas et al,23 a theoretical model based on the potential flow over a Joukowsky airfoil
was developed to determine the optimum value of 𝜖 . Here, it was concluded that for a grid resolution larger than the chord length, the choice
of the scale of the smoothing kernel is dictated by the mesh, whereas for finer discretizations, a value 𝜖 = 0.2c should be employed, which is
in fact very close to the guidelines by Shives and Crawford.22 Recently, the problem of correcting the influence of the regularization function
and the mollification of lifting line/actuator line models have been addressed by Jha and Schmitz,24 Caprace et al,25 and Martínez-Tossas and
Churchfield.26
Since the ALM is based on solving the three-dimensional Navier-Stokes equations around the actuator line, in principle, it is not needed to
introduce a tip correction. However, when crude meshes are used there is no guarantee that the loading tends to zero at the tip, and therefore,
it is sometimes seen that researchers resort to the Prandtl/Glauert tip correction, which essentially is a technique for correcting the loading for
finite number of blades effects in the momentum theory (see Sørensen27 ). A technique for correcting the use of airfoil data near the tip was
introduced by Shen et al28 by multiplying the lift coefficient by a function resembling the Prandtl tip correction. This technique certainly improves
the loading near the tip of the wing, but it is based on an empirical expression that may not be valid for all wing plan forms. An alternative
correction to the one by Shen et al28 is the decamber correction introduced by Sørensen et al.29 This correction is based on the fact that the
induction from the wake causes a curved streamline along the chord line that effectively corresponds to an additional camber of the airfoil. The
correction was shown to be important for constant chord wings but not significant for wing forms of modern wind turbines. Although there
has been a lot of progress in determining the optimum 𝜖 parameter and avoiding too high loadings near the tip when employing crude mesh
resolutions, there is still a need, as will be shown in the following, for deriving a consistent technique to determine mesh-independent solutions
of the ALM. The aim of the present work is to develop and validate such a technique.
The model we propose is based on the analogy there exists between the actuator line model and the inviscid lifting line model. Indeed, at
very high Reynolds numbers, where viscosity essentially does not play any role, the two models are expected to produce similar results, the main
difference being that the actuator line model solves the full flow field whereas the lifting line model is based on computing induced velocities
using the Biot-Savart law. In the lifting line model the wing is represented by a line consisting of a distribution of bound vortices. Each change in
the strength of the bound vortices results in the formation of a free trailing wake vortex, which is convected downstream by the local velocity
field in the wake. In the lifting line model, the wake vortices induce a velocity field on the wing that alters the inflow and which ensures a smooth
distribution of the circulation near the tip that eventually goes to zero at the very tip. Ideally, a solution to the Euler or Navier-Stokes equations
employing the ALM should result in the same circulation distribution and wake vortex system as the lifting line model. However, the size of
the smoothing kernel in the filtering function dictates the minimum resolution of the vortex kernel and acts in principle as a viscous inner core.
This modifies the induction of the wake vortices in the ALM as compared with the same wake obtained using a lifting line technique (LLT). The
technique that we propose is a relatively simple correction to the induced angle of attack to compensate for the added influence of the viscous
core. Although the correction is a method for correcting the additional induction of the smoothing kernel, we refer to it as a tip correction, as it is
at the tip that it is most pronounced.
The paper is structured as follows. In Section 2, we briefly introduce the ALM and the filtering approach. Next, in Section 3, we show some
results using ALM in its present formulation and analyze here the influence of the kernel size of the smoothing function. In Section 4, we
introduce the proposed correction technique, and in Section 5, we show some results for both planar wings and rotors to validate the method.
Finally, in Section 6 we conclude the work.

2 NUMERICAL METHODS

2.1 Large eddy simulation tool

The numerical tool used in this study is an incompressible pseudo-spectral large eddy simulation (LES) solver that has recently been developed at
DTU Wind Energy.30 The code uses Fourier series to discretize the horizontal directions and a second-order central finite difference scheme in the
vertical direction. The pseudo-spectral computation of the nonlinear terms in the momentum equations are fully dealiased with a zero-padding
operation. For the time integration, a second-order Adams-Bashforth scheme is used with a pressure-splitting technique, where the momentum
equations are used to advance in time with the pressure field from the previous time step. 31 The Poisson equation is subsequently solved directly
to obtain the new pressure to project the velocities to a divergence free field. For the subgrid-scale (SGS) modeling, a Smagorinsky model with a
constant of Cs = 0.18 is employed for the current study.32 Since Fourier series enforce periodicity in the spanwise and streamwise directions, a
buffer zone is used to gradually adjust the streamwise outgoing velocity to the desired, uniform, inflow condition. The length of the buffer zone
DAĞ AND SØRENSEN 3

is selected as 15% of the streamwise domain length.5 For the bottom and top boundaries, a stress-free condition is used. The forcing term from
the wind turbine rotor is added as an additional term to the the momentum equations as

𝜕U
= RHS + f b . (1)
𝜕t

RHS here represents the advective, diffusion, and pressure terms in the momentum equations and fb is the body force obtained with ALM. The
calculation of this term is detailed below. For more details about the code, we refer to the PhD dissertation of Dag.30

2.2 The actuator line technique

In ALM, the turbine blades are represented as distributed body forces on a line that is assumed to be located at the quarter-chord line of the
blade, as illustrated in Figure 1. The velocity of the overall flow field is described in a global coordinate system, (x, y, z), with the origin located at
the center of the wind turbine rotor, the x-axis pointing downstream along the axis of the rotor, and the y- and z-axes perpendicular to this, as
illustrated in Figure 1. With the z-axis pointing upward in the vertical direction, an angle 𝜃 defines the azimuthal position of one of the blades,
with the three blades located an angle Δ𝜃 = 2𝜋∕3 from each other. To determine the relative velocity, as it is seen from the rotor blade, a local
corotating coordinate system (r, 𝜃, x) is utilized. This coordinate system is located in the rotor plane at a radial distance, r, from the rotor axis, with
the x-axis pointing in the same direction as in the global system and with 𝜃 defining the azimuthal coordinate direction. To calculate the body
forces, tabulated airfoil data and the angle of attack are used in a blade-element approach.
First, the velocities from the flow solver are interpolated to the velocity values on the actuator line points, referred to as (ux , uy , uz ). With Ω
denoting the rotational speed of the rotor, r the blade radius, the local relative azimuthal velocity of the blade is given as

u𝜃 = Ωr − (uy cos(𝜃) + uz sin(𝜃)). (2)

Note that, for simplicity, the rotor plane is assumed to be perpendicular to the flow. The flow angle for each actuator line point is obtained as

𝜙 = tan−1 (ux ∕u𝜃 ), (3)

which together with the local twist and the pitch angle of the blade (𝛾 ) is used to obtain angle of attack values (𝛼 ) for each actuator line point

𝛼 = 𝜙 − 𝛾. (4)

The lift and drag forces for unit span is obtained with

1
fL = 𝜌 u2rel c CL , (5)
2
1
fD = 𝜌 u2rel c CD , (6)
2
√
where urel = u2𝜃 + u2x is the local relative velocity, c is the chord length, and CL and CD are the local lift and the drag coefficients, respectively. At
this stage, these forces are projected to tangential and normal components by using the local flow angles as

f𝜃 = fL sin(𝜙) − fD cos(𝜙), (7)

fn = fL cos(𝜙) + fD sin(𝜙). (8)

FIGURE 1 Illustration of the actuator line model (left) and local angles of a blade cross section (right)
4 DAĞ AND SØRENSEN

Then in grid coordinates the forces read

fx = −fn , (9)

fy = −f𝜃 cos(𝜃), (10)

fz = f𝜃 sin(𝜃). (11)

Applying these forces directly to the closest CFD grid point would produce discontinuities. To avoid this, a filtering procedure is applied by using
a 3D Gaussian filtering function, which is a function of the distance d that is determined individually for each grid-actuator line point couple as

√
dN,m
i,j,k
= (xi,j,k − xN,m )2 + (yi,j,k − y N,m )2 + (zi,j,k − zN,m )2 . (12)

Here, (i, j, k) are indices that refer to the grid points in the global coordinate system, and superscripts N and m represent the blade and actuator
line point indices, respectively. The 3D Gaussian filter kernel reads

( N,m )2
⎡ d ⎤
1 ⎢− i,j,k ⎥.
𝜂i,j,k
N,m
= exp (13)
3
𝜖3𝜋 2 ⎢ 𝜖 ⎥
⎣ ⎦

With the known kernel, the body forces fb in the momentum equations are determined as

N,m
∑∑ ⎡ fx ⎤
f b(i , j , k) = 𝜂i,j,k
N,m ⎢ f N,m ⎥. (14)
⎢ yN,m ⎥
N m ⎣ fz ⎦

3 RESULTS USING PRESENT FORMULATION OF THE ALM

To demonstrate the inadequacy of using a coarse mesh in the ALM, we here first show some results to verify the method in its present formulation.

3.1 Computation of the NREL Phase VI rotor

As verification case, the blade loading distributions of the NREL Phase VI rotor33 is computed for a steady flow case with a rotor resolution of 10
mesh points along the rotor radius. It should be noted that, as a general guideline, the minimum resolution of the wing should be around 30 to 50
grid points along the rotor radius (see Troldborg20 ). The main geometry and operational conditions of the NREL Phase VI rotor are shown in Table 1.
The size of the employed domain is [Lx , Ly , Lz ] = 180 × 60 × 60 m. The grid resolution is Δ = Δx = Δy = 0.5 m and Δz = 0.25 m, with x and y
being the axes parallel to the ground, and z the vertical axis. With this grid setting, a filter kernel 𝜖 = 3Δ = 1.5 m is used for the actuator line
forces. The resulting normal and tangential loading distributions are shown in Figure 2. To demonstrate the problems of representing correctly
the tip loading when using coarse meshes, the ALM computations are compared with results from a similar blade-element momentum (BEM)
computation27,35 and an ALM computation using the Prandtl tip correction. 36 Like the ALM, the BEM model is based on a blade-element approach
using tabulated airfoil data but with the velocities computed using momentum theory (see Glauert35 and Hansen37 ). Although it normally is
referred to as a ‘‘low-fidelity’’ model, it should be noted that the BEM model generally generates good results for rotors operating close to their
design point.38 The main difference between the loadings computed by the two models is normally found at the outer part of the blade. Here,
the BEM model needs to be modified by a tip correction37 to account for the finiteness of the blades, whereas this in principle is not needed
for the ALM, as this model obviously treats the blades as being finite. The tip correction ensures that the loading smoothly tends to zero at
the tip. However, for various reasons, which will be explored in the following, this is not the case for the ALM unless some corrections are
introduced. In lack of a better solution, many users of the ALM technique have resorted to the Prandtl tip correction as an easy fix to remedy
the tip problem. However, this merely corresponds to a curve fit, as it physically does not represent the aerodynamics of a finite rotor. It should
be recalled that the tip correction originally was introduced by Prandtl36 for correcting the momentum theory to handle the finite number of

TABLE 1 Operation conditions for NREL Phase VI33 and NREL 5-MW34 rotors Phase VI 5 MW
Rotor type 2-bladed 3-bladed
Tip speed ratio 5.39 7.55
Blade aspect ratio 7.5 15.8
Rotor diameter 10.058 m 126 m
Root cut out 0.9 m 5m
Rotational speed 7.50107 rad/s 0.9584 rad/s
Pitch angle 3◦ 0◦
◦
Cone angle 0 0◦
Wind speed 7 m/s 8 m/s
DAĞ AND SØRENSEN 5

FIGURE 2 Comparison of loading distributions

from blade-element momentum (BEM) and
actuator line method (ALM) and ALM with Prandtl
correction for NREL Phase VI rotor operating
under 7-m/s steady inflow

𝜖, m Δ, m Resolution nrGridPoints∕Span TABLE 2 Grid resolution parameters

2.5 1.25 40 × 24 × 32 8
1 0.5 100 × 60 × 80 20
0.5 0.25 200 × 120 × 160 40

blades in the blade-element approach (see, eg, Sørensen27 for a discussion). As seen in the figure, at the inner parts of the blade, the loadings
from BEM and ALM match well. However, at the tip region an overestimation of the ALM loadings is clearly visible. Furthermore, the loading
does not tend to zero at the very tip. Correcting the load distribution by the Prandtl tip correction, however, does not the remedy the situation,
as the loading becomes highly underestimated. As will be shown later, this problem only occurs when using a coarse mesh or a low resolution
of the rotor. However, it is not always possible to carry out computations on sufficiently fine meshes, and there is obviously a need for a
technique that may produce accurate results on coarse meshes that do not depend on inconsistent corrections, like tip corrections meant for BEM
techniques.

3.2 Influence of the kernel size

To further investigate the overshooting issue of the ALM, we here simplify the setup and apply the ALM on a nonrotating rectangular wing.
As reference circulation distribution, the LLT is used for the same wing and results are compared with those obtained from ALM with different
kernel sizes of the filtering function. The reason why the ALM is verified against inviscid LLT solutions is that the inviscid solution represents the
asymptotic correct solution at high Reynolds numbers when the filtering constant goes to zero. Therefore, by decreasing the filtering constant,
we expect the ALM to generate a circulation distribution that tends towards the one computed by the inviscid LLT. As lifting line model, we
employ the classical one originally developed by Prandtl.39 The rectangular wing used in this study has a constant chord length of 1 m with a
span length of 10 m. The wing is positioned at a geometrical angle of attack of 5◦ , and the incoming velocity is fixed at 8 m/s. Inviscid airfoil
data of the NACA 0012 airfoil is used for both ALM and LLT. For the ALM computations, different 𝜖 values are used together with a domain size
of 50 × 30 × 20 m, with a varying grid resolution and a fixed kernel value 𝜖∕Δ = 2. The resolution parameters of each simulation are detailed in
Table 2.
A comparison of the circulation distributions is presented below in Figure 3. Due to symmetry, only the half-span is shown in the plot.
In the LLT case, due to the strong tip vortices, the induction goes to infinity and circulation tends to zero at the very tip of the wing. This is not
the case for the ALM where the circulation for all 𝜖 values is seen to increase near the tip. On the other hand, with decreased 𝜖 values, the ALM
circulation levels converge towards the LLT results yet with a peak at the tip region. This behavior can only be explained by an underestimated
wake induction near the tip, causing an increased angle of attack, which results in a stronger circulation. Due to viscous effects, we would of
course expect minor deviations from the inviscid behavior, as those seen in Figure 3.
In order to clarify the effect of the 𝜖 parameter on the bound vortices, an infinitely long wing with constant circulations is employed as model
example. In this case, the bound vortex generated by the ALM can be studied without being disturbed by the influence of the wake vortices. The
outcome of this is shown in Figure 4, where the computed normal velocity distribution is shown for different 𝜖∕Δ values.
What is seen here is essentially the velocity distribution induced by a two-dimensional vortex, which is being filtered by the Gaussian filter
function. A high 𝜖 value tends to smear out the velocity whereas a small 𝜖 value results in a more distinct velocity distribution, tending towards
the inviscid value of Γ∕2𝜋 r. An important part of the ALM is to determine the local angle of attack as this forms the input to the airfoil data. Since
the angle of attack is determined from the local velocity, as shown in Equations (1) to (3), the induction from a vortex with a high 𝜖 value tends to
overpredict it, with the result that the loadings become overestimated. It should be emphasized that it is not the induction of the bound vortex
per se that defines the angle of attack but the induction from the wake vortices (see, eg, Shen et al40 ). However, since these vortices are born by
the bound vortex, they have the same viscous core size as the bound vortex. A quick remedy to this problem would be to use smaller 𝜖 values.
However, it is known that such an approach decreases the accuracy of the physical location of the vortex origin and causes instabilities in the
CFD results. Instead, based on the knowledge gained from the shown comparative study, we will in the next section develop a methodology for
correcting the induced velocity when determining the angle of attack.
6 DAĞ AND SØRENSEN

FIGURE 3 Comparison of circulation distributions from lifting line

technique (LLT) and actuator line method (ALM) with different 𝜖 values
for AR = 10 wing

FIGURE 4 Induction change observed with different 𝜖 values

4 CORRECTION METHODOLOGY

To account for the missing induction, it is noted that the Lamb-Oseen vortex model41 is equal to the employed filtering function and results in
the exact same induction as the one obtained from an ALM computation with a viscous core size corresponding to the size of the smoothing
kernel 𝜖 . The analytic formulation of a semi-infinite Lamb-Oseen vortex model is given as
[ [ ( )2 ]]
Γ r
wi = 1 − exp − , (15)
4𝜋r rvc

where Γ is the vortex strength and rvc is the viscous core size. From this expression, it is readily seen that the first term corresponds to the inviscid
part of the induction and the second term accounts for the influence of the viscous core. In Figure 5, it is shown that, by replacing rvc with the 𝜖
values used in an ALM computation, the vortex model gives the exactly same distribution of induced velocities.
Hence, by subtracting the viscous part from the induction of the Lamb-Oseen model, one may establish a model for correcting the unwanted
influence of the viscous core on the induced velocity,
[ ( )2 ]
Γ r
wcorr = exp − . (16)
4𝜋r 𝜖

As the bound vortex does not contribute to the induced velocity defining the angle of attack (the angle of attack is essentially determined at
the center of the bound vortex), the correction is applied on the induction from the trailing vortices in the wake, which then corrects the induced
velocity through the Biot-Savart induction law. However, to do this, it is required to know the position and induction of the wake vortices.

4.1 Translating planar wing

To demonstrate how the correction works, we here first apply it on a translating planar wing. The correction is meant to be applied at the angle
of attack calculation stage, where the total additional induction from the wake system, wcorr , is added to the local relative velocity and the angle
DAĞ AND SØRENSEN 7

FIGURE 5 Comparison of Lamb-Oseen vortex model with induction

levels obtained from computational fluid dynamics (CFD) computation
with given 𝜖 parameter

FIGURE 6 Illustration of a vortex system of a translating planar wing

of attack value is determined. To apply the correction, the location of the wake vortices and their strengths have to be defined in advance as
these are required, see Equation (16). For a translating planar wing application, wake vortices can be assumed to form a straight vortex sheet
which is parallel to the flow and which trail from the quarter-chord line of the wing (see Figure 6).
Then, for each actuator line point, the total correction from the viscous cores of the wake vortices can be calculated as

[ ( ) ]
∑ Γw( j)
N+1
d(i,j) 2
wcorr(i) = exp − . (17)
j=1
4𝜋d(i,j) 𝜖

Here, N denotes the total number of actuator line points on and Γw represents the circulation strength of the wake vortices, with magnitudes
equal to the difference between two neighboring bound circulations (see Figure 6),

Γw( j) = Γ( j) − Γ( j−1) . (18)

For the tip vortices, the magnitudes are equal to the bound circulations which are closest to the tips. In this system, the strength of a bound
circulation is a function of the corresponding chord, c, relative velocity, urel , and the lift coefficient, CL . Employing the Kutta-Joukowsky theorem,
we get

1
Γ(i) = c(i) CL(i) urel(i) . (19)
2
8 DAĞ AND SØRENSEN

FIGURE 7 Illustration of the induction correction for the tip vortex

In Equation (17), the variable d defines the distance between the control points and the points from which the trailing vortices are shed into
the wake. The definition in the given coordinate system* (see Figure 6) can be written as

d(i,j) = y(i) − yw( j) . (20)

Knowing the total correction for the induced velocity, the resulting corrected angle of attack is determined as
( )
−uz(i) + wcorr(i)
𝛼(i) = 𝛼g − sin−1 . (21)
urel(i)

In Equation (21), 𝛼 g represents the geometrical angle of attack of the wing, and urel and uz are the relative and the vertical velocity components,
respectively. An illustration of the induction correction for a single vortex line is shown in Figure 7. In the figure, the dashed blue line represents
the underestimated induction from the tip vortex using AL/CFD, which causes too large tip loadings. The thick blue line is the induction that
would occur in a LLT case. Hence, the red swept area is the induction correction, which corrects the angle of attack values at the region. The
consequence of modifying the values of angle of attack, bound vorticity, and hence, the strengths of the wake vortices change as a function of
time. Therefore, the correction requires an iterative procedure in each time step of a CFD computation. We here restrict the modeling of the
wake vortices to straight lines in order to simplify the bookkeeping. The procedure can in principle be extended to a free wake analysis. However,
as will be shown later, even a simple prescribed wake greatly improves the resulting circulation distribution.

4.2 Wind turbine rotor

The correction procedure for a turbine rotor proceeds in principle as the one for a translating planar wing. The only difference is due to the
wake structure, which adds a level of complexity to the calculation. Due to the three-dimensional formation of the wake (see Figure 8 [left]), it is
needed to define discrete vortex lines and calculate the induction correction in all three space directions. By using vector notation, the induction
correction from a vortex line is written as
( )2
⎡ wxcorr ⎤ ⃗ ⃗ ⎡ ⎡ ⃗ ⎤⎤
⎢ wycorr ⎥ = Γ dl × d ⎢exp ⎢− |d| ⎥⎥ , (22)
⎢ ⎥ 4𝜋 |d|⃗3 ⎢ ⎢ 𝜖 ⎥⎥
⎣ wzcorr ⎦ ⎣ ⎣ ⎦⎦

⃗ is the vector representing the direction and the length of the vortex line and d⃗ is the vector from the center of the vortex line to the
where dl
induction calculation point (see Figure 8 [right]). Then for each actuator line point, the total induction correction can be calculated by summing
up all the corrections from the wake system as

⎡ wxcorr ⎤
N,m
⃗ h,p,q × d⃗ N,m ⎡ ⎡ ⎛ ⃗ N,m ⎞2 ⎤⎤
⎢ wycorr ⎥
∑ ∑ ∑ Γw(h,p,q) dl ⎢ ⎢ |dh,p,q | ⎥⎥
⎢exp ⎢−⎜⎜ 𝜖 ⎟⎟ ⎥⎥ .
h,p,q
= (23)
⎢ ⎥ 4𝜋 |d⃗ N,m | ⎢ ⎢ ⎝ ⎠ ⎥⎦⎥⎦
3
⎣ wzcorr ⎦ h p q
h,p,q
⎣ ⎣

Here, the indices h, p, and q represent the blade number, span position, and the azimuthal position, respectively, for the wake vortices and N
and m represent the blade number and span position for actuator line points, respectively (see Figure 8 [right]). The circulation strengths are
calculated similar to the planar case as
1
ΓN,m = cN,m CLN,m uN,m
rel
, (24)
2

* Note that the span of the wing is aligned with the y-axis for simplicity.
DAĞ AND SØRENSEN 9

FIGURE 8 Illustration of the helical wake released from a wind turbine rotor (left) and a close-up illustration of the wake formation of the single
blade (right)

and

Γw(h,p,q) = Γ(p−1,q) − Γ(p,q) . (25)

With the calculated total induction correction for each actuator line point, the flow angle can be corrected as
( )
−1 uN,m
x
N,m
+ wxcorr
𝜙 N,m
= tan , (26)
uN,m
𝜃
N,m
+ w𝜃corr

where
N,m
w𝜃corr = wycorr
N,m
cos(𝜃 N ) + wzcorr
N,m
sin(𝜃 N ). (27)

Next, the new angle of attack values can be computed locally by Equation (4).
To generate the discrete wake, the helical pitch of the each individual trailing vortex is assumed to be constant and equal to the relative
flow angle at the position where the trailing vortex is released. Since the flow angle is updated in every time step, depending on the local flow
conditions, this gives a quasi-steady behavior for the correction. In this work, the length of the helical vortex sheet is fixed at two full revolutions,
and the azimuthal discretization of the wake is made with 2◦ intervals. The added computational expenses of including the correction are
insignificant as compared with the time-marching procedure of LES computations. Furthermore, there are various ways of making the correction
more efficient and hence reduce the added computational costs. This will be the scope of future investigations.

5 VALIDATION OF THE CORRECTION TECHNIQUE

In the following, the new correction will be validated against computations using lifting line and BEM theory. Like the BEM method, ALM is
depending on airfoil data. Therefore, any validation of the method against experimental data demands access to airfoil data corrected for 3D
effects. This makes experimental validation complicated, since this kind of data is rarely known. This is a general problem for both BEM and
ALM. However, numerous studies have shown that if adequate airfoil data can be provided, then, under usual design conditions, BEM performs
just as well as full blown boundary resolved CFD. Validation of the BEM codes are therefore typically performed by using airfoil data extracted
directly from the CFD codes or by modifying 2D airfoil data from CFD results, and then comparing BEM results with CFD. Recent examples of
this can be found in, eg, the report from the AVATAR project,38 where a comprehensive comparison between various CFD results and BEM
showed that BEM results lie within the dispersion range of different CFD methods. Another recent study by Bangga42 showed that BEM and
CFD results match extremely well, provided that correct airfoil data is being used. Therefore, it is sufficient to compare the results of the ALM
and the improvements of the new tip correction with BEM results. The same goes for the lifting line technique, which can be considered as an
inviscid version of a Navier-Stokes code.

5.1 Translating planar wing

To validate the new correction, we first compute the flow past a translating rectangular wing with a constant chord c = 1 m and aspect ratios 6
and 10. For the ALM computations, a CFD grid resolution of 100 × 60 × 80 is used. This resolution corresponds to Δ = 0.5 m in the spanwise
10 DAĞ AND SØRENSEN

FIGURE 9 Comparison of circulation distribution obtained from

actuator line method (ALM), corrected ALM (ALM*), and lifting line
technique (LLT) computations for translating planar wings with
AR = 6, 10

TABLE 3 Details of the computational fluid dynamics (CFD) cases for rotor simulations
simNr. Rotor Resolution Δ, m r∕Δ 𝜖, m 𝜖 * (𝜖∕Δ) nrAero tsMax Δt, s tsAvg Correction dataFig
1 P6 180 × 60 × 120 1 5 6 3 11 3E3 4.3E-2 250 ✓ Figure 10-upper
2 P6 180 × 60 × 120 1 5 6 3 11 3E3 4.3E-2 250 - Figure 10-upper
3 P6 180 × 60 × 120 1 5 10 5 11 3E3 4.3E-2 250 ✓ Figure 10-upper
4 P6 180 × 60 × 120 1 5 10 5 11 3E3 4.3E-2 250 - Figure 10-upper
5 P6 720 × 240 × 480 0.25 20 1.5 3 21 20E3 2E-3 1E3 ✓ Figure 10-lower
6 P6 720 × 240 × 480 0.25 20 1.5 3 21 20E3 2E-3 1E3 - Figure 10-lower
7 P6 720 × 240 × 480 0.25 20 2.5 5 21 20E3 2E-3 1E3 ✓ Figure 10-lower
8 P6 720 × 240 × 480 0.25 20 2.5 5 21 20E3 2E-3 1E3 - Figure 10-lower
9 5 MW 120 × 40 × 80 12.6 5 37.8 3 10 8E3 3.9E-2 250 ✓ Figure 11-upper
10 5 MW 120 × 40 × 80 12.6 5 37.8 3 10 8E3 3.9E-2 250 - Figure 11-upper
11 5 MW 120 × 40 × 80 12.6 5 63 5 10 8E3 3.9E-2 250 ✓ Figure 11-upper
12 5 MW 120 × 40 × 80 12.6 5 63 5 10 8E3 3.9E-2 250 - Figure 11-upper
13 5 MW 480 × 160 × 320 3.15 20 9.45 3 20 24E3 1.3E-2 1E3 ✓ Figure 11-lower
14 5 MW 480 × 160 × 320 3.15 20 9.45 3 20 24E3 1.3E-2 1E3 - Figure 11-lower
15 5 MW 480 × 160 × 320 3.15 20 15.75 5 20 24E3 1.3E-2 1E3 ✓ Figure 11-lower
16 5 MW 480 × 160 × 320 3.15 20 15.75 5 20 24E3 1.3E-2 1E3 - Figure 11-lower

and streamwise directions, and 0.25 m in the vertical direction. The filter parameter, 𝜖 , is taken as a chord length, which corresponds to 2Δ for
any given grid resolution. The inflow velocity is fixed at 8 m/s, and the geometrical angle of attack for both wings is 5◦ . In Figure 9, circulation
distributions are plotted for both wings and compared against LLT computations. In the figure, the corrected ALM computations are shown as
ALM*, and, due to symmetry, only half-span distributions are shown.
As seen from the figure, the overestimation of the tip circulations is visible in the ALM simulations, with a peak value that is higher than the
maximum circulation occurring at the middle section of the wing. Applying the new correction fixes the issue and provides comparable results
with LLT. Additionally, it is important to note that it is not only at the tip region but also at the midregions that the correction provides circulation
levels similar to LLT computations.

5.2 Wind turbine rotor

For rotor applications, in order to cover a wider range of possibilities with a minimum of computations, two different rotors are employed, namely,
the NREL Phase VI and the NREL 5-MW rotor. The two-bladed Phase VI rotor runs at a tip speed ratio of 5.39 and has an averaged aspect ratio
of 7.5, whereas the 5-MW rotor is a three-bladed type operating at a higher tip speed ratio of 7.55, with slender blades of an average aspect
ratio of 15.8. The operating conditions of the rotors are detailed in Table 1.
For this study, we have conducted eight simulation cases for each rotor. For the Phase VI rotor, a domain size of 180 × 60 × 60 m is used and
the rotor center is positioned at 20 × 30 × 30 m. For the 5-MW rotor, a domain size of 1512 × 504 × 504 m is employed and the rotor center
is positioned at 252 × 252 × 252 m. For both domains, the blockage effects are checked, and it is concluded that the current domain sizes are
sufficient. Further details of the CFD cases are provided in Table 3. In the table, Δt represents the time step size in the CFD computations, which
is restricted to be smaller than the time it takes for a blade tip to travel one grid spacing. The total number of time steps for each simulation is
shown in the table and is denoted tsMax. Before fixing the blade loadings, simulations were run long enough for the wake to develop. The number
DAĞ AND SØRENSEN 11

FIGURE 10 Comparison of normal and tangential

loadings from actuator line method (ALM) with
and without the correction and blade-element
momentum (BEM) computations for NREL Phase
VI turbine

FIGURE 11 Comparison of normal and tangential

loadings from actuator line method (ALM) with
and without the correction and blade-element
momentum (BEM) computations for NREL 5-MW
turbine

of time steps where the loading is fixed and averaged over is shown under tsAvg. The r∕Δ and nrAero columns show the number of CFD grid
points per radius and number of aerodynamic control points used in the blades, respectively. Additionally, the figure in which the corresponding
data set is presented is addressed in the dataFig column.
The resulting normal and tangential loading distributions are in Figures 10 and 11 plotted against BEM computations for the Phase VI and
the 5-MW rotors, respectively. Although experimental data are available for the Phase VI rotor, it was decided not to use them in the present
investigation. The reason is that the focus of the present work is on the correction technique and the impact it has on the loadings near the tip.
Hence, it is sufficient at this stage to validate it against the simple BEM method.
As seen from the figures, the correction greatly improves the loadings in all the tested cases. For the Phase VI rotor, the corrected results
are in good agreement with BEM computations for both coarse and fine resolutions and for two different projection widths. For the 5-MW
rotor, in coarse resolution cases, the correction lowers the peak error from 23% to 5% and from 77% to 17% for the normal and tangential loads,
respectively. In the finer grid case, the results are again in good agreement with BEM computations for both 𝜖 values.
12 DAĞ AND SØRENSEN

6 DISCUSSION AND CONCLUSIONS

The projection function that was introduced to smear out the body forces in the ALM alters the induced velocities, which in most cases results in
too high blade loadings. This is in particular pronounced near the tip where strong vortices are trailed from the blade. To overcome this problem,
a correction model has been introduced and tested. The idea behind the model is to correct the underestimated wake induction on the control
points of the actuator line. By using the Biot-Savart induction law together with the prescribed position of the wake vortices, the influence of the
projection function on the resulting induced velocity is removed. This approach results in a correction technique to correct the angle of attack,
which subsequently reduces the blade loadings. The new correction is first applied on a translating planar wing, showing that it works excellent
for this simple flow case. Next, the NREL Phase VI and 5-MW rotors were employed to test the correction. For both rotors, the correction
showed great improvement in blade loadings even for CFD grid resolutions as low as 5 grid points per rotor radius. Applying the correction on
ALM computations of the Phase VI rotor resulted in load levels very close to the results obtained from BEM computations, although a too crude
resolution may result in minor deviations of the tangential loads. For both turbines, it is seen that a CFD resolution corresponding to 20 grid
points per rotor radius is sufficient to obtain results which are in excellent agreement with BEM computations for both normal and tangential
loadings. It can be concluded that the application of this new correction results in very convincing results that indeed reduces the loadings to the
anticipated level for both translating planar wings and for wind turbine rotor configurations even with low CFD grid resolutions.

ORCID

Kaya Onur Dağ https://orcid.org/0000-0002-3909-1566

Jens Nørkær Sørensen https://orcid.org/0000-0002-1974-2675

REFERENCES
1. Sørensen JN, Shen W. Numerical modeling of wind turbine wakes. J. Fluids Eng; 2002.
2. Ivanell S, Sørensen JN, Mikkelsen R, Henningson D. Analysis of numerically generated wake structures. Wind Energy. 2009;12(1):63-80.
3. Troldborg N, Sørensen JN, Mikkelsen R. Numerical simulations of wake characteristics of a wind turbine in uniform inflow. Wind Energy.
2010;13(1):86-99.
4. Kang S, Yang X, Sotiropoulos F. On the onset of wake meandering for an axial flow turbine in a turbulent open channel flow. J Fluid Mech.
2014;744:376-403.
5. Porté-Agel F, Wu Y-T, Lu H, Conzemius R. Large-eddy simulation of atmospheric boundary layer flow through wind turbines and wind farms. J Wind
Eng Ind Aerod. 2011;99:154-168.
6. Churchfield MJ, Lee S, Michalakes J, Moriarty PJ. A numerical study of the effects of atmospheric and wake turbulence on wind turbine dynamics. J
Turbul; 2012.
7. Nilsson K, Ivanell S, Hansen KS, et al. Large-eddy simulation of the lillgrund windfarm, Wind Energy, 2014. Published online
8. Martínez-Tossas LA, Churchfield MJ, Leonardi S. Large eddy simulations of the flow past wind tturbine: actuator line and disk modeling. Wind Energy;
2015.
9. Xie S, Archer C. Self-similarity and turbulence characteristics of wind turbine wakes via large-eddy simulation. Wind Energy. 2015;18:1815-1838.
10. Marjanovic N, Mirocha J, Kosović B, Lundquist J, Chow F. Implementation of a generalized actuator line model for wind turbine parameterization in
the weather research and forecasting model. J Renew Sustain Ener. 2017;9:063308.
11. Stevens R, Martínez-Tossas LA, Meneveau C. Comparison of wind farm large eddy simulations using actuator disk and actuator line models with wind
tunnel experiments. Renew Energ. 2018;116:470-478.
12. Yu Z, Zheng X, Ma Q. Study on actuator line modeling of two nrel-5mw wind turbine wakes. J. Appl. Sci. 2018;8(434).
13. Shamsoddin S, Porté-Agel F. A large-eddy simulation study of vertical axis wind turbine wakes in the atmospheric boundary layer. Energies. 2018;9(366).
14. Akbar M, Dabiri J. Self-similarity and flow characteristics of vertical-axis wind turbine wakes: an les study. journal of turbulence. J Turbul.
2017;18(4):373-389.
15. Baratchi F, Jeans T, Gerber A. Actuator line simulation of a tidal turbine in straight and yawed flows. Int J Mar Energy. 2017;19:235-255.
16. Creech A, Borthwick A, Ingram D. Effects of support structures in an les actuator line model of a tidal turbine with contra-rotating rotors. Energies.
2017;10(726).
17. Baba-Ahmadi M, Dong P. Numerical simulations of wake characteristics of a horizontal axis tidal stream turbine using actuator line model. Renew
Energ. 2017;113:669-678.
18. Bühler M, Weihing P, Klein L, Lutz T, Krämer E. Actuator line method simulations for the analysis of wind turbine wakes acting on helicopters, In: IOP
Conf. Series: Journal of Physics: Conf. Series, vol. 1037, 2018.
19. Breton S-P, Sumner J, Sørensen JN, Hansen K, Sarmast S, Ivanell S. A survey of modelling methods for high-fidelity wind farm simulations using les.
Philosophical Transactions of the Royal Society. 2016;A 375(20160097).
20. Troldborg N. Actuator Line Modeling of Wind Turbine Wakes. PhD thesis, DTU, 2008.
21. Jha PK, Churchfield MJ, Moriarty PJ, Schmitz S. Guidelines for volume force distrubutions within actuator line modeling of wind tturbine on large-eddy
simulation-type grids. J Sol Energy Eng. 2014.
22. Shives M, Crawford C. Mesh and load distribution requirements for actuator line cfd simulations. Wind Energy. 2013.
23. Martínez-Tossas LA, Churchfield MJ, Meneveau C. Optimal smoothing length scale for actuator line models of wind turbine blades. Wind Energy; 2017.
24. Jha P, Schmitz S. Actuator curve embedding - an advanced actuator line model. J Fluid Mech. 2018;834(R2).
DAĞ AND SØRENSEN 13

25. Caprace D-G, Chatelain P, Winckelmans G. Lifting line with various mollifications: Theory and application to an elliptical wing. AIAA JOURNAL. Jan.
2019;57.
26. Martínez-Tossas LA, Churchfield MJ. Filtered lifting line theory and application to the actuator line model. J Fluid Mech. Mar. 2019;863:269-292.
27. Sørensen JN. General Momentum Theory for Horizontal Axis Wind Turbines. Springer; 2016.
28. Shen WZ, Sørensen JN, Mikkelsen R. Tip loss correction for actuator/navier-stokes computations. Journal of Solar Energy Engineering; 2005.
29. Sørensen JN, Dag KO, García NR. A refined tip correction based on decambering. Wind Energy; 2016.
30. Dag KO, Combined pseudo-spectral/actuator line model for wind turbine applications. PhD thesis, Technical University of Denmark Department of Wind
Energy, May 2017. http://orbit.dtu.dk/files/146182282/Kaya_Onur_Dag_hele_afhandlingen.pdf.
31. Canuto C, Hussaini MY, Quarteroni A, Zang TA. Spectral Methods Evolution to Complex Geometries and Applications to Fluid Dynamics. Springer; 2007.
32. Pope SB, Turbulent Flows. 2000.
33. Hand MM, Simms DA, Fingersh LJ, et al. Unsteady aerodynamics experiment phase vi: Wind tunnel test confiugrations and avaliable data campaigns,
tech. rep., NREL, 2001.
34. Jonkman J, Butterfield S, Musial W, & Scott G., Definition of a 5-mw reference wind turbine for offshore system development, techreport, National
Renewable Energy Laboratory, Golden, CO, USA, 2009. NREL/TP-500-38060.
35. Glauert H. The General Momentum Theory. Springer; 1935 Aerodynamic Theory’, Volume IV division L-III, Reprinted in 1963.
36. Betz A. Schraubenpropeller mit geringstem Energieverlust. Mit einem Zusatz von l. Prandtl. 191–217, Mathematisch-Physikalische Klasse, 1919. Nachrichten
von der Gesellschaft der Wissenschaften zu Göttingen.
37. Hansen MOL. Aerodynamics of Wind Turbines. 2nd ed. Earthscan; 2008.
38. Sørensen NN, Hansen MO, Ramos-García N, Florentie L, and Boorsma K. Power curve predictions avatar proj, tech. rep., AVATAR D2, 2018.
39. Prandtl L , Applications of modern hydrodynamics to aeronautics. (translated from ludwig prandtl (1918) tragflügeltheorie. königliche gesellschaft der
wissenschaften zu göttingen). NACA TR-116.
40. Shen WZ, Hansen MOL, Sørensen JN. Determination of the angle of attack (aoa) on rotor blades. Wind Energy. 2009;12(1):91-98.
41. Lamb H. Hydrodynamics, 6th ed. Cambridge University Press; 1932.
42. Bangga G. Comparison of blade element method and cfd simulations of a 10 mw wind turbine. FLUIDS. Dec. 2018;3.

How to cite this article: Dağ KO, Sørensen JN. A new tip correction for actuator line computations. Wind Energy. 2019;1-13.
https://doi.org/10.1002/we.2419