Anda di halaman 1dari 10

Validation of an Actuator Disc Model

P.-E. Réthoré N. N. Sørensen F. Zahle


pire@risoe.dtu.dk nsqr@risoe.dtu.dk frza@risoe.dtu.dk
Wind Energy Division · Risø DTU · Denmark

Abstract accuracy of the solution can be deeply affected


by the approach used to redistribute the forces.
Wind turbine wake can be studied in CFD with the A poor approach can slow down significantly
use of permeable body forces (e.g. actuator disc, the code at each iteration, and can require a
line, surface). This paper presents a general flexi- finer grid to obtain grid independence.
ble method to redistribute wind turbine blade forces
as permeable body forces in a computational domain.
The method can take any kind of shape discretization, Moreover, because of the general lack of
determine the intersectional elements with the com- literature on the subject, there is also a need for
putational grid and use the size of these elements to validation methods for this step. Some papers
redistribute proportionally the forces. validate this step indirectly by comparing the
The special case of the actuator disc is successfully wind turbine wake with measurements (e.g.
validated with an analytical solution for heavily loaded
[1]). By doing so, both steps (force estimation,
turbines and with a full rotor computation in CFD.
and force redistribution) are used together with
Keywords: Conway’s Actuator Disc, Bessel Laplace
a turbulence model to generate the flow field.
Integrals, CFD.
When combining several complex methods
together there is always the risk that multiple
errors compensate each other. It is therefore
1 Introduction
considered safer to validate each of the models
independently.
As computational power becomes more af-
One way of doing this is to compare with other
fordable, Computational Fluid Dynamics (CFD)
types of wind turbine wake models that can
methods become attractive solutions for mod-
provide the blade loadings, such as analytical
elling wind turbine and wind farm wakes. The
solutions or full rotor computations. Such
most cost effective CFD wind turbine wake
comparison of actuator methods with analytical
models are based on the concept of applying
solution can be found in the literature (e.g.
the wind turbine blades forces in the computa-
comparison with the heavily loaded actuator
tional domain through permeable body forces.
disc mode of Wu [13] and Conway [19]). It
There are three main categories based on this
seems, however, that there is not any paper
concept, the actuator disc, line and surface (for
that discusses the comparison of full rotor
a literature review, see [8, 15, 18, 22, 23]). All
computation with actuator methods.
these concepts follow the same two-step ap-
proach. Firstly, the blade forces are estimated
from the local flow information using different This paper describes a general method
types of methods. Secondly, they are then (Sec.2) to redistribute body forces in a com-
redistributed in the computational domain. This putational domain, which is implemented in
article focusses on the second step. EllipSys [12, 20]. The method has a low
initialization time and need a relatively small
As there is practically no physics involved in amount of domain cells to redistribute correctly
the second step, the vast majority of the papers the forces. The model is based on three
describing the different actuator methods does levels of discretization. The actual shape,
not mention this issue. However, the speed and defined by a grid, the computational domain

1
grid, and the intersectional grid between the
shape and the computational domain grids.
This independent description of the shape and
the domain gives a wide flexibility of possible
shapes and motions (e.g. disc, rotating lines,
moving kites, trees). An indirect application
of this method is to define immerse boundary
conditions such a turbulence generator grids [9].

The special case of the actuator disc model


is validated in Sec.3 with respect to the analyti-
cal solution for heavily loaded actuator discs of
Conway [6], and a full rotor computation. Both Figure 1: Illustration of the different elements con-
comparison show excellent agreement between sidered in the intersectional polygon search algorithm
the different models presented in Sec.2. This (with 2D shape cells).
shows that using actuator discs, with the proper
loading, gives an accurate description of the
wind turbine wake. Moreover, it is found that a (quadrilaterals), 3D cells (hexahedrons). Each
coarse discretization (as low as 10 cells per ro- of the points are defined in 3D. The relative po-
tor diameter) gives a relatively acceptable wake sition vector mentioned previously is relating the
development in comparison with finer meshed domain grid origin to the shape grid origin.
full rotor computation.

2.2 Initialisation Phase


2 Actuator Shape Model The initialisation phase, during which the inter-
sectional grid is redefined, is carried out each
2.1 General Idea time that the shape definition is changed. The
The actuator shape model is based on three dif- intersectional grid is built on a two-step process.
ferent types of discretization. Firstly, each of the shape grid points are
associated to the domain cell they are lo-
1. The shape is defined by a grid, where each cated into. The search routine is based on a
cells is somehow associated to a force vec- spreading recursive dichotomy methodology. A
tor. recursive dichotomy is first used to dive into the
domain grid. When one shape point is found
2. The computational grid is where the shape
in a domain cell, the neighboring shape points
forces are redistributed and solved with the
are first tested in the same domain cell, then on
Navier-Stokes equations.
the neighboring domain cells until all the points
3. The intersectional grid between the shape are found. This approach reduces dramatically
grid and the domain grid is used as a the number of iteration needed to associate all
weight to redistribute the shape forces into the shape points with domain cells, and has
the domain grid. the advantage to be reusable. When the shape
definition has changed, the previous shape
The independent shape grid and the domain point - domain cell association is used as a first
grid gives a flexibility of the shape motion in the guess to the update.
domain and a simple domain grid definition.
Secondly, based on the “shape point - domain
The shape position is defined by three vectors cell” association, the algorithm is looking for all
(one for the relative position with the domain, the “shape cell segments - domain cell faces“
and two for the orientation of the shape). These and “shape cell faces - domain cell segments“
vectors can be changed dynamically during the intersectional points (see Fig.1). These points
simulation. are defining the intersectional grid. The inter-
The discretization of the shape can be done in sectional points are then rearranged into inter-
3 different ways: 1D cells (segments), 2D cells sectional cells.

2
ghost cells between each processor is done
33 Solving through the MPI libraries. The turbulence model
Navier-Stokes
 used in this paper is Menter’s k -ω SST [11].
Redistributing Extracting
shape forces shape velocities
]]
Actuator Disc Model of EllipSys
Estimating rr
shape forces
The actuator disc model is based on the actua-
tor shape model described in the previous sec-
Figure 2: Active phase scheme. tion. The shape grid is a coplanar polar disc of
11 radial and 62 angular elements. The force
vector associated to each of the disc cells are
2.3 Active Phase found by the two other models describe in the
following sections respectively.
The active phase is executed at every solv-
ing iteration. The iterative scheme, illustrated
in Fig.2, is basically solving the Navier-Stokes Conway’s Actuator Disc
equation, extracting the shape velocities, from
which the shape forces are estimated and finally In a series of articles [4–7, 19], Conway has de-
redistributed in the computational grid. The step scribed an exact actuator disc model for a heav-
of interest, in this paper, is the redistribution of ily loaded propeller or a wind turbine. The model
the shape forces. is based on the idea of axially discretizing the
The redistribution of the forces is done by using slipstream azimuthal vorticity of a wind turbine
the intersectional grid. According to the inter- into vortex disks. As there are ways to express
sectional cell type, its relative size (i.e. length for the flow behaviour induced by vortex disks, if the
1D segments, area for 2D polygons, or volume vortex distribution of the slipstream is known, it
for 3D polyhedrons) compared to the associated is possible to derive an exact formulation of the
shape cell is used as a proportional weight to induced flow features.
scale the shape cell force vector. All the inter- This method is relatively complex to implement
sectional cell force vectors associated to a do- as it requires solving Bessel-Laplace integrals
main cell are then added together to obtain the using recursive rules in order to get an ex-
equivalent domain cell force vector (see Fig.3) . pression in terms of complete elliptic integrals
and other associated functions. This section
presents a concise explanation on how to setup
3 Validation this solution and gives some analytical expres-
sions needed to solve the special case of a
parabolic wake profile.
3.1 Methods
Conway has derived in [6] a solution for the
EllipSys Flow Solver special case of a parabolic wake profile includ-
ing the slipstream expansion, where the vorticity
EllipSys is an in-house incompressible finite distribution is taken as
volume Reynolds averaged Navier-Stokes flow
solver developed at Risø-DTU [20] and DTU- ωφ = ar, (1)
MEK [12]. The flow variables are collocated in
the mesh to facilitate complex mesh geometries. where ω is the vorticity, φ is the tangential di-
The SIMPLE algorithm [14] is used to solve the rection, a is a free parameter and r is the radial
Navier-Stokes equations. The convective terms coordinate. In this special case, the velocity pro-
are discretized using the QUICK scheme [10]. file in the ultimate wake is known to be
The pressure equation is solved using a mod-
2
ified Rhie-Chow algorithm to avoid odd/even Uz (r, ∞) = U∞ + a(R∞ − r2 )/2, (2)
pressure decoupling with body forces [15–17],
and is accelerated by a multigrid technique [20]. where R∞ is the ultimate width of the slip-
The system is parallelized in a multiblock struc- stream. In this case, the stream function Ψ and
ture, where the blocks can be solved on a differ- the axial and radial velocities Vz and Vr can be
ent processor. The communication of the block found by solving

3
(a) Shape cells. The colour indicates the force (b) Intersectional polygons between the shape
by area of each shape cell. cells and the domain domain cells. The colour
indicates the magnitude of the polygons area

Figure 3: Force discretization algorithm.

where L is the axial loading, a is the free


U∞ r 2 ar ∞
Z
Ψ(r, z) = + ζ(−1,2,1) dz ′ , (3) parameter and ρ is the density.
2 2 0
a ∞
Z
To implement them into EllipSys, they are dis-
Uz (r, z) = U∞ + ζ(0,2,0) dz ′ , (4)
2 0 tributed over the disc grid and then inserted in-
a ∞ side the computational mesh as described in
Z
Ur (r, z) = ±ζ(0,2,1) dz ′ , (5)
2 0 Sec.2.
2
ζ(λ,µ,ν) = Rw (z ′ )I(λ,µ,ν) (Rw (z ′ ), r, z − z ′ ), (6)
Full Rotor Computation
where Rw (z) is the slipstream boundary radial
position at an axial distance z from the disc The complete geometry of the rotor and nacelle
(also defined as the wake width function), U∞ is of a Nordtank 500 kW wind turbine is simulated
the freestream inflow velocity and I(λ,µ,ν) is the up to a steady state solution [25] using EllipSys
Bessel-Laplace integrals (BLI) introduced by and the k -ω SST turbulence model [11]. The
Conway [5] and defined as Eq.7. The analytical wind turbine is equipped with three LM 19.1 m
expression of I(−1,2,1) , I(0,2,0) and I(0,2,1) , as blades and has a cylindrical nacelle (2 m
well as the method used to derive them are diameter, 8.9 m length). In order to simplify the
described in the Appendix A. simulation, the geometry is lightly smoothed,
the tower is not considered and the spinner
Z ∞
I(λ,µ,ν) = e−s|z| sλ Jµ (sR)Jν (sr) ds, (7) and nacelle where rotated along with the rotor.
0 Moreover, the simulation is carried out with a
where λ, µ and ν are integers and Jα are Bessel uniform inflow without the ground boundary.
functions of the first kind.
In order to find the wake width function Rw (z),
Conway proposes in the appendix of [6] a recur- The computational mesh, illustrated in Fig.4,
sive method based on the idea that the stream is generated using Gridgen and HypGrid [21].
function Eq.3, is constant along the slipstream It is composed of 108 blocks of 323 cells (14.2
boundary, so that million cells). The y + at the solid walls is
 kept under 2. A no-slip boundary condition is
Ψ Rw (z), z = Ψ(D/2, 0). (8)
imposed at the wall of the structure.
The loads on the disc are estimated by Con-
way’s model from the stream function Eq.3, us- The Nordtank turbine rotates at 27.1 RPM.
ing Eq.9 [6]. The inflow wind speed is set to 10 m/s. The
h i whole steady-state simulation takes approxi-
L(r) = aρ Ψ(RT , 0) − Ψ(r, 0) , (9) mately 4 hours on 56 2.2Ghz nodes.

4
Figure 4: Illustration of the mesh surrounding the full-rotor computation. The colours illustrate the axial velocity
magnitude.

be x ∈ [−3D, −1D] and x ∈ [1D,3D]. The same


The blades skin friction is integrated at 11 dif- inflow parameters for both the mean velocity
ferent positions and recombined into force vec- and the turbulence are used in the actuator disc
tors. These force vectors are then smeared over model and the full rotor computation.
the corresponding annular section of the actu-
ator disc grid. Finally, they are redistributed On Fig.6, the axial Ux and tangential Uθ
in the computational domain using the method velocity components as well as the pressure
presented in Sec.2 and in Fig.3. are compared along the radial direction r, at
different positions upstream and downstream of
the rotor. The results are satisfactory in terms
3.2 Results of mean velocity and pressure distribution.
EllipSys vs Conway
However, the comparison in terms of tur-
Fig.5 presents the comparison of the CFD ac- bulence parameters shows a large difference
tuator disc model to Conway’s heavily loaded between the two models (not illustrated here).
actuator disc model. Conway’s model uses a Indeed, the turbulence generated by the actua-
vorticity factor a = −4U∞ /D2 , corresponding tor disc lacks the detailed structures generated
to CT = −0.4484 applied in the CFD-AD. The by the blade and nacelle geometry that still
model in EllipSys is used without any turbulence dominate in this region. The only turbulence
model in order to obtain a result close to the in- generated by the actuator disc is produced
viscid formulation of Conway. through the mean velocity shear at the bound-
The results show that the two models are in very ary of the wake. As the inflow turbulence is
close agreement, except for the radial velocity low, the production of turbulence is several
directly at the disc. orders of magnitude smaller than the turbulence
generated by the blades and nacelle in the full
rotor computation. Compared to atmospheric
Actuator Disc vs Full Rotor Computation
turbulence, the turbulence generated by the
The actuator disc model does not give a de- rotor is nonetheless an order of magnitude
tailed description of the flow in the direct vicinity smaller and should not play a significant role in
of the rotor blades. Instead, its application is the far wake.
to be used to model the far wake of a wind
turbine. The region of comparison between the The difference between the two actuator disc
two kind of wind turbine flow model is chosen to computations (10 cells per rotor diameter (c/D)

5
Axial velocity at various axial coordinate z (C =−0.4484) Radial velocity at various axial coordinate z (C =−0.4484)
T T
1.1 0.16
Conway: z=−0.5D
Normalized axial velocity V /U [−]

1 0.14 EllipSys: z=−0.5D

Normalized axial radial V /U [−]


Conway: z=0D


0.12
z

0.9 EllipSys: z=0D

r
0.1 Conway: z=0.5D
0.8 EllipSys: z=0.5D
0.08 Conway: z=2.5D
0.7 EllipSys: z=2.5D
0.06
0.6
0.04

0.5 0.02

0.4 0
0 0.25 0.5 0.75 1 0 0.25 0.5 0.75 1
Normalized distance from center r/D [−] Normalized distance from center r/D [−]

(a) Axial velocity (b) Radial velocity

Figure 5: Actuator disc (parabolic wake case) compared to the analytical axisymmetric solution of Conway.

and 20 c/D) is visible, yet remains relatively blade forces, the uncertainty associated with
small. estimating those velocities is going to be prop-
agated, through the forces estimation, in the
wake of the wind turbine. When considering
multiple wind turbines in cluster, the combined
4 Discussion uncertainty may then become significant.

The overall good agreements obtained in the re-


sults section demonstrate that actuator disc can The comparison between the two actuator
be a cost effective way to model wind turbine discs with 10 and 20 cells per rotor diameter in-
wake. While the vortex structures present in the dicates that using only 10 c/D is enough for ob-
close wake region are not modelled correctly, taining a good resolution of the close wake flow
the far wake region (>3D) seem to be a close features. In the far wake region, the gradients
match to a far more complex and expensive full become smaller and the cell size becomes less
rotor computation. critical. Using only 10 c/D open the possibility
to numerically carry out large wind farm compu-
The small discrepancy in the radial direction tations. For example, an hypothetical cluster of
at the rotor location probably comes from the 10 × 10 wind turbines (WT), with a wind turbine
fact that the CFD actuator disc model has spacing of 8D in each direction needs roughly
a finite thickness, corresponding to the cell 10 WT × 10D × 10 c/D =1000 cells in the two
dimensions, while the model of Conway is horizontal directions. With 128 cells in the ver-
infinitely thin. Moreover, the radial velocity tical direction, this roughly needs 128M cells in
evolves very rapidly at the disc position, which the center of the wind farm, in order to obtain a
makes it difficult to precisely interpolate the good resolutions. These types of steady-state
velocities at the correct disc position. This simulations can be solved in less than 10 hours
was also commented by Schaffarczyk and on a large computer cluster.
Conway [19] who did a similar comparison of However, please note that the studies presented
the analytical actuator disc with a CFD actuator here have been carried out without considering
disc. the large scale atmospheric turbulence. Com-
This can present a problem rarely mentioned puting the interaction between the large scale
in the paper dealing with actuator discs. The turbulence of the atmosphere and the smaller
position where the velocities are estimated, at scale of turbulence of the wind turbine wake is a
the rotor, is a place where the velocity gradients very complex task that only prohibitively expen-
are relatively high. As those velocities are sive large eddy-simulation seems to be able to
necessary to estimate accurately the related carry out [15].

6
Z=−1D Z=−1D Z=−1D
1.005 0.01 2
Full Rotor Computation

Normalized tangential velocity Ut/U∞


AD 10 cells/D
Normalized axial velocity Uz/U∞

1 1.5 AD 20 cells/D
0.008

0.995 1

Pressure
0.006
0.99 0.5

0.004
0.985 0

0.002
0.98 −0.5

0.975 0 −1
0 0.5 1 1.5 2 0 0.5 1 1.5 2 0 0.5 1 1.5 2
Normalized radial direction r/D [−] Radial direction r/D [−] Radial direction r/D [−]
Z=1D
Z=1D Z=1D 2
1.1 0.14

Normalized tangential velocity Ut/U∞


Normalized axial velocity Uz/U∞

1 0.12 1

0.9 0.1 0

Pressure
0.8 0.08
−1
0.7 0.06
−2
0.6 0.04

−3 Full Rotor Computation


0.5 0.02 AD 10 cells/D
AD 20 cells/D
0.4 0 −4
0 0.5 1 1.5 2 0 0.5 1 1.5 2 0 0.5 1 1.5 2
Normalized radial direction r/D [−] Radial direction r/D [−] Radial direction r/D [−]
Z=3D Z=3D Z=3D
1.1 0.12 0.5
Normalized tangential velocity Ut/U∞
Normalized axial velocity Uz/U∞

1
0.1 0
0.9
0.08
−0.5
Pressure
0.8
0.06
0.7
−1
0.04
0.6
−1.5 Full Rotor Computation
0.5 0.02
AD 10 cells/D
AD 20 cells/D
0.4 0 −2
0 0.5 1 1.5 2 0 0.5 1 1.5 2 0 0.5 1 1.5 2
Normalized radial direction r/D [−] Radial direction r/D [−] Radial direction r/D [−]

Figure 6: Comparison between the full rotor computation and the actuator disc model (with 10 and 20 c/D) of the
normalized axial velocity in a cross section at hub height and at different positions upstream and downstream of
the wind turbine.

5 Conclusion effectively for both the large scale atmospheric


scales and the smaller wake scales, still limits its
This paper introduced the actuator shape use. More work is therefore needed on the topic
model, a cost effective method to redistribute of multiscale turbulence in order to address this
any kind of shape forces in a computer do- problem.
main. The special case of the actuator disc is
validated through two fundamentally different
models, an analytical solution for an heavily Acknowledgement
loaded actuator disc, and full rotor computation.
Both models give a close agreement with the This work has been funded by the Danish P.S.O.
actuator shape model introduced. The results Research Project "Bottlenecks".
show that even with a coarse resolution of the
disc, the far wake region is well defined, which
gives the possibility to carry out numerically full References
wind farm wake computation. [1] I. Ammara, C. Leclerc, and C. Masson. A viscous
three-dimensional differential/actuator disc method for
This study seem to show that the actuator the analysis of wind farms. ASME J. Sol. Energy Eng.,
124(4):345–356, 2002.
disc is by itself a mature technology for the study
of wind turbine wake. However, the lack of [2] J. T. Conway. Analytical solutions for the newtonian
proper turbulence model that can account cost gravitational field induced by matter within axissym-

7
metric boundaries. Mon. Not. R. Astrom. Soc., 316: [19] A. P. Schaffarczyk and J. T. Conway. Comparison of a
540–554, 2000. nonlinear actuator disk theory with numerical integra-
tion including viscous effects. CASI J., 46:209–215,
[3] J. T. Conway. Exact solutions for the magnetic fields of 2000.
axisymmetric solenoids and current distributions. IEEE
Transactions on Magnetics, 37:2977–2988, 2001. [20] N. N. Sørensen. General Purpose Flow Solver Applied
to Flow over Hills. PhD thesis, Technical University of
[4] J. T. Conway. Application of an exact nonlinear actuator Denmark, 1994.
disk theory to wind turbines. ICNPAA 2002, Melbourne, [21] N. N. Sørensen. HypGrid2D – a 2-D mesh genera-
Florida, USA, 15 -17 May 2002. tor. Technical Report Risø-R-1035(EN), Risø National
Laboratory, 1998. [PDF].
[5] J. T. Conway. Analytical solutions for the actuator disc
with variable radial distribution of load. J. Fluid Mech., [22] N. Troldborg. Actuator Line Modeling of Wind Tur-
297:327–355, 1995. bine Wakes. PhD thesis, DTU-MEK, Denmark, 2008.
[PDF].
[6] J. T. Conway. Exact actuator disk solutions for non-
uniform heavy loading and slipstream contraction. J. [23] L. J. Vermeer, J. N. Sørensen, and A. Crespo. Wind
Fluid Mech., 356:235–267, 1998. turbine wake aerodynamics. Progress in Aerospace
Sciences, 39:467–510, 2003.
[7] J. T. Conway. Prediction of the performance of heavily [24] J. W. Wrench. Review of: Heuman lambda function.
loaded propellers with slipstream contraction. CASI J., Mathematics of Computation, 20:627–628, 1966.
44:169–174, 1998.
[25] F. Zahle and N. N. Sørensen. Characterisation of the
[8] A. Crespo, J. Hernandez, and S. T. Frandsen. Survey unsteady flow in the nacelle region of a modern wind
of modelling methods for wind turbine wakes and wind turbine. EWEC Marseille, 2009. [PDF].
farms. Wind Energy, (2):1–24, 1999.

[9] L. Gilling, N. N. Sørensen, and P.-E. Réthoré. Imposing


resolved turbulence by an actuator in a detached eddy Appendix
simulation of an airfoil. EWEC Marseille, 2009. [PDF].

[10] B. P. Leonard. A stable and accurate convective mod- A Bessel-Laplace Integrals


elling procedure based on quadratic upstream interpo-
lation. Comp. Meth. in Appl. Mech. and Engrng., 19:
Conway [2, 5] has developed a recursive
59–98, 1979.
method to express the Bessel-Laplace integrals
[11] F. R. Menter. Zonal two equation k-ω turbulence mod- (BLI) of type I(λ,µ,ν) Eq.7 using elliptic integral
els for aerodynamic flows. AIAA Journal, (93-2906), function to derive a numerical solution.
1993.
In order to solve the parabolic wake case, only
[12] J. A. Michelsen. Basis3D - a platform for development three basic BLIs are needed (I(−1,2,1) , I(0,2,0)
of multiblock PDE solvers. Technical report AFM 92- and I(0,2,1) ). While the analytical expression of
05, Technical University of Denmark, Lyngby, 1992. I(−1,2,1) is given in [3], the analytical expression
[13] R. Mikkelsen. Actuator Disc Methods Applied to Wind
of I(0,2,0) and I(0,2,1) could not be found in the
Turbines. PhD thesis, Technical University of Denmark, literature. The recursive method is therefore
Mek dept, 2003. [PDF]. used to derive them. The algorithm used to
implement this method is briefly summarized in
[14] S. V. Patankar and D. B. Spalding. A calculation proce-
dure for heat, mass and momentum transfer in three- the following.
dimensional parabolic flows. Int. J. Heat Mass Transfer,
15:1787–1972, 1972. First of all, it is interesting to notice the sym-
metrical property of the Bessel-Laplace inte-
[15] P.-E. Réthoré. Wind Turbine Wake in Atmospheric Tur-
bulence. PhD thesis, Aalborg University - Risø DTU, grals [2],
October 2009. [PDF].
I(λ,µ,ν) (R, r, z) = I(λ,ν,µ) (r, R, z). (10)
[16] P.-E. Réthoré and N. N. Sørensen. Actuator disc model
using a modified Rhie-Chow/SIMPLE pressure correc-
Furthermore, the Bessel-Laplace integrals
tion algorithm. EWEC Brussels, 2008. [PDF]. can be related to each other through recursive
relations and some initial start formulae. The
[17] C. M. Rhie and W. L. Chow. Numerical study of the tur- recursive relations are given as [2],
bulent fow past an airfoil with trailing edge separation.
AIAA Journal, 21:1525–1532, 1983. 4(ν − 1)ω
I(0,ν,ν) = I(0,ν−1,ν−1)
2ν − 1
[18] B. Sanderse. Aerodynamics of wind turbine wakes
- literature review. Technical Report ECN-E–09-016,
2ν − 3
− I(0,ν−2,ν−2) , (11)
ECN, Netherlands, 2009. [PDF]. 2ν − 1

8
(2ν − 1)|z|k 4
I(1,ν,ν) = 2

8Rr(1 − k 2 ) Λ0 (β, k) = E(k)F (β, k ′ )

I(0,ν−1,ν−1) − ωI(0,ν,ν) , (12)
π
 
+ K(k) E(β, k ′ ) − F (β, k ′ ) , (20)
r r
I(0,ν+1,ν) = I(0,ν,ν−1) + √
R 2ν  where k ′ = 1 − k 2 is the complementary
I(1,ν+1,ν+1) − I(1,ν−1,ν−1) , (13) modulus for an elliptic integral of modulus k .

which is only valid for ν 6= 0,


Using the recursive laws and the start formu-
2(ν + 1)|z| lae, it is possible to create a recursive algorithm
I(0,µ,ν) = I(0,µ,ν+1) to express any Bessel-Laplace integral in terms
r(ν + 1 − µ)
of elliptic integral functions. The procedure
2(ν + 1)R
− I(0,µ−1,ν+1) suggested is to first use the relation I(λ,µ,ν)
r(ν + 1 − µ)
Eq.15 to find an expression where λ = 0. Then
(ν + 1 + µ) use relation I(0,µ,ν) Eq.14 to reduce the gap
+ I(0,µ,ν+2) , (14)
(ν + 1 − µ) between µ and ν in order to use the relation
and I(0,ν+1,ν) Eq.13 and finally I(0,ν,ν) Eq.11.
Whenever possible, a shortcut in the recursion
R can be taken, using the symmetrical relation
I(λ,µ,ν) = I(λ+1,µ+1,ν)
2µ Eq.10. This procedure is only valid for λ ≤ 0,

−I(λ+1,µ−1,ν) , (15) yet this encompasses all the cases described
by Conway’s actuator disc model.
which is only valid for λ < 0 and where
ω = (R2 + r2 + z 2 )/(2rR) is the parame- In order to test the algorithm, one can com-
ter ofpan associated Legendre function and pare to some of the Bessel-Laplace integral
k = 4rR/[(R + r)2 + z 2 ] is the modulus of solutions given in [3].
an elliptic integral.
Using this algorithm it is possible to express
The initial start formulae are I(0,0,0) , I(0,1,0) I(−1,2,1) , I(0,2,0) and I(0,2,1) with respect to el-
and I(0,1,1) and can be found in [3], liptic integral functions. I(−1,2,1) is also given
in [3], if (r<R),
k K(k)
I(0,0,0) = √ . (16)
π rR I(−1,2,1) = E(k)ΘE K
(−1,2,1) + K(k)Θ(−1,2,1)
 
r zΛ0 (β, k)
if r < R, + 2 − |z| (21)
R 2
 
1 |z|k K(k) Λ0 (|β|, k)
I(0,1,0) = 1− √ − (17) if(r>R),
R 2π rR 2
I(−1,2,1) = E(k)ΘE K
(−1,2,1) + K(k)Θ(−1,2,1)
if r > R,
rz
+ Λ0 (β, k), (22)
  2R2
1 |z|k K(k) Λ0 (|β|, k)
I(0,1,0) = − √ + (18) where the coefficients ΘE K
(−1,2,1) and Θ(−1,2,1)
R 2π rR 2 are given as,

r 4r 8(2 − k 2 )
r  
1 1
(2 − k 2 )K(k) − 2E(k) ΘE
(−1,2,1) = − ,
I(0,1,1) = √ (19) k 2π R R 3k 2
πk rR (23)
2 2
r 
p k r (4 − k )(4 − 3k )
where β = arcsin(|z|/ (|z| − R)2 + z 2 ) is the ΘK
(−1,2,1) =
Jacobi amplitude, K(k) is the complete elliptic 2π R 3k 4
r2

integral function of the first kind, E(k) is the com- − 2 . (24)
plete elliptic integral function of the second kind R
and Λ0 (β, k) is Heuman’s Lambda function de-
fined as [2, 24],

9
I(0,2,0) is found using the recursive rules, if
(r>R)

E(k)ΘE K
(0,2,0) + K(k)Θ(0,2,0)
I(0,2,0) =
k 3 π(rR)5/2
Λ0 (β, k)|z|
− , (25)
R2
if (r<R)

E(k)ΘE K
(0,2,0) + K(k)Θ(0,2,0)
I(0,2,0) =
k 3 π(rR)5/2
 
Λ0 (β, k) − 2 |z|
+ , (26)
R2
where the coefficients ΘE K
(0,2,0) and Θ(0,2,0) are
found as Eq.27 and Eq.28.
 
ΘE R 2 −2 + k 2 r

(0,2,0) = −4rR
 
+ k 2 R + k 2 |z|2 , (27)

ΘK 2
−16 + 16k 2 − 3k 4 r

(0,2,0) = rR

− 2k 2 −2 + k 2 R


h i
+ r k 4 (r − 2R) + 4k 2 R |z|2 . (28)

I(0,2,1) is also found using the recursive rules,


if (r>R),

I(0,2,1) = E(k)ΘE K
(0,2,1) + K(k)Θ(0,2,1)
Λ0 (β, k)r
+ , (29)
2R2
if (r<R),

I(0,2,1) = E(k)ΘE K
(0,2,1) + K(k)Θ(0,2,1)
h i
2 − Λ0 (β, k) r
+ , (30)
2R2
where the coefficients ΘE K
(0,2,1) and Θ(0,2,1) are
given as,

2|z|
ΘE
(0,2,1) = √ , (31)
kπR rR
 
|z| k 2 (r − 2R) − 4R
ΘK
(0,2,1) =− √ . (32)
2kπR2 rR

10

Anda mungkin juga menyukai