Anda di halaman 1dari 112

Digital Comprehensive Summaries of Uppsala Dissertations

from the Faculty of Science and Technology 998

Fluid Mechanics of Vertical Axis


Turbines
Simulations and Model Development

ANDERS GOUDE

ACTA
UNIVERSITATIS
UPSALIENSIS ISSN 1651-6214
ISBN 978-91-554-8539-9
UPPSALA urn:nbn:se:uu:diva-183794
2012
Dissertation presented at Uppsala University to be publicly examined in Polhemssalen,
Ångströmslaboratoriet, Lägerhyddsvägen 1, Uppsala, Friday, December 14, 2012 at 13:15 for
the degree of Doctor of Philosophy. The examination will be conducted in English.

Abstract
Goude, A. 2012. Fluid Mechanics of Vertical Axis Turbines: Simulations and Model
Development. Acta Universitatis Upsaliensis. Digital Comprehensive Summaries of
Uppsala Dissertations from the Faculty of Science and Technology 998. 111 pp. Uppsala.
ISBN 978-91-554-8539-9.

Two computationally fast fluid mechanical models for vertical axis turbines are the streamtube
and the vortex model. The streamtube model is the fastest, allowing three-dimensional modeling
of the turbine, but lacks a proper time-dependent description of the flow through the turbine. The
vortex model used is two-dimensional, but gives a more complete time-dependent description of
the flow. Effects of a velocity profile and the inclusion of struts have been investigated with the
streamtube model. Simulations with an inhomogeneous velocity profile predict that the power
coefficient of a vertical axis turbine is relatively insensitive to the velocity profile. For the
struts, structural mechanic loads have been computed and the calculations show that if turbines
are designed for high flow velocities, additional struts are required, reducing the efficiency
for lower flow velocities.Turbines in channels and turbine arrays have been studied with the
vortex model. The channel study shows that smaller channels give higher power coefficients
and convergence is obtained in fewer time steps. Simulations on a turbine array were performed
on five turbines in a row and in a zigzag configuration, where better performance is predicted
for the row configuration. The row configuration was extended to ten turbines and it has been
shown that the turbine spacing needs to be increased if the misalignment in flow direction is
large.A control system for the turbine with only the rotational velocity as input has been studied
using the vortex model coupled with an electrical model. According to simulations, this system
can obtain power coefficients close to the theoretical peak values. This control system study
has been extended to a turbine farm. Individual control of each turbine has been compared to a
less costly control system where all turbines are connected to a mutual DC bus through passive
rectifiers. The individual control performs best for aerodynamically independent turbines, but
for aerodynamically coupled turbines, the results show that a mutual DC bus can be a viable
option.Finally, an implementation of the fast multipole method has been made on a graphics
processing unit (GPU) and the performance gain from this platform is demonstrated.

Keywords: Wind power, Marine current power, Vertical axis turbine, Wind farm, Channel
flow, Simulations, Vortex model, Streamtube model, Control system, Graphics processing
unit, CUDA, Fast multipole method

Anders Goude, Uppsala University, Department of Engineering Sciences, Electricity,


Box 534, SE-751 21 Uppsala, Sweden.

© Anders Goude 2012

ISSN 1651-6214
ISBN 978-91-554-8539-9
urn:nbn:se:uu:diva-183794 (http://urn.kb.se/resolve?urn=urn:nbn:se:uu:diva-183794)
To my family
List of papers

This thesis is based on the following papers, which are referred to in the text
by their Roman numerals.

I Goude, A., Lundin, S., Leijon, M., “A parameter study of the influence
of struts on the performance of a vertical-axis marine current turbine”,
In “Proceedings of the 8th European wave and tidal energy conference,
EWTEC2009”, Uppsala, Sweden, pp. 477–483, September 2009.

II Goude, A., Lalander, E., Leijon, M., “Influence of a varying vertical


velocity profile on turbine efficiency for a Vertical Axis Marine Current
Turbine”, In “Proceedings of the 28th International Conference on
Offshore Mechanics and Arctic Engineering, OMAE 2009”, Honolulu,
USA, May 2009.

III Grabbe, M., Yuen K., Goude, A., Lalander, E., Leijon, M., “Design of
an experimental setup for hydro-kinetic energy conversion”,
International Journal on Hydropower & Dams, 15(5), pp. 112–116,
2009.

IV Goude, A., Ågren, O., “Simulations of a vertical axis turbine in a


channel”, Submitted to Renewable Energy, October 2012.

V Goude, A., Ågren, O., “Numerical simulation of a farm of vertical axis


marine current turbines”, In “Proceedings of the 29th International
Conference on Offshore Mechanics and Arctic Engineering, OMAE
2010”, Shanghai, China, June 2010.

VI Dyachuk, E., Goude, A., Lalander, E., Bernhoff, H., “Influence of


incoming flow direction on spacing between vertical axis marine
current turbines placed in a row”, In “Proceedings of the 31th
International Conference on Offshore Mechanics and Arctic
Engineering, OMAE 2012”, Rio de Janeiro, Brazil, July 2012.

VII Goude, A., Bülow, F., “Robust VAWT control system evaluation by
coupled aerodynamic and electrical simulation”, Submitted to
Renewable Energy, September 2012.

VIII Goude, A., Bülow, F., “Aerodynamic and electric evaluation of a


VAWT farm control system with passive rectifiers and mutual
DC-bus”, Submitted to Renewable Energy, November 2012.
IX Goude, A., Engblom, S., “Adaptive fast multipole methods on the
GPU”, Journal of Supercomputing, DOI 10.1007/s11227-012-0836-0,
In Press, October 2012.

Reprints were made with permission from the publishers.

The author has also contributed to the following paper, not included in the
thesis:

A Yuen, K., Lundin, S., Grabbe, M., Lalander, E., Goude, A., Leijon,
M., “The Söderfors Project: Construction of an Experimental Hydroki-
netic Power Station”, In “Proceedings of the 9th European wave and
tidal energy conference, EWTEC2011”, Southampton, United Kingdom,
September 2011.
Contents

1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1. 4
1.1 Different turbine types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.2 Comparison between wind and marine current turbines . . . . . . . . . . . . . 16
1.3 Vertical axis turbine research at Uppsala University . . . . . . . . . . . . . . . . . . . 17
1.4 Extended studies within this thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.5 Outline of the thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2 Theory for vertical axis turbines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.1 Basic theory and the Betz limit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.2 Extension to include channels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.3 Theory of lift-based vertical axis turbines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.4 Angle of attack including flow curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3 Control strategy for vertical axis turbines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.1 Control of a single turbine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.2 Extension to multiple turbines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
4 Simulation models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .37
4.1 Streamtube models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .37
4.1.1 Description of model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.1.2 Including struts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.1.3 Obtaining lift and drag coefficients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.1.4 Corrections due to flow curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.1.5 Including flow expansion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.2 Vortex models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .48
4.2.1 Implementing the turbine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.2.2 Merging vortices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.2.3 Calculation of velocity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.2.4 Numerical evaluation of the velocity field . . . . . . . . . . . . . . . . . . . . . 53
5 Simulation results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .66
5.1 Evaluation of simulation tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
5.1.1 Strut modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
5.1.2 Expansion model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
5.1.3 Tip correction model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
5.1.4 Curvature modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
5.1.5 Vortex model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
5.1.6 Concluding remarks about the simulation tools . . . . . . . . . . . . 74
5.2 Results from papers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .74
5.2.1 The effects of struts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
5.2.2 The effects of a velocity profile . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
5.2.3 Design of a turbine for use in a river . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
5.2.4 Turbines in channels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
5.2.5 Turbines in an array . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
5.2.6 Simulations of control systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
5.2.7 Control of multiple turbines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
6 Conclusions ................................................................................................ 95
7 Suggestions for future work ...................................................................... 97
8 Summary of papers .................................................................................... 98
9 Errata for papers ...................................................................................... 102
10 Acknowledgments ................................................................................... 103
11 Summary in Swedish ............................................................................... 104
References ...................................................................................................... 107
Nomenclature

A m2 Turbine cross-sectional area


A∞ m2 Asymptotic area of streamtube enclosing turbine
Ac m2 Cross-sectional area of a channel
Ad m2 Area of turbine disc
Ae m2 Streamtube area far downstream/center of turbine
AR − Aspect ratio of a blade
CD − Drag coefficient
CDs − Drag coefficient for strut
CD∞ − Drag coefficient for infinitely long blade
CL − Lift coefficient
CLs − Lift coefficient for strut
CL∞ − Lift coefficient for infinitely long blade
CN − Normal force coefficient
CN0 − Normal force coefficient without curvature corrections
CNs − Normal force coefficient for strut
CP − Power coefficient
CPe − Power coefficient equivalent for extracted power
CPmax − Maximum power coefficient for a given flow velocity
CT − Tangential force coefficient
CT s − Tangential force coefficient for strut
D m Turbine diameter
F − Velocity correction factor
FD N Drag force
FD0 N Drag force at zero angle of attack
FL N Lift force
FN N Normal force
FNl N/m Normal force per meter
FR N Force in radial direction
FT N Tangential force
FT s N Tangential force on strut
Fx N Aerodynamic force from a blade on flow in a streamtube
Fxs N Aerodynamic force from a strut on flow in a streamtube
H m Channel height
J kgm2 Moment of intertia
L m Distance between struts
K N Constant to determine lift force

9
Ma − Mach number
N − Number of particles (FMM)
Nb − Number of blades
Nbox − Number of boxes (FMM)
Nd − Number particles per box (FMM)
Np − Number of panels
Ns − Number of struts
Nt − Number of turbines
Nv − Number of vortices
NFi − Set of all boxes in near field of box i (FMM)
P W Power absorbed by the turbine
Pe W Power extracted from the turbine
Pe,tot W Total power extracted from the turbines of a farm
Ptot W Power available in flow
R m Turbine radius
Rinner m Strut inner attachment point
Router m Strut outer attachment point
Re − Reynolds number
Ts Nm Torque from strut
V m/s Flow velocity
V0 m/s Flow velocity far upstream of turbine
V∞ m/s Asymptotic flow velocity
Vabs m/s Magnitude of incoming flow velocity at blade position
Vb m/s Blade velocity
Vd m/s Flow velocity at turbine disc
Ve m/s Flow velocity far downstream/center of turbine
Vi m/s Vortex velocity
Vr m/s Relative flow velocity for a blade (absolute value)
Vrs m/s Relative flow velocity for a strut (absolute value)
Vs m/s Flow velocity at strut position
Vre f m/s Reference flow velocity for estimating angle of attack
Vrel m/s Relative flow velocity for blade
Vrelz m/s Relative flow velocity for blade in its own reference frame
Vs m/s Far downstream velocity of flow passing outside turbine
Vs j m/s Flow velocity at strut segment
Vω m/s Velocity due to vortices
W m2 /s Complex velocity potential
Wb m2 /s Complex velocity potential for blade velocity
a − Axial induction factor
ai − Multipole coefficient i (FMM)
as − Slope of lift coefficient curve in CL /α plot
b m Circle radius used for conformal mapping
bi − Local coefficient i (FMM)

10
by∞ − Normalized asymptotic streamtube width
bye − Normalized streamtube width at the turbine center
bz∞ − Normalized asymptotic streamtube height
bze − Normalized streamtube height at the turbine center
c m Blade chord
cs m Strut chord
csound m/s Speed of sound
c0 m Reference chord for struts
g m/s2 Gravitational acceleration
h m Turbine height
k1 kgm2 First control system constant
k2 kgm2 Second control system constant
k3 kgm2 Third control system constant
kd1 − Constant for time estimate of direct evaluation (FMM)
kd2 − Constant for time estimate of direct evaluation (FMM)
l m Blade length in streamtube
p − Number of multipole coefficients (FMM)
p0 N/m2 Pressure far upstream of turbine
patm N/m2 Atmospheric pressure
pd1 N/m2 Pressure directly in front of turbine disc
pd2 N/m2 Pressure directly after turbine disc
pe N/m2 Pressure far downstream of turbine
r0 m Box center (FMM)
rs m Radial position on a strut
r m Arbitrary position
ri m Vortex position
s m Position on blade surface in transformed plane
t s Time
tb m Blade thickness
td s Time estimate for direct evaluation (FMM)
u − Interference factor
x m Position in the x-direction
x0 m Blade attachment point
x0r − Normalized blade attachment point
y m Position in the y-direction
y∞ m Asymptotic streamtube position in y-direction
yd m Streamtube position at turbine disc in y-direction
ye m Streamtube position far downstream in y-direction
Δy m Streamtube width
z m Position on blade surface in the blades reference frame
z0 m Position on blade surface in the turbines reference frame
zb m Blade position
Δz m Streamtube height

11
Γ p m3 /s Circulation of three dimensional point vortex
Γ m2 /s Circulation of two dimensional vortex
Δ m Cutoff radius used for vortex merging
Ω rad/s Turbine rotational velocity
Ωi rad/s Rotational velocity of turbine i
Ω1 rad/s First control system rotational velocity constant
Ω2 rad/s Second control system rotational velocity constant
Ω3 rad/s Third control system rotational velocity constant
α − Angle of attack
αb − Corrected angle of attack
αs − Angle of attack for strut
β − Direction of incoming wind
δ − Blade pitch angle
ε m Cutoff radius of Gaussian vortex kernel
η − Angle of blade relative to the vertical axis
ηs − Angle of strut relative to the horizontal plane
θ − Blade azimuthal position shifted 90 degrees
θb − Blade azimuthal position
λ − Tip speed ratio
λe − Equilibrium tip speed ratio
λmax − The tip speed ratio that gives highest power coefficient
ν m2 /s Kinematic viscosity
ρ kg/m3 Density of fluid
σ N/m2 Stress
ϕ − Angle of relative wind
φ m2 /s Velocity potential
ω 1/s Vorticity

12
Abbreviations

CPU Central processing unit


CUDA Compute unified device architecture
FEM Finite element method
FMM Fast multipole method
FVM Finite volume method
GPU Graphics processing unit
L2L Local to local translation
L2P Local to particle evaluation
M2L Multipole to local translation
M2M Multipole to multipole translation
NACA National Advisory Committee for Aeronautics
P2M Particle to multipole initialization
P2P Particle to particle interaction (direct evaluation)
RANS Reynolds-averaged Navier-Stokes equations
SIMD Single instruction multiple data
SSE Streaming SIMD Extensions

13
1. Introduction

A turbine is used to convert the energy from a moving fluid into rotational mo-
tion, which in turn can drive an electric generator. The best suited turbine for
this energy conversion depends on the characteristics of the flow. One case,
which is present in gas turbines and traditional hydro power plants, is flow con-
strained by walls, such as flow within a pipe. Here, the turbine cross-sectional
area usually covers the entire flow and the flow has a large pressure difference
that drives the turbine. A second case is the free flow, where no confining walls
are present. This is the case for wind power and often a reasonable approxi-
mation for tidal power in the ocean. In free flow, the fluid can pass around the
turbine and the available energy is the kinetic energy in the flow. This thesis
mainly treat the free flow case, but also a hybrid case where there are confin-
ing walls, but the turbine does not cover the whole cross-sectional area of the
flow, is studied in this thesis. This situation occurs in a river, where the river
cross-sectional shape area usually prevents a turbine from covering the entire
cross-section.

1.1 Different turbine types


The typical turbine design for wind power is a horizontal axis turbine, where
the rotational axis of the turbine is parallel to the flow direction [1, chapter 1].
In this thesis, however, the vertical axis turbine will be investigated. Here, the
rotational axis is perpendicular to the flow direction. This kind of turbine is
sometimes called “cross-flow turbine”, as the turbine in principle also can be
tilted 90 degrees to have a horizontal axis while still having its rotational axis
perpendicular to the flow. The traditional name “vertical axis turbine” will be
used here, even for situations where the rotational axis is tilted, since this is
the most commonly used name.
There are two different types of vertical axis turbines. The first type is
based on the drag force and is often called the Savonius rotor after the Finnish
inventor Sigurd Johannes Savonius, despite that Savonius only patented an im-
provement of older designs [2]. This improvement is neither implemented on
all present drag-based turbines. Drag-based devices rely on variation of the
drag coefficient with respect to the orientation of the object. To create a rea-
sonably efficient drag-based turbine, the drag coefficient should be high in one
direction and low in the opposite direction, which gives a torque on the tur-
bine. Drag-based devices achieve lower power coefficients than the lift-based

14
Figure 1.1. Different types of vertical axis turbines.

devices described in the section below [3, chapters 2, 7]. Another drawback is
that the amount of construction material in drag devices is quite high (as can
be seen in figure 1.1). This cost is inhibiting the construction of large turbines,
as material usage is proportional to the volume, i.e. the cube of the character-
istic length of the turbine, while the power absorption is proportional to the
cross-sectional area, i.e. the square of the characteristic length.
The second type is the lift-based turbine, which was originally invented by
the French engineer George Jean Marie Darrieus [4] in the 1920’s (approxi-
mately one year after Savonius patented his design). The patent application of
Darrieus covers both the curved blade turbine and the H-rotor (see figure 1.1),
as well as turbines with varying pitch angle and ducted turbines. It is suggested
in the patent that the designs work both for wind and tidal energy. The aim
of the curved blade design is to reduce the bending stresses in the blades due
to centrifugal forces. The North American company Flowind commercialized
in the 1980s the Darrieus turbine with the curved blade design [5, chapter 1].
During that time, the curved blade turbine was also studied by Sandia National
Laboratories, which is the main reason why much of the published work on
Darrieus turbines is on the curved blade design. This thesis will instead focus
on the straight blade H-rotor design, which currently is in development at Up-
psala University. With recent progress for light materials, composites can be
used in the turbine construction, which reduces centrifugal forces due to the

15
lighter structure. This makes the H-rotor design more feasible. The straight
blade design has the advantages that straight blades are easier to manufacture
and by attaching the blades with struts, it is possible to place the upper bear-
ing much closer to the turbine center, reducing the bending moment on the
axis. In addition, the constant radius of the straight blade design gives a larger
cross-sectional area. Disadvantages compared to the curved blade design are
the addition of extra struts and the higher bending moments due to centrifugal
forces.
The main aerodynamic advantage of vertical axis turbines, compared to
standard horizontal axis turbines, is the independence of flow direction, re-
moving the need for a yaw mechanism. For water flow, an additional advan-
tage is that the cross-sectional area can be more flexibly chosen as both height
and diameter can be varied (and the diameter can vary with height). This can
be useful in shallow water where a turbine with a large width and small height
can cover a larger area than a horizontal axis turbine, as the cross-sectional
area of a horizontal axis turbine is circular. Disadvantages of the vertical axis
turbines are the lower power coefficients and that the turbines are typically not
self-starting.
The vertical axis turbine can have its generator on the ground, which in
the wind power case simplifies maintenance, tower construction and makes
the weight of the generator less important. This is beneficial for direct driven
generators, which typically have large diameters. The use of direct driven gen-
erators further reduces the number of moving parts in the system. One major
concern for vertical axis turbines is the cyclic blade forces in each revolution,
which leads to torque oscillations and material fatigue. For further compar-
isons between horizontal and vertical axis turbines (and also between curved
and straight blade turbines) see e.g. [6].

1.2 Comparison between wind and marine current


turbines
Even though wind turbines operate in air (gas) while marine current turbines
operate in water (liquid), there are many similarities between the two. Tra-
ditionally, water is considered an incompressible fluid and can therefore be
modeled with the incompressible Navier-Stokes equations. For air, it is typi-
cally expected that compressibility effects can be neglected for Mach numbers
Ma within the range
Vrel
Ma = < 0.3.
csound
Here, Vrel is the relative flow velocity (measured in the blades rest frame) and
csound is the speed of sound in the fluid [7, chapter 9]. Note that the major con-
tribution to Vrel originates from the blades’ own motion for lift-based turbines.

16
The speed of the blades in a wind turbine is typically too low for the Mach
number to be above 0.3 and wind turbines can therefore also be modeled with
incompressible aerodynamics. Although both wind and marine turbines can be
studied with the incompressible Navier-Stokes equations, there are still some
characteristic differences. One difference is that for marine current turbines,
there is both a sea bed and a free surface that bounds the flow. Another dif-
ference is the risk for cavitation at too high flow velocities. Cavitation would
modify the flow characteristics and can cause damage to the turbine [8].
The energy absorbed by a turbine is proportional to the fluid density and
the cube of the flow velocity. As the density of water is 800 times higher than
the density of air, comparatively low fluid velocities are adequate for marine
current power generation. For equal cross-sectional area, a wind speed of
10 m/s has the same incoming kinetic power as a water flow speed of 1.1 m/s.
However, as the forces only are proportional to the square of the flow velocity,
the marine current turbine experiences approximately 9.3 times higher fluid
mechanical forces than a wind turbine at the same conditions (assuming that
the turbines are identical) and rotates 9.3 times slower. The increased forces
for marine current turbines require both stronger blades and support structure.
Many of the experimental vertical axis turbines for water have used relatively
large blades and thereby low optimal tip speed ratios [9–12].
One important parameter for the effectiveness of the turbine is the Reynolds
number, which for a blade is defined as
cV
Re = ,
ν
where V is the flow velocity, c is the blade chord and ν is the kinematic viscos-
ity. A higher Reynolds number usually decreases the drag losses and increases
the stall angle, which is beneficial for vertical axis turbines. For 20 ◦ C, the
kinematic viscosities are 15.1 μ m 2 /s for air and 1.00 μ m2 /s for water [13, ap-
pendix A]. Under the conditions of equal power extraction mentioned above
(i.e. 9.3 times higher flow velocity for the wind turbine), this would give a
63 % higher Reynolds number for the marine current turbine, which is within
the same order of magnitude as the wind turbine.

1.3 Vertical axis turbine research at Uppsala University


At the Division of Electricity at Uppsala University, three vertical axis wind
turbines have been built. The first turbine had a cross-sectional area of 6 m 2
and was later followed by a turbine with the cross-sectional area 30 m 2 and the
rated power 12 kW [14–16]. This larger turbine is used for most of the experi-
ments. A 10 kW turbine for telecom applications has also been built [17]. Fur-
ther, a 200 kW turbine has been constructed by the spin-off company Vertical
Wind AB [18]. Additionally, a marine current turbine (described in paper III)
is scheduled to be deployed by the end of 2012.

17
Several simulation tools for turbine simulations have previously been de-
veloped at the division. A two-dimensional inviscid vortex model based on
conformal mappings for the blades has been created by Deglaire et al. [19]. In
the turbine implementation, each blade is solved independently [20], allowing
for coupling to an elastic method developed by Bouquerel et al., see paper IV
in [21]. A multibody version for simulating turbines has been developed by
Österberg et al., see [22] and paper III in [21]. Two streamtube models have
also been implemented by Deglaire and Bouquerel.

1.4 Extended studies within this thesis


This thesis focuses on the fluid mechanical modeling of the vertical axis tur-
bine, and two different simulation tools have been developed. The first simu-
lation tool uses the streamtube model and the development of this tool started
from the basic streamtube model implemented by Bouquerel, which is based
on the model of Paraschivoiu [3]. All additional modeling and code develop-
ment have been developed within this thesis.
The second simulation tool uses a vortex model, and this tool has been
developed from scratch within this work. This model is based on empirical
data for lift and drag coefficients instead of the conformal mapping method
by Deglaire, which is based on inviscid theory. The computational speed is
crucial for the developed vortex model and large efforts have been put into
this. The existing implementation of the fast multipole method by Stefan Eng-
blom [23] has been significantly improved and ported to a GPU (paper IX)
Several studies have been carried out with the two simulation models. The
streamtube model has been used to study losses due to struts (paper I), the
effects of a velocity profile (Paper II) and to design a turbine for deployment
in a river (paper III). The more computationally demanding vortex model has
been used to study turbines in channels (paper IV) and turbine arrays (paper V
and VI). The vortex model is also coupled to an electrical model to study con-
trol systems for a single turbine (paper VII) and extended simulations analyze
control systems for a turbine farm (paper VIII).

1.5 Outline of the thesis


After the introduction, theory for vertical axis turbines is presented in chap-
ter 2. This is followed by an introduction to control systems in chapter 3.
The theory and implementation for the simulation models are then presented
in chapter 4, which also includes the GPU implementation of the fast multi-
pole method. The results from the simulations are given in chapter 5, where
the first part evaluates the accuracy of the simulation models and the second
part summarizes the results from the articles. The thesis ends by conclusions,

18
suggestions for future work, summary of papers, errata for papers and ac-
knowledgments.

19
2. Theory for vertical axis turbines

Given a cross-sectional area A perpendicular to a homogenous flow of a fluid,


the kinetic power that passes through this area is given by
1
Ptot = ρ AV 3 , (2.1)
2
where ρ is the density and V is the flow velocity. If the flow is not confined
by any surrounding boundaries, the kinetic power is the available power for a
wind/current turbine. The efficiency (i.e. outgoing power divided by incoming
power) would be one possible measure of how good the energy conversion
is. However, adding a turbine will change the velocity and force parts of the
flow to pass outside the turbine area and thereby change the kinetic energy
that passes through this area A. Moreover, some kinetic energy is left in the
flow and can possibly be used later. Therefore, turbine performance is usually
measured with the power coefficient instead, which is defined as
P
CP = 1
, (2.2)
2 ρ AV∞
3

where P is the power absorbed by the turbine and V∞ is the asymptotic up-
stream flow velocity. With this expression, the absorbed power is compared
to the power that would have passed through the cross-sectional area, if the
turbine would be absent, instead of compared to the power that actually passes
through the area. Since this expression is normalized against an expression
that does not change with the turbine characteristics, it is a better measure than
efficiency. Improving the power coefficient will give higher power absorption,
which is not always the case with efficiency.

2.1 Basic theory and the Betz limit


One of the most basic approximations of a turbine is the one used in the tradi-
tional Betz theory [24], where the turbine is approximated as a single flat disc
with a constant pressure drop over the whole turbine surface. All flow passing
through the disc is encapsulated in a streamtube that starts far ahead of the
turbine and ends far behind. By making the assumption that the pressure at
both ends of the streamtube is the atmospheric pressure p atm , and by using the

20
Ad
V∞ Vd Ve
patm pd1 pd2 patm
A∞
Ae

Figure 2.1. The streamtube used in the Betz limit derivation. The dashed line shows
the control volume used for momentum conservation.

Bernoulli equation before and after the turbine


1 1
patm + ρ V∞2 = pd1 + ρ Vd2 , (2.3)
2 2
1 2 1
pd2 + ρ Vd = patm + ρ Ve2 , (2.4)
2 2
combined with continuity

A∞V∞ = Ad Vd = AeVe (2.5)

and momentum conservation for the control volume in figure 2.1 (marked with
the dashed line)
ρ AeVe2 − ρ A∞V∞2 = Ad (pd2 − pd1) , (2.6)
it is possible to show that the velocity at the turbine disc Vd is equal to
V∞ +Ve
Vd = . (2.7)
2
Considering that the Betz theory assumes no losses, the power absorbed by
the turbine is given as the difference between incoming and outgoing power in
the fluid
1 1
P = ρ A∞V∞3 − ρ AeVe3 . (2.8)
2 2
If the axial induction factor a, defined as

Vd = (1 − a)V∞ , (2.9)

is combined with expression with equations (2.5) and (2.8), it can be shown
that the power will reach its maximum value for a = 1/3, and the optimal
power is given by
16 1
P= · ρ AdV∞3 (2.10)
27 2
where 16/27 is the traditional Betz limit, limiting the power coefficient to
approximately 59.3 %.

21
Vs
V0 p0

p0 Ad
V0 Vd Ve
Ac pe
A0 pd1 pd2
Ae
V0

Figure 2.2. Illustration of a streamtube confining the flow that passes through the
turbine disc for a channel of cross-sectional area A c .

2.2 Extension to include channels


The Betz theory assumes no outer boundaries in the system. For a turbine op-
erating in water, it is more common that boundaries are present. One example
is a river, where the flow is limited by the width and depth. The outer walls
prevent the flow from expanding, pushing more flow through the turbine. This
occurs in a traditional hydro power plant, where the entire flow is forced to
pass through the turbine, which results in much higher power absorption than
the Betz limit [25].
To analyze this case analytically, assume that the flow upstream of the tur-
bine has constant velocity V0 and pressure p 0 (see figure 2.2). Note that in
this case, the pressure upstream and the pressure downstream are not equal.
Instead, there will be a drop in pressure, which, for open channel flow, would
correspond to a drop in the surface level. Due to the continuity of the pressure,
the pressure inside the streamtube and outside has to be the same downstream
(pe ) . The cross-sectional area of the channel is A c and the cross-sectional area
of the turbine is Ad . In this case, the Bernoulli equation gives

1 1
p0 + ρ V02 = pd1 + ρ Vd2 , (2.11)
2 2
1 2 1 2
pd2 + ρ Vd = pe + ρ Ve , (2.12)
2 2
1 2 1 2
p0 + ρ V0 = pe + ρ Vs , (2.13)
2 2
the continuity equation gives

A0V0 = Ad Vd = AeVe , (2.14)

(Ac − A0 )V0 = (Ac − Ae )Vs (2.15)

22
and momentum conservation for a control volume that encloses the entire
channel gives
ρ AeVe2 + ρ AsVs2 − ρ AV02 = Ad (pd2 − pd1) + Ac (p0 − pe ) . (2.16)
The velocity at the turbine can be derived from equations (2.11) – (2.16) as
Ve (Vs +Ve )
Vd = , (2.17)
Vs + 2Ve −V0
which in the free flow limit (Vs → V0 ) reduces to equation (2.7). The force on
the turbine becomes
1  
Fx = Ad (pd1 − pd2) = Ad Vs2 −Ve2 , (2.18)
2
giving the power as
 
1 Ve (Vs +Ve ) Vs2 −Ve2
P = FxVd = Ad . (2.19)
2 Vs + 2Ve −V0
From equation (2.19), it can be found that the highest power absorption is
obtained when Ve = V0 /3, which actually is the same as for the free flow. The
highest power is thus given by
16 1 1 3
P= · 2 · ρ AdV0 (2.20)
27 Ad 2
1 − Ac

and the pressure drop is


 
4 AAdc 3 − AAdc
2
pe − p 0 =  2 ρ V0 . (2.21)
Ad
9 1 − Ac

From these results, it can be seen that the maximum theoretical power co-
efficient increases with the factor (1 − A d /Ac )−2 for a channel. In the limit
Ad → Ac , the power coefficient diverges, along with the pressure drop in equa-
tion (2.21). Considering that an infinite drop in pressure is unfeasible, when
the turbine area is almost as large as the channel, the available pressure dif-
ference will start limiting the maximum power coefficient, which will prevent
infinite energy extraction. This is the case for hydro power turbines, where the
power is limited by the difference in water elevation.
The model above assumes that the cross-section of the channel is constant.
An open channel will have a drop in surface level over the turbine and an
extension of the model to include this drop is given by Whelan et al. [26].
This correction has not been included in the present work, as the model is
used for comparisons with the two-dimensional vortex simulations where no
free surface is modeled.

23
R

Ω
θb

y
x

Figure 2.3. Illustration of a vertical axis turbine.

2.3 Theory of lift-based vertical axis turbines


The H-rotor is a lift-based design, which means that the aerodynamic torque is
generated by the lift force of the blades. Therefore, airfoil profiles are typically
used for the blades. Airfoils are generally designed to operate at relatively low
angles of attack, where the lift force increases approximately linearly with
the angle of attack and the drag force remains low. When the angle of attack
increases above the stall angle, which for a typical airfoil occurs for angles
around 10 – 15 degrees, the lift force is reduced, and the drag starts to increase
substantially. This work is focused towards blades with a fixed pitch angle. To
keep the angle of attack low without pitching the blades, the blades must move
with a high velocity if the wind comes from the side. To illustrate this, assume
that a blade is located at angle θ b (see figure 2.3). Using complex notation,
this gives the blade position
zb = Reiθb (2.22)
and the velocity of the blade is therefore

Vb = iθ̇b Reiθb . (2.23)

The rotational velocity θ̇b is commonly denoted Ω. The incoming wind V, if


complex, represents wind from any direction. The blade will now see a relative
wind of
Vrel = V − i ΩReiθb . (2.24)

24
Vrel
α
ϕ
δ

Vb

Figure 2.4. Definitions of angles and velocities. The positive direction for angles is
counter-clockwise, hence α and ϕ are negative for the directions of V b and Vrel in the
figure.

To obtain the angle of relative wind, rotate Vrel with the angle ie−iθb , aligning
the blade motion with the negative real axis,

Vrelz = Vie−iθb + ΩR. (2.25)

The angle of relative wind will be the argument of this complex number. In
the special case of V being real, the angle of relative wind is
cos θb
ϕ = arctan ΩR
(2.26)
V + sin θb

and the absolute value of the relative velocity is


 2
ΩR
|Vrel | = V + sin θb + (cos θb )2 . (2.27)
V

The angle of attack is given by

α = ϕ +δ, (2.28)

where δ is the blade pitch angle (see figure 2.4). Equation (2.26) shows how
the angle of attack varies during a turbine revolution and how it decreases as
the rotational velocity increases. As an example, ΩR/V = 4 gives a maximum
angle of attack of around 14 degrees, approximately where stall begins to oc-
cur. It should be noted that in equations (2.24) – (2.27), V is the flow velocity
at the blade position. This can be compared with the tip speed ratio, which is
defined as
ΩR
λ= (2.29)
V∞
where the asymptotic velocity is used. Due to the energy extracted, the veloc-
ity at the blade will generally be lower than the asymptotic velocity.

25
The turbine torque is calculated from the tangential force, which is given
by
FT = FL sin ϕ − FD cos ϕ . (2.30)
When the angle of relative wind is low, the approximations sin ϕ ≈ ϕ and
cos ϕ ≈ 1 can be applied. Assume that the pitch angle is zero, hence α =
ϕ . For symmetric blades, the blade forces can be approximated as FL ≈ K α ,
where K is a constant, and FD ≈ FD0 , where FD0 is a constant. The tangential
force can therefore be estimated as
FT = K ϕ 2 − FD0 , (2.31)
showing that when the angle of relative wind decreases, the drag force be-
comes dominating. The conclusion is that for high tip speed ratios, drag will
give a more significant contribution, reducing the power coefficient. At too
low tip speed ratios, the turbine will enter stall, where lift decreases and drag
increases, which also should be avoided. For these reasons, the turbine should
be designed to operate with a tip speed ratio close to the stall limit in order to
obtain the highest possible power coefficient.

2.4 Angle of attack including flow curvature


The expression (2.26) is only valid for infinitely small symmetric blades. The
blade performs a rotational motion, which leads to additional curvature effects,
changing the effective angle of attack. To conclude the theory section, a more
proper derivation will be performed using a rotating flat plate instead.
By the use of conformal mappings, a circle can be transformed into a flat
plate with the Joukowski transformation. The s-plane represents the circle
with radius b and the z-plane a flat plate extending between −2b to 2b giving
the blade chord c as c = 4b. The blade coordinates z in its own frame of
reference is given by
b2
z = s+ . (2.32)
s
Using the same transformation as Deglaire [20], the z 0 plane can be defined as

z0 = (z + x0) e−iδ + iR eiθ = (z + D)ei(θ −δ ) , (2.33)

where D = x0 + iReiδ . By assuming that the blade only rotates around the
center, the blade velocity is given by

Vb = iθ̇ (z + x0) e−iδ + iR eiθ = i Ω z0 (2.34)

with Ω = θ̇ . Introduce a complex velocity potential W , with complex conju-


gate W , such that
dW
= V, (2.35)
dz0

26
hence the potential can be used to calculate the flow velocity V . Assume that
the potential is given on the form
b2 iΓ
W (s) = Vabs ei(−β +θ −δ ) s +Vabs e−i(−β +θ −δ ) − log (s) +W1 (s), (2.36)
s 2π
where Vabs eiβ is the flow velocity at the blade position. Now, construct a
potential such that
dWb
= Vb , (2.37)
dz0
which gives
dWb dz0 dz0   dz
= Vb = −i Ω z0 = −i Ω z + D . (2.38)
ds ds ds ds
Note that on the boundary, z = z. Integrated, this is
 
1 2
Wb = −i Ω z + Dz
2
   
1 2 2 b4 b2
= −i Ω s + 2b + 2 + D s + . (2.39)
2 s s
The no-penetration boundary condition states that the stream function should
be constant (possibly time-dependent) on the boundary. Given that the bound-
ary is moving, the condition becomes
Im [W (s) −Wb (s)] = C. (2.40)
The first part of W (s) already fulfills the condition, while W1 (s) remains to be
determined, hence
Im [W1 (s) −Wb (s)] = C. (2.41)
The boundary condition at infinity states

dW1
= 0, (2.42)
ds s→∞
and on the boundary,
b2
s= (2.43)
s
applies. Write equation (2.41) in terms of complex conjugates
 
1 2 b4
iC = W1 (s) −W1 (s) + i Ω s + 2b2 + 2 +
2 s
 4
  2
  
1 2 2 b b b2
s + 2b + 2 + D s + +D s+
2 s s s
= W1 (s) −W1 (s) +
 4 
b b4 b2 b2 b2 b2 2
i Ω 2 + 2 + b + D + D + D + 2b . (2.44)
s s s s s s

27
Note that constants can be excluded from the potential. Therefore, W 1 (s) can
be identified as
2
b   b4
W1 (s) = −i Ω D+D + 2
s s
 2  
b4 
b iδ −iδ
= −iΩ iR e − e + 2x0 + 2 . (2.45)
s s
The Kutta condition [27] states that the velocity has to be finite at s = b where
ds/dz diverges, giving

dW
= 0⇒
ds s=b
iΓ 1
− = −Vabs ei(−β +θ −δ ) +Vabs e−i(−β +θ −δ ) −
2π b  

i Ω iR eiδ − e−iδ + 2x0 + 2b


= −Vabs 2i sin (−β + θ − δ ) − 2i Ω (−R sin δ + x0 + b)(2.46)
.

Use the reference case of a static wing


iΓ 1
− = Vre f 2i sin α ⇒
2π b
−Vabs sin (−β + θ − δ ) − Ω (−R sin δ + x0 + b)
sin α = , (2.47)
Vre f
and that for small values of x 0

Vre f ≈ (Vabs cos(θ − β ) + ΩR)2 +Vabs
2 sin2 (θ − β ). (2.48)

Now, redefine x 0 in terms of the chord as x 0 = x0r c, which means that x 0r =


0.25 is the quarter chord position and use that 4b = c
 
−Vabs sin(θ − β − δ ) − Ω −R sin δ + x0r c + 4c
sin α =
Vre f
(Vabs cos(θ − β ) + ΩR) sin δ −Vabs cos δ sin (θ − β )
= −
Vre f
 
Ω x0r c + 4c
Vre f
 
−Vabs sin(θ − β ) Ω x0r c + 4c
= sin δ + arctan − . (2.49)
V0 cos(θ − β ) + ΩR Vre f
Assuming small angles of attack, one can approximate

arcsin(α + β ) ≈ α + β (2.50)

28
which gives the simplifications
−Vabs sin (θ − β ) Ω x0r c Ωc
α = δ + arctan − − . (2.51)
Vabs cos(θ − β ) + ΩR Vre f 4Vre f

Note that at the position θ = 0, the blade is at the position θ b = π /2. The
substitution θ = θb − π /2 gives
Vabs cos(θb − β ) Ω x0r c Ωc
α = δ + arctan − − , (2.52)
Vabs sin (θb − β ) + ΩR Vre f 4Vre f

which with β = 0 would correspond to equations (2.26) and (2.28), but in-
cludes mounting position x 0r and flow curvature. As an example, for a turbine
with chord 0.25 m and radius 3 m (i.e.
 the experimental
 turbine in Marsta [16]),
at very high rotational velocities Vre f ≈ ΩR , the change in angle of attack
due to flow curvature is approximately -1.2 degrees. This gives higher angles
of attack upstream and lower downstream.

29
3. Control strategy for vertical axis turbines

Optimizing the power from a wind/marine current turbine does not only re-
quire that the turbine is designed with the highest possible power coefficient.
Another important factor is to make sure that the turbine actually runs at the
tip speed ratio associated with the peak power coefficient. Therefore, a con-
trol strategy which keeps the turbine tip speed ratio near this optimal value is
preferable.
Wind turbines in general are either controlled by pitch or stall regulation,
where the most common design today is a horizontal axis turbine with pitch
regulation [28]. The advantage with pitch control is that it introduces an ad-
ditional parameter that can be controlled, allowing for a more flexible control
system. Pitch control is mainly used in the region above rated wind speed (see
figure 3.1) to keep a smoother power and reduce mechanical loads, as the pitch
angle can be changed to reduce the blade forces [1, Chapter 8]. Stall regula-
tion, instead, reduces the tip speed ratio, which increases the angle of attack.
This will eventually cause stall, which reduces the lift force and increases the
drag force. This will be most prominent for the tangential force, and thereby
the turbine torque, due to the significant increase in the drag force. Pitch con-
trol has been used for vertical axis turbines, mainly to improve performance at
low tip speed ratios [29,30] where stall is avoided by actively altering the pitch
angle to reduce the angle of attack. The angle of attack oscillates between
positive and negative values as the blade moves between the upstream and
downstream section of the turbine. Hence, reducing the angle of attack with
active pitch requires a change in pitch angle during each revolution. An active
pitch mechanism would complicate the turbine further. No pitch mechanisms
are included in the turbines studied here to reduce the sources of mechanical
failure.
Without a pitch mechanism, the remaining parameter to control is the ro-
tational velocity, where the turbine power is controlled by regulating the tip
speed ratio and thereby the power coefficient. Even with a pitch mechanism
installed, it is common for horizontal axis wind turbines to use a fixed pitch
angle in the variable rotational speed region illustrated in figure 3.1 [1, Chap-
ter 8]. The following sections will focus on the variable rotational speed re-
gion, where the aim is to maximize the extracted power.

3.1 Control of a single turbine


One way to control a turbine is to perform real time flow velocity measure-
ments and adjust the tip speed ratio to optimal values. However, this would,

30
1.5
Constant
rotational
Cut in speed Rated wind
Fraction of rated power speed

Variable rotational speed


region, constant CP

0.5 Constant
power

0
0 2 4 6 8 10 12 14 16
Wind speed (m/s)

Figure 3.1. Example of the different control strategies for different flow velocities.
The turbine starts operating at wind speed 3 m/s and operates at optimal power coeffi-
cient (approximately constant tip speed ratio) up to rated rotational velocity (wind
speed 10 m/s), where the rotational velocity is kept constant until rated power is
achieved (wind speed 12 m/s). At higher wind speeds, power is kept constant. There
is also a cut-out wind speed where the turbine is stopped (not shown in the figure).

Turbine stops Stable region

Equilibrium
s

Pe
P
se
ea

Pe
cr

Sp
in

ee
ed

d
e

de
Sp

cr
ea
Power

se
s
s
a se
cre
de
eed
Sp

Rotational velocity

Figure 3.2. Illustration of a control system, where the extracted power only depends
on the rotational velocity.

31
rely on the accuracy of the flow measurements. An alternative approach is
to let the extracted power Pe be a function only of the rotational velocity of
the turbine. This type of control has been used for horizontal axis wind tur-
bines [31], but due to the low power coefficient at low tip speed ratios for
vertical axis turbines, some care has to be taken when transferring this control
system to a vertical axis turbine to avoid that the turbine ceases to rotate for
low tip speed ratios.
An example of a control system only using the rotational speed as input
parameter is illustrated in figure 3.2. The angular acceleration Ω̇ of the turbine
is
P − Pe
Ω̇ = , (3.1)

where J is the moment of inertia of the system. Here, extracted power Pe in-
cludes both power from the generator and electrical and mechanical losses in
the system. If the extracted power Pe is less than the turbine power P, the tur-
bine accelerates, while if the extracted power is larger, the rotational velocity
decreases. In Figure 3.2, the turbine would stop if the turbine has too low
rotational velocity as the turbine power at this low rotational velocity is very
limited (P < Pe ). This region where the turbine will stop is characterized by
a low power coefficient at low tip speed ratios, which may apply to a vertical
axis turbine. The existence of such a region depends on the power absorption
characteristics of the turbine and for high enough wind speeds, this region
typically becomes smaller.
Figure 3.2 shows that there will be an equilibrium where extracted power
equals turbine power (P = Pe ). If this equilibrium occurs at the peak of the
power curve in figure 3.2, maximum energy is extracted. If the extracted power
is normalized the same way as turbine power, the extracted power coefficient
is
Pe
CPe = 1 . (3.2)
2 ρ AV∞
3
 
Combining equations (3.1) and (3.2) gives for the equilibrium where Ω̇ = 0

CPe = CP . (3.3)

Equation (3.2) can be written in terms of tip speed ratio and rotational velocity
as
Pe λ 3
CPe = 1 3
. (3.4)
2 ρ A (R Ω)
Denote λmax the tip speed ratio with the peak power coefficient C Pmax . By
choosing λmax as the desired equilibrium, equations (3.3) and (3.4) give
3
 3
Pe λmax 1 RΩ
CPmax = 1 ⇒ Pe = ρ ACPmax = k2 Ω3 , (3.5)
3 2 λmax
2 ρ A (R Ω)

32
CP at 3 m/s
0.45
CP at 6 m/s
0.4 CP at 12 m/s
CPe
0.35
Power coefficient

0.3
0.25
0.2
0.15
0.1
0.05
0
-0.05
0 1 2 3 4 5 6 7 8
Tip speed ratio

Figure 3.3. The control strategy A from equation ( 3.5) illustrated for three different
flow velocities.

where the constant k 2 is related to λmax and CPmax according to equation (3.5).
Under the conditions that CPmax and λmax are constants, maximum power will
be extracted if the extracted power is chosen to vary with the cube of the rota-
tional velocity and the constant k 2 is chosen according to equation (3.5). This
control strategy will be denoted “strategy A”. With CPmax and λmax constant,
equation (3.5) is independent of flow velocity. However, due to the increased
Reynolds number at higher flow velocities, the airfoil performance will in-
crease, as drag is reduced and stall angle is increased. Therefore, the values
for λmax and CPmax have a small dependence on the Reynolds number, see
figure 3.3. The performance of the strategy in equation (3.5) is plotted in fig-
ure 3.3 for three flow velocities (for specifications on the simulated turbine,
see section 5.2.6). The change in power coefficient with respect to the flow
velocity causes the obtained equilibrium tip speed ratio λ e to be slightly dis-
tanced to λmax , although the obtained power coefficient is very similar to C Pmax
(see figure 3.3). For a flow velocity of 3 m/s, there is an unstable region for tip
speed ratios below 2.2, while for 6 m/s and 12 m/s, the strategy is stable in the
entire interval as long as the turbine power is larger than the mechanical and
electrical losses. Even though the strategy is stable for 6 m/s and 12 m/s, the
difference between turbine power and extracted power is low at low tip speed
ratios, causing a slow acceleration, cf. equation (3.1).
Two modifications to the devised strategy in equation (3.5) have been made,
with the intention to increase the rotational velocity at low tip speed ratios in

33
0.45
0.4
0.35
Power coefficient

0.3
0.25
0.2
0.15 CP at 3 m/s
CP at 6 m/s
0.1
CP at 12 m/s
0.05 CPe at 3 m/s
CPe at 6 m/s
0
CPe at 12 m/s
-0.05
0 1 2 3 4 5 6 7 8
Tip speed ratio

Figure 3.4. Control strategy B from equation ( 3.6) illustrated for three different flow
velocities.

0.45
0.4
0.35
Power coefficient

0.3
0.25
0.2
0.15 CP at 3 m/s
CP at 6 m/s
0.1
CP at 12 m/s
0.05 CPe at 3 m/s
CPe at 6 m/s
0
CPe at 12 m/s
-0.05
0 1 2 3 4 5 6 7 8
Tip speed ratio

Figure 3.5. Control strategy C from equation ( 3.7) illustrated for three different flow
velocities.

34
order to create a more stable and faster control strategy. Strategy B is given as

Pe = 0 Ω ≤ Ω0 ,
2
(3.6)
Pe = k1 Ω (Ω − Ω0 ) Ω > Ω0 ,

where the rotational velocity Ω0 is required before the strategy starts to extract
energy. This strategy obtains higher differences between extracted and turbine
power for low rotational velocities at the cost of having λ e further distanced
from λmax .
Strategy C combines a higher torque at low rotational velocities with the
good performance of strategy A by dividing the rotational velocities in three
regions according to


⎪ Pe = 0 Ω ≤ Ω0 ,

⎨P = k Ω2 (Ω − Ω )
e 1 0 Ω0 < Ω ≤ Ω1 ,
(3.7)
⎪Pe = k2 Ω

3 Ω1 < Ω ≤ Ω2 ,


Pe = k3 Ω2 (Ω − Ω2 ) + k2 Ω2 Ω2 Ω2 < Ω,

where
Ω1
k1 = k2
Ω1 − Ω0
due to continuity at Ω = Ω1 . This control strategy extracts minimum energy
up to Ω0 , where Ω0 for both strategy B and C is chosen to correspond to the
rotational velocity at tip speed ratio 4 and flow velocity 3 m/s. These particular
values are chosen because 3 m/s is considered a reasonable cut-in velocity and
tip speed ratio 4 is the optimal value for the chosen turbine.
The control strategy C is designed to operate according to equation (3.5)
between Ω1 and Ω2 . The region between Ω 0 and Ω1 is the increase from 0 to
the curve in equation (3.5). Hence, with increasing rotational flow velocity, the
tip speed ratio decreases in the region between Ω 0 and Ω1 until it has reached
its desired value for high power capture. At rotational velocities above Ω 2 ,
the tip speed ratio is intentionally decreased to reduce mechanical loads on
the turbine, which creates a softer transition to the constant speed region in
figure 3.1. This region is reasonable to implement on a real turbine, but it has
not been studied in detail in this work and for simplicity, k 1 = k3 was chosen.
The control system is designed to reach Ω 2 at 9 m/s. This strategy is illustrated
in figure 3.5, where it is seen that at 3 m/s, λe is higher that the peak value,
causing a small drop in power coefficient, although the difference is smaller
than for strategy B. For 12 m/s, λe is slightly lower than the peak value due
to the intentional decrease in rotational velocity above Ω 2 . Note that the value
for the constant k1 differs between strategy B and C, while the value for the
constant k2 is equal for strategy A and C. Simulations of the three control
strategies are found in section 5.2.6.

35
3.2 Extension to multiple turbines
It is common to locate several turbines in close proximity to each other in a
farm configuration, as this can give economical benefits due to synergy effects,
since some parts of the system can be commonly used. Therefore, two differ-
ent electrical topologies are suggested, where the second “mutual topology”
is designed to reduce the number of electric components [32]. For additional
possible topologies, see e.g. [33–35].
One way to control multiple turbines is to apply the model described in sec-
tion 3.1 for each turbine in the farm. This can be accomplished by using an
individual electric system for each turbine, where each turbine has a passive
diode rectifier and an inverter (“separate topology”). An alternative approach
is to connect all turbines to the same inverter, but to obtain individual control.
This would for a permanent magnet synchronous generator require some addi-
tional electrical components such as an active rectifier or a DC-DC converter.
One simplification is to connect all turbines to the same DC-bus with pas-
sive rectifiers, which makes the inverter the only parameter to control (“mutual
topology”). This reduces the number of electric components and it is therefore
of interest to study how performance is affected by this simplification.
For the mutual topology, the total power extracted from the entire turbine
farm is chosen as
Nt
Pe,tot = ∑ Pe (Ωi ) , (3.8)
i=1

where Pe (Ωi ) is calculated according to one of the strategies A – C in sec-


tion 3.1. With the total extracted power chosen according to equation (2.1),
the separate and mutual topologies extract the same power if all turbines ex-
perience identical flow velocities.
Other studies have previously been performed for horizontal axis turbines,
with a global control strategy for the entire farm [36, 37]. Farm effects have
been included in the control system in [38, 39] and a method for estimating
the flow velocity within the farm with only rotational velocities and extracted
power as input parameters is presented in [40].

36
4. Simulation models

The flow through a turbine can be simulated by solving Navier-Stokes equa-


tions coupled with the continuity equation. Navier-Stokes equations are valid
for all Newtonian fluids, i.e. both air and water, although in water, cavitation
would require additional modeling. The validity of Navier-Stokes equations
is well established, and the most accurate simulation model would be a direct
solution of these equations. However, Navier-Stokes equations are non-linear
partial differential equations and due to the computational complexity, accu-
rate direct solutions of these equations are numerically very hard to obtain,
even for the two-dimensional case. Considering that the fluid-flow in a large
turbine generally is turbulent, one common simplification is to model the tur-
bulence by e.g. Reynolds-averaged Navier-Stokes equations (RANS), which
include additional turbulence models. This has been done by e.g. [41, 42].
The most common methods to solve Navier-Stokes equations are the Finite
Element Method (FEM) and the Finite Volume Method (FVM), which are
both based on a mesh over the entire simulated volume. For this reason,
these methods are suited for confined regions. There are commercial FEM
and FVM simulation software available, but due to the long simulation times,
other methods have been chosen in this work.
Another method for solving Navier-Stokes equations is to use a vortex
method (see section 4.2). This method can also directly solve Navier-Stokes
equations, and is designed for open flows with small boundaries, which is the
case for a vertical axis turbine. Directly solving Navier-Stokes equations with
this method requires much computational time, and has not been done in this
work. An advantage of the vortex method is that there are many simplifica-
tions available, and by including models of blade forces, the simulation times
can be reduced substantially. This simplification is used within this work.
One even more simplified and very fast method is the streamtube model,
which assumes static flow, and does not give a full description of the flow
through the turbine. Due to the high computational efficiency, this model can
be used to quickly perform many simulations, and it is one of the models used
in this work, even though the simplifications of the flow limit the applicability
of the model.

4.1 Streamtube models


Streamtube models are among the fastest models used for simulating vertical
axis turbines. The first application of streamtube models, usually credited to

37
c

Ω θb

Δθ Ad

Ve Vd V∞

Figure 4.1. Illustration of the double multiple streamtube model of Paraschivoiu with-
out flow expansion, where the streamtubes are illustrated by the horizontal lines and
the vertical line in the center illustrates the transition from the upstream to the down-
stream part. Flow velocities are given for the upstream part. It is assumed that the
height of the streamtube Δz is included in the streamtube area A d .

Templin [43], used one streamtube for the entire turbine. The disadvantage
of this method is that the flow velocities at the blade positions are constant
throughout the entire turbine. A vertical axis turbine will interact with the
flow two times, one upstream and one downstream. To overcome the problem
with equal flow velocity when the blade is up- and downstream, Lapin [44] in-
troduced the double streamtube model, consisting of one streamtube upstream
and another downstream, allowing for lower flow velocities downstream. For
a real turbine, the velocity does not only vary between the upstream and down-
stream part of the turbine, but could also vary over the entire swept area of the
turbine. To account for this, Strickland introduced the idea of using multiple
streamtubes [45]. In this case, the system has to be solved for each individual
streamtube.
In Strickland’s implementation, the double streamtube model of Lapin was
not included. Later, both these two ideas were combined into the double
multiple streamtube model. Two different versions of this model were de-
veloped around the same time, one by Paraschivoiu [46] and one by Read and
Sharpe [47]. The difference between the models is that the implementation
by Paraschivoiu assumes straight streamlines including only a velocity in the
x-direction (i.e. the direction of the asymptotic flow), while the model by Read
and Sharpe assumes that the streamlines expand linearly through the turbine.
This gives the difference that in the model by Paraschivoiu, all streamtubes are
independent of each other, while in the model of Read and Sharpe, they are
coupled. Both models have shown quite good agreement with experiments for
the curved blade Darrieus turbine [3, 48].

38
4.1.1 Description of model
The double multiple streamtube model implementation used here is based on
the implementation of Paraschivoiu [3] and an illustration of this model is
given in figure 4.1. In the double multiple streamtube model, the turbine is
separated into two discs, one upstream and one downstream. In the middle of
the turbine, the pressure is assumed to be the same as the asymptotic pressure.
With this assumption, the velocity at the upstream disc will be the average of
the velocity at the middle and the asymptotic velocity, similar to equation (2.7)
in the Betz theory. In the same way, the velocity at the downstream disc be-
comes equal to the average of the velocity at the middle, and the velocity far
behind the turbine. In the implementation by Paraschivoiu, the upstream disc
is solved first, independently of the downstream disc. The downstream disc
uses the velocity in the middle of the turbine, calculated when solving the
upstream disc, as input. Except from this calculation of velocity, both discs
are solved in the same way, and most equations will only be presented for the
upstream part of the system.
This description will follow the same notation as Paraschivoiu for the direc-
tion of the flow velocity. Hence, the flow direction illustrated in figure 4.1 is
considered positive direction. The definitions of the blade angle θ b and the an-
gle of attack α will still remain the same as in section 2.3. Therefore, negative
angles of attack will be obtained when the blades are upstream, while the de-
scription by Paraschivoiu [3, Chapter 6] has positive angle of attack upstream.
The difference in the definition of the flow direction is the reason why the
expressions for flow velocity and angle of attack are different in this section,
compared to section 2.3.
The main principle behind streamtube models is the use of momentum con-
servation in each streamtube. Similar to the Betz derivation, momentum con-
servation gives
ρ AeiVei2 − ρ A∞iV∞i
2
= ρ AdiVdi (Vei −V∞i ) = Fxi , (4.1)
where A∞i , Adi and Aei are the cross-sectional areas of a streamtube and i
denotes the streamtube index (cf. figure 2.1). The difference, compared to
the Betz theory, is that in this model, the force Fxi is calculated from blade
section data. If the streamtubes are discretized to give each streamtube the
same angular distance Δθ and height Δz, the size of each streamtube becomes
Adi = Ri ΔzΔθ |cos θbi | . (4.2)
To calculate the force, the first step is to obtain the lift and drag coefficients,
which depend on the angle of attack and the Reynolds number. In the model
of Paraschivoiu, it is assumed that the direction of the flow does not change.
This gives the relative velocity as

 2
Ri Ω
Vri = Vdi − sin θbi + cos2 θbi cos2 ηi (4.3)
Vdi

39
and the angle of attack as
 
cos θbi cos ηi
αi = ϕi + δi = arctan + δi , (4.4)
sin θbi − RVi Ω
di

where η is the angle of the blade relative to the vertical axis.


When the coefficients have been determined from the empirical data, the
lift and drag forces are obtained from
1
FLi = CLi ρ ci liVri2 , (4.5)
2
1
FDi = CDi ρ ci liVri2 , (4.6)
2
where ci is the chord and li is the blade length in the streamtube. Generally,
it is the normal and tangential forces that are of interest. These forces can be
determined from the corresponding normal and tangential force coefficients
CNi = CLi cos ϕi +CDi sin ϕi , (4.7)
CTi = CLi sin ϕi −CDi cos ϕi (4.8)
and therefore, the forces are
1
FNi = CNi ρ ci liVri2 , (4.9)
2
1
FTi = CTi ρ ci liVri2 . (4.10)
2
Note that by this definition, the normal force is perpendicular to the blade, and
the force in the radial direction is
FRi = FNi cos ηi . (4.11)
To calculate how the velocity decreases in the streamtube, the forces in the
x-direction (parallel to the flow) are of interest. These are obtained as
Fxi = FNi cos θbi cos ηi − FTi sin θbi . (4.12)
Considering that there are Nb blades on the turbine, and that a streamtube only
have a blade inside it a fraction of the time, the mean force is given by
Nb Δθ
Fxi = (FNi cos θbi cos ηi − FTi sin θbi ) . (4.13)

If the length of the blade in a streamtube l i = Δz/ cos ηi is inserted into equa-
tions (4.9) and (4.10), equation (4.13) becomes
 
Nb ρ ciVri2 Δθ Δz sin θbi
Fxi = CNi cos θbi −CTi . (4.14)
4π cos ηi

40
cs
Router
Δz
ηs Streamtube i

Vei Vs Vdi
rs
Δrs

Rinner
θ
Ω
θb

Figure 4.2. Illustration of the parameters used for the strut model.

Combining this with equation (4.1) and (4.2) gives


 
Nb ρ ciVri2 Δθ Δz sin θbi
ρ Ri ΔzΔθ |cos θbi |Vdi (Vei −V∞i ) = CNi cos θbi −CTi
4π cos ηi
(4.15)
which with the substitutions
Vdi = uiV∞i (4.16)
and
V∞i +Vei
Vdi = (4.17)
2
gives
 
1 Nb ciVri2 sin θbi
1− = CNi cos θbi −CTi . (4.18)
ui 8π Ri |cos θbi |Vdi2 cos ηi

This is the equation to be solved to obtain the interference factors u i (defined


in equation (4.16)). Since Vr , CN and CT depend on u, this equation has to be
iterated. It should be noted that from equations (4.16) and (4.17), the velocity
in the center is
Vei = (2ui − 1)V∞i . (4.19)

4.1.2 Including struts


The above theory only includes the blades, but for an H-rotor, it is necessary
to attach the blades to the main shaft with struts. These struts interact with
the flow, affecting turbine performance. One strut model was implemented by
Moran [49], who derived a correction coefficient for the drag. This is a fast
solution suitable for horizontal struts, but for struts with an angle, a lift force

41
will also be generated. To increase the flexibility of the code, it was chosen to
calculate the forces directly from the lift and drag coefficients instead.
To calculate the forces, the relative velocity has to be known. Let r s be a
position along the strut
Rinner < rs < Router , (4.20)
where the strut starts at Rinner and ends at R outer , see figure 4.2. The struts
are discretized into Δr long segments in the radial direction. In the streamtube
model, the velocity is only known at the turbine disc and at the center of the
turbine. The simple approximation that the velocity varies linearly between
these positions gives the velocity Vs j of strut segment j as
Vdi −Vei
Vs j = Vei +  rs j cos θb . (4.21)
R2i − rs2j sin2 θb

The position of the blade, to which the strut is attached, is given as θ b . This
expression involves the velocity in streamtube i, in which the segment is lo-
cated. To determine which streamtube this is, the corresponding angle at the
circumference of the turbine is given by
 
rs j
θi = arcsin sin θb . (4.22)
Rj
In the vertical direction, the position of segment j will be equal to the vertical
position of the corresponding streamtube i. The relative velocity can be calcu-
lated basically in the same way as in equation (4.3), but since the discretization
is performed in the radial direction for the struts, the angle is to be given rel-
ative this direction as well (the reason for not using vertical discretization for
struts is that Fxi in equation (4.14) would diverge for horizontal struts). With
the strut angle given as η s , the velocity is given by
 2
rs j Ω
Vrs j = Vs j − sin θb + cos2 θb sin2 ηs j (4.23)
Vs j
and the angle attack α s for the strut is
cos θb sin ηs j
αs j = arctan rs j Ω
, (4.24)
sin θb − Vs j

where zero pitch angle is assumed for the strut. By defining the chord of the
strut as cs and following the same derivation as with equations (4.7) – (4.14),
the expression for the mean force in the flow direction for a section (Δr s , Δθ )
of the strut is obtained as
 
Nb cs j ρ Vrs2 j sin θb
Fxs j = Δθ Δrs CNs j cos θb tan ηs j −CT s j , (4.25)
4π cos ηs j

42
where CNs j and CT s j are the normal and tangential force coefficients of the
strut, which are defined in a similar way as for the blades

CNs j = CLs j cos αs j +CDs j sin αs j , (4.26)


CT s j = CLs j sin αs j −CDs j cos αs j . (4.27)

Here, CLs j and CDs j are the lift and drag coefficients for the struts. The differ-
ences in equations (4.24) and (4.25), compared to equations (4.4) and (4.14),
originate from the different definition of ηs j , and that the discretization is done
in the radial direction.
The interference factor ui can be calculated according to
 
1 1
1− = Fxi + ∑ Fxs j (4.28)
ui 2Ri |cos θb |Vdi2 ρ Δθ Δz j∈i

where ∑ j∈i Fxs j is the sum of the forces from all points (r si , θb ) that corre-
sponds to streamtube i. Similar expressions can be derived for the downwind
part of the turbine. The torque from the struts can be calculated in a similar
way as for the blades in a curved blade Darrieus turbine
FT s j (θb )
Ts (θb ) = ∑ rs j , (4.29)
j cos η j

where
1
FT s j (θb ) = CT s j (θb ) ρ cs j ΔrsVrs j (θb )2 .
2

4.1.3 Obtaining lift and drag coefficients


In the streamtube model, it is necessary to have the lift and drag coefficients
for a given angle of attack. Experimental data can be used to obtain these if
the Reynolds number and angle of attack are known. For several symmetrical
NACA profiles, such data can be obtained from [50], which has been used in
this work. The problem with this kind of data is that it usually is valid for very
long blades and a static angle of attack. For a vertical axis turbine, the blades
will change their angles of attack during the revolution.
When the angle of attack for a blade increases above the stall angle, the flow
will separate from the blade surface, which reduces the lift force and increases
the drag. The flow separation takes time to develop and if the blade rapidly
increases its angle of attack, the lift force can be higher than in the static case
for a short period of time. During this time, a vortex starts to form at the
surface of the blade and when this vortex detaches, the lift force is greatly
reduced. This phenomenon is called dynamic stall. When the turbine has a
low tip speed ratio, the angles of attack become high enough for the blades to
experience this in each revolution. This occurs for tip speed ratios around 4.

43
To model dynamic stall, the Gormont model has been used. This model was
originally developed by Gormont [51], later modified by Massé [52] and ad-
justed by Berg [53]. The advantages of this model is that it only requires data
for lift and drag coefficients, flow velocity, blade thickness to chord ratio, an-
gle of attack and rate of change of the angle of attack, making it easy to apply
to any blade, which is the reason why this model was chosen. Other mod-
els have been shown to give better results [3]. One example is the Beddoes-
Leishman model [54,55], but it has more parameters that have to be calibrated,
which makes it harder to apply the model to an arbitrary airfoil.

Tip effects
For a blade to obtain a lift force, there has to be circulation around the blade
(cf. Kutta Joukowski
 lift formula,equation
 (4.57)).
 Since the vorticity is di-
 
vergence free ω = ∇ × V ⇒ ∇ · ∇ × V = 0 , there are vortices generated
from the blade tips with the same circulation as around the blade. These vor-
tices are the source of induced drag, and the corresponding losses. It should be
noted that the curved blade Darrieus turbine does not have any distinct blade
tips (the blades are attached to the main axis). Here, the tip vortices have
to leave the blades more evenly distributed over the length of the blade, and
in the model of Paraschivoiu, no tip losses are applied to curved blade Dar-
rieus turbines. For straight blade turbines, one correction model, derived from
Prandtl’s theory for screw propellers, used by e.g. Sharpe [48], uses a velocity
correction factor F, which for the upstream disc is
 πa 
− i
arccos e si
Fi =  , (4.30)
πh
− 2s
arccos e i

where
πVei
si = , (4.31)
Nb Ω
h
ai = − |zai | , (4.32)
2
zai is the altitude (with zero defined in the center of the turbine) and h is the
turbine height. A corresponding expression exists for the downstream disc.
The new expressions for velocity and angle of attack are

 2
Ri Ω
Vri = Vdi − sin θbi + Fi2 cos2 θbi cos2 ηi , (4.33)
Vdi
Fi cos θbi cos ηi
αi = arctan + δi . (4.34)
sin θbi − RVidiΩ

44
In addition, Paraschivoiu [3] suggests using finite wing theory as well to cal-
culate induced drag and reduced angle of attack. This is given by
CL∞i
CLi = , (4.35)
1 − πaAR
0i

 i

0.8tbi
a0i = 1.8π 1 + , (4.36)
ci
C2
CDi = CD∞i + Li , (4.37)
π ARi
CLi
αbi = αi − , (4.38)
π ARi
where AR is the aspect ratio, tb the blade thickness, CL∞ and CD∞ lift and
drag coefficients for an infinitely long blade and α b the corrected angle of
attack. Paraschivoiu suggests that equations (4.35) – (4.38) should not be ap-
plied to dynamic stall calculations, but since this would lead to a discontinuity
in the limit when dynamic stall starts to have an effect, the current implemen-
tation uses equations (4.35), (4.36) and (4.38), then calculates the dynamic
stall corrections to the lift coefficient, and then calculates the induced drag
using equation (4.37) with the new lift coefficient. It should be noted that
equations (4.35) – (4.38) are derived using the assumption that the angle of at-
tack is constant, the tip vortices extend along a straight line behind the blade,
and the wing is elliptical, which is not the case for a vertical axis turbine. This
has to be taken into account when evaluating the accuracy of the model for
straight blade turbines.

4.1.4 Corrections due to flow curvature


As was illustrated in section 2.4, there are effects due to flow curvature that
should be taken into account. With the definition of the velocity as in the
streamtube model (V → −V ), the assumption of straight stream lines (β = 0),
using that the reference velocity Vre f is the relative velocity Vr and using the
velocity at the turbine disc Vd instead of Vabs , equation (2.52) gives
Vd cos θb Ωx0r c Ωc
α = δ + arctan − − . (4.39)
ΩR −Vd sin θb Vr 4Vr
This is included in the adaptation of Sharp [48], but with the assumptions that
the curvature force only acts in the normal direction and that the mounting
point is in the center of the blade (x 0r = 0). With the assumption
CN = as α , (4.40)
the correction to the normal coefficient becomes
as Ωc
CN = CN0 − , (4.41)
4Vr

45
ay∞ aye

aze
az∞
bze

bz∞

by∞
z
1
bye
y
Figure 4.3. Illustration of the cross-section for a streamtube at the three positions: Far
ahead of turbine (black), at the upstream disc (white) and in the center (hatched).

where CN0 is the original normal force without corrections.

4.1.5 Including flow expansion


In contrast with the model of Read and Sharpe [47], the model of Paraschi-
voiu does not include streamtube expansion in the basic implementation. One
method for implementing the expansion was used by Paraschivoiu [56] where
the streamtubes were allowed to expand, but the direction of the velocity was
not changed. In this implementation, similar to the one by Read and Sharpe,
it was assumed that streamtube expansion only occurs in the horizontal plane,
and that the streamline that intersects the turbine center is unchanged. In this
work, the assumption is instead that the flow can expand both in the vertical
and the horizontal directions. It is assumed that the direction of the expansion
depends on the ratio between the width and height of the turbine. Assume that
the streamtube size at the turbine disc is unity A d = 1, at infinity A∞ = by∞ bz∞
and at the center Ae = bye bze (see figure 4.3). By defining

ay∞ = 1 − by∞ and aye = 1 + bye (4.42)

46
and the corresponding expressions in the z direction, the assumption about the
ratio between the horizontal and vertical expansion is
D D
ay∞ = az∞ and aye = aze . (4.43)
h h
Combining equations (4.16), (4.19), (4.42) and (4.43) gives the streamtube
size

 
1 h h 1 2 h
by∞ = − + + − (1 − u), (4.44)
2 2D 2D 2 D
   
1 h h 1 2 h u
bye = − + + − 1− . (4.45)
2 2D 2D 2 D 2u − 1

With these known, the continuity equation gives the expressions in the z-
direction as
h
bzq = 1 − (1 − byq ) , (4.46)
D
where the index q can be either e or ∞. The size of the streamtube is obtained
as

Δyq = bzq Δyd and Δzq = bzq Δzd . (4.47)

Using the assumption that the center streamtube is a straight line, the position
of the streamtube at infinity and in the middle is given by the integral
yd
yq = byq dyd , (4.48)
0

which is evaluated as a sum over the streamtubes. It is assumed that the stream-
tubes located at the same vertical or horizontal position at the turbine disc, also
are located vertically or horizontally to each other at infinity and in the center
as well.
For the downstream disc, the equations are similar, with the difference that
the middle is the starting position. To connect the values between the two
discs, the vertical and horizontal positions for each streamtube at the middle
are calculated for both discs. The starting values for the downstream disc
are obtained through interpolation from the values calculated by the upstream
part. In this way, there are different sets of streamtubes upstream and down-
stream. This has the advantage that the points where the streamtubes cross
the downstream disc are fixed to the same positions as without the expansion
model (the other solution would have been to connect the streamtubes through
the whole turbine, which for two-dimensional expansion would increase the
complexity of the model significantly).

47
4.2 Vortex models
To describe the motion of a fluid, the continuity equation and Navier-Stokes
equations are used. The continuity equation
∂ρ  
+ ∇ · ρV = 0 (4.49)
∂t
represents the conservation of mass. Due to the low flow velocities present for
wind and marine current turbines, the flow can be considered incompressible.
In this case, the continuity equation reduces to

∇ · V = 0. (4.50)

Navier-Stokes equations are derived from Newton’s second law and can in the
incompressible case be written as

∂ V   ∇p
+ V · ∇ V = − +g + ν ∇2V , (4.51)
∂t ρ
where g is the acceleration of gravity and ν is the kinematic viscosity. These
are non-linear partial differential equations, which only have closed form solu-
tions in rare cases. Numerical solutions can be obtained with traditional finite
element/volume methods, where these equations are solved on a grid with the
velocity as the unknown variable. An alternative method is to use the vortex
formulation, where the vorticity

 = ∇ × V
ω (4.52)

is used instead. The vorticity equation is obtained from the rotation of equation
(4.51) according to
 
∂ω 
+ V · ∇ ω  · ∇) V + ν ∇2 ω
 = (ω . (4.53)
∂t
The left hand side represents the convection of vorticity. The first term on
the right hand side is the vortex stretching, and the second is the diffusion of
vorticity. These equations show that the vorticity moves with the flow velocity.
In the two-dimensional case, the vorticity is directed along the z-axis, while
the flow occurs in the x - y plane. Due to orthogonality, vortex stretching does
not apply in the two-dimensional case, and with ω  = ω ẑ,

∂ ω  
+ V · ∇ ω = ν ∇2 ω . (4.54)
∂t
For vortex methods, it is more common to use a Lagrangian description
rather than an Eulerian (which is used in equations (4.51) – (4.54)) [57, chap-
ter 1]. In the Lagrangian description, the vorticity is discretized into vortices,

48
which move with the flow. The use of these vortices removes the need for a
mesh (compared to e.g. the finite element method). The vortices are deter-
mined by their positions ri and their circulations Γi . If diffusion is included,
equation (4.54) is solved in two steps, one convection step, which in the La-
grangian description becomes
dri 
= Vi , (4.55)
dt
and one diffusion step, which could either change the position of the vor-
tex (like random walk [58]) or the circulation (like the vortex redistribution
method [59]). In this work, it is assumed that convection is much more im-
portant than diffusion and when solving the time evaluation of the vortices,
diffusion is ignored. This is expected to be a good approximation in the tur-
bine neighborhood outside the boundary layer of the blades. The divergence
of the vorticity is  
∇·ω  = ∇ · ∇ × V = 0, (4.56)
i.e. the vorticity is divergence free. For two-dimensional flow, this implies that
vortices are always created in pairs (or possibly higher order pairs). Kelvin’s
theorem [57, chapter 2] states that without viscosity, the circulation inside the
fluid is conserved for a vortex, and the implication is that vorticity can only be
created at the boundaries.

4.2.1 Implementing the turbine


With the vortex model, it is theoretically possible to calculate the lift and drag
forces for an airfoil. However, this requires a full solution of Navier-Stokes
equations, which is very time-consuming, and adds a large amount of vortices
to the flow at each time step. As a simplification, the same procedure as for the
streamtube models can be used, where experimental data and a dynamic stall
model are used to calculate lift and drag forces. In this implementation, the
experimental data and the dynamic stall model are the same for the streamtube
model and the vortex model. For the 3D corrections, the correction factor F in
equation (4.30) cannot be applied because the vortex model is a 2D model, but
equations (4.35) – (4.38) can be applied in the same way as for the streamtube
model, described in section 4.1.3.
The calculation of the angle of attack can either be performed by calculat-
ing the velocity at the turbine position and applying the corrections derived in
section 2.4, or by solving the potential flow problem. In this implementation,
the potential flow solution is chosen, because it can handle a varying velocity
field and flow curvature. Linear panels are used according to section 4.2.3 to
calculate the potential flow solution. These equations are not fully determined
without knowledge of the total circulation around the blade. As it is this cir-
culation that is of interest for the lift force, closure is obtained by applying the

49
Kutta condition, which states that the velocity at the trailing edge of an airfoil
should be finite [27, chapter 3]. The potential flow solution gives an infinite
velocity at an edge, unless it is a stagnation point, which therefore is enforced
with the Kutta condition. For a panel method, this can be achieved by forcing
the vorticity of the panels to be zero at the trailing edge. In the implementa-
tion, the vorticity at the end of the panel on the upper side of the blade is set to
the same vorticity as the panel on the lower side. With this condition and the
no-penetration boundary condition, the circulation can be calculated. To get
the angle of attack from the circulation, the method is calibrated by calculat-
ing the circulation for a set of stationary airfoils with known angles of attack.
These values are used to create an inverse function, which gives the angle of
attack from the circulation.
By using the conservation of vorticity, one vortex has to be released be-
hind the trailing edge every time-step with a circulation corresponding to the
change in circulation for the blade. Since experimental data and a dynamic
stall model are applied to get the lift and drag forces, the effective circulation
will not be the one calculated from the Kutta condition if the blade is stalled.
To get the circulation, the same method as in Strickland [60] is used, where
the Kutta-Joukowski lift formula
1
Γ = CL cV (4.57)
2
gives the circulation from the lift coefficient. With the circulation known, the
strength of the released vortex can be calculated. By adding an additional vor-
tex to the flow, the angle of attack changes slightly. As calculating the circula-
tion from empirical data combined with the dynamic stall model is non-linear,
the procedure to calculate the angle of attack with the Kutta condition and
recalculating the strength of the released vortex can be iterated. Considering
that equation (4.57) is derived from potential theory, this approximation is best
suited when the drag is low.
For a turbine, it is often the case that the turbine blades are quite small com-
pared to the distance between them. The first simplification with this approxi-
mation is that the circulation of each blade can be solved separately. Another
approximation is to assume that each blade can be represented with a single
point circulation during the vortex convection. With this approximation, there
is no need to calculate the velocity due to any panels on the blade surface. Be-
cause it is more expensive to calculate the velocity of a panel, approximating
the blade with a single point can significantly improve computational speed
when calculating the vortex velocities. Further, as the Kutta condition is not
satisfied as the circulation is modified by the empirical data, the flow velocity
at the trailing edge would be infinite if the panels were to be used, which is
unphysical. The approximation with a single circulation does not suffer from
this problem if a Gaussian vortex is used according to equation (4.64). Within
this approximation, a vortex can theoretically end up inside a blade when us-

50
ing the Kutta condition. If this is the case, that vortex is ignored during this
time step.

4.2.2 Merging vortices


The vortex method creates one vortex for each blade during each time step.
For long simulations, this will cause a steady increase in computational time
for each time step. However, some of the vortices that have propagated to
positions far behind the turbine can be merged without loosing any significant
accuracy.
To properly merge two vortices, there are some conservation laws that
should be satisfied. First, the total circulation should be conserved

Γ1 + Γ2 = const. (4.58)

Secondly, the center of vorticity should be conserved as well

x1 Γ1 + x2 Γ2 = const,
y1 Γ1 + y2 Γ2 = const. (4.59)

By satisfying these equations, a new vortex can be created, which is a rea-


sonable representation of the two old ones. This procedure can be performed
for all vortices, which are closer to each other than a cutoff radius Δ, where
the value of Δ is increased with increasing distance behind the turbine. The
merging is only performed on vortices with the same sign for the circulation,
where the merged vortex actually will be located somewhere in between the
old ones.

4.2.3 Calculation of velocity


Since the vortex method propagates the vortices with the flow according to
equation (4.55), the velocity needs to be calculated. The velocity field is tra-
ditionally divided into two parts (or more)
V = ∇φ + Vω , (4.60)

where the first part is the potential flow solution for the system (where ∇× V =
0), and the second part is the contribution from all vortices.

Potential flow solution


In the most simple case without any boundaries, the potential flow solution
will be the asymptotic velocity V∞ . Boundaries can be included, either by using
conformal mappings, as used by e.g. Deglaire [15], or by using panel methods.
The panel method implemented is described in [61] and uses linear panels

51
with linearly varying vortex strengths. In this work, panels have been used to
implement the blades of the turbine in the calculation of the angle of attack,
see section 4.2.1, and to implement channel walls in paper V.
When using panels, one control point in the middle of each panel is chosen.
This is the point where the boundary condition is satisfied, which means that
the boundary condition is only satisfied at N control points, where N is the
number of panels. To solve the panel equations, the velocity contribution from
each panel at each control point is used to create an N × N equation system. It
is usually desired to have a condition on the total circulation of all panels as
well (total source strength for source panels). Adding this condition leads to an
over-determined equation system. This system can be solved approximately
with the least square method, which gives a unique approximation. It should
therefore be noted that this method does not give an exact solution.
 3  The matrix
system to be solved is of size N × N, which would take O N operations to
solve, but if the relative panel
 positions are constant, L-U decomposition can
be used, which requires O N 3 operations, 2 but only has to be performed once
for each geometry. Thereafter, only O N operations are required for each
time step.
In paper V, panels have been used to create channel walls for some of the
simulations. Here, the no penetration boundary condition (zero normal veloc-
ity) is used for the side walls, while for the outflow, zero tangential velocity
is chosen. For the inflow, the normal velocity is fixed to the asymptotic ve-
locity, while the tangential velocity is set to zero. This is accomplished by
using both source and vortex strength for the inflow panels (the other panels
had only vortex strength). For channels, both the total circulation and the to-
tal source strength of the panels that belong to the channel are set to zero. In
paper IV, the channel walls have been included using the analytical expression
in equation (4.66), instead of using panels.

Vortex contribution
According to the Biot-Savart Law, the velocity at position r due to a point
vortex atri is

Vω = 1 Γ p × (r −ri ) , (4.61)
4π |r −ri |3
which in the two-dimensional case (note that a point in 2D is a line in 3D)
with complex numbers simplifies to
iΓ 1
Vω = . (4.62)
2π (r − ri )
Here, r is a complex number representing the position in the x- and y-direction
according to r = x + iy and r denotes the complex conjugate of the position.
This assumes point vortices, which have the vorticity
ω = Γδ (2) (r − ri) , (4.63)

52
where δ (2) is the two-dimensional Dirac delta function. Numerically, equa-
tion (4.62) diverges when the distance approaches zero. To avoid these un-
physical divergences in the velocity, a mollifier is usually applied to the vor-
tices. Here, a Gaussian kernel has been applied, giving the vorticity
2
Γ − |r−r2i |
ω= e ε (4.64)
πε 2
which, for a given number of vortices with strengths Γ i at positions ri gives
the velocity  
Nv
iΓi 1 |r−r |2
− 2i
Vω (r) = ∑ 1−e ε . (4.65)
i=1 2π (r − ri )

The vortex contribution in a channel


If a wall is included in the simulation domain at Im(r) = 0 and the no pene-
tration boundary condition applies at this wall, the method of images can be
used to obtain the flow velocity from a vortex at position r j by adding an ad-
ditional vortex at the mirror position r j with opposite strength. For a channel
with an additional wall at Im(r) = H, the solution will consist of an infinite
series of mirror vortices. It has been shown in [62] that this series converges
to the analytical expression
Nv
iΓ j   π   π 
Vω (r) = ∑ coth (r − r j ) − coth (r − r j ) . (4.66)
i=1 2π 2H 2H

This equation requires that the channel height is constant. This method is used
in paper IV. Otherwise, the linear panel method can be used to create channels
of arbitrary shapes [61] and this method is used in paper V.

4.2.4 Numerical evaluation of the velocity field


At every time-step, the velocities at each vortex position, and at each control
point for the panels, have to be calculated. This can be accomplished by di-
rect evaluation of equation
  2  Nv vortices and N p panels, this
4.61. Assuming
2
would require O Nv + O (Nv N p ) + O N p operations. The evaluation of the
velocity of a panel is much more expensive than the velocity of a point vortex.
This is the reason why a speedup is obtained for the approximation that the
blades are considered as single circulations during the convection phase of the
vortices, although the infinite velocity at the trailing edge also warrants this
approximation, see section 4.2.1. The velocity contributions from the panels
to the vortices are therefore only included in paper V, where panels are used
to model a channel.
To efficiently evaluate the velocity, the fast multipole method [63] can be
used. The current implementation is based on the one described in [23], but

53
with multipole expansions of the linear panels included similar to the im-
plementation in [61], which allows the total velocity to be evaluated with
O (Nv + N p ) operations instead. In the case of a channel (paper IV), a method
for implementing equation (4.66) using the fast multipole method is given
in [62], which has been implemented in combination with the models from [23,
61].

The fast multipole method


The fast multipole method is a method for rapid evaluation of pairwise interac-
tions between particles, originally developed by Greengard and Rhoklin [63].
This method works by dividing the domain, where the interaction particles are
located, into smaller and smaller boxes. The implementation used here splits
each box into four smaller boxes for each level and the number of levels is
decided from the amount of particles in the domain.
For each box, a multipole expansion can be formed around the center of the
box r0 according to [63] as

ak
V (r) = a0 log (r − r0) + ∑ k
. (4.67)
k=1 (r − r0 )

Expanding the harmonic potential


1
V (r) = q j , (4.68)
ri − r
where
iΓi
qi = , (4.69)

results in a0 = 0. The coefficients a k are given by
Nd
ak = − ∑ q j (r j − r0 )k−1 (4.70)
j=1

where Nd is the number of particles in the box and r 0 is the box center. This ex-
pansion can be used to calculate the velocity from the particles inside the box
for any point located with sufficient distance from the box. Equation (4.67)
is truncated to include coefficients up to a p . The required distance for con-
vergence of the truncated version of equation (4.67) depends on the number
of multipole coefficients p in the expansion and the required tolerance of the
algorithm.
Instead of expanding the particles of box m around the center of box m, the
expansion can instead be performed around the center of another box n, under
the condition that the distances from the center of box n to the points in box m
are large enough. This local expansion is

V (r) = ∑ bl (r − r0)l , (4.71)
l=0

54
Algorithm 4.1 Multipole to multipole translation
for k = p downto 2 do
for j = k to p do
a j := a j + (r0m − r0n ) · a j−1
end for
end for

Algorithm 4.2 Local to local translation


for k = 0 to p do
for j = p − k to p − 1 do
b j := b j − (r0m − r0n ) · b j+1
end for
end for

where
Nd
qj
bl = ∑ (r l+1
. (4.72)
j=1 j − r0 )

Similar to equation (4.67), equation (4.71) will also be truncated to include


coefficients up to b p . The main difference between multipole and local ex-
pansions is that a multipole expansion for box m is valid in all points with
sufficient distance to box m, while a local expansion only is valid inside the
box itself. Instead, all points with sufficient distance to box n can form a local
expansion inside box n. The fast multipole method combines these two prop-
erties using that it is possible to convert the multipole expansion of box m to a
local expansion in box n according to the expression (assuming a 0 = 0)
∞  
1 ak l +k−1
l ∑
bl = k
(−1)k (4.73)
(r0m − r0n ) k=1 (r0m − r0n ) k − 1

under the assumption that the distance between the two boxes is large enough
(well separated boxes). The fast multipole method operates on all levels in the
multipole tree. To convert multipole and local expansions from one level to
another, the algorithm uses that the center of both multipole and local expan-
sions can be shifted according to algorithms 4.1 and 4.2.
An illustration of the fast multipole method is given in figure 4.4, where the
boxes have been divided according to the geometric mean.
The first step consists of using equation (4.70) to create multipole expan-
sions in each box on the lowest level as illustrated for one box in figure 4.4 (a).
The second step consists of shifting all multipole expansions of the lower
level to the level above according to algorithm 4.1. This is illustrated in fig-
ure 4.4 (b), where the multipole expansions of four boxes are combined into
one. The quadrant marked 1 illustrates the first shift, the quadrant marked 2 the

55
1 3

2 3
(a) Multipole initialization (b) Multipole to multipole
shift

(c) Multipole to local shift (d) Multipole to local shift

1 2

1 3
(e) local to local shift (f) Multipole evaluation
Figure 4.4. Illustration of the fast multipole method.

56
second shift and the two quadrants marked 3 the third and final shift to the top
level. After this shift, all boxes on all levels have their multipole expansions.
The third step consists of converting multipole expansions to local expan-
sions according to equation (4.73), which is illustrated in figure 4.4 (c) and
(d). This operation is performed for all boxes, but for simplicity, the operation
is only illustrated for one box in figure 4.4. At each level, the neighboring
boxes are divided between those in the near field and those in the far field.
The boxes in the far field can be evaluated through a local expansion in the
box. The highest level only contains one box and in the homogeneous case in
figure 4.4, the second level with only four boxes also has all boxes in the near
field. As no multipole to local shifts are performed at these two levels, they are
omitted from figure 4.4. At the third level, the boxes marked with grey in (c)
are in the far field, hence local expansions of these boxes can be created in the
selected box (the arrows in the figure represent these shifts). The white boxes
represent the near field, which have to be handled at a lower level. The fourth
level is illustrated in figure 4.4 (d). Here, those boxes that were shifted at the
previous level have been marked with dark grey as they have already com-
pleted their interaction with the illustrated box. The other grey boxes interact
with the current box through equation (4.73).
As the interactions with equation (4.73) occur at different levels, the local
expansions of the higher levels have to be shifted down to the lower levels,
which is performed according to algorithm 4.2. This step is similar to the
multipole to multipole shift, but here, the shift starts from the top level and
moves downward, as illustrated by the numbers in figure 4.4 (e). It can be
noted that since there are no multipole to local shifts on the top two levels, both
the local to local shift and the multipole to multipole shifts of the two highest
levels could have been omitted, but were kept here for illustrative purposes.
When all local expansions have been formed at the lowest level, the veloc-
ity can be evaluated at each evaluation point in the box according to equa-
tion (4.71), giving the velocity contribution of all boxes in the far field. What
remains is to calculate the contribution of the boxes still remaining in the near
field at the lowest level, which is performed for each evaluation point in the
box using direct interaction with all particles in all boxes in the near field.
To obtain maximum speed for the method, a suitable number of particles
in each box Nd at the lowest level needs to be determined. The computational
time of the direct evaluation is

Nbox
td = kd1 ∑ ∑ Ndi Nd j , (4.74)
i=1 j∈NFi

where NFi is the set of all boxes in the near field of box i, N di and Nd j are the
number of particles in boxes i and j respectively, N box is the number of boxes
and kd1 is a constant. If the number of particles per box is approximately

57
constant, the expression simplifies to
td = kd2 NNd , (4.75)
where N is the total number of particles, Nd the number of particles per box
and kd2 another constant (k d2 ≈ 9kd1 for the system i figure 4.4). If Nd is too
high, an extra level can be added, reducing the N d by four. The cost is that
the number of multipole to local shifts for an additional level is approximately
four times higher than for the previous level. Hence, there will be an optimal
Nd that minimizes the total computational time.

The adaptive fast multipole method


The homogeneous implementation of the fast multipole method, as illustrated
in figure 4.4, works well for uniform particle distributions, where Nd is ap-
proximately constant for all boxes. For less uniform particle distributions, N d
will be different for different boxes as seen in figure 4.5 (a). If the particle den-
sity is very high in a neighborhood, these boxes contain many particles, which
increases the computational time according to equation (4.74). To account for
this problem, an adaptive method is necessary.
There are two common ways of implementing an adaptive fast multipole
method. One method is based on using different number of levels at dif-
ferent positions in the multipole tree (midpoint adaptivity), illustrated in fig-
ure 4.5 (b) for two different particle distributions. This adaptivity splits each
box that contains more than a given number of particles into four sub-boxes,
hence preventing too many boxes in the near field. The advantage of this
method is that all boxes still have a square shape, which is beneficial for the
convergence of the multipole and local expansions, hence keeping the number
of boxes in the near field relatively low. Fixed box sizes can also be beneficial
for the shift operations [64], as there are simplifications that can be made if the
distances of two shifts are the same. Disadvantages are that the N d will vary
between the boxes as each split divides in four parts, and that the implemen-
tations of the shift operators becomes more complicated as interactions occur
between different levels.
The second adaptive method, which is the one implemented in paper IX,
splits the boxes around the median point instead of the geometric midpoint.
This is called asymmetric adaptivity and is illustrated in figure 4.5 (c). This
adaptivity causes all boxes to have approximately equal N d , which can be
beneficial when parallelizing the algorithm. While fixed rules can be applied to
determine the near field for the midpoint adaptivity, the asymmetric adaptivity
is based on the θ - criterion, which states

Criterion 1. Let r0i and r0 j be centers of boxes containing the sets  of elements

Si andS j respectively. Define r i ≡ max S i − r 0i and r j ≡ max S j − r0 j . Let

db ≡ S j − r0 j , Rb ≡ max(ri , r j ) and rb ≡ min (ri , r j ). Then the two boxes
are well separated if R b + θ rb ≤ θ db , θ ∈ (0, 1) .

58
(a) Homogenous FMM

(b) Midpoint adaptivity

(c) Assymetric adaptivity

Figure 4.5. Comparison between different adaptivity schemes for a normal distribu-
tion and a spiral. Case (a) and case (c) have the same number of boxes (64), while
case (b) has slightly more (88 and 115) in this particular case.

59
Figure 4.6. Illustration of the interactions for asymmetric adaptivity for two boxes.
The black boxes represent the interacting boxes (one interacting box in each figure),
the white boxes represent near-field interactions according to the θ - criterion, grey
represent the far field interactions at current level and dark grey represent the far field
interactions at previous levels.

The value of θ is a design parameter, where high values for θ require many
multipole coefficients, but give few near field interactions. The value θ = 0.5
was chosen in the current work. With fixed θ , the number of multipole and
local coefficients determines the tolerance of the code.
A disadvantage with asymmetric adaptivity is that the shape of the boxes
varies and large boxes can have many boxes in their near fields according to
criterion 1. This is illustrated in figure 4.6, where the larger box has near-
field connections to more boxes than the smaller one. This is most notable in
the spiral particle distribution in figure 4.5 (c), where the median split caused
some boxes to extend over both the inner and outer part of the spiral. A further
disadvantage is that the sorting becomes more complicated as finding the me-
dian point is computationally more expensive than splitting at the geometric
midpoint.
As the symmetries described in [64] cannot be used, the size of all boxes
have been reduced, which reduces the number of boxes in the near field. Size
reduction could be applied for the midpoint adaptivity as well, at the cost of
not being able to use the symmetries in [64]. The current implementation of
the code performs all multipole to local translations within the same multipole
level (similar to the homogenous case in figure 4.4).

CPU versus GPU


The fast multipole method has been implemented both for the central pro-
cessing unit (CPU) and the graphics processing unit (GPU), and the results

60
for these architectures are analyzed in paper IX. A brief introduction to the
different systems is given here.
The traditional CPU consists of a relatively small number of CPU cores,
where each core is relatively fast. This is to be compared to the GPU, where
there are many slower computational cores. As the GPU used in Paper IX is
a Nvidia Tesla C2075, this GPU (which is based on Nvidia’s CUDA architec-
ture) will be in focus when describing the GPU architecture. This GPU has
14 multiprocessors which each have 32 cores, giving a total of 448 cores. The
GPU has several threads grouped together into a warp (where the size of a
warp is 32 threads). Within a warp, only one instruction can be executed at
the same time by all the threads in this warp, although all threads can oper-
ate on different data. This is called single instruction multiple data (SIMD).
If different threads within a warp have to perform different operations at the
same time, all instructions have to be serialized, where the CUDA cores of the
threads not executing the particular instruction become idle. If the number of
threads within the warp is not equal to 32, it will also mean that the rest of
the 32 CUDA cores become idle [65, chapter 4]. To fully use the GPU, the
algorithm has to be data parallel in a way that 32 identical instructions can
be executed all the time. To compare with the CPU case, most modern PC
processors support streaming SIMD extensions (SSE), which allows for sim-
ple algebraic operations to be performed on two double precision numbers (or
four single precision numbers) in one operation [66]. To get a good compari-
son with the GPU code, the entire multipole evaluation phase was written with
SSE instructions in the CPU case, which accelerated most parts with a factor
close to two.
In the GPU computational model, functions (called kernels) are sent from
the CPU to the GPU. Each kernel contains a set of code to be executed and
all threads execute this kernel. A kernel will be launched with a predefined
number of threads, which are grouped together into blocks. Communication
between threads is typically only performed within the block, as there is cur-
rently no synchronization method between different blocks, except for ending
the kernel and launching a new one, although some methods for sharing mem-
ory between blocks have been developed, see e.g. [67, appendix A]. Com-
munication between threads within a block is performed through a special
fast memory called shared memory, of which each multiprocessor has its own
(48 kB on the Tesla C2075). For optimal performance, the number of threads
is usually a multiple of the warp size to avoid idle cores as all threads in a warp
must belong to the same block.
The GPU programming model is based on launching a large number of
threads and each core will therefore run many threads. Several blocks can run
on the same multiprocessor simultaneously to increase the number of active
threads. The GPU is very fast at switching between threads and the large
number of threads per core is used to hide the time it takes to complete certain
instructions by issuing other instructions in the meantime. Too few threads

61
Part Time Part Time
Direct evaluation (P2P) 43 % Connectivity 1%
Sorting 30 % Multipole to multipole
Multipole to local shift (M2L) 11 % shift (M2M) <1%
Multipole initialization (P2M) 5% Local to local (L2L) <1%
Multipole evaluation (L2P) 2% Other 8%
Table 4.1. Time distribution among different parts in the GPU implementation for
optimal Nd .

executing on a multiprocessor result in lower performance. One case where


this could occur is if each thread requires too much shared memory, as the
48 kB has to be enough for all threads together.

The FMM on the GPU


The implementation of the FMM on the GPU is given in paper IX and only
a brief description is given here, where more focus is given to the more com-
putationally expensive parts. All parts of the code are implemented on the
GPU.
The sorting successively splits a box in two parts according to a chosen
pivot point. After each split, the part that does not contain the median will
be excluded for the next split until the median is found. This will require
O (log Nd ) steps to complete. For the CPU, a split can be performed in a single
pass by interchanging elements. The GPU uses a two pass algorithm, where
the first step counts the elements (each thread has its on subset of elements).
Then the cumulative sum is calculated between the threads and afterward, the
elements can be inserted into a common array. The speed of the sorting code
depends on the number of particles in each box and if the number becomes too
low, all 32 threads of the warp will no longer be active, hence reducing speed
for small boxes. The connectivity (application of criterion 1) is determined
with one thread per box making the code very similar to the CPU version.
The multipole to local shift (M2L) uses one thread block to calculate the
interaction of one box (possibly two boxes if few interactions). Two threads
per box are used, where one thread shifts the real part and the other shifts the
imaginary part. The algorithm will store the entire set of coefficients to shift
in shared memory as it is used several times during the shift. In other respects,
the implementation is similar to the CPU implementation and as the operation
can perform shifts from two boxes within a block, it can often use most of the
32 threads of the warp. Multipole to multipole (M2M) and local to local (L2L)
shifts work similarly, but each level is evaluated in its own kernel launch as the
shifts depend on the results from the shifts at the previous level, see figure 4.4.
The number of shifts per box is fixed to 4, hence 4 boxes use 32 threads and
all threads of the warp can be used.

62
25
Sort
P2M
20 L2P
P2P
15
Speedup

10

0
0 10 20 30 40 50 60 70
Nd

Figure 4.7. Computational time as a function of number of particles per box.

The direct interaction (P2P) uses one block for one box and loops through
all boxes in the near field. The particle positions are loaded into shared mem-
ory for speed issues, and 1, 2 or 4 threads are used for each particle to keep
all threads of the warp occupied throughout the calculations. With too few
particles in the box however, there will still be problems in keeping all threads
occupied, hence performance is reduced. For the CPU, the symmetry that the
interaction between two particles is mutual (if source and evaluation points
are identical) is applied, which reduces the computational time with almost a
factor of 2. Symmetries are not applied for the GPU, as it is desirable that two
threads do not write to the same memory position for a thread-safe code.
The multipole evaluation (L2P) uses one thread for each particle and caches
the coefficients. Otherwise, it is similar to (P2P). The multipole initialization
(P2M) uses one block per box and 1 or 2 threads per particle, but requires
extra thread synchronization and summation as all particles contribute to all
coefficients.

Computational times for the GPU implementation


The time distribution between the different parts in the GPU case is illustrated
in table 4.1 for N = 45 × 216 , giving Nd = 45 and 8 levels. The most time-
consuming part is the direct evaluation (P2P), followed by the sorting and
the multipole to local evaluation. For a detailed study, timing experiments of
all individual parts have been performed. The performance with respect to
the number of particles is given in figure 4.7, where the speedup is the ratio
between the time of the single threaded highly optimized CPU version and
the time of the GPU version. The trends clearly show that the GPU performs

63
25
P2M
20 M2M
M2L
Speedup

15 L2L
L2P

10

0
10 20 30 40 50
p
Figure 4.8. Computational time as a function of number of multipole coefficients.

better for more particles in each box, since more threads can participate in
each operation, hence reducing the number of idle threads. With more than 64
points for L2P (which uses one thread/particle), 32 for P2M (which uses up to
two threads/particle) and 16 for P2P (which uses 4 threads/particle) one extra
loop iteration is required on the GPU, hence the big drop in performance. The
multiple threads per particle options for P2P and P2M were added as these are
the more computationally expensive parts of the code, see table 4.1.
The second timing experiment investigates the variation in speedup with
respect to the number of multipole coefficients. Figure 4.8 shows that the de-
pendence on the number of multipole coefficients is less than the dependence
on the number of particles per box. L2P has a clear trend of increasing perfor-
mance with more coefficients, which is where the computational overhead is
less dominant than the actual calculations. The shift operators M2M, M2L and
L2L works best at around 20 coefficients, where at higher number of coeffi-
cients, the extensive usage of shared memory limits the number of blocks that
simultaneously can execute on the same multiprocessor. As an example, at 42
coefficients, M2L decreases from 4 to 3 blocks that can run simultaneously,
resulting in the drop in performance seen in figure 4.8. The time to determine
the connectivity depends only on the number of FMM levels. At one million
particles the speedup of this part is around 5 at optimal number of particles per
box.
The timing of the entire
 2  algorithm is given in figure 4.9. While the CPU
version follows the O N scaling of the direct evaluation and O (N) scaling

64
1
10

0
10

∝ N2
−1
10 ∝N
Time (s)

−2
10

GPU FMM
−3
10 CPU FMM
GPU Direct
CPU Direct
−4
10
3 4 5 6 7
10 10 10 10 10
N
Figure 4.9. Computational time as a function of number particles.

for the FMM in the entire plotted interval, the GPU direct evaluation has a
linear behavior at the beginning
  before all CUDA cores are fully occupied
and continues with an O N 2 behavior at high numbers of particles. The
computational overhead in the GPU FMM algorithm causes the algorithm to
be almost independent of the number of particles for low N and results in an
O (N) behavior at high N. At around 3500 particles, it becomes more efficient
to use the FMM than direct evaluation for the GPU.
The speedup in the direct evaluation case is asymptotically 15 for direct
evaluation and 11 for FMM if source and evaluation points are identical, al-
lowing for symmetries to be used in the direct evaluation in the CPU version.
Without symmetry, asymptotic speedup would be around 30 for direct evalua-
tion and 15 for FMM.
The GPU version of the FMM is significantly faster for large numbers of
particles. If vortex merging is applied to a simulation of a single turbine,
the number of vortices at the time of convergence would be around 1000 –
5000, depending on the accuracy. When simulating multiple turbines, both
the number of released vortices per time-step and the number of time-steps
increase before convergence. As the GPU FMM obtains its high speed up at
large numbers of particles, it is suitable for multiple turbine simulations, while
single turbine simulations works well for both GPU direct evaluation and the
CPU based FMM code.

65
5. Simulation results

The simulation results are presented in two parts. The first part contains a
comparison of the two different simulation models, and the different modifi-
cations of them. This is to illustrate how the simulation model works, which
generally is not presented in articles, and this section will therefore consist of
unpublished work. The second part consists of the results from the published
articles that are included in this thesis.

5.1 Evaluation of simulation tools


In this section, the experimental turbine located in Marsta outside Uppsala [16]
will be used as a reference case. In the simulated model, the turbine parameters
can be found in table 5.1. The struts are assumed to vary piecewise linearly
from the tip to the center, and from the center to the root. The trailing edge of
the strut is blunt at the root, and the flat end is approximately 38 mm wide
at the root (approximately 18 mm at the center, and sharp at the tip). To
account for this, a set of airfoils with blunt trailing edges was simulated in
XFOIL [68], and the increase in drag due to the blunt trailing edge was added
as an additional drag term for the struts.
As a comparison, the simulated curves are compared to experimental val-
ues, which were obtained according to [16]. The measurements were per-
formed at a constant rotational velocity of around 6 rad/s, which means that
the values for higher tip speed ratios were obtained from lower flow veloci-
ties, where external losses (e.g. mechanical/electrical losses etc) are more sig-
nificant. Considering that the power coefficient is expected to increase with
increasing flow velocity (increasing Reynolds number), which is illustrated in
figure 5.1, the measured power coefficient will likely have its peak at a slightly
lower tip speed ratio than it would if the measurements would have been per-
formed with constant flow velocity. In all simulations where comparisons are
made, the simulations are therefore performed at constant rotational velocity
as well.
For the measured data, the accuracy is consistent at low tip speed ratios
where the flow velocity is high, but shows a more scattered behavior at higher
tip speed ratios (see figure 5.1). For high tip speed ratios, the power is lower
due to the lower flow velocity, and uncertainties due to other losses, such as
bearing friction and generator losses etc. are more prominent, further increas-
ing the uncertainty in this region. Due to the cubic relation between flow ve-
locity and power, a small measurement error in wind speed will give a much

66
Number of blades 3
Radius 3m
Blade height 5m
Blade profile NACA0021
Chord at center 0.25 m
Tapered blades (from blade tip) 1m
Chord at tip 0.15 m
Strut profile NACA0025
Strut tip chord 0.2 m
Strut middle chord 0.28 m
Strut root chord 0.32 m
Strut root attachment point (from center) 0.2 m
Strut tip attachment point (from center) 1.15 m
Table 5.1. Turbine parameters

0.5
Measured points
Bins
0.4 Ω = 8 rad/s
Ω = 6 rad/s
Power coefficient

Ω = 4 rad/s
0.3

0.2

0.1

0
1.5 2 2.5 3 3.5 4 4.5
Tip speed ratio
Figure 5.1. Simulated power coefficient for different rotational velocities to illustrate
Reynolds number dependence. The curve “Bins” is the calculated average of the mea-
sured points using the method of bins [16].

67
0.5
Measured points
Bins
0.4 No struts
No flow correction
Power coefficient

Flow correction
0.3

0.2

0.1

0
1.5 2 2.5 3 3.5 4 4.5
Tip speed ratio
Figure 5.2. Simulated power coefficient with and without strut modeling compared to
measured data.

larger contribution to the power coefficient and since the measurements are 10
minute mean values, changes in flow velocity during the measurement period
also contribute to the measurement error.

5.1.1 Strut modeling


Struts can either be modeled as only giving drag (which was done in [49])
using an equivalent drag for the all struts only varying with the rotational ve-
locity, or the equations in section 4.1.2 can be used, which include lift force
on struts and corrections to the flow velocity. Lift force on struts occur if the
struts (or the flow velocity) are not horizontally aligned (in paper I of this the-
sis, the struts are horizontally aligned, and only the drag terms are included
without the flow corrections). The differences between including struts with
flow corrections or not can be seen in figure 5.2. In these simulations, curva-
ture modeling with angle correction and all tip corrections are enabled, but no
expansion model. By comparing the three simulated figures, the strut model,
including lift force on struts and flow corrections, obtains a higher power coef-
ficient at low tip speed ratios, which is closer to the measurements. All models
estimate a higher power coefficient than measurements at high tip speed ratios,
but by using strut modeling, the values are closer to the measurements.

68
0.5
Measured points
Bins
0.4 With expansion
Without expansion
Power coefficient

0.3

0.2

0.1

0
1.5 2 2.5 3 3.5 4 4.5
Tip speed ratio
Figure 5.3. Simulated power coefficient with and without expansion modeling com-
pared to measured data.

5.1.2 Expansion model


The expansion model used is described in section 4.1.5. The expansion sim-
ulations have been performed using strut modeling, all tip corrections and
curvature modeling with angle correction. Flow expansion effects are most
prominent at high axial induction factors (defined in equation (2.9)), which
can be seen in figure 5.3, where the expansion model causes the power coef-
ficient to decrease for high tip speed ratios. This is caused by a larger region
of low velocities at the downstream disc. When comparing with experiments,
the expansion model is closer to the experiments at high tip speed ratios, but
the peak value is lower than the simulated one. Neither of the two curves is
able to match the slope of the measured curve at high tip speed ratios. As
another comparison, the results are compared with vortex method results in
figure 5.4. In this comparison, struts and tip corrections are omitted, since the
two-dimensional vortex model cannot model these three-dimensional effects
properly. The blades were also simulated with uniform chord, while the real
turbine has tapered blades. The compared tip speed ratio of 4.5 was chosen
quite high to make expansion effects more prominent. In figure 5.4, the re-
gion -90 to 90 degrees in blade position correspond to the upstream disc. The
peak tangential force is approximately the same for both models, but due to
flow expansion effects, the vortex model obtains a smaller region where the
tangential force is high on the upstream disc, but a wider region on the down-
stream disc. From these results, it can be seen that the expansion model causes

69
35
Vortex model
30 Streamtube without expansion
Streamtube with expansion
25
Tangential force (N)

20

15

10

−5
−100 0 100 200 300
Blade position (degrees)
Figure 5.4. Simulated tangential force with and without expansion modeling at tip
speed ratio 4.5, compared to simulations using a vortex method.

a lower tangential force downstream, while according to the vortex model, it


is upstream that the force should be decreased.

5.1.3 Tip correction model


For the tip corrections, there are two models used for streamtube theory. Model
1 is based on screw propeller theory and model 2 comes from finite wing the-
ory. These models are described in section 4.1.3. Sharpe [48] uses only the
screw propeller model in his work, but Paraschivoiu [3] uses both models. Fig-
ure 5.5 shows that the first model gives the largest contribution to the results
and all simulations omitting this correction have too high power coefficient
for high tip speed ratios. The use of the finite wing theory corrections, as used
by Paraschivoiu, lowers the power coefficient slightly, which is better at high
tip speed ratios, but gives a too low peak power coefficient. It is therefore
inconclusive if the finite wing theory model should be included or not.

5.1.4 Curvature modeling


There are two different curvature models described in section 4.1.4. One is
based on correcting the normal force coefficient C N , which is used by e.g.
Sharpe [48], while the other one corrects the angle of attack of the blade.
Paraschivoiu does not mention any flow curvature modeling in his work [3].

70
0.5
Measured points
Bins
0.4 No tip corrections
Only model 1
Power coefficient

Only model 2
0.3 Both models

0.2

0.1

0
1.5 2 2.5 3 3.5 4 4.5
Tip speed ratio
Figure 5.5. Simulated power coefficient for different implementations of tip vortex
corrections.

0.5
Measured points
Bins
0.4 No correction
C correction
N
Power coefficient

0.3 Angle correction

0.2

0.1

0
1.5 2 2.5 3 3.5 4 4.5
Tip speed ratio
Figure 5.6. Simulated power coefficient when using curvature modeling.

71
10

5
Angle of attack (degrees)

−5
Vortex model
No correction
−10 CN correction
Angle correction

−15
−100 0 100 200 300
Blade position
Figure 5.7. Comparison of simulated angle of attack for one revolution between the
streamtube model and a vortex model.

When comparing against the measured data (figure 5.6), the curvature models
both predicts lower power coefficients than the reference without corrections.
According to figure 5.6, the CN correction appears to give a power coefficient
that more closely corresponds to the measurements than the angle correction.
Simulating without any corrections does also give a better match than with the
angle correction. As a further comparison, the angle of attack is compared
with results from a vortex simulation. Similar to section 5.1.2, tip corrections
and strut modeling were omitted and uniform chord was used for both simula-
tions. No expansion modeling was performed for the streamtube code either.
According to the results in figure 5.7, the angle correction model is more sim-
ilar to the vortex model when it comes to predicting the peak angle of attack
upstream, but for the remaining part of the revolution, running without any
corrections appears to be better. The C N correction has larger disagreement
upstream according to this figure, while being similar to using no corrections
for the rest of the revolution. It should be noted that it is this lower angle of at-
tack that causes the decrease in power coefficient in figure 4.1.4, which gives
better results in that figure. From these results, it is therefore inconclusive
which model that actually gives the best results.

72
0.5

0.4
Power coefficient

0.3

0.2

0.1 Vortex model


Streamtube without expansion
Streamtube with expansion
0
1 2 3 4 5 6
Tip speed ratio
Figure 5.8. Simulated power coefficient using vortex model compared with the
streamtube model.

5.1.5 Vortex model


The vortex model has the advantage that it actually solves the flow from
Navier-Stokes equations, instead of adding many correction models, which
is done for the streamtube models. The problem is that the current implemen-
tation is a two-dimensional model, which cannot model struts and tip effects in
the same way as three-dimensional models such as the streamtube model. To
get a good comparison between the two models, all three-dimensional effects
have been omitted from the streamtube model. The results can be found in
figure 5.8, where the vortex model is compared against the streamtube model,
both with and without expansion modeling. In this figure, flow curvature ef-
fects have been included in the streamtube model using the corrections to the
angle of attack. At tip speed ratio 4.5, the angle of attack is plotted in fig-
ure 5.7 and the tangential force in figure 5.4. The general results are that the
streamtube model predicts higher torque on the upstream disc and lower on
the downstream disc, due to flow expansion effects not properly being mod-
eled in the streamtube theory. In this particular case, the vortex model should
be more accurate, but the streamtube model does currently show the ability to
model three-dimensional effects more flexibly.

73
0.5
Measured points
Bins
0.4 Simulation
Power coefficient

0.3

0.2

0.1

0
1.5 2 2.5 3 3.5 4 4.5
Tip speed ratio
Figure 5.9. Closest matching power coefficient using streamtube model.

5.1.6 Concluding remarks about the simulation tools


From all the simulations performed, it can be concluded that the best fit to ex-
perimental data would be obtained if strut modeling, the screw propeller part
of the tip corrections, the CN correction for curvature effects and the expan-
sion model are used. For this case, the simulated power coefficient is plotted
in figure 5.9. Even though this seems good for the particular turbine, the com-
parisons with the vortex model show for example that the expansion model
does not really match the vortex model solutions of Navier-Stokes equations.
Neither do the curvature corrections show good results in the comparison with
vortex theory. For the tip corrections, further studies are necessary to validate
the model, and the strut model also requires additional validation. Further, us-
ing the power coefficient as validation is not the best method, as it cannot tell
where during the revolution the torque was generated. It should therefore not
be concluded that the model configuration that created figure 5.9 really is the
best solution. Instead, additional, and more accurate, experimental results, or
numerical results using more advanced models, are necessary to validate the
model more properly.

5.2 Results from papers


This section contains a summary of the results from the papers included in
this thesis. Papers I – III use the streamtube model, while papers IV – VIII use

74
3 blades (2.5)
6 blades (2.5)
3 blades (1.5)
6 blades (1.5)

0.4
Power coefficient

0.35

0.3
2.5 3 3.5 4 4.5
Tip speed ratio

Figure 5.10. Optimal power coefficient as a function of design tip speed ratio. This
is a comparison between turbines designed for 2.5 m/s and turbines designed for 1.5
m/s, all turbines simulated at 1.5 m/s (figure 5, paper I).

the vortex model. Paper IX studies the numerical speed of the fast multipole
method, which is used by the vortex model and the results from this paper are
presented in section 4.2.4.

5.2.1 The effects of struts


For a straight blade Darrieus turbine, the blades are attached to the main shaft
through struts. These struts cause additional losses, which increase if addi-
tional struts are required to reduce the stress in the blades. In paper I, the
strut losses are calculated from the equations in section 4.1.2, but without any
corrections to the flow, which can be considered acceptable for horizontally
aligned struts only experiencing drag forces. Further, tip losses are neglected
when calculating forces to slightly overestimate the normal force on the blade,
but are included for the power coefficient calculations. No additional correc-
tion models, except dynamic stall modeling, are included.
To calculate the maximum allowed distance between two struts, the normal
force was calculated from the streamtube model. For the structural mechanical
parts, FEM simulations have been performed to obtain the stress in a blade.
As an approximation, it is found that, if the blade is solid, the stress is approx-
imately
L2
σ ≈ 18 FNl ,
c3
75
where FNl is the normal force per meter, and L is the distance between the
struts. From this equation, the maximum allowed length between two struts
can be determined if the maximum allowed stress is given. In this study, the
blades are assumed to be made of steel, and the maximum stress is set to
100 MPa. From further FEM studies, it was found that the number of struts
approximately can be given by the equation
h
Ns = + 0.2.
L
The number of struts Ns is set to be at least 2. The studied turbine is 5×5 m and
it is designed to run in water, where hydrodynamic forces are more prominent
than centrifugal forces (both forces are included in FNl ). To determine the
necessary size of the struts, FEM studies have been performed for a blade
with chord c = 0.2 m. With this blade chord, a strut chord c s of 0.8c at the tip,
and 1.2c at the root was determined to be sufficient to keep the strees within
the limits. If the chord is increased, it is not necessary to increase the strut
chord as much, and the equations

cs = 0.8 c0 c tip,

cs = 1.2 c0 c root,

give good results for the stress in the struts. Here, c 0 is the chord used in the
FEM studies, e.g. c 0 = 0.2 m.
Two different turbine configurations have been tested, one with 3 blades and
one with 6 blades. For each configuration, one is designed to run at optimal
tip speed ratio at flow velocity of 1.5 m/s and one is designed to run at 2.5 m/s.
Due to the higher forces at 2.5 m/s, additional struts are required, especially
for a turbine with 6 blades. Results from all four combinations are found in
figure 5.10. The chord is adjusted to give the peak of the power coefficient
curve for each tip speed ratio in this figure. The sharp decreases in power
coefficient correspond to the positions where the chord is small enough to
force the addition of an additional strut. Figure 5.10 suggests that a turbine
with 6 blades can give higher power coefficients than a turbine with 3 blades,
which is a result of the higher aspect ratio for the 6-bladed turbine due to
its smaller chord. It can also be seen that the 6-bladed turbine needs more
struts due to its smaller chord. One interesting result is that if the turbine is
designed for 2.5 m/s, the additional struts required for the 6-bladed turbine
decrease the power coefficient. Therefore, the turbine designed for 2.5 m/s
has a lower power coefficient at 1.5 m/s than the turbine designed for only 1.5
m/s. The implication is that if a turbine is designed for high flow velocities, the
performance at lower flow velocities can be reduced due to additional losses.
Therefore, designing a turbine to run at high flow velocities can actually cause
a decrease in the total energy extraction if high flow velocities at the site are
rare.

76
2 1

3 2

4 3
Depth (m)

Depth (m)
5 4

6 5

7 6

8 7
1.5 1.55 1.6 1.65 1.7 0.35 0.4 0.45 0.5 0.55
Velocity profile 1 (m/s) Velocity profile 2 (m/s)

Figure 5.11. Velocity profile 1 is measured in Dalälven and profile 2 in Nordre Älv
(from figure 3, paper II).

5.2.2 The effects of a velocity profile


In a real flow, the velocity varies over the cross-sectional area of the turbine.
For wind power, this has been studied by e.g. Homicz [69] and Paraschi-
voiu [70] for the curved blade Darrieus turbine, with a velocity field varying
in the vertical direction. These studies used theoretical velocity profiles. In
Paper II, actual measurements have been performed in the river Dalälven at
Söderfors and in the river Nordre Älv at Ormo. The measured profiles are
displayed in figure 5.11 and have been used to study the turbine performance.
Four different turbine cases have been studied. One homogeneous profile,
one with vertically varying profile and the case with the turbine 90 degrees
tilted, making it horizontally aligned. For the tilted turbine, both rotational
directions have been studied. The four cases can be found in figure 5.12.
This study was performed with streamtube theory. Dynamic stall modeling,
finite wing theory corrections, and flow expansion are included, but no curva-
ture or strut losses. To simulate a velocity profile, the starting point of each
streamtube is calculated with the expansion model, and the starting velocity
in each streamtube is interpolated from the given velocity profile. To calcu-
late the power coefficient, the available power is calculated by integrating the
energy over the velocity profile, and the tip speed ratio is obtained using the
cube root of the mean cube velocity as a reference velocity.
The studied turbine is 6 m in both height and diameter, and the two chords
0.24 m (turbine A) and 0.34 m (turbine B) have been studied to illustrate how
the profile affect turbines optimized for different tip speed ratios. Turbine B

77
a) b)

6m 6m

6m

c) d)
6m 6m

Figure 5.12. Turbine cases: a) is a vertically (or horizontally) aligned turbine with
homogeneous incoming flow, b) is a vertically aligned turbine with inhomogeneous
velocity profile, c) is a clockwise rotating horizontally aligned turbine with inhomo-
geneous profile and d) is a counter-clockwise rotating with inhomogeneous profile
(figure 4, paper II).

Turbine A Turbine A
0.34
0.38 a a
Power coefficient

Power coefficient

b 0.32 b
0.36 c c
d 0.3 d
0.34
0.28
0.32
0.26
2.5 3 3.5 4 2.5 3 3.5 4

Turbine B Turbine B
0.34
0.38 a a
Power coefficient

Power coefficient

b 0.32 b
0.36 c c
d 0.3 d
0.34
0.28
0.32
0.26
2.5 3 3.5 4 2.5 3 3.5 4
Tip speed ratio Tip speed ratio

Velocity profile 1 Velocity profile 2

Figure 5.13. Power coefficient as functions of tip speed ratio for the different cases.
Figures to the left are for Dalälven and to the right are for Nordre älv. The four curves
represent the different turbine cases (from figures 8 and 9, paper II).

78
was intentionally chosen to have its peak in the dynamic stall region, while
stall should be less important for turbine A. In figure 5.13, it can be seen that
rotational direction does not matter for high tip speed ratios where stall is not
important. For lower tip speed ratios, case c) generally performs better than
case d), which can be explained from the maximum angle of attack. From
equation (2.26), it follows that the angle of attack is lower when the blade
moves against the flow direction, and that an increase in the flow velocity
increases the angle of attack. For case c), it is the region with a lower angle
of attack that gets an increased angle of attack due to the profile, while for
case d), the region with already high angles of attack gets its angle increased,
increasing the stall, which decreases performance at low tip speed ratios. In
addition to an improved performance at low tip speed ratios for case c), another
trend is that for profile 1, the power coefficient is higher than the homogeneous
reference case at high tip speed ratios, while it is lower for profile 2. The
vertical axis turbine will extract most energy close to the center, which in the
horizontal plane is the position where the angle of attack is largest (this is
valid unless stall effects are too large). In the vertical plane, the corrections
due to tip losses cause the decrease in absorbed power close to blade tips. For
profile 1, the energy available in the center is larger than the mean energy,
which explains why the turbine works better with a profile, while for profile 2,
the available energy is less than the mean in the center.

5.2.3 Design of a turbine for use in a river


Combining the work in the two previous papers, a design study of a turbine
for a Swedish river was performed in paper III. Due to site specific constraints,
the turbine height is limited to 3.5 m. For mechanical reasons, 5 blades was
chosen, which should give much lower torque oscillations than a turbine with
fewer blades, while adding more blades would increase costs and make the
blades very thin.
The study was performed with the streamtube model, including strut losses
as in section 5.2.1. Similar to section 5.2.2, a measured velocity profile was
used with the difference that the expansion model was omitted due to incom-
patibilities with the strut model.
The chord was chosen as 0.18 m, which gives a peak power coefficient
inside the dynamic stall region. This is due to structural mechanical consid-
erations. Simulations indicated that the maximum power coefficient should
start to drop for any larger chords due to stall effects, but a large chord reduces
the stress in the blade. The estimated power coefficient curve is displayed in
figure 5.14.

79
0.4
0.35

0.3
Power coefficient

0.2

0.1

0
2 3 3.5 4 5
Tip speed ratio
Figure 5.14. Estimated power coefficient for designed turbine (figure 5, paper III).

5.2.4 Turbines in channels


When a turbine is located in a channel, the channel walls will affect the per-
formance, which theoretically has been described in section 2.2. The theory
does however only describe ideal turbines. For a real turbine, there will be
losses due to drag, which decreases the performance. Further, a vertical axis
turbine does not absorb power equally over the cross-sectional area. There-
fore, a real vertical axis turbine is not expected to perform in the same way as
in the theoretical model for channels in section 2.2.
The aim of this study (paper IV) is not only to investigate the effects of
the channel geometry, but also to give guidelines on how to do simulations
on them, and how the simulations will be affected by the presence of channel
walls, which always are present in volume discretization based methods like
the finite element method.
In the simulations, no three-dimensional corrections (such as tip vortex cor-
rections) have been applied, as additional modeling can create new trends in
the results, and the aim of the study is to focus only on the channel effects.
With no three-dimensional corrections, the power coefficients will be overes-
timated (at least in the free flow case). All channel simulations in this section
use the analytical expression for the potential flow in the channel according to
section 4.2.3.
Initially, a turbine with 3 blades and a diameter of 10 m was designed for
free flow with a chord of 0.4 m and 3 degrees pitch angle (outwards). In the
free flow case, this turbine obtains its highest power coefficient of 46.4 % at a

80
120

100

80
Channel width

60

40

20

0
0 20 40 60 80 100
Revolutions
Figure 5.15. Number of revolutions until convergence as function of channel width.
Channel width is measured in terms of turbine diameters (figure 4, paper IV).

tip speed ratio of 3.4. For this case, approximately 100 revolutions are required
for convergence. The necessary number of revolutions is a function of tip
speed ratio, and increases with increasing tip speed ratio. Simulations at high
tip speed ratios show oscillatory behavior and with the chosen convergence
criterion that the power coefficient should remain within 0.001 of the final
value, these simulations do not show any good convergence at all.
The primary numerical parameter investigated is the number of revolutions
required for a given channel width. Here, all simulations have been performed
at the optimal tip speed ratio. Figure 5.15 shows that the necessary number
of revolutions increases with increasing channel width, where the increase at
the beginning is linear, while for increasing width, it approaches the free flow
case.
The second part of the study shows how the turbine performance changes
due to the channel width. According to analytical predictions, the power co-
efficient should increase with decreasing channel width, and this is indeed the
case (see figure 5.16), but the power coefficient increases slower than theory
predicts. When the power coefficient increases, the tip speed ratio, for which
the peak occurs, increases as well, illustrated in figure 5.17. It is generally not
optimal to have the peak tip speed ratio too high, since drag losses are more
dominating at high tip speed ratios, and for small channels, the power coeffi-
cient can be improved by increasing the chord. As an example, for a channel
of 4 turbine diameters, the turbine with chord 0.4 m obtains a 37 % increase

81
1

Simulated
0.9 Theoretical
Simulated/Theoretical
Power coefficient

0.8

0.7

0.6

0.5

0.4
0 20 40 60 80 100
Channel width
Figure 5.16. Changes in Power coefficient due to the channel width. Theoretical val-
ues are calculated from the theory in section 2.2. The dotted line is the ratio between
the two other curves (figure 6, paper IV).

7
Tip speed ratio

3
0 20 40 60 80 100
Channel width
Figure 5.17. Optimal the tip speed ratio as a function of channel width (figure 8,
paper IV).

82
V

Co-rotating Inner counter- Outer counter-


rotating rotating

Figure 5.18. Turbine configuration in the two turbine case (figure 1, paper V).

compared to free flow, but a turbine with chord 0.6 m in the same channel ob-
tains a 46 % increase over the chord 0.4 m turbine at free flow. The theoretical
prediction is however 78 %. Even when simulating with ideal blades, the sim-
ulated power coefficients do not reach the theoretical prediction, as the ideal
turbine obtains an increase of 48 % in the above case with chord 0.4 m, which
is quite close to the turbine with optimized chord. One implication of these
results is that the theoretical prediction seems to be an overestimation for the
vertical axis turbine. Another implication is that when designing a turbine for
a channel, the blockage ratio of the system has to be taken into account if an
optimal design is desired.
All above simulations were performed in two dimensions. In reality, the
flow can not only pass on the sides of the turbine, but also over and under
it, giving a lower blockage ratio, and according to the analytical model, it
is the blockage ratio that is important for the increase in power coefficient.
Therefore, keeping the blockage ratio correct in the simulations may be more
important than actually using the correct channel width (but it would require
three-dimensional simulations to validate this assumption).

5.2.5 Turbines in an array


For commercial applications of vertical axis turbines, it is desirable to locate
several turbines in proximity to each other in a turbine farm. For wind power,
where the flow direction can vary significantly, a relatively large distance be-
tween the turbines is commonly chosen to limit the effects of turbines op-
erating in the wake region of another turbine. For water currents, the flow
direction can be known in advance. One example is a river, where the flow
always originates from one direction. With this knowledge, it should be pos-
sible to place the turbines much closer to each other. In this study, the effects
of turbines in close proximity are investigated. Due to the close proximity, the
two-dimensional model should be reasonably accurate.
In paper V, a 3-bladed turbine with cross-sectional area 6 × 6 m and chord
0.25 m is studied, and a good blade pitch angle is determined to be 3 degrees

83
0.45

0.4
Power Coefficient

0.35

Co−rotating turbine 1
0.3 Co−rotating turbine 2
Inner counter−rotating
Outer counter−rotating
Single turbine
0.25
3 3.5 4 4.5 5 5.5
Tip Speed Ratio
Figure 5.19. Simulation using two turbines. For the co-rotating case, the individual
turbines are shown, while in the other two cases, both turbines give the same power
coefficient due to symmetry (figure 3, paper V).

8m

1
8m 8m
2
V
3
4
5
Case 1 Case 2
Figure 5.20. Turbine configuration in the five turbine case (from figure 6, paper V).

84
0.5

0.45
Power Coefficient

0.4
Turbine 1
Turbine 2
Turbine 3
0.35
Turbine 4
Turbine 5
Mean Value
0.3
3 3.5 4 4.5 5 5.5
Tip Speed Ratio
Figure 5.21. Simulated power coefficient for five turbines in a row configuration
(case 1) (figure 7, paper V).

0.6

0.5
Power Coefficient

0.4

0.3

Turbine 1
0.2 Turbine 2
Turbine 3
0.1 Turbine 4
Turbine 5
Mean Value
0
3 3.5 4 4.5 5 5.5
Tip Speed Ratio
Figure 5.22. Simulated power coefficient for five turbines in a zigzag configuration
(case 2) (figure 8, paper V).

85
outwards. Throughout the study, all turbines were run in synchronous opera-
tion (i.e. with the same rotational velocity as its neighbors). The first part of
the study concerns a two turbine system with 8 m between the turbine centers,
and the dependence on rotational direction is investigated. The three differ-
ent rotational combinations are illustrated in figure 5.18. According to figure
5.19, the differences between the different rotational directions are quite lim-
ited and for the rest of the simulations, all turbines were set to be co-rotating.
One important result is that the simulated power coefficients in the two turbine
case are higher than for the single turbine. The increase is quite moderate, but
it still shows that it actually can be beneficial for a turbine to have another
turbine in close proximity.
The second part concerns a five turbine array and the placement strategy is
investigated. One possible solution is to have all turbines on a row (case 1),
while another is to use a zigzag configuration (case 2), as illustrated in fig-
ure 5.20. In the five turbine case, the differences between the power coeffi-
cients for the individual turbines are more significant. For the row configura-
tion, the differences are quite moderate (figure 5.21), while for case 2, the two
turbines behind the other have significantly higher power coefficients, while
the middle turbine performs worst. When comparing the mean power coef-
ficient, the row configuration performs better than the zigzag pattern, and in
the row of five turbines, the power coefficients for each turbine have increased
even more that in the two turbine case. A final test was to change the incoming
flow angle 30 degrees. Here, the row configuration performs much better, and
its mean power coefficient remains close to the original value (1 % decrease),
while in case 2, the decrease is around 20 %. Because the row configuration
obtains a higher mean power coefficient and appears to be less sensitive to
changes in flow direction, this study indicates that this configuration should be
preferable, given that the turbine size is kept constant. However, if the avail-
able area for turbine placement is limited, using a zigzag pattern would allow
more turbines in the same area.
In paper VI, an array of 10 turbines placed on a straight line is studied with
respect to the turbine spacing and incidence flow angle. A 3-bladed turbine
with cross-sectional area of 10 × 10 m is used and the chord is chosen as
0.34 m to obtain the peak tip speed ratio at 4. The simulations have been
performed for 1 m/s flow velocity and all turbines were set to operate at tip
speed ratio 4 in all cases. The results in figure 5.23 show that for low incidence
angles, the dependence on turbine spacing is relatively low, although a small
improvement is seen for the lowest spacing. Large turbine spacing is beneficial
if the incidence angle is high, as less of the cross-sectional area of the turbines
will be in the wake of the turbine in front. Therefore, the required turbine
spacing depends on the variation in the flow direction. This data is compared
against flow measurements performed in a tidal stream, where the flow angle
varied 40 degrees depending on the direction of the tide (0 degrees variation
here would assume that the flow shifts 180 degrees as this angle is equivalent

86
0.45

Power coefficient
0.4
Spacing 1.5 D
Spacing 2 D
0.35
Spacing 3 D
Spacing 5 D
0.3 Spacing 7 D
Spacing 9 D
Single turbine
0.25
0 10 20 30 40 50 60 70 80
Angle of incident wind (Deg.)
Figure 5.23. Average power coefficients for array of 10 turbines with different spacing
and incidence angles (from figure 2, paper VI).

of 0 degrees for a line of vertical axis turbines). Here, the variation of the
incidence angle can be reduced to ±20 degrees and for these low angles, a low
turbine spacing is viable, which allows for many turbines in a small region of
space.

5.2.6 Simulations of control systems


The simulations in the previous sections have all been performed with constant
flow and rotational velocity. A real turbine will not experience a constant
flow velocity and it is the aim of the research at the Division of Electricity
at Uppsala University to use a variable speed rotor to obtain a higher power
extraction. Therefore, the three control strategies described in section 3.1 have
been simulated in paper VII using the aerodynamic vortex model coupled with
an electrical model that describes the generator, rectifier and extracts power
from the system through a DC load.
The constants in strategies A – C have all been chosen to make the control
strategies obtain λe = λmax at V = 6 m/s. The simulated turbine is similar to
the 200 kW turbine at Torsholm built by Vertical Wind AB [18], and has a di-
ameter of 26 m, height 24 m, chord 0.7 m and NACA0021 blades. All support
structures are omitted as the two-dimensional vortex model is not optimally
suited to simulate this. Even though this causes a too high C Pmax , the actual
value for the peak power coefficient is less important for the validity of the
control strategy, as the control strategies are calibrated to this simulated power
coefficient curve. All mechanical losses are also omitted.

87
2
Ref. A B C
1.8

1.6

1.4
λ/λmax

1.2

0.8

0.6
Windspeed (m/s)

10
6
3
0 100 200 300 400 500 600
Time (s)

Figure 5.24. Step response for the tested control strategies (figure 8, paper VII).

The power coefficient curve for this turbine is found in figures 3.3 – 3.5.
The control strategies are compared against a reference strategy, which strives
to keep the turbine at optimal tip speed ratio from preknowledge of the flow
velocity. This reference strategy can rapidly decrease the rotational velocity
by increasing the DC load, but is limited by the turbine torque for acceleration.
The first simulation is a step response where the flow velocity starts from
3 m/s, increases to 6 m/s, then up to 10 m/s, followed by a decrease back to
6 m/s and finally to 3 m/s. Aerodynamically, the step response is implemented
by a stepwise change of the asymptotic flow velocity. Figure 5.24 shows the tip
speed variation during the stepwise changes. Strategies B and C have higher
λe at 3 m/s and hence adapts faster to the increase to 6 m/s than strategy A.
The step up to 10 m/s is performed with approximately the same speed for all
three strategies A – C, although strategy B has a lower λ e than the other two.
Strategies A – C adapts approximately in the same time for both the increase
and decrease in flow velocity, while the reference strategy adapts much faster
for the step down. After the step change, there is a small amount of time where
the tip speed ratio has an overshoot and changes its value too much. This is an
aerodynamic effect as the wake is not stabilized directly after the step change.
Without a fully developed wake, the power output is temporary higher than
the steady state value, causing the increase in tip speed ratio, which is the
case for increasing flow velocities. A decreased flow velocity causes an over-
developed wake blocking the flow, hence reducing performance until it has
drifted away.

88
4
x 10
3

2.8

2.6
Delivered power (W)

2.4

2.2

Ref.
1.8
Strategy A
Strategy B
Strategy C
1.6
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04
Wind frequency (Hz)

Figure 5.25. Delivered power for oscillating wind field (figure 9, paper VII).

4.8

4.6

4.4
Tip speed ratio

4.2

3.8

3.6
Ref.
3.4 Strategy A
Strategy B
Strategy C
3.2
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04
Wind frequency (Hz)

Figure 5.26. Mean tip speed ratio for oscillating wind field (figure 10, paper VII).

89
A second study evaluates the control strategy performance for oscillating
wind conditions (figure 5.25), where the wind is set to a sinusoidal oscillation
between 3 m/s and 8 m/s. Among the three strategies A – C, strategy A has
the highest performance for steady wind conditions, which is expected as it
is the strategy that should operate closest to λ max . For unsteady conditions,
strategy C extracts more energy than strategy A and for even faster wind os-
cillations, strategy B is the best choice. The reference strategy works well for
wind frequencies below 0.02 Hz and later starts to drop rapidly. The explana-
tion is found in figure 5.26 where it is seen that the reference strategy drops in
tip speed ratio for higher wind frequencies, while strategies A – C obtains too
high tip speed ratios. The reference strategy has the ability to rapidly decrease
flow velocity, but is limited to aerodynamic torque for acceleration. When
the aerodynamic torque becomes too small to accelerate the turbine to proper
speed, the tip speed ratio drops. Strategies A – C have slightly higher accel-
erations than retardations when operating close to λ e with respect to changes
in flow velocity and hence end up at a too high tip speed ratio instead. This
higher tip speed ratio is advantageous as the turbine operates closer to λ max at
high flow velocities, which is where most energy is available.
According to the simulations, all strategies A – C should be viable options
for controlling the turbine and the best choice depends on the flow character-
istics. The reference strategy may obtain slightly higher performance in many
cases, but such a strategy would be complicated to implement as it would re-
quire full knowledge of the flow velocity through proper measurements, and
flow measurement equipment would also introduce additional sources of fail-
ure.

5.2.7 Control of multiple turbines


This section contains an evaluation of the two different electric topologies
described in section 3.2. The multiple turbine simulations with the separate
topology have been performed as four different electric simulations, while the
mutual topology simulations have been simulated as one combined electrical
simulation. Control strategy C was chosen for both topologies as it obtained
good results in the previous section. All tests were performed on a farm of
four turbines identical to the turbine tested in section 3.1.
The first test to compare the separate and mutual topologies is a simulation
where all turbines are aerodynamically separated. The flow velocity was var-
ied slowly according to figure 5.27. This velocity variation gives the rotational
velocities for the different topologies shown in figure 5.28. For the separate
topology, the different regions for strategy C is seen, where the decrease in
rotational velocity is reduced below V∞ = 4.5 m/s for turbine 4, while the in-
crease in rotational velocity is less for turbine 1 above V∞ = 9 m/s. Around
t = 1800 s for the mutual topology, turbine 4 stops delivering power and the

90
12

10

8
Flow velocity (m/s)

4
T1
T2
2
T3
T4
0
0 500 1000 1500 2000 2500 3000 3500
Time (s)

Figure 5.27. Flow velocities for independent turbines (figure 4, paper VIII).

4
Mutual
3.5 Separate

3
Rotational velocity (rad/s)

2.5

1.5

0.5

0
0 500 1000 1500 2000 2500 3000 3500
Time (s)

Figure 5.28. Turbine rotational velocities for independent turbines (figure 5, pa-
per VIII)

91
1.01

0.99
Turbine power
Indep power/mutual power

0.98 Delivered power

0.97

0.96

0.95

0.94

0.93

0.92
0 500 1000 1500 2000 2500 3000 3500
Time (s)

Figure 5.29. Ratio between power from mutual topology and separate topology (fig-
ure 7, paper VIII).

rotational velocity is determined by the electric losses. This is the point where
the difference between the two topologies is largest and the mutual topology
delivers slightly less than 93 % of the separate topology (figure 5.29). Beyond
this point, the other three turbines can accelerate more and the mutual strategy
obtains slightly better performance. Figure 5.29 also shows that the efficiency
of the electrical system is higher for the separate topology, mainly due to in-
creased core losses from the generators for the mutual topology due to higher
average rotational velocities.
A second test concerns how the different topologies behave if the differ-
ences in flow velocities are caused by the turbines mutual interaction. Tests of
four turbines in a square formation and on a line have been performed with the
turbine spacings of 3 and 6 diameters. The results in figure 5.30 show that the
difference is small between the strategies, indicating that if the flow velocity
differences are caused by turbines operating in each other’s wake, both topolo-
gies have similar performance and turbine spacing is a much more important
parameter. At the angle 45 degrees where the results differ, one of the turbines
on the diagonal is located in the wake of the other, while the two turbines on
the other diagonal are operating relatively independently. Hence, the turbines
operating independently cause an increase in rotational velocity, which is too
high for the turbine in the wake. The line configuration shows similar results
for both topologies, even at 90 degrees where three of the turbines operate
in the wake. These results show that as long as the turbines are connected
in groups where all groups have similar aerodynamic interactions within the
group, both topologies deliver nearly identical power, while for the 45 degrees

92
200

190

180
Delivered power (kW)
170

160

150

140

130
D=3, Mutual
120
D=6, Mutual
110 D=3, Separate
D=6, Separate
100
0 10 20 30 40 50 60 70 80 90
Angle of incident wind (Deg.)
Figure 5.30. Total delivered power for a square formation with four turbines (figure 11,
paper VIII).

case for the square configuration where two turbines are approximately inde-
pendent and two turbines are strongly coupled, the separate topology performs
better.
A final test was performed with vortices injected into the flow to cause a
fluctuating flow, which for the square formation is illustrated in figure 5.31.
The line configuration had a similar velocity field, although not exactly the
same due to the different turbine positions. This test was performed for the
3 diameter spacing with approximately 0 degrees incident flow angle. The
results show that for the line configuration, the separate topology delivered
1.7 % more power than the mutual topology between t = 200 s and 1400 s.
With 0 degrees incident angle, the line configuration approximately has inde-
pendent turbines and an improvement in performance is expected according
to the first test (see figure 5.29). The square configuration has two turbines
operating in the wake of the other two and for this configuration, the mutual
topology extracted 6.2 % between t = 200 s and 1400 s. These results show
that applying the separate topology with strategy C independently for each
turbine is not the best choice in case of strong aerodynamic coupling. The
mutual strategy, which kept a more constant rotational velocity, handles the
fluctuations in flow velocity better in this case.
All three tests show that the more independent the turbines are, the more
beneficial it becomes to use the separate topology, while for strongly con-
nected turbines, the mutual topology can be a viable option considering the
simplifications in the electric system.

93
15

10
Flow velocity (m/s)

T1
T2
T3
T4

0
200 400 600 800 1000 1200 1400
Time (s)

Figure 5.31. Fluctuating flow conditions for the square formation (figure 12, pa-
per VIII).

94
6. Conclusions

From the work included in this thesis, it can be concluded that:


1. It is possible to simulate a turbine with some accuracy. Even though the
simulations do not perfectly match experimental results, the simulation
codes give results that agree reasonably well with experimental values.
The simulation tools are therefore useful for designing vertical axis tur-
bines. An uncertainty is the accuracy in the experimental data available
for comparison.
2. If a turbine is designed to handle high flow velocities, the performance
of the turbine may decrease if additional struts are necessary to handle
the large forces.
3. The vertical axis turbine appears to be insensitive to the inflowing ve-
locity profile, and in some cases, it is possible that the turbine absorbs a
larger fraction of the available energy in the presence of a velocity pro-
file. For a horizontally aligned rotational axis, it is preferable to let the
blades move opposite to the flow direction at the side where the flow
velocity is highest.
4. If a turbine is located in a channel with constant flow, the power coef-
ficient increases with higher blockage ratio. A turbine will also reach
its peak performance at a higher tip speed ratio with higher blockage ra-
tio. The best turbine design is a function of the blockage ratio, and for
higher blockage ratios, the optimal design is likely to have larger chords
or more blades.
5. If several turbines are located in close proximity in a row configuration,
and the flow is perpendicular to the line intersecting the turbine centers,
simulations predict an increase in power production from the individual
turbines. If more turbines are used, placing all turbines in a row appears
to be better than placing them in a zigzag configuration. For the line con-
figuration, a larger spacing is favorable if the variation in flow direction
is large.
6. A control system that only measures the rotational velocity of the tur-
bine has been shown to be a viable control strategy for vertical axis tur-
bines. A cubic relation between extracted power and rotational velocity
gives good accuracy in obtaining peak power coefficient for the steady
state case. A reduction of the extracted power at low rotational veloc-
ities increases the response time of the control system and is therefore
beneficial for rapid flow variations.

95
7. Connecting all turbines through a mutual DC bus and only controlling
the total extracted power from a turbine farm can be a viable control
strategy. The differences between this system and a separate control of
each turbine are smaller in the coupled case than for aerodynamically
independent turbines.
8. The speed of the vortex simulations can be increased if graphics pro-
cessing units (GPU) are used to evaluate the velocity field. The increase
in computational speed is most prominent for large number of vortices.
Thus a GPU is well suited for simulations of large turbine farms.

96
7. Suggestions for future work

The simulation codes need better validation. One way would be to perform
force measurements on the blades of a turbine, which would give a much more
detailed description of the turbine than measurements of the power coefficient,
especially since it is the force that is calculated by the simulation codes.
The dynamic stall modeling should be updated to a more accurate model, as
the current model gives too low power coefficients at low tip speed ratios. The
current code uses the Gormont model, while a Beddoes Leishman model [55]
probably would be more accurate [3]. This would benefit both the vortex and
the streamtube model, since both models use the same dynamic stall model.
The expansion code in the streamtube method could be extended to include
the change in flow direction, similar to the model by Read and Sharpe [47].
This could make the streamtube results approach those of the vortex model.
The corrections due to tip vortices need to be studied further to determine
model uncertainties. This could either be performed by comparing with three-
dimensional simulations of turbines, or by measurements on a turbine with
constant diameter and chord, but with varying blade length.
To simulate tip effects, struts, wake structure etc. in more detail, a three-
dimensional vortex code should be developed. Such a model would probably
require extensive computational time and may therefore not be suited for de-
sign studies, but it could be used to create better correction models for the
two-dimensional vortex model and the streamtube code.
The current models are based on empirical data, thereby limiting the num-
ber of airfoils that can be conveniently modeled. A first principle model based
on Navier-Stokes equations would be more desirable. This requires a full so-
lution of the boundary layer which is computationally demanding. Combining
the vortex model with turbulence models or boundary layer theory could be a
way to achieve reasonably short computational times.

97
8. Summary of papers

This chapter contains a summary of all papers included in this thesis.

Paper I
A parameter study of the influence of struts on the performance of a
vertical-axis marine current turbine
A turbine for a marine current application is studied with focus on the losses
due to the struts. Hydrodynamically, a loss model for the struts is derived
within a streamtube model and empirical data is used for the drag coefficient
of the struts. Mechanically, FEM simulations are used to calculate the stresses
in the blades due to the hydrodynamic forces obtained from the streamtube
model, and when the stresses are too high, additional struts are added. Two
turbines are studies, one with 3 blades and one with 6 blades. The 6-bladed
turbine is predicted to obtain the highest power coefficient, but is also more
sensitive to the flow velocity as it needs more struts. It is shown that designing
the turbine for too high flow velocities will give lower power coefficient at
lower flow velocities due to the extra struts required. The author has been
responsible for the theory development, software development and most of the
hydrodynamic simulations as well as some FEM simulations and has written
more than half of the paper.

Paper II
Influence of a varying vertical velocity profile on turbine efficiency for a
vertical axis marine current turbine
Measurements on the velocity profile at three different locations in Swedish
rivers are performed and a streamtube model is used to simulate the perfor-
mance of a vertical axis turbine operating in these velocity profiles. The tradi-
tional configuration with the axis vertical is compared to the tilted version with
a horizontal axis. In this tilted case, both rotational directions are investigated.
These three cases are compared against the reference case with homogeneous
flow. Simulations show that the vertical axis turbine can obtain higher power
coefficients than the homogeneous case if the energy density is higher in the
center of the turbine, compared to the mean value. Simulations also show that

98
for the tilted turbine, having the blade moving against the flow at the high ve-
locity side of the turbine is beneficial at low tip speed ratios. The author has
been responsible for the theory development, software development and most
of the hydrodynamic simulations and writing more than half of the paper.

Paper III
Design of an experimental setup for hydro-kinetic energy conversion
The experimental setup that is intended to be located in Dalälven at Söder-
fors is covered in this paper. This begins with a characterization of the location
for the turbine including measurements of the velocity profile and data for the
predicted flow velocities. It also includes the electrical design of the generator
and how the generator and turbine efficiencies are coupled together. The au-
thor has been responsible for simulating and writing the hydrodynamic part in
the article, where a turbine is designed with the streamtube model to operate
at the given location including simulations of the power coefficient versus the
tip speed ratio, and how the maximum power coefficient is expected to vary
with flow velocity.

Paper IV
Simulations of a vertical axis turbine in a channel
A turbine located in a channel where there are boundaries confining the flow
is studied with a vortex model. The paper is intended to show both the benefits
of having the flow confined and the errors that will be obtained from simula-
tion models where confining boundaries are necessary. Flow confinement oc-
curs when volume discretization is used. The first part of the paper covers the
numerical aspects of performing channel simulations with the vortex model,
including the required number of revolutions required for convergence as a
function of channel width, which increases with increasing width. The second
part of the paper covers the improvement in performance due to the channel.
Here, the simulated power coefficient increases with decreasing width. The tip
speed ratio where the peak occurs also increases for a given turbine. The au-
thor has written the simulation software, performed all simulations and written
the article.

Paper V
Numerical simulation of a farm of vertical axis marine current turbines
Several turbines located in close proximity to each other are studied with
a vortex model. First, a two turbine system is simulated using all different

99
combinations of rotational directions and with different distances between the
turbines. The contribution from the rotational direction is found to be small,
and there is a slight increase in performance when the turbines are moved
closer to each other. Second, five turbine systems are studied. One case has all
turbines on a row and the other two cases use a zigzag pattern, where the last
case has larger turbines. With equal size of the turbines, the turbines on the row
performs slightly better if the flow comes straight forward, while more power
can be extracted with the zigzag pattern when the turbine radius is increased.
The row configuration is found to be less sensitive to small misalignments in
the flow direction. The author has written the simulation software, performed
all simulations and written the article.

Paper VI
Influence of incoming flow direction on spacing between vertical axis ma-
rine current turbines placed in a row
A line of ten turbines is simulated and the focus is on the turbine spacing
and how it affects the power production of the array with respect to different
incoming flow directions. The simulations are performed with a vortex model
and all turbines are simulated with constant rotational velocity independent on
flow direction. The article shows how the spacing between the turbines needs
to be increased for high power production if the direction of the incoming flow
is increased. Finally, an example with measurement data for the velocity in a
tidal flow is used and the required turbine spacing in this case is found to be
quite low. The author has written the simulation software and parts of the
article.

Paper VII
Robust VAWT control system evaluation by coupled aerodynamic and
electrical simulation
Three different control strategies, all based on the rotational velocity of the
turbine as only input parameter, are tested with respect to absorbed power
and response times to changes in flow velocity. The control strategies are
also compared against a reference strategy designed to operate at optimal tip
speed ratio. The strategies are evaluated with an aerodynamic vortex model
coupled with an electrical model. First, the different strategies are tested for a
step change in flow velocity and second, the behavior of the different control
strategies is studied under sinusoidal variations in wind velocity. The results
show that the power absorption for the three tested strategies is close to the
reference strategy. It is also shown that there is a trade off between accuracy
in obtaining the optimal tip speed ratio and the speed of the control strategy,

100
where the faster strategies perform better for the high frequency oscillations.
The author has been responsible for the aerodynamic part of the simulation
model and has written more than half of the article.

Paper VIII
Aerodynamic and electric evaluation of a VAWT farm control system with
passive rectifiers and mutual DC-bus
The control of a farm of four turbines is simulated using a coupled aero-
dynamic electrical simulation model. The article focuses on the difference
between controlling all turbines individually with the model described in pa-
per VII and connecting all turbines to a mutual DC bus by passive rectifiers.
The mutual DC bus limits the control system to set the extracted power for
the entire farm, instead of individually. For aerodynamically independent tur-
bines, less power is extracted if the flow velocities differ, but if the turbines
are aerodynamically coupled, the differences are smaller. A line and a square
configuration of the farm are tested and for simulations with varying incom-
ing flow velocity, the simulations give that if one turbine is in front of another,
the mutual DC bus can actually be better than individually optimizing each
turbine. The author has been responsible for the aerodynamic part of the sim-
ulation model and has written more than half of the article.

Paper IX
Adaptive fast multipole methods on the GPU
An implementation of the adaptive fast multipole method for a GPU using
CUDA is described. The implementation is two-dimensional and uses asym-
metric adaptivity. All individual parts of the algorithm have been ported to the
GPU and the speedup gained from using a GPU is investigated for all indi-
vidual parts of the algorithm. The comparison is made against an optimized
single threaded CPU version and the article shows a speedup of around 11
when the number of particles is large. The results also show that asymmetric
adaptivity performs well on the GPU as the non-uniform particle distributions
obtain a slightly higher speedup than the homogeneous distributions. The au-
thor has implemented the GPU version, performed many optimizations for the
single threaded CPU version and written more than half of the article.

101
9. Errata for papers

• In paper II, figure 4, the turbine diameter should be 6 m (instead of 8 m)


• In paper V, equation (7) should read
ΩR ∑ FT
CP = 1
,
2 ρ AV∞
3

i.e. the radius R should be added and the power coefficient should be
normalized against the asymptotic flow velocity

102
10. Acknowledgments

First of all, I would like to thank my supervisor Olov Ågren for his help with
writing the articles, this thesis and for all interesting discussions on many ar-
bitrary subjects. I would also like to thank my co-supervisor and head of this
division Mats Leijon for giving me the opportunity to work within this interest-
ing area of research and also Hans Bernhoff, for his support and for including
me in the wind power group.
I would also like to thank Staffan Lundin, with whom I been sharing office
with during most of my time here, for all his help, especially for proof-reading
several of my articles and for all his help with LATEX.
I like to thank Marcus Berg, Staffan Lundin, Mårten Grabbe, Jon Kjellin,
Mikael Bergkvist, Johan Bladh, Katarina Yuen, Emilia Lalander, Linnea Sjök-
vist, Fredrik Bülow, and Fredric Ottermo for proof-reading this thesis and all
the feedback you gave me.
I would also like to thank the rest of the members of the marine current
group, Karin, Johan, Nicole and also Anders N, for all your help and interest-
ing discussions
A special thanks goes also to Jon Kjellin for providing the experimental
results used to compare the simulations with in section 5.1. Also included is
the rest of the wind power group, Senad, Eduard, Stefan, Jon O, Per, Sandra
and former member Kristian for all your support and also the former mem-
bers Paul, Mathias and David, who made the foundations for the aerodynamic
research at the division.
The rest of the members of the Division of Electricity for being very nice
co-workers.
My family and all my friends, thanks for all the support you have given me,
both before, and during my time at the division.
Final thanks go to all financiers of this work, especially Uppsala University
for my salary, but also the rest of the financiers: The Swedish Research Coun-
cil, The Swedish Energy Agency, Vinnova, Ångpanneföreningen’s Foundation
for Research and Development, The J. Gust. Richert Foundation for Technical
Scientific Research and CF’s Environmental Fund.

103
11. Summary in Swedish

Vid generering av elektricitet från flödande medium, t.ex. vind eller ström-
mande vatten, används traditionellt horisontalaxlade propellerliknande turbi-
ner. Ett alternativ till dessa är den vertikalaxlade turbinen, som har fördelen
att den är oberoende av flödesriktning, samt att den vertikala axeln möjliggör
att generatorn kan placeras på marken. Aerodynamiskt kännetecknas vertika-
laxlade turbiner av att bladen, som rör sig runt i en cirkel, passerar flödet två
gånger per varv. Det medför att varje blad genererar vridmoment två gånger
under varvet. Däremellan finns två punkter där kraften är nästan noll, vilket
ger pulserande krafter.
När aerodynamiken simuleras för de vertikalaxlade turbinerna används ge-
nerellt sett tre olika metoder, nämligen strömrörsmodeller, virvelmodeller och
finita element-/volyms-metoden där endast de två förstnämnda har används i
denna avhandling. Av dessa är strömrörsmodellen den snabbaste då den inte
innehåller något tidsberoende flöde samt att hela hastighetsfältet runt turbi-
nen inte beräknas, utan endast approximeras vid turbinbladen. I denna snab-
ba modell kan turbineproblemet lösas tredimensionellt. Det möjliggör studier
av varierande hastighet över turbinarean och inkluderande av bärarmar, vilket
modellen har används till i denna avhandling.
I fallet med varierande flödeshastighet undersöktes hur turbinens effekti-
vitet påverkas av att flödeshastigheten inte är konstant över turbinarean. En
mer realistisk fördelning med lägre hastighet närmare botten användes. Enligt
simuleringarna ska den vertikalaxlade turbinen fungera bra när flödeshastig-
hetens variation har tagits som de uppmätta värdena från två svenska älvar.
Simuleringarna visar också att om flödeshastigheten varierar vinkelrätt med
turbinens rotationsaxel (vilket är fallet för liggande turbin) så bör turbinen bli
effektivare om bladen rör sig motströms där hastigheten är som högst.
För en strömkraftsturbin studerades hur många bärarmar som är nödvändi-
ga för att turbinen ska hålla vid olika flödeshastigheter. Detta kopplades sedan
till de hydrodynamiska förlusterna som bärarmarna ger upphov till. Från des-
sa resultat kan man se hur turbinens effektivitet sjunker när fler bärarmar är
nödvändiga. Detta leder till att om en turbin designas för att hålla för relativt
höga flödeshastigheter tappas effektivitet vid de lägre hastigheterna. I sådana
fall kan det vara bättre att konstruera turbinen för en lägre flödeshastighet och
stänga av den i de fåtal fall då flödeshastigheten är allt för hög.
Strömrörsmodellen har begränsningar, t.ex. hur flödet modelleras, och fun-
gerar inte för alla problem. En mer avancerad metod är virvelmetoden som
löser hela flödesfältet tidsberoende, vilket ger en mer fullständig beskrivning

104
av turbinen. Virvelmetoderna är mer tidskrävande än strömrörsmodellen, och
endast tvådimensionella simuleringar har genomförts i denna avhandling.
Med virvelmetoden har det studerats vilka effekter som uppkommer när
flödet kring en turbin innesluts av kanalväggar, vilket är fallet för t.ex. flo-
der. Studien börjar med att undersöka hur snabbt simuleringarna konvergerar
och resultatet är att man behöver simulera ett större antal turbinvarv för bre-
da kanaler innan konvergens uppnås. Sen studerades turbinen prestanda och
för kanaler ökar den erhållna effekten från turbinen när kanalens bredd mins-
kar. Detta stämmer med strömrörsteori som förutspår en effektökning, men i
simuleringarna ökar effekten inte lika snabbt som teorin förutspår. Samtidigt
kan noteras att turbinen erhåller maximal effekt vid högre löptal när kanalens
bredd minskar.
Virvelmetoden har även använts för att studera hur flera turbiner växelver-
kar med varandra. Enligt simuleringarna kan man erhålla högre effekt hos en-
skilda turbiner om de placeras nära varandra och flödet kommer rakt framifrån
(vinkelrätt mot linjen som skär turbinernas centrum). En studie genomfördes
på fem turbiner, där placering i rad eller i sicksackmönster jämfördes. När flö-
det kom rakt framifrån gav radgeometrin aningen bättre prestanda. Däremot
var rad-geometrin väsentligt bättre vid 30 graders infallsvinkel. Utifrån dessa
simuleringar kan slutsatsen dras att rad-geometrin är bättre om turbinerna är li-
ka stora. Däremot finns utrymme för flera turbiner i ett sicksackmönster, vilket
kan vara en fördel om tillgänglig yta för turbinplacering är begränsad. Utifrån
dessa resultat undersöktes rad-geometrin extra för en rad av 10 turbiner och
hur vinkeln på flödet relativt linjens riktning påverkar prestandan. Resultatet
blev att avståndet mellan turbinerna i raden behöver ökas ifall vinkeln på flödet
kan bli stor.
Virvelmetoden kombinerades även med en elektrisk modell och denna
kopplade simuleringsmodell användes för att studera kontrollstrategier för hur
man ska styra en turbin. Tre olika strategier, som alla endast använde rotations-
hastigheten som indata och styrde genom att reglera uttagen effekt, undersök-
tes och dessa strategier lyckades styra turbinen väl i avseendet att erhålla en
hög elektrisk effekt från systemet. Resultatet blev även att de strategier som
tog ut lite lägre effekt vid låga varvtal snabbare ställde in sig på angivet varvtal
och dessa kunde därmed bli fördelaktiga vid snabba förändringar i flödeshas-
tigheten.
Simuleringarna av kontrollsystem utökades sedan till att involvera en park
av 4 turbiner. Här studerades skillnaden mellan att styra varje turbin enskilt
enligt ovanstående modell, och att med passiva likriktare koppla ihop elsyste-
men via en gemensam DC-buss, vilket gör att man endast kan styra den totala
effekten från hela parken. För aerodynamiskt åtskilda turbiner var det bättre
att styra varje turbin enskilt om det var stora variationer i hastighet mellan
turbinerna. Om hastighetsvariationerna istället orsakades av att vissa turbiner
var placerade bakom andra, som då redan absorberat större delen av energin,
så blev skillnaderna mellan de olika systemen mindre. Detta visar att använ-

105
dandet av en en gemensam DC-buss kan vara ett alternativ, särskilt då det
reducerar antalet elektriska komponenter.
En studie genomfördes även på om man kan använda grafikkort för att
öka hastigheten vid beräkningen av flödeshastigheten för virvlarna, då detta
är den mest tidskrävande delen av virvelmetoden. I studien implementerades
hela snabba multipolmetoden (FMM) för grafikkortet och beräkningstiderna
jämfördes med en seriell implementation på en vanlig processor. Resultatet
visar att hastighetsökningen när man använder grafikkort är störst för ett stort
antal virvlar. Därför är denna metod lämplig för tunga beräkningar, som t.ex.
simulering av en stor park.

106
References

[1] J. F. Manwell, J. G. McGowan, and A. L. Rogers. Wind Energy Explained:


Theory, Design and Application. John Wiley & Sons, 2009.
[2] S. J. Savonius. Rotor adapted to be driven by wind or flowing water. U.S. Patent
No. 1,697,574, January 1929.
[3] I. Paraschivoiu. Wind turbine design with emphasis on Darrieus concept.
Polytechnic International Press, 2002.
[4] G. J. M. Darrieus. Turbine having its rotating shaft transverse to the flow of the
current. U.S. patent No. 1,835,018, December 1931.
[5] P. Gipe. Wind Energy Basics: A Guide to Home- And Community-Scale Wind
Energy Systems. Chelsea Green Publishing Company, 2009.
[6] S. Eriksson, H. Bernhoff, and M. Leijon. Evaluation of different turbine
concepts for wind power. Renewable & Sustainable Energy Reviews,
12(5):1419–1434, June 2008.
[7] F. M. White. Fluid Mechanics. McGraw-Hill Series in Mechanical
Engineering. McGraw-Hill, 2008.
[8] G. Kuiper. Cavitation research and ship propeller design. Applied Scientific
Research, 58:33–50, 1997. 10.1023/A:1000754928962.
[9] M. Shiono, K. Suzuki, and S. Kiho. Output characteristics of Darrieus water
turbine with helical blades for tidal current generations. In Proceedings of the
Twelfth (2002) International Offshore and Polar Engineering Conference,
Kitatkyushu, Japan, May 2002.
[10] T. Maître, J.-L. Achard, L. Guittet, and C. Ploesteanu. Marine turbine
development: numerical and experimental investigations. Scientific Bulletin of
the Polytechnic University of Timisoara: Transactions on Mechanics,
50(64):59–66, 2005.
[11] D. P. Coiro, A. De Marco, F. Nicolosi, S. Melone, and F. Montella. Dynamic
behaviour of the patented Kobold tidal turbine: numerical and experimental
aspects. Acta Polytechnica, 45(3):77–84, 2005.
[12] B. Kirke and L. Lazauskas. Variable pitch Darrieus water turbines. Journal of
Fluid Science and Technology, 3(3):430–438, 2008.
[13] C. T. Crowe, D. F. Elger, J. A. Roberson, and B. C. Williams. Engineering
Fluid Mechanics, 9th Edition Binder Ready. John Wiley & Sons, 2008.
[14] A. Solum, P. Deglaire, S. Eriksson, M. Stålberg, M. Leijon, and H. Bernhoff.
Design of a 12kW vertical axis wind turbine equipped with a direct driven PM
synchronous generator. In Proceedings of EWEC 2006, European Wind Energy
Conference & Exhibition, 2006.
[15] P. Deglaire, S. Eriksson, J. Kjellin, and H. Bernhoff. Experimental results from
a 12 kW vertical axis wind turbine with a direct driven PM synchronous
generator. In Proceedings of EWEC 2007, European Wind Energy Conference
& Exhibition, Milan, Italy, 2007.

107
[16] J. Kjellin, F. Bülow, S. Eriksson, P. Deglaire, M. Leijon, and H. Bernhoff.
Power coefficient measurement on a 12 kW straight bladed vertical axis wind
turbine. Renewable Energy, 36(11):3050 – 3053, 2011.
[17] F. Bülow, J. Kjellin, S. Eriksson, M. Bergkvist, P. Ström, and H. Bernhoff.
Adapting a VAWT with PM generator to telecom applications. In Proceedings
of the European Wind Energy Conference & Exhibition 2010, April 2010.
[18] S. Eriksson, H. Bernhoff, and M. Leijon. A 225 kW direct driven PM generator
adapted to a vertical axis wind turbine. Advances in Power Electronics, 2011,
2011. Article ID 239061.
[19] P. Deglaire, O. Ågren, H. Bernhoff, and M. Leijon. Conformal mapping and
efficient boundary element method without boundary elements for fast vortex
particle simulations. European Journal of Mechanics, B/Fluids, 27(2):150–176,
April 2008.
[20] P. Deglaire, S. Engblom, O. Ågren, and H. Bernhoff. Analytical solutions for a
single blade in vertical axis turbine motion in two-dimensions. European
Journal of Mechanics, B/Fluids, 28:506–520, 2009.
[21] P. Deglaire. Analytical Aerodynamic Simulation Tools for Vertical Axis Wind
Turbines. PhD thesis, Uppsala University, Electricity, 2010.
[22] D. Österberg. Multi-body unsteady aerodynamics in 2D applied to a
vertical-axis wind turbine using a vortex method. Master’s thesis, Uppsala
University, 2010.
[23] S. Engblom. On well-separated sets and fast multipole methods. Applied
Numerical Mathematics, 61(10):1096 – 1102, October 2011.
[24] A. Betz. Das maximum der theoretisch möglichen ausnutzung des windes durch
windmotoren. Zeitschrift für das gesamte Turbinenwesen, September 1920.
[25] C. Garrett and P. Cummins. The efficiency of a turbine in a tidal channel.
Journal of Fluid Mechanics, 588:243–251, October 2007.
[26] J. I. Whelan, J. M. R. Graham, and J. Peiró. A free-surface and blockage
correction for tidal turbines. Journal of Fluid Mechanics, 624:281–291, 2009.
[27] I. H. Abbott and A. E. V. Doenhoff. Theory of wing sections: including a
summary of airfoil data. Dover books on physics and chemistry. Dover
Publications, 1959.
[28] J. H. Laks, L. Y. Pao, and A. D. Wright. Control of wind turbines: Past, present,
and future. In American Control Conference, 2009. ACC ’09., pages 2096
–2103, June 2009.
[29] L. Lazauskas. Three pitch control systems for vertical axis wind turbines
compared. Wind Engineering, 16(5):269–282, 1992.
[30] B. K. Kirke and L. Lazauskas. Limitations of fixed pitch Darrieus hydrokinetic
turbines and the challenge of variable pitch. Renewable Energy, 36:893–897,
2011.
[31] E. A. Bossanyi. The design of closed loop controllers for wind turbines. Wind
Energy, 3(3):149–163, 2000.
[32] M. Dahlgren, H. Frank, M. Leijon, F. Owman, and L. Walfridsson.
WindformerTM. Wind power goes large-scale. ABB Review, pages 31–37, 2000.
[33] S. Lundberg. Performance comparison of wind park configurations. Technical
report, Chalmers University of Technology, 2003. Technical Report No. 30R.
[34] D. Jovcic and N. Strachan. Offshore wind farm with centralised power

108
conversion and DC interconnection. Generation, Transmission Distribution,
IET, 3(6):586 –595, June 2009.
[35] M. Parker and O. Anaya-Lara. An evaluation of collection network designs
which eliminate the turbine converter. In EWEA 2012 Scientific Proceedings,
pages 32–36, April 2012. OMAE2012-83347.
[36] A. Kusiak and Z. Song. Design of wind farm layout for maximum wind energy
capture. Renewable Energy, 35(3):685 – 694, 2010.
[37] A. Crespo, J. Hernández, and S. Frandsen. Survey of modelling methods for
wind turbine wakes and wind farms. Wind Energy, 2(1):1–24, 1999.
[38] M. Soleimanzadeh and R. Wisniewski. Controller design for a wind farm,
considering both power and load aspects. Mechatronics, 21(4):720 – 727, 2011.
[39] J. Shu, B. H. Zhang, and Z. Q. Bo. A wind farm coordinated controller for
power optimization. In Power and Energy Society General Meeting, 2011 IEEE,
pages 1 –8, july 2011.
[40] T. Knudsen, T. Bak, and M. Soltani. Prediction models for wind speed at
turbine locations in a wind farm. Wind Energy, 14(7):877–894, 2011.
[41] R. Howell, N. Quin, J. Edwards, and N. Durrani. Wind tunnel and numerical
study of a small vertical axis wind turbine. Renewable Energy, 35(2):412–422,
February 2010.
[42] O. Guerri, A. Sakout, and K. Bouhadef. Simulations of the fluid flow around a
rotating vertical axis wind turbine. Wind Engineering, 31(3):149–163, May
2007.
[43] R. Templin. Aerodynamic performance theory for the NRC vertical-axis wind
turbine. Technical report, National Research Council of Canada, 1974.
Laboratory technical report – LTR-LA-160.
[44] E. E. Lapin. Theoretical performance of vertical axis wind turbines. November
1975. ASME Paper 75-WA/ENER-1.
[45] J. H. Strickland. The Darrieus turbine: A performance prediction model using
multiple streamtubes. Technical Report SAND75-0431, Sandia National
Laboratories, Albuquerque, New Mexico, October 1975.
[46] I. Paraschivoiu. Double-multiple streamtube model for Darrieus wind turbines.
In Second DOE/NASA wind turbines dynamics workshop, pages 19–25,
Cleveland, OH, February 1981. NASA CP-2186.
[47] S. Read and D. J. Sharpe. An extended multiple streamtube theory for vertical
axis wind turbines. 2nd BWEA workshop, Cranfield, UK, pages 65–72, April
1980.
[48] D. J. Sharpe. Refinements and developments of the multiple streamtube theory
for the aerodynamic performance of vertical axis wind turbines. In Proceedings
of the BWEA Wind Energy Conference, pages 146–159, 1984.
[49] W. A. Moran. Giromill wind tunnel test and analysis. Volume 2. Technical
discussion. Technical Report C00/2617-4/2, U.S. Energy Research and
Development Administration, October 1977.
[50] R. E. Sheldahl and P. C. Klimas. Aerodynamic characteristics of seven
symmetrical airfoil sections through 180-degree angle of attack for use in
aerodynamic analysis of vertical axis wind turbines. Technical Report
SAND80-2114, Sandia National Laboratories, Albuquerque, New Mexico,
March 1981.

109
[51] R. E. Gormont. A mathematical model of unsteady aerodynamics and radial
flow for application to helicopter rotors. Technical report, USAAV Labs., May
1973. TR 72-67.
[52] B. Massé. Description de deux programmes d’ordinateur pour le calcul des
performances et des charges aerodynamiques pour des eoliennes a’axe vertical.
Report IREQ 2379, Institut de Recherche de L’Hydro–Quebec, Varennes,
Quebec, July 1981.
[53] D. E. Berg. Improved double-multiple streamtube model for the Darrieus-type
vertical axis wind turbine. In Presented at the Am. Solar Energy Soc. Meeting,
Minneapolis, 1 Jun. 1983, volume 1, pages 231–238, June 1983.
[54] J. G. Leishman and T. S. Beddoes. A generalised model for airfoil unsteady
aerodynamic behaviour and dynamic stall using the indical method. In
Proceedings from the 42nd Annual Forum of the American Helicopter Society,
pages 243–266, 1986.
[55] J. G. Leishman and T. S. Beddoes. A semi-empirical model for dynamic stall. J.
Am. Helicopter Society, 34(3), July 1989.
[56] I. Paraschivoiu, P. Fraunié, and C. Béguier. Streamtube expansion effects on the
Darrieus wind turbine. Journal of Propulsion and Power, 1:150–155,
March-April 1985.
[57] G. H. Cottet and P. D. Koumoutsakos. Vortex Methods: Theory and Practice.
Cambridge University Press, 2008.
[58] A. J. Chorin. Numerical study of slightly viscous flow. J. Fluid. Mech,
57:785–796, 1973.
[59] S. Shankar and L. van Dommelen. A new diffusion procedure for vortex
methods. J. Comput. Phys., 127:88–109, August 1996.
[60] J. H. Strickland, B. T. Webster, and T. Nguyen. A vortex model of the Darrieus
turbine: An analytical and experimental study. Journal of Fluids Engineering,
101:500–505, December 1979.
[61] P. Ramachandran, S. C. Rajan, and M. Ramakrishna. A fast, two-dimensional
panel method. SIAM Journal on Scientific Computing, 24(6):1864–1878, 2003.
[62] L. Greengard. Potential flow in channels. SIAM Journal on Scientific and
Statistical Computing, 11(4):603–620, July 1990.
[63] L. Greengard and V. Rokhlin. A fast algorithm for particle simulations. Journal
of Computational Physics, 73:325–348, December 1987.
[64] T. Hrycak and V. Rokhlin. An improved fast multipole algorithm for potential
fields. SIAM Journal on Scientific Computing, 19(6):1804–1826, 1998.
[65] R. Farber. CUDA Application Design and Development. Morgan Kaufmann
Publishers Inc., San Francisco, CA, USA, 1st edition, 2011.
[66] S. K. Raman, V. Pentkovski, and J. Keshava. Implementing streaming SIMD
extensions on the pentium III processor. Micro, IEEE, 20(4):47 –57, jul/aug
2000.
[67] J. Sanders and E. Kandrot. CUDA by Example: An Introduction to
General-Purpose GPU Programming. Addison-Wesley Professional, 1st
edition, 2010.
[68] M. Drela. XFOIL: an analysis and design system for low Reynolds number
airfoils. In Conference on low Reynolds number Aerodynamics. University of
Notre Dame, 1989.

110
[69] G. F. Homicz. Numerical simulation of VAWT stochastic aerodynamic loads
produced by atmospheric turbulence: VAWT-SAL code. Technical Report
SAND91-1124 UC-261, Sandia National Laboratories, Albuquerque, New
Mexico, September 1991.
[70] I. Paraschivoiu. Aerodynamic models and experiments for studying Darrieus
wind turbines. In Proceedings of the EWEC, pages 617–622, Herning,
Denmark, 1988.

111
Acta Universitatis Upsaliensis
Digital Comprehensive Summaries of Uppsala Dissertations
from the Faculty of Science and Technology 998
Editor: The Dean of the Faculty of Science and Technology

A doctoral dissertation from the Faculty of Science and


Technology, Uppsala University, is usually a summary of a
number of papers. A few copies of the complete dissertation
are kept at major Swedish research libraries, while the
summary alone is distributed internationally through
the series Digital Comprehensive Summaries of Uppsala
Dissertations from the Faculty of Science and Technology.

ACTA
UNIVERSITATIS
UPSALIENSIS
Distribution: publications.uu.se UPPSALA
urn:nbn:se:uu:diva-183794 2012

Anda mungkin juga menyukai