Beruflich Dokumente
Kultur Dokumente
A Thesis
Presented to
The Academic Faculty
By
Guanpeng Xu
In Partial Fulfillment
of the Requirements for the Degree
Doctor of Philosophy in Aerospace Engineering
Approved:
____________________________________
Lakshmi N. Sankar, Chairman
____________________________________
Dewey H. Hodges
____________________________________
Stephen M. Ruffin
ii
Dedicated to
my wife Hongli
iii
ACKNOWLEDGMENTS
iv
TABLE OF CONTENTS
DEDICATION
III
ACKNOWLEDGMENTS
IV
TABLE OF CONTENTS
VI
LIST OF TABLES
IX
LIST OF FIGURES
NOMENCLATURE
XIV
SUMMARY
XVIII
INTRODUCTION
1.1.
Background
1.2.
1.3.
Research Objectives
11
1.4.
12
1.5.
14
15
15
2.1.2
20
26
28
vi
2.1.4
2.1.5
2.2
2.3
2.4
3
29
30
31
33
35
36
37
Temporal Discretization
39
41
2.2.1
Temporal Discretization
44
2.2.2
Spatial Discretization
45
48
2.3.1
49
2.3.2
50
52
55
3.1
56
3.2
59
62
4.1 Introduction
62
4.2
64
4.2.1
64
4.2.2
5
Grid hierarchy
66
68
68
70
77
82
90
95
97
101
106
6.1 Conclusions
106
6.2 Recommendations
108
110
116
119
125
144
VITA
151
viii
LIST OF TABLES
Table
Page
5-1.
A-1.
111
A-2.
112
A-3.
113
A-4.
114
E-1.
141
ix
96
LIST OF FIGURES
Figure
Page
15.SketchfortheHybridMethodology
21.CellFacesandtheControlVolume
26
22.ThreePointStencil
29
23.FixedFivePointStencil
30
24.PartitioningofComputationalDomainforHOandHHTopologies
49
25.SketchforNumericalBoundaries
53
31.SketchfortheHybridMethodology
55
32.DistancebetweenaWakeElementandaPoint
58
33.EffectofInplaneVelocity
59
34.EffectoftheSkewnessofTipVortexes
60
41.SchematicofDownwindHorizontalAxisWindTubine
62
42.SchematicofaSimpleOversetGridSystem
64
43.BasicGridHierarchyforTowerShadowEffects
66
51.BodyfittedgridoverPhaseIIRotor
69
x
52.Computedvs.MeasuredPowerGenerationforPhaseIIrotor
71
53.Computedvs.MeasuredPowerGenerationforPhaseIIIrotor
72
54.SpanwiseloadingoverPhaseIIRotorwithcomparisonswithYAWDYN 73
55.EvolutionofPowerwithIterationsforPhaseIIIRotor
75
56.StreamlinesoverRotorUpperSurfaceat10m/s
76
57.StreamlinesoverRotorUppersurfaceat20m/s
77
5-8. Transition Lines on Upper Surface of the Phase III rotor in 6 m/s
77
5-9. Transition Lines on Lower Surface of the Phase III rotor in 6 m/s
78
5-10. Transition Lines on Lower Surface of the Phase III rotor in 8 m/s
79
5-11. Transition Lines on Upper Surface of the Phase III rotor in 8 m/s
80
5-12. Power Generation vs. Wind Speed for the Phase III Rotor
81
5-13. Power Generation vs. Wind Speed for the Phase III Rotor
82
83
83
85
85
86
89
91
92
xi
93
94
94
95
97
98
99
99
100
5-31. Results from YawDyn Code for 10m/s Wind and 20 Yaw
101
102
5-33. Tower Shadow for the NREL CER Turbine in 10m/s Wind
103
104
115
C1.IdealizedflowthroughanActuatorDisk
119
C2.AStreamtubeinStripTheory
121
C3.FlowVelocityDiagramataStreamtube
122
D1.CharacteristicsofthePrandtlsTipLossFunction
126
D2.TheTipLossFactorforthePhaseVIRotorbyStripTheory
128
D3.PrandtlsTipLossFactorforthePhaseVIRotorat25m/sWind
130
D4.TheVariationattheTipfromFigureD3
130
xii
D5.ThePrandtlsTipLossFactorforthePhaseVIRotorat20m/s
131
D6.ThePrandtlsTipLossFactorforthePhaseVIRotorat15m/s
131
D7.ThePrandtlsTipLossFactorforthePhaseVIRotorat7m/s
132
D8.SketchforInducedVelocitiesofaSectionofTipVortex
133
D9.CurvefittingaNewTipLossModelfromCFDfor7m/sWind
136
D10.LowSpeedShaftTorqueforthePhaseVIRotor
136
D11.PowerCurveforthePhaseIIIRotor
137
E1.CylindricalCoordinateSystemusedbyBanksandGadd
140
E2.StallDelayCorrectionto2DLiftCoefficients
141
E3.EffectsoftheConstantninCorrigansModelonLowSpeed
ShaftTorqueforthePhaseIVRotor
142
E4.EffectsoftheConstantninCorrigansModelonGeneratedPower
forthePhaseIIIRotor
143
xiii
NOMENCLATURE
Speed of sound
A, B, C
Cl
Cn
Cp
Pressure coefficient
CT
r
d l
~
e
E, F, G
FL, FR
Fk
imatch
[I]
Identity matrix
Jacobian of transformation
kmatch
xiv
lm
Mixing length
Mach number
Mtip
r
n
Pressure
Pr
Prandtl number
PrT
{q}
{qL}, {qR}
Blade radius
Re
Reynolds number
RHS
Shorthand notation for the terms on the right hand side for any
equation
Time
Temperature
T, T, T
Eigenvector matrices
u, v, w
xv
uw, vw, ww
U, V, W
r
Vw
r
VGrid
Grid velocity
V wind
Wind speed
x, y, z
Cartesian coordinates
Greek Symbols
Angle of attack
Eddy viscosity
xvi
Density
Velocity potential
Azimuthal angle
ij
Shear stresses
Vorticity
Subscripts
FP
i, j, k
grid indices
NS
Turbulent
x, y, z, t,
Freestream condition
Superscripts
n, n+1 Time level
xvii
SUMMARY
A numerical technique has been developed for efficiently simulating fully threedimensional viscous fluid flow around horizontal axis wind turbines (HAWT) using a
zonal approach. The flow field is viewed as a combination of viscous regions, inviscid
regions and vortices. The method solves the costly unsteady Reynolds averaged NavierStokes (RANS) equations only in the viscous region around the turbine blades. It solves
the full potential equation in the inviscid region where flow is irrotational and isentropic.
The tip vortices are simulated using a Lagrangean approach, thusremovingtheneedto
accuratelyresolvethemonafinegrid.Thehybridmethodisshowntoprovidegood
resultswithmodestCPUresources.
A full Navier-Stokes based methodology has also been developed for modeling
wind turbines at high wind conditions where extensive stall may occur. An overset grid
based version that can model rotor-tower interactions has been developed. Finally, a
blade element theory based methodology has been developed for the purpose of
developing improved tip loss models and stall delay models.
The effects of turbulence are simulated using a zero equation eddy viscosity
model, or a one equation Spalart-Allmaras model. Two transition models, one based on
the Epplers criterion, and the other based on Michels criterion, have been developed and
tested.
The hybrid method has been extensively validated for axial wind conditions for
three rotors NREL Phase II, Phase III, and Phase VI configurations. A limited set of
xviii
calculations has been done for rotors operating under yaw conditions.
Preliminary
simulations have also been carried out to assess the effects of the tower wake on the rotor.
In most of these cases, satisfactory agreement has been obtained with measurements.
Usingthenumericalresultsfrompresentmethodologiesasaguide,Prandtlstip
lossmodelandCorrigansstalldelaymodelwerecorrelatedwithpresentcalculations.
Animprovedtiplossmodelhasbeenobtained.AcorrectiontotheCorrigansstalldelay
model has also been developed. Incorporation of these corrections is shown to
considerablyimprovepowerpredictions,evenwhenaverysimpleaerodynamictheory
bladeelementmethodwithannularinflowisused.
xix
CHAPTER I
INTRODUCTION
1.1 Background
Wind energy is one of the cleanest sources of energy. Recent advances in airfoil
and rotor development, materials technology, power generation systems, and
manufacturing technology have made wind turbine systems in general, and horizontal
axis wind turbine (HAWT) systems in particular, economically feasible alternatives to
gas, oil, and coal based power generation systems. Brower et al. 1 and Cavallo2 discuss
the technological and economic aspects of wind energy.
Modern wind turbines have a long history which can be traced back to 200 B. C. 3
Simple, vertical-axis panemones were used for grinding grain in Persia at that time.
Primitive types of horizontal-axis windmills (consisting of up to ten wooden booms,
rigged with jib sails) were later developed, and are still in use in many Mediterranean
regions. By 1100 A. D., windmills had been in extensive use in the Middle East and were
introduced in Europe during the thirteenth century by the returning Crusaders.
After their introduction in Europe in the thirteenth century, windmills evolved
gradually in design and use. Windmills flourished in Holland beginning in the fourteenth
century, and began to be used on the American continent and most other parts of the
world in the middle nineteenth century for pumping water and generating electricity.
xx
Engineers in the United States and Europe, Denmark in particular, have greatly
contributed to the evolution of the windmill. Danish manufacturers had a 50% share of
the world market for wind turbines in 2000. Danish Wind Turbine Manufacturers
Association
provides
very
informative
web-site
for
wind
energy
at
www.windpower.org.4
Modern wind turbines were developed before World War II. In the aspect of
aerodynamics, the difference between ancient windmills and modern wind turbines is
whether the lift from the blades or the drag from the sails is used for mechanical power.
Public interest in wind energy has fluctuated with the availability of other energy
sources.
For example, wind energy was the most important energy supplement in
Denmark when it suffered an energy crisis due to the oil embargo in WWII. On the other
hand, the decline in the American windmills occurred in 1930s due to the wide spread
availability of power lines throughout the American heartland providing convenient
access to a large and ready supply of electricity. Currently the environmental problems
caused by coal and fossil fuel burning, and the limited supply and increasing cost of such
fossil fuel around the world, have rekindled interest in wind energy.
The leading
laboratory in wind technology development in the United States, the National Renewable
Energy Laboratory (NREL), funded by the Department of Energy (DOE), operates the
National Wind Technology Center and manages turbine research programs and applied
research activities.5
Wind turbines have already become economically competitive throughout the
world compared to other kinds of methods of generating electricity.6 More wind turbines
xxi
electricity are produced by wind turbines each year in the United States --- enough
electricity for a city the size of San Francisco. Figure 1-1 shows the falling cost of
electricity from wind turbines over the last twenty years.
xxii
wind resources in the United States and around the World has been carried out under the
Federal Wind Energy Program. Complete wind resource maps can be found at NRELs
wind resource database, at http://wind.nrel.gov/wind/database.html.
A horizontal axis wind turbine (HAWT) converts wind energy into mechanical
energy and electricity using one or more blades which turn a drivetrain, connected to a
generator. A turbine blade, like a fixed wing, generally uses a high lift airfoil as the blade
section. One end of the blade is connected to a hub with a low speed shaft. The wind
impinges at the airfoil lower surface, in the direction perpendicular to the rotor disk as
xxiii
viewed from ground. The airfoil section at a typical radial location r travels at a speed
r, where is the turbine angular velocity. The local air impinging velocity for a typical
airfoil is the combination of this velocity r, wind velocity, and induced velocity from
the bound circulation and the tip vortex wake, as explained later. As shown in Figure 1-3,
the resultant overall aerodynamic force is tilted so that there is a component in the
direction of travel of the airfoil. This aerodynamic force may be viewed as induced thrust
and pushes the blade forward in the direction of rotation. The resulting torque may be
converted into electrical power through the drivetrain/generator mechanism.
Wind
Figure 1-3. Mechanism for Positive Power Output from Wind Turbine
.
xxiv
The assembly of blades and their hubs form a turbine rotor. While single-bladed
rotors minimize energy loss from friction drag forces, two- or three-bladed rotors are
considered to yield the best trade-off for stability, aerodynamic performance, and cost.
Figure 1-4a shows typical components of a horizontal axis wind turbine.
xxv
xxvi
Many of the rotors found on current generation HAWT systems are designed
using a combination of 2-D airfoil tools (Tangler10, Eppler11, and Selig12) and 3-D blade
element and momentum (BEM) theory (Hansen, 13 Laino14). A number of comprehensive
computer codes using this methodology are currently available.
In these methods,
unsteady flow effects are either ignored, or modeled using a synthesis of 2-D data
(Leishman and Beddoes15), e. g. ONERA stall model stall model by strip-theory.
These methods are incapable of accurately modeling three-dimensional dynamic stall
processes, tower shadow effects, tip relief effects, and sweep effects.
These three-
dimensional effects can alter the airloads, affect the fatigue life, and significantly
influence the total cost of ownership of HAWT systems.
The first-principles-based modeling of these effects using Navier-Stokes solvers is
a viable approach, because it can capture all of the physics in great detail. However, the
high computational cost of such detailed simulations limits their use in exploratory
studies and design. Prior to the current research, Sankar and coworkers at Georgia Tech
have done extensive research on first-principles-based simulations of helicopter rotors
and propellers, in both hover and forward flight conditions.
These numerical
Georgia Tech
activities in this aspect have been documented in the thesis works of Hariharan 16 and
Berkman.17 Full Navier-Stokes simulations of HAWT configurations have also been
done by scientists at NASA Ames (Earl Duque18) and in Denmark (Michelsen19).
Tower shadow effects can be studied using overset grids. In such grids the tower
and the rotating blades are enclosed in body-fitted grid blocks. The flow field within
xxvii
each grid block is solved from first principles. Each grid block communicates with the
other grid blocks by spatial interpolation. This method, commonly referred to as the
CHIMERA Scheme, has been applied to helicopter rotor/fuselage, and propeller/fixed
wing configurations.16
Scheme, and because the interpolation and exchange of data between grids consumes
much CPU time, an overset grid based solver requires extensive hardware resources.
An alternative to full Navier-Stokes methods is a zonal approach, where viscous
flow equations are solved only around the rotor. In the present work, the flow field is
viewed as a combination of viscous regions, inviscid regions and vortices, as shown in
Figure1-5. These regions are modeled using different methodologies, making the present
approach a hybrid method. The hybrid method was motivated by the need for a firstprinciples-based analysis that will faithfully model the complexities of 3-D unsteady
viscous flow over the rotor, but remain economical enough for routine engineering use.
The present method has been previously applied to helicopter rotors in hover and in
forward flight (Berezin and Sankar20). It has also been applied to unsteady viscous flow
over oscillating wings and airfoils (Mello21).
N-S zone
Potential Flow
Zone
xxviii
Tip Vortex
The reason for choosing the hybrid methodology as the primary methodology is
discussed below:
For the horizontal axis wind turbines, the viscous flow over the rotor blades is
usually confined to small regions, even when the flow is massively separated. The
present approach models this region using 3-D Reynolds-Averaged Navier-Stokes
equations. Much of the flow field surrounding the rotor is inviscid, incompressible, and
irrotational. Modeling the inviscid region using incompressible flow equations (e. g. the
Laplace's equation) would require a costly iterative method at every time step for the
velocity field.
For this reason, the present approach models these regions using a
compressible flow equation for the velocity potential, that is elliptic in space, and
hyperbolic in time. At low tip Mach numbers, the compressible and incompressible
analyses will yield identical results, as dictated by the physics of the problem. Thus, no
error is caused by allowing compressibility effects. The tip vortices shed from the blade
affect the inflow ingested by the rotor, as well as the aerodynamic loads and torque.
These vortices are modeled in the present method as discrete entities using a Lagrangean
approach.
While this method shares some conceptual similarities with the classical viscous
inviscid interaction analyses, it does not share any of the limitations of these methods.
For example, the present method can be applied to 3-D unsteady viscous flows, whereas
conventional viscous-inviscid interaction methods are, for the most part, limited to steady
xxix
flows. The present method has no singularities at the separation point or separation line,
whereas the classical boundary-layer based interaction schemes have a strong
mathematical singularity near separation.
The hybrid solver is fast relative to a Navier-Stokes only calculation, and it can
simulate a tip vortex wake without numerical dissipation. Because of its efficiency, a
variety of flow conditions may be studied providing a better understanding of the
complex aerodynamics phenomena in the flow field around HAWT.
The current research effort focuses on (a) understanding the flow mechanisms that
affect the performance of wind turbines from axial to non-axial inflow, and (b) the
development of modern, efficient computational techniques that complement existing
combined blade element and momentum theory. The main computational effort focuses
on the extension of the hybrid method mentioned earlier to wind turbine applications. A
full Navier-Stokes solver, and an overset grid solver have also been developed for HAWT
applications.
xxx
xxxi
is done because discrepancies were found between the results from the strip theory based
design codes, which use the Prandtls tip loss model and stall delay models, and the
results from wind tunnel measurements.
spanwise load distributions than overall load estimations, hint the tip loss models and
stall delay models inherited from helicopter rotor and propeller studies may not be
applicable to wind turbine applications.
Numerical results were obtained for three rotor configurations tested at the
National Renewable Energy Laboratory (NREL) under the Combined Experiment Rotor
(CER) program. The first rotor is of rectangular planform, is untwisted, and is referred to
as the Phase II Rotor in NREL documentation. The second rotor, called the Phase IV
Rotor, has a 45 nonlinear twist distribution. The Phase II Rotor and the Phase IV Rotor
are both three bladed rotors. The third rotor configuration is two bladed, nonlinear
twisted, and linearly tapered. The third rotor is referred as the Phase VI Rotor, or the
NREL new 10-meter rotor. The present results are compared with experiments, full
Navier-Stokes simulations, and blade element and momentum theory based simulations
to establish the reliability, accuracy and efficiency of the present method. The detailed
geometry of these three rotors has been documented in Appendix A.
For the Phase II Rotor and the Phase IV Rotor, the experimental data of NREL
field test was obtained from the IEA Annex XIV: Field Rotor Aerodynamics project. 22
This project maintains an experimental database that contains measured data from five
full-scale aerodynamic test programs around the globe, including CER at NREL. The
Phase VI Rotor was tested by NREL at NASA Ames, the world largest low speed wind
xxxii
tunnel, and the simulation was done in a blind run manner. More details on the blind run
comparisons may be found in section 5.1.4.
xxxiii
CHAPTER II
This chapter explains the theory behind the Navier-Stokes equations and full
potential equation and the procedures for discretizing these equations. The algorithms for
coupling these two equation methodologies are also documented in detail.
The
methodology in this chapter has previously been used in several fixed and rotary wing
studies that were developed by Sankar and his coworkers (Sankar 23, Wake24, and
Hariharan16).
requirements to close the Navier-Stokes equations are the empirical relations for
computing viscosity and heat conductivity from pressure and temperature. Since a direct
simulation of turbulence effects is very costly, a semi-empirical turbulence model will be
used to close the Navier-Stokes equations.
xxxiv
For most fluid flow problems, especially those with a complex geometry such as a
horizontal axis wind turbine, analytical solutions to the Navier-Stokes equations are
impossible. Hence, a numerical approximation of the governing equations is required for
most applications.
The Reynolds averaged Navier-Stokes equations (RANS) are used in this study
for flow fields where all the turbulent fluctuating quantities are represented in a time
averaged manner.25, 26 The physical domain surrounding the wind turbine rotor is divided
into grid cells with the aid of a blade-fitted three dimensional grid generation procedure.
For a deforming or moving control volume, , the integral form of Navier-Stokes
equations may be written as,
d
dt
q
d
(
E
q
V
)
n
dS
E
I
V ndS
(2.1)
r
where is a deforming control volume, S is the surface of the control volume. V is the
velocity vector of the control volume surfaces, and q is the flow vector:
q v
w
(2.2)
xxxv
p
1
2
2
2
(u v w )
1 2
(2.3)
All these flow field properties are assumed to be stored at the geometry center of
the control volume formed by the grid cells.
r
r
The terms EI and Ev in equation (2.1) are the inviscid and viscous fluxes
through the surface of the control volume. They are defined as:
r
r
r
r
EI E Ix i E Iy j E Iz k
r
r
r
r
EV E Vx i E Vy j E Vzk
E Ix
E Vx
u2 p
uv
uw
u(e p)
0
xx
yx
zx
E
x5
uv
2
, E Iy v p
vw
v(e p)
xy
,
E
yy
Vy
zy
E y 5
xz
,
E
yz
Vz
zz
E
z5
xxxvi
uw
, E Iz vw
w2 p
w(e p)
(2.4)
xx 2 u x v y w z
xy u y v x
xz u z w x
yy u x 2 v y wz
yz v z w y
zz u x v y 2 w z
E x 5 u xx v xy w xz
Pr 1
E y 5 u xy v yy w yz
Pr 1
(2.5)
2
c
x
2
c
y
E z 5 u xz v yz w zz
c
Pr 1 z
(2.6)
(2.7)
These equations are given here in physical, dimensional form. In the actual
simulation, a non-dimensional form was used. Density was normalized by freestream
xxxvii
(or atmospheric) density . The velocities were normalized by the speed of sound. All
distances were non-dimensionalized by the rotor radius R. The viscosity was normalized
by the free stream viscosity, . Reynolds number of the free stream is defined as
Re cVtip / .
d
q d E I qVGrid nS
dt
faces
E
V nS
(2.8)
faces
This equation holds over a discrete control volume d, surrounded by grid cell
faces of area S and normal vector
r
n. The summation is done over the six faces of the
hexahedral grid cell in 3-D problems or over 4 line sections in 2-D problems.
Assuming that the surface integrals in equation (2.8) can be computed, equation
(2.1) or equation (2.8) can be reduced to the solution of ODEs of form,
dq
R
dt
(2.9)
where R is a residual. There are explicit and implicit procedures available for solving the
ODEs appearing in equation (2.9), such as the Runge-Kutta method.
2.1.1 Equivalence of the Integral Form of the Governing Equations and the Differential
Form of the Navier-Stokes Equations in a Generalized Coordinate System
xxxviii
While solving certain PDEs, e. g. the potential flow wave equation, an integral
form is not convenient, or readily available. For this reason, the governing equations are
also cast as PDEs in a general non-orthogonal curvilinear system.
The non-dimensional, differential form of the 3D Navier-Stokes equations in
generalized coordinates is,26
1
q~ F G H
(R S T )
Re
(2.10)
~
Here q
is the vector containing conserved flow variables in generalized
coordinates; F, G and H are the transformed inviscid flux vectors; and R, S and T are the
transformed viscous stress vectors. The are the three generalized coordinate
directions, and is the non-dimensional time in the generalized coordinate system.
~
q
1
~ v
q
J
w
(2.11)
1
y ( x z x z ) y ( x z x z ) y ( x z x z )
xxxix
1
cell
(2.12)
where cell is the volume of a grid cell and relates the physical and computational spaces.
Velocity components, u, v, and w are in Cartesian coordinate x, y and z directions
respectively. The total energy per unit volume is computed by equation (2.3).
The quantities U, V, and W, called the contravariant components of velocity are
necessary for computing the flux and viscous terms in equation (2.10). The contravariant
velocities are defined by,
U t x u y v z w
J
J
J
J
J
V t x u y v z w
J
J
J
J
J
W t x u y v z w
J
J
J
J
J
V VGrid nr S i 1/2
r
V VGrid rnS j 1/ 2
(2.13)
V VGrid nr S k1/ 2
Notice that these terms have physical explanation as the normal component of the
relative velocity between the fluid and a cell face, and the cell area.
The terms t, t, t depend on the velocity of the grid, x , y, z as seen by an
inertial observer:
t x x y y z z
t x x y y z z
t x x y y z z
xl
(2.14)
uU p
uW p
uV
p
x
x
x
1
1 vU p
1 vW p
y , G vV y p , H
F
y
J
J
J
wV p
wW p
wU z p
z
z
e p V t p
e pU t p
e p W t p
xli
(2.14)
xx
xy
xz
xx
xy
xz
x
y
z
y
z
x
1
1
R x xy y yy z yz , S x xy y yy z yz
J
J
x xz y yz z zz
x xz y yz z zz
x R5 y S5 z T5
x R 5 y S5 z T5
xx
xy
xz
x
y
z
1
T x xy y yy z yz
J
x xz y yz z zz
x R 5 y S5 z T5
where,
xlii
(2.15)
4
3
v v v
2
4
v v v
3
3
v v v
2
4
v v v
3
3
S 5 u xy v yy w yz
C P
Pr
T5 u xz v yz w zz
C P
Pr
T
T
T
x
x
T
T
T
y
y
y
C P
Pr
(2.17)
T
T
T
z
z
Intheserelationsxz,etc,aremetricsofthetransformation,PristhePrandtl
number,aisthespeedofsound,andisthespecificheatratio.Asdiscussedbefore,is
augmentedbyT,and C P Pr isaugmentedby T C P PrT inturbulentflows.
Again, Stokes hypothesis, equation (2.6), was applied, and the equation of state
for perfect gas, equation (2.7), is needed to close equation (2.10).
Notice the differential form Navier-Stokes equations (from equation 2.10 to
equation 2.17) and the finite volume form Navier-Stokes equations (equation 2.1 to
xliii
(2.16)
R5 u xx v xy w xz
v v
equation 2.7) are two forms of the same Navier-Stokes equation set and should produce
identical results. It has been shown that these two forms are mathematically equivalent. 27
~
When equation (2.10) is discretized on a general grid, the q
in equation (2.11) at
each grid point is viewed by the flow properties at that point. Suppose each grid point is
at the geometrical center of a control volume d, on which the semidiscrete finite
volumeformequation(2.8)isapplied.Thus,
u
u
~ J 1 v v q
q
cell
e
e
equivalent terms for inviscid and viscous fluxes in differential and finite volume forms
are:
xliv
uU p
vU
p
F
y E I qVGrid nS
J
wU z p
e p U p
t
uV p
vV
p
G
y E I qVGrid nS
J
wV z p
e p V p
t
uW p
vW
p
H
q
V
n
S
I
Grid
y
J
wW z p
e p W p
t
xlv
R
J
R S
xx
xy
xy
yy
xz
yz
T
z
xz
zz
i-1/2,j,k
xx
xy
1
S
J
R S T
x
xy
yy
xz
yz
xz
E
i
E
j
E
k
nS
vx
vy
vz
yz
zz
i,j-1/2,k
1
T
J
R S T
xx
xy
xy
yy
xz
yz
xz
E
i
E
j
E
k
nS
vx
vy
vz
yz
zz
i,j,k-1/2
xlvi
(2.18)
where
Finite
difference
form
1
RHS ( F G H )
( R S T )
Re
6
6
E I S E v S
i 1
i 1
Vol
Vol
(2.19)
Equation (2.18) is solved using a time marching process. That is, the solution
~
vector q
is updated from time step n to time step n+1, starting from prescribed initial
conditions. The inviscid and viscous fluxes are computed in their finite volume forms.
Computation of the inviscid and viscous fluxes is described later.
Fi 1 / 2, j ,k Fi 1 / 2, j ,k
(E VGrid ) n S
xlvii
1
i
2
(E VGrid ) n S
1
2
(2.20)
As shown in Figure 2.1, Fi+1/2,j,k is computed at cell surface (i+1/2,j,k), F i-1/2,j,k is computed
at cell surface (i-1/2,j,k) and is unity. Figure 2-1 shows a typical control volume
surrounding the node (i,j,k) with marked cell faces (i+1/2,j,k) and (i-1/2,j,k).
i1,j,k
i,j,k
i1/2,j,k
L R i+1,j,k
i+1/2,j,k
Using a technique proposed by Roe,28 the inviscid fluxes are calculated by use of
Fi 1 / 2, j ,k 0.5 ( FL FR ) C ( q L q R )
(2.21)
where qL and qR are the flow properties interpolated to (or projected to) cell face,
i , i 1, etc .
computed at the cell face (i+1/2,j,k) using the flow properties to the left and right of the
F
q
xlviii
multiplied analytically with the flow vector qR - qL. After some algebraic manipulations
the resulting form is shown to be:
u
x
v v
C(q q )
2 n y
R
L
1
1
w
n
w
z
c
e
U
h
(2.22)
where
1 C1
U c
0.5C 2
2
c
c
p
c
0.5(
)
C1
1
2
3
U c 0.5C2
2 C1
(2.23)
C 2
2
3
are defined by
The characteristic wave speeds and the contravariant velocity U
c
~
U c
~
~
~
~
~
~
~
U
;
U
c
;
U
c
2
3
~
~
~k
(u
i
v
j w
) (n x i
n y j nz k)
(2.24)
All the flow quantities with a 'tilde' represent the Roe-averaged quantities, which
for is
~
R L
(2.25a)
xlix
R / L
1
) R (
)
1 R / L
1 R / L
(2.25b)
q L = 1+ (1 ) (1 ) / 4q i
q R = 1 (1 ) (1 ) / 4qi1
(2.26)
Figure 2-2 shows the three point stencils for computing left and right primitive variables
in the 3rd order MUSCL Scheme.
qL
qR
Left
i-1
Right
i+1
i+2
qL
qR
Left
i-2
i-1
Right
i+1
i+2
li
i+3
Both the 3rd order MUSCL Scheme and the 5th order scheme are available in the
hybrid method and the full Navier-Stokes method developed here, and can be chosen by
input from the user. Most of the calculations reported here were done using the 3 rd order
MUSCL Scheme.
r
r
r
r
r
r
r
r
E
E
E
n
=
V
V
i1/ 2,j,k
V
i1/ 2,j,k
V nS i,j1/ 2,k
r
r
r
r
r
r
EV n S i, j 1/2,k EV nS i, j,k 1/2 EV n S i, j,k 1/2
(2.27)
lii
In the Baldwin-Lomax30 model, the flow field next to a solid surface is divided
into an inner and an outer layer as in the classical Cebeci-Smith model. The turbulent
viscosity in the inner layer is computed as,
Tinnerlayer l2m
(2.28)
w v 2 u w 2 v u 2
z
x
y z
x y
l m zD
(2.29)
(2.30)
Here k = 0.4 is the Von Karman constant, z is the normal distance from the wall,
and D is Prandtl-van Driest damping factor defined as,
zw w
D 1 exp
26 w
(2.31)
outer layer
K c C cp Fwake Fkleb
where
liii
(2.32)
Fmax
(2.33)
z w w
F ( z ) z 1 exp(
)
26 w
U dif
u 2 v 2 w2
max
u 2 v 2 w2
min
and
Fkleb
1
0.3 z
1 5.5
z max
(2.34)
inner layer
outer layer
has been
found. Such a place is considered as the matching point of the inner and outer layers.
liv
resolution and stiffness, and yields fairly rapid convergence to steady state. In addition,
the wall and freestream boundary conditions are trivial.31, 32
In this model, the Reynolds stresses are given by
ui u j 2 t S i j
(2.35)
S i j U i x j U j xi 2
(2.36)
where
(2.37)
where
f v1
here
~
3
3
3 And,
c v1
(2.38)
is the molecular viscosity. The quantity ~ is the working variable and satisfies
c
D
1
~
2
cb1 1 f t 2 S ~
~
~
cb 2 ~
c w1 f w b21 f t 2
Dt
f t1 U 2
(2.39)
Also,
f v2
2 2
d
(2.40)
1 f v1
(2.41)
fv2 1
lv
1 c6
f w g 6 w63
g cw3
(2.42)
where
g r cw 2 r 6 r
(2.43)
and
~
r ~ 2 2
S d
(2.44)
10
are acceptable.
The Spalart-Allmaras model has a built-in provision for driving the eddy viscosity
to zero upstream of the transition point. This is done by the f t2 function, which goes to
unity upstream of the transition point.
f t 2 ct 3 exp ct 4 2
(2.45)
The trip function f t1 is computed as follows: dt is the distance from the field point
to the trip location, which is on a wall. The quantity t is the wall vorticity at the trip
location, and U is the difference between the velocity at the field point and that at the
trip. Then g t min 0.1, U t x where x is the grid spacing along the wall at the
trip location, and
lvi
2
f t1 ct1 g t exp ct 2 t 2 d 2 g t2 d t2
U
(2.46)
lvii
u e
18.4 H 32 21.74 0.34r
log
(2.47)
lviii
This model also predicts that transition has occurred if the laminar boundary layer
separates, causing to a separation bubble near the leading edge of the rotor. The formula
for computing the boundary layer characteristics may be found in Appendix B.
22400 0.46
Rx
R 1.174 1
R x
(2.48)
In order to avoid an abrupt transition, Chen and Thyson recommend that the eddy
viscosity be multiplied by the factor:
x dx
tr 1 exp G x x tr
x tr u
e
(2.49)
The quantity
G is computed from:
3
3 u e 1.34
R x tr
2
2
C
uex
(2.50)
tr
and
lix
(2.51)
It should be noted that the quantity R x is based on the local freestream velocity
(the magnitude of the vector sum of wind speed and the blade velocity due to rotation
r ).
u x
Rx
x u local
c Vtip
cVtip
vWind
V
tip
x
c
(2.52)
Re tip
where r is the local radial distance from the hub, R is the tip radius and
x c
is the non-
tr 1 exp G x x tr
x tr u
e
ue
3
1 exp 2 R x1tr.34
C
u
local
1 exp Re
2
local
x x tr
c
x
c
x tr
c
where
Re local
vWind
Vtip
Re tip
x
d
c
u e
u local
u local c
(2.53)
(2.54)
Several explicit and implicit integral schemes are available to integrate equation
(2.18) in time.
The Runge-Kutta two- and four- step schemes are typical explicit
integration schemes.
In the present study, the following semi-implicit scheme was used to achieve
second order accuracy in time.
d
n 1
n
(qV) RHSI RHSV
dt
(2.55)
lxi
n 1
F
n
n
2
F
q O( )
q
Fn 1 Fn A n q n
n
G
n
2
Gn 1 Gn
q O( )
q
Gn 1 Gn Bn q n
(2.56)
H
n
2
Hn 1 Hn
q O( )
q
Hn 1 Hn Cn q n
where q is volume weighted primitive property, and A, B and C are the 55 flux Jacobian
matrices.
After substituting equation (2.55) into Equation (2.56) and transferring all of the
known quantities at time step n to the right hand side, one obtains,
I + t(A
i 1 / 2 E
A i1 / 2 E B j1 / 2 E B j1 / 2 E C k 1/ 2 E Ck 1 / 2 E q n1
t(RHS )
(2.57)
E q i , j , k qi 1, j ,k
(2.58)
(2.59)
lxii
Chaussee37 may be used to further approximate the block tri-diagonal matrices to scalar
tri-diagonal matrices. This option was used in the present study. Computation time is
reduced by 30% without sacrificing the formal spatial and temporal accuracy with this
approach. Details of the diagonalization process are given in Bangalore38 and Hariharan16
and will not be repeated here.
The right hand side contains a low pass filter term (Roe Scheme) to prevent oddeven errors. The left hand side also requires a similar treatment. Therefore, the matrices
are modified by adding an implicit filter. For example, the matrix Ai 1 2 , j , k is replaced
by A I
i 1 , j, k
2
, where
behind the addition of the filter terms are given in classical CFD texts (e. g. Anderson,
Tannehill) and are not presented here.
The full potential flow equation can be obtained from the mass conservation
under the assumptions of inviscid, irrotational flow fields.
The mass conservation equation in conversation form is
lxiii
t u x v y wz 0
(2.60)
For wind turbine applications, the velocity in the full potential zone is
decomposed into three parts: the wind velocity, the perturbation potential velocity, and
the induced velocity due to the embedded wake. Equation (2.36) below shows these
three components,
V Vwind Vi
u u wind x u i
v v wind y v i
(2.61)
w wwind z wi
Here the terms ui, vi, wi are the induced vortical velocity components associated
r
with the embedded wake elements. Details on calculation of this vortical velocity, Vw ,
are given later in Chapter III. The solution procedure given in this section was originally
developed by Sankar et al.39 for application to helicopter rotor. It was applied to hover
and forward flight conditions for helicopter rotors by Berkman 17 and Yang40 and modified
in this study for wind turbine applications.
The density terms in equation (2.60) can be expressed using the velocity potential,
and its derivatives (i. e., the potential velocity components), and the isentropic gas
relation.
a 2
2
a
1
1
lxiv
(2.62)
a2
u 2 v 2 w2
a2
V2
t
1
2
1 2
(2.63)
Combiningequation(2.61),(2.62)and(2.63),asecondorderhyperbolicpartial
differentialequationfor is obtained:
x xt y yt z zt x x y
a 2 tt
y z z
(2.64)
This equation is readily transformed from the physical space (x, y, z, t) to the
computational space (, , , ):
U V W
Q
2
J J J
aJ
(2.65)
HereQisasourcetermandisafunctionofgridmotionsanddeformations.Thedetailed
derivationaboutequation(2.64)andequation(2.65)maybefoundinSankar,Maloneand
Tassa.41Thesourcetermvanishesforgridsinrigidmotion,asinthepresentstudies.The
termsU,V,andWarethecontravariantcomponentsofvelocityasbeforeandJisthe
JacobianofthetransformationbetweentheCartesianandthetransformedcoordinates.
lxv
U t A 1 A 2 A 3
V t A2 A 4 A 5
(2.66)
W t A 3 A 5 A6
where t, t, and t depend on (x,y,z) as before.
Also,
2
2
2
A 1 x y z
A 2 x x y y z z
A 3 x x y y z z
A 4 2x 2y 2z
(2.67)
A 5 x x y y z z
A 6 2x 2y z2
lxvi
2
a
J
n 1
i , j ,k
nU n 1
U nn 1 V nn 1 W nn 1
nV n 1
nW n 1
(2.68)
where
n 1
i , j ,k
n 1
n ) ( n n 1 ) i , j ,k
n 1
i , j ,k
(2.69)
n 1 n 1 n
The mixed time and space terms in equation (2.68) are discretized using two-point
upwind spatial differences in and two point backward temporal differences. For example,
U
n 1
i , j ,k
2t
2t
(2.70)
nU n 1
i , j ,k
nV n 1
i 1 / 2 , j , k
nV n 1
J
nU n 1
i 1 / 2 , j , k
i , j ,k
nU n 1
lxvii
i , j 1 / 2 , k
nV n 1
i , j 1 2, k
(2.71)
nW n 1
i , j ,k
nW n 1
i , j , k 1 2
nW n 1
J
i , j , k 1 2
Introducing equation (2.69) through equation (2.71) into equation (2.68), and
using n 1 n the following system of equations results at any point/cell (i,j,k):
n 1
n 1
n 1
n
n
n
a ni, j,k in1
,j,k 1 b i,j,k i,j 1,k c i, j,k i 1, j,k di, j,k i,j,k
n 1
n
n 1
n
n
e ni,j,k ni1
1, j,k f i, j,k i, j1,k gi, j,k i, j,k 1 R i,j,k
(2.72)
The
R ni, j,k
coefficients
a ni, j,k , b ni,j,k , c ni, j,k , d ni,j,k , e ni, j,k , f i,n j,k , g i,j,k ,
and
velocities, the density , the speed of sound a, and the time step t as shown below.
a i,j,k
n
W n
A 6
z 2
a J i,j,k J i,j,k 1 / 2
n
V n
A4
b i,j,k y 2
U n
A1 n
c i,j,k x 2
a J i,j,k J i1 / 2,j,k
lxviii
d i , j ,k 2
2
a J
A1
J
A6
i 1 / 2 , j , k
n
i , j , k 1 / 2
2
a J
i , j ,k
A1
J
A6
i 1 / 2 , j , k
2
a J
i , j ,k
A4
2
a J
i , j ,k
i , j 1 / 2 , k
i , j ,k
A4
i , j 1 / 2 , k
(2.73)
i , j , k 1 / 2
U n
A1
e i,j,k (x 1) 2
a J i, j,k J i 1/ 2, j,k
fi,j,k
V n
A 4
( y 1) 2
g i, j,k
n
n
W
A6
( z 1) 2
a J i, j,k J i,j,k 1/ 2
R i,j,k
1
U
U
n
i, j,k
2 2
J i 1/ 2, j,k J i 1/ 2, j,k
a J i,j,k
n
V
V
W
W
J i,j1 / 2,k J i,j1 / 2,k J i, j,k 1/ 2 J
n
i, j,k1 / 2
(2.74)
where x, y, z are switching terms from the upwind differences. For example, x = 0
for U < 0 and x = 1 for U > 0.
Equation (2.72) can be expressed in a sparse septa-diagonal matrix system:
M i, j,k
n 1
lxix
R i,j,k
(2.75)
M 1 d i, 1j ,k M 2 d i, 1j ,k M 3 n 1 R in, j ,k
(2.76)
where
M 1 d i , j ,k ci , j ,k E ei , j ,k E
M 2 d i , j ,k bi , j ,k E f i , j ,k E
(2.77)
M 3 d i , j ,k ai , j ,k E g i , j ,k E
n 1 inequation(2.76).Finally,
n 1 n n
(2.78)
To correctly simulate the flow field, flow field properties need to propagate from
the outer potential zone into the inner Navier-Stokes zone, and from the inside zone to the
outside zone. False reflections at the boundary will most likely cause divergences in the
unsteady solving procedure. A method had been developed and validated by Sankar and
his coworkers for fixed wing (Sankar et al. 42) and rotor flows (Berezin and Sankar20). In
the present integrated Navier-Stokes/Full Potential/Free Wake method, a similar scheme
is used to pass flow information between Navier-Stokes and potential flow zones.
lxx
Figure 2-4 shows the computational domain that consists of the Navier-Stokes and
potential flow zones. Two grid blocks, both structural H-O grid, cover the overall
computational domain surrounding a reference blade. For each block (upper and lower)
there are three interfaces that surround the Navier-Stokes zone. The plane k = k match
corresponds to the interface between the inner zone and the outer zone. The planes i =
imatch1 and i = imatch2 are the interfaces upstream and downstream of the blade. These three
planes extend all the way in the spanwise direction. While the normal vectors for these
three interfaces are different, these three interfaces are treated in the same manner as
shown below.
Kmatch
Block 1
Imatch 2
Imatch 1
N. -S.
FP
Block 2
Figure 2-4. Partitioning of Computational Domain for H-O and H-H Topologies
lxxi
u x x x u w u wind
v y y y v w v wind
(2.79)
w z z z ww wwind
u2 v2 w 2
V2
C PT
t C PT
2
2
(2.80)
(2.81)
lxxii
equation. At a typical interface between the inner and outer zones, there are three types
of waves that cross the boundary: acoustic, vorticity, and entropy waves. The acoustic
waves are passed to the outer zone. The vortical waves are accounted for through vortex
elements or markers that are tracked in a Lagrangean fashion as discussed in the next
section. The entropy wave carries entropy (i. e. changes in total pressure) at a velocity
Vn, and has to be ignored in the isentropic potential flow region.
The normal component of velocity at the interface in the potential flow zone is set
equal to the normal velocity component from the Navier-Stokes (inner) zone for the
acoustic waves:
Vn
FP
Vn
(2.82)
NS
W
J
NS
W
J
FP
(2.83)
or,
W
J
A3 A5 A6
J
NS
(2.84)
Here can be solved from equation (2.83). Using the central difference formula at the
boundary,
lxxiii
i, j,k match 1
(2.85)
the interface condition for the full potential zone can be expressed as,
i,j,k match 2
2 W NS A3 A5
A6
i, j,k match
(2.86)
Initial conditions are set equal to the properties of the freestream fluid. In the
present calculations involving multiple moving grids, an inertial reference system is
adopted. The coordinate system is fixed to the body, and the body moves with respect to
the stationary surrounding fluid.
The non-dimensional flow vector q is initialized by:
u
wind
q v wind
wwind
(2.87)
numerical boundary as shown in figure 2-5. They are, far field boundaries at j=jmax,
lxxiv
k=kmax for the upper block, and k=1 for the lower block; rotation center boundaries at
j=1; the cyclic boundaries at i=1 and i=imax; the internal boundaries at the interface
between two blocks; and the blade surface boundary.
Blade Surfaces
Internal Boundaries
Cyclic Boundaries
The blade grid for a reference blade covers azimuthal angle 360 N b , where Nb
is the number of blades. At the upstream and downstream boundaries, (I=1 and I=Imax)
one of the following strategies was used. If a full Navier-Stokes solution is done, and if
the wind is along the axial direction, then the solutions at these boundaries are periodic
lxxv
and
V V wind V Induced
, where
V Induced
is
0
n
n
(2.88)
lxxvi
CHAPTER III
As described earlier, in the present work, the flow field is divided into three
regimes: (a) a small viscous region surrounding individual rotor blades where the NavierStokes equations are solved, (b) a potential flow region which carries the acoustic and
pressure waves generated by the rotor to the far field and (c) a Lagrangean scheme for
capturing the vorticity that leaves the viscous region and convecting it away to the far
field as shown below.
N-S zone
Potential Flow
Zone
Tip Vortex
lxxvii
A wake model is needed to carry the shed wake and generate an induced velocity
field for the potential zone nodes. The Biot-Savart law is used to calculate the induced
velocities due to wake elements.
neglecting the inboard vortex sheet and the trailing vorticity. However, it must be noted
that the very near wake effects are captured in the Navier-Stokes zone and neglecting
trailing vorticity in the far wake is a common approach followed in numerous free or
rigid wake models (Johnson43). The justification comes from observations of several
researchers that the vortex sheet influence is weak compared to that of tip vortex. The tip
vortex that leaves the Navier-Stokes zone, and enters the potential flow zone, was
converted into 200 to 300 piecewise line segments, which together form a skewed shape.
The spatial positions of the wake were subsequently tracked in time in a Lagrangean
fashion. Vortices from all the blades were modeled in this manner.
A prescribed wake model is used to determine the wake shape. The tip vortex is
typically modeled up to 10 revolutions in this study. For the simulation of HAWT,
because the rotor is wind driven, and the axial far field velocity is much larger than the
induced velocities and the wake filaments are separated by large distances from the rotor.
So a prescribed rigid wake model is sufficient.
For the rigid wake model, the radius of the wake is equal to the rotor radius. The
distance between two adjacent revolutions is equal to the product of the time for a
revolution and the average normal velocity at the rotor disk. The average normal velocity
is derived from momentum theory as,
lxxviii
V vi
V
V
2
2
T
2 A
(3.1)
Once the tip vortex shape is defined, the next step is the selection of a physically
correct vortex strength. The vortical strength for each of these markers was assumed to
be the maximum bound circulation at the blade at the time the marker is released. In the
hybrid methodology, the vorticity strength is calculated based on the information from
the Navier-Stokes analysis. Since the sectional lift coefficient along the blade is known
from the computed surface pressures, the Kutta-Joukowski theorem gives the bound
vortex strength,
U t
1
U t2 cCT
2
(3.2)
The peak value of was taken as the tip vortex strength. Here Ut is the total local
velocity, and consists of the inflow wind velocity, the tangential velocity component
r at
that blade section, and the wake induced velocities. C T is the normal force
coefficient. This relation is different than the method used for helicopter rotor, which
uses tangential velocity component velocity and normal force coefficient, C N. For the
simulation of a HAWT, because the inflow wind speed is large (especially at the inner
span of the blade), the above method can avoid the singularity near the blade root, where
the normal forces do not come from inviscid mechanisms, but leading edge separation.
Given the vorticity strength and geometry, the induced velocity can be evaluated
by the Biot-Savart law, viz.,
Vw
elements
dl r
4 r 3
lxxix
(3.3)
where r is the spatial vector from a vortex element to the point where the induced
volocity Vw is calculated. dl is the length vector that defines the wake element as shown
in Figure 3-2:
P
dl
A Rankine vortex core model with core radius of one tenth of the blade radius is
used. This model forces the induced velocity to approach zero, whenever the point where
the induced velocity is being calculated lies within the vortex core.
Each element induces velocity on all of the potential nodes. Thus computation of
induced velocities consumes considerable CPU time. However, such a calculation can be
done once every 4 degrees of blade rotation. At each time step the induced velocities are
updated based on the changes to tip vortex strength.
lxxx
A numerical procedure for modeling non-axial wind (yaw) conditions has been
developed. As in the case for axial flow conditions, the yaw calculations only need to
model the aerodynamics of a single blade. Other blades will experience the same load
and flow pattern 1/N revolutions later, where N is the number of blades. For a threebladed rotor the computational domain covers a 120 portion of the rotor disk. The
present procedure thus retains the efficiency of the hybrid method even for yaw
conditions. In contrast to the hybrid method, a full Navier-Stokes solver would require
the modeling of all blades, significantly increasing the computational effort.
When developing a first-principles-based analysis for modeling rotors in cross
flow, there are three kinds of non-axial flow (yaw) effects that should be addressed. First
is the difference in the flow between the advancing and retreating sides due to the
velocity component in the plane of rotor disk. As shown in figure 3-3, the turbine blade
has a higher relative velocity with respect to the wind when it is on the advancing side,
and a lower relative velocity on the retreating side. This fluctuation in the chordwise
velocity produces fluctuations in the blade loads and the power generated.
Advancing side
Vx
Retreating side
rotor plane. Furthermore, the vorticity strangth in the wake will vary with time, as the
loads on the blade vary with time. This is in contrast to axial flow, where the blade
loading is independent of the azimuthal location of the blade.
Vwind
Axial
vI(r,)
Edgewise
Finally, the analysis must include aeroelastic deformation of the wind blades, as
well as blade teetering and flapping movement, if any. The rotors tested by NREL are
assumed rigid, without any cyclic pitching or flapping of the blades.
The present methodology can readily simulate yaw conditions. As before, only a
single blade was solved.
mechanism. The present methodology greatly reduced the necessary computer memory
and CPU time for yaw simulation.
When there is only an axial wind present, physically, and computationally, the
flow field, the loads, and the power become asymptotically steady. When cross wind or
lxxxii
yaw effects are present, only an unsteady simulation may be done, and the loads will be
periodic from cycle to cycle.
lxxxiii
CHAPTER IV
4.1 Introduction
For a downwind wind turbine as used in the CER test, the blades are downwind of
the rotor. Thus, the tower and the nacelle will notably influence the flow field of the
wind rotor, the unsteady blade loads, and the power output quality. These factors must be
accurately computed so that the life span (fatigue life) of the wind turbine may be
estimated.
turbines.
Wind
Tower
Blade Rotation
Figure 4-1. Schematic
of Downwind Horizontal Axis Wind Tubine
lxxxiv
Tower effects on the wind rotor were studied using overset grids. A full NavierStokes methodology was used. The tower and each blade were enclosed by body-fitted
grids. The blade grid is rotating while the tower/nacelle grid and a background grid
encompassing the entire flow domain (if applicable) is stationary.
Information is
This is achieved by
interpolating the numerical boundary conditions based on the flow properties from the
other grid blocks into the current grid. Such numerical boundary conditions are needed
for the outmost grid layer and the fringe points, which are defined below. The process of
lxxxv
(ii)
(iii)
Blade Rotation
C
Minor Grid
Figure 4-2: Schematic of the Top View of a Simple Overset Grid System
lxxxvi
The tower grid (larger grid) in the background is identified as the major grid, and
the rotor grid (inner submerged grid) is identified as the minor grid.
Again, for
simplicity, the minor grid boundary does not intersect with the tower geometry.
A boundary needs be defined where transfer of flow field properties between
major and minor grids occurs. This boundary is called in CHIMERA terminology as the
hole boundary. The solid enclosed curve C shown in figure 4-2 is the hole boundary.
Either a selected grid layer or a closed analytical surface can be used as the hole
boundary. However, the hole boundary should be sufficiently far away from the minorbody (the rotor blade in figure 4-2).
Further, the grid points of the tower grid enclosed by the hole-boundary (curve C)
are set as the hole points for that grid. These hole points are set to invalid and are not
solved. All the points at the fringe of the region enclosed by the hold-boundary and
adjacent to a hole point are defined as the fringe points. The present methodology treats
all of the fringe points as a boundary. The information at these fringe points in the tower
grid is prescribed as follows. First, the present methodology identifies the grid cell in the
rotor grid encompassing a tower grid fringe point, and extracts flow information at the
four (eight for 3D) rotor grid points forming the grid cell.
variables are updated at the major-grid fringe point by interpolation. This process is
repeated for all fringe points. In order to have sufficient minor-grid points available for
such interpolation, all the fringe points in the tower grid need to be at the inside of the
rotor grid. This completes the solution process for the major grid.
lxxxvii
In this simple configuration the entire minor grid lies within the interior of the
major grid. The information at the outmost boundary of the rotor grid is derived from the
tower grid by an interpolation process similar to that for tower gird fringe points.
Rotor Grid
Tower Grid
Each blade is enclosed in a body fitted C-H grid rotating inside the tower grid. In
the present simulation, which is similar to the tower/rotor configuration in section 2.1, the
blade grids do not touch the tower, nor intersect with the outmost layer of the tower grid.
With these constraints, there is no hole point in blade grid, and the interference between
the overset boundary and outer boundary/body need not to be computed. This hierarchy
lxxxviii
is chosen because the computation of grid connectivity consumes much CPU time. It
must be noted that there is no limitation in present CHIMERA methodology with regard
to the complexity of grid hierarchy, and the grid system can be constructed as complex as
needed.
lxxxix
CHAPTER V
A number of calculations have been carried out for a three-bladed horizontal axis
wind turbine system tested at NREL (Simms et al 44). NRELs Combined Experiment
Rotor is a 3-bladed downwind HAWT. Three configurations were computed. The first,
known as the Phase II Rotor, is a rectangular planform blade with no twist. It has a rotor
radius of 5.05 meters and chord of 0.4172 meters. The rotor has a fixed hub with the
blade pitch set to 12 degree with a pre-cone of 3.5 degrees. The second rotor, known as
the Phase III Rotor or the Phase IV rotor, has a planform identical to Phase II, but has a
45 non-linear twist. The blade pitch was set to 3 degrees referenced to the tip pitch. The
third rotor studied here, is known as the NREL new 10-meter rotor, or as the Phase VI
xc
Rotor. This is a 2 bladed rotor with twisted, tapered blades. The S809 airfoil was used in
all three rotors. The configuration details of these rotors can be found in Appendix A.
Figure 5-1 shows a typical H-O grid. When the wind velocity is directed along
the axis of the rotor, the flow properties are periodic from one blade to the next, and
viscous and inviscid flow zones over only a single blade need to be considered.
However, the tip vortices from all blades must be modeled. The user may specify the
number of cells along the chord, and along the radial and normal directions.
Overall Grid
Near Grid
The present methodology is quite flexible. One can choose the size of the NavierStokes zone, and the potential flow region at the start of the flow analysis. The NavierStokes zone should be large enough to enclose the boundary layer over the blade, and any
separated flow regions. In the cases studied here, the Navier-Stokes zone extended 1/4
xci
chord ahead of blade leading edge, and 1 to 2 chords behind blade trailing edge, and 2
radii beyond the blade tip. The Navier-Stokes zone extended about 1/3 chord length
above the upper surface, and 1/3 chord length below the lower surface. The outer
boundary of a potential flow zone was a cylinder of radius equal 2 blade radii, and a
height of 2.2 blade radii (1.1 radii above the rotor disk and below). The baseline grid
dimension for present calculations is 1104380, with a total of 380,000 grid points. A
viscous zone with above size consists approximately 100,000 grid points.
Both the hybrid methodology and a stand-alone full Navier-Stokes methodology
have been tested for these three configurations. The full Navier-Stokes methodology uses
exactly the same equations and numerical procedures as the viscous flow method in
hybrid procedure. The only difference is that the Navier-Stokes equations are solved over
the entire flow field.
5.1.1 Validation of the Baldwin Lomax Turbulence Model and Baseline Hybrid
Methodology
Figure 5-2 shows the Navier-Stokes code and hybrid code predictions, along with
measured data for the Phase II rotor. In general, very good agreement was observed with
measured data from NREL aerodynamics experiments at attached flow conditions (for
example, at 10 m/s) and under stalled conditions. In these calculations, the transition
point was assumed to be at 40% chord, based on the 2-D calculations carried out by
Wolfe et al.46,47 Downstream of the transition location, and in the wake, the eddy viscosity
was calculated using the Baldwin-Lomax turbulence model.
xcii
Generator Power[kw]
20
15
10
NREL experiment
N-S Solver
Hybrid Code
BEM Theory
5
0
0
-5
10
15
20
25
Wind Speeds[m/s]
Figure 5-2. Computed vs. Measured Power Generation for Phase II rotor
xciii
solver would have yielded results that are quantitatively equivalent to the hybrid code.
As in the case of the Phase II rotor, the BEM theory predicts the trends in the power
generation well, but the magnitudes are overestimated.
Generator Power[kw]
20
NREL experiment
BEM Code
Hybrid code
Navier-Stokes
15
10
0
0
10
Wind Speed[m/s]
15
20
Figure 5-3. Computed vs. Measured Power Generation for Phase III rotor
An attempt was made to identify the source of the discrepancy between the
present first-principles-based methods, and the BEM theory code. Figure 5-4 shows the
spanwise lift distribution for the Phase II rotor from the full Navier-Stokes code and the
BEM code. The agreement is reasonable at stations outboard of 60% blade radius. These
outboard stations operate at high dynamic pressures and are primarily responsible for the
power generation. In the inboard regions, the low rotational velocity r of the blade
xciv
sections yields low dynamic pressures, exaggerating the differences in the sectional lift
coefficients between the BEM theory and the Navier-Stokes code.
Figure 5-4. Spanwise loading over Phase II Rotor with comparisons with YAWDYN
Although the BEM theory and the first-principle codes have comparable blade
loads at the outboard stations, the induced velocity (or the inflow through the rotor) was
not the same for the two codes. The BEM theory uses an analytical expression for the
inflow velocity based on the combined blade element-momentum theory. The NavierStokes code and the hybrid code compute the inflow from first principles. The power
generation adversely depends on the induced drag, which is roughly the product of the
sectional lift forces and the inflow velocity. Thus the two methods (BEM theory and
present method) predict somewhat different values for the power generation.
xcv
Figure 5-5 shows the evolution of power coefficient with iterations for three wind
conditions, using the hybrid code. It is seen that the calculations rapidly settle down to
steady state values in 2000 iterations or less. A typical iteration requires 7 seconds on a
SGI Octane 2 work station by single processor for a 1104380 grid. Thus, reliable
steady state results can be obtained in about 4 CPU hours or less on a SGI Octane 2
workstation class systems. On faster multiprocessor machines such as the Cray Y/MP,
and the SGI Origin 2000 systems, the CPU time may be effectively reduced to 1 hour or
less by a combination of faster clock speeds and more processors after this code is
parallelized. Such a fast turn-around will be needed for industry use of this firstprinciples-based methodology in a design environment. However, The Navier-Stokes
zone requires more CPU time per point per time step, compared to the full potential zone.
As a result, if the grid is highly clustered near the blade, the Navier-Stokes portion of the
simulations will dominate the overall CPU time requirements. For example, if the grid
points are doubled in the direction normal to the blade, the CPU time will increase by a
factor of 4, instead of the anticipated factor of 2, because there will be more nodes in the
Navier-Stokes zone.
xcvi
20
Power(kw)
16
12
10 m/s
8 m/s
6 m/s
0
0
1000
2000
3000
Iterations of code
4000
5000
Figure 5-5. Evolution of Power with Iterations for Phase III Rotor
It may be noted that the hybrid solver converges quickly, while the full NavierStokes simulation requires 10,000 time steps or more. This apparent discrepancy is
explained by the fact that the hybrid solver starts with an initial guess for several
revolutions of the tip vortex shed into the wake, and its associated inflow velocity. In the
present simulations this starting wake is assumed to be a helix. Thus, the hybrid code
spends much of the calculations relaxing the tip vortex shape, and adjusting the tip vortex
strength based on the blade loads. The full Navier-Stokes calculations, on the other hand,
start from a uniform flow. The blade has to spin several revolutions (for several thousand
time steps) to generate the tip vortex, and establish the inflow through the rotor.
In addition to yielding engineering quantities of interest to the designer, the
present first-principles-based simulations provide useful information on the flow features.
Features such as flow separation, flow unsteadiness, and the radial migration of the low
momentum fluid due to centrifugal pumping effects may be studied. The designer may
xcvii
be able to use these visualizations to design rotors with minimum flow separation.
Figures 5-6 and 5-7 show the particle traces over the Phase II rotor, computed by
releasing particles into the flow field, and tracking their subsequent motion in time. A
coordinate system attached to the rotor is used for clarity. At lower wind conditions (e. g.
10 m/s) the flow is well attached over the outer 50% of the rotor, as shown in figure 5-6.
Only the inboard stations are separated. The radial migration of the fluid particles along
spiral trajectories, due to centrifugal pumping effects is also seen. At higher wind
conditions (e. g. 20 m/s wind) much of the flow over the rotor is extensively separated,
except near the tip where the strong inflow due to the tip vortex reduces the angle of
attack and keeps the flow attached. As a result, significant amounts of power are
generated in the tip regions at this wind condition.
xcviii
-3
-2
1_eqn; Eppler
0_eqn; Eppler
Leading Edge
Root
-1
Tip
R
0
1
0_eqn; Michel
1_eqn; Michel
2
3
0
10
12
Figure 5-8. Transition Lines on Upper Surface of the Phase III Rotor in 6 m/s
xcix
-3
0_eqn; Eppler
-2
1_eqn; Eppler
Leading Edge
R
Root
-1
Tip
0
1
2
0_eqn; Michel
1_eqn;Michel
3
0
10
12
Figure 5-9. Transition Lines on Lower Surface of the Phase III Rotor in 6 m/s
implemented in the present approach, first checks to see if laminar boundary layer has
separated. If so, Epplers model assumes that transition has occurred. Note that the
inflexion point on the separated flow boundary layer will cause Tollmien-Schlichting
instability to develop, causing transition. The Michel criterion, on the other hand,
bases its transition criterion primarily on the boundary layer thickness. At this wind
speed, the boundary layer has to grow up to 55% of the chord length, before Michels
criterion detects transition.
b) On the lower surface, the pressure gradients tend to be more favorable than on the
upper side. This leads to a thinner boundary layer and separation aft of the 40%
chord. As a consequence, both these criteria predict that transition will occur aft of
the corresponding upper surface locations.
c
c) The Reynolds number near the root is less than 105. Both models predict that the flow
will remain laminar all the way to the trailing edge, near the root region.
d) The transition line location appears insensitive to the turbulence model used.
-3
0_eqn;Eppler
-2
1_eqn;Eppler
Leading Edge
R
Root
-1
Tip
0
1
0_eqn;Michel
1_eqn;Michel
3
0
10
12
Figure 5-10. Transition Lines on Lower Surface of the Phase III Rotor in 8 m/s
Figure 5-10 shows the transition lines on the lower surface of the CER Phase III
rotor at 8m/s. Even though the overall pattern of the transition lines is similar to the 6m/s
case, the following differences may be observed:
a) Michels model predicts that the transition phenomenon over much of the lower
surface is delayed, compared to the 6m/s case. This is attributable to the higher local
angle of attack the blade sections operate in, and the favorable pressure gradients that
exist on the windward (i. e. lower) side of the rotor.
b) The Eppler transition model, on the other hand, predicts transition lines that are
similar at the 8m/s and 6m/s conditions, presumably because laminar separation is
detected in the vicinity of 40% chord at both these wind conditions. Notice that the
ci
maximum thickness location for the S-809 airfoil is near 40% chord. The pressure
gradient tends to be favorable from the leading edge up to 40% chord, after which it
becomes adverse at both these wind conditions.
c) At the higher wind speed, a larger region near the root on the windward side remains
laminar.
d) The transition line predicted using the Michels transition model in conjunction with
the Sparlart-Allmaras turbulence model has a kink near 33% radius. The reason for
this behavior is not known at this writing.
-3
0_eqn;Eppler
-2
1_eqn;Eppler
Leading Edge
R
Root
-1
Tip
0
1
0_eqn;Michel
1_eqn;Michel
3
0
10
12
Figure 5-11. Transition Lines on Upper Surface of the Phase III Rotor in 8 m/s
Figure 5-11 shows the transition lines on the upper surface at 8m/s. Epplers
model predicts that transition will occur near the leading edge, as a result of leading edge
separation. Michels model, on the other hand, predicts transition around 50% chord,
with a considerable radial variation in the transition location, especially near the root.
cii
The large difference observed in the upper surface transition pattern of the rotor
between the 8 m/s and the 6 m/s cases is physically meaningful. For the CER Phase III
rotor at 72 rpm, as the wind speed increase to around 8m/s wind, the operating state
switches from a wind turbine state to a turbulent wake state. The next section shows
what happens to the operation of a wind turbine as the wind speed is increased.
Figure 5-12 shows the effect of the turbulence model for the Phase III rotor. The
Spalart-Allmaras model predicts a lower power output compared with the Baldwin
Lomax zero-equation turbulence model. This difference is because the Spalart model
predict much larger turbulence viscosity, so that the drag is larger than for the BaldwinLomax turbulence model.
20
NREL experiment
BEM Code
Hybrid; Boldwin Lomax Model
Hybrid; Spalart Model
Generator Power[kw]
15
10
0
0
10
Wind Speed[m/s]
15
20
Figure 5-12. Power vs. Wind Speed for the Phase III Rotor
ciii
20
NREL experiment
BEM
Hybrid
Generator Power[kw]
15
10
0
0
10
Wind Speed[m/s]
15
20
Figure 5-13. Power Generation vs. Wind Speed for the Phase III Rotor
It turns out that the answers to these questions are linked to a concept called the
rotor state. The theory and phenomenology of rotor states was first studied by Glauert 48
for propellers, and extended by Wilson and Lissaman49 to wind turbines (Eggleston50). As
illustrated by Eggleston, several wind turbine states can occur as the blade pitch angle
changes from a relatively large positive value to negative values. The wind rotor states
civ
are: propeller state, zero-slip state, windmill state, turbulent windmill state, vortex ring
state, and propeller brake state.
In the CER experiment, the wind turbines rotate at a constant speed, and the blade
pitch angle is set to a designer-specified value. This angle is (usually) the optimum pitch
angle at which maximum power will be generated, over a variety of wind conditions
encountered in a wind farm.
r
V0
Lift
Zero-lift
line
vi
Vtota
l
r
V wind
wind
Zero-lift line
cv
Figures 5-14 through 5-18 show some of the possible rotor states. As shown in
figure 5-14, the symbol represents local impinging angle, and is the pitch angle with
regard to the zero-lift line. The propeller-state exists when is greater then . In this
state, the rotor accelerates wind in the same direction it was originally going, adding
energy to the wind. The induced velocity vi is in the same direction as the oncoming
wind. Power must be supplied to maintain the rotor rpm when the rotor operates in the
propeller-state.
When = , the rotor is in zero-slip state, and the induced velocity is zero, as
shown in figure 5-15.The rotor does not extract energy from the wind, nor supply energy.
Propeller aircraft pilots normally adjust the blade pitch angle so that this condition
occurs, if the engine should fail.
When the wind speed increases further, for < , the windmill state occurs. It is
the designed operational state of wind turbine.
0 vi
V0
. The rotor extracts power from wind. For this condition, the flow field over
2
the rotor is steady in a rotating coordinate system. Figure 5-16 shows sketches for the
windmill state.
cvi
Lift
r
vi
Vwind
Zero-lift line
V total
When the wind speed V0 increases further, vi may be larger than V0/2, but still
small enough, so that is still less than . Then the rotor is in the turbulent windmill
state. Under the turbulent windmill state, the rotor still can generate power. To an
observer, the rotor looks like an opaque disk, and the unsteady vortex structure looks like
a turbulent wake. The flow around the rotor is highly non-uniform, and unsteady, and
can not be studied by momentum theory.
cvii
As mentioned above, in the present study, the relative variation between the local
pitching angle and local impinging angle was due to the increase of wind speed. On
the other hand, similar variation can be obtained by increasing the collective pitch of the
rotor. According to Wilson and Lissaman 49, if the collective pitch is increased while the
wind speed is constant, vi may be of the same order of V0, and the vortex-ring state
occurs. The tip vortices are pushed up by the oncoming wind, and pushed down by the
induced velocity. These vortices, therefore, get trapped in a vortex ring. As more and
more vorticity is captured in a doughnut-shaped vortex ring, the ring distends and
periodically bursts leading to an unstable flow near the rotor disk. Whenever the ring
bursts the trapped vorticity is released and carried away by the wind, and away from
the rotor, the process of vorticity accumulation inside the vortex ring begins all over
again. Thus, the vortex-ring-state is an unsteady, periodically recurring state.
If the collective pitch is increased further from vortex-ring state, while the wind
speed is constant, goes negative, a point is reached where the torque goes to zero.
Beyond that, the propeller brake state occurs. The rotor acts as a brake, actively pushing
the wind back from whence it came, as shown in figure 5-18.
cviii
While the vortex ring state and the propeller brake state will definitely occur if the
collective pitch is increased for a constant wind speed, they may not occur for a wind
turbine rotor at an operational collective pitch and rpm while the wind speed increases, e.
g. the case studied. This is because the induced velocity cannot be higher than the wind
speed if the collective pitch is held constant at an operational condition, and wind speed
is increased further from a significant wind speed for windmill state.
In the present cases, the collective pitch is set to a constant value so that the rotor
is at a favorable windmill state at an operational wind speed. While the wind speed
increases from zero to a significantly high value, only the propeller state, zero slip state,
windmill state, and turbulent windmill state may occur.
When the wind speed increased further from the turbulent windmill state, a fully
stalled turbulent windmill state occurs. The rotor blade stalls and acts like a plate at an
angle of attack, found on ancient windmills. The positive power output observed is due
to drag generated by the blades, and not due to lift. The ratio of power generated to the
wind kinetic energy is low, and the efficiency is low, because the wind turbine is driven
by drag, as ancient windmills were, hundreds years ago.
The windmill state and the turbulent windmill state are the only two states of
interest to wind turbine designers. The vortex ring state can cause fatigue and should be
avoided by adjusting the blade pitch angle.
Returning to figure 5-13, we can answer the questions previously raised as
follows:
a) Why does the measured power curve has a distinct break at 10m/s?
cix
From 5 m/s to 10 m/s the windmill-state occurs. Between 10m/s and 13 m/s the
rotor operates in a turbulent windmill state. The induced velocity v i has two
branches near vi = V0/2, which must be correctly chosen based on physical
considerations, i. e. the rotor operating state. Simple wake theories (e. g. that
used in YawDyn) may completely miss this break in the branch and produce a
smooth, incorrect power prediction curve, as shown in figure 5-13.
b) Why is the power not measurable between 13 m/s and 17 m/s for the Phase III Rotor?
At about 13m/s, the rotor operates in a strong unsteady state changing from the
turbulent windmill state to fully stalled turbulent windmill state. Since the local
impinging angle is oscillating around the stall angle, separated and attached flow
conditions may occur alternately due to the perturbation in the wind turbulence.
This phenomenon is very similar to dynamic stall effects, and can occur even
under quasi-steady conditions. In the next section, the wind turbulence in field
tests is shown in figure 5-28, and it can be seen that the wind oscillation may be
as high as 10% of the mean wind. The power generated is highly unsteady, and
fluctuates widely. The CER experiment reports no quasi-steady measured data in
this region for the Phase III/IV Rotor until the rotor is in the second turbulent
windmill state at 17 m/s. Such unsteady phenomena may not occur if the inflow
wind is steady, as in the wind tunnel test. The measurements from NRELs NASA
Ames wind tunnel tests did not show such an unsteady behavior. However, since
wind turbines are used in the open atmosphere, this phenomenon, dynamic stall
due to inflow turbulence, deserves further studies.
cx
Untwisted
Blades
Optimal Twisted
Blades
A. O.
A
Stall AOA
Wind Speed
Increases
Root
Tip
cxi
defined dynamic stall phenomenon for most of the blade radius simultaneously,
causing the measured power oscillate drastically during stall. Thus, optimally
twisted rotors are not as stable as rotors using untwisted blades near stalled wind
speed under naturally fluctuating wind conditions.
c) In what region can the hybrid code reliably predict the power output?
The present numerical studies show that it is hard for the hybrid methodology to
converge in the stalled turbulent windmill state and higher wind speeds. Under
these states the basic assumption, i. e. the division of the flow field in to potential
region, Navier-Stokes region, and tip vortex wake, breaks down. Furthermore,
the strong turbulent wake from stall interacts with the full potential field, making
it difficult for the hybrid solver to converge. In this regime, either the full NavierStokes solver must be used or a numerical smoothing has to be used at the match
surface between the viscous and inviscid regions to eliminate the turbulent
information from the Navier-Stokes region.
Note that the hybrid method can predict the rotor performance in the turbulent
windmill state well before full stall conditions occur. The reason is that the
vorticity is generally confined to a narrow region, and the geometry of tip vortex
wake may be determined by empirical (e. g. Glauert48) methods.
cxii
wind turbine measurements were acquired in the NASA Ames National Full Scale
Aerodynamics Complex 80120 wind tunnel. To assess the fidelity, robustness, and
efficiency of existing wind turbine computational models, NREL invited developers of
wind turbine aerodynamics codes to participate in a blind run comparison in December
2000. The present hybrid methodology and full Navier-Stokes methodology were used in
the Blind Run Comparison for 6 axial flow conditions.
This new wind turbine is referred to in NREL literature as the NREL new 10meter wind turbine, or the Phase VI Rotor. The Phase VI Rotor is a 2 bladed rotor with
tapered and twisted blades. The NREL S809 airfoil was used. The rotor diameter is
10.06m. Figure 5-20 shows the taper and twist distribution for the blades of the Phase VI
Rotor. Details of the geometry of the Phase VI Rotor are documented in Appendix A.
30
25
20
15
10
5
0
0.0
1.0
2.0
3.0
4.0
5.0
6.0
-5
The data submitted in the form required by NREL was extracted by following
method:
cxiii
Air speeds and flow angles, at radius locations of 0.30R, 0.47R, 0.63R, 0.80R and
0.95R, and points that are 0.80c ahead of the leading edge on the chord line, were
extracted by computing the sectional dynamic pressures and impinging angles. The
pressure distributions at these spanwise locations were integrated to yield sectional
aerodynamic forces and moments. Sectional aerodynamics coefficients were computed
using the sectional dynamic pressures measured at these locations. The location where
the speed and inflow angle measurements at a typical radial location are required is
shown below as figure 5-21.
-0.5
C
0.8C
0.03m
Measured Point
0.5
The air velocity components were interpolated at this location from CFD results.
No further correction was applied. The aerodynamic coefficients and sectional loads
were computed based on the equations provided by NREL in the Blind Comparison
Overview. The normal force coefficients below are defined as sectional force coefficients
normal to the local chord.
Figure 5-22 shows the normal force coefficients vs. wind speed at 95% span. The
hybrid methodology was used for wind speeds 7m/s, 10m/s and 15m/s, and the full
cxiv
Navier-Stokes methodology was used for higher wind speeds. The results from present
methodologies compare favorably with measurements.
At the time these results were submitted to NREL for blind run comparison, one
of the calculations, at 13m/sec wind speed had not been carried out for full convergence.
These calculations have been redone, and fully converged results are shown here.
1.5
NREL
Hybrid
Navier-Stokes
1
0.5
0
5
10
15
20
25
30
Figure522.The95%RNormalForceCoefficients
In our simulations, the grid is usually clustered near the tip where much of the
power generation occurs. Figure 5-23 shows the 30% radius normal force coefficients.
The grid is very sparse near the root, and the local velocity component due to rotor
rotation is much lower than at the tip. For these reasons, the correlation between the
measurements and experiments was the worst near the root.
cxv
2.5
2
1.5
NREL
Hybrid
Navier-Stokes
1
0.5
0
5
10
15
20
Wind Speed (m/s)
25
30
Figure523.30%SpanNormalForceCoefficients
5000
4000
3000
NREL
Hybrid
Navier-Stokes
2000
1000
0
5
10
15
20
25
30
Figure524.FlapBendingMomentforOneBlade
Figure 5-24 shows the flap bending moment for one of the blades at the hub
connection. The hybrid methodology over-predicted the flap bending moment at 10 m/s,
cxvi
while the full Navier-Stokes methodology under-estimated the bending moment at fully
stalled wind speeds.
Upwind Configuration, Zero Yaw
5000
4000
3000
NREL
Hybrid
Navier-Stokes
2000
1000
0
5
10
15
20
25
30
Figure525.95%SpanLocalDynamicPressure
Figure 5-25 shows the local dynamic pressure at 95% span location for all these
cases. The Navier-Stokes simulation under-estimated the dynamic pressure especially at
higher wind speeds.
cxvii
The y+ is computed using the velocity u at the grid points next to body surface
with the normal distance to the body is y:
y
yu
The y+ for the baseline grid is in the range of 10 to 20, while the y+ for the grid
refined in normal (k) direction is between 3 to 5.
Table 5-1 compares the simulated power for the Phase VI Rotor for the 7 m/s
wind condition. As is the grid is refined, the solution does approached value as expected.
Refining the grid in the normal direction has the most beneficial effects because this
refinement leads to a better prediction of the skin friction, and associated profile power
loss. It is recommended that further grid studies be done to arrive at optimum grid
spacings that give accurate results in a computationally efficient manner.
Table 5-1. Grid-sensitivity Study for the Phase VI Rotor at 7 m/s Wind
Power(kw)
Baseline
1104380
2azimuthal
2194380
2radial
1108580
2normal
11043158
7.048
6.774
6.717
6.355
Navier-Stokes
Zone increased two
fold
6.413
Measured
Value
6.033
A study was also done on the effects of the Navier-Stokes zone size on the
computed results. These calculations also shown in table 5-1. It is seen that increasing the
Navier-Stokes zone size only has a minor effect on the predictions. For this reason, all
cxviii
the remaining results were done with the baseline Navier-Stokes zone, which covers
roughly one chord length around the blade section, as discussed on pages 70 and 71.
5.2. Yaw Effects
The hybrid method discussed previously has been modified to account for three
yaw effects as discribed in the section of yaw methodology. The Phase III/IV rotor was
studied for 10 m/s wind, and a 20-degree fixed yaw condition. Figure 5-26 shows the
power generated by all three blades, and takes into account the phase difference among
the blades.
superimposed on a steady state. The power fluctuations are about 4% of the timeaveraged power. The time-averaged values are in good agreement with NREL data.
7.65
7.60
Power (KW )
7.55
7.50
7.45
7.40
7.35
7.30
7.25
0
50
100
150
200
250
300
350
400
Azimuthal Angle
cxix
Figure 5-26 is the result after Fourier filtering of the data because the raw
numerical results contain noise from the numerical procedure. For example, in the above
case, the wake induced velocity is updated every 10 degrees. So there is numerical noise
of wave number 36, and harmonics higher than 36 are meaningless. Figure 5-27 shows
the Fourier coefficients of the higher harmonics.
harmonics 3, 9, 15, all of which bear a common factor of 3, the number of blades. Its
interesting that the effects of harmonic 15 are higher than harmonic 12.
Amplitude
0.06
0.05
0.04
0.03
0.02
0.01
0
0
10
15
20
25
30
Wave Number
Measurements for the Phase IV rotor of NREL were comparied with results from
the Hybrid solver. The Phase IV rotor of NREL has the same geometry as the Phase III
rotor, but has improved measurement devices. A time series of unsteady measurements
cxx
last 16 seconds, or 18 revolutions of the rotor. The measured data blend the effect of yaw
and unsteady inflow wind, and other noise from other sources, such as tower effect and
ground wind shear effect. Figure 5-28 shows the unsteadiness of the natural inflow wind.
Inflow wind(3)
inflow wind(4)
11.0
10.5
10.0
Wind
9.5
9.0
8.5
8.0
0
10
12
14
16
Time(sec)
Figure 5-28. Natural Inflow Wind in the Time Series for 18 Revolutions
Phase IV Measured Power Vs. Time at 20 degree Yaw
(NREL data for a typical wind condition)
8.5
8.0
7.5
Power
7.0
Present Calculations 10 m/s
6.5
6.0
0
45
90
135
180
225
270
315
360
cxxi
Figure 5-29 compares the unsteady hybrid results with the measured data for the
most typical revolutions. Approximately the same amplitude of osscilation is observed.
From the experimental results, a wave number 3 may be found to be dominant.
Figure 5-30 shows the unsteady power curve vs. azimuthal angle for the Phase III
rotor as predicted by the YawDyn code, which is based on BEM theory. The numerical
condition corresponds to an 8 m/s wind and 5 degree yaw. YawDyn code can predict the
amplitude of power oscillation reasonably for attached flow. Furthermore, YawDyn is
fast even for yaw computation. However, when the dynamic stall happens due to large
yaw angle and large wind speed, YawDyn requires better 2-D curve fits for dynamic stall.
The curve-fit should consider 3-D effects and the time/frequency factors. Figure 5-31
shows the unsteady power curve vs. azimuthal angle for a 10 m/s and 20 yaw by
YawDyn, using quasi-steady aerodynamics.
6.895
6.890
Power
6.885
6.880
6.875
6.870
6.865
6.860
360
410
460
510
560
610
Azimuthal Angle
cxxii
660
710
760
Figure 5-30. YawDyn Code for 8m/s Wind and 5 degree Yaw
Power Predicted
10.08
10.07
10.06
10.05
10.04
10.03
10.02
0
50
100
150
200
250
300
350
Azimuthal Angle
Figure 5-31. Results from YawDyn Code for 10m/s Wind and 20 Yaw
The overset grid method (CHIMERA Scheme) has been used to study the tower
shadow effect for the Phase III/IV rotor.
independently. Dr. Earl Duque in NASA Ames offered the detail geometry for the NREL
HAWT with the Phase II Rotor.
simulation is derived from Earl Duques grid. The Phase III/IV rotor geometry was used.
cxxiii
Because of the extensive CPU resources needed, calculation to date have been on
a very coarse grid. The nacelle is omitted, and very limited grid points were used for
each block. The blade grid dimension is 613324 for each blade. The tower grid was
generated using GRIDGEN software, its dimension is 513130. With such a grid, the
present overset grid based methodology can easily run on the SGI Origin 2000, or the
SGI Octane 2 workstation and provide limit cycle (i.e. periodic) solutions in
approximately 20 to 40 hours of CPU time.
The tower shadow effects for the Phase IV rotor in 10 m/s wind have been
studied. First the velocity field downstream of the tower was examined. Figure 5-33
shows the the tower shadow causes 15% variation of mean wind speed. However, Figure
cxxiv
5-33 does not show the turbulent wake downstream of the tower. This is because the
tower grid is too sparse downstream of the tower cylinder. There are only 30 grid points
in the circumferential direction, which means the adjacent grid point is 12 away. Also
the large grid spacing in the radial direction makes the simulation of the boundary layer
difficult.
Figure 5-33. Tower Shadow for the NREL CER Turbine in 10m/s Wind
Converged results were obtained after the rotor rotated 4 revolutions. Figure 5-43
shows the pressure coefficient distribution at 80% radius for the Phase IV rotor in a
10m/s wind. The numerical results were extracted from the three blades that are at three
azimuthal positions in rotor tip path plane, when the reference blade is behind tower
shadow. The field measurements contain significant amount of turbulence. This is the
reason it was represented by range bars. Figure 5-34 highlights that the sectional lift
cxxv
decreases when the wind blades are in the tower shadow. This decrease in lift primarily
comes from the pressure increase on the upper airfoil surface.
Correlation of CHIMERA Simulation with Measurement for the Phase III/IV Rotor
Cp Distributation at 0.8R at 10m/s Inflow Wind
-1.0
-0.5
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.5
Tower Shadow
+120 degree
-120 degree
1.0
1.5
In the present case, the changes in pressure distribution are from the decrease of
transient local impinging angle, as the tower shadow slows down the wind speed. The
rotor rotates at 72 rpm, which means a blade rotates a revolution in less than one second,
and the blade tip pass tower shadow in less than 0.015 second. From this point of view,
one may be tempted to conclude that the primary tower shadow effects are on the blade
fatigue life and noise issues other than mean power performance. Unfortunately this is
not true.
According to Robinson et al.,51 Song et al.52, and Schreck et al.,53 the tower
shadow not only influences the transient local angle of attack behind the tower, but it also
cxxvi
significantly influences the dynamic stall circle and stall delay phenomena. This is
important when the local impinging angle is close to the stall angle. For the wind turbine
at operational conditions, especially for the inner untwisted blade, the local impinging
angle is large enough to cause strong interaction between local high-lift flow and the
turbulent tower wake. The stall pattern initiated by the blade passing the tower shadow
may persist for a much longer time than when it is in the shadow, reducing the overall
power output.
cxxvii
CHAPTER 6
combination of viscous regions, inviscid regions and vortices. The method solves the
costly unsteady Reynolds averaged Navier-Stokes (RANS) equations from first principles
only in the viscous region around the turbine blades. It solves the full potential equation
in the inviscid region where flow is irrotational and isentropic. The tip vortices are
simulated using a Lagrangean approach.
A full Navier-Stokes based methodology has also been developed for modeling
wind turbines at high wind conditions where extensive stall may occur. An overset grid
based version that can model rotor-tower interactions has been developed. Finally, a
blade element theory based methodology, similar in approach to such public domain
comprehensive codes, YawDyn, has been developed for the purpose of improving tip loss
and stall delay models.
The effects of turbulence are simulated using either a zero equation eddy viscosity
model, or a one equation Spalart-Allmaras model. Two transition models, one based on
cxxviii
the Epplers criterion, and the other based on Michels criterion, have been developed and
tested.
This method has been extensively validated for axial wind conditions for three
rotors NREL Phase II, Phase III, and Phase VI configurations. A limited set of
calculations has been done for rotors operating under crosswind or yaw conditions.
Preliminary simulations have also been carried out to assess the effects of the tower wake
on the rotor. In most of these cases, satisfactory agreement has been obtained with
measurements.
Based on the work done to date, the following conclusions and recommendations
may be drawn:
1. The Hybrid methodology is an efficient means of studying the HAWT flow
phenomena for both axial and yaw conditions. This approach captures the tip vortices
using a Lagrangean approach, removing the need to accurately resolve them on a fine
grid, and gives good results with modest CPU resources.
2.
3. In the present methodology transition models of Eppler and Michel give comparable
transition locations. Epplers model also assumes that transition will occur if there is
cxxix
6.2 Recommendations
1. Only a limited number of calculations have been done in this study for wind turbines
operating in yaw conditions, and turbines operating in tower shadow. One reason for
this is the lack of good quality field data. However, the comprehensive set of data
obtained by NREL for the Phase VI rotor, will greatly aid the validation and further
cxxx
distributing the workload among the processors must be explored, leading to further
reduction in CPU time, and more importantly, the design cycle.
It is hoped that this work will serve as a useful stepping stone for pursuing further
studies in this exciting branch of renewable energy systems.
cxxxi
APPENDIX A
In this appendix the geometric characteristics of the NREL Phase II, III and IV
rotors are given.
The Phase II and III Rotors
Field measurements for the NREL Phase II Rotor, the NREL Phase III Rotor and
the NREL Phase IV Rotor were stored into the database managed by IEA Annex XIV:
Field Rotor Aerodynamics.45,54,55 The primary differences of the Phase III Rotor and the
Phase IV Rotor are the local flow angle (LFA) measurement devices and the span
locations instrumented with LFA sensors. The geometry of the Phase III Rotor and that
of the Phase IV Rotor are the same.
Overall characteristics:
Number of blades: 3
Blade profile: NREL S809 as in table 4 at the end of Appendix A
Rotor diameter: 10.06 m
Hub height: 17.03 m
Type rotor: fixed
Rotor angular speed: 71.63 rpm
Tilt: 0
Cone: 3.42
Location of rotor: downwind
Rotor Geometry:
Twist []
44.67
39.39
32.39
26.56
21.95
18.19
15.10
12.52
10.35
8.50
6.91
5.52
4.32
3.25
2.30
1.45
0.69
0.00
cxxxiii
Table A-2. Distribution of Blade Taper/Twist for the NREL Phase VI Rotor
Radial
Distance (m)
0.000
0.724
0.838
0.968
1.258
1.522
1.798
2.075
2.352
2.628
2.905
3.181
3.458
3.735
3.772
4.011
4.288
4.565
4.841
Chord (m)
Twist ()
Hub
Hub
Not supplied
Not supplied
0.737
0.710
0.682
0.654
0.626
0.598
0.570
0.542
0.514
0.486
0.483
0.459
0.431
0.403
0.375
0.00
0.00
30.00
27.59
20.05
14.04
9.67
6.75
4.84
3.48
2.40
1.51
0.76
0.09
0.00
-0.55
-1.11
-1.55
-1.84
cxxxiv
5.030
0.356
-2.00
y/c
0.000000
-0.004980
-0.012720
-0.021620
-0.031440
-0.041990
-0.053010
-0.064080
-0.074670
-0.084470
-0.093260
-0.100600
-0.105890
-0.108660
-0.108420
-0.104840
-0.097560
-0.086970
-0.074420
-0.061120
-0.047920
-0.035580
-0.024660
-0.015590
-0.008590
-0.003700
-0.000750
0.000540
0.000650
0.000240
0.000000
0.000000
x/c
0.000000
0.000370
0.005750
0.016260
0.031580
0.051470
0.075680
0.103900
0.135800
0.171030
0.209200
0.249870
0.292590
0.336890
0.382230
0.428090
0.473840
0.520050
0.568010
0.617470
0.667180
0.716060
0.763140
0.807560
0.848540
0.885370
0.917630
0.945230
0.967990
0.985280
0.996230
1.000000
Lower Surface
y/c
0.000000
0.002750
0.011660
0.021330
0.031360
0.041430
0.051320
0.060820
0.069720
0.077860
0.085050
0.091130
0.095940
0.099330
0.101090
0.101010
0.098430
0.092370
0.083560
0.073790
0.064030
0.054620
0.045780
0.037610
0.030170
0.023350
0.016940
0.011010
0.006000
0.002450
0.000540
0.000000
Upper Surface
cxxxv
cL
cD
. 000
0.1748
. 230
0.2116
. 460
0.3172
. 494
0.4784
. 510
0.6743
. 486
0.8799
. 415
1.0684
. 302
1.2148
. 159
1.2989
. 000
1.308
-. 159
1.2989
-. 302
1.2148
-. 415
1.0684
-. 486
0.8799
-. 510
0.6743
-. 494
0.4784
-. 560
0.3027
-. 670
0.3069
-. 790
0.1928
-. 840
0.0898
-. 700
0.0553
-. 630
0.039
-. 560
0.0233
-. 640
0.0131
-. 420
0.0134
-. 210
0.0119
. 050
0.0122
. 300
0.0116
. 540
0.0144
. 790
0.0146
. 900
0.0162
. 930
0.0274
. 920
0.0303
. 950
0.0369
. 990
0.0509
1.010
0.0648
1.020
0.0776
To be continued
cxxxvi
cM
0.0000
0.4000
0.1018
0.1333
0.1727
0.2132
0.2498
0.2779
0.2933
0.2936
0.2933
0.2779
0.2498
0.2132
0.1727
0.1333
0.0612
0.0904
0.0293
-0.0090
-0.0045
-0.0044
-0.0051
0.0018
-0.0216
-0.0282
-0.0346
-0.0405
-0.0455
-0.0507
-0.0404
-0.0321
-0.0281
-0.0284
-0.0322
-0.0361
-0.0363
cL
1.000
. 940
. 850
. 700
. 660
. 705
. 729
. 694
. 593
. 432
. 227
. 000
-. 159
-. 302
-. 415
-. 486
-. 510
-. 494
-. 460
-. 230
. 000
cD
0.0917
0.0994
0.2306
0.3142
0.3186
0.4784
0.6743
0.8799
1.0684
1.2148
1.2989
1.308
1.2989
1.2148
1.0684
0.8799
0.6743
0.4784
0.3172
0.2116
0.1748
cxxxvii
cM
-0.0393
-0.0398
-0.0983
-0.1242
-0.1155
-0.2459
-0.2813
-0.3134
-0.3388
-0.3557
-0.3630
-0.3604
-0.3600
-0.3446
-0.3166
-0.2800
-0.2394
-0.2001
-0.1685
-0.5000
0.0000
APPENDIX B
Boundary layer characteristics used in section 3.4 are computed from the
chordwise pressure distribution at each span location. These boundary properties include
the momentum thickness, ; energy thickness, 3; the boundary layer edge velocity, u e;
the shape factor, H, and ratio H32 = 3/.
The edge velocity, ue, is computed using the potential flow relation,
cp 1
where
u e2
V2
(B-1)
the radius, r. The pressure coefficient CP is extracted from the present CFD simulations.
It has been observed that most laminar boundary layers obey the following
relationship56
ue d 2
2 du e
A B
dx
dx
(B-2)
Thawaites recommends A= 0.45 and B=6 as the best empirical fit. The above equation
may be analytically integrated yielding
0.45
u e6
2
u e6 x 0
5
u
dx
e
u e6 x
x 0
x
cxxxviii
(B-3)
For blunt bodies such as airfoils, the edge velocity u e is zero at x=0, the stagnation
point. For sharp-nosed geometries such as a flat plate, is zero at the leading edge.
Thus, the term in the square bracket always vanishes.
Starting from the leading edge, ue is computed at each grid point on the airfoil
surface,
du e
is computed by central differences, as shown in figure A2-1. For example,
dx
dx
S
(B-4)
S
S
i-1
i+1
i+2
u
Re e
V
ue
x0 V
0.45
6
cxxxix
x
c
(B-4)
where
for
0.0147
0.107
0 0.1
for 0.1 0
(B-5)
2 du e
.
dx
(B-
6)
2
H 79.870845 89.582142 H 32 25.715786 H 32
(B-7)
u
3 1
ue
wall
dy
ue
and where u(y) is the velocity profile as a function of y, the distance from the wall.
cxl
(B-8)
APPENDIX C
The strip theory combines the actuator disk model, and the blade element method
(BEM). It is an intermediate theory bridging the gap between actuator disk models of
wind turbines and a rigorous vortex theory. The strip theory presented here has been
sometimes called the modified blade element theory (Glauert56).
(C-1)
where
T = thrust force on the disk (N)
A = Disk area (m2)
= air mass flow rate through the disk (kg/s)
M
a
(C-2)
where Pu and Pd are pressures upwind and downwind of the disk, respectively (N/m2).
Applying the Bernoulli equation between the free-stream and the upwind side of
the rotor disk and again between the downwind side of the turbine and the far wake,
equation (C-2) becomes
T
1
AU 2 V12
2
(C-3)
(C-4)
U V aU
(C-5a)
Define
Then
U V1 2aU
(C-5b)
cxlii
(C-6)
T
1
AU 2
2
41 a a
(C-7)
1
1
AU 2 V12 V AV U V1 U V1
2
2
(C-8)
P
1
U 3 A
2
4 a 1 a
(C-9)
Stream Tube
Figure C-2. A Streamtube in Strip Theory
The other two assumptions are that the spanwise flow is negligible, and the flow
conditions do not vary in the circumferential direction.
cxliii
(C-10)
where is the angular velocity in the fluid at rotor, is the rotor speed (rad/s).
A relationship between the rotation factor, a, and axial induction factor, a, is
given by, 58
a
1 3a
4a 1
cxliv
(C-11)
As shown in figure C-3, Vn is the wind velocity normal to the rotor disk; v is the
in-plane velocity component from both the rotor and the wake rotation. The total velocity
is Vr. The sectional thrust coefficient, Ct, for the streamtube can be written as,
Ct
dT
(C-12)
1
U 2 2r dr
2
R
Referring to figure C-3, the thrust increment, dT, for a rotor with N b blades and of
chord c(r) is,
dT
1
Vr2 N b c C L cos C D sin dr
2
(C-13)
Using equation (C-12), and referring to figure C-3, C t for the annulus can be
written as,
Ct
Nb c
1
2
C L cos C D sin
1 a
2
2 r
sin
(C-14)
Introducing Prandtls tip loss factor, F, as shown by Wilson and Walker,59 equation
(C-7) can be evaluated as,
C t 4aF 1 a
C t 4 F a c2 1 2a c a
if a a c
(C-15a)
if a a c
(C-15b)
where, ac 0.2
Further details of the Prandtls tip loss factor F are documented in Appendix D.
Solving Procedure:
cxlv
1
3. Compute tan
4. Compute after finding local pitch angle, from tip pitch, tip, and twist
distribution, tw(r).
5. Find CL and CD from S809 aerodynamic characteristics, Table A-4.
6. Compute Ct using equation (C-14).
7. Find a new axial induction factor, anew.
8. If anew equals a, convergence has been found;
Otherwise set a equals anew, go to step 2.
cxlvi
APPENDIX D
Because first-principles-based schemes such as the one discussed in this thesis are
computationally expensive, wind turbine designers will continue to use comprehensive
analysis tools such as AeroDyn. These comprehensive codes use a blade element theory.
They rely on empirical models for the tip losses (i. e. loss of lift near the tip), tip vortex
roll-up, and dynamic stall/stall delay. It is very hard to develop appropriate models for
these important effects from measurements alone.
recommendations based on the present study is made for improving these models.
f r
Nb
2
r
1
R sin R
cxlvii
(D-1)
(D-2)
where R is the angle between the relative wind vector and the plane of rotation at
the tip, as shown in figure C-3 in Appendix C. In practice, Prandtls expression for tip
loss factor is often modified to:
f r
Nb
2
r 1
R sin
(D-2a)
Figure D-1 shows the radial variation of F at a number of wind speeds and
number of blades conditions for the Phase VI Rotor. Prandtls model assumes that the tip
loss factor is zero at tip, and asymptotically approaches unity as r goes to zero. This
model tends to predict higher tip losses for a small number of blades, and at high wind
speeds. The behavior of this model is, however, physically correct for attached flow
conditions.
Nb=4; 7m/s
0.8
Nb=2; 7m/s
Nb=4; 25m/s
0.6
F
Nb=2; 25m/s
0.4
0.2
0
0
0.2
0.4
0.6
0.8
r/R
cxlviii
This model is used in blade element theory in one or more of the following ways:
a) The bound circulation estimated from 2-D strip theory is corrected by :
3 D strip theory F
(D-3)
where strip theory = bound circulation of a rotor with infinte number of blades.
b) The inflow velocity predicted by momentum theory is corrected:
vinduced ( r ) F vmomentum theory
(D-4)
(D-5)
Figure D-2 shows the spanwise distribution of the Prandtls tip loss factor, F, for
the Phase VI Rotor using equation (D-1) and (D-2a). This model predicts tip losses keep
increasing with the wind speed even under post-stall conditions.
cxlix
0.8
25m/s
0.6
F
20m/s
0.4
15m/s
7m/s
0.2
0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
r/R
Figure D-2. The Tip Loss Factor for the Phase VI Rotor by Strip Theory
inboard stations, the blade sections often experience stall delay due to
c
, as shown
r
in Appendix E, are higher for inbound stations. In this section this stall delay
model has not been applied yet.
cl
L r
Q r cL c r
(D-6)
where sectional lift, L(r), Local impinging angle, (r), and local dynamic
pressure, Q(r), are extracted from the CFD results. CL is found from the S809
airfoil table, as documented in table A-4 in Appendix A.
c) Stall delay effects need to be considered for post-stall conditions. A constant
stall delay of 1 (near the tip) is assumed for all post-stall conditions.
Figure D-3 compares the F from the strip theory with F from CFD results for the
Phase VI Rotor at 25m/s. It should be noted that this is a fully stalled condition. The
Prandtls tip loss model with the strip theory over-estimates tip losses in the tip region,
0.8 r 1 . To highlight the variation of tip loss at tip, figure D-4 shows ratio of these
R
two F factors near the tip. The CFD results are seen to predict a higher lift near the tip,
and the sectional lift does not become zero gradually toward the tip. Significant values of
lift are generated up to 99.7% radius, beyond which the lift linearly drops to zero.
cli
0.8
0.6
0.4
F from CFD
F from Strip Theory
0.2
0
0.8
0.85
0.9
0.95
r/R
Figure D-3. Prandtls Tip Loss Factor for the Phase VI Rotor at 25m/s Wind
Ratio of F factors for the Phase VI Rotor at 25m/s
8
Fcfd/Fstrip
0
0.8
0.85
0.9
0.95
r/R
Figure D-5 shows the correlation of the Prandtls tip loss factor, F, for the Phase
VI Rotor at 20m/s. This is also a post-stall condition. Discrepancies between strip theory
and CFD results similar to the 25 m/s case are observed.
clii
0.8
0.6
0.4
F from Strip Theory
F from CFD
0.2
0
0.8
0.85
0.9
0.95
r/R
Figure D-5. The Prandtls Tip Loss Factor for the Phase VI Rotor at 20m/s
0.8
0.6
F from CFD
F from Strip Theory
0.4
0.2
0
0.8
0.85
0.9
0.95
r/R
Figure D-6. The Prandtls Tip Loss Factor for the Phase VI Rotor at 15m/s
Figure D-6 shows the correlation of CFD and the strip theory with Prandtls
model for 15 m/s. This is a wind speed just a little bit higher than the speed at which the
rotor stalls. CFD simulations show that the tip losses occur only beyond 95%R. The
cliii
CFD-based factor F is higher than 1 at some places in CFD results, indicting that an
assumed 1 stall delay near the tip is not sufficient for this condition.
Figure D-7 shows the correlation for 7m/s, which is an attached flow condition.
At this wind speed, at the inboard stations, the tip loss factor F from original Prandtls
model is higher than F from CFD. This is a reversal of what was observed earlier at poststall conditions. This discrepancy of the inboard sections does not significantly affect
predicted power production, since the sectional power (i.e. power per foot of span) varies
as
0.8
0.6
0.4
F from CFD
F from Strip Theory
0.2
0
0.8
0.85
0.9
0.95
r/R
Figure D-7. The Prandtls Tip Loss Factor for the Phase VI Rotor at 7m/s
The discrepancy between the Prandtls tip loss model and the CFD simulations is
due to the tip vortex stretching, as discussed below.
cliv
Tip
P2
Root
P1
r1
r2
Tip Vortex
Center Line
As shown in figure D-8, considering the induced velocities at two points, P 1 near
the tip and P2 at inbound location. The ratio of induced velocities at these two points
v P1
vP2
is proportional
r2
induced velocities, is thus inversely proportional to their distances from the tip vortex
section. If the tip vortex section is closer to the rotor disk, higher lift losses at the tip
region, and lower lift losses at the inboard region will be observed, and vice versa.
The operational wind speed for the Phase VI rotor is about 10m/s. According to
the actuator disk model (equation C-9) in Appendix C, the highest power occurs when the
axial induced velocity factor, a, equals to 0.3. This corresponds to a wind speed of 7 m/s
at the rotor disk. The tip rotating velocity for the Phase VI Rotor is about 38m/s. Thus
the ratio of the wind speed at the rotor to the tip speed is about 0.18.
clv
a
16
32
r 1
a
(D-7)
where a is the sectional lift curve slope, a = 5.7; solidity, , is defined as the ratio
of blade area to the rotor area, generally in the order of 0.1; is the impinging angle,
generally less than 10; r is the radius normalized by rotor radius R.
Thus, the highest spanwise value h is less than 0.07, for hoving helicopter rotors.
The mean induced velocity is even less than this value. Thus the tip vortices are closer to
the rotor for helicoptor in hover than the Phase VI rotor at operational conditions.
Prandtls model was developed for rotors in hover which typically have closely
spaced vortices as discussed above, and predicts large tip losses. The present CFD
mothodologies correctly models the larger spacing of the tip vortices, and predicts lower
tip losses.
As seen earlier, the CFD results indicates significant level of lift at both pre-stall
and post-stall conditions. This behavior is consistant with physics as ciscussed below.
The pressure at the upwind side of the rotor is higher than the pressure at the
downwind side of the rotor. This is true for both pre-stall and post-stall conditions. The
fluid moves around the tip from the blade lower surface, which is at the upwind side of
the rotor, toward the upper surface of the blade, separates, and forms a vortex. Such a tip
vortex puts more kinetic energy into the upper surface flow field, and results in more lift
near the tip. For attached flow conditions, the circulation in the tip vortex also consists of
clvi
the bound circulation of the blade, while, for post-stall conditions, the circulation is
primarily from the separation at the chordwise tip end.
It may be noted that this higher lift levels near the tip have been experimentally observed.
Gray65,66 observed similar tip loading variation in measurements of a model helicopter
rotor. According to McAlister et al67, such phenomena also occur for wing tips.
1 F r 0 .7
r
R
1 R 0.7
Fnew
for 0.7 r 1
R
for r 0.7
R
(D-8a)
0.8
Fnew
for 0.8 r 1
R
for r 0.8
R
(D-8b)
Figure D-9 correlates the Prandtls model, and the new model above with CFD
results for an attached flow condition, at 7m/s wind conditions.
clvii
0.8
0.6
0.4
F from CFD
Prandtl's Model
New Tip Loss Model
0.2
0
0.8
0.85
0.9
0.95
r/R
Figure D-9. Curve-fitting a New Tip Loss Model from CFD for 7m/s Wind
Figure D-10 shows the low speed shaft torque for the Phase VI Rotor under axial
wind conditions. The new model predicts equivalent torque at attached flow conditions,
but predict better results for post-stall conditions.
measurements and the strip theory (with the new tip loss model) is attributable to the
absence of stall delay model. An improved stall delay model is proposed in Appendix E.
2000
1500
1000
500
0
0
-500
-1000
10
15
20
25
-1500
Wind Speed
Figure D-10. Low Speed Shaft Torque for the Phase VI Rotor
clviii
Because the improved tip loss model was generated by curve-fitting the CFD
results without further tuning for the Phase VI Rotor, it needs to be applied to other
configurations to see how it performs. For this reason this model was next applied to the
Phase III Rotor, a three-bladed rotor configuration.
generated by the Phase III Rotor based on different models. The new model predicts
power curves for pre-stall conditions equivalent to the old Prandtls model, and better
results for post-stall conditions. Further improvements can be achieved with an improved
stall delay model. However, the sharp kink at the wind speed of 13m/s indicates that
there is still a significant amount of tip loss at this speed. It may be noted that the rotor is
close to stall at this speed (13 m/s). The kink at 13m/s can be corrected by smoothly
connecting equation (D-8a) and (D-8b) to ensure the tip losses decrease gradually when
the wind speed changes through the stall speed.
20000
strip; no tiploss
NREL field test
15000
strip;Prandtl's tiploss
New Model
10000
5000
0
0
10
15
20
-5000
clix
APPENDIX E
The accuracy of strip theory based procedures, such as the YawDyn code and the
present one, is dependent on airfoil characteristics, as shown in Table A-4 in Appendix A.
These airfoil characteristics consist of: the lift coefficient, CL vs. ; CL vs. CD drag polar;
and moment coefficient, CM vs. CL. These characteristics are usually obtained from 2-D
wind tunnel measurements. When the 2-D airfoil static characteristics are directly used,
it is known that strip theory based procedures under-predict the rotor power under poststall conditions, as shown in figure D-10 and figure D-11 in appendix D. These tablelook up procedures will also fail to predict the unsteady power under dynamic stall (yaw)
conditions, as shown in figure 5-31 in Chapter V.
The discrepancies between the strip theory predictions and the measurements are
attributable to centrifugal pumping effects.
clx
pressure gradient exists that also contributes to radial flow. This effect is hard to
quantify, and include in strip theory based models.
Analytical studies (Banks and Gadd71, MaCroskey and Yaggy72, Dwyer and
MaCroskey73) have shown that the centrifugal component is a function of the ratio of the
local chord to local radius, (c/r). A few stall delay models are available, such as, the
method of Eggers and Digumarthi74, and the method of Corrigan and Schillings.75
Corrigan and Schillings model is able to quantify the centrifugal pumping effects and the
effects of radial pressure gradient, and is simple to be integrated into strip theory based
methodologies. Corrigans model has been evaluated by Tangler76 et al.
The Corrigans stall delay model predicts a stall delay angle, ,
Cl , max Cl 0
K TE
0.136
l 0
l , max
(E-1)
, and the zero lift angle of attack,
(E-2)
clxi
An analytical justification for this model may be found from Corrigan 75 and
Tangler76. For n=0, equation (E-1) reduces to the characteristic of 2D data, where the stall
delay is zero. Corrigan75 indicates that a value of n between 0.8 to 1.6 provides best
correlation with most test data, and that n = 1 provides good results for most cases.
Tangler76 uses n = 1 for the evaluation of several rotor configurations.
Once the stall delay angle, , is obtained, the lift coefficient, CL, at post-stall
conditions is computed as,
C L , 3 D C L , 2 D C L, 2 D
(E-3)
where, C L , 2 D is the slope for the linear part in the CL ~ curve, and is taken to be 0.1
per degree for the NREL S809 airfoil. Figure E-2 shows the stall delay correction to 2-D
airfoil lift coefficients.
clxii
3-D
CL
2-D
The CFD simulations discussed in Chapter 5 were used to assess and improve
Corrigans model as follows. The 3-D lift coefficient, C L,3-D, and the local impinging
angle, , are extracted from CFD results. The stall delay angle, , is found using the 2D airfoil table, Table A-4 for S809 airfoil. The parameter, n, in Corrigans model is next
determined using equation (E. 1). Table E-1 shows the values of n obtained from CFD
results for the Phase VI Rotor at 20m/s wind.
r/R
0.57
0.743
()
34.47
31.37
CL,3-D
0.9773
0.9195
c/r
()
2.7
0.200
2.1
0.130
K
0.77492
1.15305
KTE/0.136
1.13959
1.10218
n
1.79
1.925
Figure E-3 shows the results for the Corrigans stall delay model for the Phase VI
Rotor using the original value n=1 suggested by Corrigan, and the value of n=1.85,
suggested by the CFD simulations. The new tip loss model discussed in Appendix D has
been applied. It must also be noted that these results were obtained using the strip theory
discussed in Appendix C.
2000
NREL NASA Ames
New Tip Loss Mode;no stall delay
Corrigan Model; n=1
Corrigan Model; n=1.85
1500
1000
500
0
0
10
15
Wind Speed
20
25
clxiv
20,000
NREL field test
strip;Prandtl's tiploss
New Tip loss;Corrigan, n=1
New Tip loss; Corrigan, n=1.85
Generated Power(w)
15,000
10,000
5,000
0
0
10
15
20
-5,000
Wind Speed (m/s)
clxv
REFERENCES
Brower, M. S., Tennis, M. W., Denzler, E. W. and Kaplan, M. M., (1993). Powering
the Midwest: Renewable Electricity for the Economy and the Environment.
Cambridge, MA: Union of Concerned Scientists; 188 pp.
Cavallo, A. J., Hock, S. M. and Smith, D. R. (1993). Wind Energy: Technology and
Economics. Chapter 3 in Renewable Energy: Sources for Fuels and Electricity.
Edited by T. B. Johansson, H. Kelly, A. K. N. Reddy, and R. H. Williams.
Washington, DC: Island Press; pp. 121-156.
Walter Frost and Carl Aspliden, Characteristics of the Wind, Chapter 8 in Wind
Turbine Technology. Edited by David A. Spera, New York, ASME Press, 1994.
10 Tangler, J. L., Smith, B. and Jager, D. (1992). SERI Advanced Wind Turbine
Blades, NREL/TP-257-4492.Golden, CO.
11 Eppler, R. (1990). Airfoil Design and Data, New York, NY: Springer-Verlag; 562 pp.
12 Selig, M. S., Donovan, J. F. and Fraser, D. B. (1989). Airfoils at Low Speeds,
Soartech 8.Virginia Beach, VA: H. A. Stokely.
clxvi
clxvii
26 Hirsch, C., Numerical Computation of Internal and External Flows, Volume I and
Volume II, John Wiley and Sons, 1987.
27 Vinokur, M., An Analysis of Finite Difference and Finite Volume Formulations of
Conservation Laws, Journal of Computational Physics, Vol. 81, 1989, pp. 1-52.
28 Roe, P. L., Approximate Riemann Solvers, Parameter Vectors, and Difference
Schemes, Journal of Computational Physics, Vol. 43, 1981 pp. 357-372.
29 van Leer, N., Upwind Difference Methods for Aerodynamic Problems Governed by
the Euler Equations, Lectures in Applied Mathematics, Vol. 22, 1985.
30 Baldwin, B. S. and Lomax, H. Thin Layer Approximation and Algebraic Model for
Spearated Turbulent Flow, AIAA Paper 78-0257, Jan. 1978.
31 P. R. Spalart and S. R. Allmaras, "A One-Equation Turbulence Model for
Aerodynamic Flows," AIAA-92-0439.
32 M. Shur, M. Strelets, A. Travin, and P. R. Spalart, Turbulence Modeling in Rotating
and Curved Channels: Assessment of the Spalart-Shur Correction Term, 36 th
Aerospace Sciences Meeting & Exhibit, January, 1998, AIAA 98-0325.
33 R. Michel, D. Arnal, and E. Coustols, "Stability Calculations and Transition Criteria
on Two- or Three-Dimensional Flows," Conference Paper, Laminar-Turbulent
Transition. Novosibirsk, USSR July 09-13, 1984.Published by Springer-Verlag,
Berlin, West Germany and New York, NY, USA pp. 455-461.
34 Chen, K. K., and Thyson, N. A., "Extension of Emmons' Spot Theory to Flow on
Blunt Bodies," AIAA Journal, Vol. 9, 1971, pp. 821-825.
35 Tuncer Cebeci, "Essential Ingredients of a Method for Low Reynolds-Number
Airfoils," AIAA Journal Vol. 27, No. 12, December 1989.pp. 1680-1685.
36 Beam, R. and Warming, R. F., An Implicit Finite Difference Algorithm for
Hyperbolic Systems in Conservation Law Form, Journal of Computational Physics,
Vol. 22, Sept 1976.
37 Pulliam, T. H. and Chaussee, D. S., A Diagonal form of an Implicit ApproximateFactorization Algorithm, Journal of Computational Physics, Vol. 39, 1981.
38 Bangalore, A., Computational Fluid Dynamic Studies of High Lift Rotor Systems
Using Distributed computing, Ph.D. Dissertation, School of Aerospace Engineering,
Georgia Institute of Technology, Atlanta, GA, May 1994.
clxviii
39 Sankar, L. N. and Prichard, D., Solution of Transonic Flow Past Rotor Blades Using
the Conservative Full Potential Equation, AIAA Paper 85-5012, October 1985.
40 Yang, Z., A Hybrid Flow Analysis for Rotors in Forward Flight, Ph.D. Dissertation,
School of Aerospace Engineering, Georgia Institute of Technology, August 2000.
41 Sankar, L. N., Malone, J. B., Tassa, Y., An Implicit, Conservative Algorithm for
Steady and Unsteady Transonic Potential Flows, Proceedings of the AIAA 5 th
Computational Fluid Dynamics Conference, 1981.
42 Sankar, L. N., Bharadvaj, B. K., Tsung, F. L., A Three-Dimensional NavierStokes/Full Potential Coupled analysis for Viscous Transonic Flow, AIAA Journal,
Vol. 31, No. 10, 1993.
43 Johnson, W., Development of a Comprehensive Analysis for Rotorcraft I. Rotor
Model and Wake Analysis, Vertica, Vol. 5, 1989, pp. 90-130.
44 Giguere, P., Selig, M. S. (1999), Design of a Tapered and Twisted Blade for the
NREL Combined Experiment Rotor, March 1999, NICH Report No. SR-500-26173.
45 Simms, D. A.; Hand, M. M.; Fingersh, L. J. ; Jager, D. W. (1999). Unsteady
Aerodynamics Experiment Phases II-IV: Test Configurations and Available Data
Campaigns. 178 pp.; NICH Report No. TP-500-25950.
46 Wolfe, W. P. and Ochs, S. S., "CFD Calculations of S809 Aerodynamic
Characteristics," AIAA paper 97-0973, January 1997.
47 Wolfe, W. P. and Ochs, S. S., "Predicting Aerodynamic Characteristics of Typical
Wind Turbine Airfoils Using CFD," Sandia Report SAND96-2345, September 1997.
48 Glauert, H. Airplane Propellers, From Div. L, Aerodynamic Theory, ed. W. F.
Durand. Berlin: Springer Verlag, 1935.(Reprinted by Peter Smith, Glouster,
Mass.,1976)
49 Wilson, R. E., and Lissaman, P. B. S. Applied Aerodynamics of Wind Power
Machines, Oregon State University, 1974.
50 David M. Eggleston and Forrest S. Stoddard, Wind Turbine Engineering Design, Van
Nostrand Reinhold, 1987.
51 Robinson, M. C.; Hand, M. M.; Simms, D. A.; Schreck, S. J. (1999). Horizontal Axis
Wind Turbine Aerodynamics: Three-Dimensional, Unsteady, and Separated Flow
Influences. 13 pp.; NICH Report No. CP-500-26337.
clxix
52 Song, Y. D.; Robinson, M. (2000). Guest Editorial for the Special Issue on Wind
Turbines: Dynamics, Control and Monitoring. Journal of Wind Engineering and
Industrial Aerodynamics. Vol. 85, 2000; pp. 209-210; NICH Report No. 28690.
53 Schreck, S.; Robinson, M.; Hand, M.; Simms, D. (2000). HAWT Dynamic Stall
Response Asymmetries Under Yawed Flow Conditions. 16 pp.; NICH Report No.
CP-500-27898.
54 Simms, D. A.; Fingersh, L. J.; Butterfield, C. P. (1995). NREL Unsteady
Aerodynamics Experiment Phase III Test Objectives and Preliminary Results. Wind
Energy 1995: Proceedings of the Energy and Environmental Expo '95, the EnergySources Technology Conference and Exhibition, 29 January - 1 February 1995,
Houston, Texas. SED-Vol. 16, New York: American Society of Mechanical Engineers,
pp. 273-275.
55 Fingersh, L. J.; Simms, D. A.; Butterfield, C. P.; Jenks, M. D. (1995), Overview of
the Unsteady Aerodynamics Experiment Phase III Data Acquisition System and
Instrumentation, Musial, W. D.; Hock, S. M.; Berg, D. E., eds. Wind Energy 1995:
Proceedings of the Energy and Environmental Expo '95, the Energy-Sources
Technology Conference and Exhibition, 29 January - 1 February 1995, Houston,
Texas. SED-Vol. 16.New York: American Society of Mechanical Engineers; pp. 277280.
56 Glauert, H., 1935, Airplane Propellers, Aerodynamic Theory, Vol. 4, Div. I, W. F.
Durand, ed., Berlin: Julius Springer, pp. 169-360.
57 Eppler, A Computer Program for the Design and Analysis of Low Speed Airfoils,
NASA TM 80210, August 1980.
58 Robert E. Wilson, Aerodynamic Behavior of Wind Turbines, Chapter 5 in Wind
Turbine Technology. Edited by David A. Spera, New York, ASME Press, 1994.
59 Wilson, R. E., and S. N. Walker, 1984, Performance Analysis of Horizontal Axis
Wind Turbines, Corvallis, Oregon: Oregon State University.
60 Prandtl, L. 1919, Appendix to [Betz 1919] see below: pp. 213-217.
61 Betz, A., 1919, Schraubenpropeller mit geringstem Energieverlust, Gottinger
Nachrichten, mathematisch-physikalische Klasse: pp. 193-213.
62 Goldstein, S., 1929, On the Vortex Theory of Screw Propellers, Proceedings, Royal
Society, A 123: pp. 440-465.
clxx
clxxi
76 Tangler J. L. and Selig, M. S., An Evaluation of an Empirical Model for Stall Delay
due to Rotation from HAWTS, Presented at Windpower 97, Austin, Texas, June 1518, 1997.
clxxii
VITA
Guanpeng Xu, son of Litian Xu and Songyun Yang, was born in Heilongjiang
Province, China on February 3, 1971.
He finished high school and enrolled into the Department of Aerodynamics,
NanjingUniversityofAeronauticsandAstronautics(NUAA)inChinaattheageof14.
HegraduatedwiththeBachelorofEngineeringSciencedegreefromNUAAin1989.
Aftergraduation,Mr.XuworkedonwindtunnelexperimentsatHarbinAerodynamic
ResearchInstitute(HARI)fortwoyears. Hethenenrolledinajointmasterprogram
betweenBeijingUniversityofAeronauticsandAstronautics(BUAA)andtheChinese
AeronauticsEstablishment(CAE)andreceivedtheMasterofEngineeringSciencedegree
in1994.HeworkedforanotherthreeyearsasaCFDengineeratHARIbeforejoiningthe
School of Aerospace Engineering at Georgia Institute of Technology as a Graduate
ResearchAssistantinSeptember1997.SinceMay2000hehasinternedasaResearch
Associate at the Center of Excellence of Information System and Management in
TennesseeStateUniversity.Mr.XuhaspresentedscientificpapersatseveralAmerican
InstituteofAeronauticsandAstronautics(AIAA)conferences,ofwhichheisamember.
Mr.XualsohaspublishedintheJournalofSolarEnergyEngineering.
InMay2000,Mr.XumarriedtheformerHongliHeinAtlanta,Georgia.
clxxiii