You are on page 1of 29

AIAA 2002-0037

Challenges in Modeling the


Unsteady Aerodynamics of Wind
Turbines
J. Gordon Leishman
Department of Aerospace Engineering,
Glenn L. Martin Institute of Technology,
University of Maryland at College Park,
Maryland 20742.

21st ASME Wind Energy Symposium and the 40th


AIAA Aerospace Sciences Meeting, Reno, NV
For permission to copy or republish, contact the author or the American Institute of Aeronautics and Astronautics
1801 Alexander Bell Drive, Suite 500, Reston, VA 201914344

Challenges in Modeling the Unsteady


Aerodynamics of Wind Turbines
J. Gordon Leishman
Department of Aerospace Engineering,
Glenn L. Martin Institute of Technology,
University of Maryland at College Park,
Maryland 20742.
Abstract
Many of the aerodynamic phenomena contributing to
the observed effects on wind turbines are now known, but
the details of the flow are still poorly understood and are
challenging to predict accurately. Issues discussed herein
include the modeling of the induced velocity field produced by the vortical wake behind the turbine, the various
unsteady aerodynamic issues associated with the blade sections, and the intricacies of dynamic stall. Fundamental
limits exist in the capabilities of all models, and misunderstandings or ambiguities can also arise in how these
models should be properly applied. A challenge for analysts is to use complementary experimental measurements
and modeling techniques to better understand the aerodynamic problems found on wind turbines, and to develop
more rigorous models with wider ranges of application.

Nomenclature
a
a
A
Ai
bi
c
C
Cl
Cm
Cl
CT
h
M
Nb
q
R
s
S
S

sonic velocity, ms1


pitch axis location, measured from mid-chord
rotor disk area, R2 , m2
coefficients of indicial functions
exponents of indicial functions
chord, m
Theodorsen function
lift coefficient
moment coefficient about 1/4-chord
lift curve slope, rad 1
rotor thrust coefficient, T /(2 R4 )
plunge displacement, m
free-stream Mach number
number of blades

nondimensional pitch rate, = c/V


rotor radius, m

distance in semi-chords, = (2/c) 0t V dt
Sears function (referenced to airfoil mid-chord)
Sears function (referenced to airfoil leading-edge)

t
T
UR
UT
vi
V
Vg
V
V
Vex
Vind
w
x, y, z

g
i

Abbreviations
CFD
Computational fluid dynamics
HAWT Horizontal axis wind turbine
NREL National Renewable Energy Laboratory

Professor.

Email: leishman@eng.umd.edu
Paper 2002-0037. Presented at the 21st ASME Wind Energy Symposium and the 40th AIAA Aerospace Sciences Meeting, Reno,
NV, Jan. 1417, 2002.
c 2002 by J. G. Leishman. Pubished by the American Institute of Aeronautics and Astronautics,
Copyright 
Inc. and the Institute of Mechanical Engineers, with permission.

time, s
rotor thrust, N
velocity component parallel to blade, ms1
velocity component perpendicular to blade, ms1
average induced velocity, ms1
airfoil velocity, ms1
gust convection velocity, ms1
velocity vector at a Lagrangian marker, ms1
free-stream velocity, ms1
external velocity field, ms1
induced velocity, ms1
gust velocity induced normal to airfoil, ms1
Cartesian coordinates, m, m, m
angle of attack, rad
effective angle of attack, rad
blade flapping angle, rad
incremental quantity
Wagner indicial response function
vortex filament strength (circulation), m2 s1
gust speed ratio, V /(V +Vg )
wavelength of the gust
nondimensional inflow, vi /R
flow sweep angle
blade pitch angle, rad
flow density, kg m3
dummy variable of integration
rotor solidity, Nb c/R
rotor rotational velocity, rad s1
Kussner function
blade azimuth, t, rad
vortex filament age, rad

Introduction
Renewable energy devices such as wind turbines are
playing a significantly increasing role in the generation of
electrical power, both in the United States and Europe. The
aerodynamics of a wind turbine, however, are extremely

1 OF 28
A MERICAN I NSTITUTE OF A ERONAUTICS AND A STRONAUTICS PAPER 2002-0037

4000

Mostly Periodic

Wind Inflow
speed

Yaw

Turbine torque output, Nm

Flowfield Structure

Mostly Aperiodic

Tower
Wind
Wake
shadow turbulence dynamics

Blade/
wake
interactions

Fig. 1 Summary of the various aerodynamic sources that


contribute to the airloads on a wind turbine.

complicated and in many ways parallel the problems found


with helicopter rotors. Such problems include the challenges in understanding and predicting the unsteady blade
airloads and rotor performance, as well as predicting the
dynamic stresses and aeroelastic response of the blades.
Wind turbines are also subjected to complicated environmental effects such as atmospheric turbulence, ground
boundary layer effects, directional and spatial variations in
wind shear, thermal stratifications, and the possible effects
of an upstream unsteady wake from a support structure
(tower shadow).
Figure 1 summarizes the various aerodynamic sources
that may affect the airloads on a wind turbine, which can be
decomposed into a variety of essentially periodic and aperiodic contributions. The net effect is that the wind turbine
operates in an adverse, unsteady aerodynamic environment
that is both hard to define using measurements and also to
predict using mathematical models. The overall difficulties in predicting the performance and structural loads have
led to higher capital investment and operating/maintenance
costs for wind turbines, making it difficult for wind energy
devices to compete with other forms of renewable and nonrenewable energy sources.1
Because the blade loads and performance of a wind
turbine are directly determined by unsteady aerodynamic
forces, a better understanding of the underlying fluid dynamics is essential if accurate modeling of the rotor aerodynamics and acceptable predictions of the turbine loads
and power generation are to be made. A better definition
of the airloads will also define the structural requirements
and will allow optimal strength, light-weight blades to be
designed. It is clear that better predictive tools are critical if more efficient and lower cost wind turbines are to be
designed in the future.
Recently, the National Renewable Energy Laboratory
(NREL) invited the international community to participate
in an in the blind prediction of the loads and performance
of a comprehensively instrumented wind turbine that was
tested under controlled conditions in the 80by120 foot
(24.4by36.6 meter) wind tunnel at NASA Ames.2 The
primary objective of those experiments was to create a
definitive set of airloads and performance measurements
over a wide range of operating conditions that was free of
the uncertainties caused by the various atmospheric effects
that are always found in field tests with turbines. These

Note: Results are shown for 8 representative


turbine analyses spanning the extremes
of the predictions

3500
3000

NREL Experiments

2500
2000
1500
1000
500
0
5

10

15
Wind speed, m/s

20

25

Fig. 2 Representative in the blind predictions of turbine


power output as a function of wind speed compared to experimental measurements.2

wind tunnel results provide the analyst with an opportunity


to better understand the physics of wind turbine aerodynamics, and gives a definitive data resource for validating predictive methods and perhaps resolving outstanding
modeling issues.
Results from the NREL blind comparisons were found
extremely mixed,3 with considerable deficiencies noted between the predictions for blade loads and power output
from the wind turbine even for the simplest unyawed, unstalled operating conditions see Fig. 2. The results for
power (torque) output ranged from a 60% underprediction
to more than a 150% overprediction. Even using similar
predictive methods with essentially the same medley of
sub-component models, there were significant differences
between the results. This suggests unresolved deficiencies
in the models, perhaps even at a first-order level. However,
it is clear that at least some part of the differences can be
attributed to inconsistencies in empirical input parameters,
such as assumed two-dimensional airfoil characteristics.4, 5
For operations in yawed flow and/or for higher wind
speed conditions where unsteady effects become important and the turbine begins to stall, the modeling of the
rotor aerodynamics becomes much more challenging and
the NREL blind comparisons suggested major deficiencies
in the models for these conditions. Unlike a helicopter rotor
where the onset of stall is a hard boundary severely limiting its performance, fixed pitch wind turbines may have
to operate continuously with considerable amounts of flow
separation and blade stall.6 Even for pitch controlled turbines, because of changing wind and flow directions, unsteady aerodynamics and stall effects can still be important
contributors to the blade airloads and wind turbine performance.
One unsteady, nonlinear aerodynamic problem of particular significance on wind turbines is dynamic stall. This
is a transient stall effect that can result in unsteady aerodynamic forces being produced that are considerably in
excess of what would be expected or predicted under steady
(static) conditions. Results from the NREL blind comparisons have shown that when the wind turbine was op-

2 OF 28
A MERICAN I NSTITUTE OF A ERONAUTICS AND A STRONAUTICS PAPER 2002-0037

erating under dynamic stall conditions, most models were


deficient in terms of predicting net power output, ranging
from a 50% underprediction to a 200% overprediction, and
were particularly deficient in terms of predicting both the
amplitude and phasing of the blades loads that are associated with the occurrence of dynamic stall. In many cases
dynamic stall onset was predicted by the models where
clearly the experimental results showed there was none, or
dynamic stall was not predicted at all whereas the experimental data suggested extensive regions of stall over the
rotor. The problem of dynamic stall, however, cannot be
understood or predicted in isolation, and it is necessary to
consider the coupled effects of the self-generated vortical
wake on the problem. The vortical wake behind the turbine produces a very nonuniform inflow velocity and angle
of attack distribution over the disk, and so proper treatment of the wake is part of the key to properly defining
the aerodynamic environment over the blades and defining
the conditions that produce the onset of stall.
The objective of this paper is to review some of the challenges involved in the better understanding and prediction
of the unsteady airloads on wind turbines. The focus of
the paper has an emphasis on modeling, with a reminder
that many of the basic tools needed to model the unsteady
airloads of rotors and wind turbines are already on hand.
However, there are also fundamental limits in the capabilities of existing models, and there may be misunderstandings or ambiguities in how these models should be applied.
A challenge for the analysts of the future is to better understand the physics of the aerodynamic problems found on
wind turbines, to develop more rigorous models with wider
ranges of application, and to better integrate and validate
these models with reference to good quality experimental
measurements.

Defining the Unsteady Operating


Environment
Wind turbines operate for most of their time in an unsteady flow environment.6, 7 The airloads on each blade
element vary in time because the turbine is usually yawed
with respect to the oncoming wind, and also because of
shear in the ambient wind, ambient turbulence, blade flapping and vibratory displacements, and other factors such as
tower shadow. While unsteady effects are important on
many different types of rotating machinery,8 the issues associated with wind turbines are particularly acute because
large amplitude flow perturbations and high effective reduced frequencies may be involved. For example, because
of the relatively low rotational velocity of wind turbines
(about one-second per rotor revolution) and low tip speeds,
changes in wind speed or atmospheric fluctuations can result in significant changes in angle of attack at the blade
elements. Also, when the turbine is yawed with respect to
the wind, there are large fluctuations in the relative velocity
at the leading-edge blade element compared to the velocity induced by blade rotation alone. These large amplitude
displacements in local angle of attack and onset velocity

contribute significantly to the unsteady flow environment


on the blades. Furthermore, large amplitude displacements
may often stretch the limits of the unsteady aerodynamic
modeling capabilities, which may be derived under the assumption of small perturbations.
While dynamic stall is certainly a significant unsteady
flow problem on many types of wind turbines, it is important to recognize that unsteady airloads will be produced
even in the absence of dynamic stall. It is well known, both
from experimental and theoretical perspectives, that significant unsteady effects can be produced on lifting surfaces
during unsteady motion but when operating with fully attached flow. These effects manifest as amplitude and phase
changes in the airloads compared to what would be obtained under quasi-steady conditions. Both circulatory and
noncirculatory contributions to the airloads are always involved. A prerequisite to predicting the onset of dynamic
stall, of course, is to adequately predict the unsteady airloads under attached flow conditions. Despite its apparent
simplicity compared to the stalled case, however, this is a
nontrivial problem for which the development of general
models that are valid for the rotor environment still continue to challenge the analyst.
A particularly interesting aspect of the wind turbine
aerodynamics problem is a quantification of the unsteadiness associated with the flow field and the turbine operating
state. The reduced frequency, k, can be viewed as measure
of the degree of unsteadiness of the flow, and can be written as k = b/V = c/2V , where is some characteristic
physical frequency of the flow, c is a characteristic length
such as blade chord, and V is an average flow velocity.
Because of the typically low rotational velocities of wind
turbines, this means that perturbations at the blade elements
may have high effective values of reduced frequency, and
especially so on moving inward along the blade from its
tip because V becomes smaller. While many of these flow
disturbances may be transient and aperiodic in nature, it
is convenient to characterize the significance of unsteady
problems, in general, in terms of an effective reduced frequency. For example, under yawed conditions the 1/rev
fluctuations in the component of flow velocity normal to
the blade can occur at reduced frequencies greater than 0.1,
and considerably higher inboard on the blade. Passage
of the blade through the tower shadow results in transient
changes in angle of attack that may involve effective values
of k that exceed 0.2.
Another and perhaps less obvious issue, is the treatment
of flow compressibility and the potential effect on the unsteady airloads. Generally, compressibility effects manifest
as additional amplitude and phasing effects compared to
what would be produced in a nominally incompressible
flow. Although the flow velocities are found to be relatively
low on wind turbines, especially compared to a helicopter
rotor, an incompressible flow justification requires that not
only must the local velocities and Mach numbers be low,
Only for k < O(0.01) can the flow be assumed steady or
quasi-steady.

3 OF 28
A MERICAN I NSTITUTE OF A ERONAUTICS AND A STRONAUTICS PAPER 2002-0037

but the frequency of the sources of unsteady effects must


be small compared to the sonic velocity, i.e., the product
c/a << 1, where a is the speed of sound and is a
characteristic frequency. This means that the characteristic
reduced frequency must also be small. The reduced frequency can be written as k = c/2Ma, so that Mk << 1 to
justify the assumption of incompressible flow. Therefore,
there may be situations found on wind turbines where the
treatment of compressibility is necessary if modeling efforts are to be successful. However, to include the effects
of compressibility, it is required to raise the complexity of
the models, usually with substantial increases in computational overhead.9

Modeling Strategies for Wind Turbine


Aerodynamics
The aerodynamic modeling of wind turbines has
spanned the gamut from parsimonious models based on
simple momentum and blade element/momentum (BEM)
theories, through engineering models based on blade element and inflow or vortex wake methods, to computational
fluid dynamics (CFD) methods that solve numerically the
Euler or Navier-Stokes equations. In this paper, the focus
will be on the use of engineering methods, which represent
the majority of the methods currently being used to design
wind turbines. While the other approaches using BEM, and
perhaps even CFD, can play a useful role in the design process, it is the engineering models that currently give the
highest predictive confidence levels for design purposes.
BEM theories generally give acceptable approximations
to the axisymmetric distribution of inflow and loads found
under conditions where the wind is normal to the plane of
the rotor (i.e., the turbine is unyawed with respect to the
oncoming wind), and there are no dynamic stall effects.3, 4
The idea of this essentially analytical approach is to solve
for the rotor inflow based on a combination of a momentum balance on successive annuli of the rotor disk and a
blade element representation of the sectional aerodynamics.9 Because the basis of the BEM method is strictly a
two-dimensional theory, three-dimensional effects such
as the physical roll-off in the lift as the blade tip is approached must be treated using corrections, such as with
the use of Prandtls tip loss function. This correction can
approximately account for a finite number of blades, and
also the effects of blade planform and twist distribution
through the effect on the inflow angle. Static airfoil characteristics, such as the nonlinear changes in the lift with
the onset of flow separation, can be incorporated into the
BEM theory using table look-up or other such strategy.
However, Tangler4 suggests that there may be considerable
uncertainty involved in the incorporation of nonlinear airfoil characteristics into BEM type models.
Despite the various assumptions and approximations
used with the BEM, validation studies have shown it can
The

early work with the technique as it applies to propellers


is reviewed by Glauert see Durand.10

Fig. 3 Photograph of the vortical wake behind a horizontal axis wind turbine rendered visible using smoke injection.
Courtesy of NREL.

give good preliminary predictions of turbine loads and performance,1 and considerable insight into basic design parameters affecting the turbine performance, including blade
geometric parameters such a planform and twist. Under
yawed conditions, however, the BEM is less able to model
the physics of the turbine aerodynamics because of the
strong non-axisymmetric flow and three-dimensional nature of the airloads produced over the rotor disk. A major source of this three-dimensionality is from the vortical
wake system produced behind the rotor, an example of
which is shown in Fig. 3. Like a helicopter, the rotor wake
is comprised of vortices that trail from each of the blade
tips, and these are rendered visible in the photograph by
the use of smoke generators. Various modifications to the
BEM approaches have been used to represent these threedimensional wake effects, such as by using various types
of inflow models.1 While improvements in predictions
of net power are usually obtained using these methods,
they have not found great satisfaction in the process of predicting blade loads.4 Therefore, for most non-axisymmetric
operating conditions it is necessary to treat the problem of
the turbine wake and the blade aerodynamics more rigorously.
At the other end of the modeling scale, are the CFD
methods. This class of methods have the potential to provide a consistent and physically realistic simulation of the
turbine flow field. The huge computational costs, large
memory requirements, and numerous numerical issues associated with CFD methods have, at least so-far, also delayed their practical use for many rotating-wing problems,
including helicopter and wind turbine applications. The
field of CFD applied to rotating-wing problems is reviewed
by McCroskey11 and Landgrebe,12 which contain many
relevant references. In particular, problems involving flow
Because

a wind turbine extracts energy from the wind, the


wake expands downstream of the turbine. It may also distort as a
result of the wind gradient and turbine yaw angle.

4 OF 28
A MERICAN I NSTITUTE OF A ERONAUTICS AND A STRONAUTICS PAPER 2002-0037

separation, such as dynamic stall, have proved extremely


challenging for Navier-Stokes based methods, in part, because of a need to develop better turbulence models. Also,
the prediction of the three-dimensional vortical wake behind a rotor has proved just as daunting for CFD methods.
This is because vortical wake formation is a result of complex three-dimensional separated flow, and also because
the numerical methods have difficulty in preserving concentrated vorticity as it is convected downstream. While
it is clear that many have begun to take up the challenge
of modeling wind turbine problems using CFD,13, 14 the
capabilities of these methods have not yet been validated
sufficiently to be assigned the confidence levels that are
necessary for design purposes. With faster computers and
with improved numerical methods, CFD methods will ultimately prevail, and will become increasingly integrated
into the design of wind turbines and helicopter rotors.
Bridging the BEM theories and CFD approaches are a
class of engineering methods based on blade element theory combined with either inflow models or vortex models
that represent the nonuniform induced effects associated
with the vortical wake trailed from the turbine. One advantage of this class of predictive methods is the flexibility
to include a wide range of validated sub-component models representing various physical effects that are difficult to
model from first principles. For example, faithful predictions of the nonlinear unsteady aerodynamics and stall of
the blade sections must be modeled semi-empirically, such
as with simplified equations or table look-up. The modularity of the sub-component approach has many advantages,
including the ability to validate the sub-component models
with idealized laboratory experiments, and also the flexibility to progressively upgrade the models as a deeper understanding of the physics are obtained. However, there are issues in the proper coupling together of the sub-component
models, and sometimes different numerical coupling strategies may give different net results. Also, these methods
are by no means computationally inexpensive compared to
BEM models, but in principle should give better predictive success in predicting the details of the blade airloads.
This potentially extends theor use beyond preliminary design purposes. In particular, they can be combined with
structural dynamic models of the blades and rotor to produce powerful aeroelastic tools for the detailed structural
design of wind turbines.1 As in the case of all predictive
models, however, sustained validation of the approach with
experimental measurements is essential, despite the inherent difficulties and lengthy nature of the process.

Unsteady Wake Modeling


Unsteady aerodynamics effects are, in part, a consequence of the time-history of the induced velocity from the
vorticity contained in the shed wake, coupled with the induced velocity contributed by the circulation contained in
the trailed wake. In blade element models there are two
aspects of the problem of defining and modeling the aerodynamic environment. The first is an outer problem, which

is to model the effects of the induced velocity field produced by the vortical wake trailed from behind each blade.
The second, the inner problem, is to model the resulting
local unsteady aerodynamic response at each of the blade
elements. Physically, of course, these two parts are intrinsically connected together, but for both the understanding
of the problem and also for modeling purposes, it is convenient to treat them separately.
A photograph of the wake visualized downstream of a
one-bladed HAWT has been shown previously in Fig. 3.
The induced effects produced by this cycloidal wake have
been modeled using two general approaches: dynamic
inflow methods and vortex wake methods. Both approaches have advantages and disadvantages. Because of
the better flexibility to handle a broad range of operating
conditions, it is in the area of vortex wake modeling that
most future challenges lie for the wind turbine analyst. Yet,
dynamic inflow models have attractive mathematical forms
and low computational overheads that will always be appealing for certain types of rotor analyses.
Dynamic Inflow Modeling

The principles of the dynamic inflow approach can be attributed to Carpenter & Freidovitch,15 where the idea is to
consider the unsteady aerodynamic lag of the inflow development over the rotor disk in response to changes in
blade pitch inputs or changes in rotor thrust. The mathematics of the approach formally embody the concepts of
the BEM theory. The equations describing the distribution
of inflow are written in the form of ordinary differential
equations, with a time constant (or constants) representing
the dynamic lag in the build-up of the inflow.
One attraction of dynamic inflow models for use in wind
turbine work is their mathematical form and relative numerical efficiency, and also the prior predictive success in
using these methods for various applications in the helicopter field. The main advantage of representing the aerodynamic model as ordinary differential equations is that
it is appealing for many forms of structural dynamic and
aeroelastic analysis of the rotor, so that the entire coupled
problem can be solved simultaneously using the same numerical methods. Crews et al.,16 Ormiston,17 Peters18 and
Curtiss19 have shown that using dynamic inflow models
can have a significant influence on predictions made in helicopter flight dynamics and rotor aeroelasticity.
The most popular model of dynamic inflow for both helicopter and wind turbine work is that of Pitt & Peters,20
which has seen several further developments for helicopter
applications see Gaonkar & Peters21 and Peters and colleagues.22, 23 For a discussion and background of dynamic
inflow models in regard to their possible application to
wind turbine problems, see and Hansen & Butterfield,1
Bierbooms,24 Snel & Schepers.25 Other variations of the
dynamic inflow theory and applications thereof have appeared in the literature see Johnson.26
The principle of the dynamic inflow approach can be illustrated as follows. Using the blade element momentum

5 OF 28
A MERICAN I NSTITUTE OF A ERONAUTICS AND A STRONAUTICS PAPER 2002-0037

theory, the rotor thrust can be related to the inflow using




2
T = ma vi + Avi vi + R
(1)
3


1
1 vi
1
=
Nb cCl 2 R3

2
3 2 R 3
or in nondimensional form the former equation becomes


ma 
2 
CT =
+

(2)
+2

i
i
R3
3


1

i
=
Cl
2
3 2 3
where ma was associated by Carpenter & Friedovich with
the apparent mass of an impermeable (solid) circular disk
accelerating in a stagnant fluid. They suggest this to be
63.7% the mass of a sphere of fluid with
equal

a radius
to the rotor radius,27 i.e., ma = 0.637 43 R3 . The ma vi
term, therefore, represents the additional force on the rotor
disk because of the accelerating inflow. The dynamic inflow equation can be written nondimensionally as




ma 

i
=
i
Cl
R3
2
3 2 3


2 
2i i +
(3)
3
To find the inflow, i , this differential equation must be
solved together with the blade flapping equations of motion.
The left-hand side of Eq. 3 contains the nondimensional
apparent mass component, which is found to be 0.85. An
approximate time constant of the dynamic inflow equation
can be estimated using only the linear portion of Eq. 3, i.e.,


0.85 i
i.e.,

0.85 4 
i
Cl

1
= Cl i
4
= i

(4)

Assuming Cl = 2/rad, the inflow time constant can be


identified as

0.54
radians

0.086
rotor revolutions

(5)

Therefore, the time constant for the unsteady inflow development through a typical wind turbine is relatively long at
approximately one to 1.5 rotor revolutions. This suggests
that transients introduced into the aerodynamics can take at
least 5 to 8 rotor revolutions to begin to settle out and have
the flow come back to nominally steady-state conditions.
One of the less satisfying aspects of the simplest forms of
the dynamic inflow theory, is that the time constants of the
dynamic inflow theory have been developed using the concept of the apparent mass or inertia of the flow surrounding

the rotor (i.e., a noncirculatory effect) as opposed the lag in


the dynamic evolution of the vortical wake (i.e., a circulatory effect). The apparent mass concept applied to the rotor
also assumes an equivalence between the apparent force on
a solid disk accelerating in a stagnant fluid and the force
on a fluid accelerating through a permeable actuator disk,
which is certainly not a rigorous analogy.
Bierbooms,24 Snel & Schepers,25 suggest alternative values of the wake time constants for wind turbine work, based
on empirical evidence of the induction lag. The complexity of the physics of the unsteady wake problem, however,
cannot be underestimated. Yet, if the time constant(s) of
the model can be obtained, say using time-accurate vortex
wake calculations, and suitably validated with experiments,
the dynamic inflow method provides one form of parsimonious model that relates changes in rotor loads to the lag in
the inflow development.
Vortex Wake Models

A more explicit treatment of the turbine wake requires


a method that can represent the strengths (circulation) and
spatial locations of the vortical elements that are trailed by
each blade and convected into the downstream wake. This
can be done using vortex methods, and over the last 20
years these types of methods have become popular tools for
the analysis of a wide range of aerodynamic problems.28, 29
The development of a vortex wake model is usually based
on the assumption of an incompressible potential flow, with
all vorticity being assumed concentrated within vortex filaments. From the strengths of the vortex filaments (which
in the case of a rotor requires a coupling to the blade lift
distribution), the induced velocity field can be determined
through the application of the Biot-Savart law.
Vortex methods applied to rotor wake problems encompass a variety of different approaches, ranging from prescribed to free vortex techniques. A prescribed vortex
technique is where a discrete representation is made of the
vorticity field, but the positions of the vortical elements are
specified a priori based on semi-empirical rules. Prescribed
vortex methods can be viewed more as postdictive methods
because they use experimental results for formulation purposes and so are strictly limited in scope to the conditions
that encompass the range of measured operating conditions
for which they were originally formulated. A free vortex
method is a predictive method because the elements are
allowed to convect and deform freely under the action of
the local velocity field to force-free locations. While a
disadvantage of all types of vortex methods for modeling
the wake is the relatively higher computational overheads
(mainly because of the need to evaluate the Biot-Savart
law a very large number of times), they have an inherently
more satisfying appealing physical approach for modeling
the wake than the so-called inflow models.
Prescribed vortex wake models have been developed
for helicopter applications30 but have seen some use for
wind turbines.31, 32 Development of more generalized
prescribed wake models for wind turbine applications is a

6 OF 28
A MERICAN I NSTITUTE OF A ERONAUTICS AND A STRONAUTICS PAPER 2002-0037

significant undertaking because not only do the positions


of the wake vortices need to be documented, but the experimental conditions must encompass a wide range of rotor
geometric (e.g., blade shape and twist) and operating states
(wind speed, yaw angle, etc.). Therefore, a trade-off must
be drawn between the costs of doing more experiments and
collecting more data to build the model, and the costs of
developing a better and more flexible model that can be
validated against existing data. Prescribed wake models are
also only strictly applicable when the operating conditions
are nominally steady-state, i.e., in a steady wind.
Free vortex methods have fewer potential limitations
when applied new turbine designs and/or more general operating states. Such methods have been widely developed
for use in helicopter rotor analyses, dating from the work
of Clark & Leiper33 in the late 1960s, but are still yet
to see significant use for wind turbine applications. Free
vortex methods are based on discretized, finite-difference
representation of the governing equations for the wake, and
when solved, track the evolution of discrete vortex elements through the flow. The number of discrete elements
per vortex filament can be very large, making the tracking
process memory intensive and computationally demanding, although considerably less demanding than when using
CFD methods.
Experience with helicopter rotors suggest that free vortex methods generally give better predictions of blade loads
than is possible with prescribed or rigid wake models.
They can also, in principle, deal with transient problems
such as when the turbine yaws into and out of the wind
all the wake dynamics of can be accounted for in a
time-accurate manner without having to make assumptions
about the effects on the wake geometry or the time constants of the flow. Under transient conditions, the rotor
may yaw momentarily into its own wake, resulting in powerful blade/wake interactions. The flexibility of free vortex
methods, coupled with the computing power available today, makes the further development of such methods very
attractive to the analyst.
In a typical (discretized) free vortex wake scheme, Lagrangian markers are placed on the vortex filaments trailing
from the blades see Fig. 4. Any two successive markers
on the vortex filaments are linked together, usually with
a piecewise linear reconstruction. The convection of any
Lagrangian fluid marker on a trailed vortex filament is described by
dr 
(6)
= V (r,t), r(0) =r0
dt
where r0 is the initial position vector of the marker. The
relationship in Eq. 6 can be formalized as
dr(, ) 
= V (r(, ))
dt

(7)

wherer(, ) defines the position vector of a marker lying


on a vortex filament that is trailed from the blade when it
was located at an azimuth at time t. Let t0 denote the time
when the element was first formed and when the blade was

Curved vortex
filament

l +1

l +2

Lagrangian
markers

r
Blade, N

Straight line segment


approximation

Blade, N-1

Induced velocity from


element of vortex trailed
by blade N-1

x
h

Wind velocity

Fig. 4 Schematic showing a Lagrangian discretization of vortex filaments trailing from the blades into the downstream
wake.

located at (). Because = t and () = t0 , then


= (t t0 ). The left-hand side of Eq. 7 can be rewritten
so that Eq. 7 reduces to a partial differential equation of the
form
r(, ) r(, )
1
+
= V (r(, ))
(8)

Equation 8 is the governing equation for the free vortex


wake problem.34
A numerical solution to the free vortex problem dictates a discretization of both the left and right-hand sides
of Eq. 8. This discretization results in a set of finite difference equations, which can be solved using various types
of numerical integration techniques.35, 36 One challenge in
in this process arises from the numerical evaluation of the
velocity term on the right-hand side of Eq. 8, which is a
highly nonlinear function of the wake geometry. The contributing elements can be written as
V = V + Vind + Vex

(9)

so that the total induced velocity field at each Lagrangian


marker is determined by the aggregate of the ambient free
stream velocity, V , the contributions induced by all of the
other vortex filaments in the rotor wake, Vind , as well as the
induced velocities produced by the blades and any other
sources of velocities such as the turbulence in the ambient
wind via the term Vex .
The Vind term is the most complicated and numerically
expensive contribution to V because the induced velocity
at any point in the flow must be determined by integrat-

7 OF 28
A MERICAN I NSTITUTE OF A ERONAUTICS AND A STRONAUTICS PAPER 2002-0037

ing numerically along the length of each curvilinear filament. To do this, segmentation of the vortex filaments into
small straight elements is normally used because the induced contribution of each element can then be evaluated
exactly from the Biot-Savart law. While this discretization
is formally second-order accurate,37 there is a relatively
high cost incurred by the repetitive numerical evaluations
of the Biot-Savart law. This cost, however, can be offset to
some extent by acceleration schemes such as velocity field
interpolation,38 which can result in an order of magnitude
reduction in computational effort with little loss of net accuracy.
The formulation of an integration methodology for the
free vortex wake equations that is numerically accurate,
stable and versatile as well as computationally efficient has
proven difficult. As a result, the numerical methods used
for free vortex wake models are still the subject of considerable research.35, 36 Two main classes of solution methodologies have evolved, namely: iterative or relaxation methods and time-marching methods. Time-marching free vortex methods potentially offer the best level of approximation to the problem, and with the fewest restrictions in
application. However, these methods has proven susceptible to types of instabilities resulting from the initiation
of numerical microstructures,39 thereby reducing the confidence that a physically realistic solution has been obtained.
For example, discretized evaluation of the induced velocity field through the Vind term can result in the inclusion of
higher-order anti-dissipative terms into the solution. This
can result in unbounded, non-physical growth of numerical
disturbances. Properly distinguishing between the known
physical instabilities of rotor wakes40 and those that are numerical in origin have proven to be a major hindrance in the
development of reliable and robust time-accurate free vortex wake models.36 Relaxation or iterative methods,35, 41
which explicitly enforce periodicity of the evolving wake,
are applicable only to steady, non-transient problems, but
are generally found to be free of the numerical issues that
can compromise time-marching solutions.
While the convection of vortex elements is the primary
task handled by vortex methods, the methodology may
also include some representation of vorticity intensification
through the stretching of the vortex filaments and also viscous diffusion. The representation of viscous effects in free
vortex wake methods are usually based on the use of semiempirical rules, such as using developments of Lamb-like
viscous core growth models.42, 43 Furthermore, although
not explicitly defined in the governing equations for either the prescribed or free vortex wake problem, it will be
appreciated that the wake solution must be coupled with
the aerodynamic loading on the blades, as well as accounting for rigid and elastic blade motions (blade dynamics)
such as flapping and bending. This means that boundary
conditions must be specified to ensure that the physics of
the problem are correctly modeled; specifically all of the
vortex filaments convected into the rotor wake must originate at the locations of the blade(s) with appropriate initial

(a) Side view

Wind speed

(b) Top view

Wind speed

Fig. 5 Free vortex wake calculations of a three-bladed wind


turbine operating at a steady yaw angle of 0 . (a) Side view of
wake, (b) Top view of wake.

strengths.44 In most cases, the tip vortex roll-up and the


velocity field close to the vortex core will be modeled semiempirically.
Two examples of time-accurate free vortex wake calculations for a HAWT are shown in Figs. 5 and 6. Results
for the wake geometry are shown for a 3-bladed Grumman
Wind Stream 33 downwind HAWT operating at 72 rpm in
a steady wind speed. A small surface wind gradient was
imposed. Figure 5 shows the side and top views of the
wake geometry for zero yaw angle. Note that the wake
expands downstream of the disk because the rotor extracts
energy from the wind. The wake structure is also relatively
axisymmetric, apart from a mild axial distortion resulting
from the wind gradient.
Figure 6 shows a more interesting case where the turbine
suddenly yaws 30 out of wind. While this is a somewhat
artificial case in practice, it serves to illustrate the dynamics of the wake adjustments that take place when the turbine
yaws into and out of the wind. Initially, just after the yaw
starts, the turbine moves into its own wake. This is a situation where there can be relatively powerful interactions
between the blades and discrete vorticity in the wake, leading to locally high unsteady blade loads. Note that for
increasing time, the adjustments to the developing wake
take place relatively slowly, and only after about 9 rotor
revolutions does the solution become essentially periodic,

8 OF 28
A MERICAN I NSTITUTE OF A ERONAUTICS AND A STRONAUTICS PAPER 2002-0037

(a) Top view, time = 0 + revs.

(b) Top view, time = 3 revs.

Angle of attack & pitch


rate (blade pitch &
yaw response)

Plunge (blade flapping


& bending)

z
Wind speed

z
Wind speed

Time-varying indicident
velocity or in-plane gusts

(c) Top view, time = 6 revs.

(d) Top view, time = 9 revs.

V(t)

Normal velocity perturbations


(from rotor wake etc.)

V
z
Wind speed

z
Wind speed

Fig. 7 Decomposition of flow field into constituent elements


of net aerodynamic forcing.
Attached Flow Conditions

(e) Top view, time = 12 revs.

(f) Top view, time = 15 revs.

z
Wind speed

z
Wind speed

Fig. 6 Time-accurate free vortex wake calculations of a threebladed wind turbine suddenly yawing 30 out of wind. Top
views of wake: (a) Time=0+ , (b) Time = 3 revs., (c) Time = 6
revs., (d) Time = 9 revs., (e) Time = 12 revs., (f) Time = 15 revs.

a result consistent with the dynamic inflow model. In this


condition, the component of wind velocity parallel to the
plane of the disk causes the wake to roll-up along its top
and bottom edges, essentially forming two dominant vortex bundles.

Blade Section Unsteady Aerodynamics


The inner loop of the blade element approach usually
uses a two-dimensional model for the unsteady aerodynamic behavior of the blade sections. At the blade element
level, the term unsteady aerodynamics is often considered by some as being synonymous with dynamic stall.
However, it is important to appreciate that significant unsteady effects may be produced on lifting surfaces even in
the absence of dynamic stall. The essential physics of nonsteady airfoil problems can be observed from experiments,
and interpretations of the behavior supported by relatively
parsimonious theoretical or numerical models.

It is important for the analyst to recognize that many of


the tools to model unsteady aerodynamic effects on airfoils
have already been laid down. Two-dimensional, unsteady
aerodynamic theories describing unsteady airfoil behavior
in fully attached flow have formed the basis for many types
of rotor analyses. The tools for the analysis of incompressible, unsteady problems were developed by 1940,4547 with
the extension to compressible flows complete by 1950.48, 49
At the blade element level, the various effects described
in Fig. 1 can be decomposed into perturbations to the local
angle of attack and velocity field, as shown in Fig. 7. At
low angles of attack with fully attached flow, the various
sources of unsteady effects manifest primarily as moderate
amplitude and phase variations relative to the quasi-steady
airloads. If, however, the effective reduced frequencies become high, much larger effects on both the amplitude, and
particularly, the phase may be produced. Also, one must be
cautious to appreciate that the unsteady effects associated
with various types of forcing shown Fig. 7 cannot be treated
in the same way because they cannot all be lumped into
a response from a net effective change in angle of attack;
the aerodynamic response to each component of forcing
must be considered separately and combined through the
superposition. This is a fundamental principle that can be
overlooked in the application of these types of unsteady
aerodynamic models to complex flow environments, such
as those encountered on rotors.
Results for incompressible, unsteady airfoil problems
have been formulated in both the frequency domain and
the time-domain, primarily by Wagner,50 Theodorsen,45
Kussner, and von Karman & Sears.46 An authoritative
source documenting these classical theories is Bisplinghoff
et al.,48 with further details of their application given by
Leishman.9 These solutions have the same root in unsteady
thin-airfoil theory, and give exact analytic solutions for
airloads for different forcing conditions, i.e., for airfoil
motions or for externally imposed velocity fields. While
these methods are valid for two-dimensional, incompressible flows, and were primarily intended for fixed-wing ap-

9 OF 28
A MERICAN I NSTITUTE OF A ERONAUTICS AND A STRONAUTICS PAPER 2002-0037

plications, they have also formed the foundation for several extensions to subsonic compressible flow9, 49, 51 and,
with certain other assumptions, to various types of threedimensional rotating-wing problems. It is up to the analyst
to explore these tools, to examine their capabilities, and to
determine their limitations. Only when this is done, will
the analyst be in a position to decide what aspects of the
modeling require further development.

Cl

it +
= 2 (F[1 + ik] + G[i k]) e


k
it
k i
e
2

(11)

Notice that for a plunging motion the circulatory part of the


lift response leads the forcing displacement h by a phase angle of
/2. Also, the apparent mass force leads the circulatory part of
the response by a phase angle of /2, or the forcing by an angle
of .

Circulatory part
Noncirculatory part
Total
Experimental data

Lift amplitude

2
1.5
1
0.5
0
0

0.1

0.1

0.2

0.3

0.4

0.5

0.2
0.3
0.4
Reduced frequency, k

0.5

180
160
Phase of lift - deg.

Theodorsens Theory
The capabilities of the unsteady thin-airfoil theory as
a tool for modeling the sectional aerodynamics can be
demonstrated using Theodorsens solution45 for harmonic
variations in forcing. This simulates the quasi-periodic first
harmonic variations in angle of attack, which are representative of the conditions found at a blade element on a wind
turbine.
Comparisons between Theodorsens incompressible unsteady theory and measurements of the unsteady lift produced on airfoils oscillating in plunge (vertical translation
or heaving motion) and pitch oscillations (in angle of attack) are shown in Figs. 8 and 9, respectively. The results
are plotted as the first harmonic normalized amplitude of
the lift and the corresponding phase angles as functions of
k.
it the lift coefficient is
For a plunge forcing, h = he
 h

Cl = 2k(iF G) k2 eit
(10)
b
where complex valued Theodorsens function is C(k) =
F(k) + iG(k). The first term inside the brackets in Eq. 10
is the circulatory term, and the second term is the noncirculatory or apparent mass contribution. In the case of
the plunge oscillations, the lift is zero when the airfoil is
stationary, but increases rapidly as the reduced frequency
increases. This is an example where the addition of the
apparent mass forces actually decreases the lift amplitude;
this is because the circulatory and noncirculatory components are combined by vector addition. The results in Fig. 8
show good agreement between the linear theory and the
measurements. The sign of the phase angle changes from
a lag (less than 90 ) to a lead (greater than 90 ) at higher
reduced frequencies because the noncirculatory effects become more dominant.
For harmonic pitch oscillations, terms involving appear in the equations for the aerodynamic response. The
it , and the pitch rate by
forcing is now given by = e
it

= ie
. Theorodorsens approach gives the lift coefficient for pitching about the 1/4chord as

2.5

140
120
100
80
60
40

Fig. 8 Unsteady lift response to harmonic variations in pure


angle of attack (plunge forcing).

As shown in Fig. 9, the lift amplitude initially decreases


with increasing k because of the effects of the shed wake,
which is a circulatory effect. The amplitude begins to increase again for k > 0.5 as the noncirculatory forces begin
to dominate the airloads. Again, this is particularly shown
by the phase angle, which exhibits an increasing lead for
k > 0.2. It is significant that a reduced frequency of up to
0.6 was obtained in these experiments, which gives a good
opportunity to examine the applicability of the linearized
theory at higher reduced frequencies.
Overall, the correlation obtained between the linearized,
incompressible unsteady theory and experimental measurements for airfoils oscillating in pitch and plunge is relatively good. Note that both the circulatory and noncirculatory effects must be included to ensure that both the
amplitude and the phase is accurately predicted. These results give considerable support to the validity of unsteady
thin-airfoil theory, at least for low free-stream Mach numbers and up to moderate values of reduced frequency. Extensions of these linearized unsteady aerodynamic models
to subsonic compressible flows is reviewed in Ref. 9. In
most cases, the mathematical elegance and computational
simplicity of the linearized models will be attractive to
the wind turbine analyst. However, in some cases the assumption of linearity may be difficult to justify. This is
where CFD methods can help the analyst define the limits
of linearity, and may also give new guidance in developing

10 OF 28
A MERICAN I NSTITUTE OF A ERONAUTICS AND A STRONAUTICS PAPER 2002-0037

i 2
wg = wo e g (Vt - x)

0.3

Circulatory part
Noncirculatory part
Total
Experimental data

0
0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

k =10.0

kg=0.0

k =4.0

k =0.5

-0.1

k =10.0

-0.2

k =0.02
g

k =5.0

k =0.1

k =2.0
g

C (k)

k =0.2
g

k g=1.0

-0.3

Phase of lift - deg.

k =1.0

k =8.0

100
80

S (k)

0.1

k =2.0

0.2
Imaginary part

Lift amplitude

S' (k)

k =0.5
g

k =0.2
g

-0.4
-0.2

60

0.2

0.4
Real part

0.6

0.8

40

Fig. 10 Sears function referenced to airfoil leading-edge and


mid-chord plotted as real and imaginary parts.

20
0

Sears. In this case, the result for the lift can be written as
w 
0
Cl = 2
(13)
S(kg )
V

-20
0

0.1

0.2 0.3 0.4 0.5 0.6


Reduced frequency, k

0.7

0.8

Fig. 9 Unsteady lift response to oscillatory pitch forcing.

improved unsteady aerodynamic models for future use in


blade airloads prediction and wind turbine design.
Sears Problem
In the rotor environment, one of the largest sources of
changes in angle of attack come from the nonuniform induced velocity over the rotor disk. Von Karman & Sears46
analyzed the problem of a thin-airfoil moving through a
sinusoidal vertical velocity (gust) field. Like Theodorsens
solution, this is also a frequency domain solution and it
is useful because it shows the fundamental differences between the results obtained on an oscillating airfoil in angle
of attack and an airfoil moving through an oscillatory vertical velocity field; the latter results in a nonuniform angle
of attack over the airfoil chord.
The gust field can be considered as an upwash velocity
that is uniformly convected by the free-stream, as shown in
Fig. 10. The perturbation velocity normal to the chord is

g x 
wg (x,t) = sin gt
V

(12)

where g is the gust frequency. There are two cases of


interest. First, if the gust is referenced to the airfoil leadingedge then x = 0 and Eq. 12 becomes wg (t) = sin gt. Second, if the gust is referenced to the mid-chord, then x = c/2,
which is equivalent to a phase shift. The mid-chord was the
reference point used in the original work of von Karman &

where S(kg ) is known as Sears function. The gust encounter frequency is kg = 2V /g , where g is the wavelength of the gust see Fig. 10.
Notice that the spiral shape of the Sears function arises
only when the gust front is referenced to the mid-chord. If
the gust response is computed relative to the leading-edge,
then a different function is obtained. The gust front reference point is frequently confused, with clear consequences.
While the differences are small at low reduced frequencies,
the errors become significant for kg > 0.2. At low reduced
frequencies the Sears and Theodorsen functions converge,
but for reduced frequencies greater than 0.1 the differences
become increasingly large so it is important to be sure that
the correct function is applied for the type of problem being
considered.
Indicial Response Theory

Exact frequency domain theories, such as those of


Theodorsen and Sears, have found use in many problems
in both fixed-wing and rotating-wing aerodynamics. However, they are less useful in a rotor environment because
the velocity at a blade element is not usually constant so
the reduced frequency argument, k is an ambiguous parameter. Therefore, a theory formulated in the time-domain is
more general, and is usually more powerful, despite the fact
that the problem must be solved numerically in most cases.
For rotor analyses, the indicial lift response is a useful
starting point in the development of a general time-domain
By

definition, an indicial function is the response to a disturbance that is applied instantaneously at time zero and held
constant thereafter; that is a disturbance given by a step function.

11 OF 28
A MERICAN I NSTITUTE OF A ERONAUTICS AND A STRONAUTICS PAPER 2002-0037

Wagner function,

see Bisplinghoff et al.,48 or Leishman.9 When this is


done, a whole series of practical numerical tools for computing the unsteady aerodynamics can be unleashed. One
exponential approximation to the Wagner function is attributed to R. T. Jones52 is written as

1 for s=

0.8
0.6

(s) 1.0 0.165e0.0455s 0.335e0.3s for s > 0 (17)

0.4
0.2

This approximation is found to agree with the exact solution to an accuracy of within 1%, which is sufficient
accuracy for most practical purposes.

Wagner function (exact)


R. T. Jones' approximation
Algebraic approximation

0
0

5
10
15
Distance traveled in semi-chords, s

20

Fig. 11 Wagners unsteady lift function compared to exponential approximation.

unsteady aerodynamic theory. If the indicial response is


known, then the unsteady loads to arbitrary changes in angle of attack can be obtained through the superposition of
indicial aerodynamic responses using Duhamels integral.
In fact, this approach is used as the root of several of the
semi-empirical dynamic stall models used in helicopter and
wind turbine work (discussed later).
Wagner Function
Wagner50 has obtained a solution for the indicial lift on
a thin-airfoil undergoing a step change in angle of attack in
incompressible flow. The circulatory part of the lift coefficient in response to an arbitrary variation in angle of attack
can be written as


 s
d
Clc (t) = 2 (0)(s) +
(14)
W (s )d
0 dt
= 2e (t)
where W is Wagners function, which is shown in Fig. 11,
and e simply represents an effective angle of attack and
contains within it all of the time-history effects on the lift.
The reduced time, s, is given by
s=

1
b

 t

V dt =
0

2
c

 t

V dt

(15)

which represents the relative distance traveled by the airfoil in terms of semi-chords. Notice that if V = constant,
then s = 2V t/c. In addition, the appropriate noncirculatory
terms must be added to Eq. 14 to get the total lift9 where
Clnc (t) =


b
h ab

2
V

(16)

For incompressible flow the apparent mass terms are proportional to the instantaneous motion, so they all appear
outside the Duhamel integral.
Although the Wagner function is known exactly, it is not
in a convenient analytic form. Therefore, it is usually replaced by a simple exponential or algebraic approximation
 This is implied by the infinite value of the Wagner function
at s = 0.

Kussner Function
As already mentioned, the rotor wake produces a highly
nonuniform induced velocity across the plane of the disk.
Therefore, a typical blade element encounters a nonuniform vertical upwash/downwash field as it rotates about
the shaft. It is important to distinguish properly the effects on the airloads arising from angle of attack changes
from blade motion (in-effect, a plunging and pitching motion at the blade element) from the effects resulting from
the rotor wake induced velocity field (in essence, a nonuniform vertical gust velocity normal to the blade element).
This distinction is important, and should not be overlooked
in the mathematical modeling of wind turbine problems.
For example, the tower shadow problem (see later) is correctly viewed as vertical gust problem with a spatial and
temporal nonuniform velocity relative to the chord dimensions of the blade.
The problem of finding the transient lift response on a
thin-airfoil entering a sharp-edged vertical gust (that is, a
sharp change in vertical upwash velocity) was tackled by
Kussner and properly solved by von Karman and Sears.46
In this problem, the upwash velocity, wg , is defined as
wg = 0 for s < 0 and wg = w0 for s 0, as shown in
Fig. 12. Recall that in Wagners problem, the angle of attack changes instantaneously over the whole chord at s = 0.
In Kussners problem, however, the quasi-steady angle of
attack changes progressively as the airfoil penetrates into
the front of the change in vertical velocity; only at s = 2
is the airfoil fully immersed in the gust. Compared to the
Wagner function, which has a value of 1/2 for s = 0+ , it
will be seen that the Kussner function builds from an initial
value of zero.
The Kussner function can also be used with Duhamel
superposition to find the lift response to an arbitrary vertical upwash velocity field. The total lift coefficient can be
obtained using


 s
dwg
2
Cl (t) =
wg (0)(s) +
(s )d
(18)
V
0 dt
To enable practical calculations using Duhamel superposition the Kussner function, like the Wagner function, is
usually replaced by an exponential or algebraic approximation. One exponential approximation is given by Sears
& Sparks53 as
(s) 1 0.5e0.13s 0.5e1.0s

12 OF 28
A MERICAN I NSTITUTE OF A ERONAUTICS AND A STRONAUTICS PAPER 2002-0037

(19)

x0 =

s
2| |

0.6

0.4

Vt

Vg = V
||

w0

1.5

x, x

w0

Normalized lift, Cl / 2

Kussner function,

1 for s=

0.8

Kussner function (exact)


Exponential approximation
Algebraic approximation

0.2

0
0

5
10
15
Distance traveled in semi-chords, s

20

Convecting Gusts: Miles Problem


Results for the lift on a thin, two-dimensional airfoil encountering a traveling vertical gust have been obtained by
Miles54 in terms of the gust speed ratio, , i.e.,
=

V
(V +Vg )

(20)

where V is the velocity of the airfoil and Vg is the gust


convection velocity relative to the airfoil. The assumption
made in most rotor aerodynamic analyses is that the wake is
stationary with respect to the rotor blades (i.e., rigid, nonconvecting wake assumption), so that = 1 at all blade
elements over the rotor disk. However, the self-induced
velocities generated by the vortex wake system results in
a continuously changing and nonuniform convection of the
induced velocity field with respect to the rotor, and this may
produce values of less than or greater than unity.
Miles54 showed that as the propagation speed of the traveling gust front increased from zero to ( decreases from
1 to 0), the solution for the unsteady lift changes from
the Kussner result to the Wagner result, with a variety of
intermediate transitional results being obtained. Miles results were later generalized by Drischler & Diederich55 and
by Leishman56 using the reverse flow theorems. Both approaches make use of approximations to the Wagner function to facilitate numerical solutions.
Consider an airfoil traveling with velocity V and subject to a vertical sharp-edged gust field of magnitude w0
convecting downstream with velocity Vg = (1 1)V , as
shown in Fig. 13. Note that when the gust field is stationary = 1, and when traveling toward the airfoil at infinite
speed = 0. For the sharp-edged gust, the primary boundary condition is that the downwash, w, is zero on the part of
the airfoil that has not reached the gust front. In either case,
it will be seen that, like the Kussner problem, the angle of
attack on the airfoil changes progressively as a function of
time as the airfoil penetrates into the gust front.
Results for the unsteady lift for downstream traveling
sharp-edged gusts are shown in Fig. 13. For = 0 (Vg =
), the results lead to the Wagner function. For = 1

=1
=1.5
=2

0.5

Fig. 12 Kussners
function compared to exponential approximation.

=0
=0.25
=0.5
=0.75

2
3
4
5
Distance traveled, s (semi-chords)

Fig. 13 Unsteady lift for downstream traveling sharp-edged


gusts. For = 0 (Vg = ) the results lead to the Wagner

function. For = 1 (Vg = 0) the results reduce to the Kussner


function. For intermediate values of a different series of
results are obtained as the gust propagation speed increases.

(Vg = 0), the results reduce to the Kussner function. For


intermediate values of , a different series of results are
obtained as the gust propagation speed increases from zero
( = 1). The noncirculatory term is responsible for the very
large peaks in the lift produced as decreases. The lift
reaches a maximum at the point when the airfoil is about
half way into the gust. It can be seen that the magnitudes
of these peaks are often larger than the steady-state lift coefficient of 2 per radian.
Numerical Integration Methods
Equation 14 or 18 is usually solved numerically for discrete values of time. For a discretely sampled system at
times s = s, 1 , 2 , ....i , then using Eq. 14 e (s) can be
written as

d
(i )(s i )i
i=1 ds

e (s) = (0)(s) +

= (0)(s) +  (1 )(s 1 )1 +
 (2 )(s 2 )2 + ...
+ (i )(s i )i + ...

(21)

with the summation extending over all inputs that have


acted up to the instant s. Therefore, the result for e (s)
requires the storage of  (s),  (s 1 ), ... at all previous
time steps, and the repeated re-evaluation of the indicial
function for each s i at each new time step.
Fortunately, if the indicial function is written in exponential form, the evaluation can be conducted easily using
a recurrence solution. If a general two term exponentially
growing indicial function is assumed such that
(s) = 1 A1 eb1 s A2 eb2 s

13 OF 28
A MERICAN I NSTITUTE OF A ERONAUTICS AND A STRONAUTICS PAPER 2002-0037

(22)

then the Duhamel integral in Eq. 14 can be written as




s d
e (s) = (0)(s) +
(s )d
0 ds


= (0) 1 A1 eb1 s A2 eb2 s +
 s

d 
1 A1 eb1 (s) A2 eb2 (s) d
0 ds
= (0) A1 (0)eb1 s A2 (0)eb2 s +
 s
 s
d b1 (s)
d(s) A1
d
e
0
0 ds
 s
d
A2
(23)
()eb2 (s) d
0 ds

Notice that the terms A1 (0)eb1 s and A2 (0)eb2 s containing the initial value of are short term transients and
can be neglected. Therefore, the Duhamel integral can be
rewritten as
e (s) = (s) X(s) Y (s)

(24)

where the X and Y terms are given by


X(s) = A1
Y (s) =

 s
d b1 (s)
d
e
0

ds

ds

 s
d b2 (s)
A2
e
d

(25)
(26)

Assuming a continuously sampled system with time step


s (which may be nonuniform), then the one-step recursive
formulas can be derived from the previous equations to give
X(s) = X(s s)eb1 s + A1 s

(27)

Y (s) = Y (s s)e

(28)

b2 s

+ A2 s

which will be called Algorithm-1. Notice that recursive


functions X and Y contain all the time-history information of the unsteady aerodynamics and are simply updated
once at each time step, thereby providing numerically efficient solutions to the unsteady lift for arbitrary variations
in . Obviously, the above results can be extended to other
modes of forcing, such as variations in vertical velocity,
gusts etc., and to any number of exponential terms that may
be used to represent the indicial function.
To minimize errors associated with the larger time steps
that would be typical of an actual rotor calculation, various alternative sets of recursive formulas can be obtained.9
If s (or the products b1 s or b2 s) are large, another
approximation based on the mid-point rule can be used
(Algorithm-2). In this case
X(s) = X(s s)eb1 s + A1 s eb1 s/2 (29)
Y (s) = Y (s s)eb2 s + A2 s eb2 s/2

(30)

This method gives errors of less than 1% from an exact solution if each of b1 s and b2 s are less than 0.25, which is a
much more practically realizable option in a rotor analysis.

A third method (Algorithm-3) based on the Simpsons


rule can also be used.9 In this case
X(s) = X(s s)eb1 s +


A1
s 1 + 4eb1 s/2 + eb1 s
6
Y (s) = Y (s s)eb2 s +


A2
s 1 + 4eb2 s/2 + eb2 s
6

(31)

(32)

Errors of less than 0.05% from an exact solution will be


obtained using this algorithm if each of b1 t and b2 s are
less than 0.5. Therefore, despite some minor additional
computational overhead, this will normally be the preferred
method if s must be chosen to take larger values, say to
minimize computational costs in the overall rotor analysis.
Tower Shadow Modeling

Tower shadow is a good example of a problem where


there is a need to properly distinguish the effects on the
airloads arising from changes in angle of attack from the
effects resulting from a nonuniform velocity field. In an essentially incompressible flow, the former can be viewed as
the net response to changes in angle of attack using superposition with the Wagner function, which is equivalent to
assuming that at any instant the angle of attack is constant
over the chord of the airfoil. This assumption is typical of
what is used in many unsteady aerodynamic models for all
modes of forcing (see later). The latter can be viewed as the
net response to a nonuniform vertical component of the velocity field using superposition with the Kussner function,
which is the correct assumption.
Tower shadow manifests as a velocity deficit in the flow
behind the support tower. It can be modeled as a spatial
variation in the velocity normal to the surface of the blade.
In a two-dimensional model, the airfoil section rapidly
moves through a velocity field as function of time, producing an attenuation in unsteady lift and a phase lag compared
to the quasi-steady lift values. The problem has been modeled here by assuming an airfoil traveling at unit velocity in
a normal velocity field given by w = 0.08 + 0.02 cos(5)
where is in radians defined for the range 144
216 and w = 0.1 elsewhere. The results in Fig. 14 show
the differences in the unsteady airloads using the two assumed models clearly there are significant differences in
the results; the differences are larger than that can be accounted for by uncertainties in static airfoil characteristics
alone, i.e., Reynolds number effects on lift-curve-slope etc.
Therefore, attention to detail in the modeling does matter.
This point is further reinforced in Fig. 15. Results are
shown for the same tower shadow simulation, but with
different azimuthal step sizes and choices of superposition algorithm. Figure 15(a) shows the results that would
be obtained using the indicial model with superposition
Algorithm-1, which is the algorithm found in many of the
unsteady aerodynamics and dynamic stall models used in
helicopter and wind turbine simulations. Notice there is
considerable sensitivity to step size, which manifests as

14 OF 28
A MERICAN I NSTITUTE OF A ERONAUTICS AND A STRONAUTICS PAPER 2002-0037

Unsteady lift coefficient

0.7
0.6
0.5
0.4
Quasi-steady
Using Kussner function
Using Wagner function

0.3
0.2
90

120

150
180
210
Blade azimuth - deg.

240

270

Fig. 14 Prediction of the unsteady lift during a simulated


tower shadow encounter.

Unsteady lift / quasi-steady lift

(a)

Unsteady lift / quasi-steady lift

(b)

1.3
1.2
1.1
1
0.9
0.8
0.7
90

Recurrence algorithm 1
= 1 o
= 5 o
= 10 o

120

150

180

210

240

270

150
180
210
Blade azimuth - deg.

240

270

1.3
1.2
1.1
1
0.9

Recurrence algorithm 3
= 1 o

0.8
0.7
90

= 5 o
= 10 o

120

Fig. 15 Effect of sampling on lift prediction during a simulated tower shadow encounter using numerical solution to
Duhamel integral. (a) Algorithm-1, (b) Algorithm-3.

both amplitude and phase errors. The situation is dramatically improved with the use of Algorithm-3, which despite
some minor additional computational overhead, will be the
preferred algorithm.
Non-uniform Incident Velocity

In yawed flow, a representative blade element on a wind


turbine will encounter a time-varying incident velocity.
Under these conditions, there are additional unsteady aerodynamic effects to be considered. With a time-varying
incident flow velocity, the shed wake behind the airfoil is
convected at a nonuniform velocity, and this causes sev-

eral complications in the mathematical treatment of the


wake induced velocity using the two-dimensional models.
Isaacs57 derived a closed form solution for the additional
unsteady aerodynamic effects of a harmonically varying
free-stream velocity. Greenberg58 published a similar frequency domain theory, but made a high frequency assumption about the shed wake behind the airfoil to obtain a
solution in terms of the Theodorsen function alone. Johnson26 has also discussed the general problem of a varying
velocity onset on the unsteady loads.
The problem of a time-varying free-stream velocity on
the unsteady aerodynamic response has been examined
by Van der Wall & Leishman,59 both from the classical frequency-domain approach and also from the timedomain. In the time-domain the effect of a time-varying
velocity in an incompressible flow can be modeled using
Duhamel superposition with the Wagner function. However, the appropriate apparent mass forces must also be
included in the solution. In this case


b d(V )
Cl (t) = 2 h +
ab
+
V
dt


 s
2
d(V )
V (0)(s) +
W (s )d (33)
V
dt
0
The distance traveled through the flow is now proportional
to the area under the velocity, as given by Eq. 15.
Results calculated using Eq. 33 are shown in Fig. 16 for
the lift on an airfoil at a constant angle of attack, 0 , experiencing a harmonic oscillation of velocity of the form
V (t) = V (1 + sin t) at a reduced frequency, k, of 0.2
with = 0, 0.2, 0.4, 0.6, 0.8. The Duhamel integral was
solved by means of a finite-difference approximation using
the numerical scheme described previously. The lift coefficient in Fig. 16 is shown as fraction of the quasi-steady lift
coefficient, (Cl )qs = 20 . Also shown in this figure are
results computed by means of a finite-difference method
based on a CFD (Euler) solution. It will be seen from
Fig. 16 that at a given time the primary effect of unsteadiness in this case is either a further increase or attenuation of
the unsteady lift compared to the quasi-steady result, which
is a result already noted for oscillations in angle of attack.
It is fair to say that for wind turbine applications, which
may involve large values of , the modeling of this effect
will be very important.
State-Space Models

For some applications, it is more convenient if the blade


element unsteady aerodynamic model can be written in the
form of ordinary differential equations, i.e., in state-space
form. In this case, the problem can be solved using any
standard algorithm that can numerically solve differential
equations. The state equations describing the behavior of a
two-dimensional unsteady airfoil can be obtained through
direct application of Laplace transform methods to the indicial response.60
Consider a general indicial lift response, , which is to
be approximated by a two term exponential function given

15 OF 28
A MERICAN I NSTITUTE OF A ERONAUTICS AND A STRONAUTICS PAPER 2002-0037

Unsteady lift / quasi-steady lift, cl / (cl) qs

V(t) = V (1 + sin t), k = 0.2

2.5

Indicial, =0.2
Indicial, =0.4
Indicial, =0.6
Indicial, =0.8
Euler method

2
1.5
1
0.5
0

45

90

135

180 225
t - deg.

270

315

360

Fig. 16 Effect of unsteady free-stream velocity fluctuations


on the unsteady lift of an airfoil at constant angle of attack.

by Eq. 22. The lift response to an input (t) can be written


in state-space form as



 

x1
x1
0
= [A]
+
(t)
(34)
1
x2
x2
where

[A] =

0
2V 2

b1 b2

1

(b1 + b2 ) 2V
c


(35)

and the corresponding equation for the lift coefficient is




x1
Cl (t) = Cl [C]
x2


+ Cl (1 A1 A2 ) (t)

[C] = (A1 + A2 )b1 b2

2V 2
c

(A1 b1 + A2 b2 )

2V

(36)


(37)
It should be appreciated that Theodorsen and Sears solutions, the Wagner and Kussner solutions with Duhamel
superposition, and the above state-space model are simply different mathematical realizations of the same aerodynamic system.60, 61

Dynamic Stall
Dynamic stall will occur on any airfoil or other lifting
surface when it is subjected to time-dependent pitching,
plunging or vertical translation, or other type of motion,
that takes the effective angle of attack above its normal
static stall angle.8, 62, 63 Under these circumstances, the
physics of the onset of flow separation and the development of stall is distinctly different to the stall mechanism
exhibited by the same airfoil under static (quasi-steady)
conditions.
Dynamic stall is, in part, characterized by a delay in the
onset of flow separation to a higher angle of attack than
would occur statically see Fig. 17. However, various laboratory tests on two-dimensional airfoils and experiments
with rotors have shown that the most distinguishing feature
of dynamic stall is the shedding of a concentrated vortical disturbance from the leading-edge region of the airfoil

section. This disturbance is subsequently swept over the


chord, and can result in significant changes in the sectional
forces that are considerably different to the static loads.
Dynamic stall is also accompanied by much larger phase
variations in the unsteady airloads as a result of significant hysteresis in the flow developments; that is, the values
of the airloads at the same angle of attack may be very
different depending on whether the flow is separating or
reattaching.
Dynamic stall has been extensively studied experimentally, mostly using oscillating two-dimensional airfoils in
wind tunnel experiments. The majority of the documented
experimental results are for airfoils oscillating in pitch (angle of attack), but there are some limited amounts of data
available for other types of motions, such as plunging oscillations and constant angular rate (ramp) type motion.
Different modes of forcing are very desirable in the study of
dynamic stall because the problem may contain nonlinear
physics of great importance that cannot be understood by
examining the airfoil behavior in response to only one type
of forcing. For example, plunging experiments are useful because pitch rate effects (i.e., terms) can be isolated
from the problem; ramp (constant pitch rate) experiments
are useful because acceleration effects (i.e., terms) can
be eliminated.
If dynamic stall becomes sufficiently severe on a wind
turbine or a helicopter rotor, it can produce loads that may
quickly exceed the structural fatigue limits. Therefore, consideration of dynamic stall in the rotor design will more
accurately define the operational and performance boundaries.6 However, the accurate prediction of the combination
of conditions that will produce dynamic stall onset, as well
as the prediction of the subsequent effects of dynamic stall
on blade loads and performance, is certainly not an easy
task for the analyst. The complicated nonlinear physics of
dynamic stall means that the behavior can only be completely modeled by means of numerical solutions to the
Navier-Stokes equations. This, like other CFD problems
that involve unsteady, separated flows, is a formidable challenge that is not yet practical. However, rapid increases in
computer resources has enabled considerable progress to
be made12, 64
Currently, dynamic stall must be modeled using more
parsimonious, semi-empirical models, for which a number
of different approaches have been developed for helicopter
applications and adapted for wind turbine use. While often
giving good results, these models are not strictly predictive
tools, and can really only be used confidently for conditions
that are bounded by their validation with experimental data.
Semi-Empirical Models of Dynamic Stall

There are several dynamic stall models available that can


be adapted by the wind turbine analyst. A few of the models of dynamic stall that are in current use, especially in
the helicopter community, are a form of resynthesis of the
measured unsteady airloads, which are based on results
from two-dimensional oscillating airfoils in wind tunnel

16 OF 28
A MERICAN I NSTITUTE OF A ERONAUTICS AND A STRONAUTICS PAPER 2002-0037

oefficient, C
Lift
Cll
L ift ccoefficient,

1.5

Stage 1: Airfoil exceeds static stall angle, then


flow reversals take place in boundary layer.

Unsteady
Static

0.5

Stage 2: Flow separation at the leading-edge, followed


by the formation of a 'spilled' vortex. Moment stall.

5
0

Cm
oeffic ient , C
Moment
m
Moment ccoefficient,

10

15

2
0

25

0.1
5

Stage 2-3: Vortex convects over chord, it induces


extra lift and aft center of pressure movement.

0
1

-0.1
Unsteady
Static

-0.2
-0.3

3
10

15

20

25

Stage 3-4: Lift stall. After vortex reaches trailing-edge, the


flow over upper surface becomes fully separated.

0.6

coefficient, C
Drag
Cd
D rag coefficient,
d

Unsteady
Static

0.5
0.4
0.3

Stage 5: When angle of attack becomes low enough, the flow


reattaches to the airfoil, front to back.

4
0.2
0.1
0

5
0

1
10

15

20

25

Angle of attack - deg.

Fig. 17 Schematic showing unsteady airloads and flow physics for a two-dimensonal airfoil undergoing dynamic stall.

experiments. Other semi-empirical models of dynamic


stall contain representations of the essential physics using
sets of linear and nonlinear equations for the lift, drag,
and pitching moment. The nonlinear equations may have
several empirical coefficients, most of which must be deduced from unsteady airfoil measurements. However, the
root of all of these models is based on classical unsteady
thin-airfoil theory, the basis of which has been discussed
previously. The development of the nonlinear part of such
models are more subjective, and requires skillful interpretation of experimental data. Most of the experimental results
of dynamic stall have been obtained for the types of airfoils
used on helicopters, but there are also good results for the
classes of airfoils used on wind turbines.65 Most of these
models, however, remain in a perpetual state of flux as the
level of detail is refined and/or more experimental data become available for formulation and/or correlation purposes.

when using parsimonious models.

While the qualitative features of dynamic stall have been


found to be the same for a wide range of airfoil sections
and operating conditions (Reynolds number, Mach number), the quantitative behavior of the airloads show more
subtle variations, especially for different airfoil shapes. It
is these more subtle aspects of the dynamic stall problem
that make its accurate prediction difficult for the analyst

This is an another resynthesis method that was initially


developed by Gross & Harris68 and Gormont.69 In the
Gamma Function method, the influence of airfoil motion
is determined by computing an effective angle of attack.
Unsteady effects are first accounted for using Theodorsens
theory applied at the appropriate reduced frequency of the
forcing. From this, a correction is applied using the instan-

UTRC , A, B Method
This is a pure resynthesis method, with the approach being described by Carta et al.66 and Bielawa.67 The basis of
this method is that in attached flow the airloads can be ex
pressed in terms of the forcing parameters , A = c/2V
2 /4V 2 . In an attempt to isolate the nonlinear
and B = c
dynamic contributions to the airloads, the static coefficients
are subtracted from the total airloads. By appropriate crossplotting and interpolating for given instantaneous values of
the A and B parameters, the contributions to the dynamic
stall airloads can be reconstructed and added to the static
values. The method has met with limited success and large
data tables must generated for each airfoil.
Boeing-Vertol Gamma Function Method

17 OF 28
A MERICAN I NSTITUTE OF A ERONAUTICS AND A STRONAUTICS PAPER 2002-0037

taneous value of the gamma function; this gamma function


is obtained empirically as a function of Mach number from
two-dimensional oscillating airfoil tests on the appropriate airfoil. The corrected angle of attack is then used
to obtain values of the airloads from the static force and
moment curves. This has the effect of delaying the onset of
stall to higher angles of attack with increasing pitch rate, a
result observed experimentally. Reasonable predictions of
the unsteady airloads seem possible with this method, but
the quantitative accuracy is not good, especially near stall
onset or for light dynamic stall.
Beddoes Time-Delay Method
Beddoes70 has developed a time-domain based model for
dynamic stall. Unlike the resynthesis methods, the philosophy behind this method is to try to model the key causal
factors of the dynamic stall process itself. The behavior of
the airloads in attached flow are obtained from Duhamel superposition using the Wagner indicial response function, as
described previously. A key feature of this model is the use
of two nondimensional time delays. These time delays represent periods of nondimensional time that exist between
identifiable dynamic stall flow states; the first delay represents the period in the onset of separation after the static
stall angle has been exceeded and the time required for
the initial separation to develop.71 The second time delay
represents the time during which the leading-edge vortex
shedding process occurs. These time delays have been obtained from a statistical analysis of many airfoil tests over a
relatively wide range of Mach number. Similar studies have
been done by Galbraith et al.72 for the lower Mach numbers
appropriate to wind turbine applications. The results from
the time-delay model have been shown to give good predictions of dynamic stall. The method also requires relatively
few empirical constants.
Gangwanis Method
Gangwani73 has developed a synthesized airfoil method
for the prediction of dynamic stall. This method is also
based in the time-domain. To model the airloads in attached flow, a pseudoWagner function is used with a
finite-difference approximation to the Duhamel superposition integral, very much in the same manner as for the
Beddoes time-delay model. In the nonlinear part of the
model, a number of equations with empirical coefficients
are used to represent the forces and moments produced by
the various dynamic stall events. These equations are based
on the determination of delayed angles of attack, with the
coefficients being derived from steady and unsteady airfoil
data. A disadvantage with this method is the relatively large
number of empirical coefficients, nearly all which are derived from oscillating airfoil data.
Johnsons Method
Johnson74 has developed a parsimonious dynamic stall
model for the sectional airloads. Experimental data were
used to develop the model in a form that could be used to
correct the static stall lift and pitching moments as func-

tions of pitch-rate. Stall onset was represented by defining


vortex shedding to occur just above the static stall angle of
attack, assuming that vortex shedding produced increments
in lift and nose-down moment that increased linearly to a
peak value over a finite time, followed by a decay back to
the static loads. Reasonable predictions of the unsteady lift
seem possible with this method, although it is not particularly well validated.
ONERA Method
This model describes the unsteady airfoil behavior using
a set of nonlinear differential equations. The model was
first described by Tran & Pitot,75 Tran & Falchero76 and
McAlister et al.,77 with various suggested modifications by
Peters.78 A version of the ONERA model has been evaluated by Reddy & Kaza.79 A later version of the model
(the ONERA Edlin model) is documented by Petot.80 The
coefficients in the equations are determined by parameter
identification from experimental measurements on oscillating airfoils. Like several other models, the airloads are
expressed as a sum of two components; a component associated with the linear (attached flow) behavior, and an
increment that represents a deviation from the linear value
resulting from stall. The model requires a significant number of empirical coefficients. The later ONERA BH model,
as documented by Truong,81 requires slightly fewer coefficients, and also adapts some elements of the Leishman
Beddoes model.
LeishmanBeddoes Method
While many of the fore noted semi-empirical models are
adequate for most rotor design purposes, they usually lack
rigor and generality when applied to different airfoils. Another major problem with some of the models is that a
significant number of empirical coefficients may need to
be derived from unsteady airfoil measurements. The nonlinearity of the dynamic stall problem means that these
coefficients may be non-unique. Generally, a set of coefficients for the model must be derived for each and every
airfoil, and also over the appropriate range of Reynolds
numbers and Mach numbers, assuming such measurements
are available. In cases where unsteady airfoil measurements are not available, the models cannot be used with
the same confidence levels to predict the nonlinear airloads. Unfortunately, a problem is that usually the full
range of Reynolds numbers, reduced frequencies, and to
some extent, airfoil shapes, cannot be studied in the same
test facility or wind tunnel. Therefore, the problems and
uncertainties in data quality that have been pointed out by
McCroskey82 in regard static airfoil measurements also apply to the dynamic case, and these uncertainties can be
carried forward into the models.
Leishman & Beddoes83, 84 have developed a model capable of representing the unsteady lift, pitching moment and
drag characteristics of an airfoil undergoing dynamic stall.
This model has been developed to overcome certain shortcomings of other unsteady aerodynamic models that were

18 OF 28
A MERICAN I NSTITUTE OF A ERONAUTICS AND A STRONAUTICS PAPER 2002-0037

previously available for use in helicopter rotor analysis.


The emphasis in this model is on a more complete physical
representation of the overall unsteady aerodynamic problem, but still keeping the complexity of the analysis down
to minimize computational overheads. The model was initially developed by Beddoes,85 with various developments
documented by Leishman86 and Tyler & Leishman.87 The
latest versions of the model encompass elements described
previously, including the effects of different modes of forcing (i.e., pitch versus plunge versus nonuniform vertical
velocity fields), unsteady free-stream velocity effects, improved numerical methods, etc. The model has also been
adapted and modified by Pierce & Hansen65 for the class
of airfoil sections used on wind turbines.
The LeishmanBeddoes model consists essentially of
four subsystems: 1. An attached flow model for the unsteady (linear) airloads based on Duhamel superposition,
2. A separated flow model for the nonlinear airloads, 3. A
dynamic stall onset model, 4. A dynamic stall model for
the vortex induced airloads. The sub-models are connected
as an open-loop system in the form of a Kelvin chain
see Fig. 18. An important feature is that rigorous representations of compressibility effects, which are essential
for helicopter applications, are included in the model. The
model has also been developed as a set of ordinary differential equations,60, 88 which, as mentioned previously, can
be useful for some applications. The treatment of nonlinear aerodynamic effects associated with flow separation on
the airfoil are derived from Kirchhoff/Helmholtz theory,89
which can be used to relate the airfoil lift to the angle of
attack and an effective trailing-edge separation point. In
application, the experimental static lift characteristics are
used with the Kirchhoff/Helmholtz model to define this
effective separation point variation that can then be generalized empirically as a function of angle of attack and used
to accurately reconstruct the nonlinear static airloads.
To represent the effects of dynamic stall, a further subsystem emulates the dynamic effects on the airloads of the
accretion of vorticity into a concentrated leading-edge vortex, the passage of this vortex across the upper surface of
the airfoil, and its eventual convection downstream. The
dynamic stall process begins when an equivalent leadingedge pressure parameter reaches a Mach number/Reynolds
number dependent critical value indicative of leading-edge
or (for high free-stream Mach numbers) shock induced separation.71, 85 The lift then continues to build in a manner
that is related to the rate of change of movement of the
separation point. A first-order dynamic system with an empirically derived time constant governs the accumulation
of vortex lift, and in the limit when the changes in angle
of attack become small, the vortex lift dissipates and the
airloads return to their static (nonlinear) values. The corresponding pitching moment during the vortex shedding
process is obtained using a center of pressure function estimated from correlation studies with unsteady airfoil measurements in dynamic stall.
An advantage of the LeishmanBeddoes model is that

Forcing (input)

Unsteady attached
flow module

Nonlinear trailing-edge
separation module
Time
constant
modifications

Time
constant
modifications

Leading-edge flow
separation module

Vortex shedding
module

Output of airloads

Fig. 18 Flow chart showing elements of the Leishman


Beddoes dynamic stall model.

it uses relatively few empirical coefficients, with all but


four being derived from static airfoil data. There are two
time constants used in the second subsystem of the model,
and one in the dynamic stall subsystem. The fourth parameter is a nondimensional time period related to the
duration of the dynamic stall process. The first time constant is used in the stall onset model, and recognizes that
the pressure distribution on the leading-edge of the airfoil
is not in phase with the unsteady lift. This behavior is
also modeled as a first-order dynamic system, and the time
constant is derived from experimental measurements by examining the relationship between the unsteady lift and the
pressure response near the leading-edge. The second time
constant represents unsteady effects on the boundary layer
response and the movement of the separation point; this
time constant has been obtained through a combination of
unsteady boundary layer theory and experimental measurements. The third time constant is used for the dynamic
lift subsystem, which has been described previously. A
fourth coefficient is used in the center of pressure function,
and represents the time period (in semi-chords of airfoil
travel) between the initiation of vortex shedding from the
leading-edge and the point when the vortex reaches the
trailing-edge of the airfoil. This coefficient is obtained statistically from correlation studies using a variety unsteady
airfoil measurements in dynamic stall, and simply recognizes that despite the type of airfoil motion the dynamic
stall process occurs (on average) over a common timescale. To simulate the effects of the complex changes in the
flow topology during dynamic stall, the two time constants
involved in the behavior of the dynamic stall subsystem and
the trailing-edge separation subsystem are modified in a

19 OF 28
A MERICAN I NSTITUTE OF A ERONAUTICS AND A STRONAUTICS PAPER 2002-0037

= 12.45 + 3.14 sin t; k = 0.116, M = 0.4

1.5

1.5

0.5

0.5

10

15

20

Test data
Model
Static data

Loop of
negative
damping

Loop of
positive damping

-0.1

10

15

20
Loop of
negative
damping

-0.1
Net damping is negative

-0.2

Loop of
positive damping

0.1

0.1

= 12.29 + 4.94 sin t; k = 0.124, M = 0.4

Cn

Test data
Model
Static data

eq

5
10
15
Equivalent angle of attack, (deg.)

Net damping is approximately zero

-0.2

20

eq

5
10
15
Angle of attack, (deg.)

20

Fig. 19 Predictions of unsteady lift and pitching moment using the LeishmanBeddoes model for an airfoil undergoing
oscillatory plunging motion.

Fig. 20 Predictions of unsteady lift and pitching moment using the LeishmanBeddoes model for an airfoil undergoing
oscillatory pitching motion.

logically determined feed-back and feed-forward sequence


(Fig. 18), again based on correlation studies with experimental results.
Representative results made using the Leishman
Beddoes model are shown in Figs. 19 and 20 for an
NACA 23015 airfoil undergoing plunging and pitching
oscillations, respectively. Notice that the general features
of dynamic stall are evident, the results showing a qualitative similarity for both types of forcing. The hysteresis
loops predicted by the dynamic stall model are in good
agreement with the experimental measurements, especially so bearing in mind that these measurements were
not used to develop the model. In both cases, stall onset
is clearly postponed to above the static stall angle of attack, followed by vortex shedding. Both cases suggest that
an organized vortex is formed and transported downstream
over the chord, although for the plunge case the degree of
stall penetration is slightly less. While less important for
wind turbine applications, the pitching moment is a sensitive indicator to the complex changes in the chordwise
pressure distributions occurring during dynamic stall, and
so the ability to predict the pitching moment can be viewed
as a success indicator for the modeling.
Carta90 has postulated that dynamic stall occurred on the

airfoils during certain pitch oscillation cases but not in the


corresponding plunge cases, even though the same equivalent angle of attack history was imposed. It appears that
this behavior can, in part, be traced to the (inviscid) pressure distribution at the leading-edge of the airfoil, and perhaps also to the additional effects of the unsteady boundary
layer response. For pitch oscillations the pitch rate lowers
the leading-edge pressure gradient91 so that the conditions
which delimit attached flow should be met at a higher angle of attack than for the plunge oscillations. Tyler &
Leishman87 have confirmed that the stall onset behavior is
related to the additional effect of pitch rate contributions to
the unsteady airloads during pitch oscillations.
Overall, it can be seen from the results shown in Figs. 19
and 20 that the dynamic stall model provides fairly satisfactory predictions of the unsteady airloads, and for
both pitching and plunging oscillations. These predictions
can only be considered as representative, and other semiempirical models may provide different or better results.
Correlation studies, therefore, are important if measurements are available. None of the current models, however,
can be considered as the last word on the problem, and it
is likely that improved semi-empirical models of dynamic
stall for wind turbine applications will continue to be developed in the future.

For a plunging motion, the equivalent angle of attack is given


.
by eq = h/V
Many of the measurements show mean or steady offsets
which have been attributed to uncertainties caused by thermal drift
of pressure transducers. Such offsets are apparent in the experimental data shown in Figs. 19 and 20.

Three-Dimensional Effects
In high winds, when much of a wind turbine blade
can be stalled, existing performance methods can predict
power outputs that are considerably lower than those actually measured. While part of the problem is the imprecise

20 OF 28
A MERICAN I NSTITUTE OF A ERONAUTICS AND A STRONAUTICS PAPER 2002-0037

treatment of delayed stall as a result of unsteady aerodynamic effects, there are also subtle three-dimensional
effects that contribute to the problem.92 While a full understanding and modeling of these problems is the subject
of ongoing research, they can, in most cases, be traced to
three-dimensional boundary layer developments on the rotating blades.
One aspect of the three-dimensional aerodynamic problem is the centrifugal and Coriolis effects acting on the
boundary layer in a rotating flow environment. Several
experimental and computational studies have provided insight into the problem.9396 From an order of magnitude
analysis of the three-dimensional boundary layer equations
applied to a rotating flow environment, Snel94 has found
that the Coriolis acceleration terms can act to alleviate adverse pressure gradients and so may delay the onset of
flow separation and stall. The effects are powerful at the
inboard part of the blade, for which experimental results
have shown significant increases in sectional maximum
lift coefficients beyond what would be expected based on
two-dimensional static measurements. Similar results have
been suggested using CFD methods.95 The experiments of
Dwyer & McCroskey96 also suggest favorable effects on
the spanwise development of the boundary layer on a rotating blade, which tend to delay the onset of flow separation
to a higher angle of attack. Various ad hoc methods to
model the observed effects have been developed,97, 98 but
a rigorous approach is lacking. Also, none of the models
have been validated under unsteady conditions or during
dynamic stall.
Another quasi three-dimensional effect can be traced to
the direction of the incident flow velocities on the rotating
blades. Because of the relatively low rotational velocity
of wind turbines, the local sweep angle of the flow at any
blade element on a wind turbine can be very significant,
especially when the turbine is yawed with respect to the
oncoming wind. As shown in Fig. 21, a sweep angle, ,
can be defined at each blade element in terms of the normal
and radial velocity components UT and UR , respectively.
The radial component of flow velocity, UR , can affect the
development of the three-dimensional boundary layer development and the onset of stall, leading to a somewhat
more complicated behavior than would be predicted under
nominally two-dimensional conditions.
In the classical blade element theory, one neglects the
effect of any sweep angle on the airloads. This is in accordance with the independence principle. However, when
an airfoil is operated at high angles of attack near stall, this
may not be a valid assumption. For example, the effect of
sweep angle on the static lift characteristics of a wing100
is shown in Fig. 22. The results are presented in the conventional blade element format that is, in terms of angle
of attack and velocities normal to the leading-edge of the
airfoil section. These results confirm that, at least in the
attached flow regime, the independence principle is a valid
See,

for example, Jones & Cohen.99

Velocity component normal


to blade ( U cos)
Incident velocity

Radial velocity
component ( U sin)

UT

UR

Fig. 21 Definition of sweep angle with respect to a blade element.

assumption. However, it will be noted that in the high angle of attack region much larger maximum lift coefficients
are obtained with increasing sweep angles.
The upshot of these observations is that if these sweep
effects are carried forth into the blade element environment
of a wind turbine, then they will help to delay the onset
of stall. Various early studies of the problem applied to
helicopter rotors, including the work of Harris101 and Gormont,69 suggest that improvements in rotor performance
prediction can be obtained by including a static stall model
for sweep effects. However, it must be remembered that
when the onset of stall occurs on either a wind turbine or a
helicopter rotor, the stall is usually dynamic in nature. The
fundamental question is whether the delay in stall and increase in static lift shown in Fig. 22, and three-dimensional
boundary layer effects in general, are also carried forth into
the dynamic stall regime.
To examine this problem, a series of experiments have
been conducted by St. Hillaire & Carta102, 103 where a
NACA 0012 airfoil was oscillated in angle of attack at different (constant) sweep angles of 0 and 30 . From these
results it appears that for unsteady flows when the flow is
fully attached the independence principle also applies; any
unsteady effects associated with sweep are small and probably smaller than uncertainties associated with the measurements themselves. Figure 23 shows an example of the
dynamic stall results. Compared with the static case where
approximately a 20% higher maximum Cn was attained for
= 30 , the unsteady case shows a delay in dynamic lift
stall to a higher angle of attack, but not to a significantly
higher value of lift. Also, for = 30 somewhat narrower
lift hysteresis loops are produced, and the mean value of lift
is somewhat higher. Thus, it appears that modeling the effects of sweep may serve to produce higher values of power
output on a wind turbine compared to predictions obtained
when sweep effects are not modeled.
Further insight into the physics of the dynamic stall process with sweep can be deduced from the behavior of the
pitching moment. The divergence in the pitching moment
(moment stall) is noted to occur at the same nominal value
of angle of attack for both the = 0 and = 30 cases.

21 OF 28
A MERICAN I NSTITUTE OF A ERONAUTICS AND A STRONAUTICS PAPER 2002-0037

Normalized lift coefficient, C

= 0o unsteady
= 30o unsteady

NACA 0012

= 0 steady
= 30o steady

1.5

Sweep angle - deg.

10

20

30

40

50

0.5
= 15o + 8o sin t; k = 0.075, M = 0.4

60

10

15

20

25

10

15

20

25

10
15
20
Angle of attack, - deg.

25

0.1

Normalized moment
coefficient, C

-1
-10

0
15
30
35
40
45
60

-0.4
-10

-0.2

-0.1
-0.2

10

20

30

40

50

60

-0.3
0.6

1
d

0.4
C

Normalized drag coefficient, C

1.5

0.2
0.5

0
0
-10

10

20

30

40

50

60

Angle of attack normal to leading-edge - deg.

Fig. 22 Effect of sweep angle on the static lift characteristics of a wing.100 The results are presented in the conventional
blade element format.

However, the slope of the moment curve during the next


part of the cycle is clearly less for the = 30 case. Also,
the minimum pitching moment is reached at a higher angle of attack. This suggests that the delay in dynamic lift
stall to a higher angle of attack in swept flow is, in part,
because of a lower velocity at which the shed leading-edge
vortex is convected over the chord. The analysis performed
on the airfoil pressure time histories103 also supports this
observation.
Modeling efforts of this three-dimensional sweep problem have actually met with good success.86 With the
LeishmanBeddoes model, the static airfoil behavior in
swept flow can be used to deduce a modified effective
trailing-edge separation point. A modification is also made
to the time constant for the convection of the stall vortex
over the chord, in accordance with the experimental observations. The net effect is that the predictions of the
unsteady airloads are in good quantitative agreement86 with
the measurements.

Fig. 23 Representative example of the dynamic stall behavior


of a swept, nominally two-dimensional airfoil.102, 103
Challenges in Developing Better Dynamic Stall Models

Existing dynamic stall models represent, to engineering


levels of approximation, the integrated forces and moments
produced on an airfoil during dynamic stall. While in many
conditions these models give credible predictions, there
are many lessons that one can learn from past experiences
in using these models. Reddy & Kaza79 independently
have compared and contrasted some of these models, from
which the general quantitative capabilities and deficiencies were documented. Tan & Carr104 have presented a
summary of results for most semi-empirical dynamic stall
models, as well as CFD approaches to the dynamic stall
problem. Results have been compared for both an oscillating two-dimensional airfoil and an idealized 3-D problem
in the form of an oscillating cantilevered finite-wing. Used
with caution, it would seem that most of the semi-empirical
models are adequate for use in a wide variety of rotor analysis; much comes down to the confidence levels that can
be established through correlation studies with measurements, both at a two-dimensional level as well as when
included inside the rotor models. With appropriate validation, and the appropriate exclusion of uncertainties and

22 OF 28
A MERICAN I NSTITUTE OF A ERONAUTICS AND A STRONAUTICS PAPER 2002-0037

other unwanted noise from the measurements, it seems that


very credible predictions of the airloads can be obtained.
There are several issues, however, that must be addressed
if improved representations of the unsteady airloads are to
be obtained for wind turbine applications. The issues can
be broadly placed into a category that addresses a better
understanding of the physical issues, and another category
that addresses numerical modeling.
Physical Issues
One limitation with modeling includes the accuracy with
which the stall onset can be predicted; that is, the prediction of the combination of unsteady angle of attack and
Reynolds number (or Mach number) that produce the onset of dynamic leading-edge flow separation. The physics
defining the onset of dynamic stall is relatively complicated, and may involve the role of compressibility even
at low free-stream Mach numbers.71 The onset, however,
is closely correlated with the attainment of a critical value
of the leading-edge pressure and/or leading-edge pressure
gradient.85 Therefore, a prerequisite to predicting the onset
of dynamic stall is good predictions of the airfoil behavior
under unsteady attached flow conditions, a point alluded to
previously.
Most of the dynamic stall models in current use have
been validated for the types of airfoil sections used on helicopters, and also over the range of Reynolds numbers and
Mach numbers typical of helicopter applications. The airfoil sections for helicopter use are designed for high maximum lift coefficients, and so under steady conditions they
tend to stall by either the leading-edge or abrupt trailingedge stall mechanism.9 The operation of many wind turbines are stall regulated, so that wind turbine airfoils are
often designed to have a static stall behavior with a more
progressive onset of trailing-edge separation; these airfoils
may operate over a substantial range in angle of attack with
a pronounced amount of trailing-edge separation. Even for
pitch regulated turbines, which may use more conventional
airfoils, dynamic stall may still be an issue in the operation
of the turbine. Therefore, a prerequisite for the dynamic
stall models is to accurately represent the strong nonlinearities present in the static stall behavior of the types of
airfoils used on wind turbines.
While under unsteady conditions it is known that
trailing-edge flow separation is suppressed by the favorable
effects of positive pitch rate, the development of significant
trailing-edge separation limits the build-up of the suction
pressure and adversity of the pressure gradient over the
leading-edge of the airfoil. The onset of dynamic stall
for these airfoils is still found to be leading-edge dominated, but the quantitative behavior of the airloads is somewhat different. This suggests some interaction between
the trailing-edge separation and the leading-edge flow conditions that ultimately dictate the onset of dynamic stall.
Therefore, it will be necessary to further revisit the modeling criteria used in the models to better define the onset
of dynamic stall and leading-edge vortex shedding for the

airfoil sections used in wind turbine applications.


Similar arguments can be made for the modeling of flow
reattachment after the dynamic stall process, which requires as much care as for modeling dynamic flow separation and vortex shedding, especially if accurate predictions
of the flow hysteresis and aerodynamic torsional damping
are an objective. Airfoils that exhibit pronounced trailingedge separation statically are likely to show substantial
delays on the reattachment of the flow during parts of the
dynamic stall cycle where conditions may be conducive to
flow reattachment see Green et al.? Some of the dynamic
stall models show relatively poor correlation with experimental measurements during flow reattachment, even for
the conventional airfoils used in helicopter applications,
suggesting that this is one area where further basic research
may be most fruitful.
Three-dimensional effects on the problem of dynamic
stall are still poorly understood, and a better understanding of these effects is where many future challenges lie
for the analyst. The mechanisms contributing to the threedimensional aspect of the dynamic stall problem still require a much deeper understanding of the flow before modeling efforts can be successful. It is likely that this is where
the use of results from the recent NREL experiments will
prove very beneficial.92
Numerical & Modeling Issues
The numerical methods used to model unsteady aerodynamics of a blade element operating in fully attached flow
have been previously described. These may be based on a
finite-difference solution to the Duhamel superposition integral (i.e., discrete time form) or developed in state-space
form (i.e., continuous time form). The approaches are numerically equivalent if solved to the same level of accuracy.
The discrete time is often the preferred format and, except
for the ONERA model, forms the basis for most of dynamic
stall models currently being used in helicopter and wind
turbine applications. The nonlinear (dynamic stall) part of
the model is built on top of the attached flow solution, and
this is where substantial empiricism may be included. The
coding of the nonlinear part of the model must be done with
extreme care to ensure that logic or conditional branching
in the algorithm does not cause non-physical transients in
the predictions of the unsteady airloads, especially if large
time (azimuth) steps are involved in the rotor simulation.
This undesirable behavior may produce erroneous predictions, and would be considered unacceptable in a rotor
design analysis.
An example is shown in Fig. 24, where the azimuthal
step size in the dynamic stall calculation has been degraded
from about = 5 to = 20 , the latter being a worst
case situation for a rotor analysis. While in this case the
model exhibits no erroneous behavior, degrading the step
size results in two primary effects. First, the angle of attack
for stall onset is not predicted accurately. Second, there are
insufficient samples during the vortex shedding process, so
the peak nose down pitching moment and maximum drag

23 OF 28
A MERICAN I NSTITUTE OF A ERONAUTICS AND A STRONAUTICS PAPER 2002-0037

2.5
Model, high sampling
Model, average sampling
Model, low sampling
Experiment

accura cy
Mod eling accuracy
Modeling

2
1.5
1
0.5

"Ockham's Hill"
Best
model
Too much noise

= 10 o + 10 o sin t; k = 0.1, M = 0.3

-5

10

15

20

Too little physics

25

0.1

Increasing model complexity

Fig. 25 Predictive accuracy increases with increasing modeling complexity to a point, and then decreases again exhibiting
an Ockhams Hill.

-0.1
-0.2
-0.3
-0.4
-5

-5

10

15

20

25

20

25

0.6

0.4

0.2
0
5
10
15
Angle of attack, - deg.

Fig. 24 Effect of sampling (step size) on predictions of


dynamic stall

is greatly underpredicted. Therefore, it is important for the


analyst to build-up a confidence level with any model selected for the design process, with proper consideration of
the potential effects of step size in the rotor application.
The need to use some empirically defined coefficients
in the dynamic stall models is unavoidable. Empiricism,
however, is not a negative concept, if suitably justified. Existing dynamic stall models have used from as many as 50
empirical coefficients to as few as four; these coefficients
may also be a function of Reynolds number and/or Mach
number. Most modeling efforts follow the relatively simple philosophy of enhancing prediction by using equations
to amplify a pattern in the experimental results and also filtering noise. Complicated models always have a greater
probability of modeling more and more of the unwanted
noise and the uncertainties that are omnipresent in the experimental data. Therefore, one objective for the analyst
should be to balance the complexity of the model by using a minimum number of equations and coefficients, while
maximizing the predictive accuracy and minimizing noise.
One strategy toward this end is that all (most) of the coefficients should have a physical meaning, and should be

easily derived from either steady or unsteady airfoil measurements. Obviously, with a large number of coefficients
it is hard to assign a physical significance to all of them.
More importantly, however, with complex models the
unique identification of the empirical coefficients becomes
difficult and substantially increases the probability of unwanted noise. It is clear that most of the features of
dynamic stall are a result of a few key causal factors,
and so in principle can be modeled using relatively parsimonious models; the stall-delay model of Beddoes70 and
Johnsons dynamic stall model74 are good examples. In
attempts to extend the generality of dynamic stall models, say to more general airfoil shapes or to wider ranges
of conditions, the complexity of the model must necessarily be increased and parameters added. The accuracy of
both postdiction and prediction generally increase quickly
as parameters are added to the simplest models. Postdiction
does not distinguish between the causal features present
in the data and the omnipresent experimental uncertainties and noise, so postdictive accuracy always increases
as the model becomes more complex. Parsimonious models, however, filter noise and so accuracy increases with
increasing modeling complexity to a point, and then decreases again, exhibiting an Ockhams Hill, and leveling
out as the model becomes more complex.106 These effects
are summarized in Fig. 25. The challenge in developing
the best engineering models for dynamic stall is obviously
to emphasize both accuracy and parsimony.

Concluding Remarks
Many strides have been made in the understanding and
modeling of wind turbine aerodynamics. Like all knowledge, however, our understanding of aerodynamics is not
absolute and can be viewed as tentative, approximate, and
always subject to revision. The recent NREL experiments
using an instrumented research wind turbine have provided
a definitive set of airloads and performance measurements
over a wide range of operating conditions, free of the uncertainties caused by various atmospheric effects. Initial

24 OF 28
A MERICAN I NSTITUTE OF A ERONAUTICS AND A STRONAUTICS PAPER 2002-0037

modeling efforts by the international community, using the


NREL measurements as a reference, have been found to
be relatively poor, with considerable deficiencies noted between the predictions for blade loads and power output
from the turbine even for unyawed, unstalled operating
conditions. Understanding deficiencies with the models
obviously provides the analysts with many challenges for
the future, and perhaps a new opportunity to better understand the physics of wind turbine aerodynamics.
This paper has emphasized two key areas that need continued serious consideration in wind energy modeling: the
modeling of the rotor wake and the modeling of unsteady
aerodynamics of the blade sections. The modeling of the
rotor wake using dynamic inflow and vortex methods has
been discussed. Despite their limitations, dynamic inflow
models have attractive mathematical forms and good computational efficiency that will always be appealing for certain types of rotor analyses. However, it is in the area of
vortex wake modeling and the incorporation of these models into wind turbine analyses that many future challenges
lie for the wind turbine analyst. Vortex wake methods
are attractive because of their appealing physical nature
and flexibility to handle a broad range of steady and transient operating conditions. While prescribed vortex wake
models have seen some use in wind turbine applications,
they have limited scope and today the rapid increases in
computer power mean that they are fast being surpassed by
free vortex schemes.
This paper is not meant to be a tutorial on unsteady aerodynamics. Rather, it serves as a reminder to the analyst that
many of the basic tools to model unsteady aerodynamic
effects on airfoils have already been laid down. It has
been emphasized that significant unsteady effects may be
produced on lifting surfaces even the absence of dynamic
stall. The most parsimonious and fundamental approach
to understanding the problem and predicting the effects is
using two-dimensional, unsteady thin-airfoil theories. This
gives a good level of analysis of the problem, and it has
been shown that excellent quantitative predictions can be
achieved for the unsteady airloads on two-dimensional airfoils if the correct models are used and both the circulatory
and noncirculatory components are included. It is up to
the analyst to use these theoretical tools, to explore their
capabilities, and to determine the limitations in their use
through validation studies. Only when this is done, will
the analysts be in a position to decide what aspects of the
airfoil modeling require further development.
Dynamic stall remains a serious concern for wind turbines, even for pitch controlled machines. The present
understanding of dynamic stall is based mostly on twodimensional airfoil experiments. It is clear that the basic features of dynamic stall are a result of a few key
causal factors, and it has been shown that the effects of the
phenomenon can be modeled relatively successfully using
parsimonious models. However, it is known from experiments that when dynamic stall does occur on a turbine
blade it is far from being two-dimensional, with adjacent

parts of the blade being stalled to a lesser or greater degree. Three-dimensional effects on the problem of dynamic
stall are still poorly understood, and a better understanding of these effects is where future challenges lie for the
analyst. Some of the potential mechanisms contributing to
the three-dimensional aspect of the problem have been proposed, although the exact mechanisms at play still require
a much deeper understanding of the flow. To model these
more general three-dimensional effects, the complexity of
the models must necessarily be increased and parameters
added. One must be cautious though to emphasize modeling accuracy while still retaining the parsimony necessary
for use in wind turbine design analyses.

Acknowledgments
My thanks to Dr. Scott Schreck for inviting me to write
and present this paper, and to NREL for their technical and
financial support.

References
1 Hansen.

A. C., and Butterfield, C. P., Aerodynamics of


Horizontal-Axis Wind Turbines, Annual Review of Fluid Mechanics, Vol. 25, pp. 115149.
2 Fingersh, L. J., Simms, D., Hand, M., Jager, D., Cotrell,
J., Robinson, M., Schreck, S., and Larwood, S., Wind Tunnel
Testing of NRELs Unsteady Aerodynamics Experiment, AIAA
Paper 2001-0035, 39th Aerospace Sciences Meeting, Reno, NV,
Jan. 811, 2001.
3 Simms, D., Schreck, S., Hand, M., and Fingersh, L., NREL
Unsteady Aerodynamics Experiment in the NASA-Ames Wind
Tunnel: A Comparison of Predictions to Measurements, National Renewable Energy Laboratory, NREL/TP-500-29494, June
2001.
4 Tangler, J. L., The Nebulous Art of Using Wind Tunnel Airfoil Data and Blade-Element Momentum Theory for Predicting
Rotor Performance, Paper 2002-0040, 21st ASME Wind Energy
Symposium and the 40th AIAA Aerospace Sciences Meeting,
Reno, NV, Jan. 1417, 2002.
5 Coton, F. N., Wang, T., and Galbraith, R. A. McD., An Examination of Key Aerodynamic Modeling Issues Raised by the
NREL Blind Comparison, Paper 2002-0038, 21st ASME Wind
Energy Symposium and the the 40th AIAA Aerospace Sciences
Meeting, Reno, NV, Jan. 1417, 2002.
6 Huyer, S. A., Simms, D. A., and Robinson, M. C., Unsteady
Aerodynamics Associated with a Horizontal-Axis Wind Turbine,
AIAA Journal, Vol. 34, No. 7, 1996, pp. 14101419.
7 Robinson, M. C., Galbraith, R. A. McD., Shipley, D., and
Miller, M., Unsteady Aerodynamics of Wind Turbines, Paper
95-0526, 33rd Aerospace Sciences Meeting, Reno, NV, Jan. 1995.
8 McCroskey, W. J., Some Current Research in Unsteady
Fluid Dynamics, ASME Freeman Scholar Lecture, Journal of
Fluids Engineering, Transactions of the ASME, 1975, pp. 838.
9 Leishman, J. G., Principles of Helicopter Aerodynamics,
Cambridge University Press, New York, 2000.
10 Durand, W. F. (ed), Aerodynamic Theory, Vol. IV, Divisions
JM, Springer, Berlin, 1935.
11 McCroskey, W. J., Vortex Wakes of Rotorcraft, Paper
95-0530, 33rd AIAA Aerospace Sciences Meeting and Exhibit,
Reno, NV, Jan. 912, 1995.
12 Landgrebe, A. J. New Directions in Rotorcraft Computational Aerodynamics Research in the U.S., AGARD CP552,
1994.

25 OF 28
A MERICAN I NSTITUTE OF A ERONAUTICS AND A STRONAUTICS PAPER 2002-0037

13 Duquet, E. P. N., van Dam, C. P., Brodeur, R. R., and Chao,


D. D., Navier-Stokes Analysis of Time-Dependent Flows About
a Wind Turbine, Proceedings of the 3rd ASME/JSME Joint Fluids Engineering Conference, July 1823, 1999.
14 Iida, M, Arakwawa, C., and Matsumiya, H., Three Dimensional Navier-Stokes Flow Field Computations Through a Horizontal Axis Wind Turbine Blade, Paper 2001-0058, 20th ASME
Wind Energy Symposium and the 39th AIAA Aerospace Sciences
Meeting, Reno, NV, Jan. 1811, 2001.
15 Carpenter, P. J., and Fridovich, B., Effect of A Rapid BladePitch Increase on the Thrust and Induced-Velocity Response of a
Full-Scale Helicopter Rotor, NACA TN 3044, Nov. 1953.
16 Crews, S. T., Hohenemeser, K. H., and Ormiston, R. A., An
Unsteady Wake Model for a Hingeless Rotor, Journal of Aircraft,
Vol. 10, No. 12, 1973, pp. 758759.
17 Ormiston, R. A., Application of Simplified Inflow Models
to Rotorcraft Dynamic Analysis, Journal of the American Helicopter Society, Vol. 21, No. 3, 1976, pp. 3437.
18 Peters, D. A., Hingeless Rotor Frequency Response with
Unsteady Aerodynamics, AHS/NASA Specialists Meeting on
Rotorcraft Dynamics, NASA SP-362, 1974.
19 Curtiss, H. C., Stability and Control Modeling, 12th European Rotorcraft Forum, Garmisch-Partenkirchen, Germany, Sept.
2225, 1986.
20 Pitt, D. M., and Peters, D. A., Rotor Dynamic Inflow
Derivatives and Time Constants from Various Inflow Models, 9th
European Rotorcraft Forum, Stresa, Italy, Sept. 1315, 1983.
21 Gaonkar, G. H., and Peters, D. A., Effectiveness of Current
Dynamic Inflow Models in Hover and Forward Flight, Journal of
the American Helicopter Society, Vol. 31, No. 2, 1986, pp. 4757.
22 Peters, D. A., Karunamoorthy, S., and Cae, W.-M., Finite
State Induced Flow Models Part I: Two Dimensional Thin Airfoil, Journal of Aircraft, Vol. 32, No. 2, 1995, pp. 313322.
23 Peters, D. A., and He, C. J., Finite State Induced Flow Models Part II: Three-Dimensional Rotor Disk, Journal of Aircraft,
Vol. 32, No. 2, 1995, pp. 323333.
24 Bierbooms, W., A Comparison Between Unsteady Aerodynamic Models, Proceedings of the European Wind Energy Conference, Amsterdam, The Netherlands, 1991.
25 Snel, H., and Schepers, J. G., Engineering Models for
Dynamic Inflow Phenomena, Proceedings of the European Wind
Energy Conference, Amsterdam, The Netherlands, 1991.
26 Johnson, W., Helicopter Theory, Princeton University Press,
1980.
27 Munk, M. M., Some Tables of the Factor of Apparent Additional Mass, NACA TN 197, July 1924.
28 Lewis, R. I., Vortex Element Methods for Fluid Dynamic
Analysis of Engineering Systems, Cambridge University Press,
New York, 1991.
29 Cottet, G-H., and Koumoutsakos, P. D., Vortex Methods:
Theory and Practice, Cambridge University Press, New York,
2000.
30 Kocurek, J. D., and Berkovitz, L. F., Velocity Coupling:
A New Concept for Hover and Axial Flow Wake Analysis and
Design, AGARD CP-334, May 1982.
31 Korcurek, D., Lifting Surface Performance Analysis for
Horizontal Wind Axis Turbines, SERI/STR-217-3163, 1987.
32 Coton, F. N., and Wang, T., The Prediction of Horizontal
Wind Turbine Performance in Yawed Flow Using an Unsteady
prescribed Wake Model, Proceedings Institute of Mechanical Engineers, Part A, Journal of Power and Energy, Vol. 213, 1999,
pp. 3343.
33 Clark, D. R., and Leiper, A. C., The Free Wake Analysis
A Method for Prediction of Helicopter Rotor Hovering Performance, Journal of the American Helicopter Society, Vol. 15,
No. 1, Jan. 1970, pp. 311.

34 Bagai, A., and Leishman, J. G., Rotor Free-Wake Modeling using a Psuedoimplicit Relaxation Algorithm, Journal of
Aircraft, Vol. 32, No. 6, Nov.-Dec. 1995, pp. 12761285.
35 Bagai, A., and Leishman, J. G., Rotor Free-Wake Modeling Using a Relaxation Technique Including Comparisons with
Experimental Data, Journal of the American Helicopter Society,
Vol. 40, No. 3, July 1995, pp. 2941.
36 Bhagwat, M., and Leishman, J. G., Stability, Consistency
and Convergence of Time Marching Free-Vortex Rotor Wake Algorithms, Journal of the American Helicopter Society, Vol. 46,
No. 1, Jan. 2001, pp. 5971.
37 Bhagwat, M., and Leishman, J. G., Accuracy of StraightLine Segmentation Applied to Curvilinear Vortex Filaments
Journal of the American Helicopter Society, Vol. 46, No. 2, April
2001, pp. 166169.
38 Bagai, A., and Leishman, J. G., Adaptive Grid Sequencing
and Interpolation Schemes for Rotor Free-Wake Analyses, AIAA
Journal, Vol. 36, No. 9, September 1998, pp. 15931602.
39 Bhagwat, M. J., and Leishman, J. G., Stability Analysis of
Rotor Wakes in Axial Flight, Journal of the American Helicopter
Society, Vol. 45, No. 3, 2000.
40 Leishman, J. G., and Bagai, A., Challenges in Understanding the Vortex Dynamics of Helicopter Rotor Wakes, AIAA
Journal, Vol. 36, No. 7, July 1998, pp. 11301140.
41 Miller, W. O., and Bliss, D. B., Direct Periodic Solutions
of Rotor Free Wake Calculations, Journal of the American Helicopter Society, Vol. 38, No. 2, April 1993, pp. 5360.
42 Ogawa, A., Vortex Flow, CRC Series on Fine Particle Science and Technology CRC Press Inc., 1993.
43 Bhagwat, M. J., and Leishman, J. G., Correlation of Helicopter Tip Vortex Measurements, AIAA Journal, Vol. 38, No. 2,
Feb. 2000, pp. 301308.
44 Rule, J. A., and Bliss, D. B., Prediction of Viscous Trailing
Vortex Structure from Basic Loading Parameters, AIAA Journal,
Vol. 36, No. 2, 1998, pp. 208218.
45 Theodorsen, T. 1935. General Theory of Aerodynamic Instability and the Mechanism of Flutter, NACA Report 496.
46 von Karman, Th., and Sears, W.R., Airfoil Theory for NonUniform Motion, Journal of the Aeronautical Sciences, Vol. 5,
No. 10, 1938, pp. 379390.
47 Sears, W., Operational Methods in the Theory of Airfoils in
Nonunform Motion, Journal of the Franklin Institute, Vol. 230,
1940, pp. 95111.
48 Bisplinghoff, R.L., Ashley H., and Halfman, R.L., Aeroelasticity, Addison-Wesley Publishing Co., Reading, MA, 1955.
49 Lomax, H., Heaslet, M.A., Fuller, F.B., and Sluder, L., Two
and Three Dimensional Unsteady Lift Problems in High Speed
Flight, NACA Report 1077, 1952.
50 Wagner, H., Uber

die Entstehung des dynamischen


Auftriebes von Tragflugeln, Zeitschrift fur Angewandte Mathematik und Mechanick, Vol. 5, No. 1, 1925.
51 Lomax, H., Indicial Aerodynamics, Chapter 6, AGARD
Manual on Aeroelasticity, 1968.
52 Jones, R. T., The Unsteady Lift of a Wing of Finite Aspect
Ratio, NACA Report 681, 1940.
53 Sears, W. R., and Sparks, B. O., On the Reaction of an Elastic Wing to Vertical Gusts, Journal of the Aeronautical Sciences,
Vol. 9, No. 2, 1941, pp. 6451.
54 Miles, J. W., The Aerodynamic Force on an Airfoil in a
Moving Gust, Journal of the Aeronautical Sciences, Vol. 23,
No. 11, 1956, pp. 1044-1050.
55 Drischler, J. A., and Diederich, F. W., Lift and Moment Responses to Penetration of Sharp-Edged Traveling Vertical gusts,
with Application to Penetration of Weak Blast Waves, NACA
TN 3956, 1957.

26 OF 28
A MERICAN I NSTITUTE OF A ERONAUTICS AND A STRONAUTICS PAPER 2002-0037

56 Leishman, J. G., Unsteady Aerodynamics of Airfoils Encountering Traveling Gusts and Vortices, Journal of Aircraft,
Vol. 34, No. 6, 1997, pp. 719-729.
57 Isaacs, R., Airfoil Theory for Flows of Variable Velocity, Journal of the Aeronautical Sciences, Vol. 12, No. 1, 1945,
pp. 113117.
58 Greenberg, J. M., Airfoil in Sinusoidal Motion in a Pulsating Stream, NACA TN 1326, 1947.
59 Van der Wall, B., and Leishman, J. G., The Influence of
Variable Flow Velocity on Unsteady Airfoil Behavior, Journal of
the American Helicopter Society, Vol. 39, No. 4, 1994, pp. 288
297.
60 Leishman, J. G., and Nguyen, K. Q., A State-Space Representation of Unsteady Aerodynamic Behavior, AIAA Journal,
Vol. 28, No. 5, 1990, pp. 836845.
61 Dinyavari, M. A. H., and Friedmann, P. P., Application of
Time-Domain Unsteady Aerodynamics to Rotary-Wing Aeroelasticity, AIAA Journal, Vol. 24, No. 9, 1986, pp. 14241432.
62 Beddoes, T.S., A Qualitative Discussion of Dynamic Stall,
AGARD Report 679, 1979.
63 McCroskey, W. J., The Phenomenon of Dynamic Stall,
NASA 81264, 1981.
64 Srinivasan, G. R., Ekaterinas, J. A., and McCroskey, W.
J., Dynamic Stall of an Oscillating Wing Part:1 Evaluation of
Turbulence Models, Paper 93-3403, AIAA 11th Applied Aerodynamics Conference, Monterey, CA, Aug. 911, 1993.
65 Pierce, K., and Hansen, A. C., Prediction of Wind Turbine
Rotor Loads Using the Beddoes-Leishman Model for Dynamic
Stall, Journal of Solar Energy Engineering. Vol. 117, 1995.
66 Carta, F. O., An Analysis of the Stall Flutter Instability of
Helicopter Rotor Blades, Journal of the American Helicopter Society, Vol. 12, No. 4, 1967, pp. 118.
67 Bielawa, R. L., Synthesized Unsteady Airfoil Data with Applications to Stall Flutter Calculations, 31st Annual Forum of the
American Helicopter Society, Washington DC, May 1315, 1975.
68 Gross, D. W., and Harris, F. D., Prediction of In-Flight
Stalled Airloads from Oscillating Airfoil Data, 25th Annual
Forum of the American Helicopter Society, Washington DC, May
1416, 1969.
69 Gormont, R. E., A Mathematical Model of Unsteady Aerodynamics and Radial Flow for Application to Helicopter Rotors,
USAAVLABS TR 72-67, 1973.
70 Beddoes, T. S., A Synthesis of Unsteady Aerodynamic Effects Including Stall Hysteresis, Vertica, Vol. 1, 1976, pp. 113
123.
71 Beddoes, T. S., Onset of Leading Edge Separation Effects
under Dynamic Conditions and Low Mach Number, 34th Annual
Forum of the American Helicopter Society, Washington DC, May
1517, 1978.
72 Galbraith, R. A. McD, Niven, A. J., and Seto, L. Y., On the
Duration of Low Speed Dynamic Stall, Paper ICAS-86-2.4.3.,
Proceedings of the ICAS, 1986.
73 Gangwani, S. T., Synthesized Airfoil Data Method for Prediction of Dynamic Stall and Unsteady Airloads, Vertica, Vol. 8,
No. 2, 1984, pp. 93-118.
74 Johnson, W., Comparison of Three Methods for Calculation
of Helicopter Rotor Blade Loading and Stresses Due to Stall,
NASA TN D-7833, 1974.
75 Tran, C. T., and Petot, D., Semi-Empirical Model for the
Dynamic Stall of Airfoils in View of the Application to the Calculation of the Responses of a Helicopter Blade in Forward Flight,
Vertica, Vol. 5, No. 1, 1981, pp. 3553.
76 Tran, C. T., and Falchero, D., Application of the ONERA Dynamic Stall Model to a Helicopter Rotor Blade in
Forward Flight, 7th European Rotorcraft Forum, GarmischPartenkirchen, Sept. 2225, 1981.

27

77 McAlister, K. W., Lambert, O., and Petot, D., Application


of the ONERA Model of Dynamic Stall, NASA Technical Paper
2399, AVSCOM Technical Report 84-A-3, 1984.
78 Peters, D. A., Toward a Unified Lift Model for Use in Rotor
Blade Stability Analysis, Journal of the American Helicopter Society, Vol. 30, No. 3, 1985, pp. 3242.
79 Reddy, T. S. R., and Kaza, K. R. V., A Comparative Study
of Some Dynamic Stall Models, NASA TM-88917, 1987.
80 Pitot, D., Differential Equation Modeling of Dynamic
Stall, La Reserche Aerospatiale, No. 1989-6,1989.
81 Truong, V. K.,A 2-D Dynamic Stall Model Based on a Hopf
Bifurcation, 19th European Rotorcraft Forum, Cernobbo, Italy,
Sept. 1416, 1993.
82 McCroskey, W. J., A Critical Assessment of Wind Tunnel
Results for the NACA 0012 Airfoil, NASA Technical Memorandum 100019, 1987.
83 Leishman, J. G., and Beddoes, T. S., A Generalized Method
for Unsteady Airfoil Behavior and Dynamic Stall Using the Indicial Method, 42nd Annual Forum of the American Helicopter
Society, Washington DC, June 1986.
84 Leishman, J. G., and Beddoes, T. S., A Semi-Empirical
Model for Dynamic Stall, Journal of the American Helicopter
Society, Vol. 34, No. 3, 1989, pp. 317.
85 Beddoes, T.S., Representation of Airfoil Behavior, Vertica,
Vol. 7, No. 2, 1983, pp. 183197.
86 Leishman, J. G., Modeling Sweep Effects on Dynamic
Stall, Journal of the American Helicopter Society, Vol. 34, No. 3,
1989, pp. 1829.
87 Tyler, J. C., and Leishman, J. G., An Analysis of Pitch
and Plunge Effects on Unsteady Airfoil Behavior, Journal of the
American Helicopter Society, Vol. 37, No. 3, 1992, pp. 6982.
88 Leishman, J. G., and Crouse, G. L., State-Space Model for
Unsteady Airfoil Behavior and Dynamic Stall, AIAA Paper 891219, 1989.
89 Thwaites, B., Incompressible Aerodynamics, Oxford University Press, Oxford, 1960.
90 Carta, F. O., A Comparison of the Pitching and Plunging
Response of an Oscillating Airfoil, NASA CR-3172, 1979.
91 Carta, F. O., Effect of Unsteady Pressure Gradient Reduction on Dynamic Stall Delay, Journal of Aircraft, Vol. 8, No. 10,
1971, pp. 839840.
92 Schreck, S. J., Robinson, M. C., Hand, M. M., and Simms,
D. A., Contrasting Blade Flow Field Archetypes on the Horizontal Axis Wind Turbine Operating Regimes, Journal of Solar
Energy Engineering, Vol. 123, Nov. 2001. (To appear.)
93 Madsen, H., and Christensen, H., On the Relative Importance of Rotational, Unsteady & Three-Dimensional Effects on
the HAWT Rotor Aerodynamics, Wind Engineering, Vol. 14,
No. 6, 1990, pp. 405415.
94 Snel, H., Scaling Laws for the Boundary Layer Flow on
Rotating Wind Turbine Blades, 4th IEA Symposium on Aerodynamics for Wind Turbines, Rome 1991.
95 Narramore, J. C., and Vermeland, R., Navier-Stokes Calculations of Inboard Stall Delay Due to Rotation, Journal of
Aircraft, Vol. 29, No. 1, Jan. 1992, pp. 7378.
96 Dwyer, H. A., and McCroskey, W. J., Crossflow and Unsteady Boundary-Layer Effects on Rotating Blades, AIAA Journal, Vol. 9, No. 8, 1971, pp. 14981505.
97 Corrigan, J. Empirical Model for Stall Delay Due to Rotation, American Helicopter Society Aeromechaniscs Specialist
Meeting, San Francisco, CA, Jan. 1921, 1994.
98 Du, Z. H., and Selig, M. S., A 3-D Stall Delay Model for
HAWT Performance Prediction, Paper 89-0021, 1998.
99 Jones, R. T. and Cohen, D., Aerodynamics of Wings at High
Speeds, Vol. VII of High Speed Aerodynamics and Jet Propulsion, Section A, Chapter 1, pp. 3648. Aerodynamic Components

OF

28

A MERICAN I NSTITUTE OF A ERONAUTICS AND A STRONAUTICS PAPER 2002-0037

of Aircraft at High Speeds. Donovan & Lawrence (eds.), Princeton University Press, NJ, 1957.
100 Purser, P. E., and Spearman, M. L., Wind Tunnel Tests at
Low Speeds of Swept and Yawed Wings Having Various Planforms, NACA TN 2445, 1951.
101 Harris, F. D., Preliminary Study of Radial Flow Effects
on Rotor Blades, Journal of the American Helicopter Society,
Vol. 11, No. 3, 1966, pp. 121.
102 St. Hillaire, A. O., Carta, F. O., Fink, M. R., and Jepson, W.
D., The Influence of Sweep on the Aerodynamic Loading of a
NACA 0012 Airfoil, Vol. 1, NASA CR 3092, 1979.
103 St. Hillaire, A. O., Carta, F. O., Analysis of Unswept and
Swept Wing Chordwise Pressure Data from an Oscillating NACA
0012 Airfoil Experiment, Vol. 1, NASA CR 3567, 1983.
104 Tan, C. M., and Carr, L. W., The AFDD Int. Dynamic
Stall Workshop on Correlation of Dynamic Stall Models with 3-D
Dynamic Stall Data, NASA TM-110375, USAATCOM TR-96A-009, 1996.
105 Green, R. B., and Galbraith, R. A. M., Dynamic Recovery
to Fully Attached Aerofoil Flow from Deep Stall, AIAA J., Vol.
33, No. 8, pp. 1433-1440, 1995.
106 Gauch, H. G., Jr., Prediction, Parsimony and Noise, American Scientist, Vol. 81, 1993, pp. 468478.

28 OF 28
A MERICAN I NSTITUTE OF A ERONAUTICS AND A STRONAUTICS PAPER 2002-0037

You might also like