Energy 89 (2015) 1001e1009

Contents lists available at ScienceDirect

Energy
journal homepage: www.elsevier.com/locate/energy

Aeroelastic coupling analysis of the flexible blade of a wind turbine

Wenwei Mo, Deyuan Li*, Xianneng Wang, Cantang Zhong
School of Electromechanical Engineering, Guangdong University of Technology, Guangzhou 510006, China

a r t i c l e i n f o a b s t r a c t

Article history: This paper presents an aeroelastic coupling analysis of the flexible blade of a large scale HAWT (hori-
Received 31 January 2015 zontal axis wind turbine). To model the flexibility of the blade more accurately, ‘SE’ (super-element) is
Received in revised form introduced to the blade dynamics model. The flexible blade is discretized into a MBS (multi-body system)
22 April 2015
using a limited number of SEs. The blade bending vibration and torsional deflection are both considered
Accepted 15 June 2015
when calculating the aerodynamic loads; thus, the BEM (blade element momentum) theory used in this
Available online 21 July 2015
study is modified. In addition, the BeL (BeddoeseLeishman) dynamic stall model is integrated into the
BEM-modified model to investigate the airfoil dynamic stall characteristics. The nonlinear governing
Keywords:
Wind turbine
equations of the constrained blade MBS are derived based on the theory of MBS dynamics coupling with
Flexible blade the blade aerodynamics model. The time domain aeroelastic responses of the United States NREL (Na-
Super-element tional Renewable Energy Laboratory) offshore 5-MW wind turbine blade are obtained. The simulation
Multi-body system results indicate that blade vibration and deformation have significant effects on the aerodynamic loads,
Dynamic stall and the dynamic stall can cause more violent fluctuation for the blade aerodynamic loads compared with
Aeroelastic coupling the steady aerodynamic model, which can considerably affect the blade fatigue load spectrum analysis
and the fatigue life design.
© 2015 Elsevier Ltd. All rights reserved.

1. Introduction based on the unsteady aerodynamic model of a flexible blade is

Under the excitations of mechanical loads, such as aerodynamic
loads and gravity, random vibration of wind turbine blades would
occur. With the increasing unit capacity of wind turbines, blades
are becoming longer (e.g., the NREL (National Renewable Energy
Laboratory) 5-MW wind turbine blade has very large aspect ratio)
[1]. Furthermore, blades are usually made of composite materials
with low Young's modulus and stiffness; therefore, blade magni-
tude of deformation is beyond the scope of assumption for small
blade deflection [2]. In this case, the blade operates in unsteady
conditions because of the blade vibration. The unsteady conditions
can cause the following effects: the feedback of blade vibration and
the relatively large deflection on the aerodynamic loads cannot be
ignored, and the feedback needs to be quantifiably analyzed. The
blades generally work in a high angle of attack or even in severe
stall conditions when the vibration and deformation increase. The
unsteady characteristics of the blade airfoil dynamic stall need to be
simulated accurately, which is the basis of the blade dynamic
aerodynamic load calculation [3]. Aeroelastic coupling analysis
* Corresponding author.
E-mail address: lidey@gdut.edu.cn (D. Li).
essential for the blade aerodynamics and structural design as well
as the fatigue life design.
When operating in a random wind condition, the angle of attack
of a blade airfoil is time varying. If the blade torsional deformation
and vibration increase, a hysteresis phenomenon and dynamic stall
will occur [4]. In this case the aerodynamic coefficients will
considerably deviate from their static values. For example, the error
between the predicted data and the measured data is approxi-
mately 15%e20% for a NREL Combined Experiment turbine [5],
which means that the energy output reduces by up to 20%. The
hysteresis of aerodynamic loads will in turn cause more violent
vibration, resulting in the decline of blade fatigue life. Research
conducted by Shipley [6] showed that a dynamic stall could cause
more violent fluctuation for aerodynamic loads. Due to the
complication of airflow flow, there have not been any completely
accurate theoretical models to simulate the airfoil unsteady aero-
dynamic response. The semi-empirical models based on the
experimental summary are often used [7e9], including the
incompressible Theodorsen thin wing model [10], the Onera Edlin
model and the BeL (BeddoeseLeishman) model, etc. [11,12].
Although the development of CFD (computational fluid dynamics)
makes it possible to calculate the airfoil aerodynamic response
under operating conditions, its practical application in the blade

http://dx.doi.org/10.1016/j.energy.2015.06.046
0360-5442/© 2015 Elsevier Ltd. All rights reserved.
1002 W. Mo et al. / Energy 89 (2015) 1001e1009
aeroelastic coupling analysis of wind turbines does not seem
possible in the near future because of its extensive cost of calcu-
lation and complicated solving process [13]. At present, applying
semi-empirical models to calculate the unsteady aerodynamic
loads are still the dominant method, among which the BeL model
receives much attention and recognition [14]. Liu et al. [3,4,7] and
Dai et al. [13] adopted the BeL model to analyze the dynamic
aerodynamic loads of blades and investigated the unsteady
characteristics.
Wind turbine blades have complicated time-varying MBS
(multi-body system) coupling nonlinear airflow flow and blade
elastic deformation. In addition, the interactive effects between
blade vibration and aerodynamic loads involve the simultaneous
solution for the blade dynamics equations, which are a set of time-
varying DAEs (differential and algebraic equations). Hence the
blade mechanics model needs to be analyzed with fewer DOFs
(degrees of freedom). To discretize the components, which undergo
spatial motion and elastic deformation, Molenaar [15], Zhao [16]
and Holierhoek [17] introduced a ‘super-element;’ thus, it is suit-
able for the mechanical modeling of MBS with flexible components
[18]. Li et al. [19] discretized flexible components such as wind
turbine blades and towers into a series of rigid bodies connected
with joints, springs and dampers by applying the super-element
mentioned above. They then established the aeroelastic coupling
equations of the rigid-flexible MBS of wind turbines using the
theory of MBS dynamics, the Roberson-Wittenburg modeling
methodology, and the BEM (blade element momentum) theory.
Eventually, the time domain aeroelastic coupling response was
obtained for a wind turbine via numerical simulation.
Taking account the aeroelastic coupling, the present studies
introduce the airfoil vibrational velocity and the torsional deflec-
tion of a flexible blade in the calculation of the inflow angle and the
angle of attack. Thus, the BEM model is correspondingly modified.
To investigate the unsteady aerodynamic characteristics, the BEM-
modified model incorporates with the BeL dynamic stall model.
Based on the blade MBS model and the aerodynamics model, an
analysis on the time domain aeroelastic coupling responses of the
blade is completed. The effects of blade torsional deflection and
bending vibration on the aerodynamic loads are quantifiably
analyzed. The comparison and analysis on the simulation results
clearly indicate the significance and necessity of considering the
blade flexibility and the unsteady aerodynamic characteristics in
the blade aerodynamic load calculation during the blade design
stage.
2. BEM modified model and BeL model
Wind speed changes with time and altitude during blade
operation. The airstream flows along the spanwise and chordwise
directions and flow separation and vortex shedding may occur. If
the blade motion combines with the blade bending vibrations (i.e.,
in-plane vibration and out-of-plane vibration) and the torsional
vibration, a blade unsteady aerodynamic effect will occur. The un-
steady aerodynamic loads can be numerically calculated via the
BEM modified model integrating the BeL dynamic stall model.
2.1. BEM modified theory (used in steady analysis)
Usually, blade profiles consist of a series of airfoils with different
aerodynamic characteristics along the span. Conventionally, the
BEM modified theory is applied to compute the steady aero-
dynamic loads. This theory combines the blade element theory and
the momentum theory. It is widely used in aerodynamic analysis of
wind turbines due to its relatively simple calculation process and
reliable calculation result.
The blade is discretized into a series of rigid bodies along the
span, and it is assumed that the profile of each rigid body consists of
the same airfoil. According to the blade element theory, when the
spanwise length of the blade element is small enough, the spanwise
flow of the airstream can be neglected, and aerodynamic loads (lift
and drag) are uniformly distributed along the span; they act on the
quarter chord point or the AC (aerodynamic center). When the
angle of attack is obtained, the airfoil steady lift CL(a), drag CD(a)
and moment coefficients CM(a) can be computed by interpolating
the steady aerodynamic data. Then, the lift Li, drag Di and moment
Mi loads per unit length on each blade rigid body can be calculated
according to the blade element theory. Thus, the lift FL, drag FD and
moment Maero exerted on the AC of the rigid body Bi can be ob-
tained as
FL ¼ Li  li ; F D ¼ D i  li ; Maero ¼ Mi  li (1)
where li is the spanwise length of rigid body Bi.
When the infinitesimal span dr, an infinitesimal span of the
spanwise length of a rigid body, rotates in a circle about the rotor
axis, the thrust dT and the torque dQ on this annular element are
obtained by the equations below:
dT ¼ 0:5BrcW 2 ðCL ðaÞcos f þ CD ðaÞsin fÞdr (2)
dQ ¼ 0:5BrcW 2 ðCL ðaÞsin f  CD ðaÞcos fÞrdr (3)
where B is the number of blades, c is the sectional chord length, r is
the density, W is the sectional relative velocity of airflow, r is the
radial distance from the rotational axis to the infinitesimal span dr,
and f is the inflow angle.
The momentum theory further introduces the 3-D airstream
flow, the tangential induction factor a0 and the axial induction
factor a. Furthermore, the tip loss and hub loss models by Prandtl
are introduced to correct the aerodynamic load calculation when
considering the airstream vortex characteristics on the blade tip
and hub. Thus, the thrust dT and the torque dQ on this annular
element are given as:
2
dT ¼ 4prrU∞ ð1  aÞaFdr (4)
0
dQ ¼ 4pr 3 rU∞ Uð1  aÞa Fdr (5)
where F denotes the combined factor which includes the blade tip
loss and hub loss, whose value can be found in Ref. [20]. The
aerodynamic parameters involved in this BEM theory are shown in
Fig. 1.
In this article, the effects of blade torsional deformation and
bending vibration on the aerodynamic forces embody in the effects
of out-of-plane and in-plane velocities on the inflow angle f and
the angle of attack a, yielding:
Fig. 1. Aerodynamic parameters of airfoil cross section and body-fixed coordinate
system of the aerodynamic center of rigid body Bi.
 W. Mo et al. / Energy 89 (2015) 1001e1009 1003

U∞ ð1  aÞ þ Ve­op fCa ðtÞ ¼ A0  A1 expð  b1 sÞ  A2 expð  b2 sÞ (12)

tan f ¼ 0 (6)
Urð1 þ a Þ þ Ve­ip
where constantsA0, A1, A2, b1 and b2 can be obtained by the
a ¼ f  qðtÞ (7) experimental data fitting with values of 1.0, 0.3, 0.7, 0.14 and 0.53,
respectively [4]. s is the non-dimensional time constant.
where Ve-op and Ve-ip are the out-of-plane and in-plane velocities,

respectively, and q(t) is the torsional angle of the blade airfoil. In the s ¼ 2 1  M 2 Wt (13)
conventional steady analysis of aerodynamic loads, q(t) is the local
blade pitch, which is the combination of the pitch angle and the blade where M is the Mach number; W is the combined speed; t is the
twist. If the value of the pitch angle remains zero, q is usually constant. airfoil relative thickness.
However, in this paper, q(t) is time-varying because it also includes the The non-circulatory normal force coefficient CNI has the impul-
blade torsional deformation caused by the time-varying aerodynamic sive characteristic as
torsional moment, which is neglected in the conventional steady
analysis. Eqs. (6) and (7) reflect the essence of aeroelastic coupling. 4 I
By combining Eqs. (1) through (7) and the lift and drag aero-
CNI ðtÞ ¼ f ðtÞDa (14)
M a
dynamic data of the 2-D airfoil, the converged value of the angle of
attack at each time step can be obtained through iteration. The where fIa ðtÞ is the non-circulatory indicial function,
detailed iteration process was introduced in Ref. [20]. The aero- fIa ðtÞ ¼ expðt=TJ Þ, and TJ ¼ c/vc. c is the sectional chord length, and
dynamic loads can be computed by integrating the aerodynamic vc is the local sound velocity.
loads of the annular element along the entire span. The normal force coefficient CN can then be obtained by adding
the circulatory normal force coefficient CNC and the non-circulatory
2.2. BeddoeseLeishman dynamic stall model (used in unsteady normal force coefficient CNI .
analysis)
CN ðtÞ ¼ CNC ðtÞ þ CNI ðtÞ (15)
The original BeL model includes the 2-D inviscid wake, the The tangential force coefficient CT is determined by the circu-
leading edge separation, the trailing edge separation and the latory term,
compressibility effects [11]; therefore, it is widely used in helicopter
aerodynamics. The Danish Risø National Laboratory developed a CT ðtÞ ¼ CNa ðtÞa2E (16)
more simplified and practical version for the original BeL model
that neglected the compressibility effects and flow separation
initiating from the leading edge, according to the operating condi-
2.2.2. Dynamic decay in the movement of the separation point of
tion of wind turbine airfoils [13]. The body-fixed coordinate system
unsteady separated flow
XYZ on the airfoil AC is used in the BeL model, so the aerodynamic
The flow separation on the airfoil surface can lead to circulation
force components are the normal force coefficient CN and the
loss and consequently to the reduction of the aerodynamic co-
tangential force coefficient CT. Hence, the steady aerodynamic data
efficients (compared with the attached flow). The position of the
CL and CD need to be transformed into the AC coordinate system.
separation point on the airfoil surface can be represented by a
variable f. f ¼ x/c, where x is the point of flow separation measured
CNst ¼ CL cos a þ CD sin a (8)
from the leading edge, and c is the sectional chord length. Ac-
cording to the Kirchhoff theory, the nonlinear functional relation-
CTst ¼ CL sin a  CD cos a (9) ship between the aerodynamic force coefficients and flow
separation point is defined as:
2.2.1. Unsteady attached flow state pffiffiffi!2
For the attached flow, the wake separation effect and the 1þ f
CN ðtÞ ¼ CNa ða  a0 Þ (17)
airstream acceleration effect on the normal force coefficient are 2
mainly considered. These effects are represented by the circulatory
normal force coefficient CNC and the non-circulatory normal force pffiffiffi
coefficient CNI , respectively. CT ðtÞ ¼ CNa ða  a0 ÞtgðaÞ f (18)
The lay characteristic of the circulatory normal force coefficient
CNC due to the change in the angle of attack is represented as: where CNa is the normal force coefficient curve slope of the steady
aerodynamic data, and a is the angle of attack.
CNC ðtÞ ¼ CNa ðaE  a0 Þ (10) For the steady analysis, the values of normal and tangential
steady separation point positions corresponding to a series angle of
where aE is the effective angle of attack, a0 is the zero-lift angle of attacks can be calculated by solving the inversion equations of Eqs.
attack, CNa is the normal force coefficient curve slope of the steady (17) and (18).
aerodynamic data, which is a function of both the angle of attack, a, In the BeL model, there exists the unsteady boundary layer
and the Reynolds number, Re. CNa can be obtained by fitting the response. Its effect on the static effective separation point fs can be
steady airfoil characteristic curve. The effective angle of attack aE represented by a first order lag equation:
can be calculated using the following equation:
df fs  f
¼ (19)
aEn ¼ an1 þ fCa ðtÞDa (11) ds Tf

where aEn is the effective angle of attack at time step n, an1 is the where f is the dynamic separation point position. Tf is the semi-
angle of attack at the time step n1, Da ¼ anan1, and fCa ðtÞ is the empirical time constant in s space, and Tf ¼ 3.0 in the present pa-
circulatory indicial function as: per [4].
1004 W. Mo et al. / Energy 89 (2015) 1001e1009

To obtain the dynamic f, the effective angle of attack needs to be
computed first. Hence, the offset of angle of attack due to the lag in
the leading edge pressure response is taken into account. A first
order lag equation is used to represent the lag effect on the normal
force coefficient of the attached flow.
dCN0 ðtÞ CN ðtÞ  C 0 N ðtÞ
¼ (20)
ds Tp
where Tp is an empirical time constant set to 1.7 [4].
Then, the effective angle of attack af can be obtained by the
equation below:
CN0 ðtÞ
af ¼ þ a0 (21)
CNa
Once af is obtained from Eq. (21), the value of the static effective
separation point fs can be calculated by interpolating the steady
separation points. Then, the dynamic separation point f can be
obtained from Eq. (19). The normal and tangential force coefficients
CNf and CTf including the unsteady separation effect can be obtained
by solving Eqs. (17) and (18), respectively, when the steady real-
time angle of attack a is substituted by the effective angle of
attack aE.
2.2.3. Vortex effect of the unsteady separated flow
When flow separation occurs, the effects of vortex build-up and
shedding should be considered. The magnitude of the vortex
normal force coefficient cv(t) is defined as the difference between
the unsteady circulatory normal force coefficient CNC and the un-
steady normal force coefficient CNf .
2.2.4. Numerical calculation process of the BeL model
At the initial moment of calculation, CL and CD use the static
values. The computed results of the previous loop are the initial
values for the next loop. The converged, unsteady values CL and CD
can be obtained after a period of iteration. Fig. 2 displays the flow
chart of the numerical calculation of the BeddoeseLeishman
model [4].
3. Multi-body dynamics model of a flexible blade
Large-scale wind turbine blades have a long, thin and twisted
profile with a sectional shape and size that varies along the span.
The blade is discretized into a MBS using the ‘super-element,’ and
the adjacent rigid bodies of the system are connected with cardan
or revolute joints with springs and dampers [15e18]. Thus, the
lateral and torsional deformation can be described by a limited
number of DOFs. In this article, the multi-body dynamics model of a
blade is established by applying the Roberson-Wittenburg stylized
modeling methodology of multi-body dynamics.
3.1. Discretization of a flexible blade
For NREL 5-MW HAWT (horizontal axis wind turbine) blade, 4
super-elements are used to build the blade topological configura-
tion, as shown in Fig. 3. Since the last rigid body in a super-element
is rigidly connected to the first rigid body of the next adjacent
super-element, these two rigid bodies can be merged into one rigid
body (the merged bodies are shown in Fig. 3 as B4, B7 and B10).
Consequently, a blade is divided into 13 rigid bodies with 21 total
DOFs. The expressions of spring stiffness coefficients including the
bending and torsional springs are introduced in Ref. [16], while the
values of these coefficients can be found in Ref. [19].

cv ðtÞ ¼ CNC ðtÞ  CNf ðtÞ (22)

3.2. Dynamics equations of MBS of a constrained blade
Furthermore, the total accumulated vortex normal force coeffi-
cient is allowed to exponentially decay with time, but it may also be Using the rotational DOFs of each joint in the blade as gener-
updated with new vortex lift feed. Hence, the variation character- alized coordinates, the vector of generalized coordinates describing
istic of CNv can be represented by the following equation: the topological configuration of the blade is obtained as:
qðtÞ ¼ ðq1 ; …; q21 ÞT (29)
dcv ðtÞ
dCNv ðtÞ ds
 CNv ðtÞ
¼ (23)
ds Tv

where Tv is an empirical time constant set to 6.0 in this study [4].

The vortex effect on the tangential force coefficient is given by

CTv ðtÞ ¼ CNv ðtÞaf ð1  tv Þ (24)

where tv is the non-dimensional time parameter that is estimated

from experimental data and experience.
From the above, the total normal force coefficient and the
tangential force coefficient can be written as:

CNd ðtÞ ¼ CN ðtÞ þ CNv ðtÞ þ CNI ðtÞ

f
(25)

CTd ðtÞ ¼ CT ðtÞ þ CTv ðtÞ

f
(26)

The unsteady lift and drag coefficients CLd and CDd are then
calculated from resolving CNd and CTd into components perpendic-
ular and parallel to the direction of the sectional relative velocity of
airflow, and adding the minimum steady drag coefficient CD0.

CLd ðtÞ ¼ CNd ðtÞcos a þ CTd ðtÞsin a (27)

CDd ðtÞ ¼ CNd ðtÞsin a  CTd ðtÞcos a þ CD0 (28) Fig. 2. Flow chart of the numerical calculation of the BeddoeseLeishman model.
 W. Mo et al. / Energy 89 (2015) 1001e1009 1005

According to the virtual work principle, the dynamics differen-

tial equations of the blade MBS are derived as:

€z¼0
Zq (30)
Ref. [21] introduced the detailed expression and derivation
method for the generalized mass matrix Z and the generalized force
vector z. Hence, they are not introduced in this paper.
During operation, the blade rotational speed is subject to a
driving constraint, so the constraint equations are introduced into
the dynamics equations. The number of independent constraint
equations of the system s is:

F ¼ ðF1 ; …; Fs ÞT ¼ 0 (31)
The number of independent variables of the system-generalized
coordinates is d, d ¼ 21s. Coupling Eqs. (30) and (31), the
dynamical equations of the blade MBS with Lagrange multipliers
can be given as:

€ ¼ z  FTq l
Zq (32)
Fðq; tÞ ¼ 0

where Fq is the system constraint Jacobian matrix, l is the Lagrange Fig. 4. Flow chart of the aeroelastic coupling analysis.

multiplier vector, and FTq l represents the contributions from all

ideal constraint forces between bodies. The dynamical equations
velocity at the next time step. The obtained velocities are then
are closed DAEs (differential-algebraic equations). The BSM
applied to find the aerodynamic forces at the next time step. With
(Baumgarte stabilization method) [21] is used to solve the DAEs. All
this method, the aerodynamic forces and the structural response at
of the generalized coordinates and Lagrange multipliers are treated
each time step can be determined, and the code will run until the
as unknown variables; then, the DAEs can be written as a closed set
termination time tend.
of equations with more variables than Eq. (30).
In this study, it is assumed that the blade rotates at the rated
     angular speed, so the constraint equation can be written as
Z FTq €
q z
¼ _ b F
2 (33)
Fq 0 l x  2aF
F ¼ q1  ue t ¼ 0 (34)
where x ¼ ðFq qÞ_ q q_  2Fqt q_  Ftt and F and F_ are the default of
where the generalized coordinate q1 is the rotation of the blade
the coordinate and velocity, respectively. The correction terms, 2aF _
with respect to the axis of rotor and ue is the rated angular speed
and b2F, will take effect automatically when the default occurs. The (12.1 rpm in this work). For variable speed wind turbines, the blade
empirical values of a or b are between 5 and 50. When a ¼ b, the rotational speed varies with the wind speed, so Eq. (34) will be
stabilization response time is much shorter [21]. The process for changed into a nonlinear constraint equation. In this study,
numerically solving Eq. (33) is as follows: q € and l are obtained by
a ¼ b ¼ 10, while the integral accuracy is 104.
solving Eq. (33) using the method for solving algebraic equations;
then, q_ and q can be numerically obtained by integrating q €.
5. Numerical results and analysis of aeroelastic coupling

4. Aeroelastic code development
The blade aeroelastic coupling equation must be solved by nu-
merical integration due to its high nonlinear property. Fig. 4 dis-
plays the coupling process between the blade MBS model and the
aerodynamics model. During the simulation, unsteady aero-
dynamic forces are calculated at each time step. These forces are
treated as external forces and applied to the MBS model to compute
the values of the generalized coordinate and the generalized
Fig. 3. Regular mark numbers of each rigid body, the inertial coordinate system XYZ
and the blade coordinate system X0 Y0 Z0 .
5.1. Feedback of torsional deformation and vibration of flexible
blades on the aerodynamic loads
For the conventional analysis of blade aerodynamic loads, the
feedback of torsional deformation and the vibration of flexible
blades on the aerodynamic loads are generally neglected. However,
during the operation of flexible blades of large-scale wind turbines,
the random aerodynamic and mechanical loads will cause consid-
erably large torsional deformation and vibration, which will couple
with the blade aerodynamic loads. Thus, the aeroelastic-coupling
phenomenon of flexible blades will be more obvious. Two
different blade models are built in the present article: the first blade
model (the 1st model) includes the aeroelastic coupling (i.e., taking
into account the feedback), whereas the second blade model (the
2nd model) neglects the aeroelastic coupling. The blade of a NREL
5-MW wind turbine is chosen as the research subject. The aero-
dynamic load differences under rated operating conditions be-
tween these two models are quantifiably analyzed to investigate
the feedback degree of blade vibration and the deformation on the
aerodynamic loads. The post-process module of the simulation
code can analyze the internal forces and the deformation of
dangerous cross sections. The blade structural and aerodynamic
1006 W. Mo et al. / Energy 89 (2015) 1001e1009

properties can be found in Ref. [1]. It is assumed that the mean wind
speed at the height of the hub is the rated wind speed (11.4 m/s),
and the impulsive variation of random wind is also taken into
account.
The feedback of the blade torsional deformation and the bending
vibration on the aerodynamic loads are analyzed in terms of the
blade root bending moments, the displacements of the blade tip and
the variation of the angle of attack of each rigid body within the
blade MBS. For the second blade model, the simulation code intro-
duced in Section 2 in this paper must be partly modified as follows:
(a) All values of the spring stiffness coefficients of the revolute
joint in each super-element must be set to extremely large values in
the simulation code (i.e., neglecting the torsional deformation). In
this case, the expression of the angle of attack, Eq. (7), is written as:

ai ðtÞ ¼ fðtÞ  qi0 (35)

where ai(t) is the real-time angle of attack of rigid body Bi, and qi0 is
the twist of the airfoil section within the rigid body.
(b) The blade bending vibrational velocities are not considered
when computing the inflow angle fðtÞ of the second blade model.
Eq. (6) is written as:

U∞ ð1  aÞ
tan f ¼ 0 (36)
Urð1 þ a Þ
The dynamic simulations of the random aeroelastic responses of
the two blade models are carried out based on the BEM modified
theory. Fig. 5a and b shows the time domain responses of in-plane
and out-of-plane deflections of the blade tip, respectively. From
these two figures it is clear that the bending deflections and vi-
bration amplitudes of the blade model, neglecting the aeroelastic
coupling, are considerately larger than those of the other blade
model.
Fig. 6 shows the curves of the time-varying angle of attacks of a
cross section of these two blade models. The mean of the calcula-
tion result of the blade model neglecting the aeroelastic coupling
Fig. 5. The time domain responses of the blade tip: (a) in-plane deflection; (b) out-of-
increases by 0.53 compared with the first blade model.
plane deflection.
Fig. 7a and b displays the time domain responses of the edge-
wise and flapwise blade root bending moments, respectively.
Table 1 shows the analysis results on the aeroelastic coupling effect. coefficients will deviate from their static values. Therefore, it is
From Table 1, the blade root flapwise bending moment of the sec- necessary to establish an accurate unsteady aerodynamic model to
ond blade model increases by 18.81% compared with the first blade compute the dynamic aerodynamic loads of wind turbines. In this
model, while the displacement of the blade tip increases by 25.63% article, the numerical simulation of the airfoil dynamic stall char-
with more violent fluctuation. The considerable differences mainly acteristics of a HAWT is realized based on the BeddoeseLesihman
occur because the blade vibration and torsional deformation have model. By combining the blade MBS model and the BEM modified
significant effects on the angle of attack. Thus, the aerodynamic
forces computed by these two blade models are different, resulting
in the considerable differences of the blade bending deflections and
the blade root bending moments.
For the large-scale wind turbines, the coupling between the
flexible blade torsional deformation and vibration and the aero-
dynamic loads should be considered to more accurately predict the
aerodynamic loads. This is essential for the optimum design of
wind turbines.

5.2. Calculation of airfoil dynamic aerodynamic loads

Accurate analysis of dynamic aerodynamic loads is the basis for

the design and verification of large-scale wind turbines, while the
precise modeling of blade unsteady aerodynamic characteristics is
the prerequisite for dynamic aerodynamic load analysis for wind
turbines. When an airfoil undergoes random motion, a complicated
stall decay phenomenon occurs, commonly referred as ‘dynamic
stall.’ In such case the stall angle of attack and the lift and drag Fig. 6. The curves of time-varying angle of attacks of a blade cross section.
 W. Mo et al. / Energy 89 (2015) 1001e1009 1007

Fig. 8. The curves of the airfoil lift coefficients versus angle of attack: (a) rigid body B6;
(b) rigid body B9.

optimum lift over drag ratio during operation. The variation range
of angle of attack of each airfoil is also different.
As shown in Fig. 8, the lift coefficients of two typical blade cross
sections are respectively compared with steady data provided by
NREL to illustrate the dynamic stall characteristics of an airfoil. For
the rigid body B6, its cross section is the DU25_A17 airfoil, whose
static stall angle of attack is approximately 10 . In the unsteady
aerodynamic load simulation, the variation range of the angle of
attack of the airfoil operating at the rated wind speed (11.4 m/s) is
9.30 e9.77, which is near the airfoil stall region. When a dynamic
stall occurs, the lift coefficient increases considerably due to the
flow separation; however, when the angle of attack decreases, the
airfoil dynamic flow field does not quickly revert to the initial state.
There is a time lag in the lift coefficient response against the step
change of the angle of attack; hence, the hysteresis loop appears in
the unsteady lift coefficient curve, as shown in Fig. 8a. For the rigid
body B9, its cross section is the DU21_A17 airfoil, whose static stall
angle of attack is approximately 9 . The variation range of the angle
of attack of the airfoil in the unsteady aerodynamic load simulation
is 5.31 e6.37, which is near the linear region. In this region, there
is no obvious flow separation, and the lift coefficient is dominantly
affected by the attached flow. Hence, the unsteady value of lift
coefficient basically matches its static value, as shown in Fig. 8b.

5.2.2. Comparison between the steady and unsteady aerodynamic

Fig. 7. The time domain responses of the blade root bending moments: (a) edgewise
direction; (b) flapwise direction.
model, which incorporates with the BeL model, the blade dynamic
aerodynamic loads can be computed.
5.2.1. Comparison with steady data
The research subject, the blade of a 5-MW HAWT, consists of a
series of different airfoils. The twists of the different airfoil cross
sections are different, and they decrease gradually from blade root
to blade tip. This enables each airfoil cross section to obtain the
loads
In these studies the blade MBS model combines the BEM
modified model (i.e., the steady aerodynamic model) and the
modified BEM model integrating the BeL dynamic stall model (i.e.,
the unsteady aerodynamic model). The blade root moments under
turbulent wind speeds are analyzed. The means of the turbulent
wind speeds are 11.4 m/s, 14.5 m/s, 18 m/s and 20 m/s, respectively.
The variation curves of an airfoil angle of attack of rigid body B9
operating under impulsive wind speed with the mean value of
20 m/s are plotted in Fig. 9a, while the curves of the unsteady and
steady lift coefficients of the airfoil are compared in Fig. 9b. From
Fig. 9a, it is obvious that the airfoil angle of attack is in the stall

Table 1
The aeroelastic coupling effect on the blade aerodynamic loads.

Contrast data The 1st model The 2nd model Percentagea (%)
6 6
Blade root edgewise bending moment (N m) Mean 2.46  10 2.34  10 4.94
Variance 4.99  1012 6.94  1012 39.21
Blade root flapwise bending moment (N m) Mean 8.17  106 9.70  106 18.81
Variance 1.05  1012 1.27  1012 21.81
Edgewise displacement of blade tip (m) Mean 1.221 1.145 6.23
Variance 0.103 0.152 48.26
Flapwise displacement of blade tip (m) Mean 4.593 5.770 25.63
Variance 0.359 0.455 26.54
a
In the “Percentage” column, the minus denotes that the value of the second blade model is smaller than that of the first blade model.
1008 W. Mo et al. / Energy 89 (2015) 1001e1009

6
x 10
5

Edgewise bending moment [N·m]

0

-5

-10 steady data

unsteady data

-15
10 15 20 25 30 35 40 45 50
Time [s]
x 10
7 (a)
1.5
steady data

Flapwise bending moment [N·m]

1.4 unsteady data

1.3

1.2

1.1

0.9
10 15 20 25 30 35 40 45 50
Time [s]
(b)

Fig. 10. The curves of the blade root bending moments at 20 m/s mean wind speed: (a)
edgewise direction; (b) flapwise direction.

fluctuation amplitude and frequency of the load spectrum are

Fig. 9. Aerodynamic parameter curves of the blade cross section for B9 at 20 m/s mean
wind speed: (a) angle of attack; (b) lift coefficient.
essential for the blade fatigue characteristic analysis and the fatigue
life design. From the comparison between the simulation results, it
is significant and necessary to introduce the dynamic stall model
into the blade aerodynamic load analysis.

region because its mean is larger than the static stall angle of attack
(9 ). As seen in Fig. 9b, the unsteady lift coefficient deviates
considerably from the steady data, and the maximum deviation of
the two coefficients is up to 40.53%. This means that the dynamic
stall phenomenon can have significant effects on the unsteady
aerodynamic load calculation, and the same conclusion was also
drawn in literature [4] and [7].
Fig. 10 shows that the variation amplitude of the flapwise and
edgewise bending moments of the blade root computed by the
unsteady aerodynamic model are larger than those by the steady
aerodynamic model.
Table 2 shows the analysis on the means and variances of the
flapwise and edgewise blade root moments. As the mean wind
speed increases, the variation percentage of the means obtained by
applying the two aerodynamic models changes little, but the
variation percentage of the variances increases dramatically. When
the wind speed is low, the blade airfoils mainly work in the linear
region; therefore, there is little difference between the calculation
results of the steady and unsteady models. Nevertheless, when the
mean wind speed approaches to 20 m/s, most of the blade airfoils
work in stall conditions, so the dynamic stall makes the values of
the lift and drag coefficients fluctuate more violently. Consequently,
both the fluctuation amplitude and the frequency of the unsteady
aerodynamic loads increase considerably.
When wind turbines operate under random loads, most of the
failure is caused by the fatigue failure of the components, and the
6. Conclusions
In this article, the aeroelastic coupling equations of a large-scale
wind turbine blade were derived by the computational dynamics of
multi-body systems theory and the aerodynamics theory. The
aeroelastic code was developed, and then the aeroelastic coupling
simulation analysis of flexible blades was carried out based on the
simulation code. The feedback of the blade torsional deformation
and vibration on the aerodynamic loads were quantifiably
analyzed. The BeddoeseLeishman dynamic stall model was inte-
grated into the BEM-modified model in the present study to
investigate the unsteady blade airfoil aerodynamic characteristics.
The simulation results between the steady and unsteady aero-
dynamic models were compared and analyzed. It can be concluded
that:
a. The out-of-plane vibration and torsional deformation of a flex-
ible blade have significant effect on the aerodynamic loads. The
analysis on the time-domain responses of a 5-MW wind turbine
blade working under rated wind speed indicates that the ob-
tained means and variances of the flapwise blade root bending
moment and the blade tip out-of-plane deflection from the
model that neglects the aeroelastic coupling are approximately
20% greater than those from the model that includes the aero-
elastic coupling, respectively. This shows that the blade design
 W. Mo et al. / Energy 89 (2015) 1001e1009 1009

Table 2
The dynamic stall effect on the blade aerodynamic loads.

Wind speed Comparative data Edgewise bending moment at the blade Flapwise bending moment at the blade
root (N m) root (N m)

Mean Variance Mean Variance

6 12 6
11.4 m/s Steady data 2.46  10 4.99  10 8.16  10 1.04  1012
Unsteady data 2.41  106 5.14  1012 8.21  106 1.09  1012
Percentagea (%) 1.99 3.13 0.53 4.83
14.5 m/s Steady data 3.38  106 5.08  1012 9.85  106 5.32  1011
Unsteady data 3.34  106 5.29  1012 9.94  106 6.26  1011
Percentage (%) 1.10 4.12 0.88 17.80
18 m/s Steady data 4.31  106 5.25  1012 1.10  107 4.45  1011
Unsteady data 4.27  106 5.82  1012 1.12  107 5.57  1011
Percentage (%) 0.75 10.69 1.25 25.18
20 m/s Steady data 4.83  106 5.42  1012 1.16  107 4.20  1011
Unsteady data 4.86  106 1.07  1013 1.19  107 9.79  1011
Percentage (%) 0.58 97.48 2.52 133.17
a
In the ‘Percentage’ rows, the minus denotes that the magnitude of the bending moment mean obtained by utilizing the unsteady airfoil data is smaller than that obtained
by utilizing the steady airfoil data.

not based on the aeroelastic coupling analysis tends to be
conservative.
b. The BeL dynamic stall model can reflect the effect of dynamic
stall characteristics of blade airfoils on blade aerodynamic loads.
When a dynamic stall occurs, the circulatory hysteresis char-
acteristics in the lift and drag coefficients will cause more vio-
lent blade vibration. The simulation results show that there is
little difference in the mean values of blade root bending mo-
ments between the unsteady and steady models, but the fluc-
tuation amplitude and vibrational frequency considerably
increase in the unsteady model. Particularly when the mean
wind speed approaches 20 m/s, the variance values of the blade
root bending moments of the unsteady model exceeds the
steady model by 100%. Therefore, the effect of dynamic aero-
dynamic loads must be taken into account when carrying out
blade fatigue load spectrum analysis.
c. An analysis conducted by Hansen demonstrates that airfoil dy-
namic stall characteristics and out-of-plane vibration have
considerable effects on aerodynamic damping. The established
aeroelastic model in this article can be used to predict the un-
steady aerodynamic force as well as the work done by the force
in one oscillating period, which provides a basis for further
analysis on the aeroelastic damping and stability of flexible
blades.
Acknowledgments
The work is supported by National Natural Science Foundation
of China (Grant No. 51276043).
References
[1] Jonkman J, Butterfield S, Musial W, Scott G. Definition of a 5-MW reference
wind turbine for offshore system development. NREL Technical report, No TP-
500e38060. National Renewable Energy Laboratory; 2009. Available from:
http://www.osti.gov/scitech/biblio/947422.
[2] Lin W, Liu XW, Nathalie R, Matthew S, George MH. Nonlinear aeroelastic
modelling for wind turbine blades based on blade element momentum theory
and geometrically exact beam theory. Energy 2014;76:487e501.
[3] Liu X, Liang S, Chen Y, Zhang SQ, Chen C. Dynamic stall simulation of wind
turbine airfoils. Eng Mech 2015;32(3):203e11.
[4] Liu X, Zhang XM, Chen Y, Ye ZQ. Transient aerodynamic load prediction of
horizontal axis wind turbine based on the BEDDOES- LEISHMAN model. J Acta
Energiae Solaris Sinica 2008;19(12):1449e55.
[5] Du ZH, Selig MS. A 3-D stall-delay model for horizontal axis wind turbine
performance prediction. In: 36th AIAA Aerospace Sciences Meeting and
Exhibit. Reno, Nevada: University of Illinois at Urbana-Champaign; 1998.
[6] Shipley DE, Miller MS, Robinson MC. Dynamic stall occurrence on a horizontal
axis wind turbine blade. Technical report, No TP-442e6912. National
Renewable Energy Laboratory; 1995. Available from: http://www.osti.gov/
scitech/biblio/111926.
[7] Liu X, Zhang XM, Li DY, Chen Y, Ye ZQ. Unsteady aerodynamic force analysis of
wind turbine airfoil. J Basic Sci Eng 2010;18(3):503e16.
[8] Larsen JW, Nielsen SRK, Krenk S. Dynamic stall model for wind turbine airfoils.
J Fluids Struct 2007;23:959e82.
[9] Holierhoek JG, Vaal JB, Zuijlen AH, Bijl H. Comparing different dynamic stall
models. Wind Energy 2013;16(1):139e58.
[10] Theodorsen T. General theory of aerodynamic instability and the mechanism
of flutter. In: Jones RT, editor. Classical aerodynamic theory. Washington D.C.:
National Aeronautics and Space Administration; 1979. p. 291e311.
[11] Dat R, Tran CT. Investigation of the stall flutter of an airfoil with a semi-
empirical model of 2-D flow. Vertica 1983;7(2):73e86.
[12] Leishman JG, Beddoes TS. A semi-empirical model for dynamic stall. J Am
Helicopter Soc 1989;34(3):3e17.
[13] Dai JC, Hu YP, Liu DS, Long X. Aerodynamic loads calculation and analysis for
large scale wind turbine based on combining BEM modified theory with dy-
namic stall model. Renew Energy 2011;36:1095e104.
[14] Hansen MH, Gaunaa M, Madsen HA. A Beddoes-Leishman type dynamic stall
model in state-space and indicial formulations. Denmark: Risø National Lab-
oratory; 2004.
[15] Molenaar DP. Cost-effective design and operation of variable speed wind
turbines. Delft: Delft University of Technology; 2003.
[16] Zhao XY, Peter M, Wu JY. A new multibody modeling methodology for wind
turbine structures using a cardanic joint beam element. Renew Energy
2007;32:532e46.
[17] Holierhoek JG. Aeroelasticity of large wind turbines. Delft: Delft University of
Technology; 2008.
[18] Rauh J, Schiehlen W. Various approaches for the modeling of flexible robot
arms. In: Proceedings of the Euromech-Colloquium 219 on Refined Dynamical
Theories of Beams, Plates, and Shells and their Applications, Kassel. Berlin
Heidelberg: Springer- Verlag; 1986e87. p. 420e9.
[19] Li DY, Mo WW, Yan XH, Zhang XW. Aeroelastic analysis of horizontal axis
wind turbine based on multi-body model. J Mech Eng 2014;50(12):140e50.
[20] Li C, Ye Z, Gao W, Jiang Z. Modern land-sea wind turbine calculation and
simulation. Shanghai: Shanghai Science and Technology Press; 2012.
[21] Hong JZ. Computational dynamics of multibody systems. Beijing: Higher Ed-
ucation Press; 1999.