Sei sulla pagina 1di 6

Simulation of Aerodynamic Divergence and Flutter on

Wind Turbines using ANSYS-CFX


Drishtysingh Ramdenee 1 , Sorin Ion Minea1 and Adrian Ilinca1
1

Wind Energy Research Laboratory, Universit Du Qubec Rimouski


Email: Author@eng.uwo.ca

ABSTRACT
The recent development of large wind turbines poses
new challenges with regard to understanding the
mechanisms surrounding unsteady flow-structure
interaction. The larger and more flexible blades
imply risks from an aeroelastic point of view and
urge the need to properly understand and model these
phenomena. Due to limited experimental data
available in this field, Computational Fluid dynamics
(CFD) techniques provide an invaluable alternative to
identify and model aerodynamic and aeroelastic
phenomena around the wind blades. The study is part
of the coupling between aerodynamic and elastic
models of the commercial code - CFX with ANSYS,
respectively. In this paper we are modeling the
aeroelastic divergence. This article presents the the
studies aiming at modeling divergence. In this article
the results of the divergence modeling using ANSYSCFX will be presented and compared with results
from Jennifer Heeg [13]. The study is realized on
the NACA0012 airfoil for which experimental data
are available in literature. The ANSYS workbench is
used for the fluid structure interaction to simulate the
divergence phenomenon which is a structural
response imposed by aerodynamic loads due to
transient fluid flow.

1. INTRODUTION
An aeroelastic instability occurs when the variation in
the aerodynamic forces resulting from the blade
displacement tends to amplify the latter. The most
commonly occurring aeroelastic phenomena with
wind turbines are stall induced vibrations. Flutter is
another instability problem that needs to be inspected
and understood in an aim to mitigate it. Stall induced
vibrations occur in dynamic stall controlled wind
turbines. The gradient of the power coefficient curve
becomes negative when part of the blade is
subjected to stall, having as result a local negative
aerodynamic damping of the blade movement in the
direction of the lift. If the global aerodynamic
damping for a particular vibration mode is negative,

and exceeds (in magnitude) the structural modal


damping, then the oscillations can be amplified from
any initial perturbation independent of the ratio
between the normal frequency of the vibration mode
and the excitation frequencies. The first mode in each
direction is more prone to such a behavior as the
structural damping increases with the frequency
whereas the aerodynamic damping decreases. On the
other hand, flutter is a very dangerous phenomenon
which results from an interaction between elastic,
inertial and aerodynamic forces. It occurs when
structural damping becomes insufficient to damp
aerodynamically induced vibration motions. Flutter is
caused by the superposition of two structural modes
pitch and plunge. The pitch mode is described by a
rotational movement about the elastic centre of the
airfoil whereas the plunge mode is a vertical up and
down motion at the blade tip. As wind speed
increases, the frequencies of these modes coalesce to
create the flutter motion. This is called flutter
resonance. Usually, flutter is initiated as the airfoil is
subjected to an initial rotation. As a divergence like
phenomena arises, the torsional stiffness of the blade
reacts to achieve zero rotation again. On the other
hand, the resistance to bending tries to restore a
neutral position and sets the airfoil in a nose down
rotation position. The amplified force causes
plunging and torsional stiffness to restore zero
rotation. With time, though [1] the plunging motion
tends to damp out, the rotation motion diverges till
failure.

2.

CHALLENGE IN MODELING
AEROELASTIC PHENOMENA

Modeling of aeroelastic phenomena requires coupling


of aerodynamic and structural equations and such
represents a particular challenge. The need to obtain
solutions for several combinations of the parameters
of the structure and the fluid require very precise
modeling of the fluid model solutions and this is very
demanding from a computational point of view. For
several years the finite element model size has been
reduced. This has been possible by first finding the
Eigen modes and using a discrete set of values to

rebuild a discrete structure upon which the modeling


is performed. Lagrangian equations of classical
dynamics are made use of and the computational
requirements are appreciably reduced. Usually a
finite element structural model with some thousands
of degree of freedom (DoF) can be reduced into one
of some ten DoF.[2]. Though, this method is highly
convenient in terms of computational needs, it is very
tedious and requires refined modeling at each step.

These surfaces are made use of in order to transfer


the solicitation between the structural and fluid
domains. In the case of the divergence, it is the fluid
that imposes the solicitations on the solid such that
the CFX code will be the first to be solved followed
by the ANSYS code.
.

3. ANSYS-CFX COUPLING

There has been little, if not, no article that pondered


on the modeling of divergence modeling as from subcritical conditions till failure. This can be explained
by the difficulties encountered by software to model
acute gradients in the proximity of divergence and,
also, due to the fact that the primary importance is
only to analyze stability and sub-critical speed and to
design the system in order for the latter to stay within
these conditions at all time. [3] has illustrated the
dramatic reduction in the divergence speed for a
reversed arrow wing. [4] went further with the works
presented in [3] and proposes an analytical method to
calculate the divergence speed of the airfoils. [5]
clearly illustrated the destructive nature of divergence
and the possibility of controlling the latter. [6]
realized wind tunnel experiments which showed the
fundamental relationship between the angle of the
blade, the orientation of the material fibers and the
divergence speed. Rodney H.Ricketts, and Robert
V.Doggett, Jr [7] makes use of flat plate models
with varying geometries and their subcritical
response testing techniques were formulated and
evaluated for accuracy in predicting static
divergence. Sefic, Walter J., and Cleo M.Maxwell
[8] did experiments to correlate flight data with
predicted structural stability and determination of
aeroservoelastic stability margins. Stanley R. Cole,
James R. Florance, Lee B. Thmpson, Charles
V.Spain and Ellen P.Bullock [9] made use of
experimental data obtained from supersonic tests at
the Unitary Plan Wind Tunnel at the NASA Langley
Research Center to examine the divergence of all
moveable parts. [10] has performed experiments to
correlate flight results with predicted structural
stability and the determination of aerservoelastic
stability. A more detailed literature review is
available in previous works [11] and [12]. The
literature surrounding the divergence phenomenon is
very large and includes several models all mostly
aiming at modeling sub-critical velocity, dynamic
pressure and the modal frequency within the stability
zone. Most works dealing with experimental analysis
have lengthily discussed on the errors and the
graphical results provide a domain of divergence
rather than a fixed value. The consulted literature has
been very broad, though, not exhaustive, and the

In order to realize the fluid structure coupling study,


we make use of ANSYS multi domain (MFX). This
module was primarily developed for fluid-structure
interaction studies. On one side, the structural part is
solved using ANSYS Multiphysics and on the other
side, the fluid part is solved using ANSYS CFX. The
study needs to be conducted on a 3D geometry.
However, if the geometries used by ANSYS and CFX
need to have common surfaces (interfaces), the
meshes of these surfaces need not be identical [3].
The ANSYS code acts as the master code and reads
all the multi-domain commands. It recuperates the
interface meshes of the CFX code, creates the
mapping and communicates the parameters that
control the timescale and coupling loops to the CFX
code. The ANSYS generated mapping interpolates
the solicitations between the different meshes on each
side of the coupling. Each solver realizes a sequence
of multi-domain time steps and coupling iterations
between each time steps. For each iteration, each
solver recuperates its required solicitation from the
other domain and then solves it physical domain.
Each element of interface is initially divided into n
interpolation faces (IP) where n is the number of
nodes on that face. The 3D IP faces are transformed
into 2D polygons. We, then, create the intersection
between these polygons, on one hand, the solver
diffusing solicitations and on the other hand, the
solver receiving the solicitations. This intersection
creates a large number of surfaces called control
surfaces as illustrated in figure 1.

Figure 1: Transfer Surfaces and resolution scheme

4. STATE OF THE ART

fundamental philosophy has been the same: the


objective of most aeroelastic work have most of the
time underlined the consequences of divergence and
flutter, the sub-critical behavior of structures in order
to be able to know how far we are from the critical
zones in order to avoid them. Modeling of these
phenomena has only been a very limited part of the
works.

The mass of the considered configuration of the


original model is 2.2864 kg, and the mass of our
model is 262.5 times smaller, that is 0.00871 kg. The
moment of inertia is such that our model has the
same fundamental frequency as the original that is
7.88 Hz. Figure 2 illustrates the model built on
ANSYS.

5. EXPERIMENTAL VALIDATION
In 2000, Jennifer Heeg [13] realised aeroelastic
experiments at the Duke University wind Tunnel.
The aim was to validate analytical calculations of non
critical characteristic modes and explicitly examine
the aerodynamic divergence phenomenon. The
simplest model was built and tested. A NACA 0012
airfoil was used and the structure was only allowed a
single DoF, i.e., in torsion. A chord length of 8 inches
and a blade length of 21 inches were made use of.
The wing was made out of an aluminum shell of 1/32
inches in thickness. A spring constant of 5.826
Nm/rad was used. The structural dynamic
parameters for this model are summarized in table 1:
Elastic

Natural

Natural

Damping ratio

Constant K

pulsation

Frequency f

[Nm/rad]

[rad/sec]

[Hz]

5.8262

49.5

7.88

Details about the used fluid model are available from


[14].

7.

RESULTS

In a preliminary stage, prior to divergence and zerofrequency flutter modeling, the lift coefficient curve
for the model at a velocity of 20 m/s was constructed
using ANSYS and was compared to results from
[13]. Furthermore, the ANSYS-CFX coupling was
verified.

0.053

Table 1: Structural dynamic parameters associated to


the model used for the wind tunnel experiment from
[13]

The objectives of the studies in [13] were to : 1)


calculate the dynamic pressure at divergence, 2)
examine the modal characteristics of non critical
modes, 3) examine the behavior of Eigen vectors.The
aim of our simulations is to be able to simulate the
same using ANSYS- CFX coupling.

7.1 Construction of the

SIMULATION MODEL

The model of the experiment was simulated at a


reduced scale, in order to reduce the calculation time
by reducing the dimensions of the fluid domain. The
span of the airfoil was reduced 262.5 times, from 21
inches to 0.08 inches or 2.032 mm, while the chord of
the airfoil was maintained at 8 inch or 203.2 mm. We
used a cylinder to simulate the torsion spring used in
the configuration of the experiment that was detailed
in the previous section. The constant of the original
spring is K = 5.8262 Nm/rad and since we used a
reduced model, with an span 262.5 times smaller than
the original, the dimensions and properties of the
cylinder are such that:
= 0.022195 N m/ rad

curve

The curve was constructed for a velocity of 20 m/s,


which is between a velocity of 19.15 m/s estimated
by [13] and 20.16 m/s which [13] used as comparison
from experimental studies. For the validation of our
simulation, we have used experimental data from
[15] for Re numbers of Re = 1.7x105 and 3.3x105
from which we interpolated values for Re =2.6x105
corresponding to values used in our ANSYS-CFX
simulation. The result is shown in figure 3:
Lift coe fficie nt - CL()
1,2

0,8

CL

6.

Figure 2: Meshed ANSYS simulated model

0,6

0,4

0,2

0
0

10

12

14

Figure 3: Lift coefficient curve of the NACA 0012 for Re =


2.6x105 calculated using ANSYS-CFX and experimental
results from Sheldahl & Klimas (blue circle for increasing
angle and yellow triangle for decreasing angle)

The results from ANSYS-CFX are in close


accordance with experimental data till an AoA of
12. At 2.5, the results from CFX are inferior by
13%, at 4.5 by 20% and at 6.5 by 6%. The
maximum value of the lift coefficient is achieved
around 12 and is 0.96 and 0.9 at 10.5. As for CFX
achieved results, the maximum value is again
achieved at around 12 and is 0.924 and 0.899 at
10.5. The difference in the values can be explained
by the fact that Sheldahl & Klimas have extrapolated
the obtained results from experiments using the
software PROFILE to estimate the aerodynamic
coefficients from other Re numbers. Linearization of
the polars for low AoA had, also, been performed.

the simulation. A time step of 3.8 10-4 s is used


though it would have been more interesting to use a
variable time step to adjust with the Courant number.
Figures 5 and 6 illustrate the response of the blade
provided by ANSYS-CFX. The represented variable
is the AoA of the airfoil.

7.2 Verification of the Coupling


In order to verify the coupling between Mechanical
APDL and CFX, we performed a simulation at a
constant flow below divergence. The profile in [13]
was fixed at 0 = 4, restricted to one DoF and
subjected to vair = 15 m/s as a shock wave. Figure 4
illustrates the ANSYS-CFX generated response. The
represented variable is the vertical displacement of
the trailing edge. The equilibrium position induced
force obtained from CFX is FL = 0.0363787 N. The
AoA corresponding to a displacement of 0.006 m is,
= arcsine (sin (0) + 0.006 / y0) = 7.022, where y0
= 0.1143 m. At this angle, the elastic moment is ME =
K ( 0) = 0,022195 0.05275 = 0.00117 Nm.
We obtain the same aerodynamic moment value from
CFX.

Figure 5: nstability simulation obtained for an AoA


of 5 for configuration #2 from [13]

Figure 6: Instability simulation obtained for an AoA


of 50 for configuration #2 from [13] with details

Figure 4: Oscillatory response of the profile subjected


to a sudden shock wave of 15 m/s

7.3 Divergence and flutter simulation


The model from [13] was restricted from all DoF and
fixed to an AoA of 5. For convergence needs, it was
subjected to a constant velocity of 1 m/s till
stabilisation of the flow. The fixation is then removed
leaving a rotation about the elastic axis and the
velocity is increased according to the expression V =
15.84 14.84*exp (-3*t) till 15.805 m/s at the end of

The AoA of the airfoil increases progressively until it


enters a region of stall controlled dynamic
equilibrium. As the speed tends to 15.84 m/s, the
AoA stays temporarily small, resting in equilibrium
between the aerodynamic moment and the elastic
moment between 10 and 10.5. The onset of flutter
is very quick due to the high gradient of the used
velocity profile which avoids any possibility of stall
stabilization. The simulation was stopped at 150
because the Courant number ( ) reached 81.84
which is much larger than the recommended value
such that with the used time step, the analysis cannot
go beyond 15.

7.4 Discussion
Due to numerical computation limitations, we can
only verify the frequency of the movement in a

velocity range between 15.6 and 15.8 m/s


corresponding to a small error of 1.26 % which we
judge very close to experimental data. The obtained
frequencies in these conditions vary between 5.822
Hz and 6.326 Hz which is close to the 6 Hz obtained
from the experiments conducted in [13]. In this case,
it will be more difficult to determine the flutter speed
as the airfoil achieves equilibrium in the stall zone
with the flow velocity still increasing. Due to this
perturbation, the flutter will start at a speed smaller
than that used in the experiments. In order improve
on the results, the simulation running time should be
longer and the simulation parameters more precise.
However, due to computational frontiers, we had to
limit ourselves to the presented simulation.

7.5 Divergence and flutter illustration


Figures 7 to 9 illustrate the airflow and the airfoil
movement at certain time steps of the simulation with
the air speed as the represented variable. In order to
ease eddy visualization, we have used a narrow speed
range of 14 m/s to 18 m/s for the color scale. In the
dark blue zones, the airflow is less than 14 m/s and in
the red regions, the speed is larger than 18 m/s.
Figure 7 illustrates the flow at the moment noted 1on
figure 6, when the oscillation is its minimum,
=
6.53 and the maximum airflow, U= 26.95 m/s.
Figure 8 illustrates the airflow at moment noted 2 on
figure 6 at an intermediate point of the oscillation,
where
= 10.780 and U = 33.65 m/s. Figure 9
illustrates timestep at moment 3, at a crest of the
oscillation with = 14.580 and U = 215.38 m/s.

Figure 8: Instability Simulation at t=1.88822 s,


AoA=10.780, U=33.65 m/s

Figure 9: Instability Simulation at t=1.93154 s,


AoA=14.580, U=215.38 m/s

8.

Figure 7: Instability Simulation at t=1.8449 s,


AoA=6.530, U=26.95 m/s

CONCLUSION

This article pondered on the divergence and zerofrequency flutter phenomena. The ANSYS-CFX
coupling to model fluid-structure interaction has been
very useful and we have been able to satisfactorily
model these phenomena within limits bounded by
computational capacity. It seems that the threshold
between divergence and flutter is very narrow. The
modeling of such phenomena is very complex and we

see that it is, still, very difficult to produce closely


reproducible results. The problem lies in
computational capacities and, also, some refinements
to be added into analytical modeling. The obtained
results are very encouraging in anticipating good
results for a 3D blade analysis. The expected
precision for a 3D blade will mainly depend on the
mesh size, the turbulence model and the
computational capacity that will allow us to use very
refined time steps.

REFERENCES
[1]http://www.cs.wright.edu/~jslater/SDTCOutreach
Website/aerodynamic_flutter_banner.pdf
[2] Dowell, e.a., A Modern Course in Aeroelasticity,
ed. G.M.L. GLADWELL. 2005, Dordrecht:
KLUWER ACADEMIC PUBLISHERS.
[3] Diederich, F.W., Budiansky, B. : Divergence of
swept wings NACA/Technical Note no.1680, August
1948
[4] Diederich, F.W., Foss, K.A. : Static Aeroelastic
Phenomena of M-, W- and
- Wings
NACA/Research Memorandum, February 9, 1953
[5] Krone, N.J., Jr. , Divergence Elimination with
Advanced Composites. August 1975, AIAA Paper no.
75-1009
[6] Blair, M.: Wind tunnel Experiments on the
Divergence of Swept Wings with Composite
Structures, Air Force Wright Aeronautical
Laboratories/Technical Report AFWAL-TR-82-3018,
October 1982
[7] Rodney H.Ricketts, and Robert V.Doggett, Jr,
Wind tunnel Experiments on Divergence of ForwardSwept wings
[8] Sefic, Walter J., and Cleo M.Maxwell, X29-A
Technology
demonstrator
Flight
Test
programOverview
[9] Stanley R. Cole, James R. Florance, Lee B.
Thmpson, Charles V.Spain and Ellen P.Bullock,
Supersonic Aeroelastic Instability Results for a
NASP- like Wing model
[10] Sefic, W.J., Maxwell, C.M.: X29-A Technology
Demonstrator Flight Test Program Overview
NASA/Technical Memorandum 86809, May 1986
[11] D.Ramdenee et al. Numerical Simulation of the
Divergence phenomenon on a NACA 4412

Airfoil_part 1 Canadian Societyof Mechanical


Engineers Conference, University of Victoria, British
Columbia. June 2010
[12] I.S Minea Analyse de la rponse dynamique
des pales des turbines oliennes soumises des
charges arodynamiques laide des logiciels
commerciaux ANSYS et CFX Masters thesis,
Universit du Qubec Rimouski. February 2011
[13] Heeg, J.: Dynamic Investigation of Static
Divergence: Analysis and Testing NASA/Technical
Publication NASA/TP-2000-210310, November
2000
[14] D.Ramdenee et al. Numerical Simulation of
the Divergence phenomenon on a NACA 4412
Airfoil_part 2 Canadian Societyof Mechanical
Engineers Conference, University of Victoria, British
Columbia. June 2010
[15] Sheldahl, R.E., Klimas, P.C.: Aerodynamic
Characteristics of Seven Symmetrical AirfoilSections
Through 180-Degree Angle of Attack For Use In
Aerodynamic Analysis of Vertical Axis Wind
Turbines Sandia National Laboratories/Report
SAND80-

Potrebbero piacerti anche