Validation and comparison of a newly developed aeroelastic design code for VAWT

Abstract

The open source simulation code QBLADE, based on a Lifting Line Free Vortex Wake formulation to evaluate the unsteady aerodynamics, recently integrated the PROJECT-CHRONO FEA library that, by using Euler-Bernoulli beams in a corotational formulation, solves for the structural dynamics to achieve an aeroelastic coupling. To validate the newly implemented structural model its performance is compared to literature data and two other structural computer codes. The comparison is based on a modal analysis, steady- and aeroelastic simulations of the SNL 34m VAWT testbed for which the aerodynamic and structural properties are well known. The structural loads are obtained from IEC 61400-1 design load cases. In one of the calculated load cases an aeroelastic instability could be observed which confirms similar observations that have previously been reported in the literature.

Full text

American Institute of Aeronautics and Astronautics
1
Validation and comparison of a newly developed aeroelastic
design code for VAWT
D. Marten1, M. Lennie2, G. Pechlivanoglou3, C.O. Paschereit4
TU Berlin, Berlin, 10623, Germany
N.V. Dy5, I. Paraschivoiu6
Polytechnique Montréal, Quebec, H3T 1J4, Canada
F. Saeed7
University of Dammam, Dammam, 34212, Saudi Arabia
The open source wind turbine simulation code QBLADE, based on a Lifting Line Free
Vortex Wake formulation to evaluate the unsteady aerodynamics, recently integrated the
PROJECTCHRONO FEA library that, by using Euler-Bernoulli beams in a corotational
formulation, solves for the structural dynamics to achieve an aeroelastic coupling. To validate
the newly implemented structural model its performance is compared to literature data and
two other finite element computer codes. The comparison is based on a modal analysis, steady-
and aeroelastic simulations of the SNL 34m VAWT testbed, for which the aerodynamic and
structural properties are well known. The structural loads are obtained from IEC 61400-1
design load cases. In one of the calculated load cases an aeroelastic instability could be
observed which confirms similar observations that have previously been reported in the
literature.
Nomenclature
AC = Actuator Cylinder Method
Cp = Power Coefficient
DLC = Design Load Case
DMST = Double Multiple Streamtube Method
ETM = Extreme Turbulence Model
EWM = Extreme Windspeed Model
HAWT = Horizontal Axis Wind Turbine
LLFVW = Lifting Line Free Vortex Wake Method
SNL = Sandia National Laboratories
TSR = Tip Speed Ratio
VAWT = Vertical Axis Wind Turbine
I. Introduction
NE can say that the major Vertical Axis Wind Turbines (VAWT) research stalled in the mid 90’s [1] when the
HAWT concept was adopted by the industry. Recently, after almost 20 years of absence from the agendas of the
large wind turbine research institutes, interest in VAWT technology is increasing again. The reason is that for deep
water offshore applications, where floating structures become a necessity, the VAWT concept offers some apparent
advantages over HAWT. These could lead to smaller floating structures, reduced logistics and materials costs and
ultimately to a lower Cost of Energy. Lately several research projects, involving the large European wind energy
1 Research Associate, Chair of Fluid Dynamics, TU Berlin, Berlin, 10623, Germany
2 Research Associate, Chair of Fluid Dynamics, TU Berlin, Berlin, 10623, Germany
3 Research Associate, Chair of Fluid Dynamics, TU Berlin, Berlin, 10623, Germany
4 Professor, Chair of Fluid Dynamics, TU Berlin, Berlin, 10623, Germany
5 Research Associate, Département de génie mécanique, Polytechnique Montréal, Quebec, H3T 1J4, Canada
6 Professor, Département de génie mécanique, Polytechnique Montréal, Quebec, H3T 1J4, Canada
7 Associate Professor, Basic Engineering Department, University of Dammam, Dammam, 34212, Saudi Arabia
O
AIAA Paper 2017-0452. Presented at the 35th Wind Energy Symposium at the
AIAA SciTech Forum, Grapevine, Texas, USA, Jan. 9-13, 2017

American Institute of Aeronautics and Astronautics
2
institutes, were started to investigate the VAWT offshore concept, such as the Deepwind, Nova, Inflow or S4VAWT
projects [2,3,4,5].
However, due to the lack of research focus in the past 20 years, there is a only a small number of design, simulation
and certification tools available when compared to the numerous tools that exist for HAWT. Also, no detailed large
scale aerodynamic measurement campaigns (comparable to the MEXICO or NREL Phase VI HAWT experimental
measurement campaigns) have been undertaken to provide the necessary data for their validation.
To further complicate the issue, no dedicated guidelines for VAWT certification exists and, due to a lack of
experience and tools, it is common practice for manufacturers [6] to follow the IEC 61400-2 standard under chapter
7.6, where the aeroelastic calculations are replaced with load measurements when the size of the wind turbine allows
it. The computation of the full set of Design Load Cases (DLC), as required by the IEC standard, for a VAWT is still
challenging for certifiers [6], some load cases appear redundant and others need to be modified [5].
Furthermore, the number of readily available coupled aeroelastic design tools that could be used for VAWT
certification is very limited. To the authors knowledge the only available aeroelastic VAWT tool, capable of being
used in a certification context, is DTU’s HAWC2. The number of available codes that model VAWT aero- or structural
dynamics separately is slightly larger and a (non-exhaustive) list of the most prominent tools is presented below:
A. VAWT aerodynamic codes
 QBLADE: Lifting Line with Free Vortex Wake (3D), developed at TU Berlin [7]
 HAWC2: Actuator Cylinder Flow Model (2D), developed at DTU [8]
 CARDAAV: Double Multiple Streamtube (2D), developed at Iopara [9]
 UMPM: Vortex Panel Method with Free Vortex Wake (3D), developed at TUD [10]
 CACTUS: Lifting Line with Free Vortex Wake (3D), developed at SNL [11]
B. VAWT structural codes
 HAWC2: Multibody formulation with Timoshenko beams, developed at DTU [8]
 DYNAMIST: Euler beam element code with gyroscopic and spin stiffness terms, from Iopara
 OWENS toolkit: Multibody formulation with Timoshenko beams, developed at SNL [12]
 QBLADE with CHRONO: Multibody formulation with Euler beams, in development at TU Berlin
The limited availability of computer codes for VAWT aeroelastic analysis was the main motivation for the new
development of a structural model for VAWT, and its integration with the unsteady aerodynamics module of the
QBLADE code. To be of practical use in the context of VAWT certification the target was to generate a reliable,
highly computationally efficient aeroelastic simulation tool. In this context, the exemplary runtime for a 630s long
simulation run is given in section 5.C of this document. In the following paragraphs a short overview of the
aerodynamic and the structural model formulation is given and a brief validation study is presented.
II. The aeroelastic model formulation in QBLADE
A. The aerodynamic model
As an aerodynamic model for the fluid-structure coupled analysis QBLADE uses the Lifting Line Free Vortex
Wake method (LLFVW). A large benefit of using a free vortex wake simulation code over the double multiple
streamtube (DMST) or actuator cylinder (AC) codes is that the rotor (blades and struts), and the wake can be modeled
in 3D, using the true rotor geometry, and no reduction of the problem into 2D stacked cylinders (AC) or streamtubes
(DMST) is necessary. This simplifies the setup of the problem greatly, especially if large deflections of structural
components, such as the rotor blades, occur during the simulation. Furthermore, as the “flow history” is contained in
the 3D vortex representation of the wake (see Figure 1), the LLFVW method is not limited to steady state solutions
of the flow field but is also able to capture the highly transient aerodynamics and forces that are inherent in VAWT
rotors – this includes the interaction of discrete vortices, shed at upwind blade positions, with rotor blades during the
downwind passage.
Overall, vortex methods offer a physical sound representation of VAWT aerodynamics with far less assumptions
than the AC or DMST methods. These benefits come at a significantly higher computational cost, but in recent
publications [13,14] it could be shown that, due to the ever-growing processing power of modern PC’s and by using

American Institute of Aeronautics and Astronautics
3
massive GPU parallelization and adaptive wake element reduction techniques, running full load calculations on a
single workstation in a reasonable timeframe is very feasible.
The formulation of the LLFVW method, implemented in QBLADE, is loosely based on the formulation as
described by van Garrel [15] and its implementation in QBLADE is used to simulate both HAWT and VAWT rotors.
Aerodynamic forces are evaluated at the quarter chord line of the blade panels using tabulated lift and drag polar data.
Blade struts are also modeled with aerodynamic panels that, if possessing a lift generating profile, are also modeled
through tabulated airfoil data. The wake is discretized with vortex line elements, shed at the blades (and struts) trailing
edge during every time step, that undergo free convection behind the rotor. The vortex elements are de-singularized
using the van Garrel’s cut off method [15] with the vortex core size, considering viscous diffusion via the vortex core
growth rate that is modeled through a turbulent vortex viscosity coefficient.
Dynamic stall effects play a major role in VAWT aerodynamics [16] and they are introduced [17] via the ATEFlap
aerodynamic model [18] that reconstructs lift and drag hysteresis curves from a decomposition of the lift polars. The
implemented ATEFlap formulation has been further adapted to work under the intricate conditions of VAWT
exhibiting large fluctuations of the angle of attack when rotating at low tip speed ratios (TSR).
To increase the computational efficiency the wake convection step is GPU parallelized using the OpenCL
framework. To further prevent the computational cost from growing exponentially over time, and to reduce costs,
different wake reduction schemes are implemented [19].
QBLADE can read and import wind input files in various formats, such as the AeroDyn hub height (*.hh) or the
binary turb sim (*.bts), etc.) to integrate realistic inflow conditions, according to the IEC 64100-1 standard, into the
aerodynamic simulations.
Figure 1. Visualization of the three dimensional freely evolving wake behind a VAWT, showing shed- and
trailing vorticity, as calculated with the LLFVW method in QBLADE
B. The structural model; integration with PROJECTCHRONO
The structural VAWT model that has been integrated in QBLADE code is based on the open-source multi-physics
simulation engine PROJECTCHRONO [20], developed at the University of Parma and the University of Wisconsin-
Madison. CHRONO is an object-oriented library with a C++ API and includes, among others, components for parallel
computation of nonlinear FEA analysis. CHRONO::FEA contains several finite element formulations to define
structural problems. To model the different structural components of VAWT in QBLADE the CHRONO
implementation of Euler-Bernoulli beams, in a co-rotational formulation [21], was chosen. These types of elements
are well suited for flexible multibody systems consisting of slender components such as blades, tower, struts and guy-
cables. The corotational formulation splits the dynamics of the system into rigid body motions and component
defections that are measured in the co-rotated frame. Furthermore, terms for geometric stiffening effects are already
accounted for in the Euler-Bernoulli beam elements tangent stiffness matrix. Using the constraint formulations,
provided in CHRONO, the turbine system can then be assembled from the different components. Constraints are also
used to impose certain boundary conditions, such as fixed or bearing nodes or constant rotational speeds of
components during the modal analysis.
For time marching simulations CHRONO contains several different solvers and time-integrators. In the case of
modal analysis, which is currently not supported in CHRONO, the global tangent stiffness-, damping-, mass- and
constraints matrices are exported (for different constant rotational speeds) and assembled outside of CHRONO to
solve the generalized eigenvalue problem.

American Institute of Aeronautics and Astronautics
4
D. Pre-processor for structural VAWT meshes and properties
To facilitate the generation of structural meshes for VAWT rotors of various configurations (Figure 2), a pre-
processor has been implemented that automatically generates the structural mesh in CHRONO from the aerodynamic
blade definitions in QBLADE’s blade design module. The pre-processor is also used to create possible guy-cable
elements. Blade tower connections can be attained directly or via struts. Material properties for the different
components are specified via input files resembling the blade file format of NREL’s FAST code.
Figure 2. VAWT geometries created with the pre-processor in
QBLADE
. Showing aerodynamic panels in red
and structural elements and nodes in black
E. Aeroelastic coupling
A loose coupling scheme, with independent time step sizes, is employed in the coupling of the aerodynamic and
the structural simulation (Figure 3). The advantage, when compared to a tight coupling scheme, is that both solvers
can operate in their own framework and no details of the respective simulation processes need to be communicated
during their operation. After each aerodynamic time step the aerodynamic forces and the current aerodynamic time
step size are communicated to the structural solver. The structural solver then advances the structural simulation with
its own temporal discretization (and constant aerodynamic forces) until its simulation time reaches that of the next
aerodynamic time step.
In case of sub-time steps of the structural simulation, where the rotation of the rotor is advanced in small
increments, the aerodynamic forces (split up into a normal and a tangential component) assigned at the beginning of
the structural step are rotated with the rotor coordinate system to more closely approximate their real orientation in
space. When the structural sub-simulation is finished, the nodal positions of all components are passed to the
aerodynamic simulation, panel positions are updated and relative element velocities due to rotations or component
deflections are computed to be used by the aerodynamic solver.
Figure 3. The implemented loose coupling scheme between the aerodynamic and structural module
As the aerodynamic discretization is independent from the structural discretization all values (forces and positions)
are linearly interpolated between both meshes, based on a non-dimensionalized length of the sub structures. As the
forces resulting from the aerodynamic simulation are linearly distributed over the panels they must be lumped to be
mapped onto the discrete nodal positions. Overall, the resulting aeroelastic coupling is very flexible and easily allows
to refine one or both meshes or temporal discretizations independently until the desired accuracy is achieved.

American Institute of Aeronautics and Astronautics
5
Figure 4 shows the sensitivity of the aeroelastic coupling towards the temporal discretization of the aerodynamic
sub module. In this comparison, the flapwise deflection of the midspan node for one revolution of the rotor is shown.
It can be seen that azimuthal discretizations of up to 5° yield sufficiently accurate results. At 10° azimuthal
discretization the distribution of the flapwise deflection start to deviate significantly from the reference calculation.
Due to performance considerations of the aerodynamic step a 5° azimuthal discretization is used in all subsequent
calculations of the aerodynamic sub-module.
Figure 4. Flapwise deflection of the blade 1 midspan node (inwards towards the tower is positive) over the
azimuthal angle for different aerodynamic time steps; dt_struct is constant at 0.001s; the REFERENCE line
was computed with dt_aero = 0.0001s and dt_struct=0.0001s
Figure 5 shows the sensitivity of the aeroelastic coupling towards the time step of the structural sub-module. The
distribution of the flapwise deflection, in magnitude and phase, is very sensitive towards the structural time step. A
value of 0.001s is used in all subsequent simulations – resulting in approximately 30 structural sub-calculations per
aerodynamic time step in this example. Without decreasing the size of the aerodynamic time step a finer structural
time step does not improve the accuracy of the prediction.
Figure 5. Flapwise deflection of the blade 1 midspan node (inwards towards the tower is positive) over the
azimuthal angle for different structural time steps; dt_aero constant at 0.03s; the REFERENCE line was
computed with dt_aero = 0.0001s and dt_struct=0.0001s

American Institute of Aeronautics and Astronautics
6
IV. Setup of the structural models used in the validation
As a first step in its validation, the performance of the newly implemented aeroelastic model in QBLADE has been
compared to the results of other simulation codes and to experimental data. The comparison was carried out based on
the SNL 34m testbed. As the range of aeroelastic codes for VAWT is very limited, the aerodynamic and structural
sub-models are first benchmarked individually. The aerodynamic performance of QBLADE was compared to the
computer code CARDAAV. The structural sub-component of QBLADE has been compared with the codes
DYNAMIST and ANSYS. The SNL 34m experiment and the computer codes used in the comparison are shortly
described in the following. Details of the modeled turbine geometry and the structural parameters that were used to
setup the models are given in the appendix.
A. The SNL 34m testbed and measured data
As a benchmark for the code comparison the SANDIA 34m turbine [22] is chosen. The reason for this selection is
that the structural properties are publicly available [23] and the SNL 34m testbed has already been employed in other
numerical studies [24][25], which enables a further comparison of the computed results.
The SNL 34m testbed was built in the late 80’s as a research-oriented vertical axis wind turbine and tested in
Bushland, Texas. It has been equipped with a large number of strain sensors, accelerometers and pressure tabs to
measure its mechanical and aerodynamic performance. Moreover, the test side was also equipped with two
meteorological towers to measure wind speed and direction. The test results, including aerodynamic performance
curves, measured natural frequencies and structural stresses, have been presented in several publications [26, 27].
A. CARDAAV
CARDAAV is a computer code based on the Double-Multiple Streamtube model with variable upwind and
downwind induced velocities in each streamtube (DMST). A unique feature of the CARDAAV code, apart from the
two actuator disks model, is that additional parameters or input options can be included, for example: to analyze the
influence of the blade geometry, the airfoil type and the secondary effects such as dynamic stall effects, the rotating
tower or the presence of struts on the performance of the VAWT.
B. DYNAMIST
DYNAMIST is a MATLAB code using the lumped mass approach to model the structure of a VAWT. The rotor
blades, struts or blade connectors, and the turbine tower are divided into simple straight 3D Euler-Bernoulli beam
elements. Each element is modeled as a second order differential equation that includes mass-, gyroscopic-, stiffness-
and spin stiffness matrices and an aerodynamic force vector on the RHS. Kinematic constraints are implemented by
fixing the respective degrees of freedom (DOF) of single element nodes. DYNAMIST is capable of modal analysis of
non-spinning and spinning VAWT and of forced response time marching simulations.
C. ANSYS
A simplified 3-D finite element model was created using the general purpose commercial finite element analysis
software ANSYS Mechanical. The blades and tower were modeled using uniaxial 3-D 3-node beam elements, based
on Timoshenko beam theory, with 6 degrees of freedom at each node; rigid beam elements were used to model the
rigid constraints between the blades and the tower near the rotor’s extremities; and the cables were modeled using 3-
D spar (uniaxial tension-compression) elements with 3 degrees of freedom at each node.
Figure 6. Overview of the structural models used in the comparison; from left to right; DYNAMIST: 82
blade, 7 tower, 8 connection, 3 guy cable elements; ANSYS: 126 blade, 30 tower, 4 connection, 3 guy cable
elements;
QBLADE
: 40 blade, 19 tower, 4 connection, 3 guy cable elements

American Institute of Aeronautics and Astronautics
7
V. Sub model validation and comparison
A. Aerodynamic Comparison / Validation
Figure 7 shows a comparison between the measured and the simulated aerodynamic performance of the SNL 34
testbed. Both codes, QBLADE and CARDAAV, used the same lift and drag polar data. Furthermore, the influence of
the tower was included in both simulations. CARDAAV employs a modified Gormont dynamic stall model while
QBLADE implements the ATEFlap unsteady aerodynamic model with a Beddoes-Leishman type dynamic stall
model.
As can be seen, when comparing to the experimentally measured data, both codes predict the constant rotational
speed power curves remarkably well, but QBLADE slightly overpredicts the rated power output. On the other hand,
when looking at the power coefficient (Cp) – tip speed ratio curves (TSR), CARDAAV overpredicts the maximum
Cp value. The reason for this discrepancy is the different implementation of the dynamic stall model. The polar
decomposition, as realized in QBLADE for the SNL 34m testbed, and the different dynamic stall model result in
slightly larger dynamic stall hysteresis loops – and thus in an overall stronger effect of the dynamic stall model on the
resulting lift and drag coefficients. This leads to an increase of performance for low TSR, where large angles of attack
are experienced by the rotor blades that cause a dynamic lift overshoot. At high TSR the more pronounced hysteresis
effects leads to an increase of the dynamic drag – which reduces the performance.
Figure 7. Aerodynamic performance of the SNL 34m testbed; top row: power curves at different rpm; bottom
row: power coefficient at different tip speed ratios
B. Modal Analysis Comparison / Validation
A modal analysis was performed with all three structural codes. Results for the parked rotor with engaged brakes
are compared in Figure 8 and numeric values for the computed frequencies of the first 10 eigen modes, with the
relative error of the prediction, are given in Table 1. The key that is used for the mode shape abbreviations is given in
Table 2.
From the comparison, it can be found that all blade modes (edgewise and flapwise) are predicted reasonably well
by the different codes with relative errors that are below 5% (comparing with the SNL measurements as a baseline).
The largest differences can be found in the prediction of the tower modes (and the PR mode that includes torsion of
the tower). One reason for the difficulties to model the tower might be the explicit integration of the guy cables as pre-
tensioned beam elements into the structural models. In similar setups, it has been common practice to model the cables
as springs. Overall the QBLADE and the ANSYS model show slightly more consistent results when comparing with
the experimental data.

American Institute of Aeronautics and Astronautics
8
Table 1. Comparison between predicted and measured modal frequencies
Type Experiment
[Hz]
QBLADE
[Hz]
ERROR
[%]
ANSYS
[Hz]
ERROR
[%]
DYNAMIST
[Hz]
ERROR
[%]
1FA 1,040 1,041 -0,06 1,011 2,83 1,010 2,89
1FS 1,040 1,043 -0,29 1,013 2,63 1,010 2,89
1PR 1,520 1,612 -6,07 1,660 -9,18 0,883 41,92
1BE 1,810 1,815 -0,26 1,806 0,24 2,047 -13,09
2FA 2,060 2,029 1,50 2,025 1,72 2,047 0,63
2FS 2,160 2,031 5,98 2,036 5,72 2,063 4,50
1TI 2,490 2,491 -0,04 3,127 -25,59 2,063 17,14
1TO 2,600 2,501 3,81 3,214 -23,61 2,453 5,66
3FA 3,450 3,378 2,10 3,366 2,44 3,363 2,52
3FS 3,450 3,388 1,79 3,370 2,32 3,363 2,52
Figure 8. Comparison of the first 20 modal frequencies, for the parked rotor, as computed by DYNAMIST,
ANSYS and QBLADE and measured in the SNL 34m experiment
The capability of the structural models to accurately predict the
rotationally augmented mode shapes, taking into account the geometric
stiffness terms, is compared in Figure 9. Simulations were carried out
for a range of constant rotational speeds, without considering
aerodynamic forces, and the resulting structural forces were applied to
the FEA model. The modal analysis was then carried out for the
stiffness, mass and damping matrices linearized around these points.
Figure 9 shows the eigen frequencies for the first three flapwise mode
shapes at different rotational speeds. Both the QBLADE model and the
ANSYS model predict growing frequencies with larger rotational
speeds and align well with the experimental data. The model setup in
the DYNAMIST code could not capture this behavior, spin-softening
was overpredicted, and lower frequencies resulted for the whole range
of rotational speeds.
1FA first flapwise antisymmetric
1FS first flapwise symmetric
1PR first propeller mode
1BE first edgewise
2FA second flapwise antisymmetric
2FS second flapwise symmetric
1TI first tower in plane
1TO first tower out of plane
2FA second flapwise antisymmetric
2FS second flapwise symmetric
Table 2. Key for the mode shape
abbreviations

American Institute of Aeronautics and Astronautics
9
Figure 9. Comparison of the rotationally augmented frequencies for the first three flapwise blade modes as
predicted by the simulation codes and as measured in the SNL 34m experiment
C. Stress distribution due to gravitational forces
To validate the distribution of steady beam element forces and bending moments, calculated from QBLADE, a
simulation without aerodynamic forces was carried out. After convergence was obtained the bending moments were
converted into stresses on the blade surface based on the simple formula for bending stresses in isotropic beams.
Figure 10 shows the comparison of the calculated stresses with measured and simulated SNL data. The predicted
stresses match those from the SNL calculations and measurements.
Figure 10. Blade stress distribution due to gravitational forces; SNL measurements and calculations against
QBLADE predictions

American Institute of Aeronautics and Astronautics
10
VI. Aeroelastic simulations of two IEC design load cases
To benchmark its applicability to real life simulation scenarios, the application of newly developed structural
model in QBLADE in two design load cases (DLC) from the IEC 61400-1 certification standard is demonstrated. The
motivation is twofold; the capability of the structural model to produce accurate results for structural loads shall be
confirmed in the DLC 1.3 (Power Production) simulation and the ability to predict aeroelastic instabilities will be
demonstrated based on a simulation from DLC 6.1. Furthermore, exemplary runtimes will be given at the end of this
section to give the reader an estimation for the computational cost associated with the aeroelastic simulations.
A. DLC 1.3: Power production with the extreme turbulence model (ETM)
An aeroelastic simulation of one time series from DLC 1.3 is carried out with QBLADE and compared to a “quasi”
aeroelastic simulation conducted with the structural model in ANSYS. As ANSYS is not directly coupled to an
aerodynamic solver the aerodynamic loads, obtained from a simulation of a rigid rotor in QBLADE, are prescribed to
the structural model. If deflections are relatively small and no aeroelastic instabilities occur, the results between the
fully coupled aeroelastic- and the prescribed loads simulations should agree reasonably well, creating a simple
validation case for the load predictions from QBLADE. One 630s simulation from DLC 1.3 with a constant rotational
speed of 34rpm and a mean inflow velocity of 20m/s with turbulence generated after the ETM (Class II), which is
expected to create significant structural loads, is compared in the following graphs. Only a time series of 200s is shown
for clarity.
Figure 11 shows the wind speed, measured at the height of the blade midspan. The resulting axial forces at the
mid-span node of blade 1, as computed by ANSYS and QBLADE, are shown in Figure 12. In force magnitude
QBLADE predicts a slightly larger oscillation than ANSYS. The matching in phase is remarkable, which is an
indication that in this particular case the prescribed loads approach with ANSYS can adequately reproduce the results
from the aeroelastic model. A histogram of the force oscillations is shown in Figure 14.
Figure 11. Windspeed measured at the height of rotor midspan; generated with the extreme turbulent wind
model (ETM) with a mean velocity of 20m/s
Figure 12. Axial element force acting on blade 1 midspan element, from QBLADE and ANSYS

American Institute of Aeronautics and Astronautics
11
FFigure 13 shows the deflection of the blade midspan position along the vertical global z-axis (aligned with the
tower). Under the highly turbulent inflow conditions the mid-span element undergoes oscillations with highly varying
amplitudes. A close observation shows that there is reasonable matching in the results from ANSYS and QBLADE
with only small differences in phase and magnitude. This observation can be confirmed when looking at the histogram
that is shown in Figure 14.
Overall, for this limited case, the calculations with ANSYS could reasonably confirm the capability of QBLADE
to predict elemental forces and displacements in unsteady aeroelastic simulations.
Figure 13. Z deflection of blade 1 midspan element, from QBLADE and ANSYS
Figure 14. Histogram of axial force (left) from QBLADE and ANSYS; histogram of z-displacement (right)
from QBLADE and ANSYS
B. DLC 6.1 & DLC 6.3: Parked (locked) turbine with the extreme wind speed model (EWM)
In a recent publication of Galinos et al. [5, 25] an aeroelastic instability
was observed for a parked and locked VAWT at some azimuthal positions.
It was also proclaimed that due to these oscillations DLC 6.1 & 6.3 could
be a future design driver for VAWT. Especially because these findings
were “…considered to be quite uncertain as the aerodynamics may not be
valid in this particular simulation” (from [5]), this present a very
interesting case, not only to demonstrate the capability of QBLADE to
predict aeroelastic instabilities but also to confirm the finding of Galinos
et al. with a different aerodynamic model (the AC model was used by
Galinos).
Galinos found the instabilities for positions of the locked rotor around
the 20° azimuthal rotor positions. Simulations with QBLADE were
carried out around this position with both turbulent and steady inflow
conditions representing the extreme wind speed model (EWM for Turbine
Figure 15
. Definition of azimuthal
rotor positions

American Institute of Aeronautics and Astronautics
12
class II) with 50-year recurrence period. The numerical damping has a large effect on this particular simulation. For
all subsequent simulations, the Rayleigh parameter was tunes to obtain a 3% logarithmic decrement for the first two
mode shapes.
Figure 16 shows the z-displacement of the blade 1 midspan node. The parked rotor was simulated with turbulent
EWM50 inflow conditions. An aerodynamic time step of 0.01s was used and 5 different azimuthal positions (10°, 15°,
20°, 25° and 30°) were investigated. It can be seen that the simulations at 20° and 15° exhibit significantly larger
oscillations of the blade mid span which could be an indication for an aeroelastic flutter instability. Although not
shown here larger oscillations in this case are also associated with larger structural blade loads.
Figure 16. Displacement of mid span node during simulation of locked, parked rotor with EWM_50 Class
II turbulent inflow
To reduce the damping effect from turbulent fluctuations and to isolate the fluctuations a second set of simulations
has been carried out were the turbulent inflow conditions were replaced with steady laminar inflow after the formula:
()= 1.4 (
). (1)
A reference velocity of 42.5m/s was used for wind turbine class II. From the results in Figure 17 it is apparent that
an aeroelastic flutter instability causes the blade to oscillate heavily for the 20° case of the locked rotor. This confirms
the observation made by Galinos. For the other two investigated rotor positions no flutter instability could be observed.
Although not shown here, the simulations with steady inflow were repeated for a range of different inflow velocities.
The oscillating behavior could be detected for velocities as low as 30m/s.
Figure 18 shows the oscillation of the rotor during the flutter instability for one cycle. The snapshots were taken
from the steady simulation case. From these snapshots, the instability occurs at the first symmetric blade flapwise
mode. That the same mode is causing the oscillations in Figure 16 can be deducted when looking at the amplitude
spectra of the oscillating deflections in turbulent wind that is shown in . A clear a peak at a frequency of around 1Hz
can be seen, which is also the first flapwise eigenfrequency of the rotor.
Figure 17. Displacement of mid span node during simulation of locked, parked rotor with EWM_50 Class
II steady inflow

American Institute of Aeronautics and Astronautics
13
Figure 18. The cycle of one oscillation for the instability observed in steady inflow case (after Eq. 1)
For an additional comparison, the simulations of
the parked rotor at 20° azimuth with turbulent
inflow were repeated with the ANSYS model.
Figure 20 compares the results of the aeroelastic
QBlade simulations with the “quasi” aeroelastic,
prescribed loads model in ANSYS (that was also
applied to DLC 1.3 above). As there is no direct
aeroelastic coupling in the ANSYS model it is not
capable of capturing the, potentially critical, flutter
instability. This comparison highlights the
importance of using a fully coupled aeroelastic tool
in the certification process.
Figure 20. Displacement of mid span node during simulation of locked, parked rotor with EWM_50 Class
II turbulent inflow at 20° azimuthal position; comparison between QBLADE and ANSYS prediction
C. Exemplary runtime
All QBLADE simulations have been carried out on a single workstation with an Intel Core i7-5930K CPU @
3.50GHz, 32GB of RAM and a GeForce GTX1070 GPU.
The ratio between simulated real time and the wall clock time of the simulation is also a function of the rotational
rate of the rotor. For the aerodynamic calculation at 34rpm, each rotation of the rotor is discretized into steps of 5°
azimuthal angle (0.025s). The structural computations are discretized with a time step size of 0.001s. For the case of
the SNL 34m testbed, rotating at 34rpm an aeroelastic simulation of 630s took 3337s with the above specs, which
yields a ratio of ~1/5 between simulated real- and wall clock time.
Figure 19. Amplitude spectra of the z-
displacements
simulated with EWM_50 inflow

American Institute of Aeronautics and Astronautics
14
VII. Conclusion
An aeroelastic coupling between the LLFVW aerodynamic formulation and an Euler-Bernoulli beam multibody
code, based on the open-source PROJECTCHRONO library was realized in QBLADE. An aerodynamic vortex
method for VAWT was applied in the aeroelastic simulation of IEC DLC for the first time in this work. Validation of
the aerodynamic and structural sub-models, and their combination, produced promising results for the small number
of validation cases that were investigated.
For future work it is planned to extend the validation study and to compare results with a fully coupled aeroelastic
simulation tool such as HAWC2. Furthermore, work is already underway to implement generator models and turbine
controllers into the simulation to provide a tool with the full aero-servo-elastic capabilities that is needed for thorough
VAWT design and certification, according to the IEC standards.
Appendix
As the geometric and structural properties for the SNL 34m testbed are given in imperial units in the original
publications (the conversion was rather tiresome and prone to misinterpretations) the most important properties that
were used to set up the aerodynamic and the structural model are given here in SI units for convenience.
Table 3. Blade geometry and airfoils
HEIGHT CHORD RADIUS AIRFOIL
m m m -
0,00 1,22 0,00 NACA 0021
0,51 1,22 0,78 NACA 0021
5,54 1,22 8,53 NACA 0021
6,02 1,07 9,11 SAND 0018/50
6,68 1,07 9,89 SAND 0018/50
7,37
1,07
10,63
SAND 0018/50
8,08 1,07 11,36 SAND 0018/50
8,81 1,07 12,06 SAND 0018/50
9,57
1,07
12,74
SAND 001
8/50
10,35 1,07 13,39 SAND 0018/50
10,94 1,07 13,86 SAND 0018/50
11,59
0,91
14,26
SAND 0018/50
12,61 0,91 14,84 SAND 0018/50
13,67 0,91 15,34 SAND 0018/50
14,75 0,91 15,77 SAND 0018/50
15,87 0,91 16,13 SAND 0018/50
17,00 0,91 16,41 SAND 0018/50
18,16 0,91 16,61 SAND 0018/50
19,32 0,91 16,73 SAND 0018/50
20,49 0,91 16,77 SAND 0018/50
21,66 0,91 16,74 SAND 0018/50
22,82 0,91 16,62 SAND 0018/50
23,97 0,91 16,42 SAND 0018/50
25,11 0,91 16,15 SAND 0018/50
26,23 0,91 15,80 SAND 0018/50
27,32 0,91 15,37 SAND 0018/50
28,37 0,91 14,87 SAND 0018/50
29,05 1,07 14,53 SAND 0018/50
29,68 1,07 14,09 SAND 0018/50
30,50
1,07
13,50
SAND 0018/50
31,31 1,07 12,88 SAND 0018/50
32,09 1,07 12,23 SAND 0018/50
32,85
1,07
11,56
SAND 0018/50
33,58 1,07 10,86 SAND 0018/50
34,30 1,07 10,14 SAND 0018/50
34,82
1,22
9,58
NACA 0021
41,29 1,22 0,91 NACA 0021
41,88 1,22 0,13 NACA 0021

American Institute of Aeronautics and Astronautics
15
Table 4. Structural properties of the blades
BlFract AerCtr StrTwt BMassDen
FlpStff EdgStff GJStff EAStff
[-] [-] [°] [kg/m] [N/m] [N/m] [N/m] [N]
0,00 0,25 0,00 99,30 1,534E+07 2,710E+08 1,538E+07 2,534E+09
0,01
0,25
0,00
99,30
1,534E+07
2,710E+08
1,538E+07
2,534E+09
0,13 0,25 0,00 99,30 1,534E+07 2,710E+08 1,538E+07 2,534E+09
0,14 0,25 0,00 57,00 4,884E+09 1,140E+08 5,133E+09 1,454E+09
0,16 0,25 0,00 57,00 4,884E+09 1,140E+08 5,133E+09 1,454E+09
0,18 0,25 0,00 57,00 4,884E+09 1,140E+08 5,133E+09 1,454E+09
0,19 0,25 0,00 57,00 4,884E+09 1,140E+08 5,133E+09 1,454E+09
0,21 0,25 0,00 57,00 4,884E+09 1,140E+08 5,133E+09 1,454E+09
0,23 0,25 0,00 57,00 4,884E+09 1,140E+08 5,133E+09 1,454E+09
0,25 0,25 0,00 57,00 4,884E+09 1,140E+08 5,133E+09 1,454E+09
0,26 0,25 0,00 57,00 4,884E+09 1,140E+08 5,133E+09 1,454E+09
0,28 0,25 0,00 45,30 2,914E+09 6,760E+07 3,179E+09 1,156E+09
0,30 0,25 0,00 45,30 2,914E+09 6,760E+07 3,179E+09 1,156E+09
0,33 0,25 0,00 45,30 2,914E+09 6,760E+07 3,179E+09 1,156E+09
0,35 0,25 0,00 45,30 2,914E+09 6,760E+07 3,179E+09 1,156E+09
0,38 0,25 0,00 45,30 2,914E+09 6,760E+07 3,179E+09 1,156E+09
0,41 0,25 0,00 45,30 2,914E+09 6,760E+07 3,179E+09 1,156E+09
0,44 0,25 0,00 45,30 2,914E+09 6,760E+07 3,179E+09 1,156E+09
0,46 0,25 0,00 45,30 2,914E+09 6,760E+07 3,179E+09 1,156E+09
0,49
0,25
0,00
45,30
2,914E+09
6,760E+07
3,179E+09
1,156
E+09
0,52 0,25 0,00 45,30 2,914E+09 6,760E+07 3,179E+09 1,156E+09
0,55 0,25 0,00 45,30 2,914E+09 6,760E+07 3,179E+09 1,156E+09
0,57
0,25
0,00
45,30
2,914E+09
6,760E+07
3,179E+09
1,156E+09
0,60 0,25 0,00 45,30 2,914E+09 6,760E+07 3,179E+09 1,156E+09
0,63 0,25 0,00 45,30 2,914E+09 6,760E+07 3,179E+09 1,156E+09
0,65
0,25
0,00
45,30
2,914E+09
6,760E+07
3,179E+09
1,156E+09
0,68 0,25 0,00 45,30 2,914E+09 6,760E+07 3,179E+09 1,156E+09
0,70 0,25 0,00 57,00 4,884E+09 1,140E+08 5,133E+09 1,454E+09
0,71 0,25 0,00 57,00 4,884E+09 1,140E+08 5,133E+09 1,454E+09
0,73 0,25 0,00 57,00 4,884E+09 1,140E+08 5,133E+09 1,454E+09
0,75 0,25 0,00 57,00 4,884E+09 1,140E+08 5,133E+09 1,454E+09
0,77 0,25 0,00 57,00 4,884E+09 1,140E+08 5,133E+09 1,454E+09
0,79 0,25 0,00 57,00 4,884E+09 1,140E+08 5,133E+09 1,454E+09
0,81 0,25 0,00 57,00 4,884E+09 1,140E+08 5,133E+09 1,454E+09
0,82 0,25 0,00 57,00 4,884E+09 1,140E+08 5,133E+09 1,454E+09
0,83 0,25 0,00 99,30 1,534E+07 2,710E+08 1,538E+07 2,534E+09
0,99 0,25 0,00 99,30 1,534E+07 2,710E+08 1,538E+07 2,534E+09
1,00 0,25 0,00 99,30 1,534E+07 2,710E+08 1,538E+07 2,534E+09
Table 5. Structural properties of the tower (tower height = 50m)
TFract TMassDen FlpStff EdgStff GJStff EAStff
[-] [kg/m] [N/m] [N/m] [N/m] [N]
0 353,43 9,16E+09 9,16E+09 6,80E+09 8,21E+09
1 353,43 9,16E+09 9,16E+09 6,80E+09 8,21E+09
Table 6. Structural properties of a single guy cables (3 sets of 2 cables are used)
Metallic Area Cable Stiffness Cable Pretension Angle with Ground
[m^2] [N/m] [N] [°]
2,23E-03 7,97E+06 4,14E+05 35
Acknowledgments
The present work would not have been possible without the PROJECTCHRONO library and the whole team
behind this project. The authors would also like to express their gratitude for the great help that was experienced on
the PROJECTCHRONO website forum from A. Tasora, D. Negrut and others. Under no other circumstances did we
ever get such fast and accurate answers to all our numerous questions.

American Institute of Aeronautics and Astronautics
16
References
[1] H.J. Sutherland, D.E. Berg, T.D. Ashwill, A Retrospective of VAWT Technology, SANDIA Report, SAND2012-0304, New
Mexico, USA, 2012
[2] U. Schmidt Paulsen, H. Aagaard Madsen,P.H. Nielsen, K.A. Kragh et al. “DeepWind. From idea to 5 MW concept” 2014.
EERA DeepWind 2014 - 11th Deep Sea Offshore Wind R&D Conference, Trondheim, Norway, 2014
[3] F.J. Savenije “S4VAWT; Floating vertical axis wind turbines with pitched blades” ECN Report, ECN-M—16-049, Presented
at Winddays 2016, Rotterdam, Netherlands, 2016
[4] M. Collu, F.P. Brennan, M.H. Patel “Nova Project: Lessons Learnt During the Conceptual Phase of the Design of a Floating
Support Structure for an Offshore Vertical Axis Wind Turbine” Ocean Renewable Energy: Ocean Space utilization Vol. 7,
2016
[5] C. Galinos, T.J. Larsen, H.A. Madsen, U.S. Paulsen, Vertical axis wind turbine design load cases investigation and comparison
with horizontal axis wind turbine, 13th Deep Sea Offshore Wind R&D Conference, Trondheim, Norway, 2016.
[6] A. Jain, T. Hewitt, J. Spossey, M. Hudon, Challenges in Performance and Safety Certification of Small-scale Vertical-axis
Wind Turbines, 4th World Summit for Small Wind, Husum Germany, 2013.
[7] D. Marten, M. Lennie, G. Pechlivanoglou, C. N. Nayeri and C. O. Paschereit, Implementation, Optimization and Validation
of a Nonlinear Lifting Line Free Vortex Wake Module Within the Wind Turbine Simulation Code QBlade, J. Eng. Gas
Turbines Power 138(7), 2015
[8] H. Aagaards Madsen, T.J. Larsen, U. Schmidt Paulsen, L. Vita: “Implementation of the Actuator Cylinder Flow Model in the
HAWC2 code for Aeroelastic Simulations on Vertical Axis Wind Turbines”, Proceedings of 51st AIAA Aerospace Sciences
Meeting including the New Horizons Forum and Aerospace Exposition, USA, 2013
[9] I. Paraschivoiu, Double-Multiple-Streamtube Model for Studying Vertical-Axis Wind Turbines, Journal of Propulsion and
Power, Vol.4, No.4, pp.370-377, 1988.
[10] K.R. Dixon: “The Near Wake Structure of a Vertical Axis Wind Turbine”, PhD thesis, Technical University of Delft, 2008
[11] J.C. Murray, M. Barone: “The Development of CACTUS, a Wind and Marine Turbine Performance Simulation Code”, 49th
AIAA Aerospace Sciences Meeting including the New Horizons Forum and Aerospace Exposition, USA, 2011
[12] Aeroelastic Modeling of Large Offshore Vertical -axis Wind Turbines: Development of the Offshore Wind Energy Simulation
Toolkit, 54th AIAA Structures, Structural Dynamics and Materials Conference, Boston, USA, 2013
[13] D. Marten, M. Lennie, G. Pechlivanoglou, C.N. Nayeri, C.O. Paschereit: “Nonlinear Lifting Line Theory Applied To Vertical
Axis Wind Turbines: Development of a Practical Design Tool”, Presented at ISROMAC 2016, 2016
[14] D. Marten, G. Pechlivanoglou, C.N. Nayeri, C.O. Paschereit: ”Integration of an unsteady nonlinear lifting line free wake
algorithm in a wind turbine design framework”, EWEA 2015 Annual Event and Exhibition, 2015
[15] A. Van Garrel: “Development of a wind turbine aerodynamics simulation module”, ECN Report, ECN-C-03 079, 2003
[16] D. Marten, A. Bianchini, et al., Effects of Airfoils’ Polar Data in the Stall Region on the Estimation of Darrieus Wind
Turbine Performance, GT2016-56685, ASME Turbo Expo 2016, Seoul, South Korea, 2016.
[17] J. Wendler, D. Marten, G. Pechlivanoglou, C.N. Nayeri, C.O. Paschereit, Implementation and Validation of an
Unsteady Aerodynamics Model for Horizontal and Vertical Axis Wind Turbines Within the Simulation Tool
QBlade, GT2016-57184, ASME Turbo Expo 2016, Seoul, South Korea, 2016
[18] L. Bergami, M. Gaunaa: “ATEFlap Aerodynamic Model, a dynamic stall model including the effects of trailing edge flap
deflection”, Risø DTU Report, Risø-R-1792(EN), 2012
[19] D. Marten: “QBlade v0.95 Guidelines for Lifting Line Free Vortex Wake Simulations”, Technical Report, TU Berlin, 2016
[20] A. Tasora, R. Serban, H. Mazhar, A. Pazouki, D. Melanz, J. Fleischmann, M. Taylor, H. Sugiyama, D. Negrut: “Chrono: An
Open Source Multi-Physics Dynamics Engine”, Springer’s Lecture Notes in Computer Science, 2016
[21] A. Tasora: ”Euler-Bernoulli corotational beams in Chrono::Engine”, Chrono::Engine technical documentation, 2016, available
from http://www.projectchrono.org/assets/white_papers/FEA/euler_beams.pdf, last accessed 09.12.2016
[22] D.E. Berg, Structural Design of the Sandia 34-Meter Vertical Axis Wind Turbine, SANDIA Report SAND84-1287, 1985.
[23] W.A. Stephenson: “Test Plan for the 34 Meter Vertical Axis Wind Turbine Test Bed Located at Bushland Texas”, SANDIA
Report, SAND86-1623, 1986
[24] B. Roscher, Structural Optmization Of A Vertical Axis Wind Turbine With Aeroelastic Analysis, Masters of Science Thesis,
Technical University of Denmark, 2014.
[25] C. Galinos: “Study of Design Load Cases for Multi-Megawatt Onshore Vertical Axis Wind Turbines”, Masters Masters of
Science Thesis, Technical University of Delft, 2015.
[26] T.D. Ashwill: “Measured Data for the Sandia 34-Meter Vertical Axis Wind Turbine”, SANDIA Report SAND91-2228, 1992.
[27] T.D. Ashwill: “Initial Structural Response Measurements and Model Validation for the SANDIA 34-Meter VAWT Test Bed”,
SANDIA Report, SAND88-0633, 1990