You are on page 1of 69

Graduation report

Flexible wind turbine blades:


a FEM-BEM coupled model approach
David Verelst, January 2009

1
2
Summary
The final Master project at the Delft University of Technology (TU Delft) emphasizes on writing a thesis
during a period of 9 months. This document is the result of that process. The project proposal was
formulated by Alex De Broe, an engineer at the renewable energy consultant 3E. Consequently, this thesis
work was performed at 3E HQ in Brussels.

The global scope of the investigation considers the modelling of flexible wind turbine blades. Due to the
increase in blade size and ever present push for better blades (in any way), less stiff blade design are
(re-)entering the horizon of engineers. It is argued that due to the complexity of a flexible blade design,
current modelling techniques have to be adapted or at least reviewed against new designs. In order to be
able to capture the complex deformations of flexible blades (which in this case also includes structural
couplings due to manipulated fibre lay-up strategies in the composite glass fibre epoxy material of the
blade), a finite element analysis is required. For the aerodynamic modelling, the blade element and
momentum theory is expected to give satisfactory results under fast computation times. Computational
fluid dynamics is considered too time consuming (with respect to both computation times and modelling)
if to be considered in an engineering friendly environment which focuses on design and analysis of
multiple cases.

For this thesis work, the finite element program Abaqus is coupled to the blade element momentum
theory program WT_Perf. An iterative approach assures that structural deformations and consequently
changing aerodynamic forces (and vice versa) are in balance so the aeroelastic response of the flexible
blade is captured correctly. The existing AOC 15/50 blade and 4 variations have been modelled in Abaqus
by researchers at the University of Leuven. They are conveniently re-used for this investigation. The
aerodynamic calculations require lift, drag and moment coefficients at different angles of attack for all the
relevant airfoil sections of the blade. In many cases wind tunnel data is used to this end, for the AOC
15/50 blade those are not available however. XFOIL was then used to obtain these. Later on in the project
it became apparent that XFOIL overestimated in this case some crucial parameters of the aerodynamic
coefficients (stall and post stall behaviour), thanks to some valuable discussions with James Tangler, the
original designer of the considered airfoils.

In a previous research project, which involved 3E, the University of Leuven and the Brussel based
aerodynamic consultant Numeca, the Abaqus models were coupled with Numeca's CFD solver Fine/Hexa,
through the coupling interface MpCCI of the Fraunhofer SCAI. The results were used to benchmark the
previously mentioned coupling between Abaqus and WT_Perf, created for this research topic. It is
concluded that the deformations caused by the Abaqus-WT_Perf tandem, are significantly overestimated
with respect to the Abaqus-CFD. Since the considered flow conditions resulted in a blade under full stall, it
is a difficult comparison: stalled rotor flow is challenging to predict correctly, for both CFD solvers and the
combination XFOIL/WT_Perf in this case. Results show that a flexible blade with complex geometry
(sweep) and structural couplings significantly influences blade performance and loading. It is thus
confirmed that modelling techniques for wind turbine blades should take the full structural and
geometrical complexity of the blade.

It is concluded that with respect to a blade design cycle, the coupling between a finite element package
and blade element and momentum code under steady and static conditions can be useful. Especially when
a tight integration between this coupled approach and a dynamic simulation tool could be established,
more advanced flexible blade designs are ready to be analysed for a new generation more flexible wind
turbine blades.

3
4
Acknowledgements

This thesis has been written under supervision of Alex De Broe (3E) and Thanasis Barlas (TU Delft). I would
like to thank them both for their patience and valuable inputs throughout the entire process. Also the final
support and feedback of Prof.dr.ir. G.J.W. van Bussel is greatly acknowledged.

During these months I have had interesting discussions with some TU Delft staff members. I greatly
appreciated the general talks on aeroelasticity with Roeland De Breuker. The working environment at the
3E HQ in Brussels was motivating and enriching. From time to time enjoyable unexpected social events
would pop-up. I had a great time in Brussels.

It sounds like a cliché, but it is true. The support of close family and friends, as always, have been quite
important during the sometimes quite stressful process of writing this thesis. Thank you so much for that.

5
6
Table of Contents

Summary..................................................................................................................................................................................3
Acknowledgements...........................................................................................................................................................5
Nomenclature........................................................................................................................................................................9
1 Introduction......................................................................................................................................................................11
2 Literature review...........................................................................................................................................................12
2.1 Introduction.....................................................................................................................12
2.2 Bend-twist coupled blades to increase energy capture...................................................12
2.3 Behavioural aspects of bend-twist coupled blades..........................................................13
2.4 Obtaining bend-twist couplings.......................................................................................15
2.4.1 Off-axis fibres in the composite lay-up................................................................15
2.4.2 Swept blade planform...........................................................................................17
2.5 Modelling flexible blades.................................................................................................18
2.5.1 Introduction..........................................................................................................18
2.5.2 Load application...................................................................................................19
2.5.3 The “Maheri” approach: efficient power curve calculations.................................20
2.6 Conclusions......................................................................................................................21

3 Modelling flexible wind turbine blades: procedures, techniques and scope...............................22


3.1 Introduction and scope....................................................................................................22
3.2 Aerodynamic modelling...................................................................................................23
3.2.1 Blade Element Momentum theory........................................................................23
3.2.2 Lifting line, panel and vortex methods................................................................24
3.2.3 Computational Fluid Dynamics............................................................................24
3.3 Structural modelling........................................................................................................25
3.3.1 Introduction..........................................................................................................25
3.3.2 Modal representation...........................................................................................25
3.3.3 Multi Body System (MBS)......................................................................................25
3.3.4 Cross-sectional modelling of blades for a MBS....................................................26
3.3.5 Finite Element Method..........................................................................................26
3.4 Solution strategies for a flexible wind turbine blade......................................................27
3.5 Structural data for a MBS analysis....................................................................................27
3.6 Concluding: a FEM-BEM coupled approach......................................................................28

4 Modelling flexible wind turbine blades: coupled FEM-BEM approach..............................................29


4.1 Introduction.....................................................................................................................29
4.2 General procedure...........................................................................................................29
4.3 Blade Element and Momentum theory.............................................................................30
4.4 Finite Element Method.....................................................................................................32
4.5 Software components......................................................................................................32
4.5.1 Commercial FEM package Abaqus (Simulia, Dassault Systems)..........................32
4.5.2 Python...................................................................................................................33
4.5.3 WT_Perf.................................................................................................................33
4.6 Program outline and coupling procedures......................................................................34
4.6.1 Introduction..........................................................................................................34
4.6.2 Detailed program work flow.................................................................................34
4.6.3 Coupling scheme..................................................................................................34
4.6.4 Transferring parameters between WT_Perf and Abaqus model...........................35
4.6.5 Program files structure.........................................................................................38
4.7 Simulation cases..............................................................................................................39
4.7.1 Blade structural model.........................................................................................39
4.7.2 Profile aerodynamic data......................................................................................41
4.7.3 Simulation conditions...........................................................................................41

7
4.7.4 Sensitivities...........................................................................................................42
4.8 Simulation results and comparison with CFD-FEM coupling...........................................42
4.9 Conclusions and and recommendations.........................................................................50

Annex A Program flowcharts .....................................................................................................................................52


Annex B AOC 15/50 aerodynamic blade data and BEM model verification........................................55
Annex C AOC 15/50 HAWC2 model.........................................................................................................................60
Annex D AOC 15/50 structural blade data..........................................................................................................62
Annex E WT_Perf model and configuration.........................................................................................................63
Annex F Computation times........................................................................................................................................65
References...........................................................................................................................................................................66

8
Nomenclature

Latin symbols

Ncrit amplification factor


fax aerodynamic force component perpendicular to the rotor plane
ftan aerodynamic force component in the rotor plane
R rotor radius
r local section radius
cl, cd, l, d 2D section lift- drag coefficients and forces
dT, dQ total section thrust and torque
CP, CM, CT total rotor pressure-, moment- and thrust coefficients
a axial induction factor
a' tangential induction factor
F Prandtl tip correction factor
T, Q, P total rotor thrust, torque and power

Greek symbols

α angle of attack
Φ inflow angle
θr section twist angle
Ω rotational speed

Abbreviations

ac aerodynamic centre
API Application Programming Interface
BEM Blade Element Momentum theory
CFD Computational Fluid Dynamics
DOF Degree Of Freedom
FE Finite Element
FEA Finite Element Analysis
FEM Finite Element Method
GL Germanischer Lloyd
HAWT Horizontal Axis Wind Turbine
LE Leading edge
MBS Multi Body System
NREL National Renewable Energy Laboratory, USA
RPM Rotations per minute
TE Trailing edge
TUD, TU Delft Delft University of Technology

9
10
1 Introduction
Until recently, wind turbine blades had a relative high rigidity and small deformations. This allowed for
modelling techniques which assumed a simplified aeroelastic response. The established methodologies
regarding aeroelasticity used in the traditional aerospace sector could be applied in this case. However,
ongoing research and further increase in blade length, revealed that a more flexible blade imposes new
and different challenges in the areas of modelling, design, construction and operation. Recent reports
have shown that an aeroelastic optimised flexible blade can offer a number of advantages over the more
rigid variant: higher energy yield and/or shedding loads (increasing fatigue life).

To further investigate flexible blades, different modelling techniques and design methodologies are
required when compared with the traditional approach. Non linearities regarding geometrical (e.g. large
deformations), structural (e.g. structural couplings, non constant stiffness matrix) and aerodynamic (e.g.
stall, dynamic inflow) parameters have to be considered.

For this thesis, three goals are formulated:

• to identify modelling techniques for flexible wind turbine blades

• to give a brief overview of how modelling techniques relate to software tools and design
procedures

• to create a coupled FEM-BEM approach, calculate the aeroelastic response and compare the
results obtained from a previously coupled FEM-CFD study

The thesis report is structured as follows:

• chapter 2 Literature review: overview of relevant articles on flexible, bend-twist coupled wind
turbine blades, selected based on following key questions: why, how do they behave, how to
obtain bend-twist couplings and how to model them correctly

• chapter 3 Modelling flexible wind turbine blades: procedures, techniques and scope: describing
the relation between modelling techniques, modelling procedures, software tools and the place
in a general wind turbine design cycle

• chapter 4 Modelling flexible wind turbine blades: coupled FEM-BEM approach: the technical
details about the coupling procedure, between a FEM program (Abaqus) and a BEM code (NREL's
WT_Perf). Discussion of the results and a comparison with the results from a previous FEM-CFD
coupled study

• In Annex A: Program flowcharts , the program structure is outlined in detail. Annex B: AOC
15/50 aerodynamic blade data and BEM model verification elaborates on the results of the
calculated aerodynamic coefficients and the comparison of the WT_Perf BEM code with HAWC2.
Annex C: AOC 15/50 HAWC2 model reviews some basics results of the HAWC2 model and in
Annex D: AOC 15/50 structural blade data, the structural properties of the different fibre lay-ups
applicable for the Abaqus model are considered. The WT_Perf configuration is outlined in Annex
E: WT_Perf model and configuration. Finally Annex F: Computation times discloses some
information regarding the required computation times for the coupling program

11
2 Literature review

2.1 Introduction
In order to better understand and place the research area into perspective, a literature review is
conducted. A selection of relevant papers to the author is shortly reviewed. This investigation served as a
basis and starting point for several decisions made during the research project.

2.2 Bend-twist coupled blades to increase energy capture


[5] presents a review (dated 1998) of several articles considering aeroelastic blades on constant and
variable-speed stall-regulated machines. It states that it would be difficult at best to attain perfect power
regulation by passive bend-twist feathering. The analogy with helicopter blades is investigated in
literature. The conclusion states that general aeroelastic tailoring techniques are similar, although design
and mission objectives of a wind turbine and a helicopter do differ significantly.
A more detailed review of the steady effects on an aeroelastic blade, reveal an increased average power
generation (reported energy gains around 10%) for a constant speed, stall-controlled machine with a twist
to stall aeroelastic blade. It is explained that the blade will stall at lower wind speeds and therefore will
have a lower rated power. As a result, larger blades for the same machine can be deployed leading to
higher energy capture in the lower wind speed regions, having the same rated power level when compared
to a stiff blade.
For the variable speed, stall regulated machine deployed with twist to stall blades, the question is raised
whether a passively bend-twist coupled blade can replace an active pitch system. Only a 4-5% percent
energy capture increase is mentioned for an aeroelastic coupled blade (with respect to stiff blades).
Results for flap-twist coupling were almost identical to centrifugal-twist coupling in this application.

In [6] an existing stiff blade of a constant-speed, stall-regulated machine is modified into a more flexible
one, incorporating an aeroelastic twist-to-stall behaviour. Maximizing the energy capture was the objective
in this study by changing the pre-twist distribution, rotor radius and structural coupling properties. For
the modified blade, an increased energy capture of 15% is reported (depending on site conditions, see
figure 1). Figure 2 shows the original and optimized pre-twist distribution. More information on the
modelling approach will be explained in more detail in chapter 2.5.3 on page 20. It was concluded that
bend-twist coupled blades can increase the energy capture of a wind turbine, if a thorough and automated
(evaluating a large number of options) design procedure is followed.

Figure 1: Energy capture enhancement at different Figure 2: Pre-twist distribution of a flexible and
site average wind speeds [6] rigid blade [6]

12
A different approach in [7] was considered. The main objective was increasing the energy capture of an
existing blade design for a variable speed, variable-pitch machine. A swept blade which twists to feather is
deployed to this end. For the conceptual design of the swept planform, conclusions stated by [8] are based
upon the load shedding requirement of a 4° to 7° induced twist at the blade tip. The lift coefficients of the
swept blade were changed (due to the presence of the sweep angle) and the chord and pre-twist
distributions optimized with respect to the peak power coefficient. The structural design was then based
on simplified extreme and operating loads. The boundary conditions where set by the maximum bending
deflections (cfr tower clearance). At this point an optimization effort for the sweep geometry was
performed. The probability density function of the expected peak flap bending moments for several sweep
geometries where plotted versus the normalised flap moment (see figure 3 of [7]). It was concluded that
by displaying data in this manner, a certain sweep shape could be selected based on optimal load
reductions. Conclusions regarding the power output stated that the load shedding due to twist to
feathering, allowed a larger rotor (56m instead of the original 50m) while the loading increased only
slightly. As a result, the energy capture in below rated wind conditions was increased (5% to 10%), while
the power rating and load envelope remained identical.

Figure 3: Histogram of flap bending moment for 10 min 10m/s turbulent simulation; BASE6 indicates the
elongated 56m blade without sweep, exp2 and exp3.3 are two sweep variants of the 56m blade and the
50-m indicates the original 50 meter blade [7]

2.3 Behavioural aspects of bend-twist coupled blades


In [9] the concept of aeroelastic blades is explored by using a simple FE (Finite Element) model of the
Combined Experiment Blade (CEB) rotating at 72 rpm. The publication expresses the amount of coupling
between bending-twisting (bending due to thrust loading of the rotor) and extension-twisting (extension
due to centrifugal loading) as α and theoretically ranges between -1 (twisting towards feather) and 1
(twisting towards stall).

The influence of the coupling terms on several aspects are evaluated below in figure 4 (as done by [9]).
These aspects include: extension-twist (A), bend-twist (B), flap-twist distribution for bend-twist coupling
coefficient of α = -0.8 (C), natural frequencies (D), aeroelastic stability boundaries (flutter and divergence)
(E) and damping coefficients (F). The results in figure 4 present an indication of the influences of changing
elastic coupling configurations. The aeroelastic loading terms are based on Theodorson's equations for lift
and moment (linear airfoil theory, blade stall is not considered here) on an oscillating airfoil in terms of its
plunging and pitching motions (bending and twisting for HAWT blade).

13
For the bend-twist case, an induced blade tip twist between -3° and 10° for varying coupling coefficients
between -0.8 and 0.8 (figure 4, B) is found. It is concluded that these induced twist angles could be useful
for enhanced energy capture, passive power control or load alleviation. However, the induced twist angles
obtained from an extension-twist coupling (ranging from 0.1° to 1.0° twist at the tip, figure 4, A) are
considered not sufficient to this end. For the stability analysis, the conclusion states that the flutter and
divergence boundaries (figure 4, E) where always above the rotor design speed. Further it is mentioned
that the divergence limits are much more critical and closer to the operating range than the flutter
boundaries.

Figure 4: Plots for different responses and stability related issues made in [9]

14
2.4 Obtaining bend-twist couplings

2.4.1 Off-axis fibres in the composite lay-up

When the fibre orientations are manipulated to obtain a bend-twist coupling, the stiffness properties of the
composite material are altered. The stress in the laminate tend to follow the fibre direction, causing a
resultant shear component which is responsible for creating a torsional moment while under bending. This
principle is outlined in [10] and is illustrated figure 5.

A FE model is employed by [10] to compare some structural lay-outs and the obtained bending-torsion
coupling. Based on considerations of possible failure mechanisms typical for certain configurations and
local stress concentrations, alternative structural blade concepts are considered. These suggestions are
summarized in figure 7 from [10], together with advantages and disadvantages regarding that
configuration. The coupling is obtained by adding off-axis fibres to the spar cap (see figure 6). It should
be noted that the remaining blade has a soft skin, in order not to increase the blade torsional stiffness
again. It was concluded that the induced twist of the conventional coupled configuration (two skin shells
with two spar webs, no ribs) would be sufficient to increase the energy output.

Figure 5: Fibre-orientation in conventional (a) and bending-torsion (b) coupled blade [10]

15
Figure 6: Blade layouts as used in [10]

Figure 7: Comparison between structural blade design concepts [10]

[11] and [12] considered several concepts with identical cross sections (see figure 8), but the spar caps
where composed out of different fibre lay-up strategies (seven in [11] and nine in [12] ) resulting in a
bend-twist coupled behaviour. For practical applications, it is noted that the off-axis fibres should not
extend to the leading and trailing edges, since it would introduce too large strains at those locations. By
using a FE model, each configuration is loaded based on a standardized loading case representing typical
peak flapwise bending conditions, applied as a tip load. The spar cap thickness was iteratively adjusted
until certain design strain levels at the midspan where achieved. Material and manufacturing costs where
considered for all the evaluated cases as well in order to assess the price of the attained coupling.

Figure 8: Blade configuration for [11] and [12]

16
2.4.2 Swept blade planform

For a swept blade planform, the tip of the blade is moved backwards in the plane of rotation. In doing so,
the aerodynamic loading at the swept blade parts will induce a moment around the pitch axis. When the
blade is swept back (against the direction of rotation), this will introduce a pitch-to-feather moment. If the
blade is swept forward (same direction as the rotation) it will introduce a pitch to stall moment. See also
figure 9 from [7].

Figure 9: The swept blade concept, from [7]

The concept of a swept blade to induce blade twist was assessed in [8]. Starting point was an existing 30m
blade which was then redesigned with a curved planform (sweep). In a first iteration, the outboard blade
thickness was reduced (i.e. lower torsional rigidity) and the spar cap thickness increased to keep the blade
flapwise bending stiffness constant. Leaving length, planform, and airfoil selections unchanged. It
appeared that the original blade had already a close to the maximum spar cap thickness of 10% airfoil
thickness (imposed boundary condition by the authors). As a result, the airfoil thickness reduction was
limited and the induced twist was only 0.9° at the tip.

Figure 10: Glass vs carbon spar cap construction comparison at station 65% r/R [8]

It was deduced that the initial circular arc sweep curve produced a lot of twisting moment on the inboard
part of the blade (which was torsionally rather stiff) and much less at the outboard part of the blade (with

17
a lower torsional rigidity). When a power law shaped bend curve with an exponent of 4 was used, the outer
part of the blade experienced a larger twisting moment. The circular arc sweep curve produced an induced
tip twist of 1.2°, while the power law shaped sweep curve (with exponent 4) produced nearly 4° of tip
induced twist. A further increase in induced tip twist was achieved by replacing the glass fibres in the spar
caps with carbon, starting from the tip till 75%, 55% and 35% r/R blade span positions. In doing so, the
airfoil thickness could be reduced further while retaining flapwise bending stiffness (see figure 10).
Results are summarized in figure 11 (as extracted from [8]).

Weighted Tip Max Blade Carbon to


Twist Twist Shear Weight Tip Weight
(deg) (deg) (psi) (lbs) (lbs)
AllGlass 2.28 3.90 1974 6020
Carbon 75%-> 2.48 4.70 1974 5906 122
Carbon 55%-> 3.20 6.15 2364 5715 432
Carbon 35%-> 4.03 7.30 2842 5279 1035

Figure 11: Comparing different variants of swept blade with a sweep curve exponent of 4 [8]

In all cases, the loading in [8] was assumed to have a linear changing thrust distribution with zero thrust
at the blade root and maximum thrust at the tip. However, for a tip twist of 7°, about 2/3 of the section lift
would be gone, and this approximation is clearly no longer accurate. To keep the results of this study on a
basis where comparisons could be easily made and understood, it was decided to retain the same loading
distribution for all twist calculations. [8] concludes that for other follow up studies the loading should be
corrected for the effect of induced structural twist.

2.5 Modelling flexible blades

2.5.1 Introduction

A flexible wind turbine blade introduces a coupled problem: the deformations of the blade influences the
aerodynamic loading and vice versa; so it can truly be called an aeroelastic problem. The field of
aeroelasticity is not new, in the aerospace industry this field is already covered for over decades now. In
the wind turbine industry the aeroelastic response of more flexible and smart blades where considered
only recently.

To capture the aeroelastic response with a model correctly, following aspects have to be taken into
account: aerodynamic forces, structural deformations and the coupling between both of them. The
coupled behaviour can be captured by solving uncoupled aerodynamic and structural equations separately
and iteratively till certain convergence criteria are met (staggered approach) or by solving both the
coupled aerodynamic and structural equations at the same time. The relevance of such an iterative
coupling (changing aerodynamic loads due to a deforming blade) is illustrated in [13] where for a blade
with certain flexibility, the performance for both a single step (SS) simulation (no iterations between
aerodynamic loads and structural deformations) and a coupled aero structure approach (CAS) is showed
(see figure 12). The difference between SS and CAS is significant and so for a flexible blade, a CAS
procedure should be followed.

18
Figure 12: Comparing Single Step (SS) versus Coupled Aero Structure approach (CAS) [13]

Since blades are made of composites, the structural deformation can behave in a non-linear way and the
material properties can be anisotropic. By changing for instance fibre layer orientations, a structural
coupling between bending and torsion can be obtained. To capture the behaviour (including structural
couplings) of many different untested and unknown configurations of flexible composite blades, only 3D
FEM models can produce satisfactory results (according to [13]).

Computation time is in engineering applications an important factor. When considering wind turbine
blades, it is likely that many iterations of a single design will follow in order to find an optimum. Further,
wind turbine simulations are also required to be run in the time domain. Certification bodies stipulate
simulation cases spanning several minutes in turbulent wind fields. If for instance fatigue damage would
be a design driver, a time domain simulation is required for many design variations. This further stresses
the importance of modelling techniques and short computation times.

2.5.2 Load application

The BEM code only delivers 2-dimensional loads (lift, drag and aerodynamic moment) acting at the
aerodynamic centre of a certain profile section. In [20] the comparison was made between the results of a
FEM blade model loaded by a pressure distribution (pressure applied per element of the FE model) and a
discrete sectional BEM loading. The concluding remarks stated that for detailed stress/strain analysis the
pressure distribution delivers much better results. However, almost no difference was noted with regard to
the general deflections. So to compute the correct aeroelastic response, it can be stated (based on [20])
that a discrete 2D BEM loading will have a marginal effect on the displacements (see figure 14). It should
be noted that the considered blade had no significant structural couplings.

Figure 13: Actual and discretized system of BEM loading on profile [13]

In [13] all the nodal points of the selected profile are loaded, based on the actual and known pressure
distribution, as can be seen in figure 13. The approach discretises the pressure distributions over the
available nodes of the profile section, in which the original 2D lift and drag forces remain equal to the
summed discretized ones. The resultant aerodynamic moment is not necessary the same in both cases, so
the difference is then distributed over the available nodes in order to assure a correct loading.

19
Figure 14: Deflection plots for 2 loading cases (discrete vs pressure distribution) [20]

2.5.3 The “Maheri” approach: efficient power curve calculations

Maheri's modelling approach ([6], [13], [29], [30] and [28]) emphasises the development of a fast and
efficient FE model of the blade, coupled to a BEM code (staggered coupling approach). Constant wind
speeds are considered (static simulations only). For every part of the code (meshing, load application,
convergence accelerators, element choice), its efficiency was considered. On top of this, a generic
algorithm was constructed to perform optimisations with respect to performance. A FE model specifically
for wind turbine blades was written from scratch, in order to construct an accurate and state of the art
analysis, however faster than currently possible with a commercial FEM package.

The FEM-BEM coupled code is called WTAB (see [13]) and can be summarized as follows (as in [29]): “using
a force-adaptive semi-automatic mesh generator, applying a convergence accelerator in the BEM-based
aerodynamic load predictor and employing a sophisticated element based on natural mode FE
formulation”. It is considered beyond the scope of this document to elaborate on the exact
implementation of the FE methodologies.

When considering the performance of a wind turbine, the aeroelastic response for every wind speed along
the power curve should be known. If for every wind speed a FEA is to be made, computation time can be
significant. In [29], a method is presented to reduce the required time to calculate the performance of a
wind turbine with flexible blades. A 95% reduction is reported when comparing to a full FEA for all wind
speeds, without a significant loss in accuracy (see figure 15). Some important assumptions used in this
combined analytical/FEA modelling approach are the following:

• edgewise slope of the blade is assumed small compared to flap-wise slope

• twist aerodynamic moment (due to off axis aerodynamic loading of the blade) is small compared
to the torque produced by the bend-twist coupling

20
Figure 15: Relative difference between full FEA and the combined analytical/FEA approach. Reference
wind speed for the combined FEA/analytical analysis: left graph 10 m/s, right graph 14 m/s [29]

In order to further decrease calculation time, the FEA can be completely replaced by a simplified analytical
model of the normalised induced twist [30]. An important assumption used is that the normalised
stiffness K* can be represented by a material and geometrical part:

K *≈ K *Material K *Geometry

In which the material dependent stiffness is only a function of the airfoil thickness. Further, it is assumed
that the mechanical properties of the blade material (for instance fibre orientation, shell thickness, spar
cabs) are not varying span wise.

2.6 Conclusions
From the literature review, some important conclusions can be drawn:

• geometrical non-linearities (e.g. large instead of small angle deflections) have to be taken into
account

• torsional degree of freedom (DOF): a flexible blade should not be considered torsionally stiff

• structural non-linearities: structural couplings have to be accounted for

• structural non-linearities: the tangential stiffness matrix of the composite material will not
remain constant when larger deformations are considered

• passive bend-twist coupled blades can increase energy capture and/or reduce fatigue damage

• passive bend-twist couplings are structurally feasible by introducing a swept blade planform or
off-axis fibres in the laminate

• accurate modelling of the fluid-structure interactions are required: large deformations will
influence the aerodynamic forces considerably

• Blade Element Momentum theory (BEM) enhanced with engineering corrections (e.g. tip loss,
dynamic inflow, dynamic stall) remains a fast but reasonable accurate aerodynamic code for wind
turbine rotors

• a swept blade will have an influence on lift and drag forces, the aerodynamic coefficients should
be corrected for that

21
3 Modelling flexible wind turbine blades:
procedures, techniques and scope

3.1 Introduction and scope


In wind turbine engineering, usually two types of analyses (with respect to the blades) are considered.
First, a reduced model is solved in the time or frequency domain to capture all relevant behavioural
aspects, stability issues and to determine the loads on the turbine (dynamic simulation). Secondly,
detailed (or full) FE models of the blades are constructed and loads from the aeroelastic simulations are
transferred to the FE model for detailed structural analysis. Usually, the detailed FE model is built in a later
stage of the design process, since it requires a higher level of detail.

In the dynamic simulation, the fluid-structure interactions, inertial and gravitational forces are taken into
account. Since many load cases are considered, computation time has to be short: a BEM code and a
reduced structural model (modal approach or beam element) is usually used at this stage. For flexible
blades it is important that the structural element is capable of capturing the deformations accurately,
since a deformed blade will have a significant influence on the aerodynamic loading.

A FEA tackles the detailed structural aspects: locate stress/strain concentrations, evaluate different failure
criteria, perform local optimisations etc. After converting the reduced structural model (as used in the
dynamic analysis) to a full 3D FE model, some benchmarks or tests could be useful to assure both models
have the same characteristics. The (extreme) loads as obtained in the dynamic simulations, have to be
translated to the FE model. If the dynamic analysis software tool provides external loads, this is a
straightforward process. If not, a recovery analysis based on the relevant output has to be performed.

It is possible that after a FEA, a new design iteration will require a new dynamic analysis. In that case, it is
also important to have a good conversion algorithm between the FE model and the reduced model (beams,
mode shapes) as well. Again, both structural models should have comparable global characteristics.

In figure 16, relations between modelling techniques and analysis are outlined. Currently, different
software tools cover only partially or entirely the modelling steps and procedures of figure 16. An
interesting subject for further investigation could be the comparison between different modelling
procedures with respect to both modelling efforts and accuracy of obtained analysis results.

Chapters 3.2 and 3.3 (Aerodynamic modelling and Structural modelling) are based on several remarks as
stated in the overview paper of [14].

22
Figure 16: Conceptual representation of different blade modelling procedures. Red box is a model, grey
box an analysis type, blue box output of an analysis

3.2 Aerodynamic modelling

3.2.1 Blade Element Momentum theory

For wind turbine rotors, the Blade Element Momentum theory (BEM) offers a fast, simple but quite robust
way to simulate the aerodynamic forces on the blades. Enhanced engineering corrections are applied to
increase its accuracy. The method's accuracy depends upon the knowledge of the aerodynamic
characteristics of the deployed airfoils, most often obtained in wind tunnel tests.

Advantages:

• computation time is very short, suitable for dynamic time domain simulations

• simple model, yet accurate and verified under different circumstances

Disadvantages:

• accurate aerodynamic wind tunnel data of airfoil profiles required (CL, CD, CM as function of angle
of attack, ranging from -180° to 180°)

Some basics concerning BEM theory is explained in chapter 4.3 Blade Element and Momentum theory.

23
• highly 3-dimensional flow characteristics have to be accounted for by engineering corrections,
especially when a stalled flow is considered. These corrections vary according to the deployed
airfoil and/or correction model

3.2.2 Lifting line, panel and vortex methods

Panel and vortex methods can offer a slightly higher accuracy over the BEM method [15][16][17] at the
cost of a higher computational effort, both in terms of cpu power and programming effort. They have
been developed in an attempt to obtain a more detailed description of the 3D flow that exits around a
wind turbine and the influence it has on its performance. Panel and vortex methods could also take into
account wake information of nearby turbines or other objects disturbing the flow. Questions around these
more advanced aerodynamic codes still remain. For engineering applications, their usefulness is not yet
conclusive.

Advantages:

• a higher accuracy over BEM in some cases

• more accurate representation of the 3D flow

• suitable for dynamic time domain calculations

Disadvantages:

• increased model complexity and computational requirements

• inviscid flow assumption, stalled flow is not included in the model

3.2.3 Computational Fluid Dynamics

Computational Fluid Dynamics (CFD) offers a very powerful yet computationally expensive tool to simulate
and capture the complete aerodynamic behaviour of a wind turbine rotor. Currently, CFD calculations are
not a widespread practice when designing a wind turbine. Certainly in a design and optimisation cycle
(where numerous load cases for different configurations are simulated), it is impractical because the
aerodynamic calculations for one case are too time consuming. It is mentioned by [18], that a viable
solution of an aeroelastic performing rotor (using a FEM and CFD code in a staggered and iterative way),
needs tedious tuning of the numerical parameters for the CFD solver. At the moment, these are dependent
on the structural behaviour and flow parameters. So apart from the huge computational effort to perform
a CFD analysis, fine tuning of the model and thus CFD expertise is required.

Advantages:

• simulation method, requiring only geometric information of the blade

• full 3D flow of the rotor is considered

Disadvantages:

• computationally intensive, not suitable for dynamic time domain calculations

• numerical fine-tuning makes it less practical for engineering applications

24
3.3 Structural modelling

3.3.1 Introduction

Some aeroelastic codes available today for wind turbine blade simulation, assume small deformations,
which is not advisable when considering flexible blades and corresponding larger displacements. [19]
performs a detailed study on the effects of modelling with and without the small deformations
assumption. It is concluded “that non-linearities due to large blade deflections are present and must be
considered in the design of very light and flexible wind turbines”.

3.3.2 Modal representation

A model of the blade, constructed with assumed mode shapes enables to calculate the structural
response. The accuracy depends on the quality and number of the assumed mode shapes. It will require
real life blade testing or advanced modelling techniques to obtain accurate mode shapes of a blade
structure. If only standardised and known blade structural layouts are used (for which the mode shapes
are determined easily or are known from real life experiments), this is a fast and robust modelling
technique. For more complex structures, is quite hard to obtain an accurate modal representation when
no data of experiments exists.

Advantages

• simple and fast modelling technique

• suitable for dynamic time domain calculations

Disadvantages

• for each component, the mode shapes have to be found by either more complex modelling
techniques or real life experiments

3.3.3 Multi Body System (MBS)

From the perspective of the complete wind turbine, a multi body representation is often applied to
account for all the relevant components (gearbox, yaw and pitch mechanisms, blades, tower etc). The
components are split up in several bodies (either rigid or flexible), subjected to the appropriate boundary
conditions fitting them into the complete configuration. Although more complex than a modal approach,
the MBS is less computational expensive as a FEA.

Advantages

• robust technique to model the structural dynamics of an entire wind turbine

• computational effort is acceptable

• suitable for dynamic time domain calculations

Disadvantages

• for each component or sub-assembly, a suitable MBS representation has to be constructed,


requiring an extra layer of complexity (see next paragraph)

25
3.3.4 Cross-sectional modelling of blades for a MBS

In [1], several computer tools for generating cross-sectional data (inertial and structural properties) for
wind turbine blades are assessed. By modelling only certain cross-sections of the blade, the required
modelling time is reduced significantly when compared to creating a full 3D FE model. The structural
details required for the FE model are in many cases only available late in the design phase. It is argued
that dynamic multi-body simulations of a wind turbine, can deliver results which are comparable to a 3D
FE analysis in terms of global behaviour and point-wise 3D stress/strain distributions [2]. Further, [1]
compares and benchmarks different cross-sections in PreComp, VABS, FAROB, CROSTAB and BPE. The
conclusion states that only VABS consistently provides good results for the different assessed cross-
sections. The author also point out that more extensive validation of modelling techniques and software
tools is required in the wind industry.

Advantages

• fast structural modelling technique when compared to full 3D FE model

• computational effort is acceptable

• suitable for dynamic time domain calculations in combination with a MBS

Disadvantages

• not all available software tools deliver satisfactory results, model validation could be an issue

3.3.5 Finite Element Method

When considering complex structures such as flexible composite blades with bending-twist couplings, a FE
model can account for the relevant behaviour and non linearities. Computation efforts are considerable.
As a result, it is impractical to include the entire wind turbine (gearbox, blades, generator, bearings)
within a single FE model.

Advantages

• high fidelity technique which can capture the complex (non linear) behaviour of a complex
composite structure, only requiring the geometric and material properties

Disadvantages

• computation time and model construction efforts can be considerable

• not suitable for dynamic time domain calculations

Some basics concerning FEM theory is explained in chapter 4.4.

26
3.4 Solution strategies for a flexible wind turbine blade
When dealing with an aeroelastic problem, both the structural and aerodynamic behaviour have to be
modelled. Two possible solution techniques are:

• Solve aerodynamic forces and structural deformations separately (staggered approach)

• Solve aerodynamic forces and structural deformations at the same time

A possible advantage of the staggered approach is that two existing codes could be coupled to each
other. Especially when considering mature and well developed commercial software products, this can be
an advantage. However, in some cases, programming such an iterative coupling procedure between an
aerodynamic and structural code can be very complex. If a closed source program is chosen (which is true
for many commercial FE packages), the source code is not available to change some parts of it to facilitate
a coupling procedure with another program.

3.5 Structural data for a MBS analysis


As pointed out by reference [1] (see chapter 3.3.4, generating a set of accurate structural blade
parameters for a MBS analysis can be difficult. Different software tools and modelling techniques can,
depending on the blade structure, give quite different results. However, [1] also mentions that defining a
blade at certain cross-sections only, requires less modelling time when compared to a FE model of an
entire blade. Therefore, it could be beneficial to use this approach in an early design and optimisation
phase.

Since both a reduced and full structural model will exist at a certain point in the design phase, the FE
model can be used to calibrate or validate the reduced model. Due to the uncertainty of the quality of the
reduced model, this verification process can not be ignored for the flexible rotor concept. As a result, the
complexity of the blade modelling procedure is further increased: dynamic analysis, cross-sectional
parameter generation (reduced structural model), extract external loads from the dynamic analysis for
extreme load verification with FE (static), FE analysis and conversion/validation between reduced and FE
model (see also figure above 16).

Although such a modelling procedure could be a powerful tool in a design and optimisation scheme, it is
considered beyond the scope for this research topic.

27
3.6 Concluding: a FEM-BEM coupled approach
For this study, a coupled FEM-BEM approach is pursued, in order to be able to investigate how significant
the blade deformations are as a result of structural couplings and a swept blade planform.

As mentioned before, a FE program is best suited to accurately capture the response of a flexible blade. By
coupling a BEM code iteratively to such a FE model, the static behaviour of a flexible blade under an
aerodynamic load can be determined. Following benefits are applicable to such an approach:

• no complex interactions between full and reduced structural model to determine the static loads
and estimate the static performance

• optimize blade structure both locally (stress concentrations, failure criteria) and globally (power
curve, maximum deflections for tower clearance, blade weight, eigenfrequencies)

• the “Maheri approach” (as briefly explained in chapter 2.5 on page 20) can be used to further
minimize computation time to estimate the power curve

• accurate response (no possible loss of accuracy due to converting process) and detailed
structural assessment of the blade, no need to model other wind turbine parts

• coupling 2 existing codes should minimize programming efforts on the condition that both
programs can easily exchange the required variables

However, following limitations apply:

• no dynamic simulations (will lead to an excessive amount of computation time)

• no fatigue damage assessment due to time-dependent loads

• reducing the FE model for the dynamic analysis is required, although it is expected less iterations
between both of them are required due to previous design iterations

• computation time could be an issue

• detailed blade model has to be constructed early in the design phase

• interactions with other wind turbine parts are not simulated

A FEM-BEM coupling could also be used to verify the reduced structural model used in the dynamic
simulations. An other key element is that since the design space of flexible blade is very large, it seems
useful to be able to isolate the blade design procedure from the rest of the turbine in an early stage.

28
4 Modelling flexible wind turbine blades:
coupled FEM-BEM approach

4.1 Introduction
The behaviour of a flexible blade can be obtained by using a structural and aerodynamic solver in a
staggered approach: both are solved independently from each other. As established in the previous
chapters, the aerodynamics and structural deformations have mutual influences. Therefore it is required
to have a coupling between both solvers: a deforming blade will influence the aerodynamic forces, the
updated aerodynamic forces will change the deformations again etc. This process will continue until
convergence to a final deformed state (when operating below the unstable divergence area), where the
changing aerodynamic forces due to a new deformed blade state are below a certain threshold
(convergence criteria).

For this thesis, the results from a previous study involving a coupling between Abaqus and the CFD code
FINE/Hexa are available [35] and will be used as a benchmark. This study was performed by Numeca (a
software developer and solution provider for fluid dynamics simulations), the Catholic University of Leuven
(KUL) and 3E (renewable energy consultant). The set-up included a FE model of the AOC 15/50 blade
created by the KUL in the FE software package Abaqus, coupled trough the code coupling interface MpCCI
of the Fraunhofer SCAI [49]. The available data features 5 Abaqus blade models (see chapter 4.7.1), which
are loaded by the pressure forces from the final FEM-CFD coupling step on each element of the model.

4.2 General procedure


Abaqus and WT_Perf have quite different model definitions. WT_perf's blade model is defined by some
simple parameters (see chapter 4.5.3). Furthermore, the WT_Perf output simply consist of a 1-dimensional
distributed aerodynamic loading (in axial and tangential directions). The Abaqus model on the other hand
is more complex: a mesh covering the exact geometry of the blade composed of around 11,000 nodes.
Forces and moments are placed on the nodes or elements (= a collection of 4 nodes forming a quasi
rectangle) and have a 3-dimensional position, direction and magnitude.
So in order to let WT_Perf and Abaqus communicate with each other, the simple 1D blade model existing
in the BEM environment has to mapped onto a detailed 3D mesh.

The most simple representation of the coupling between WT_Perf and Abaqus can be represented as
follows:

1. Obtain chord length, twist distribution and coning angle from the Abaqus model

2. Create WT_Perf input (text file) which corresponds to the Abaqus model

3. Run WT_Perf analysis

4. Translate WT_Perf output (spanwise thrust and torque distribution) to multiple discrete loads
with corresponding position, direction and magnitude on the 3D blade mesh

5. Run Abaqus analysis

29
6. Obtain chord length, twist distribution and coning angle from the Abaqus analysis

7. Check if twist angles are converged, if not go back to step 2

WT_Perf can only be made aware of some effects of the deformed blade: a changed twist distribution and
a flap-wise bending approximated by a straight line under an angle (which is the coning angle). Therefore
the convergence criteria can only be based on the twist angle (a change in twist angle is the significant
parameter affecting a change in aerodynamic loads, see figure 27). Coning angle has only a marginal
effect on the output, as can be seen in figure 28, see chapter 4.7.4.

4.3 Blade Element and Momentum theory


This chapter will briefly explains the very basic theory behind blade element and momentum theory.

BEM is actually the combination of blade element (also known as strip theory) and momentum theory.
Blade element theory divides the blade into discrete 2D sections, for which the aerodynamic lift and drag
forces per unit length are calculated based on local values of pitch angle, angle of attack, chord length,
airfoil section lift/drag coefficients, induction and wind speed. Note that the wind speed is the vectorial
sum of the free stream velocity and the rotational induced velocity. Further, the aerodynamic coefficients
of the 2D airfoil section have to be known as function of angle of attack.
The momentum theory relates rotor thrust to the induction over the rotor plane. The induction could be
interpreted as the change in wind speed conditions due to the presence of the lift and drag generating
rotor blades.
When combining both blade element and momentum theories, an iterative process starts in which rotor
thrust has to be equal according to both theories, that is an induction factor has to be found that results
for both theories in the same rotor thrust. An example of such an iterative process is outlined in fig. 17.

Since a BEM approach has many limitations, some corrections are used to improve the accuracy of the
approach. An example of such a correction factor is the Prandtl tip correction (which is also applied in the
example flowchart in figure 17). It accounts for the aerodynamic losses near the blade tip, due the
complex 3D interactions in the flow.

For a more comprehensive explanation and the derivation of the equations, see for instance the general
wind energy handbooks [36] and [37].

30
Calculate tangential and axial forces for all radial stations

Angle of relative wind START


r = R0
=arctan
U 1−a 0

r 1a ' 0   Initial estimate
for a0 and a'0

Effective angle of attack


eff =−r
and corresponding cl from
2D airfoil data
c l  eff 

Radial station blade lift and drag Prandtl tip correction


l =c l 12 V 2eff c d =cd 12 V 2eff c d=
2 R U 1−a 
B  R2U 2 1−a 2
corresponding thrust (inc tip corr) R−r
2 −
d
dT =F l cos d sin   r F r = cos−1 e 

and torque (inc tip corr)


dQ=F l sin −d cos  r  r

no Next step,
the same for a'
Update tangential induction factor
a 0a1
dQ r=r  r a new=
a '1= r R 2
4  r 3  U 1−a 0  r 
yes a 0=a new

Moments and forces


r= R r=R
T = ∑ dT Q= ∑ dQ P=Q
r=R 0 r= R0
no
yes
a 1≈a 0 END
Corresponding coefficients
P
C P= 1 3
 U  R−R0 2
2

Q Glauert correction for heavily


C Q=
1 loaded turbines and solve for a1
U 2  R−R0  2 R
2
CT =4 a1 1−a 1 a0 ≤1/ 3
T
C T= 1 2 5−3 a 1
 U  R−R02 C T =4 a 1 1− a 1 a0 1/ 3
2
4

Figure 17: Example flowchart of BEM code. The grey box is repeated until all discrete 2D radial airfoil
sections are calculated. This part represents the blade element theory. The next two steps are summing
up already known discrete 2D forces for the blade element theory and non-dimensionalize them. Now with
the known value for the thrust coefficient (CT), the induction can be calculated according to the
momentum theory (box indicated with the Glauert correction in the lower right corner).
R: rotor radius, r: local section radius, cl, cd, l, d: 2D section lift- drag coefficients and forces, dT, dQ:
total section thrust and torque, CP, CQ, CT: total rotor power-, torque- and thrust coefficients, a: axial
induction factor, a': tangential induction factor, F: Prandtl tip correction factor, T, Q, P: total rotor thrust,
torque and power, Φ: inflow angle, θr: section twist angle, Ω: rotational speed

31
4.4 Finite Element Method
The finite element method refers to a numerical technique to solve partial differential or integral
equations. In engineering, the term FEM or FEA is often used to indicate the numerical analysis of
structures. The numerical structural analysis is much more complex when compared to the relative simple
BEM theory as explained in the previous chapter. The same holds for the corresponding software packages
used for this research project: Abaqus (the FE package) is vastly more complex when compared to
WT_Perf.
Some simplified basic concepts applicable for a FEA are only be treated briefly in this chapter. One could
break down a FEA into following steps:

1. Definition of the objects geometry

2. Meshing the geometry, dividing the geometry into very small but finite elements, defined by
nodes. Since different element types can be considered (triangles, rectangles, cubes, shells,...),
this step involves the need to know which element is most suitable for the considered geometry
and analysis with respect to accuracy and computation time. For a thin walled wind turbine
blade, shell elements are often used

3. By collecting all the relations between the different nodes (or elements so to say) in matrix form,
the master stiffness equation can be constructed:
f=Ku
where f is a vector containing the nodal forces, K is a stiffness matrix and u is the nodal
displacement vector.

4. Numerically solve assembled master matrix equations and corresponding boundary conditions.
The set-up and

5. Post processing of the results to obtain the general displacements, strains and stresses.

In many software packages available today, the applied methodology for steps 3 and 4 are classified as
the direct stiffness method. For a more comprehensive introduction to FE methodologies, see for instance
[38] and [39].

4.5 Software components

4.5.1 Commercial FEM package Abaqus (Simulia, Dassault Systems)

Since the blade model was already available in Abaqus, re-using this model eliminates the need to remodel
or export/import the blade (and corresponding issue's) into another FE package. By keeping the structural
solver and model the same in both the CFD and BEM coupled simulations, differences due to coupling
strategies and aerodynamic modelling can be evaluated. As a result, Abaqus was selected as preferred FE
environment.

Besides the practical issues mentioned in the previous paragraph, Abaqus is a powerful FE package.
Especially for a complicated and flexible blade structure, Abaqus offers several relevant modelling
options:

• Non-linear analysis: the composite blade is expected to encounter large deformations, contains
structural couplings and could have a changing tangential stiffness matrix

32
• Shell elements for efficient modelling of a composite, thin-walled, blade structure. Fibre
orientations of the different plies can be easily defined and manipulated

• Integrated Python API (Application Programming Interface): allows the user to completely
manipulate Abaqus models and initiate analysis all from the Abaqus Python script command line.
As a result, a Python script can act as the bonding element between Abaqus and any other code
or program

Abaqus version 6.8 is used during this investigation.

4.5.2 Python

Python is a high level - open source - interpreted - language, designed to have a clear and readable syntax.
It is platform independent (runs on Windows, Linux, MAC OSX and others) and has an extensive user base
and developers community which is involved in maintaining, improving and documenting the project.

Due to its open architecture and the many optional modules available, Python can be fully customized to
perform efficiently specific tasks. Numpy and SciPy are examples of optional Python modules (or libraries)
which upgrade the Python stack to Matlab-Maple alike environments: 2D and 3D plotting capabilities,
complex matrix and vector operations, symbolic solvers etc. These optional Python libraries are
extensively used in the coupling procedure which is developed for this research topic.

Python version 2.4 [46] is used in Abaqus v6.8. Since the Python environment in Abaqus is not completely
the same as a normal stand alone Python installation, a workaround has to be set up in order to install the
optional required modules (NumPy [47] and SciPy [48]). This problem can be solved by installing a
standard Python environment (having the same version as the Abaqus integrated Python environment) and
putting the optional modules directory into the appropriate OS environment variable. In doing so, all extra
modules installed in the standard Python installation are also available in the Abaqus Python environment.

4.5.3 WT_Perf

WT_Perf [41] is a BEM code, capable of predicting the aerodynamic loads on the rotor blades. NREL made
the source code freely available and it can be downloaded as a-ready-to-use windows executable. The
simple WT_Perf ASCII (text) input and output makes it easy and straightforward to let Python format the
input, read its result and translates it to a specified Abaqus model.

Some of WT_Perf characteristics (see also Annex E or reference 41):

• Static wind field only (no flow dynamics nor turbulence) and it can include vertical wind shear

• Includes tip and hub loss corrections, inclusion of swirl effects and skewed-wake correction

• Blade sections are defined by their aerodynamic properties, chord length, position relative to the
centre of rotation and local twist angle. The blade is thus defined as a straight line, no edge-wise
or flap-wise variations possible.

• The rotor position has a coning, tilt and yaw angle.

• Output data includes thrust and torque per unit length along the blade. Although cm (airfoil
pitching moment coefficient) can be included in the airfoil data and the local flow conditions are
outputted as well, the airfoil aerodynamic pitching moment per unit length is not included in the
output.

33
4.6 Program outline and coupling procedures

4.6.1 Introduction

A general overview of the program has been given above in chapter 4.2. In this chapter a more detailed
explanation is presented.

4.6.2 Detailed program work flow

The structure of the program is visualized in the flowcharts in Annex A.

The mentioned function correct_nodeset_array_reduced() is used to account for some apparent


inconsistencies in the nodal coordinates as outputted by Abaqus. When comparing the leading edge,
trailing edge and loading point coordinates of the deformed blade structure, it was noticed that some
nodes where completely out of place and not located in between neighbouring points, as expected. The
correction function replaces these bogus points by interpolating between two correct neighbouring nodes.
However, it is possible that the sorting routine is responsible for this error. It could be caused by a slightly
different organised mesh database in Abaqus between the undeformed model and deformed output
databases. It is recommended to reassess the sorting function in a second programming iteration.

4.6.3 Coupling scheme

In Abaqus, the concept of analysis steps are defined and allows to brake down an entire analysis into
subsequent parts. The starting point of one step is the model state at the previous step. As a result, the
updated loads can be applied on the deformed model of the previous step. Further, Abaqus allows to stop
an analysis, update the model (e.g. create new step and update the aerodynamic loads in a newly created
step) and restart the analysis at a predefined step (e.g. the newly created step).

Three different couplings schemes are outlined in figure 18. For this study, coupling scheme A (see figure
18) is used: the updated aerodynamic forces are applied on the deformations caused by the aerodynamic
forces from the previous iteration. The centrifugal forces are applied at the beginning of the first iteration.
Optionally, these can be updated after the aerodynamic forces are converged. Due to stability problems
with the Abaqus-Python interface, this feature could not be used.
If however the centrifugal forces have a significant impact on the deformed blade (due to the aerodynamic
forces), coupling scheme B or C might be more appropriate. It is expected that scheme's B and C will
require more iterations and it is difficult to say beforehand how the convergence process will behave.
Coupling scheme's B and C are not implemented in the code and are as such not evaluated for this
research topic.

34
Figure 18: Sequential coupling schemes

4.6.4 Transferring parameters between WT_Perf and Abaqus model

The WT_Perf blade model consists out of radial segments, defined by chord length, twist angle and airfoil
characteristics (see also Annex E: WT_Perf model and configuration). In Abaqus, the blade is modelled with
shell elements, consisting out of a little more than 11,000 nodes. As a result, the spanwise load
distribution of the aerodynamic centre as outputted by WT_Perf, has to be translated to specific nodes on
the real blade structure. In the Abaqus blade model node sets are created which group the leading edge,
trailing edge, loading points, blade sections, etc.

a) WT_Perf sections for the Abaqus model

Since the mesh of the Abaqus model is well structured, organizing it in correspondence with a valid set of
WT_Perf sections is straightforward (see also further below in figure 28). Each profile can be defined by a
series of mesh nodes with almost the same spanwise x-coordinate (corresponding to a 2D airfoil profile).

The Abaqus airfoil sections are created according to the mesh nodes: each section consists out of a
predefined number of of spanwise mesh nodes (see figure 19). The next step is to calculate the RElm
(WT_Perf definition for an elements radial position, see Annex E) value's for each section. Note that the
RElm value's do not necessarily coincide with a mesh node position, although it would be located close to
one.

35
Figure 19: Matching WT_Perf's and Abaqus' grids

b) Chord length, aerodynamic centre and twist angle

In Abaqus, the section parameters, chord length and twist angle, are calculated based on coordinates of
the leading and trailing edge of specific section (see figure 20).

Figure 20: Determining aerodynamic centre, chord length and twist angle for the Abaqus model

c) Load application and moment correction

Since the aerodynamic centre is not a real loading point in the Abaqus model (hollow blade section), the
following translating strategy is considered in the program:

• One loading point per section (above the sparweb) for the axial and tangential forces. Since the
aerodynamic forces are now translated from the aerodynamic centre to the loading point, a

36
correction moment has to be applied in order compensate for the changed aerodynamic moment
around the aerodynamic centre (see figure 21). The correction moment is then defined as:

mcorr = f ax  y − f tan  z −1 [Nm]

Note that in figure 21, Δy is negative and Δz positive.

Figure 21: Aerodynamic loading in Abaqus (loading point) and its relation with the aerodynamic centre

• The aerodynamic moment is in both cases distributed over all available section nodes. The
correction moment is added to the aerodynamic moment and distributed over all section nodes.
The aerodynamic moment is defined positive in the pitch up direction, which coincides with the
the positive moment direction around the spanwise x-axis in the Abaqus model.

The aerodynamic moment is not outputted by WT_Perf. However, the moment coefficients are available in
the airfoil files, the local inflow angle and wind speed is known from the WT_Perf output. With the given
effective angle of attack, the corresponding value of cm can be looked up from the airfoil data tables. As a
result, the aerodynamic moment per unit length can be calculated as follows:

1
maero=  V 2res c 2 c m [N]
2

d) Coning angle

WT_Perf allows to set a constant blade coning angle. The flap-wise bending deflection of a flexible blade
could results in an increased coning angle. A simple least-square fit on the spanwise distribution of the
loading points is used to determine the coning angle. For large deformations this is not an accurate
representation for the blade bending, since the coning angle in that case is expected to be changing
significantly along the blade.

Although small coning angles have a limited influence on for instance the thrust force on the rotor (see
figure 28), the code takes it into account since implementation was almost trivial.

e) Units in Abaqus

The units in Abaqus are not defined. Instead, the defined parameters (mesh node coordinates, material
stiffness and mass properties, forces and moments) should be consistent. The current Abaqus model has
following input units:

37
Abaqus input SI conversion
length [mm] 10-3 [m]
Material constants (E, G) [Pa] [Pa]
density [g/mm3] 106 [kg/m3]

As a result, the units for force, moment and pressure in Abaqus are:

Abaqus input SI conversion


pressure [Pa] [Pa]
2
force [g mm/s ] 10 [kg m/s2] = 10-6 [N]
-6

moment [g/s2] 10-9 [kg/s2] = 10-9 [Nm]

These conversions have to be taken into account when passing on data between the WT_Perf and Abaqus
representations of the blade models.

4.6.5 Program files structure

The program is constructed around 3 files:

• coupling.py is the main file: run from the Abaqus Python command line. The relevant functions
from abaqus_modules.py are structured and called from here.

• abaqus_modules.py: grouping all the different functions in one file

• parameters.py: configuration file (Abaqus model name, twist convergence criteria, maximum
number of FEM-BEM iterations, ...)

38
4.7 Simulation cases

4.7.1 Blade structural model

Together with the baseline blade, four other variations of the AOC 15/50 case are defined and modelled in
Abaqus by [35] (see Annex D for more detailed information on fibre lay-up properties):

• Case 1: baseline AOC blade (figure 22)

• Case 2: changed torsional stiffness by new fibre lay-up strategy, some glass fibres layers are
replaced with carbon

• Case 3: reduced torsional stiffness by changed fibre lay-up strategy

• Case 4: swept blade planform (figure 23) with the same fibre lay-up strategy as defined in case 2

• Case 5: swept blade planform with the same fibre lay-up strategy as defined in case 3

For the coupling between Abaqus and WT_Perf, no structural changes where made to these models. Some
minor changes consists out of labelling and grouping mesh coordinates to identify different blade
sections for a structured communication procedure with WT_Perf (as explained in chapter 4.6.4a) ).

Figure 22: Case 1, 2 and 3 straight baseline blade model

Figure 23: case 4 and 5 swept blade model

39
Where the leading and trailing edges of the straight blade where swept backwards in the xy plane
according to the following power law:

y= A
x−x root
xtip −x root   where A = 1 and e = 3

The blade twist and chord distributions are for all 5 cases identical:

Twist distribution and chord length for all cases


10
9
8
chord length [m*1e-01]

7
twist angle [deg]

6
5 tw ist (θ)
4 chord (c)

3
2
1
0
0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0
radial position [m]

Figure 24: Twist and chord distributions for cases 1 through 5. Note that chord length is in meters*10

Plotting the two different blade planforms:

Swept and straight blade planform


1.0
leading and trailing edge positions [m]

0.8
0.6
0.4
0.2
LE c4, c5
0.0
TE c4, c5
-0.2 0 1 2 3 4 5 6 7 8
LE c1, c2, c3
-0.4
TE c1, c2, c3
-0.6
-0.8
-1.0
-1.2
-1.4
radial position [m]

Figure 25: Comparing the straight (case 1, case 2, case 3) and swept blade planform (case 4, case 5)

40
4.7.2 Profile aerodynamic data

According to [32], the NREL profiles S821, S819, S820 (from root to tip) are used on the AOC blade.
However, an older report [34] mentions 3 other airfoils: S814, S812 and S813. Later in this research
project, it became apparent that the S821, S819, S820 where intended at first, but the S814, S812 and
S813 where finally used on the AOC 15/50 blade. According to former NREL airfoil designer James
Tangler, both airfoils are very similar and are expected to behave accordingly. For this report, it is
assumed the AOC blade uses the S821, S819 and S820 airfoils. Regarding the comparison with the CFD
results this assumption remains valid since the available Abaqus model is created based on these airfoils.

Since no wind tunnel data exists for these NREL airfoils, XFOIL [43] is used to calculate the aerodynamic
coefficients for the attached flow conditions, cl,max, stall angle and some post stall angles. NREL's
AirfoilPrep [42] spreadsheet can, based on the XFOIL input in this case, create a full -180° + 180° angle of
attack data set and add 3D stall delay corrections. A second data set became available later on in the
project. James Tangler provided the aerodynamic coefficients for 10 AOC blade sections (using the S821,
S819 and S820 airfoils), as calculated by the Eppler [45] code (note: the Eppler results where calculated
several years ago). For more details on these airfoil data sets see Annex B.

For the swept blade model (cases 4 and 5), following correction on the lift coefficients is applied (as
suggested in [7], which is based on the formula for swept wings by Hoerner):

C l =C l cos2 

Which creates the following difference when compared to the normal straight blade (figure 26):

Lift Coefficients
2
lift and drag coefficients

1.8

1.6

1.4 Cl XFOIL
Cl sw ept
1.2

0.8
1 2 3 4 5 6 7 8
radius [m]

Figure 26: lift coefficient distributions for the straight and swept undeformed blade, as calculated by
WT_Perf

4.7.3 Simulation conditions

Following conditions where set during the CFD (using the Reynolds Averaged Navier-Stokes equations, for
more information on the CFD analysis see [35]) and will also be used in the BEM-FEM coupling procedure:

• 18 m/s steady wind field, no wind shear and rotor aligned with the flow

• rotor speed set at 80 RPM

41
4.7.4 Sensitivities

In figure 27, the influence of the pitch angle in WT_Perf on power and rotor thrust is displayed.

As mentioned before in chapter 4.2 General procedure, the coning angle has a limited influence on the
total thrust force on the rotor, as can be seen in figure 28.

The sensitivity regarding the aerodynamic coefficients is outlined in Annex B AOC 15/50 aerodynamic
blade data and BEM model verification.

Power and rotor thrust for different pitch settings


5%
differenence wrt 0 deg pitch [%]

0%
-5 -4 -3 -2 -1 0 1 2 3 4 5
-5%
Pow er
Thrust
-10%

-15%

-20%
pitch angle [deg]

Figure 27: Influence of the blade pitch angle on power output and rotor thrust for case 1 in WT_Perf at 18
m/s and 80 RPM. Results normalised with respect to 0 pitch angle

Influence coning angle


6%
Thrust loss [%]

4%

2%

0%
0 1 2 3 4 5 6 7 8 9 10
coning angle [deg]

Figure 28: influence of the coning angle in WT_Perf for case 1. Comparing total rotor thrust under
different coning angles with respect to zero coning

4.8 Simulation results and comparison with CFD-FEM coupling


When comparing the CFD and BEM cases, induced twist angle, flap and edge-wise deflections are
considered. Ideally, also the aerodynamic forces of the CFD and BEM case should be compared. However,
the integrated pressure forces of the CFD analysis are not available. The individual pressure force
magnitudes on the mesh elements are available. An attempt obtain the integrated pressure forces by
calculating the element normal vector (based on the coordinates of the 4 mesh points), did not deliver
satisfactory results. Due to time constraints, no further attempts where made. As a result, the integrated
aerodynamic forces of the CFD cases are not considered for this research topic.

42
Case 1 BEM-FEM: baseline blade

deflection: [m] twist: [deg]


1
0.8
0.6 Edge-w ise
0.4 Flap-w ise
Ind. Tw ist
0.2
0
-0.2 1 2 3 4 5 6 7 8

radial position [m]

Error Case 1: comparing BEM-FEM with CFD-FEM


60%

50%

40%
error [%]

edge
30%
flap
20% tw ist
10%

0%
1 2 3 4 5 6 7 8
-10%
radial station [m]

Figure 29: Positive induced twist angle is towards feathering. Positive error indicates an overestimation on
the BEM-FEM side, relative to the CFD-FEM results

For case 1, the induced twist angles are small and are close to the CFD-FEM case. However, from the error
in the edge and flap-wise deflections, it is clear that the aerodynamic forces originating from the WT_Perf
are significantly overestimated with respect to the CFD loads.

Figures 34, 35 and 36 compare the aerodynamic loading as calculated by WT_Perf for the undeformed
blade and deformed blade state. They quantify the importance of the fluid structure interactions.

The small deformations result in a very small change in aerodynamic rotor output (0.03 %), see figure 35.
When looking at the distributed thrust forces and torque in figures 34 and 36, it can be concluded that the
blade deformations due to the aerodynamic loading has an insignificant influence on the performance.
Consequently, it could be argued that a BEM-FEM approach is not necessarily to calculate the performance
for the baseline case.

Although the total blade deflection (and consequently the aerodynamic loading) is highly overestimated
on the BEM side (compared to the CFD case), the induced twist angles are in fact relatively close to each
other. From this one can conclude that for the stiff baseline blade, the structural couplings are very weak
and can be safely ignored.

Note that a positive induced twist is towards blade feathering, or a locally decreased angle of attack. The
errors where calculated by taking the absolute difference of the BEM-FEM and CFD-FEM results (so positive
means larger deformations on the BEM-FEM side), normalised by the latter.

43
Case 2 BEM-FEM: reduced fibre lay-up scheme
deflection: [m] twist: [deg] 1
0.8
0.6 Edge-w ise
0.4 Flap-w ise
Ind. Tw ist
0.2
0
-0.2 1 2 3 4 5 6 7 8

radial position [m]

Error Case 2: comparing BEM-FEM with CFD-FEM


60%
50%
40%
30%
error [%]

edge
20%
flap
10% tw ist
0%
-10% 1 2 3 4 5 6 7 8

-20%
-30%
radial station [m]

Figure 30: Positive induced twist angle is towards feathering. Positive error indicates an overestimation on
the BEM-FEM side, relative to the CFD-FEM results

The comments for case 1 are also valid for case 2, which has an increased torsional stiffness due to the
introduced carbon fibres (see figure 46 in Annex D), even though the total number of layers was reduced.

The induced twist angles are almost halved when compared to case 1. However, the considered angles are
too small to be relevant in this case, since both aerodynamic power output, thrust and torque distribution
are very close to each other (see figures 34 and 36). The induced twist angle error is especially large near
the tip, but due the small induced angles not relevant.

44
Case 3 BEM-FEM: mirror lay-up (increased bend-twist coupling)
3.6
3.4
3.2
3
2.8
2.6
deflection: [m] twist: [deg]

2.4
2.2
2
Edge-w ise
1.8
Flap-w ise
1.6
Ind. Tw ist
1.4
1.2
1
0.8
0.6
0.4
0.2
0
1 2 3 4 5 6 7 8

radial position [m]

Error Case 3: comparing BEM-FEM with CFD-FEM


50%
40%
30%
error [%]

edge
20% flap
tw ist
10%
0%
1 2 3 4 5 6 7 8
radial station [m]

Figure 31: Positive induced twist angle is towards feathering. Positive error indicates an overestimation on
the BEM-FEM side, relative to the CFD-FEM results

The mirror lay-up results in a much stronger coupling between bending and torsion modes. The induced
twist angle is slightly overestimated, flap and edge-wise deflections remain significantly overestimated,
although the error is smaller (roughly 40%) when compared to cases 1 and 2.
The aerodynamic power is affected, but since it is small (reduction of 3.53 % when compared to the blade
undeformed state, see figure 35), the significance would depend on the required accuracy. The distribute
aerodynamic loading changes noticeably (see figures 34 and 36). When considering detailed blade
structural design and/or analysis, the influence of the blade deformations on the loading might proof to
be important.
Although the structural coupling has a clear influence on the loading and a more limited effect on the
aerodynamic power output, the rather large overestimated BEM forces do not result in the same error on
the induced pitch angles. It could be concluded that the structural couplings for case 3 do not follow
exactly the same trend as the flap and edge-wise deflections.

45
Case 4 BEM-FEM: Case 2 + sweep
compare
2.8
2.6
2.4
2.2
deflection: [m] twist: [deg]

2
1.8
1.6 Edge-w ise
1.4 Flap-w ise

1.2 Ind. Tw ist

1
0.8
0.6
0.4
0.2
0
1 2 3 4 5 6 7 8

radial position [m]

Error Case 4: comparing BEM-FEM with CFD-FEM


40%

30%
error [%]

20% edge
flap
10% tw ist

0%
1 2 3 4 5 6 7 8
-10%
radial station [m]

Figure 32: Positive induced twist angle is towards feathering. Positive error indicates an overestimation on
the BEM-FEM side, relative to the CFD-FEM results

The swept blade cases (as explained in chapter 4.7.2) results in a decreased, but still significant, error
when considering the edge and flap-wise deflections, indicating an overestimation of the aerodynamic
forces. The induced twist angle is now slightly under estimated.

Due to the increased torsional stiffness of the fibre lay-up (as indicated by the case 2 results), the swept
blade planform (which produces now most of the twist induced torsion) results in lower induced twist
angles when compared to the structural bend-twist couplings of case 3 (as can be seen in figure 31).

The induced twist angles result in a reduced aerodynamic power output of 1.43 % (figure 35) when
compared to the undeformed blade state. The swept blade planform causes a significant difference in
aerodynamic power output when compared to the unswept case: it drops from 168 kW (case 1) to 141 kW
(case 4), which is reduction of 13%.

46
Case 5 BEM-FEM: mirror lay-up + sweep
compare
4.6
4.4
4.2
4
3.8
3.6
3.4
3.2
3
deflection: [m] twist: [deg]

2.8
2.6 Edge-w ise
2.4 Flap-w ise
2.2 Ind. Tw ist
2
1.8
1.6
1.4
1.2
1
0.8
0.6
0.4
0.2
0
1 2 3 4 5 6 7 8

radial position [m]

Error Case 5: comparing BEM-FEM with CFD-FEM


40%

30%
error [%]

20% edge
flap
10% tw ist

0%
1 2 3 4 5 6 7 8
-10%
radial station [m]

Figure 33: Positive induced twist angle is towards feathering. Positive error indicates an overestimation on
the BEM-FEM side, relative to the CFD-FEM results

Case 5 presents itself as the best match between CFD-FEM and BEM-FEM approach. Flap- and edge-wise
errors are still significant.

The induced twist angles are close to 4.6 degrees near the tip. Notice that the induced twist angles in case
5, which has both the swept planform and the mirrored off-axis fibres to create a feathering moment, is
clearly not equal to the summation of the swept induced twist angles of case 4 and the structural induced
couplings (mirrored off-axis fibres) of case 3.

47
At this stage it is difficult to conclude whether the swept BEM correction on the lift coefficients (see
chapter 4.7.2) are to severe (resulting in a smaller error with the CFD case) or that the CFD calculations do
not handle the swept blade correctly. Further investigations are required to come to a conclusion with this
respect.

Aerodynamic power output is affected most when comparing the deformed and undeformed blade states:
a decrease of 4.86% (figure 35). The blade distributed loading is also affected most noticeably of all 5
cases (see figures 34 and 36).

Rotor thrust inital and final coupling step

1400

1200

1000
Thrust distribution [N/m]

800

600
initial step case 1,2,3
initial step case 4,5
400 final step case 1
final step case 2
final step case 3
200
final step case 4
final step case 5
0
1 2 3 4 5 6 7 8

radial position [m]

Figure 34: Comparing the aerodynamic thrust (as calculated by WT_Perf) for the undeformed blade (initial
iteration step) and the deformed blade (final iteration step). The bend-twist coupling has clearly a
significant influence on the performance (@ 80 RPM, wind speed 18 m/s)

Note that normally, a twist to feather situation for a stall controlled machine should lead to an increased
power output, since reducing the angle of attack means higher lift coefficients (at the considered operatig
conditions of 18 m/s wind speed at 80 RPM). In this case however, the outer part of the blade is close to
maximum angle of attack (see figures 39 and 40) and hence a smaller effective angle of attack
corresponds to a lower lift coefficients and corresponding loading (see also figures 34 and 36). The inner
part of the blades does however encounter a small increased aerodynamic loading, since the effective
angles of attack are indeed far in the stalled region (see figures 39 and 40) and thus a smaller effective
angle of attack leads to larger lift coefficients.

48
Aerodynamic power output initial and final coupling step
170
initial
165 final
160

155
Power [kW]

150

145

140

135

130
Case 1, 0.03% Case 2, 0.26% Case 3, -3.53% Case 4, -1.43% Case 5, -4.86%

Figure 35: Influence of the deformations on power production (@ 80 RPM, wind speed 18 m/s). The
difference between both of them is indicated in percentages on the horizontal axis.

Rotor torque inital and final coupling step


2500

2000
Torque distribution [N]

1500

1000
initial step case 1,2,3
initial step case 4,5
final step case 1
500 final step case 2
final step case 3
final step case 4
final step case 5
0
1 2 3 4 5 6 7 8

radial position [m]

Figure 36: Comparing the aerodynamic torque (as calculated by WT_Perf) for the undeformed blade
(initial iteration step) and the deformed blade (final iteration step)

49
4.9 Conclusions and and recommendations
From the current results, following conclusions can be drawn:

• As pointed out in literature, this coupled FEM-BEM study confirms that a bend-twist coupled
blade can induce significant twist angles which influences blade loading and power output

• WT_Perf overestimates the aerodynamic forces quite significant. As a result the blade flap- and
edge-wise deflections are off with 40% and 50% respectively (with respect to the CFD
simulations). Blade twist angles are on average within a 5 % error range. The overestimations are
caused by incorrect stall and post stall airfoil characteristics as calculated by XFOIL. It should be
noted however that the CFD simulations are not verified against wind tunnel tests or other CFD
studies. The considered case (80 RPM, wind speed 18 m/s) has a fully stalled blade (see also
figure 40 for the span-wise variation of the effective angle of attack for the BEM simulations),
which can be difficult to predict correctly for a CFD formulation as well

• WT_Perf's over prediction is, although to a lesser extent, caused by the somewhat rough model
of the blade, having an airfoil shaped root section instead of an ovoid one (see Annex B for
comparison with HAWC2 and WT_Perf)

• Although the swept cases (4 and 5) still have significant differences with the CFD simulations,
their error is smaller when compared to the straight blade planforms (cases 1, 2 and 3). This
could indicate that the sweep corrected model is closer to the CFD results, the sweep correction
is too severe (which reduces the error due to the overestimated BEM loads) or that the CFD
simulations did not captured all the relevant aerodynamic effects regarding a swept blade layout
correctly either

• Errors due to 3D stall correction (NREL spreadsheet AirfoilPrep-v2p2): since 3D stall correction
leads to higher maximum lift coefficients, this could also contribute to the difference with the
CFD simulations

Many issues remain unanswered. Some of the following recommendations can also be placed in the
broader context of a blade design and analysis cycle (see figure 16 on page 23):

• Detailed information and verification regarding the CFD simulations could further help
interpreted the current results. The turbulence model used for the CFD simulations is Spalart-
Allmaras. A further in depth CFD simulation could include investigating and comparing the
influence of the turbulence model on aerodynamic rotor performance.

• Use the coupled FEM-BEM approach for a blade with known and verified airfoil data

• In the current model, the centrifugal forces are applied based on the undeformed structure.
However, for cases 3, 4 and 5, the large deformations are expected to influence the centrifugal
forces significant as well (and vice versa)

• Investigate influence of coupling scheme's on the convergence characteristics and run-time of


the coupling program

• Reassess the mesh node sorting function in the coupling program (residing in the function
read_setname() )

50
• Implement the “Maheri approach” to estimate the power curve based on a coupled FEM-BEM
simulation for one wind speed only

• Investigate the influence of chord wise loading distribution for bend-twist coupled blades and
compare with the current single loading point approach

• Construct an automated verification procedure between a full FE blade model and a reduced
structural model for dynamical simulations

• Investigate the influence on the dynamics and fatigue loading of a flexible blade

• Transfer loads from the dynamic analysis back to the FEM program, to investigate extreme
loading cases in detail

• Investigate the influence of a swept rotor blade planform on the aerodynamic performance: it is
expected that the additional radial flow over the blade has an influence on the affected sections'
stall behaviour (comparable to 3D stall delay)

• Design and optimisation procedure for a flexible blades, which could consider following design
parameters: blade sweep curve, fibre lay-up strategies (structural couplings), to check against:
power output, fatigue damage and ultimate loads of the blade

51
Annex A Program flowcharts
On the following two pages, the coupling Python scripts between Abaqus and WT_Perf are outlined as flow
charts. Two views are presented: the general program structure and a more detailed representation of the
different functions with their respective input and output variables.

52
Determine WT_Perf input
Read LE, TE, LP coordinates Create Abaqus set sections for BEM: Create for each section a moment arm due to
Define coning angle parameters and AC of
and sort spanwise create_BEM_element_sections() single loading point loading at LP instead of
mdb model calc_coning() section loading points
read_setname() create_BEM_node_sections() create_WT_Perf_sections() AC: corr_moment_arm
database derive_blade_state()
An Abaqus model
Read undeformed blade state from mdb model database: blade_profile_data()

csv file
save coordinates of LE, LP and
TE (REDUCED and ALL),
blade_state_array and AC
coordinates

Abort on error

Apply loading in new Abaqus Gather required Manually catch CM from Calculate translate Abaqus model to
Read WT_Perf output
step, calculate Maero parameters in airfoil files, based on AoA BEM loading WT_Perf input
read_WT_Perf_output()
create_WT_Perf_load_set() nodelabel_loading_array from WT_Perf output exe_WT_perf() create_WT_Perf_input()
read_aero_moment()

Submit Abaqus
create aerodynamic loading in Abaqus BEM: create input, execute WT_Perf, read output
analysis
csv file
save WT_Perf output

odb database
output database with
results from analysis

Re-Configure
model (mdb object)
for restart analysis Read LE, TE, LP coordinates and sort Correct section node sets (LE, LP, Determine moment arm
Define coning angle Determine WT_Perf input
spanwise (ALL nodes and REDUCED) TE) due to load application
calc_coning() parameters and AC of
read_odb_setname_fieldOutputs() correct_nodeset_array_reduced() at LP instead of AC
section loading points
derive_blade_state()
read deflected blade state from odb database: blade_profile_data_fieldOutputs()

csv file
UPDATED save coordinates of LE, LP and
odb database TE (all nodes uncorrected),
blade_state_array (corrected) Check convergence
output database with
results from analysis and AC coordinates (corrected)

Is
no twist deformation
converged?

Abort if time exceeds


Update Save as maxAnalysisTime or if
Apply loading in new Abaqus step, yes centrifugal previous blade
number of iterations yes
Submit restart apply centrifugal forces forces? state
analysis, at the exceeds max_iter
update_load_set_centrifugal()
newly created load
step
update centrifugal loading in Abaqus no Update
yes centrifugal
forces?

Abort on error no
Apply loading in new Abaqus Gather required Manually catch C_M from Calculate translate Abaqus model to
Read WT_Perf output
step, calculate Maero parameters in airfoil files, based on AoA BEM loading WT_Perf input
read_WT_Perf_output()
update_load_set() nodelabel_loading_array from WT_Perf output exe_WT_perf() update_WT_Perf_input()
read_aero_moment()

update aerodynamic loading in Abaqus BEM: update input, execute WT_Perf, read output
odb object odb object
partially aeroelastic aeroelastic deformed
csv file deformed blade blade
save WT_Perf output for each
iteration
blade_profile_data (mdb/odb) A read_setname A1
calc_RElm 2
A2
calc_coning (for LP) 3
A3

input------------------------------------------------------ input------------------------------------------------------ input------------------------------------------------------ input------------------------------------------------------


db_model: an odb or mdb object db_model: an odb or mdb object NodeSet_array: [NodeLabel, x, y, z, 0] NodeSet_array: [NodeLabel, x, y, z, x]
iteration: current iteration set_name: the specified set name (string) plot_title : string for the plot title update_WT_Perf_input create_WT_Perf_input
output---------------------------------------------------- output---------------------------------------------------- plot_switch : [on or off] to plot or not to plot input------------------------------------------------------ input------------------------------------------------------
blade_state_array: array_header: output---------------------------------------------------- output---------------------------------------------------- theta_r_array: twist state of the blade input_file_Perf: name of the BEM input file
[RElm, Twist, Chord, AFfile, PrntElem] accompany the csv print of the array NodeSet_array: [NodeLabel, x, y, z, x] coning_angle_deg input_file_Perf: name of the BEM input file blade_state_array:
blade_state_AC_array: NodeSet_array: coning angle coning_angle_deg [RElm, Twist, Chord, AFfile, PrntElem]
[RElm, x_ac, y_ac, z_ac] [NodeLabel, x, y, z, 0] RotorRad
nodelabel_profile_array: documentation---------------------------------------- documentation---------------------------------------- documentation---------------------------------------- output---------------------------------------------------- HubRad
[LE(NodeLabel), LP(label), TE(label)] read all the nodes contained in set_name, Calculate RElm (for WT_Perf input): the Calculate an assumed coning angle to NumSeg: int, # of radial sections per blade coning_angle_deg
RotorRad : float sort them on x coordinate (spanwise), output BEM loading points, the distance along the compromise for the bending deflection in the NumSect: int, # of azimuthal sections for the output----------------------------------------------------
HubRad : float them as an array containing NodeLabel, x, y, blade from the centre of rotation (only for the x-z plane (out-of-rotor-plane, flapwise): least rotor plane NumSeg: number of blade segments (int)
coning_angle_deg_LP : float z, 0. The database can be an odb x-coordinate, otherwise WT_Perf complains). square fit on data points and its angle with NumSect: number of azimuth sectors (int)
corr_moment_arm : [ arm_y , arm_z ] (coordinate member) or mdb object. This This is done considering all the spanwise the vertical plane Rho: air density from WT_Perf input (float)
WTPerfSections : [RElm, segmenth_length] function considers the original coordinates nodes. documentation----------------------------------------
csv file output----------------------------------------- documentation---------------------------------------- Use the template_file_Perf to create the
mdb coordinates from LE, TE, LP 4
A4 5
A5 6
A6 update the new twist angles in the WT_Perf correct input file for WT_Perf: RotorRad,
blade_state_array for WT_Perf sections create_BEM_element_sections create_WT_Perf_sections create_BEM_node_sections input file to recalculate the BEM loading, the HubRad, coning_angle_deg,
documentation---------------------------------------- input------------------------------------------------------ input------------------------------------------------------ input------------------------------------------------------ twist array should be equal to the number of blade_state_array and airfoil files (from
read the LE, TE, LP from the mdb with mdb_model: mdb object WTPerfSections: mdb_model: mdb object sections defined in the WT_Perf input airfoil_file_path)
read_setname(), calculate the coning angle NodeSet_array : [NodeLabel, x, y, z, x] [RElm, segment length, nr of nodes]
based on the LP nodes calc_coning(), add NodeSet_reduced: [NodeLabel, x, y, z, x]
RElm value for each node (x-coordinate) output---------------------------------------------------- output---------------------------------------------------- output----------------------------------------------------
calc_RElm(), print LE, LP, TE node information WTPerfSections: NodeSet_reduced_array:
to csv file, create_BEM_element_sections() [RElm, segment length, nr of nodes] [NodeLabel, x, y, z, RElm] exe_WT_Perf 1
documentation---------------------------------------- documentation---------------------------------------- documentation---------------------------------------- input------------------------------------------------------
(elements) and create_BEM_node_sections()
Create sections consisting out of elements. Create a reduced NodeSet_array for the Convert element sections into node sections. input_file_Perf: NodeLabel, x, y, z, RElm 3
(nodes). Reduce LE, LP, TE nodes to number read_aero_moment
Spacing is according to the variable WT_Perf sections. Find the Abaqus node Remove overlapping nodes between output----------------------------------------------------
of WT_Perf sections input------------------------------------------------------
SectionProfileDelta (except the first closest to the actual WT_Perf RElm point. adjacent sections and add TE-LE nodes. WT_Perf_cmd_out:
create_WT_Perf_sections() and derive blade wt_perf_out_array:
element, which is 9 elements wide). This The RElm value is the same for LE, LP and string containing WT_Perf exit msg [element nr, azimuth angle, Trust/len (N/m),
state at these sections derive_blade_state().
method only works for an even number of TE sets. documentation---------------------------------------- Torque/len (N), AlfaD, Loc Vel, 0, 0]
Calculate correction moment arm (loading in LP
sections (NumberOfSectoins). segment length = 2*delta_n Execute WT_Perf with given input file, output blade_state_array:
instead of AC). Write blade_state_array and
segment length = 2*delta_n a variable with the the exit message [RElm, Twist, Chord, AFfile, PrntElem]
blade_state_AC_array to csv file.
(containing possible errors) Rho:
output----------------------------------------------------
read_WT_Perf_output 2 wt_perf_out_array:
input------------------------------------------------------ [element nr, azimuth angle, Trust/len (N/m),
input_file_Perf: name of the BEM input file Torque/len (N), AlfaD, Loc Vel, CM, Rho]
NumSeg: number of blade segments documentation----------------------------------------
Read from the wt_perf_out_array the
NumSect: number of azimuth sectors
output---------------------------------------------------- section inflow angle (AlfaD) and the local
B 1
B1 2
B2 wt_perf_out_array: velocity (Loc Vel). Get from the section airfoil
blade_profile_data_fieldOutputs (odb) read_odb_setname_fieldOutputs calc_coning (for LP) file the CM value at this AlfaD. Add CM and
input------------------------------------------------------ input------------------------------------------------------ input------------------------------------------------------ [element nr, azimuth angle, Trust/len (N/m),
Torque/len (N), AlfaD, Loc Vel, 0, 0] Rho to the wt_perf_out_array.
nodelabel_profile_array: odb: an odb object NodeSet_array: [NodeLabel, x, y, z, x]
[ LE(NodeLabel), set_name: string plot_title : string for the plot title documentation----------------------------------------
LP(NodeLabel),TE(NodeLabel) ] iteration: int plot_switch : [on or off] to plot or not to plot read the BEM output file (BEM loading)
Odb: Abaqus output database object step: string Check convergence
Iteration: int output---------------------------------------------------- output----------------------------------------------------
RotorRad: float array_header: coning_angle_deg
Step: string accompany the csv print of the array coning angle
WTPerfSections: NodeSet_array: NodeLabel, x, y, z, 0 documentation----------------------------------------
[RElm, segmenth length, nr of nodes] documentation---------------------------------------- Calculate an assumed coning angle (using
deformed blade state odb fieldOutput

read all the odb fieldOutputs contained in the full LP set) to compromise for the
update_load_set_centrifugal update_load_set create_WT_Perf_load_set
output---------------------------------------------------- set_name, at the last frame of step. sort bending deflection in the x-z plane (out-of- input------------------------------------------------------ input------------------------------------------------------ input------------------------------------------------------
blade_state_array: them on x coordinate (spanwise), output rotor-plane, flapwise): least square fit on
mdb_model: a mdb object mdb_model: a mdb object mdb_model: a mdb object
[RElm, Twist, Chord, AFfile, PrntElem] them as an array containing NodeLabel, x, y, data points and its angle with the vertical
numberCentrifugal: Abaqus name nodelabel_loading_array: [NodeLabel_LP, nodelabel_loading_array: [NodeLabel_LP,
blade_state_AC_array: [RElm, x_ac, y_ac, z, 0. plane
centrifugal force(string) RElm, segment length, chord length] RElm, segment length, chord length]
z_ac]
iteration: wt_perf_out_array: wt_perf_out_array:
coning_angle_deg_LP: float
correct_nodeset_array_reduced 3
B3 derive_blade_state 4
B4 output---------------------------------------------------- element nr [-], azimuth position [deg], element nr [-], azimuth position [deg],
corr_moment_arm: [ arm_y , arm_z ] trhust/len [N/m], torque/len [Nm/m] trhust/len [N/m], torque/len [Nm/m]
CorrTextLog: string stating how many input------------------------------------------------------ input------------------------------------------------------
NodeSet_array: NodeLabel, x, y, z, RElm LE_nodes: [NodeLabel, x, y, z, RElm] reduced documentation---------------------------------------- NumSect: integer, number of azimuthal NumSect: integer, number of azimuthal
nodes where corrected Create a new step, apply centrifugal forces sections for the rotor plane sections for the rotor plane
deltaBogusLow: lower limit difference for LP_nodes: [NodeLabel, x, y, z, RElm] reduced
two successive points (float) TE_nodes: [NodeLabel, x, y, z, RElm] reduced in the newly created step and set iteration: int iteration: int
csv file output----------------------------------------- FieldOutputRequests
deltaBogusHigh: upper limit difference for RotorRad: float corr_moment_arm: [ arm_y , arm_z ] corr_moment_arm: [ arm_y , arm_z ]
odb fieldOutput coordinates from LE, TE, LP
two successive points (float) force_name_list: Abaqus force names output----------------------------------------------------
blade_state_array for WT_Perf sections
checkBogusCoord: 0, 1 or 2 (x, y, z moment_name_list:Abaqus moment names force_array: [thrust, torque_force, torque]
respectively) output---------------------------------------------------- force_name_list: Abaqus force names
documentation----------------------------------------
output---------------------------------------------------- output---------------------------------------------------- force_array: [thrust, torque_force, torque] moment_name_list:Abaqus moment names
read the LE, TE, LP from the odb fieldOutputs
NodeSet_array: NodeLabel, x, y, z, RElm blade_state_array: documentation---------------------------------------- documentation----------------------------------------
with read_odb_setname_fieldOutputs() (print
textLog: one line text message containing [RElm, Twist, Chord, AFfile, PrntElem] Create new Abaqus step and set Create new Abaqus step and set
LE, LP, TE node information to csv file) for both
number of corrected nodes (string) blade_state_AC_array: FieldOutputRequests. Change the loads in FieldOutputRequests. Create new loads in
the full (SetName) and reduced set profile_details B4
[RElm, x_ac, y_ac, z_ac] the newly created step from the the newly created step from the
(reducedSetName), add the RElm values from input------------------------------------------------------
documentation---------------------------------------- documentation---------------------------------------- wt_perf_out_array (load name as given in wt_perf_out_array (on one node per
WTPerfSections, calculate the coning angle SetNodes : [ TE_x, TE_y, TE_z
Scan the reduced array for inconstancies on Chord length, AC position, twist angle for the force_name_list), calculate M_aero and section), calculate M_aero and M_corr and
based on the LP nodes calc_coning(), LP_x, LP_y, LP_z
the checkBogusCoord (x, y or z). If the reduced set via profile_details() and find M_corr and update the moments as given in apply (create loads) them on all section
correct_nodeset_array_reduced() to filter out LE_x, LE_y, LE_z ] nodes.
difference between successive nodes is not the correct corresponding airfoil, based on its moment_name_list.
bogus points and afterwards output----------------------------------------------------
derive_blade_state(). Calculate correction in the defined range deltaBogusLow- radial position (percentage).
x_ac, y_ac, z_ac: float
moment arm (loading in LP instead of AC). deltaBogusHigh, replace them with an
theta_r_deg, chord_length: float
Write blade_state_array and interpolation (maximum of 3 successive
documentation----------------------------------------
blade_state_AC_array to csv file. bogus points) between the correct
Determine AC, chord and twist angle based
. neighbouring points
on LE, LP, TE.
Annex B AOC 15/50 aerodynamic blade data
and BEM model verification
Since no aerodynamic data of the AOC 15/50 blade is available, XFOIL and AirfoilPrep are used to obtain a
data set of aerodynamic coefficients (cl, cd and cm) over a -180 +180 angle of attack range.

From the original NREL design report considering the S821, S819 and S820 airfoils [31], the XFOIL
formatted .dat files (a labelled coordinate file, as explained in the XFOIL 6.9 documentation [44]) where
generated describing the upper and lower surface coordinates in y/c as function of x/c positions. This
means that the file starts with the name of the airfoil. The following lines are assembled of the x/c and y/c
coordinates, starting from the TE, over the upper surface to the LE and back to the TE via the lower
surface. The origin of this coordinate system is the LE.

Figure 37: The NREL profiles of the AOC 15/50 blade as used in the Abaqus model and XFOIL calculations

For Ncrit (boundary layer amplification factor), the standard XFOIL value of 9 is chosen, which corresponds
to the average wind tunnel conditions [44]. Ncrit will influence the laminar-turbulence-transition point and
as a consequence will effect the stall angle. Determining Ncrit is not a straightforward process and is
considered beyond the scope of this thesis. However, since in many BEM computations the cL-cD data
originates from wind tunnel measurements, it is appropriate to set Ncrit to 9. Reynolds and Mach number
are set to 1e06 (one million) and 0 respectively.

The airfoils are blended by AirfoilPrep to smoothen the differences between airfoil sections. For this end,
4 stations are added between the different airfoils, using following weight fractions: 20%-80%, 40%-60%,

55
60%-40%, 80%-20%. In doing so, the relative difference in cL between the airfoil sections is reduced globally
to several percentages instead of several ten's of percentages. AirfoilPrep also adds 3D stall delay and
extends the angle of attack range from -180° to 180° based on flat plate theory.

In a later stage, the XFOIL results were re-calculated using the local flow conditions (as outputted by
WT_Perf, so including induction factors) to determine local Reynolds and Mach numbers, see figure 38.
For the air density, speed of sound and dynamic viscosity, the standard atmosphere values are used.
Further, the number of panels in XFOIL is changed to account for possible numerical errors. Obtained
results are comparable with the first iteration, although cL,max is slightly higher.

Reynolds and Mach number @ 80 RPM, V=18 m/s


2.50E+06

2.00E+06
Re [-], M*1e07

1.50E+06 Re
Mach*1e7
1.00E+06

5.00E+05
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Radial position [-]

Figure 38: Reynolds and Mach number for the AOC blade at 80 RPM, wind speed = 18 m/s

When comparing both XFOIL results with a dataset calculated by James Tangler using the Eppler code, the
former has a higher cL,max and stall angle than the later (see figure 39). It should be noted that these
results are computed years ago with an older version of the Eppler code. In both cases the angle of attack
range was extend (to cover the 180 to -180 degree range) and the lift and drag coefficients where
corrected for 3D stall by using AirfoilPrep. Running only WT_Perf with the XFOIL and Eppler data sets
results in a very different loading and power output for the 80 RPM at 18 m/s case (same conditions as
used in the CFD). See figures 40, 41 and 42. The comparison with the Eppler results are included here to
illustrate that the obtained aerodynamic coefficients varies significantly depending on the methodology. In
the same figures, some HAWC2 results are included as well. They serve as verification of the WT_Perf
results. See Annex C for more information on the HAWC2 model and results.

Nando Timmer form the TU Delft kindly provided the lift coefficients of the S819 profile, which where
calculated with RFOIL, a for wind turbine rotors optimized version of XFOIL. It is included here just to
point out the difference with XFOIL [40].

An other indication that XFOIL over predicts both cL,max and the stall angle, is given in [33]. A comparison
between XFOIL and wind tunnel tests is made. Further, the authors show that the characteristic “drag
bucket” in the cl cd plots is not accurately reproduced with XFOIL.

As can be seen from table 1, the Eppler and XFOIL data sets generate significant different rotor power,
torque and thrust. HAWC2 indicates an even larger difference between both aerodynamic sets when
compared to WT_Perf. The simplified structural HAWC2 model has a limited influence on rotor
performance (comparing HAWC2 XFOIL and HAWC2 XFOIL stiff in table 1).

56
WindSpeed TSR RotSpeed Pow er Torque Thrust
WT_Perf m/s - rpm kW kN-m kN
XFOIL 18 3.68 80 168.80 20.15 13.03
Eppler 18 3.68 80 124.94 14.91 10.59
XFOIL no3Dstall 18 3.68 80 163.29 19.49 12.22
XFOIL noHubLoss 18 3.68 80 168.01 20.05 12.83
HAWC2
XFOIL 18 3.68 80 156.91 18.73 12.29
Eppler 18 3.68 80 98.13 11.71 9.61
XFOIL stiff 18 3.68 80 154.46 18.44 12.12

Table 1:Significant difference in blade loading (as calculated by WT_Perf) when using XFOIL or Eppler
generated aerodynamic data

The main difference between HAWC2 and WT_Perf results is caused by the modelling differences on the
blade. In HAWC2, the blade shape starts as a circular section at the root and is automatically blended into
the first airfoil section at 40% radial position (interpolating aerodynamic coefficients of a circular cross-
section and the respective airfoil). In WT_Perf however, the inner sections are all modelled as the same
airfoil at 40% radial position. See figures 41 and 42 for the effect it has on lift coefficients and thrust
distributions. The integrated properties however (see table 1) show that the influence of this modelling
difference is limited but relevant.

Although HAWC2 does not include a hub loss model, the influence in WT_Perf of this correction factor is
almost irrelevant in this case (see table 1).

Figure 39: Comparing XFOIL and Eppler (as obtained by James Tangler with an older version of the Eppler
code) lift coefficients. No 3D corrections are applied for these plots.

57
Effective angle of attack
50

45

40
angle of attack [deg]

35

30 XFOIL
XFOIL HAWC2
25 Eppler
20 XFOIL no3Dstall

15

10

5
1 2 3 4 5 6 7 8
radius [m]

Figure 40: Comparing the effective angle of attack as calculated by WT_Perf and HAWC2 for case 1. Note
that HAWC2 gives higher angles effective angles of attack in general

Lift and Drag Coefficients


2
1.8
1.6
lift and drag coefficients

cl XFOIL
1.4
cd XFOIL
1.2
Cl Eppler
1
Cd Eppler
0.8
cl XFOIL no3Dstall
0.6 cd XFOIL no3Dstall
0.4 cl XFOIL HAWC2
0.2
0
1 2 3 4 5 6 7 8
radius [m]

Figure 41: Lift and drag coefficients as function of radial position for XFOIL and Eppler for case 1. Note
that the modelling differences are clearly affecting the lift coefficients close to the root. For sections past
3.5m, HAWC2 and WT_Perf have comparable results. Further, the 3D stall correction for XFOIL, as
computed with the NREL spreadsheet AirfoilPrep shows a significance influence. The Eppler airfoil
characteristics where provided with 3D stall corrections

58
Thrust and Torque / unit length
2500

2000
torque/m [N], thrust/m [N/m]

Thrust/Len XFOIL
1500
Torque/Len XFOIL
Thrust/Len Eppler
Torque/Len Eppler
1000 Thrust XFOIL no3Dstall
Torque XFOIL no3Dstall
Thrust XFOIL HAWC2

500

0
1 2 3 4 5 6 7 8

radius [m]

Figure 42: Thrust [N/m] and Torque [N] per unit length blade length for case 1. Although the lift
coefficients indicated some differences between WT_Perf with and without 3D stall corrections and
HAWC2, the actual thrust distributions are quite similar. The most significant difference is caused by the
over prediction of cl,max, as is indicated by the comparison with the older Eppler data set.

59
Annex C AOC 15/50 HAWC2 model
A simplified blade structural model of the AOC 1550 blade is set-up in HAWC2, based on data provided by
[32]. The aerodynamic coefficients are the same as used for WT_Perf. Since HAWC2 performs a dynamic
analysis, the results below are average values under steady conditions (no dynamic stall, tower shadow,
turbulence or wind shear).

The total deflections in HAWC2 are higher, although the loads are slightly lower (see figure 42 and table
1), when compared to the BEM-FEM coupling created for this research topic (comparing figure 44 and 29
on page 43). However, the magnitude of the deflections has the same order of magnitude and can
therefore be considered as an acceptable verification at this point. Note that the HAWC2 structural model
is set-up fast and it is considered beyond the scope to find a good match between the structural FE and
HAWC2 model for this topic.

Figure 43: Effective angle of attack for the HAWC2 AOC 1550 simplified model

60
Figure 44: Flap and edge-wise deflections for the AOC 1550 model, the stiff blade variant shows to be stiff
indeed with a near zero deflection. Although it is considered beyond the scope for this research topic (see
chapters 3.3.4 and 3.5), when comparing HAWC2 deflections with the coupled BEM-FEM results (see figure
29), HAWC2 results in larger deflections, although the loads are smaller.

Figure 45: Thrust forces using XFOIL and Eppler data sets for the AOC 1550 simplified model in HAWC2.
Both in agreement with WT_Perf (see Annex B).

61
Annex D AOC 15/50 structural blade data

case ID case 1 case 2 case 3 case 4 case 5


“ reduc e
LAY -UP: “ referenc e” “ reduc e t o rsio n” “ mirro r” t o rsio n” “ mirro r”
Adapt Reduce
[+-45/0i/+-45/0i/ Torsion:
[+-45/0 i/+-45/0 i/ +45/0i/0i] DB120:+-45º
+45/… one half of the +-45 → 130:20º
...+45/0 i/-+45/0i/- remov ed (nearest to A130:0º
Spar Cap +45] neutral axis) → A130:20º Idem c ase 2 Idem c ase 3
+-45 lay e rs remov e d
A130 → M55J/M18
(glass laminate
re plac ed by c arbon
laminate that is about Idem Red uce
Spar Web [+-45/0 2/+-45] s 2 times thinner) Torsion Idem c ase 2 Idem c ase 3
+-45 lay e rs remov e d
A130 → M55J/M18
(glass laminate
re plac ed by c arbon
laminate that is about S par flange
Spar Flang e [+-45/0 2/+-45] s 2 times thinner) fully remov e d Idem c ase 2 Idem c ase 3
SWEPT
GEOM ETRY : No No No Yes Yes

Figure 46: Fibre lay-up configurations of different blade sections

The 3D blade geometry is generated with AutoBladeTM [50], based on the known geometry of the airfoil
sections (S820, S819, S820), a scaled down tip section (based on the S820) and an assumed symmetric
ovoid root section. Twist varies linearly from the 37% root section to the tip with 7° and a pitch angle of
2.15 is further assumed.

Baseline case 1 (original AOC 1550 blade) is also considered and described in [32]. The published results
where used to verify the Abaqus model by the University of Leuven [35].

62
Annex E WT_Perf model and configuration
WT_Perf's interpretation of the blade length and the defined stations is visualized in figure 47. Each
segment has a length 2*δn for which the spanwise distributed aerodynamic force (thrust and tangential),
local angle of attack and local velocity is calculated. WT_Perf allows for a non constant value for the
different δ. However, it does not allow for overlaps between segments or that the segments at the root
and the tip do not extend beyond the hub and rotor radius.

Figure 47: Relation between hub radius, rotor radius, radial position (RElm) and segment length as used in
WT_Perf

All available corrections in WT_Perf where enabled for the simulations considered during this research
project.

The WT_Perf input file:

----- WT_Perf Input File -----------------------------------------------------


WT_Perf Test01 input file. AOC 15/50 blade.
Compatible with WT_Perf v3.00f
----- Input Configuration ----------------------------------------------------
True Echo: Echo input parameters to "<rootname>.ech"?
True DimenInp: Turbine parameters are dimensional?
True Metric: Turbine parameters are Metric (MKS vs FPS)?
----- Model Configuration ----------------------------------------------------
1 NumSect: Number of circumferential sectors.
5000 MaxIter: Max number of iterations for induction factor.
1.0e-6 ATol: Error tolerance for induction iteration.
1.0e-6 SWTol: Error tolerance for skewed-wake iteration.
----- Algorithm Configuration ------------------------------------------------
True TipLoss: Use the Prandtl tip-loss model?
True HubLoss: Use the Prandtl hub-loss model?
True Swirl: Include Swirl effects?
True SkewWake: Apply skewed-wake correction?
True AdvBrake: Use the advanced brake-state model?
True IndProp: Use PROP-PC instead of PROPX induction algorithm?
True AIDrag: Use the drag term in the axial induction calculation?
True TIDrag:Use the drag term in the tangential induction calculation?
----- Turbine Data -----------------------------------------------------------
2 NumBlade: Number of blades.
7.90 RotorRad: Rotor radius [length].
0.79 HubRad: Hub radius [length or div by radius].
0.0 PreCone: Precone angle, positive downwind [deg].
0.0 Tilt: Shaft tilt [deg].
0.0 Yaw: Yaw error [deg].
20.0 HubHt: Hub height [length or div by radius].
44 NumSeg: Number of blade segments (entire rotor radius).

63
RElm Twist Chord AFfile PrntElem
0.9472 9.20353 0.4895 1 True
1.1775 9.23081 0.5212 1 True
1.3305 9.25567 0.5420 1 True
1.4905 9.27842 0.5640 1 True
1.6491 9.29932 0.5852 1 True
1.8064 9.35897 0.6058 1 True
1.9637 9.42343 0.6259 1 True
2.1210 9.49169 0.6455 1 True
2.2783 9.54950 0.6641 1 True
2.4357 9.59444 0.6813 1 True
2.5930 9.62713 0.6971 1 True
2.7508 9.63234 0.7115 1 True
2.9092 9.65505 0.7246 1 True
3.0678 9.63198 0.7356 1 True
3.2264 9.57839 0.7443 1 True
3.3850 9.49454 0.7505 1 True
3.5436 9.38023 0.7544 2 True
3.7022 9.23849 0.7560 2 True
3.8608 9.07271 0.7555 2 True
4.0194 8.88131 0.7529 3 True
4.1780 8.66941 0.7485 3 True
4.3366 8.43400 0.7423 3 True
4.4952 8.19385 0.7345 3 True
4.6538 7.93592 0.7254 4 True
4.8124 7.66548 0.7149 4 True
4.9710 7.38836 0.7033 4 True
5.1296 7.10388 0.6907 5 True
5.2882 6.81625 0.6773 5 True
5.4468 6.52864 0.6632 5 True
5.6054 6.24135 0.6485 5 True
5.7640 5.95927 0.6334 6 True
5.9228 5.70883 0.6180 6 True
6.0824 5.42262 0.6024 6 True
6.2426 5.16921 0.5867 7 True
6.4028 4.92546 0.5709 8 True
6.5630 4.69006 0.5548 8 True
6.7232 4.46204 0.5386 9 True
6.8834 4.23906 0.5220 9 True
7.0407 4.02065 0.5054 10 True
7.1941 3.79160 0.4889 10 True
7.3465 3.59062 0.4720 10 True
7.5016 3.10841 0.4544 11 True
7.6604 2.59702 0.4355 11 True
7.8201 2.34961 0.4160 11 True
----- Aerodynamic Data -------------------------------------------------------
1.225 Rho: Air density [mass/volume].
0.00001625 KinVisc: Kinematic air viscosity
0.000 ShearExp: Wind shear exponent (1/7 law = 0.143).
True UseCm: Are Cm data included in the airfoil tables?
11 NumAF: Number of airfoil files.
"_r_R_0.400_.dat" AF_File: List of NumAF airfoil files.
"_r_R_0.470_.dat"
"_r_R_0.540_.dat"
"_r_R_0.610_.dat"
"_r_R_0.680_.dat"
"_r_R_0.750_.dat"
"_r_R_0.790_.dat"
"_r_R_0.830_.dat"
"_r_R_0.870_.dat"
"_r_R_0.910_.dat"
"_r_R_0.950_.dat"
----- I/O Settings -----------------------------------------------------------
True TabDel: Make output tab-delimited (fixed-width otherwise).
False KFact: Output dimensional parameters in K (e.g., kN instead on N)
True WriteBED: Write out blade element data to "<rootname>.bed"?
False InputTSR: Input speeds as TSRs?
Mps SpdUnits: Wind-speed units (mps, fps, mph).
----- Combined-Case Analysis -------------------------------------------------
1 NumCases: Number of cases to run. Enter zero for parametric analysis.
WS or TSR RotSpd Pitch Remove following block of lines if NumCases is zero.
18 80 0
----- Parametric Analysis (Ignored if NumCases > 0 ) -------------------------

64
Annex F Computation times
For each part of the program, a simple python stopwatch routine is used to be able to pinpoint the time
consuming procedures. Looking closer into these (see figure 27).

Following observations are made:

• The initial model configuration in which the blade segment's are configured in Abaqus, requires
31 seconds (read mdb object original). It could be possible to further streamline this process in
order to decrease computation time

• Executing WT_Perf and transferring the loads to Abaqus only, can be considered insignificant
with respect to the time required by the other processes

• As expected, the most time consuming part is the FEA (first analysis 136 seconds, second
iteration, 162 seconds). It is also worthwhile nothing that due to the coupling scheme and the
nature of how the analysis is set-up in Abaqus (adding a new analysis step for each iteration),
computation time approximately increases with 20-30 seconds per iteration. As a result, the FEA
already requires twice the computation time at iteration 5 when compared to the initial analysis.

Figure 48: Partial time log of the coupling procedure

single core 1.7 GHz, 2GB laptop dual core, 2.6 GHz, 2GB Desktop

computation computation
case iterations case iterations
time [sec] time [sec]

Case 1 582.3 3 Case 1 779.9 3


Case 2 448.5 2 Case 2 651.0 3
Case 3 3040.2 5 Case 3 1356.0 6
Case 4 2385.1 4 Case 4 1169.0 5
Case 5 4212.4 6 Case 5 1930.2 8

Figure 49: computation times for various cases on two different machines. Note that the computation
times on the desktop (left side of the picture) are a result from an older version of the BEM-FEM coupling,
hence the difference in number of iterations. It is included here to indicate the impact the machine has on
computation times.

65
References

Scientific articles

[1] Hui Chen, Wenbin Yuy, Mark Capellaro: “Critical assessment of VABS against other wind turbine
blade tools”, May 18th, 2009,
http://hifi-comp.googlegroups.com/web/AssessmentTurbineBlades.pdf

[2] Yu, W: “Efficient High-Fidelity Simulation of Multibody Systems with Composite Dimensionally
Reducible Components”, Journal of the American Helicopter Society, vol. 52, no. 1, 2007, pp. 49-57,
http://www.engineering.usu.edu/mae/faculty/wenbin/papers/BucklingVibration.pdf

[3] Jieun Ku, Vitali V. Volovoi, and Dewey H. Hodges: “Multilevel-Multiphase Optimization of Composite
Rotor Blade with Surrogate Model ”, 48th AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics,
and Materials Conference 2007, Hawaii, AIAA-2007-1900

[4] D. H. Hodges, W. Yu: “A rigorous engineer-friendly approach for modelling realistic composite
rotor blades”, Wind Energy 2007 10:179–193, http://dx.doi.org/10.1002/we.215

[5] Paul Veers, Gunjit Bir, Donald Lobitz: “Aeroelastic tailoring in wind-turbine blade applications”
Presented at Windpower '98, American Wind Energy Association Meeting and Exhibition, May 1998,
http://www.sandia.gov/wind/other/AWEA4-98.pdf

[6] Alireza Maheri, Siamak Noroozi, John Vinney: “Application of combined analytical/FEA coupled aero-
structure simulation in design of wind turbine adaptive blades”, Renewable Energy 32 (2007) 2011-
2018, http://dx.doi.org/10.1016/j.renene.2006.10.012

[7] Scott Larwood, Mike Zuteck “Swept wind turbine blade aeroelastic modeling for loads and dynamic
behavior”, Windpower 2006 Conference (USA),
http://flight.engr.ucdavis.edu/~smlarwood/documents/Larwood_Windpower_2006.pdf

[8] Mike Zuteck: “Adaptive blade concept assessment: Curved Planform Induced Twist Investigation”,
SANDIA report 2002 (SAND2002-2996), http://www.sandia.gov/wind/other/022996.pdf

[9] Don W. Lobitz, Paul S. Veers: “Aeroelastic behavior of twist-coupled HAWT blades”, Aerospace
Sciences Meeting and Exhibit, 36th, and 1998 ASME Wind Energy Symposium, Reno, NV, Jan. 12-15, 1998,
Collection of Technical Papers (A98-16844 03-44), AIAA-98-0029

[10] W.C. De Goeij, M.J.L. van Tooren, A. Beukers: “Implementation of bending-torsion coupling in the
design of a wind turbine rotor blade”, Applied Energy 63 (1999) 191-207,
http://dx.doi.org/10.1016/S0306-2619(99)00016-1

[11] Thomas D. Ashwill, Paul S. Veers, James Locke, Ivan Contreras, Dayton Griffin, Mike D. Zuteck:
“Concepts for adaptive wind turbine blades”, AIAA-2002-0028

[12] Dayton A. Griffin (principal investigator): “Evaluation of Design Concepts for Adaptive Wind
Turbine Blades”, SANDIA report 2002 (SAND2002-2424), http://dx.doi.org/10.2172/801399

[13] Alireza Maheri, Siamak Noroozi, Chris A. Toomer, John Vinney: “WTAB: a computer program for
predicting the performance of horizontal axis wind turbines with adaptive blades”, Renewable Energy
31 (2006) 1673–1685, http://dx.doi.org/10.1016/j.renene.2005.09.023

66
[14] M.O.L. Hansen, J.N. Sørensen, S. Voutsinas, N. Sørensen, H.Aa. Madsen: “State of the art in wind
turbine aerodynamics and aeroelasticity”, Progress in Aerospace Sciences 42 (2006), 285-330,
http://dx.doi.org/10.1016/j.paerosci.2006.10.002

[15] Gerard Schepers, Arne Van Garrel: “AWSM: free wake lifting line model for wind turbine
aerodynamics”, EWEC 2006 conference proceedings,
http://www.ewec2006proceedings.info/pdf/servlet01cgi/ewec.php?id=260

[16] D. Kaufer: “Coupling a Free Vortex Wake Model and a Multi-Body-Simulation Code for Aero-Elastic
Simulation of Wind Turbines“,
Student Thesis 12/2007, SWE/IAG -Universität Stuttgart

[17] S. Hauptmann, D. Kaufer, M. Kühn: “Wind Turbine Simulation Using a Coupled Free Wake and
Multibody System Code”, DEWEK 2008 conference presentation and abstract
http://08.dewek.de/fileadmin/pdf/boa_web_2008.pdf (page 105)

[18] S. Streiner, S. Hauptmann, M. Kühn, E. Krämer: “Coupled Fluid-Structure Simulations of a Wind


Turbine Rotor”, DEWEK 2008 conference presentation and abstract,
http://08.dewek.de/fileadmin/pdf/boa_web_2008.pdf (page 106)

[19] Anders Ahlström: “Influence of Wind Turbine Flexibility on Loads and Power Production”, Wind
Energy 2006; 9:237–249, http://dx.doi.org/10.1002/we.167

[20] Timothy J. Knill: “The application of aeroelastic analysis output load distributions to finite
element models of wind”, Wind Engineering Volume 29 Number 2 March 2005 pp. 153-168(16),
http://dx.doi.org/10.1260/0309524054797104

[21] Marshall L. Buhl Jr, Alan D. Wright, Kirk G. Pierce: “Wind turbine design codes: a comparison of the
structural response”, 2000 ASME Wind Energy Symposium, 19th, Aerospace Sciences Meeting and
Exhibit, 38th, Reno, NV, Jan. 10-13, 2000, Collection of Technical Papers (A00-16885 03-44), AIAA 2000-
0022

[22] J.G. Schepers, J. Heijdra, K. Thomsen, T. Larsen, D. Foussekis, R. Rawlinson Smith, I.Kraan, B. Visser,
S. Oye, H. Ganander, I. Carlen S. Voutsinas, M. Belessis, L. Drost: “Verification of European wind turbine
design codes”, ECN report 2001, http://www.ecn.nl/docs/library/report/2001/rx01059.pdf

[23] J.G. Schepers, J. Heijdra, D. Foussekis, S. Øye, R. Rawlinson Smith, M. Belessis K. Thomsen, T. Larsen,
I. Kraan, B. Visser, I. Carlen, H. Ganander, L. Drost: “Verification of European WindTurbine Design
Codes, VEWTDC; Final report”, ECN report 2002: ECN-C-01-055,
http://www.ecn.nl/fileadmin/ecn/units/wind/docs/Other/finrep.pdf

[24] Jason Jonkman, Marshall Buhl, “New Developments for the NWTC's FAST Aeroelastic HAWT
Simulator”, 42nd AIAA Aerospace Sciences Meeting and Exhibit, Reno, Nevada, Jan. 5-8, 2004, AIAA-2004-
504, or http://www.nrel.gov/docs/fy04osti/35077.pdf

[25] M.L. Buhl Jr, A. Manjock: “A Comparison of wind turbine aeroelastic codes used for certification”,
NREL report 2006: NREL/CP-500-39113, 44th AIAA Aerospace Sciences Meeting and Exhibit Reno, Nevada
January 9–12, 2006 http://www.nrel.gov/docs/fy06osti/39113.pdf

[26] N.P. Duineveld: “Structure and possibilities of the FOCUS design package”,
Delft Wind Workshops October 2008,
http://www.windworkshops.tudelft.nl/files/presentaties/4B/Duineveld/DWW2008%20Duineveld
%20Structure%20and%20possibilities%20of%20the%20FOCUS%20design%20package.pdf

67
[27] Andrew T Lee, Riachard GJ Flay: “Compliant blades for wind turbines”,
IPENZ Transactions, 1999, Vol 26 no 1/EMCh,
http://www.ipenz.org.nz/ipenz/Publications/transactions/Transactions99/EMCh/Lee.PDF

[28] Alireza Maheri: “aero-structure simulation and aerodynamic design of wind turbines utilising
adaptive blades” Phd Thesis 2006, Faculty of Computing, Engineering and Mathematical Sciences,
University of the West of England-Bristol

[29] Alireza Maheri, Siamak Noroozi, John Vinney: “Combined analytical-FEA-based coupled aero
structure simulation of a wind turbine with bend–twist adaptive blades”,
Renewable Energy 32 (2007) 916-930, http://dx.doi.org/10.1016/j.renene.2006.04.007

[30] Alireza Maheri, Siamak Noroozi, John Vinney: “Decoupled aerodynamic and structural design of
wind turbine adaptive blades”, Renewable Energy 32 (2007) 1753-1767,
http://dx.doi.org/10.1016/j.renene.2006.11.004

[31] D.M. Somers, “The S819 S820 and S821 Airfoils”, NREL report (1993),
http://www.nrel.gov/docs/fy05osti/36334.pdf

[32] Ladean R. McKittrick, Douglas S. Cairns, John Mandell, David C. Combs, Donald A. Rabern, and R.
Daniel Van Luchene: “Analysis of a composite blade design for the AOC 1550 wind turbine using a
finite element model”, SANDIA report 2001 (SAND2001-1441), http://prod.sandia.gov/techlib/access-
control.cgi/2001/011441.pdf

[33] Peter Baek, Peter Fuglsang, “Experimental Detection of Transition on Wind Turbine Airfoils”, EWEC
2009 conference proceedings,
http://www.ewec2009proceedings.info/allfiles2/302_EWEC2009presentation.pdf

[34] J.L. Tangler, D.M. Somers: “NREL Airfoil famillies for HAWTs”, AWEA Windpower '95 conference
proceedings, http://wind.nrel.gov/designcodes/papers/NREL%20Airfoil%20Families%20for%20HAWTs.pdf

[35] M. Mezine, A. Willems, B. Léonard, M. Pottiez, D. Vandepitte and A. De Broe: “Fluid structure
interaction on flexible wind turbine blades using FINE/Hexa and Abaqus” 2009

[36] Tony Burton, David Sharpe, Nick Jenkins, Ervin Bossanyi: “Wind Energy Handbook”,
ISBN: 0-471-48997-2

[37] J.F. Manwell, J.G. McGowan, A.L. Rogers: “Wind Energy Explained”, ISBN 978-0-470-01500-1

[38] Carlos A. Felippa, Lecture notes: “Introduction to Finite Element Methods (ASEN 5007)”,
Department of Aerospace Engineering Sciences, University of Colorado at Boulder (USA)
http://colorado.edu/engineering/CAS/courses.d/IFEM.d/

[39] M.A. Crisfield: “Non-linear Finite Element Analysis of Solids and Structures Volume 1 :
Essentials”, Wiley, 1991, ISBN: 978-0-471-92956-7

[40]. B. Montgomerie, A. Brand, J. Bosschers, R. van Rooij: “Three-dimensional effects in stall”,


ECN-C--96-079, The Energy Research Center of the Netherlands, ECN, 1997

68
Software and manuals

[41] NWTC Design Codes (WT_Perf by Marshall Buhl), “WT_Perf v3.10, 17-December-2004 ”,
http://wind.nrel.gov/designcodes/simulators/wtperf/

[42] Dr. Craig Hansen, NWTC Design Code AirfoilPrep


http://wind.nrel.gov/designcodes/preprocessors/airfoilprep/ Last modified 16-January-2007

[43] Mark Drela and Harold Youngren, XFOIL version P4-v6.94,


http://web.mit.edu/drela/Public/web/xfoil/

[44] Mark Drela and Harold Youngren, “XFOIL 6.9 User Primer”, XFOIL documentation last updated on 30
November 2001

[45] “Eppler Airfoil Design and Analysis Code”, http://www.airfoils.com/eppler.htm

[46] Python version 2.4.4, http://www.python.org/download/releases/2.4.4/

[47] NumPy version 1.2.1, numpy-1.2.1-win32-superpack-python2.4.exe,


http://sourceforge.net/projects/numpy/files/

[48] SciPy version 0.7.0, scipy-0.7.0-win32-superpack-python2.4.exe,


http://sourceforge.net/projects/scipy/files/

[49] Code coupling interface MpCCI, http://www.mpcci.de/mpcci.html

[50] Numeca International AutoBlade, http://www.numeca.com/index.php?id=28

69

You might also like