Ships and Offshore Structures

ISSN: 1744-5302 (Print) 1754-212X (Online) Journal homepage: http://www.tandfonline.com/loi/tsos20

Simulation of the fluid-structure interaction of a

floating wind turbine

Bjarne Wiegard, Lars Radtke, Marcel König, Moustafa Abdel-Maksoud &

Alexander Düster

To cite this article: Bjarne Wiegard, Lars Radtke, Marcel König, Moustafa Abdel-Maksoud &
Alexander Düster (2019): Simulation of the fluid-structure interaction of a floating wind turbine,
Ships and Offshore Structures, DOI: 10.1080/17445302.2019.1565295

To link to this article: https://doi.org/10.1080/17445302.2019.1565295

Published online: 11 Jan 2019.

Submit your article to this journal

Article views: 3

View Crossmark data

Full Terms & Conditions of access and use can be found at

http://www.tandfonline.com/action/journalInformation?journalCode=tsos20
SHIPS AND OFFSHORE STRUCTURES
https://doi.org/10.1080/17445302.2019.1565295

Simulation of the fluid-structure interaction of a floating wind turbine

a
Bjarne Wiegard , Lars Radtkea, Marcel Königa, Moustafa Abdel-Maksoudb and Alexander Düstera
a
Institute for Ship Structural Design and Analysis, Hamburg University of Technology (TUHH), Hamburg, Germany; bInstitute for Fluid Dynamics and
Ship Theory, Hamburg University of Technology (TUHH), Hamburg, Germany

ABSTRACT ARTICLE HISTORY

In this paper, we present a method for the analysis of the motion behaviour and also the structural Received 2 October 2018
response – stresses and deformations – of a floating wind turbine. A partitioned approach is chosen to Accepted 28 December 2018
solve the fluid-structure interaction problem. Therefore, our C++ library comana, developed to couple
KEYWORDS
various solvers, is enhanced to couple the fluid solver panMARE and the structural solver ANSYS. Floating wind turbine; fluid-
Initially, a simple elastic finite element model is used in the coupled analysis. This finite element model structure interaction;
is described, and some results of the coupled simulation are presented. In order to use a more detailed partitioned solution
finite element model without drastically increasing the computation time, superelements can be approach; substructuring;
employed. This procedure is described and applied to an example, a floating buoy in waves, to superelement
demonstrate its applicability in fluid-structure interaction simulations.

Nomenclature Sk Operator representing structural field solution

tk Time at time step k
Aik Operator to include convergence acceleration tk+1 Traction at time step k + 1
ca Added mass coefficient tAk+1 Traction in aerodynamic field at time step k + 1
C Damping matrix tH Traction in hydrodynamic field at time step
k+1
Ĉ Reduced damping matrix k+1
dk+1 Discretized displacements at time step k + 1 T Transformation matrix
dik+1 Discretized displacements at time step k + 1 and u Displacement vector
i
iteration i û Reduced displacement vector
d̃k+1 Modified discretized displacements at time step u̇ Velocity vector
k + 1 and iteration i

u̇ Reduced velocity vector
eabs Absolute convergence tolerance ü Acceleration vector
erel Relative convergence tolerance

ü Reduced acceleration vector
F Load vector um Displacement vector containing master node
F̂ Reduced load vector displacements
Fm Load vector containing master node loads us Displacement vector containing slave node
Fs Load vector containing slave node loads displacements
Fk Operator representing fluid field solution Vm Volume of mooring
FkA Operator representing aerodynamic field rc Corrected mooring density
solution rm Mooring density
FkH Operator representing hydrodynamic field rw Water density
solution vi Relaxation factor
g Gravity constant Dt Time step size
gc Corrected gravity
Gsm Sub matrix of transformation matrix T
I Identity matrix 1. Introduction
K Stiffness matrix
K̂ Reduced stiffness matrix In recent years, floating wind turbines (FWTs) have evolved as
Kmm Stiffness matrix containing values for master an attractive alternative to conventional fixed-foundation tur-
nodes bines, which are limited to on- or near-shore areas. FWTs, in
Kms Stiffness matrix containing values for master and
slave nodes contrast, can be installed in greater water depths and are able
Ksm Stiffness matrix containing values for master and to harvest the stronger and steadier winds further away from
slave nodes the coast. However, they may also experience significantly
Kss Stiffness matrix containing values for slave higher hydrodynamic forces and accelerations due to their
nodes motion in the seaway. In order to ensure their structural integ-
ma Added mass
M Mass matrix rity and also assess their performance, it is essential to take the
M̂ Reduced mass matrix fluid-structure interaction (FSI) into account.
rik+1 Residual displacement at time step k + 1 and FSI problems can generally be solved with monolithic or
iteration i partitioned solution approaches. In monolithic approaches,

CONTACT Bjarne Wiegard [email protected] Institute for Ship Structural Design and Analysis, Hamburg University of Technology (TUHH), Am Schwar-
zenberg Campus 4 C, Hamburg D-21073, Germany
© 2019 Informa UK Limited, trading as Taylor & Francis Group
2 B. WIEGARD ET AL.

the coupled problem is solved as a whole, usually by applying model to generate so-called superelements is summarised. A
one numerical method (Bazilevs et al. 2013). Partitioned sol- simple example serves to demonstrate its applicability in the
ution approaches usually employ specialised numerical FSI. The method will be utilised in the next stage of the project,
methods for the individual subproblems and allow to reuse to enable the use of a detailed FE model of the FWT in the
existing software or solvers (Bungartz and Schäger 2006; Bun- coupled simulation.
gartz et al. 2010). The coupling is often realised by a coupling
manager, through which boundary values are exchanged. To
solve an FSI problem, a structural solver solves for the displace- 2. Floating wind turbine concept
ments at the structural model’s boundary, which are sent over
to the fluid solver. The fluid solver then computes the traction, At water depths greater than about 50 m, a fixed foundation of
which in turn is applied to the structural model. This procedure an offshore wind turbine (OWT) becomes uneconomical
continues until a global equilibrium is obtained. The iterative (European Wind Energy Association 2013). In deeper waters,
coupling procedure, however, make partitioned solution OWTs can be installed economically by using floating foun-
approaches more prone to instabilities as opposed to mono- dations – thus opening up a large potential for wind energy
lithic ones. Yet, novel convergence acceleration methods such use. Currently, there are three main types of floating foun-
as the quasi-Newton least squares (QNLS) method are able to dations for FWTs: the spar buoy, the tension leg platform
stabilise the procedure and reduce the number of coupling iter- (TLP), and the semi-submersible platform (semi-sub). An over-
ations significantly, see e.g. Küttler and Wall (2009) and view is shown in Figure 1. The spar buoy is a very large cylinder.
Degroote et al. (2009). In view of this and the increased flexi- Here, stability is achieved with heavy ballast in the lower part of
bility, a partitioned solution approach seems to be an appropri- the cylinder. Due to the very large draft of the spar buoy, this
ate choice for the FSI simulation of FWTs. type of FWT-foundation can only be installed in deeper waters.
In the joint research project ‘Hydrodynamic and Structural The tension leg platform is a platform with large buoyancy,
Optimization of a Semi-submersible Offshore Wind Turbine’ anchored with mooring lines which are tensioned in order to
(HyStOH), a new FWT concept (Cruse 2016) is developed – add stability. In the case of the semi-sub, stability is realised
and several aspects are investigated in subprojects. This paper by a large semi-submersed platform that is partly filled with bal-
is primarily based on the results so far achieved within the sub- last water. Further information on these three concepts is given
project ‘Fluid-Structure Interaction and Optimization of a in European Wind Energy Association (2013).
Floating Platform for Offshore Wind Turbines’ (FSIOpt). The The 6 MW FWT shown in Figure 2 is currently being devel-
aim within this subproject is to employ the partitioned solution oped in the joint research project. Its concept is based on a
approach to analyze the FSI of FWTs. semi-sub foundation. Unlike conventional systems, the entire
For the analysis of the stresses within the structure of an FWT including the platform self-aligns in the wind. This
FWT, an elastic finite element (FE) model is required. How- becomes possible because the platform is attached to the moor-
ever, the use of an elastic FE model significantly increases the ing at a single point with a revolute joint. The power cable is
computation time compared to a rigid model, because the connected with a sliding contact allowing the FWT to rotate
number of degrees of freedom is significantly increased. The without a limit. Furthermore, the FWT is a so-called downwind
direct use of a detailed FE model of a wind turbine in an FSI turbine, which means that the rotor is positioned behind the
simulation is therefore numerically quite expensive. In this nacelle in the wind – which leads to a large self-aligning
paper, a method is presented that benefits from the aforemen- moment acting on the entire FWT. This has the following
tioned high flexibility of the partitioned solution approach advantages: A yaw drive to adjust the nacelle to the wind direc-
because well-developed fluid and structural solvers can be tion is not needed, which significantly reduces the weight of the
coupled and associated powerful FE software can be used. nacelle and, hence, also the counterweight in the platform. In
This allows the use of the substructuring method (generation addition, the shape of the tower can be optimised because,
of superelements), which keeps the computation time in the
coupled simulation low while enabling a high resolution of
the FE model at the same time.
For the fluid problem, a fluid solver based on potential the-
ory and employing the boundary element method (BEM) is
chosen, which enables a fast computation. In the further course
of the project, the fluid model presented in this paper will be
replaced by a more detailed fluid model, which is being devel-
oped in another subproject with the focus on the specifics of
wind turbines (Netzband et al. 2017).
This paper is structured as follows. First, in Chapter 2, the
FWT concept is presented. The partitioned solution approach
and the coupling of the fluid and the structural solver are
explained in Chapter 3. Then, in Chapter 4, the initially used
coarsely meshed FE model and the fluid model are described
and some simulation results are presented. In Chapter 5, a
method to condense degrees of freedom in a detailed FE Figure 1. FWT foundation types.
 SHIPS AND OFFSHORE STRUCTURES 3

subproblem is known at time tk , a solver computes the state

of the respective subproblem at time tk+1 = tk + Dt (Dt is the
time step size) in terms of boundary values at time tk+1 . For
each subproblem, an operator formulation is introduced.
The solution of the structural subproblem is then abbre-
viated as

dk+1 = Sk (tk+1 ), (1)

where dk+1 represents the discretized displacements and tk+1

are the discretized tractions. An evaluation of the operator Sk
corresponds to the solution of one time step of the structural
subproblem and may accordingly involve a computationally
expensive solution of a large nonlinear system. For a detailed
introduction to the FEM in the context of nonlinear structural
mechanics, the reader is referred to standard text books such as
Wriggers (2008) or Bonet and Wood (2008). Here, it is
sufficient to mention that the tractions tk+1 have to be provided
at the quadrature points of the finite element mesh, as usual.
The displacements dk+1 are evaluated at the positions of the
Figure 2 . 6 MW FWT (also: FOWT = floating offshore wind turbine) with self-align-
nodes of the fluid mesh drawing on the shape functions of
ing capability (source: CRUSE Offshore GmbH). the finite element model.
The solution of the fluid subproblems is abbreviated as

k+1 = Fk (dk+1 )
tH H
due to the self-alignment capability, the direction of maximum (2)
load does not change significantly. As a result, the tower weight
can be reduced and the geometry can be designed to achieve a and
low aerodynamic resistance. Moreover, the concept offers
tAk+1 = FkA (dk+1 ), (3)
advantages in terms of manufacturing and installation. Most
of the offshore work, such as mooring and installation, can where tH A
k+1 and tk+1 denote the hydrodynamic and the aerody-
be done using conventional tugs. For major maintenance and namic tractions, respectively. For convenience, a combined
repair work, the FWT can be detached from its mooring in fluid operator
order to pull it to a protected area or to a shipyard. This signifi-  H 
cantly reduces work time under harsh conditions. Two rotor Fk (dk+1 )
tk+1 = Fk (dk+1 ) = (4)
blades are used to reduce the height of the centre of gravity FkA (dk+1 )
and to ensure optimal aerodynamic efficiency.
is introduced. The concatenation in Equation (4) results from
the fact that each fluid subproblem delivers loads that only
3. Coupling of fluid and structural solver act on the respective part of the structure. Again, an evaluation
of the operator corresponds to a solution of one time step of the
The coupling manager comana, introduced in König et al. subproblems, based on potential flow theory and discretized
(2016), realises the coupling between the structural solver using the BEM. The details of the method are explained in
ANSYS (ANSYS Academic Research Mechanical, Release Netzband et al. (2017). The BEM provides the tractions at
18.2) and the fluid solver panMARE, see Berger et al. (2014) the panel centres of the fluid mesh, which are then interpolated
and Netzband et al. (2017). ANSYS is a commercial FEM soft- using mesh-based inverse distance weighting in order to obtain
ware with various functions for linear and nonlinear static and the values at the structural mesh’s quadrature points, as
dynamic FE analyses. It includes a preprocessor, various sol- required (König et al. 2016).
vers, and a postprocessor. The fluid solver panMARE is based The staggered coupling algorithm employed here can now
on the boundary element method (BEM), which eases the be stated in terms of a fixed-point iteration (FPI)
meshing process and drastically reduces the computation
time as compared to methods based on volume discretization. i+1
dk+1 = Sk (Fk (dik+1 )). (5)
During the coupling procedure, the solvers are treated as black
boxes – as shown in Figure 3. Given that the state of a Every time step starts with a prediction of the displacements
d0k+1 based on the solution of previous time steps. Then, the
fluid subproblems are solved to obtain the tractions, which
are then sent to the structural solver that computes new displa-
cements. A residual

rik+1 = dk+1
i+1
− dik+1 (6)

Figure 3. Interaction of the structural solver ANSYS and the fluid solver panMARE. is used to check whether the coupling iterations have
4 B. WIEGARD ET AL.

converged. Once rik+1 falls below a given tolerance eabs or To validate the results of the coupled simulation, several
benchmark studies were conducted. Numerical examples for
rik+1 partitioned FSI simulations using comana are presented in
, erel , (7)
r0k+1 König et al. (2016). The coupling of ANSYS and panMARE
was verified by comparing results from coupled simulations
the iteration is stopped and the simulation proceeds with the of rigid floating bodies with the results produced by panMARE
next time step. Due to the fact that a standard FPI as in utilising an included rigid body solver (Netzband et al. 2017).
Equation (5) is prone to instabilities, the displacement is
modified by a convergence acceleration method before it is
passed on to the fluid solver. The resulting modified FPI can 4. Coupled fluid-structure interaction simulation
be written as with floating offshore wind turbine
 i  In accordance with the procedure described in the previous
k+1 = Sk Fk (d̃k+1 ) ,
di+1 (8) chapter, a coupled FSI simulation of the FWT is performed.
The aim is to calculate the motion of the wind turbine caused
where
by sea and wind, and to analyze the associated stresses and
i+1   deformations in the structure. The structural and the fluid
d̃k+1 = Aik di+1
k+1 (9)
model are described in the following sections. Afterwards, we
denotes the modified displacement. The operator Aik represents will outline the FSI simulation and present some of the simu-
an arbitrary convergence acceleration method. While a simple lation results. Detailed results will be published in a later project
relaxation with stage when the fluid and structural model have been further
developed. Now, the focus is on the method.
i+1  
d̃k+1 = dik+1 + vi di+1
k+1 − dk+1
i
(10)
4.1. Structural model
may already yield stable coupling iterations, we prefer the
QNLS method from Degroote et al. (2009) because it serves Figure 4 shows a general overview of the FWT structure. The 80
to significantly reduce the number of coupling iterations, as m long and 55 m wide platform has a draft of 13 m. It consists
demonstrated in König et al. (2016). To further improve the of four 23.5 m high floaters, which are connected to each other.
efficiency, we apply the modifications of the QNLS method The mooring is attached at the fore below a floater. The wind
introduced in Haeltermann et al. (2016) and Radtke et al. turbine is positioned on the aft floater. In order to reduce the
(2016), namely a filtering and a resetting technique, respect- bending moment in the ca. 80 m high tower, struts are arranged
ively. However, it has to be noted that dynamic relaxation to distribute the load into the middle floaters. Each rotor blade
methods that adaptively compute optimised relaxation factors has a length of 70 m. The platform is partly filled with ballast
vi – such as the Aitken relaxation-like method introduced in water to achieve the necessary restoring moment to ensure
Irons and Tuck (1969) and investigated in Küttler and Wall longitudinal and transverse stability. Because of the self-align-
(2009) – are reasonable alternatives in view of a reduced ing capability of the entire FWT, the ballast water levels don’t
implementation effort. have to be controlled in service. In the following paragraphs,

Figure 4. General overview on the FWT structure (source: CRUSE Offshore GmbH).
 SHIPS AND OFFSHORE STRUCTURES 5

the FE model is described. More detailed structural data will be pressure, but this FE model focuses on the global structural
published at a later stage of the project when the structural response.
design is completed. The mooring consists of five mooring lines of 350 m length,
The FE model of the FWT is shown in Figure 5. The plat- arranged with an angle of 72° to each other. The mooring line
form including its inner structure is modelled with 4-node arrangement is illustrated in Figure 6. The water depth is
underintegrated shell elements, based on the Reissner-Mindlin assumed to be 40 m. The mooring is anchored at the seabed
theory. The maximum edge length of the shell elements is 5 and attached to the fore floater of the platform with a buoy
m. The hydrostatic and hydrodynamic loads acting at the type that is commonly used in the offshore oil and gas industry.
outer surface of the platform are calculated in panMARE. In moderate sea state and wind conditions, the largest part of
After interpolating and evaluating the panMARE results as the mooring lines lies on the sea bed. In the FE model, the
described in Section 3, nodal forces are computed using Gaus- mooring is modelled with beam elements based on the
sian quadrature. This process is done by comana. Using Timoshenko beam theory with low bending stiffness. Alterna-
ANSYS, the forces are then applied at the corresponding shell tively, link elements with or without compression stiffness
element’s nodes. The ballast water mass is included with solid can be used. The mooring elements are connected to the plat-
elements placed inside the platform. These solid elements form shell elements with multi point constraints that transfer
share nodes with the shell elements representing the platform the forces. As the platform is free to rotate around the mooring,
structure. The modulus of elasticity assigned to these elements no moments are transferred. The sea bottom is modelled with
is chosen very low so as to prevent a stiffness contribution. An one large shell element. Contact is defined between this shell
enhanced model of the ballast water would be necessary to ana- element and the beam element nodes of the mooring. So far,
lyze the local structural response due to hydrostatic water the mooring was not included into the fluid model. The inertial

Figure 5. FE model of the FWT without (left) and with aerofoils (right) (mooring not displayed).

Figure 6. Mooring to platform connection with 5 mooring lines and power cable (source: CRUSE Offshore GmbH) (left) and FE model, including mooring in initial static
equilibrium state (right).
6 B. WIEGARD ET AL.

fluid reactions on the mooring elements are represented by an are coupled with multi point constraints to the nearest beam
added mass. With reference to Antonutti (2015), the added nodes. The surface effect elements follow the deformation of
mass ma is assumed as the beam elements without contributing to the mass and stiff-
ness of the structure. The same procedure is followed to realise
ma = ca rw Vm (11) the load transfer at the rotor’s surface.
where ca is the added mass coefficient, rw is the water density, Consequently, the global deformations of the FWT due to
and Vm is the volume of the mooring lines. The added mass the aerodynamic and hydrodynamic loads can be calculated,
coefficient for mooring lines is assumed to be ca = 1. This is and the associated stresses can be analyzed. The FE simulation
in most cases correct for the added mass in normal direction, of the entire FWT is carried out geometrically nonlinear
whereas the added mass in tangential direction is usually smal- because large displacements and rotations occur.
ler (Antonutti 2015). This modelling of the added mass is sub-
ject to further improvement. The added mass is included into
the model by replacing the mooring density rm with an 4.2. Fluid model
increased density rc (Antonutti 2015): The simulation is carried out using the inhouse code pan-
MARE, which is a three-dimensional first-order panel method.
rc = rm + ca rw (12)
For each fluid subproblem, the entire surface of the structural
The buoyancy is considered by a correction of the gravitational model is discretized using polygonal panel elements, see Katz
acceleration used for the mooring (Antonutti 2015): and Plotkin (2001) and Netzband et al. (2017). In the aerody-
  namic subproblem, we further discretize vortex sheets
ca + 1 (wakes) which shed from sharp trailing edges of the airfoils
gc = g 1 − (13)
ca + rm /rw (rotor blades, struts and tower). Here, the Kutta condition is
enforced. It states that the flow leaves a trailing edge smoothly
Rayleigh damping with coefficients αd = 0.1 (mass proportional
and that the velocity at this point is finite. The resulting mesh is
damping) and βd = 0 (stiffness proportional damping) is used to
shown in Figure 7. The kink apparent in the wake discretization
represent the damped behaviour of the mooring. In a later stage
is a result of the initial conditions and is transported out of the
of the research project, the mooring will be included into the
mesh within the first time steps of the simulation.
fluid model in order to enhance the simulation of its behaviour.
The solution process is based on potential flow theory. In
Prior to the coupled simulation, a static simulation is carried
short, the Laplace equation for the velocity potential is solved
out in ANSYS to compute the initial static equilibrium state of
approximately using the BEM. The surface of the immersed
the mooring.
body is represented by sources and doublets. In addition, the
The tower, struts, and rotor are modelled with 2-node beam
free form surfaces of the wake panels are simulated only by a
elements based on the Timoshenko beam theory, exhibiting a
doublet distribution. The pressure field is related to the velocity
tapered cross-section to represent the box girder structure.
potential through the unsteady Bernoulli equation. Based on
The beam element length ranges from 4 to 8 m in the tower/
the panel elements, the resulting equation is solved approxi-
struts and from 0.3 to 0.9 m in the rotor. The connection
mately by collocation as usual in BEMs. The collocation points
between the beam elements of the tower or the struts, respect-
are the element centres.
ively, and the shell elements of the platform are realised with
The underlying potential theory implies an incompressible,
multi point constraints in order to connect the lowest beam
irrotational, and inviscid flow. Accordingly, only pressure loads
node with all shell element nodes on top of the floater. This
acting in the normal direction are directly obtained from the
way, a realistic bending stiffness at the transition into the plat-
solution of the fluid subproblem. Drag forces acting on the
form is realised. The rotor is also connected with multipoint
body are computed by special means, namely a friction correc-
constraints, where a revolute joint is realised by coupling all
tion method based on the local Reynolds number and near wall
degrees of freedom of the node at the rotor centre and the con-
velocity.
necting node at the tower, in such a way that only rotation
The densities of water and air are set to 1026 and 1.2 kg/m³,
around the rotor axis is possible. The generator moment is
respectively. For the friction correction, a viscosity of
applied to this revolute joint, depending on the actual rotor fre-
1.48·10−5 Ns/m² is assumed for the air.
quency. The nacelle with its equipment is modelled as a point
mass connected to the tower.
As described above, the tower is aerodynamically optimised.
4.3. Coupled FSI simulation
Thus, a cover is installed around the box girder cross section of
the tower structure. This cover has an aerofoil form. The aero- The three subproblems, see Figure 8, modelled and coupled as
dynamic forces calculated in panMARE are acting on this aero- described above, are solved in order to simulate the motion and
foil surface. In order to enable an efficient and precise load the structural response of the FWT. Regular waves are mod-
transfer to the beam element nodes, this aerofoil is also elled with an amplitude of 3 m and a period of 8 s. The wind
included in the FE model (displayed in red in Figure 5). The speed is 12.3 m/s. The rotor is designed to rotate with ca.
aerofoil is meshed with so-called surface effect elements. 13.6 rounds per minute at this wind speed. This rotational
These elements can be used to represent the outer surfaces, speed is defined as the initial rotational speed in the structural
and they are able to transfer the aerodynamic load to the model. In the fluid model, an initial wake is created. The wind
beam nodes. Therefore, the nodes of the surface effect elements and wave direction are both aligned with the x-axis, which
 SHIPS AND OFFSHORE STRUCTURES 7

Figure 7. Fluid mesh.

Figure 8. Partitioning of the FSI problem.

matches the initial position of the FWT. Thus, no self-align- Detailed results will be published at a later stage of the research
ment of the FWT is investigated in this load case. project, when the structural model is further developed and all
The simulated time is 200 s. In general, different time step innovations in the fluid model achieved within the joint project
sizes can be defined for each subproblem. Here, a common are implemented.
time step size of Dt = 0.1 s is chosen. The implicit Euler time
integration method is used for the fluid problems, and the New-
5. Substructuring in the structural model
mark method is used for the time integration in the structural
problem. The Newmark parameters are defined as a = 0.4225 The FE model described in Section 4.1 was set up as simple as
and d = 0.8. possible. In order to enable the use of a detailed FE model, it is
The convergence acceleration with the QNLS method is possible to use a method known as substructuring. This method
based on the solution of the 10 previous time steps. Resetting – which is often used in the aerospace industry, but is also suit-
according to Radtke et al. (2016) is performed after 25 coupling able for FE analyses of ship (Wiegard et al. 2018) and offshore
iterations. The filtering tolerance for the technique described by structures – will be described in the following section. Then, its
Haeltermann et al. (2016) is set to 10−6 . The absolute and rela- use in the FSI simulation will be demonstrated with a simple
tive convergence tolerances for the coupling iterations are set to example.
eabs = 10−12 and erel = 10−3 .
Figure 9 shows the results at the start of the simulation and
5.1. Substructuring method
at t = 20 s, where a deformation of the rotor and a surge (recog-
nisable by the tension caused in the left mooring line) and a The substructuring method was developed to simplify large
small pitch motion can be seen. The wave elevation and the complex FE models and, thus, to reduce the computational
1st principal stress in the platform structure are plotted. time. For this purpose, the entire structure is divided into
8 B. WIEGARD ET AL.

Figure 9. FSI simulation results of FWT in waves and wind at t = 0 s and t = 20 s; the wave elevation (in m) and the 1st principal stress (in Pa) in the platform structure are
plotted (beam stresses in the tower, struts and rotor are not displayed).

parts, the so-called substructures. A detailed FE model is cre- so that us can be substituted in Equation (16), yielding
ated for each substructure. In the next step, the nodes of the
FE model are distinguished into master and slave nodes. [Kmm − Kms K−1 −1
ss Ksm ] um = Fm − Kms Kss Fs . (19)
Nodes that are important – e.g. because loads are acting Equation (19) can then be expressed as
there, or because these nodes are needed for connections to
other parts of the overall FE model – should be defined as mas- K̂ û = F̂ (20)
ter nodes. Other nodes can be defined as slave nodes. The aim is
where
to remove all degrees of freedom of the slave nodes from the
equation system by static condensation (Petyt 2010). First, K̂ = Kmm − Kms K−1
ss Ksm , (21)
this process is first explained for the static case – based on
the following equations published in ANSYS (2017). F̂ = Fm − Kms K−1
ss Fs (22)
The basic static equation
and
Ku=F (14)
û = um . (23)
with stiffness matrix K, displacement vector u, and load vector
If the slave nodes are chosen in such a way that there are no
F is sorted by master degrees of freedom (subscript m) and
forces acting on these nodes, the reduced force vector in
slave degrees of freedom (subscript s):
Equation (22) can be simplified to
 
Kmm Kms um Fm
= (15) F̂ = Fm . (24)
Ksm Kss us Fs
A transformation matrix
This equation system can be divided into    
I I
Kmm um + Kms us = Fm (16) T= = (25)
−K−1
ss Ksm Gsm
and
can be defined to calculate the reduced stiffness matrix, reduced
Ksm um + Kss us = Fs . (17) load vector, and reduced displacement vector as follows:

Now, Equation (17) is solved for us K̂ = TT K T (26)

us = K−1 −1
ss Fs − Kss Ksm um (18) û = TT u (27)
 SHIPS AND OFFSHORE STRUCTURES 9

with the mass M and damping matrix C transformed as follows

F̂ = TT F. (28) (Guyan 1965; ANSYS 2017):
As a result, the slave degrees of freedom are removed from M̂ = TT M T (31)
the equation system, which reduces the computational effort
during the solution of the overall equation system. Only the Ĉ = TT C T. (32)
degrees of freedom of the master nodes remain in the system
of equations. The reduced stiffness matrix K̂ can be saved, The substructuring method can be used in large rotation ana-
and it can be used in FE models to represent the stiffness lyses, as the exterior rotations can be separated from the
of the entire substructure as if the elements of the substruc- interior superelement deformations. However, the interior
ture were condensed into one single element. Thus, the term superelement deformations are limited to small deformations
superelement is also common in this context. The process because a linear behaviour, including material linearity, is a
where the reduced stiffness matrix is derived from the precondition for the applicability of the condensation process
detailed FE model is called generation pass. This process is explained above.
part of the preprocessing of an FE analysis with The substructuring method will be applied to the FSI simu-
superelements. lation of the FWT in the next stage of the research project. The
In the so-called use pass, the superelement is then used as advantage is that a detailed FE model can be used to generate
part of a larger FE model, similarly to a conventional one or more superelements in order to reduce the number of
element. The simulation result can be calculated faster due degrees of freedom to a number suitable for a partitioned FSI
to the lower number of degrees of freedom. Finally, an evalu- simulation involving many iterations – while a detailed solution
ation of stresses in the substructure can be done in the can be calculated in the post processing for all time steps or for
detailed model again. For this, the results at the slave time steps where maximum forces are acting. The substructur-
nodes are calculated based on the master node results by a ing method can be applied to the platform and the tower, but
backward substitution. This sub-process is called expansion not to the rotor – as the rotor deforms significantly, requiring
pass. a geometrical nonlinear solution.
The substructuring technique can also be applied to
dynamic load cases. Therefore, the general equation
5.2. Example: substructuring in FSI simulation of a
M ü+ Cu̇ + Ku = F (29) floating buoy
is transformed to Here, the applicability of the substructuring method in FSI
simulations is demonstrated by an example. We analyze the
ü+ Ĉ

M̂
u̇ + K̂û = F̂ (30) motion behaviour and the structural response of the buoy

Figure 10. FE model of the buoy with dimensions.

10 B. WIEGARD ET AL.

sketched in Figure 10. The lower part of the buoy is a stiffened

cylinder with a wall thickness of 10 mm, a diameter of 2 m, and
a height of 5 m. The cylinder is filled with heavy ballast, shifting
the centre of gravity downwards below the center of buoyancy.
A 5 m high mast rises out of the cylinder. The mast exhibits a
rectangular cross section with 5 mm wall thickness. A 1000 kg
point mass is positioned at the top of the mast, causing signifi-
cant bending stresses in the mast.
In the reference FE model (no superelements are used), the
cylinder including the inner structure and the mast are mod-
elled with 4-node reduced integrated shell elements. The ballast
is modelled with 8-noded solid elements sharing nodes with the
shell elements of the cylinder. The mast is meshed relatively
fine with 1964 nodes, using shell elements. The mass on top
of the mast is modelled with a mass element.
An FSI simulation is performed with a time step size of Dt =
0.1 s. The simulated time is 30 s. The buoy is floating in regular
waves with an amplitude of 1 m and a period of 10 s, see Figure
11. Only motions within the x-z-plane are enabled; out-of-
plane motions are suppressed in this simulation. In Figure Figure 12. FE model with superelement generated to represent the mast (left) and
13, the x- and z- positions of the mast top (point A, see Figure to represent entire buoy (right); the edges and master nodes of the superelement
are displayed.
10) are plotted over time. While the buoy is floating, significant
stresses occur in the mast at the transition to the cylinder. The
surface of the cylinder. The number of master nodes at the
first principal stress at points B and C (see Figure 10) is plotted
outer surface of the cylinder may be further reduced, but a cer-
over time in Figure 14.
tain number of nodes are necessary for the correct application
In addition to this FSI simulation, two simulations with FE
of the tractions calculated by panMARE.
models incorporating superelements are carried out. In one FE
The movement of the mast top is compared to the reference
model, see Figure 12 left, a superelement is generated for the
simulation in Figure 13. The results of both FE models match
mast. In this case, 80 nodes of the mast are defined as master
each other almost perfectly (curves are almost exactly on top
nodes, and 1884 nodes of the mast are defined as slave nodes.
of each other). The stresses in the mast at the transition to
Most of the master nodes are necessary for the connection
the cylinder structure are calculated in the post-processing
between mast and cylinder. The other master nodes are needed
(expansion pass), and a very good agreement is achieved here
for the connection to the point mass. In a further FE model, see
as well, see Figure 14.
Figure 12 right, the entire buoy, except the point mass, is con-
Consequently, it is also possible to use the substructuring
tained within a superelement. Here, the number of nodes is
method in FSI simulations – and it will be employed in FSI
reduced from 3145 nodes in the reference model to 658 master
simulations of the described FWT in the scope of the research
nodes plus the node for the mass element. Hereof, 16 master
project. Since the FE model in this example is quite small, a
nodes are defined at the mast, which are needed to connect
reduction in the computation time is not considerable. In the
the mass element to the mast with multi-point constraints.
case of the FWT, in contrast, the reduction of the degrees of
The remaining 642 master nodes are defined at the outer

Figure 11. Floating buoy in regular waves at t = 0 s (left) and t = 14.8 s (right); the wave elevation (in m) and the 1st principal stress (in Pa) in the buoy structure are
plotted.
 SHIPS AND OFFSHORE STRUCTURES 11

Figure 13. x- and z-displacement of mast top (point A) over time calculated with reference model and with FE models where superelements are used for mast only or for
the entire buoy.

Figure 14. 1st principal stress at point B (left) and point C (right) over time, calculated with reference model and with FE models where superelements are used for mast
only or for the entire buoy.

freedom will be substantial, leading to a significant reduction in Disclosure statement

solution time. No potential conflict of interest was reported by the authors.

6. Conclusions/Outlook Funding
The use of the partitioned solution approach for FSI simu- This research was funded by the Federal Ministry for Economic Affairs and
lations allows the coupling of various solvers. The coupling is Energy of Germany (BMWi) in the joint project HyStOH (project number
managed by our coupling software comana. In order to over- 03SX409C).
come instabilities, which can occur in partitioned solutions,
novel convergence acceleration methods such as the quasi-
ORCID
Newton least squares method are utilised. Due to the flexibility
within the partitioned solution approach, it is possible to use Bjarne Wiegard http://orcid.org/0000-0002-4155-9199
well-developed solvers and corresponding software. For the
FSI simulation of the presented FWT, the structural solver References
ANSYS and the fluid solver panMARE are coupled. The use
of the substructuring method implemented e.g. in ANSYS ANSYS. 2017. ANSYS 18.2 documentation: theory reference: 15. Analysis
procedures: 15.6 Substructuring analysis. Southpointe: ANSYS.
promises great opportunities in the simulation of the FWT Antonutti R. 2015. Numerical study of floating wind turbines: hydro- and
and in FSI simulations in general. The number of degrees of aero-mechanics [PhD thesis]. Edinburgh: University of Edinburgh.
freedom of detailed FE models can be reduced significantly, Bazilevs Y, Takizawa K, Tezduyar TE. 2013. Computational fluid-struc-
which enables the use of these models in the iterative coupled ture interaction: methods and applications. Chichester: Wiley.
solution process. Nonetheless, in the course of post-processing, Berger S, Druckenbrod M, Pergande M, Abdel-Maksoud M. 2014. A two-
stage optimization method for full-scale marine propellers working
a solution can be calculated for all slave nodes based on the sol- behind a ship. Ship Technol Res. 61(2):64–79.
utions of the master nodes. This enables a detailed analysis of Bonet J, Wood R. 2008. Nonlinear continuum mechanics for finite element
the structural response of FWTs. analysis. New York: Cambridge University Press.
12 B. WIEGARD ET AL.

Bungartz HJ, Schäger M, editors. 2006. Fluid-structure interaction, model- Katz J, Plotkin A. 2001. Low-speed aerodynamics. 2nd ed. Cambridge:
ling simulation and optimization. (Lecture notes in computational Cambridge university press.
Science and Engineering; vol. 53). Berlin: Springer. König M, Radtke L, Düster A. 2016. A flexible C++ framework for the par-
Bungartz HJ, Mehl M, Schäfer M, editors. 2010. Fluid-structure interaction titioned solution of strongly coupled multifield problems. Comput
ii, modelling simulation, optimisation. (Lecture notes in computational Math Appl. 72(7):1764–1789.
science and engineering; vol. 73). Berlin: Springer. Küttler U, Wall WA. 2009. Vector extrapolation for strong coupling
Cruse J. 2016. Floating wind turbine promises cost reductions. HANSA Int fluid-structure interaction solvers. J Appl Mech. 76(2):021205-1–
Maritime J. 153(9):173. 021205-7.
Degroote J, Bathe KJ, Vierendeels J. 2009. Performance of a new parti- Netzband S, Schulz S, Abdel-Maksoud M. 2017. A fully coupled simulation
tioned procedure versus a monolithic procedure in fluid-structure method for floating offshore wind turbine dynamics using a boundary
interaction. Comput Struct. 87(11):793–801. element method in the time domain. 10th International Workshop on
European Wind Energy Association. 2013. Deep water: the next step for Ship and Marine Hydrodynamics IWSH 2017.
offshore wind energy. Brussels, Belgium: European Wind Energy Petyt M. 2010. Introduction to finite element vibration analysis.
Association (EWEA). Cambridge, UK: Cambridge University Press.
Guyan RJ. 1965. Reduction of stiffness and mass matrices. AIAA J. 3 Radtke L, Larena-Avellaneda A, Debus E, Düster A. 2016. Convergence
(2):380–380. acceleration for partitioned simulations of the fluid-structure inter-
Haelterman R, Bogaers A, Scheufele K, Uekermann B, Mehl M. 2016. action in arteries. Comput Math Appl. 74(7):1675–1689.
Improving the performance of the partitioned QN-ILS procedure for Wiegard B, Ehlers S, Klapp O, Schneider B. 2018. Bonded window panes
fluid-structure interaction problems: filtering. Comput Struct. 171 in strength analysis of ship structures. Ship Technol Res. 65(2):102–
(Supplement C):9–17. 121.
Irons B, Tuck RC. 1969. A version of the Aitken Accelerator for Computer Wriggers P. 2008. Nonlinear finite-element-methods. Berlin and
implementation. Int J Numer Meth Eng. 1:275–277. Heidelberg: Springer-Verlag.