Journal of Wind Engineering & Industrial Aerodynamics 177 (2018) 117–135

Contents lists available at ScienceDirect

Journal of Wind Engineering & Industrial Aerodynamics

journal homepage: www.elsevier.com/locate/jweia

A multiply-partitioned methodology for fully-coupled computational

wind-structure interaction simulation considering the inclusion of arbitrary
added mass dampers
M

at
e P
entek *, Andreas Winterstein, Michael Vogl, P
eter Kup

as, Kai-Uwe Bletzinger,
Roland Wüchner
Lehrstuhl für Statik, Technische Universit€
at München, Arcisstr. 21, 80333 München, Germany

A R T I C L E I N F O A B S T R A C T

Keywords: Recent advances of the numerical wind tunnel result in a flexible methodological framework, which enables the
Fully-coupled simulation fully-coupled simulation of wind-structure interactions considering the arbitrary and modular inclusion of
Computational wind-structure interaction additional devices into the system – such as added mass dampers – for the mitigation of vibrations caused by the
Multiply-partitioned methodology wind flow. This feature could be seen as an add-on to what can be done with experimental methods and enables
Added mass dampers further possibilities when the predictive character of such simulations is used for structural design or even for the
Vibration mitigation
development and optimization of the additional devices themselves. The procedure promotes an approach fully-
solved in time domain using a multiply-partitioned concept to be able to deal with the respective components in
an efficient way. It is crucial to adopt a coupled approach and analysis in time as the involved systems can heavily
influence each other. The systematic preparation of each module accompanied by coupled simulations is pre-
sented, which aims to ensure and enhance the quality of the overall solution strategy. The effectiveness and
industrial relevance of the concept is presented on a tall building undergoing dynamic excitation due to vortex
shedding with an integrated vibration mitigation device.

1. Introduction the advances in both hardware as well as simulation techniques permit

Many recent developments have lead to numerical methods being
available for both research as well as industry on an everyday basis. On
the one hand, the unquestionable improvement in hardware and
lowering of computational costs has permitted the presence of powerful
computers. On the other hand, the discrete mathematical and algorithmic
part permits the modeling of individual physical phenomena. The
respective numerical implementations can be found in commercial and
research codes, the latter being frequently open-source.
It can be noted, that the simulation approaches for single field
problems, like structural mechanics or fluid dynamics, have reached a
significant level of maturity and recent research efforts concentrate on
the numerical treatment of coupled multi-field scenarios. These should
provide more detailed insight into existing – typically coupled – natural
phenomena by removing the need of simplifying assumptions of the
considered investigated systems. For each part involved, the most
adequate model ought to be selected, whilst the multiply-coupled
framework should permit the proper co-functioning of these. Hence,
the investigation of complex physical problems. The present contribution
discusses computational wind-structure interaction (WSI) with the
possible modular inclusion of various further devices in order to mitigate
wind-induced structural response.
In general, for the wind engineer, the number of features provided by
various numerical frameworks has been growing, sometimes over-
whelmingly. The result is, that more and more advanced numerical
techniques, realized as modules/parts, permit additional detailing com-
bined with improved modeling and insights into wind-induced effects.
These improvements range from the more accurate modeling of wind (for
e.g. using specific inlet generator accompanied with suitable numerical
modeling of the flow field) to allowing a more realistic quantification of
the coupled fluid-structure interaction (FSI) phenomena (permitting the
usage of structural models of various complexity and also the inclusion of
additional added mass devices to serve as a kind of controller). Numerical
coupling is mostly addressed in a partitioned manner, which allows the
individual implementation and availability of proper numerical models
(wind flow, structure, added device – here added mass damper AMD) and

* Corresponding author.
E-mail address: mate.pentek@tum.de (M. P
entek).

https://doi.org/10.1016/j.jweia.2018.03.010
Received 1 November 2017; Received in revised form 23 February 2018; Accepted 7 March 2018
Available online 24 April 2018
0167-6105/© 2018 Elsevier Ltd. All rights reserved.
M. P
entek et al.
an adequate interaction between these.
The current work discusses a systematic approach for the simulation
of multiply-coupled (and possibly multi-physical) problems in case of
computational wind engineering (CWE). Whereas most are familiar with
how the individual components should be modeled separately, one needs
special attention in order to obtain a co-functioning, which is both effi-
cient as well as physically accurate. A conscious choice and proper setup
of the individual components needs to be undertaken, which implies
comparisons with various other physical and numerical experiments, as
well with analytic results, where possible. This investigation needs to be
extended to the communication or coupling between the modules, that in
turn needs to be done in a smart and efficient way, so that it is fast,
reliable, and last but not least, scalable. The latter aspect should not be
neglected, as detailed numerical simulations are still relatively compu-
tationally expensive and lengthy. For practical purposes, an adequate
choice of models and selection of proper software-hardware is a must,
otherwise the common usage of such simulations remains utopic.
A systematic approach is proposed, which ensures quality showcased
with the example of a simplified generic highrise building geometry in
constant wind inflow undergoing excitation due to vortex shedding. The
resonant amplification is mitigated using an AMD (here various devices
being investigated, such as a classical tuned mass damper TMD and a
semi-active variation SATMD). The ideas and framework can be extended
to other structures, such as bridges, towers and so on, permitting the
usage of other control devices and algorithms as well.
The present contribution begins with the presentation of the WSI as
part of coupled physical problems in a numerical context. Modules are
identified, notions and specifics of the co-simulation are presented.
Winterstein et al. (2016) discuss the control of FSI in a computational
context with a strong emphasis on various control algorithms and
respective methodology, as well as its application to a well known
benchmark problem. P
entek et al. (2017) layout the formalized premise
for control by AMDs in a partitioned manner for usage in CWE, accom-
panied by initial studies. These previous contributions serve as a base for
the current discussion. Consequently, the numerical models of various
AMDs are introduced in the existing scheme. This is followed by the
description of the individual components. Various comparative studies of
these separate modules intend to ensure the appropriateness and quality,
these being backed up by selected results. A chosen example shows how
this methodology is relevant and usable for wind engineering (WE)
purposes, namely in case of a skyscraper undergoing heavy oscillations
due to vortex shedding. The numerical wind tunnel thus aims to
contribute to the knowledge provided by the more established physical
one, striving to provide additional insight and options where possible.
The challenges and potentials of the numerical approach are thoroughly
outlined and discussed in recent contributions by Kareem (2017) and
Bletzinger et al. (2016).
Current progress builds upon and ought to be related to previous
works which address various aspects of high-rise buildings excited by
wind forces. Our chosen example focuses on across-wind excitation. For
such cases, structural response on various buildings has been predicted
from early on using various mathematical models, such as forcing
Fig. 1. Key components and the possible extension of the numerical wind tunnel.
118
Journal of Wind Engineering & Industrial Aerodynamics 177 (2018) 117–135
functions suggested by Kareem (1983). Liang et al. (2002) further
contribute to this line of thought with an application on rectangular
buildings. Structures of such shape and slenderness generally are ex-
pected to exhibit dangerous behavior in low approaching turbulence, so
more and more accurate methods should be developed. Yang et al.
(2004) propose a benchmark problem for the control of wind-excited
tall buildings. Their work includes wind tunnel measurements for the
retrieval of loading time history on rigid models. These represent the
forces applied onto a simplified structural model. As such, an interac-
tion between the flow field and the building deformation is neglected.
This early work elaborates on the usage of control devices and
respective algorithms for mitigation purposes. The authors in Yang
et al. (2004) humbly note that the interaction between structure and
controller is also not considered, which might not be suitable for certain
use cases. We currently aim to address and remedy some of these
aforementioned shortcomings – for e.g. such as neglecting the inter-
action between various components – whilst contributing to and
exploiting the potentials of recent developments in numerical tools and
methods.
2. The modular extension of the numerical wind tunnel
A conceptual discussion aims to present the current state as well as
potential enhancements of the numerical wind tunnel achieved through
the extension of the staus quo. Focus lies on outlining the various existing
components of the numerical wind tunnel with an outlook to possible
enrichment. A brief part already discusses partitioned approaches
available for WSI simulation.
2.1. Components
A numerical wind tunnel is sketched in Fig. 1 depicting the main
building blocks. The fluid solver uses methods from computational fluid
dynamics (CFD) to simulate the airflow around the structure. The
availability of an inlet boundary condition (BC) or wind inlet generator
permits a more realistic inflow. The usage of simplified structural models
for fluid-structure interaction (FSI) assessment is proposed to reduce
modeling effort and computational necessities leading to the improve-
ment of the overall efficiency. The established parts can be extended with
other modules, in the current work we specifically refer to the inclusion
of various AMD models for vibration mitigation.
The involved modules need to be properly chosen and correctly setup
in order to function as intended. Also, an adequate interaction between
these components has to be ensured. These steps are detailed in the up-
coming sections.
2.2. Partitioned wind-structure interaction simulation
The assessment of wind effects using wind tunnels already represents
a certain level of abstraction. For experimental ones, this implies scaling
effects, for the numerical setting, going from the continuous (”real”)
physical world to the discrete one. Nonetheless, a flow domain is defined,
M. P
entek et al.
Fig. 2. Different levels of complexity and simulation of wind effects on structures; schematic representation using a vertical 2D cut model, adapted from Hanzlicek
(2015) and Wüchner et al. (2016).
a structural model is provided. BCs (inlet, outlet, on walls) need to be set
appropriately and the structural response is to be quantified. There is a
need to distinguish between the main approaches to how fluid loads are
measured and applied on structural models. A pure CFD simulation or a
typical wind tunnel measurement would only provide forces at the base,
for e.g. base shear Fx (here noting streamwise) and bending moment My
(due to Fx ) exerted on a rigid model. If an elastic numerical structure
model is at hand, the recorded forces could be applied in a so-called one-
way coupling scenario to be able to assess characteristic displacements
d in time t. This already has the advantage of including the inertial effects
of the body oscillator. The most complex approach would be the two-way
coupling. This implies the usage of (aero)elastic structural models
(whether physical or numerical, the latter being simulated using
computational structural dynamics CSD), which can strongly interact
with the surrounding flow.
Fig. 2 presents the interaction between structure and fluid flow
for the described FSI approaches, with their different level of
complexity. The upper part of the figure qualitatively shows how the
various physical fields are being simulated, more precisely how, for
example, the flow field looks like and whether the structure deformations
are computed and considered or not. The lower part already hints the
differences in methodology and numerics with the help of discrete time
steps tnðiÞ . Going from left to right one should note that: for a CFD there is
only a fluid solve involved in between these time steps; one way coupling
additionally includes a structural solve based upon wind-induced loads,
but no information about the deformations of the structure is received by
the fluid domain; the two-way coupling comprises of a fluid solve to
receive realistic loads on the structure as well as an update of the flow
domain as a result of structural deformations. The latter shall be
considered the highest level of complexity.
3. Choice and setup of the numerical models
This section presents the preliminaries for choosing and setting up the
various models for the proposed methodology. Selected results for each
component are shown. Throughout the workflow, the Kratos Multi-
physics open-source framework (for a detailed description refer to Dad-
vand et al. (2010)) is used to accommodate our concepts. Own extensions
enable some of the additional functionalities being addressed.
119
Journal of Wind Engineering & Industrial Aerodynamics 177 (2018) 117–135
3.1. Fluid domain
The chosen domain for the wind flow is based upon the research by
Daniels et al. (2013), who perform CFD analysis on a generic building
(according to Commonwealth Advisory Aeronautical Council CAARC)
proposed by IAWE (2009). We use the same Building B model as a
structure and setup the CFD model accordingly. The chosen numerical
technique for the flow is a large eddy simulation (LES) type approach
using a finite element method (FEM) based formulation called variational
multiscale (VMS). Consequently, large scale eddies are resolved by the
chosen spatial (mesh size) and temporal (time step) discretization,
whereas the energy contribution of vortices smaller than the grid is
modeled. An extensive theoretical and technical discussion of the chosen
formulation is provided by Cotela et al. (2016). The fluid domain is
modeled with FractionalStep elements (for technical details refer to the
Kratos Multiphysics project (Kratos Development Team, 2017)) with the
solver using a second-order accurate BDF2 time discretization scheme.
The setup is carried out in such a way, that values can be compared to
results by various other research groups submitted to the International
Association for Wind Engineering in accordance with the description in
IAWE (2009). The chosen geometry of the building is frequently used in
case of wind tunnel simulations, and, more recently, it has also become a
common choice for calibrating numerical models in CFD. The shape
consists of a simple rectangular section described by the planar di-
mensions length in streamwise direction D ¼ 45 m and width B ¼ 30 m,
whereas the height is H ¼ 180 m. Note, that this corresponds to the θ ¼
90∘ case, which we extensively use and refer to in the current study. The
size of the fluid domain and corresponding BCs are presented in Fig. 3.
These are kept as consistent as possible with the parameters in Daniels et
al. (2013).
The total length of 12H ¼ 2160 m in streamwise direction consists of
an inflow region of 4H (to allow a sufficient length for the development
of the wind flow) and an outflow of 8H (which in turn is needed to have
sufficient space and time for the dissipation of the vortices). 4H is chosen
on both sides. The guidelines provided by Franke et al. (2007) recom-
mend a blockage ratio of maximum 3 %, here currently kept at  0:5 %.
The meshing of the domain is done using four bounding boxes, gradually
increasing the refinement. There is a closest region created around the
building, which is a rectangular box 5H long, H wide and 1:5H tall. The
M. P
entek et al.
Fig. 3. Setup for the fluid domain. Note here the definition of the flow-attached
axes x'  y'  z'.
mesh size has a variation from 0:0375D (on the surface of the building)
up to 0:125D (on the edges of this refinement box). This area with the
higher refinement extends from 4H upstream (so from the inlet) to H
distance downstream of the building. In the rest of the bounding boxes
the mesh size increases gradually to a maximum of 0:77D. The total
number of elements add up to circa  3e7.
Empty channel simulations are carried out to check the main flow
parameters. These mostly aim to analyze the development of the wind
flow velocity profile in time and along the domain in streamwise direc-
tion. The target power law wind profile (with Vx';ref ¼ 40 m=s streamwise
velocity component at the top of the building) prescribed in IAWE (2009)
is used as an inlet BC

γ
z enough to capture at least 5  15 periods of vortex shedding (as seen in
Vx' ¼ Vx';ref ; (1)
zref
where zref ¼ 180 m is at the top of the building and γ ¼ 0:25 an empir-
ically derived coefficient that varies depending on the stability of the
atmosphere.
Fig. 4 shows a good match between the target values and the ones
sampled at the site of the building. These are in close match with results
Fig. 4. Comparison of the time-averaged stream-wise velocity profile over the height of the domain: the target profile and out own result (on the left) compared with
the data from other groups (on the right).
120
Journal of Wind Engineering & Industrial Aerodynamics 177 (2018) 117–135
reported by other groups (Holmes and Tse, 2012).
We use an inlet without fluctuations, which slightly differs from the
setup in IAWE (2009) and Daniels et al. (2013). Nonetheless, here we
check time-averaged values, which are still comparable despite the slight
difference in inlet BC. This modification of the inlet is consciously made
as the fluid domain is prepared for an FSI, where we aim to have a worst
case scenario for vortex-shedding. This ought to be more severe in a wind
flow with lower upstream turbulence intensity.
The characteristics of the air is kept at the standard condition of an
assumed temperature of Tair ¼ 20∘ C with a density of ρair ¼ 1:2 kg=m3
and a kinematic viscosity of νair ¼ 15:11e  6 m2 =s. These parameters
are considered constant over the height of the building. This assumption
is deemed adequate for common problems in wind engineering where a
neutral (Tamura and Kareem, 2013) wind velocity profile (as defined in
Equation (1)) is used because thermal/buoyancy effects are usually a lot
weaker than the mechanical effects driving the flow. It should be more so
suitable for a prototypical discussion carried out in the current paper.
The Reynolds number is computed based on the free stream velocity
at reference height
Vx'ref B
Re ¼ ¼ 1:0e8: (2)
ν
Results of the current work are compared to the ones presented by
Holmes and Tse (2012). This mostly aims at global aerodynamic force
resultants at the base of the building for various angles of attack θ. The
fluid domain and respective characteristics are defined in the
fluid-attached coordinate system noted x'  y'  z' (which is kept fixed),
whereas this comparison presents values of moments with respect to the
body-attached axis x  y  z (which is being rotated together with the
structure). In the current paper values for the angle θ are shown ranging
from 0∘ to 90∘ in increments of 10∘ . Our results consistently fit with the
works of others (Holmes and Tse, 2012), showing that the setup, based
upon similar works (Daniels et al., 2013) and best practice guides (Franke
et al., 2007), is appropriate in correctly capturing the aerodynamic force
resultants at the base of the building. Whilst Holmes and Tse (2012)
present and discuss the whole bandwidth of results from 7 groups, Fig. 5
only marks the minimum, maximum and average of those values,
including the results of our work.
For the CFD a total simulated time of t ¼ 120 s is chosen, which is
Fig. 6 for θ ¼ 90∘ ), that ought to be relevant and sufficient for the current
discussion. An initial ramp-up phase of 20 s is used to gradually increase
the inlet velocity from 0 to the prescribed value over the height of the
domain. This permits avoiding any shock effects early in the simulation,
which could have a quite negative influence during the FSI simulations.
The Courant Friedrichs-Lewy (CFL) number is set to CFL ¼ 1. Conse-
quently, the time-step size is calculated by
M. P
entek et al.
Fig. 5. Comparison of the moment resultants at the base of the structure as a function of the angle of attack θ.
Fig. 6. Aerodynamic forces acting in the direction of the weak bending and shedding frequency for θ ¼ 90∘ : in time (upper row) and frequency domain (lower row).
CFL Δxmin
Δt ¼ (3)
Vx';ref
so that is satisfies this conditions. The grid resolution is kept the same for
both cases of Vx';ref ¼ 40 m=s and Vx';ref ¼ 85 m=s. This leads to a time
step value of Δt40 ¼ 0:04 s and Δt85 ¼ 0:025 s, respectively.
For the main part of this study, the 90∘ angle case is used, as this
represents the most unfavorable wind flow condition for the building.
This setup serves as the basis for discussion during the upcoming sec-
tions.
The oscillatory pattern of the forming vortices can be observed well
when plotting the resultant forces on the sides of the building over time,
as in Fig. 6. A fast Fourier transform (FFT) can be applied to determine
the shedding frequency fsh around the building, resulting fsh;40 ¼ 0:83 Hz
and fsh;85 ¼ 0:2 Hz. The chosen time step sizes also satisfies the recom-
mended minimum (Franke et al., 2007) of ð10  20ÞΔt per the shedding
period Tsh .
As the inlet velocity varies over the height with a logarithmic profile,
the flow around the bottom of the building is influenced by the roughness
of the terrain. Quantification of dimensionless shedding by means of the
Strouhal number is here computed as
121
Journal of Wind Engineering & Industrial Aerodynamics 177 (2018) 117–135
fsh;crit B
St ¼ ¼ 0:806: (4)
Vx';ref
The European Committee for Standardization (2005) determines the
value for the St number as a planar characteristic of the geometrical
shape for a mostly 2D flow scenario around a rectangular cross sections
with sharp edges. For the current setup this would mean St ¼ 0:9.
However, specifications and adjustments considering a varying inlet over
the height and additional 3D effects are not provided.
We aim to setup the flow conditions, such that vortex-induced vi-
brations (VIV) lead to the resonance of the Building B along the weak
axis. The eigenfrequency for the structure in this mode is fn;S ¼ 0:2 Hz.
Rearranging the equation for St and solving for the velocity provides the
critical value
fsh;target B
Vx';ref ;crit ¼ ¼ 85 m=s; (5)
St
where fsh;target ¼ fn;S . The CFD domain has been setup according to works
and recommendation from others as well as our own experience with
similar simulation. For further discussions it is crucial to use a setup
which correctly captures the vortex shedding. As a further check, two
M. P
entek et al. Journal of Wind Engineering & Industrial Aerodynamics 177 (2018) 117–135

Fig. 7. Convergence of the shedding frequency fsh as a function of mesh refinement. Case of θ ¼ 90∘ and Vx';ref ¼ 40 m=s.

variations of the mesh have been studied, namely a medium (with the
element size on the surface of the building  2:5 m) and a very fine one
(with the respective smallest element  1:0 m). Fig. 7 shows the FFT of
the aerodynamic force Fy , which is in direct relation to fsh , as a function of
the mesh. It is concluded that the setup labeled as fine is deemed accurate
for our further purposes.
We also include a more graphical depiction of the shedding phe-
nomenon, because VIV represents the flow scenario we intend to deal
with. A zero turbulence intensity in the inflow region leads to almost
purely harmonic vortex shedding, as it can be seen in Fig. 8. This figure
clearly shows the strong pattern of the vortex shedding with the help of
the Q-criterion, which is a velocity gradient tensor-based method aiding
the visual detection and quantification of vortical structures in a flow
field (Jeong and Hussain, 1995).
This setup of the flow domain serves as the test case for the upcoming
sections.

Fig. 8. Visualization of vortex shedding using the Q-criterion (iso surfaces for Q ¼ 0:01 s2 ) colored by the magnitude of the velocity: axonometric (left) and top view
(right) for given time instances. Development depicted using respective phases during a harmonic vortex shedding period Tsh ¼ 5 s over the time t ¼ 45:25  50:25 s
for Vx';ref ¼ 85 m=s.

122
M. P
entek et al.
Fig. 9. Overview of the investigated structural models for the highrise building.
3.2. Structural model
Generally, the correct modeling of any structure, let alone highrise
buildings, involves detailed information about material properties and
the geometry. In WE, one of the main relevant factors is the outer hull
(i.e. the geometry), as this heavily influences the pressure distribution
and arising aerodynamic forces on the surface. Moreover, for FSI simu-
lations some dynamic characteristics have to be taken into account. The
current work mostly resumes to accurately capturing the first eigenmode
around the weak bending axis. Consequently, structural parameters of
the Building B, outlined in the comparative studies in IAWE (2009), are
used to setup the model. The source provides information about the ge-
ometry and main dynamic properties (such as the first three eigenmodes
– two sways and one twist) as well as (equivalent or average) material
characteristics. These serve quite well the setup of simplified (or equiv-
alent) structural models.
The dimensions of the Building B are already introduced for the CFD
simulations. For the structural dynamics part additional information
about the mass and stiffness properties is needed. The average density is
taken as ρ ¼ 160 kg=m3 and its first eigenfrequency for bending around
the weak axis (referred to as sway in the source) to be fn;S ¼ 0:2 Hz
(IAWE, 2009). The building can be modeled straight forward as a
cantilever beam with constant Young's modulus E and mass distribution
μ. By assuming a constant distribution of the structural mass within the
building volume the mass distribution can be computed as
μ ¼ ρ B D ¼ 2:16e5 kg=m: (6)
An equivalent bending stiffness EI results from the equation for the
natural frequency fn of the continuous system (here for a cantilever
beam), given by Petersen (2000) as
sffiffiffiffiffiffiffi
1:8752 EI
fn ¼ : (7)
2π H 2 μ
The moment of inertia I can be computed from the dimensions of the
cross section leading to the elasticity modulus E ¼ 2:861e8 N=m2 .
Several discretizations of the cantilever are used to show the ver-
satile applicability of the structure-AMD coupling scheme on a broad
range of structural models. These include a 3D model with solid ele-
ments and a 2D one using shells, both modeled with Kratos Multi-
physics (Kratos Development Team, 2017). Furthermore, a 2D beam
and a single-degree-of-freedom (SDoF) model are implemented in the
Python programming language. This gives the possibility to easily and
directly access and append their system matrices, a functionality
necessary for the upcoming session. Fig. 9 gives an overview of the
employed models.
The structural models used here focus on correctly capturing the
eigenmode and -frequency for the first mode shape – at 0:2 Hz – so
around the weak bending axis. This is of utmost relevance for the
shedding-induced across-wind response which is simulated.
123
Journal of Wind Engineering & Industrial Aerodynamics 177 (2018) 117–135
3.3. Added mass dampers
The task of reducing structural vibrations due to dynamic loads is one
of the major concerns in the field of WE. Consequently, different ap-
proaches have been developed to lessen the structural oscillations
without having to introduce significant changes to the main structural
system. A widely applied strategy is to employ inertia effects of devices,
which are added to the structure for this purpose. These mitigation de-
vices are mostly known as added mass or inertia dampers (Tamura and
Kareem (2013)). The effect of such devices is often interpreted as
changing the frequency response of the building, but it can also be un-
derstood as a kind of an open-loop controller, which takes the kinematics
(displacement, velocity and acceleration) of the structure at its connec-
tion point(s) and returns a reaction force, which reduces the structural
response. See a similar discussion involving control algorithms by Win-
terstein et al. (2016).
In the present contribution two different AMDs, a passive and a semi-
active one, are treated. These help to show how the proposed partitioned
approach makes it possible to deal with a range of practically relevant
devices in a straightforward manner. The mathematical models and
respective solution strategies for these systems are discussed in the
following.
3.3.1. Tuned mass damper
The TMD, as originally proposed by Den Hartog (1934), is used in this
study as benchmark example for passive AMDs. The basic idea implies
the addition of a SDoF oscillator to the structure. Its eigenfrequency and
damping characteristics are tuned in such a manner that it effectively
reduces the structural response.
For the derivation of the optimal tuning, the main structure is
abstracted as a SDoF system, using the generalized mass mS and stiffness
kS of the targeted vibration mode, as sketched in Fig. 10. The eigenform,
used to obtain the generalized quantities, has to be properly normalized
for this purpose depending on the position of the damper (Petersen,
2000). For the given structural system, with constant mass distribution
and a damper positioned at a height of 165 m, the equivalent mass is
evaluated and is equal to 1:244e7 kg, which represents 32 % of the mass
of the whole structure. A damper mass mAMD of 5:832e5 kg is chosen for
the TMD, which corresponds to 1:5 % of the complete structure mass, this
being the recommended value for cantilever type structures (Petersen,
2000).
From the mass ratio
mAMD
χ¼ (8)
mS
the optimal spring stiffness kAMD and damping ratio ζAMD of the added
oscillator are determined (Den Hartog, 1934):
χ
kAMD ¼ kS ; (9)
ðχ þ 1Þ2
M. P
entek et al. Journal of Wind Engineering & Industrial Aerodynamics 177 (2018) 117–135

Fig. 10. Reduction process of the structure model for the tuning of the TMD (principal visualization of the tuning procedure adapted from Petersen (2000). Note, that
ds and dAMD are the absolute displacements of the structure at the connection point pc and the displacement of the AMD mass, respectively.

sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi replaced passive damping element, meaning that cmin equals the optimal
3χ damping ratio of the TMD as given by Equation (11). The on-state is
ζ AMD ¼ : (10)
8ð1 þ χ Þ3 modeled by assigning cmax ¼ 50 cmin (Moutinho, 2015). This is a very
rough approximation for the behavior of a MR-damper, but it has been
Based on the damping ratio ζAMD the optimal damping coefficient is shown that an implementation based on such a model can effectively be
computed as applied to mitigate structural vibrations (Moutinho et al., 2015).
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi The control algorithm proposed is based on the idea of locking the
cAMD ¼ 2ζAMD kAMD mAMD : (11)
damper mass to the structure at certain time instances to compel the
The EoM for the two-degree-of-freedom (2DoF) system depicted in damping force and structure displacement to be in phase. This is achieved
Fig. 10 is by following the control law in Equation (13):
 
  
 
mS 0 cS þ cAMD
aS cAMD vS dS ðvAMD  vS Þ  0⇒cAMD ¼ cmin
þ : (13)
0 mAMD aAMD cAMD cAMD vAMD dS ðvAMD  vS Þ > 0⇒cAMD ¼ cmax
 

kS þ kAMD kAMD dS
þ
kAMD kAMD dAMD 3.3.3. Comparison of the behavior for various AMDs in case of prescribed

 movement
fS
¼ : (12) An AMD can be seen as a device which takes the kinematic state of the
0
target structure at its connection point pc as an input and returns a force
Based on Equation (12), we already hint how the coupled system to it. The corresponding differential equation for the TMD subjected to a
(consisting of structure-AMD) can be partitioned and how the derived movement of its pc , depicted in Fig. 11, is given as
approach can be directly extended to various structures with the flexible
and modular inclusion of arbitrary models of AMDs. mAMD aAMD þ cAMD ðvAMD  vS Þ þ kAMD ðdAMD  dS Þ ¼ 0; (14)

3.3.2. Semi-active tuned mass damper which can be rewritten

A major drawback of the TMD described above is its sensitivity to
(and dependency on) an exact frequency, so-called narrowband tuning. As
shown by Petersen (2000), relatively small deviations in frequency from
the optimal tuning may lead to a significant decrease in performance. To
overcome this problem, SATMDs have been developed, which are able to
adjust their properties to enhance broadband effectiveness according to
the actual vibrational behavior of the target structure. Thus, additional
advantage is achieved, which leads to better performance at frequencies
close to but shifted from the targeted eigenfrequency of the structural
model. This renders it superior to a comparable passive TMD. The term
semi-active is used to describe devices with mechanical properties which
can be controlled, but cannot increase the mechanical energy of the
controlled system, meaning that these are only able to produce dissipa-
tive forces, see Dyke et al. (1998). A widely used type of semi-active
devices is the magneto-rheological (MR) damper. This is a device filled
with MR fluid, which changes its rheological behavior (more specifically
its apparent viscosity) when subjected to an altering magnetic field.
In our contribution a simpler variation of a semi-active control
strategy for AMDs with MR-dampers (based on a phase control proposed
by Moutinho (2015)) is used to show how such a control strategy can be
flexibly accommodated by the proposed coupling scheme.
The idea is to replace the passive damping element of the TMD by a
MR(-like)-damper, which is controlled by an open-loop control strategy
of on/off type (Moutinho, 2015). This means, that either no magnetic
field is applied to the MR-damper or only the maximal possible one. The
corresponding behavior of the MR-device is modeled by assigning a
damping coefficient cAMD ¼ cmax in Equation (12) for the on- and a
damping coefficient cAMD ¼ cmin for the off-state. The MR-damper is
assumed to be designed such, that it behaves in the off-state as the
mAMD aAMD þ cAMD vAMD þ kAMD dAMD ¼ cAMD vS þ kAMD dS : (15)
Here vS and dS denote the velocity and displacement of the cantilever
type structure at the connection point pc of the TMD.
The TMD and SATMD are subjected to harmonically oscillating
movements of their respective pc . Different excitation frequencies are
investigated in order to highlight the differences of their behavior.
Fig. 12 shows the resulting displacements of the mass point and the
corresponding reaction forces for the two damper types. The respective
excitation frequency is given as multiple of the eigenfrequency fn;AMD of
the spring-mass system. It can be observed how the SATMD element
works by changing from off-to on-state (modeled by assigning cAMD ¼
cmax in Equation (15)). For a detailed description of the mechanisms
leading to the depicted reaction forces the reader is referred to (Mou-
tinho, 2015).
4. Setup and simulation of the (multiply-)partitioned wind-
structure interaction
In section 2 a conceptual discussion of the modular numerical wind
Fig. 11. TMD subjected to movement of the connection point.

124
M. P
entek et al.
Fig. 12. Mass displacement and corresponding reaction forces for TMD and SATMD subjected to movement of connection point dS ¼ 0:63 msinð2π fext t) with different
frequencies fext . These are given in relation to the eigenfrequency of the mass-spring system fn;AMD . The chosen ratios – 0.9 and 1.1 – seem to have the largest influence
and subsequently result the biggest difference in performance, whereas at 1.0 (resonance) the outcome is identical.
tunnel was presented, followed by the setup and methodological
description of each involved component analyzed individually, as seen in
section 3. The current section connects the concepts and methods pre-
viously outlined, this step being necessary for the proper co-functioning
of the models. As a result, this part deals more in-depth with the
(multiply-)partitioned approach used for the simulation of various in-
teractions. The available modules, which need to be connected, are wind
or generically fluid , structure and , the latter possibly
being either modeled as a further structure (S) (for the current work) or
even more generically as a controller (C) (Winterstein et al., 2016). A
generalization of the methodology ends the discussion substantiating the
flexibility, modularity and scalability of the approach.
4.1. Wind-structure interaction
The numerical simulation of WSI (or generically FSI) can be done
via a monolithic or partitioned approach. For fundamental de-
velopments and analytic studies there is a strong need for the former
one. Nonetheless, the usability and generic formulation for a variety of
situations is strongly limited. For these cases, a partitioned approach
(based upon the description in Felippa et al. (1999) referring to the
spatial separation of the main physical subsystems) is more viable.
This implies a separate and more adequate implementation of the fluid
and structural solvers, respectively. It furthermore requires a fitting
communication pattern between the models to ensure the fulfillment
of the coupling conditions. The main differences at a conceptual level
are shown in Fig. 13. The classical SDoF system is used as an analogy,
the stiffness k of the system is mostly related to the solid (or structural)
model whereas the damping c is contributed by the fluid domain
(structural models assumed to have very low to negligible damping
compared to fluids). Both contribute to the total mass m in motion, the
respective participation being noted by the factor α. At the interface,
125
Journal of Wind Engineering & Industrial Aerodynamics 177 (2018) 117–135
the moving mass has the characteristic displacement dðtÞ. The color
coding ( for solid or structure and for fluid) is kept to further
visualize the respective system matrices, which need to be assembled
and solved. For the monolithic approach the strong coupling is
implicitly enforced by the formulation. A partitioned approach does
need more components to be able to work, but it is preferred for in-
dustrial applications due to its scalability and versatility. In this sense,
it contributes to the efficiency of the workflow. In the following, it is
shown how this partitioned approach can be further extended for the
inclusion of other components (in this case AMDs) to mitigate the
structural response under wind.
Fig. 13 clearly depicts that the partitioned approach needs an addi-
tional step, namely the coupling of the physical fields. For our case, fluid
forces are transferred to the surface of the structure in a conservative
manner and serve as Neumann BC, whereas the nodal displacements (or
generally kinematics) of the structure are transferred back to the fluid
domain in a consistent way representing a Dirichlet BC, thus rendering it a
Dirichlet-Neumann type coupling. The convergence of the coupling is
achieved via a fixed-point approach with a dynamic under-relaxation factor
computed using the Aitken method. An in-depth discussion related to this
aspect is available from Küttler and Wall (2008) and Degroote et al.
(2008), whereas Heil et al. (2008) provide a detailed comparison of the
two approaches for FSI.
As a consequence, a fully-coupled WSI simulation is used to show the
effect of the consideration of structural displacements on the wind flow
domain, more specifically on the development of the aerodynamic forces.
Fig. 14 presents how the motion of the structure only slightly effects the
streamwise force Fx component (for the θ ¼ 90∘ case corresponding to
the body-attached strong bending axis My ), whereas a rather large in-
fluence can be seen on the aerodynamic force Fy (affecting the bending
over the weak axis Mx ). Without a coupled analysis, these implications
would be hard to foresee and even more difficult to quantify.
M. P
entek et al. Journal of Wind Engineering & Industrial Aerodynamics 177 (2018) 117–135

Fig. 13. Comparison of monolithic and partitioned approaches for FSI, based upon Dettmer et al. (2015) and Winterstein et al. (2016): conceptual illustration of the
coupled systems (left) and the resulting systems of equations to be solved (right).

Fig. 14. Comparison of the base force resultants due to aerodynamic forces arising when running a pure CFD compared to the two-way FSI. Relate to the conceptual
differences in Fig. 2. Note the additional effects of the coupled phenomenon. The dashed line at t ¼ 20 s marks the end of the inlet velocity ramp-up. Fx relates to My
and Fy to Mx , as seen in the definition of axis in Fig. 5.

4.2. Structure-added mass damper interaction
Similarly to the partitioning method for WSI described in subsection
4.1, this subsection presents the coupling of an arbitrary AMD to a
structural model. The added device needs information about the kine-
matics of the connection point relating it to the movement of the struc-
tural model. As the structure moves, the displacement, velocity and
acceleration of a certain defined point of connection pc is transmitted to
the added device, according to the concept already shown in Fig. 11. As
typical in numerical structural dynamics, prescribed deformations are
generally transformed into equivalent forces. Consequently, we end up
with a Neumann BC for the AMD. Analogously to the effect of the fluid
solver in FSI, the AMD transmits forces back to the structure, this also
serving as a Neumann condition, so in case of the S-AMD we have a
Neumann-Neumann coupling (a discussion and assessment of the
different types of coupling is done extensively in Sicklinger (2014)).
Consequently, the overall solution procedure for the coupled problem,
including iterations, convergence measure and relaxation schemes (as
already presented in subsection 4.1), can be extended and applied to this
new context in a straightforward manner. Once again, one should note
the flexibility provided by partitioned approaches.
In the present contribution only a single AMD with a single connec-
tion point is coupled to the structure, but in general the discussed setup
ought to be modular enough to be able to treat structures with multiple

126
M. P
entek et al.
AMDs, where each of them may have several connection points. The
main motivation is to show why it is advantageous to treat a structure
with generic AMD(s) in a partitioned way. For the TMD there would be
clearly no need for this and the usual approach would be a monolithic
treatment, since the resulting equation system can be easily treated by
structural solvers and the respective elements should be available in most
codes. In case of CSD, this would imply the extension of the mass, stiff-
ness, damping matrices and adding some additional degrees of freedom.
However, in case of dampers based on more complex mathematical
models, the partitioned solution strategy should be adopted, since this
makes it possible to setup the structure part conveniently with available
FEM software, while enabling the engineer to opt for modeling AMDs
with more refined governing equations and, possibly, other software
modules. The proposed partitioned approach makes a wider range of
choices for special damper cases possible.
The TMD is taken as starting point for the investigation of the parti-
tioned scheme. Due to its simplicity, monolithic solutions for the coupled
problem structure-AMD are easily attainable. Additionally, since the
TMD is a linear-time-invariant system, analytic solutions in time for the
monolithic setup can also be obtained. The analytic solutions serve as
basis unaffected by temporal discretization, whilst the monolithic ver-
sions are only affected by time discretization. The comparison of the
results of both solution procedures and the partitioned setup make it
possible to quantify the error introduced by the partitioned scheme.
Note, that for the sake of simplicity, we refer to the monolithic solution
by numerical means just as monolithic, while we denote the analytic
solution (which is also inherently monolithic) as analytic.
The governing equations of the TMD present an adequate demon-
stration how to obtain an iterative solution procedure in a partitioned
way. Therefore Equation (12) is reframed into
 and shell models the line load is uniformly distributed to the nodes at the
mS aS þ cS vS þ kS dS ¼ fS ðtÞ þ fC2S
; (16)
mAMD aAMD þ cAMD vAMD þ kAMD dAMD ¼ 0 þ fC2AMD
with the coupling terms
 (19)
fC2S ¼ ðcAMD vAMD þ kAMD dAMD Þ  ðcAMD vS þ kAMD dS Þ
: (17)
fC2AMD ¼ cAMD vS þ kAMD dS
The second line of Equation (16) in combination with the coupling
forces fC2AMD in Equation (17) is the differential equation for an AMD
subjected to a movement of the connection point already introduced in
Equation (15).
Both differential equations in Equation (16) can be discretized in time
and solved using a numerical time integration scheme, as these represent
the equation of motion (EoM) for 2 SDoF systems (one being the structure
model S, the other the TMD). The problem is, that they are coupled by the
terms given in Equation (17), which are initially unknown. To obtain a
consistent solution for the coupled problem, Equation (16) is treated in
an iterative Gauss-Seidel type manner (Winterstein et al., 2016) as part of
a fixed-point scheme (Küttler and Wall, 2008). Fig. 15 displays the
principal steps of the iteration procedure.
The algorithm starts by passing a prediction for the state of the
structure to the TMD. The most simple choice here is to use the present
state of the structure. The equation of the TMD is then solved, the
resulting reaction forces are passed back to the structure and the struc-
ture is solved with these new coupling forces, giving a new approxima-
tion for the structural state. By repeating the steps described above in an
iterative manner, as depicted in Fig. 15, the states of structure and
damper converge to the desired (prescribed) accuracy.
A more detailed discussion of setup and results follows, which aims to
highlight the accuracy and robustness of the partitioned approach when
compared to monolithic and analytic solutions, where possible. Own
implementations of the SDoF and Euler-Bernoulli beam models make all
three approaches possible, whereas for the 2D FE shells and 3D FE solid
only partitioned examples are at hand. For the test cases in this section,
the load on the structure has been chosen to be of harmonic type with a
127
Journal of Wind Engineering & Industrial Aerodynamics 177 (2018) 117–135
Fig. 15. Iteration sequence and the evolution between discrete time steps of the
partitioned schemed for the coupled simulation structure and AMD. The pro-
cedure is to be followed along the steps [1]–[16].
constant amplitude of F0 ¼ 5:0e4 N=m over the height, which roughly
represents the order of magnitude of the base force given by the CFD
simulation, as in Fig. 14. The load function is given as
Fðz; tÞ ¼ F0 cosð2π fext tÞ: (18)
These assumptions are justified to be adequate, as the aim is to check
the accuracy of the coupled scheme, which is considerably simplified by
the chosen load model, especially for the analytic solutions. For the solid
boundary of the structure.
In case of the SDoF model the amplitude of the generalized load for
the first vibration mode is determined, according to Petersen (2000), as
Fgen ¼ ∫ 0 F0 ηðzÞ dz:
H
where η is the properly normalized eigenform as shown in Fig. 10.
Equation (19) already takes into account that the load Fðz; tÞ is constant
over the height. This leads to Fgen ¼ 3:98e5 N.
For practical cases in structural engineering, a critical damping ratio
of under 5 % is realistic, thus we mostly assume either the worst case of
0 % or 5 %. The latter value is included to briefly show the mostly
negligible influence of structural damping. The time step size for the
following simulations is 0:025 s. This value is chosen to equal the time
step size used in the CFD simulations, which is then later also applied for
the coupled fluid-structure-AMD simulation. For the first eigenmode of
the structure with the corresponding eigenfrequency of fn;S ¼ 0:2 Hz,
which dominates the structural behavior, this step size results in a fine
time discretization of 200 time steps per period. The same holds for the
AMD, as its natural frequency of fn;AMD ¼ 0:191 Hz is very close to the
structural one. For the numerical time integration – of TMD, SATMD and
the SDoF and beam model of the structure – the generalized-α method
(Chung and Hulbert, 1993) is chosen (an implicit scheme), using the
parameter settings for second order accuracy and unconditional stability
discussed in the reference. ρinf ¼ 1:0 is the parameter that influences the
high-frequency dissipation, meaning no numerical damping is applied.
The FE models are solved using an α modification of Newmark's method
as discussed in Wood et al. (1980), yet another implicit scheme of second
order accuracy and unconditional stability.
Since detailed information about the design and placement con-
straints for an AMD are not available, we choose a region of 30  30 
30 m (in case of the 3D FE building; for the 2D and 1D structural models
and equivalency is made) for our example to represent the geometric
location of the AMD during further discussions. The reaction forces from
damper to structure are distributed to the nodes in this region.
M. P
entek et al.
Fig. 16. Displacements at the TMD's connection point due to a harmonically oscillating load (resonant case) for the SDoF and beam model computed by monolithic,
analytic and partitioned approaches.
Fig. 17. Displacements at the TMD's connection point due to a harmonically oscillating load (resonant case) for different structure models with attached TMD
simulated in a partitioned manner.
We compare the displacement results of the monolithic, analytic and
partitioned setups of the SDoF and beam model with TMD. This is done in
order to get a measure of the error introduced by the partitioned scheme
in comparison to the error introduced by temporal discretization. The
excitation frequency fext for these simulations is chosen to be in reso-
nance with the first eigenfrequency of the structure. The results depicted
in Fig. 16 display a good congruence for all three computational setups. A
closer examination shows that the difference between partitioned and
monolithic setup is significantly smaller than the difference between the
monolithic and analytic setup. For the setup leading to the results in
Fig. 16 the error introduced by the partitioning of structure (compare
monolithic vs. partitioned) and damper is negligible compared to the
error due to time discretization (analytic vs. monolithic).
For the 2D FE shells and 3D FE solid models with TMD only a parti-
tioned solution is at hand. To get a rough measure of accuracy for these
setups, we compare them in Fig. 17 to the partitioned solutions of the
Beam and SDoF model from Fig. 16. The results for the different structure
models are very similar. As the solutions of the SDoF and Beam model are
checked by analytic and monolithic schemes, this confirms the accuracy
of the partitioned approach for the FE models.
128
Journal of Wind Engineering & Industrial Aerodynamics 177 (2018) 117–135
Since Fig. 16 only represents a single combination of possible input
parameters, an extended parameter study is carried out to show the
accuracy of the partitioned setup for a broader range of input values.
This is done for the SDoF model of the structure part and the 2DoF
system (as given by Equation (12) for the monolithic and analytic so-
lution), respectively. The varied parameters are the mass ratio χ (see
Equation (8)) and the frequency of the harmonically oscillating load fext
in relation to the natural frequency of the structure fn;S . The values
taken into consideration for the mass ratio range from 0:01  0:2. The
latter corresponds to a damper mass of 2:488e6 kg, which is far above
typical values and has been chosen to point out that the partitioned
treatment is not limited in this respect. The range for the excitation
frequency has been chosen to cover 0.5 to 1.5 times the natural fre-
quency of the structure. Both domains have been sampled with 25
linearly spaced values and, for the respective combinations, the struc-
tural displacements of the partitioned, monolithic and analytic solution
have been evaluated. As a scalar valued substitute for the errors
occurring between a signal and a reference signal, we use a relative
error based on the L2-norm of absolute error and reference signal (of the
displacement)
M. P
entek et al. Journal of Wind Engineering & Industrial Aerodynamics 177 (2018) 117–135

Fig. 18. Relative displacement error as defined by Equation (20) for monolithic solution compared against analytic and partitioned solution compared against
monolithic for different combinations of mass ratio χ ¼ mAMD =mS and frequency ratio fext =fn;S ; The respective simulation times have been chosen to be 50 s.

 
d  dr j
rel: error ¼   2 ;
 
dr j2
where d denotes the signal and dr the reference.
Fig. 18 depicts the results of the parameter study for the displacement
of the structure at the connection point of the TMD. A comparison of both
graphs shows, that the error between partitioned and monolithic solution
by time integration is for all cases much smaller than the error due to
time discretization, given by the comparison of monolithic and analytic
solution. Going from analytic to monolithic (so considering the influence
(20) with a maximum deflection at the top of the structure of dtip ¼ H=250 ¼
0:72 m.
For the simulations in Fig. 19 it has been assumed that the structure
itself is undamped, which would be a worst case scenario and the
simulation time for each case has been chosen to be 50 s. For the forced
oscillation cases the root-mean-square (RMS) of the displacement signal
is used as measure for the magnitude of the occurring displacements. It is
computed for the discrete signal dS ¼ ðdS;1 ; dS;2 ; …; dS;n Þ as
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1X n

of temporal discretization) results in a maximum relative error of 0:8 % RMSðdS Þ ¼ d2 : (22)

over the considered range, whereas subsequent partitioning leads to only
a small fraction of that, namely 0:002 %.
To gain further insights on the performance of the different AMDs, the
3D FE model with and without AMD has been studied for the case of a
free oscillation with an initial displacement as well as for the case of
forced oscillations due to a prescribed harmonic load. For the free
oscillation the initial deflection is chosen as a third order polynomial
dtip z2 ð3H  zÞ
dðzÞ ¼
2H 3
n i¼1 S;i
The results for the free and forced oscillation cases are depicted in
Fig. 19. It can be observed that the application of TMD or SATMD
strongly reduces the free oscillation within four periods. Concerning the
forced vibration case, the effect of the AMDs is most significant in the
vicinity of the natural frequency of the structure. The chosen simulation
time of 50 s is sufficiently long for an almost complete decay of the free
oscillation for a structure with AMD. The results depicted in Fig. 17 show,
that this time span is also long enough for the resonantly excited struc-
(21)
ture to reach an almost steady state of vibration.

Fig. 19. Displacement of the 3D structure model at the connection point of the AMD due to an initial tip deflection of H=250 ¼ 0:72 m (corresponding to deflection at
the connection point of 0:63 m and RMS of the structural displacement due to a harmonically oscillating load for frequency ratios fext =fn;S of 0.5, 0.75, 0.9, 1.0, 1.1,
1.25 and 1.5. The respective simulation times for the evaluation of the RMS values have been chosen to be 50s.

129
M. P
entek et al.
Fig. 20. Displacement of the SDoF structure model due to an initial deflection of 0:63 m for different structural damping ratios ζS and RMS of the structural
displacement due to a harmonically oscillating load for frequency ratios fext =fn;S of 0.5, 0.75, 1.0, 1.25 and 1.5. The respective simulation times have been chosen to
be 50 s.
The influence of a low, but existing structural damping is investigated
in Fig. 20. Therefore, the setups discussed for Fig. 19 are simulated for
different damping ratios ζS . For simplicity, the SDoF oscillator is chosen
as the structural model for these studies. As one would expect, the
structural damping has a significant influence on the decay for free os-
cillations and on the amplitude in the case of resonant excitation by a
harmonic load for a structure without AMD, but an almost negligible
influence on the results for the structure with AMD. Consequently, the
assumption of an undamped structure can be seen of minor influence for
the results obtained for a structure with AMD.
Moreover, one should note the difference in performance of TMD
compared to SATMD, as seen in Fig. 20. In case of free vibrations, the
oscillations happen primarily with the eigenfrequency, thus both devices
behave similarly. However, in case of forced vibrations there is a notable
difference for excitation frequency ratios of 0.9 and 1.1.
4.3. Enhancement of the FSI to a multiply-partitioned simulation with
included arbitrary AMDs
The coupling of F-S and S-AMD has been discussed in subsection 4.1
and subsection 4.2. These individually represent the bi-coupling of two
physical fields or modules. Putting these together leads to a generaliza-
tion. The proposed methodology permits a generic extension of the
partitioned concept. At a conceptual level, this double-partitioned
approach, presented in Fig. 13 for WSI, is thus adjusted to accommo-
date multiple partitions, as seen in Fig. 21. The current work limits itself
to the inclusion of a single AMD. We connect the three distinct physical
fields, such that [F-[S-AMD]] arises, implying an outer iteration [F-[inner
group]] and an inner one [S-AMD]. Depending on the type of interaction
and chosen methodology, three systems could interact in another manner
as well, as described in Winterstein et al. (2016), but this is out of the
current scope.
Fig. 21. Suggested communication patterns for multiply-partitioned numerical simulations, graphics and naming adapted from Bungartz et al. (2015).
130
Journal of Wind Engineering & Industrial Aerodynamics 177 (2018) 117–135
An AMD is introduced into the existing partitioned scheme. As there
already exists a coupling between fluid (F) and structure (S), this
extension leads to a multiply-coupled scenario. A similar manner is used
to model the interaction of S-AMD. The communication pattern depicted
in Fig. 21 uses an outer loop between F and the inner bi-coupling group,
whereas AMD-S needs an inner loop when implemented. The values that
are ”communicated” between solvers are: fluid forces fF ; forces from the
AMD fAMD (as we are here applying control by force); displacements from
the structure dS;pc (value from the point of connection pc ) which serves as
input data needed by the AMD and dS;F on the interface between the
solid-fluid domains for F. The simulation of the fluid solver is the
computationally most expensive, so this pattern strives towards a mini-
mal number of evaluations for this component. For CWE, the used fluid
solvers typically already have parallel implementations. In case the
subsystems have been programmed in such a manner as well, the
framework could further exploit this leading to an overall improvement
in computational efficiency. The extension of the approach towards the
inclusion of other control devices and algorithms is enabled by the
inherent modularity.
Whilst a monolithic solution to such problems could still be possible,
this may mostly be restricted to small systems which can serve as a
baseline case for benchmarking and validation. Upscaling to industrially
relevant examples leads to the partitioned solution approach, which of-
fers such versatility that it renders it as the favorable choice. Fig. 22 (this
extension from Fig. 13 not only shows this on a conceptual level, but also
accentuates the differences of complexity on the matrix level. Numerical
methods, in the end, resume to solving algebraic systems, so the assem-
bly, conditioning and stability of these is crucial. The monolithic system
matrix is heavily interwoven and could become badly conditioned due to
very diverse numerical properties of the involved systems. The parti-
tioned approach permits adequate assembly and solution routines for
each of the subsystems separately. For our case, this would mean that for
M. P
entek et al. Journal of Wind Engineering & Industrial Aerodynamics 177 (2018) 117–135

Fig. 22. Comparison of the monolithic and partitioned approaches for 3 coupled subsystems.

e.g. the fluid field is solved in a segregated fashion using two iterative results can be found in the subsequent work by P
entek et al. (2017).
solvers for the velocity and pressure fields, the structural model can even In Fig. 24 we have the classical case of a vibration absorber (i.e. the
rely on direct linear solvers, whereas the solution of the AMD here ends structural solver and AMD) with the given external excitation of the
up being a scalar problem. coupled system coming from a fluid solver fF ðtÞ. For the SDoF structural
Clearly, the partitioning will need an additional coupling of the in- model the following parameters were chosen: density ρair , eigenfre-
terfaces. For the current framework, a Gauss-Seidel type fixed-point quency fn;S , damping ratio(s) ζS used for the determination of the
coupling is used for both partitions [F-[S-AMD]] (see details for each equivalent viscous damping coefficient cS . These followed the order of
groups in subsection 4.1 and subsection 4.2, respectively). The main magnitude of the typical values proposed for benchmarking in IAWE
aspect of this numerical component can be seen in Fig. 23. Herein, the
algorithm for coupling the three subsystems is shown in a graphical way,
where the consequential steps [1]–[26] are to be followed. The arrows
with the numbers attached to them show the data transfer between fluid,
structure and AMD. The [S-AMD] group is nested and ik is the typical
number of inner iterations needed to ensure the equilibrium of this inner
group. im denotes steps necessary for the convergence of the outer group
[F-[inner group]]. For now, the framework implies having the same time
step size for all systems. Further extensions could include accommoda-
ting variable time steps, if deemed necessary. In Winterstein et al. (2016)
an assessment for the two other different possibilities for coupling three
subsystems with a Gauss-Seidel scheme is done. It has to be kept in mind,
that this kind of coupling works best for the case of the active and
semi-active control units, which are generally open-loop systems. On the
other hand, when dealing with closed loop systems, another type of
coupling pattern, depending on the controller type, might be a better
choice. This is due to the fact that a closed-loop controller might cause
instabilities by its feedback value, even if the open-loop system has been
stable.

4.4. Application and showcase of the new methodology

The conceptual description of the new methodology in subsection 4.3

could translate into the 2D simplified situation presented in Fig. 24,
which stands for a 2D horizontal cut of a building. This is the case of an
SDoF undergoing VIV. Another 2D scenario is presented in Fig. 25, which
is a more generic example of a MDoF or FE-type structure under Fig. 23. Iteration sequence and the evolution between discrete time steps of the
buffeting-like wind load. The current contribution uses these examples to partitioned schemed for the fully-coupled simulation with 3 coupled fields,
substantiate and prepare the 3D case, here providing the most relevant namely fluid, structure and AMD. The procedure is to be followed along the
parameters as well as the visuals. The detailing of the exact setup and steps [1] - [26].

131
M. P
entek et al.
Fig. 24. Conceptual setup of the VIV FSI scenario and the partitioned S-AMD in case of a SDoF (P
entek et al., 2017).
Fig. 25. Conceptual setup of the a buffeting FSI scenario and the partitioned S-AMD in case of a FE model (P
entek et al., 2017).
(2009). The TMD was accordingly setup for an optimal passive perfor-
mance for the mitigation of peak displacement. The mass ratio χ , the
ratios of frequencies fn;TMD =fn;S , the damping ratios of the TMD ζopt;TMD
were calculated according to Den Hartog (1934). Again, the damping
ratio ζopt;TMD leads to the equivalent viscous damping coefficient copt;TMD .
For the SATMD, the damping is chosen as cmin and cmax used for the
on/off-type control law. This setup procedure has been thoroughly dis-
cussed in subsection 3.3 and subsection 4.2.
The simulation setup models a simplified 2D flow around a square
cross section of a building using a constant inlet. Cross-wind forces occur
due to vortex shedding. The structure model can only perform rigid body
motion in the degree of freedom depicted and is considered as a point
mass. Consequently, the forces are reduced to the center of the oscillator.
The parameters inflow velocity Vx' and the rectangle size B of the
structure are chosen so that the shedding is close as possible to the nat-
ural frequency of the structure. The vortex shedding and motion of the
oscillator interact and this situation can lead to lock-in.
In case of applying the previous framework to more generic FE
structures, as previously discussed in subsection 3.2, structural solvers,
Equation (16) and Equation (17) become more generalized. This leads to
Equation (23), where we see that the structure is characterized by a
matrix-vector equation
 standing of the coupling for the more generic FE case. Here a cantilever
MS AS þ CS VS þ KS DS ¼ FF ðtÞ þ FC2S
; (23)
mAMD aAMD þ cAMD vAMD þ kAMD dAMD ¼ 0 þ fC2AMD
with the coupling terms

 An extension to 3D is straightforward based upon the previous con-
FC2S →fC2S;nodal ¼ ðcAMD vAMD þ kAMD dAMD Þ  cAMD vS;pc þ kAMD dS;pc nnodes
: siderations. For this scenario the, 3D FEM structural model without
fC2AMD ¼ cAMD vS;pc þ kAMD dS;pc
(24)
The structure still receives a concentrated force, so mathematically a
132
Journal of Wind Engineering & Industrial Aerodynamics 177 (2018) 117–135
scalar value fC2S which is divided by the number of nodes considered as
connected to the AMD nnodes . Ultimately, the vector notation FC2S repre-
sents the ”collection” of nodal forces fC2S;nodal . Here a uniform distribution
is assumed. Relate to Fig. 25, as it shows how the setup is done for a
simplified 2D case exemplifying a horizontal cut through a 3D flow
domain.
This setup has been included to show how all the components (F-S-
AMD) function in a more generic manner. A proposed inlet BC would be
constant along the height of the domain and would fluctuate harmoni-
cally around the given mean Vx' with the natural frequency (seen as
simplified buffeting) of the generic FE structure. The AMD “needs” a
certain point of connection pc (which for the SDoF is given by definition,
whereas for the FE model it needs to be explicitly chosen). The added
damper takes the base point excitation from pc and sends back a reaction
force to it. This is then uniformly ”spread” over the relevant points in the
structural model, here marked with the light red filled region at the top of
the building in Fig. 25. The loads from wind flow FF ðtÞ are applied to each
node of the FE structure which lie on the interface to the fluid domain.
Otherwise, communication is exactly as in case of the SDoF. In both cases,
the only communication is either between F-S or S-AMD. The AMD
neither receives information directly from the wind flow nor does it
directly influence the flow conditions. Fig. 26 is used to aid the under-
type structure modeled with quadrilateral thick shell element is set to
vibrate freely. The evolution over a period of oscillation is illustrated
with green arrows representing the nodal force component fC2S;nodal as
vectors resulting from the distribution of fAMD onto the region affected.
structural damping has been chosen to be on the unfavorable side. The
setup of the fluid domain and parameters of the CFD simulation are those
described in subsection 3.1. Note, that the inflow condition has
M. P
entek et al. Journal of Wind Engineering & Industrial Aerodynamics 177 (2018) 117–135

Fig. 26. The vector plot of forces fAMD ⇒FC2S in case of the first period Tn ¼ 5 s of free vibration around x-axis with initial displacement are depicted. Note, that
displacements are upscaled 25 times. Relate the concept to Fig. 25. Here the coupling forces act as point loads fC2S;nodal on various nodes of the 2D FEM model. These
nodes are part of the region, which are assumed to be directly acted upon by the AMD.

Fig. 27. Comparison of the aerodynamic forces and structural displacement in case of FSI. The aerodynamic force Fy from the pure CFD is marked with grey in the first
row (CFD not in the legend of the plot). The deformation with respect to the weak bending axis is shown in the second row, here the displacement dy ¼ dpc of the point
of connection pc is depicted. The 3D FEM structural model is undamped.

practically no turbulence rendering this case as a worst condition due to
which vortex shedding is expected to be an almost pure harmonic with
the frequency close to or exactly matching the natural vibration of the
Building B model, as in subsection 3.2. Here results of the numerical
simulations of WSI for a 3D scenario with the inclusion of the AMDs (as
detailed in subsection 3.3) are presented. The coupling scheme has been
thoroughly described in subsection 4.3. The fully-coupled interaction of
the three subsystems is resolved in time.
Fig. 27 shows the relevant FSI results with and without the inclusion
of the AMDs. First, one should note that the movement of the structure
does affect the arising base force Fy in the direction of the weak axis.
Whereas the force magnitude for the pure CFD is in the order of 0:75e7 N,
for the FSI case without an AMD it grows up to 2:5e7 N, again being
reduced to around 1:5e7 N, when vibration mitigation devices limit the

133
M. P
entek et al.
Fig. 28. Conceptual generalization of the structure-controller concept: enabling
of efficient exchange of arbitrary types of added devices due to the modular
multiply-partitioned simulation concept. The sketch depicting the structure
model is adapted from (Tamura and Kareem, 2013).
deformations. The subplot in the second row shows the evolution in time
of the displacement component dy in the direction of the weak bending
for the point of connection pc (this being at the height 165 m). The FSI
case without an AMD is a structure that has no structural damping, so the
displacement could theoretically grow unbounded. In case of WSI, there
is an added damping effect from the fluid, which here (for the considered
simulation time of 70 s) seems to be negligible. Thus, the deformations
grow up to almost 10 m, but are limited to approximately 2 m with the
usage of additional devices. The high deformations of the undamped and
uncontrolled structure are attributed to the choice of the setup, namely a
worst case scenario from the point of view of the fluid domain (low
turbulence, very high flow velocity Vx';ref ¼ 85 m=s for this highly un-
likely limit case) combined with the zero structural damping and con-
stant cross section over the height with periodic the vortex shedding
happening over the weak bending axis. Nonetheless, the AMDs serve
their purpose well and the simulation of the 3D case proves to be suc-
cessful in substantiating the usefulness of the proposed methodology
even for large scale examples. More so, the computational effort of
running an FSI without or with AMDs is comparable due to two reasons:
the additional effort of solving the added device is relatively low when
compared to the solving S, and very low compared to F (due to the dif-
ferences in the number of degrees of freedom of the respective models
and the governing equations); the AMDs reduce the motions of the
structure, so the FSI solve is made easier as the incremental deformations
of the fluid domain are smaller in between consecutive two time steps.
4.5. Generalization of the methodology for arbitrary domains and complex
structural models
The new methodology presented is prepared to accommodate various
industrially-relevant devices. For now, the current work thoroughly
discusses and analyzes the usage of a TMD, which serves as an important
choice for the testing of the setup. The SATMD variation helps to argu-
ment why a (multiply-) partitioned scheme is a must for a robust and
modular approach. Possible additional AMD-type controllers would also
fit well into this approach. Thus, more complex SATMD variations such
as a more complex MR model as described by Weber and Ma
shlanka
(2012) or a certain fractional order controller (FOC) as in Muresan et al.
(2016) could be included. Most of these devices have been tested in an
experimental setting adequate for earthquake engineering. This might
not be sufficient when aiming to investigate the effect of such controllers
on structures undergoing coupled WSI phenomenon. Whereas dynamic
excitation due to earthquakes can be mostly considered one-way (as the
construction does not considerably effect the excitation forces), this
might not be sufficient for various wind-induced loads.
Additional inclusions could imply the usage of tuned liquid dampers
(TLDs) as well as AMDs featuring inerter elements (as in Giaralis and
Petrini (2017)). The partitioned approach would enable in both cases the
treatment of devices, which are usually not available in most structural
simulation codes. The generalization of the concept is shown in Fig. 28.
This contribution currently concentrates on the framework and meth-
odology itself as well as the benefits of fully-coupled simulations for CWE
applications. However, one should bear in mind the potential advantages
134
Journal of Wind Engineering & Industrial Aerodynamics 177 (2018) 117–135
provided for the design of additional devices by such an approach: due to
modularity and scalability, the possibility arises to test out new control
devices in realistic scenarios with a small to negligible computational
overhead compared to already existing FSI simulations in WE.
5. Conclusions
A numerical methodology for the partitioned simulation of fully-
coupled WSI has been presented. This solution strategy enables the in-
clusion of further devices, such as AMDs, in order to further raise the
added value provided by such simulations. Thus, the procedure provides
additional information about the behavior of the structures co-
functioning with the added devices in wind. The present contribution
outlined the proper choice and setup of the involved modules in the
preparation of the proposed coupling approach. For each individual
component (fluid domain, structural model, AMD) the modeling was
done such that an adequately and sufficiently complex modeling is used.
These were first tested individually to quantify the behavior. Various
components were also connected in order the check the correct co-
functioning leading to a multiply-partitioned 3D simulation. Such a
systematic approach substantiates the trustworthiness of the results. The
novelty comes from the versatility and flexibility aspect. Various com-
ponents can be co-simulated using this modular manner. For simulations
in computational WSI with AMDs, this nested approach seems to be the
best fit. Various AMDs can be setup, tested and designed without creating
a large overhead, as the methodology is quite independent of the exact
functioning of these devices. Due to its modularity, the coupling scheme
should be robust enough to accommodate various added devices or even
open and closed-loop control algorithms. The showcase of the 3D simu-
lation of a generic highrise building should render this approach highly
viable for practical examples and industrially-relevant cases in WE. The
described modular linking of various elements of the numerical wind
tunnel could be further extrapolated. Multiple devices could be used
simultaneously. Optimization algorithms could assess the improved
placement of such controllers. These additional aspects should be further
investigated to exploit the predictive capabilities of numerical methods
and further assist design.
Acknowledgements
The authors gratefully acknowledge the support of the PhD project by
the DAAD (German Academic Exchange Service). Gratitude is expressed
for valuable discussions with Michael Andre and Alexander Michalski as
well as supportive works by the students Pratyush Jha and Dennis Kasper.
References
Bletzinger, K.-U., Wüchner, R., Andre, M., P
entek, M., Michalski, A., 2016. Der
numerische Windkanal im Bauwesen – Potentiale und Herausforderungen am
Beispiel flexibler Tragwerke. In: 20. Dresdner Baustatik Seminar: Realit€at –
Modellierung – Tragwerksplanung. Institut für Statik und Dynamik der Tragwerke,
Technische Universit€at Dresden.
Bungartz, H.-J., Lindner, F., Mehl, M., Uekermann, B., 2015. A plug-and-play coupling
approach for parallel multi-field simulations. Comput. Mech. 55 (6), 1119–1129.
Chung, J., Hulbert, G.M., 1993. A time integration algorithm for structural dynamics with
improved numerical dissipation: the generalized-α method. ASME J. Appl. Mech. 60
(3), 371–375.
Cotela, J., O~
nate, E., Rossi, R., 2016. Applications of Turbulence Modeling in Civil
Engineering. Monograph CIMNE M159, ISBN 978-84-945077-2-4.
Dadvand, P., Rossi, R., O~ nate, E., 2010. An object-oriented environment for developing
finite element codes for multi-disciplinary applications. Arch. Comput. Meth. Eng.
https://doi.org/10.1007/s11831-010-9045-2.
Daniels, S.J., Castro, I.P., Xie, Z.-T., June 2013. Peak loading and surface pressure
fluctuations of a tall model building. J. Wind Eng. Ind. Aerod. 120, 19–28. https://
doi.org/10.1016/j.jweia.2013.06.014.
Degroote, J., Bruggeman, P., Haelterman, R., Vierendeels, J., 2008. Stability of a coupling
technique for partitioned solvers in FSI applications. Comput. Struct. 86 (23–24),
224–2234.
Den Hartog, J.P., 1934. Mechanical Vibrations. McGraw Hill.
Dettmer, W., Kadapa, D., Peri
c, C., Liang, R., Loosten, M., 2015. On Partitioned Solution
Strategies for Computational Fluid-structure Interaction. Technical presentation at
the Technical University of Munich.
M. P
entek et al.
Dyke, S.J., Spencer Jr., B.F., Sain, S.J., Carlson, S.J., 1998. An experimental study of MR
dampers for seismic protection. Smart Mater. Struct. 7 (5), 693.
European Committee for Standardization, 2005. Eurocode 1: Actions on Structures - Part
1-4: General Actions - Wind Actions.
Felippa, C., Park, K., Farhat, C., 1999. Partitioned Analysis of Coupled Mechanical
Systems. Technical report. College of Engineering, University of Colorado.
Franke, J., Hellsten, A., Schlünzen, H., Carissimo, B., 2007. Best Practice Guideline for the
CFD Simulation of Flows. COST Office, Brussels.
Giaralis, A., Petrini, F., Sep 2017. Wind-induced vibration mitigation in tall buildings
using the Tuned Mass-Damper-Inerter. J. Struct. Eng. ASCE 143.
Hanzlicek, I., 2015. Staggered Algorithms for Fluid-structure Interaction. Master’s thesis.
Lehrstuhl für Statik, Technische Universit€at München.
Heil, M., Hazel, A.L., Boyle, J., 2008. Solvers for large-displacement fluid-structure
interaction problems: segregated versus monolithic approaches. Comput. Mech. 43,
91–101. https://doi.org/10.1007/s00466-008-0270-6.
Holmes, J.D., Tse, T.K.T., Aug 2012. International high-frequency base balance
benchmark study. In: Proceedings of the the 2012 World Congress on Advances in
Civil, Environmental, and Materials Research (ACEM’ 12). International Association
of Structural Engineering and Mechanics.
International Association for Wind Engineering (IAWE), 2009. Benchmark buildings for
an international HFBB comparison. http://www.iawe.org/committees/HFBB-spec.
pdf.
Jeong, J., Hussain, F., 1995. On the identification of a vortex. J. Fluid Mech. 285 (1),
69–94.
Kareem, A., 1983. Model for predicting the acrosswind response of buildings. Eng. Struct.
6.
Kareem, A., 2017. Computational fluid dynamics, simulation and computational tools. In:
Webinar by SimCenter Series, vol. 101. Natural Hazards Engineering online: accessed
12 June. https://www.designsafe-ci.org/learning-center/training/simcenter/
webinar-170612/.
Kratos Development Team, 2017. The Kratos Multiphysics Open-source Project. https://
github.com/KratosMultiphysics.
Küttler, U., Wall, W.A., 2008. Fixed-point fluid structure interaction solvers with dynamic
relaxation. Comput. Mech. 43 (1), 6172.
Liang, S., Liu, S., Li, Q.S., Zhang, L., Gu, M., 2002. Mathematical model of across-wind
dynamic loads on rectangular tall buildings. J. Wind Eng. Ind. Aerod. 90, 201–251.
135
Journal of Wind Engineering & Industrial Aerodynamics 177 (2018) 117–135
Moutinho, C., 2015. Testing a simple control law to reduce broadband frequency
harmonic vibrations using semi-active tuned mass dampers. Smart Mater. Struct. 24
(5).
Moutinho, C., Cunha, A., Martins de Carvalho, J., June 2015. Implementation of a semi-
active tuned mass damper to reduce vibrations in a slender footbridge. In:
Proceedings of the 7th ECCOMAS Thematic Conference on Smart Structures and
Materials.
Muresan, C.I., Dulf, E.H., Prodan, O., 2016. A fractional order controller for seismic
mitigation of structures equipped with viscoelastic mass dampers. J. Vib. Contr. 22
(8).
P
entek, M., Wüchner, R., Bletzinger, K.-U., Jul 2017. A novel fully-coupled computational
wind-structure interaction approach for the design of added mass dampers. In:
Proceedings of the 7th European and African Conference on Wind Engineering.
A.I.V., l'Association pour l'Ing
enierie du Vent, on behalf of the International
Association for Wind Engineering (Digital version).
Petersen, C., 2000. Dynamik der Baukonstruktionen. Vieweg Verlag, ISBN 3-528-08123-
6.
Sicklinger, S., 2014. Stabilized Co-simulation of Coupled Problems Including Fields and
Signals. PhD thesis. Lehrstuhl für Statik, Technische Universit€at München.
Tamura, Y., Kareem, A., 2013. Advanced Structural Wind Engineering. Springer.
Weber, F., Ma
shlanka, M., 2012. Frequency and damping adaptation of a TMD with
controlled MR damper. Smart Mater. Struct. 21 (5).
Winterstein, A., Lerch, C., Wüchner, R., Bletzinger, K.-U., Oct 2016. Computational
closed-loop control of fluid-structure interaction FSCI for lightweight structures.
PAMM Proc. Appl. Math. Mech. 16 (1), 15–18. https://doi.org/10.1002/
pamm.201610005. Joint Annual Meeting of DMV and GAMM 2016.
Wood, W.L., Bossak, M., Zienkiewicz, O.C., 1980. An alpha modification of Newmark's
method. Int. J. Numer. Meth. Eng. ISSN: 1097-0207 15 (10), 1562–1566. https://
doi.org/10.1002/nme.1620151011.
Wüchner, R., P
entek, M., Michalski, A., 2016. Lecture Notes in Structural Wind
Engineering/Windingenieurwesen. Lehrstuhl für Statik, Technische Universit€at
München.
Yang, J., Agrawal, A., Samali, B., Wu, J., 2004. Benchmark problem for response control
of wind-excited tall buildings. J. Eng. Mech. 130 (4), 437–446.