Renewable Energy 182 (2022) 1152e1170

Contents lists available at ScienceDirect

Renewable Energy
journal homepage: www.elsevier.com/locate/renene

Platform and mooring system optimization of a 10 MW

semisubmersible offshore wind turbine
Giulio Ferri a, *, Enzo Marino a, Niccolo

Bruschi a, Claudio Borri a
a
Department of Civil and Environmental Engineering, University of Florence, Via di S. Marta 3, 50139, Firenze, Italy

a r t i c l e i n f o a b s t r a c t

Article history: In this paper, an optimization procedure is proposed to find platform and mooring system configurations
Received 30 June 2021 which most effectively reduce the dynamic response of a semisubmersible 10 MW Floating Offshore
Received in revised form Wind Turbine (FOWT). This is done by developing an efficient frequency domain simulation model able
28 September 2021
to account for the viscous drag forces and the contributions to the equation of motion stemming from
Accepted 21 October 2021
turbine and mooring lines. The objective function is the value of the Response Amplitude Operator (RAO)
Available online 30 October 2021
at the eigenfrequency of the selected degree of freedom (DoF) of the system. Both parked and power
production states are investigated. Feasibility constraints related to mean displacements and moorings
Keywords:
Floating offshore wind turbines
layout are considered. Results show that optimized configurations can be found with better perfor-
Frequency domain model mances and smaller platform dimensions with respect to the configuration obtained by scaling up the
Semisubmersible platform 5 MW geometry.
10 MW Wind turbines © 2021 Elsevier Ltd. All rights reserved.
Large floating platform
Platform optimization

1. Introduction sufficiently accurate optimization procedures. Within this context,

a first parametric study was carried out by Sclavounus et al. [7].
Floating Offshore Wind Turbines (FOWTs) are the leading They performed one of the first parametric design study of a FOWT
innovation in the wind energy sector. Higher wind speed, reduced and its mooring system based on a coupled Frequency Domain (FD)
turbulence, abundance of exploitable sites, and almost absence of model [8]. It was adopted a six-degree of freedom (DoF) platform
visual impact are some of the most important aspects which make model in which mooring lines and turbine contribution were
FOWTs competitive with respect to onshore or fixed-bottom modelled as added inertial, damping and stiffness matrices ob-
offshore technologies. On the other hand, installing a wind tur- tained from a linear analysis performed with FAST [9] around the
bine (WT) on a floating platform gives rise to a number of addi- steady state operating point of the system. Linear hydrodynamic
tional complexities associated with multiple fluid-structure contributions were calculated by the potential flow solver WAMIT
interaction problems involving simultaneously very slender, highly [10] at different frequencies. The optimization was performed with
deformable bodies, such as rotor blades and mooring lines, and the intent of reducing significant quantities, such as nacelle accel-
bluff bodies, such as the floating platform. To properly reproduce eration and platform mass. Similar FD models were presented in
the behaviour of such structural systems, the effects of nonlinear Refs. [11,12]. Karimi et al. [13] performed a FD-based multi-objec-
wave kinematics [1e3], as well as fatigue loads, resonance phe- tive optimization of a 5 MW FOWT aimed at controlling construc-
nomena, and wind-wave misalignment [4e6] have been found to tion costs. Dou et al. [14] proposed a FD-based optimization in
be of primary importance for reliably predictions of FOWTs which the nacelle displacement in the wind/wave direction and the
response especially in harsh sea states. The large number of design platform motions were considered as DoFs. Optimization proced-
variables involved, as well as the need for pursuing contrasting ures were also focused on mooring lines. A FD model was used in
targets, such as reducing the overall cost and enhancing structural Ref. [15] to reduce the cable length of a semisubmersible platform.
performances and reliability, raises the need for robust and More recently, Bruschi et al. [16] and Barbanti et al. [17] investi-
gated the optimal amount of mass and position of clump weights in
order to minimize the platform response.
* Corresponding author. The dramatic increase of key dimensions for next generation
E-mail addresses: giulio.ferri@unifi.it (G. Ferri), enzo.marino@unifi.it (E. Marino), WTs, such as rotor diameter, tower height, and (possibly) water
niccolo.bruschi@unifi.it (N. Bruschi), claudio.borri@unifi.it (C. Borri).

https://doi.org/10.1016/j.renene.2021.10.060
0960-1481/© 2021 Elsevier Ltd. All rights reserved.
G. Ferri, E. Marino, N. Bruschi et al. Renewable Energy 182 (2022) 1152e1170

depth, rises the need for new design techniques. So far, design 2. Methodology
approaches based on power ratios rational upscaling [18,19] have
been proposed for the dimensioning of platforms suitable for larger A FOWT is a complex multibody system consisting in the turbine
wind turbines. Also new mooring lines configurations [20] and and the floating support, which is exposed to a multi-physics fluid-
platform shapes have been studied both numerically [21] and structure interaction involving wind and water (waves and cur-
experimentally [22]. Although several critical aspects, such as rents). To properly reproduce the dynamic response of a FOWT,
short-term and long-term response [19], wind-wave misalignment nonlinear coupled aero-hydro-servo-elastic numerical simulations
[20] and extreme loads [23] have already been studied, the design in the time-domain are usually necessary. The opensource code
of optimal supporting platform for 10 MWþ WTs is still an open FAST [9], developed at the National Renewable Energy Laboratory
problem. Indeed, overly simplistic upscaling design procedures (NREL), is widely adopted by researchers for this purpose. Non-
may lead to overdesigned platforms and uneconomic designs. In linearities, for example stemming from viscous drag and higher
Ref. [24], the authors performed a preliminary parametric study order hydrodynamics [25e27] are neglected or approximated. Also,
aimed to identify the optimal dimension of a semisubmersible transient effects cannot be considered and only stationary load
platform suitable for a 10 MW wind turbine. Response Amplitude conditions can be simulated. For a given wave spectrum, frequency
Operators (RAOs) were chosen as target to be minimized. It was domain models allow to evaluate the RAO, i.e. the linearized
found that the optimized platform behaves better than one ob- Transfer Function (TF) of the dynamic system, which is often cho-
tained with a direct geometric upscaling procedure. However, in sen as objective function in the optimization procedures. The FD
Ref. [24] the parametric study featured two limitations: (i) only two model developed for the present study adopts ANSYS AQWA [28] to
design variables were considered, i.e. the columns diameter and solve the potential flow problems of radiation and diffraction. FAST
platform radius; (ii) a single constraint on the mean value of the v7 [9] linearization is adopted to estimate wind turbine and
Pitch angle of the platform at the rated wind speed was enforced. mooring lines contribution to the equation of motion of the floating
In the present work, a Genetic Algorithm (GA)-based optimi- system. Morison forces [29] are linearized and calculated on
zation is performed with the aim of identifying the best- slender elements and Heave plates.
performing semisubmersible platform suitable for a 10 MW WT.
An in-house developed FD model, described in detail in Section 2, is
implemented in an evolutionary-based minimization procedure. 2.1. Hydrodynamic model
FD models lead to approximated solutions with respect of time
domain models and cannot account for nonlinear effects [5,23]. In this section, the hydrodynamic model of the floating plat-
Several studies have been carried out to deeply investigate the form, which is based on the combination of potential flow theory
complex fluid structure interaction between waves and the floating and Morison equation [29], is summarized.
substructure. Bachynski and Moan [5] investigated the effect of Assuming an inviscid and incompressible fluid, a velocity po-
ringing, a forcing load in the proximity of the resonance frequency, tential Fðx; y; z; tÞ exists. Under the hypothesis of small displace-
which is caused by higher order harmonics on Tension Leg Platform ments and rotations, the body boundary condition and the free
FOWTs. They found that such a nonlinear effect is determinant for a surface condition can be linearized as follows:
reliable estimation of short-term fatigue loads on tower and ten-
v4
dons. However, when a very large number of simulations needs to u2 4 þ g ¼0 (1)
be performed, e.g. in optimization procedures, the computational vz
cost of time-domain models becomes unaffordable. However, the
where u is the circular frequency, g is the gravity acceleration, z is
extraordinary efficiency of FD simulations, provided they are suf-
the vertical coordinates and 4 is the time-independent part of the
ficiently sophisticated to capture the coupled hydro-aero-elastic
velocity potential Fðx; y; z; tÞ. Considering a steady-state condition,
response of the system, makes them ideal for optimization pro-
the velocity potential can be written as follows:
cedures which are fundamental in early-stage design phases and
require a very large number of simulations. Response Amplitude
Fðx; y; z; tÞ ¼ 4ðx; y; zÞeiut ¼ ð4D þ 4R Þeiut (2)
Operators (RAOs) of significant DoFs of the floating platform are
chosen as objective functions, allowing to obtain an optimal solu- where 4D is the diffraction potential collecting the incident wave
tion which is independent from specific wind and wave inputs. In and scattering potentials, and 4R is the radiation potential. They can
addition to columns diameter and platform radius, also platform be calculated by solving two separate Boundary Value Problems
draft and mooring system properties are included in the set of (BVPs) which differ in the boundary condition at the wetted surface
design variables. Geometrical constraints related to mooring lines of the platform that reads as
and to the mass of the floating substructure are considered. Firstly,
single-objective optimizations are carried out considering design v4S v4
¼  0 (3)
variables related only to the floater geometry, namely side column vn vn
diameter, platform radius and draft, aiming to minimize the
response of one single DoF at a time. Secondly, another set of single- v4j
objective optimizations is performed considering both floater and ¼ nj ; j ¼ 1; 2; …; 6 (4)
vn
cable design variables. The purpose of this procedure is to improve
the results obtained by including the mooring system in the opti- 4S is the scattering velocity potential, 40 is the incident wave
v is the normal derivative in the direction of
velocity potential and vn
mization process.
The paper is structured as follows: in section 2 the FD model is the outward normal to the surface of the body. 4j is the radiation
presented and verified through comparisons with results of FAST velocity potential and nj is the direction of the jth DoFs. In the
simulations; in section 3 GA optimizations set-up are presented; diffraction problem, the body, fixed in the reference position, is
results of single-objective optimizations are reported in sections 4 subjected to an incident wave. In radiation problem, the fluid stays
and 5; comparisons between the optimized platforms also still and body oscillates in its six DoFs h ¼
considering the effect of power production are shown in section 6; ½ h1 h2 h3 h4 h5 h6 , namely Surge, Sway, Heave, Roll,
conclusions and future works are drawn and discussed in section 7. Pitch, Yaw. The dynamic pressure p at the wetted body surface is
1153
G. Ferri, E. Marino, N. Bruschi et al. Renewable Energy 182 (2022) 1152e1170

then calculated by means of the linearized Bernoulli equation:

2 3
Ax 0 0 Sx;Ay Sy;Ax 0
vFðx; y; z; tÞ 60 07
pðx; y; z; tÞ ¼  r Ay 0 0 0
vt
(5)
rffiffiffiffi 6 7
1 8 60 0 Az 0 0 07
where r is the water density. Integrating the dynamic pressure on BMor
¼ rCD s 6 7
2 p u 6 Sx;A 0 0 Ix;Ay þ Ix;Az 0 07
the wetted surface, the added mass (Equation (6)), radiation 4S y

0 0 0 Iy;Ax þ Iy;Az 0
5
damping (Equation (7)), and diffraction and Froude-Krilov forces y;Ax
(Equation (8)) are evaluated, respectively, as follows: 0 0 0 0 0 0
8 9 8 9 (13)
>
< >
= >
< >
=
Ajk ðuÞ ¼ Re ∬ pR , ndS ¼  rRe ∬ 4jR nk dS (6) where Ax ; Ay ; Az are the floater area projected on plane yx, xz, xy,
>
: SB >
; >
: SB >
; respectively. Sx;Ay ; Ix;Ay and Sy;Ax ; IyAx are the first and second
moment of area Ay and Ax ; with respect to x axis (roll) and y axis
8 9 8 9 (Pitch), respectively. Ix;Az and Iy;Az are instead the second moments
>
< >
= >
< >
= of area Az . The calculation of such geometrical quantities is detailed
Bjk ðuÞ ¼ Im ∬ pR ndS ¼  ruIm ∬ 4jR nk dS (7) described in Ref. [33].
>
: SB >
; >
: SB >
; Similarly, the hydrodynamic loads on the Heave plates are
described with a modified Morison equation [30], able to take into
account inertial, viscous and dynamic effects:
Xj ðu; bÞ ¼ ∬ pD ndS ¼  ru∬ 4D nj dS (8)
SB SB 1 p
dFMor
z ¼ rCAz VR D2 ðw € z Þ þ rCDz D2 ðw  q_ 3 Þjw  q_ 3 j
_ q
2 4
Such contributions are dominant on bluff elements, such as p p 2 
barge-like or cylinders where l/D < 5, l being the wavelength and þ D2 pb  D  D2c pt
4 4
D, the cylinder diameter. When the viscosity of the fluid flow
(14)
cannot be neglected, namely on slender elements, inertia and
viscous drag forces become dominant. Hydrodynamic loads per where CAz ¼ 1 is the added mass coefficient in the Heave direction;
unit-length can be expressed by Morison equation for moving CDz is the drag coefficient in the Heave direction; w is the vertical
cylinders as [30] component of the wave particle velocity; q_ 3 is the vertical velocity
of the Heave plates; D is the diameter of the Heave plates; Dc is the
p p 1 diameter of the upper column; and pb and pt are respectively the
dFMor ¼ rCM D2 u_  rCA D2 q€ þ rCD Dðu  qÞju
_ _
 qj (9)
4 4 2 dynamic pressures, projected in the direction of the normal vectors,
at the bottom and at the top of the Heave plates. VR is the reference
where u is the wave velocity component normal to the member volume, i.e., the volume of fluid that is displaced by the oscillations
_ q€ are the cylinder velocity and acceleration, CM is the inertial
axis, q, of the Heave plates, calculated as [30]
coefficient, CA is the added-mass mass coefficient and CD is the
viscous drag coefficient. The drag term in the right-hand side of Eq. A33 ð0Þ
VR ¼ (15)
(9) is expressed as a function of the six DoFs of the platform 3r
regarded as a rigid body. The viscous drag term being quadratic
requires a linearization. The Borgman linearization [31] is herein where A33(0) is the zero-frequency added mass coefficient in the
adopted, namely, the nonlinear drag term is approximated with an Heave direction. The viscous drag term is linearized according to
equivalent linear term written as Ref. [35] as follows:

rffiffiffiffi 1
1 1 8 dFHeaveplates y rCDz D2 uaðw  q_ 3 Þ (16)
dFDrag ¼ rCD Dðu  qÞju  qjy rCD D
_ _ s ðu  qÞ
_ (10) Drag;z 3
2 2 p u
where u is the wave frequency and a is the average of all the Heave
where su is the standard deviation of the relative velocity. In this plates oscillations amplitude. Only the damping terms of the
study, the exciting wave velocity is not considered for calculating equation is herein considered. By expressing the vertical velocity of
su. Denoting with h the vector of the six platform DoFs, neglecting the Heave plates as a function of the six platform DoFs, the
the inertial terms and the forcing terms of Eqs. (9) and (10) (see for contribution of Heave plates to the overall damping matrix is
example [32e34]), the linearized Morison force can be rewritten as calculated.
a function of h as follows:

rffiffiffiffi rffiffiffiffi 2.2. Hydrostatic, wind turbine and mooring lines contribution
1 8 1 8
dFMor y rCD D s q_ ¼  rCD D s Τh_ (11)
2 p u 2 p u Wind turbine and mooring lines contribution are estimated
directly from FAST linearized simulation around a steady state
where T is a matrix, which permits expressing the velocity of the operating point [8]. The hydrostatic stiffness matrix is calculated
members as a function of the six platform DoFs. Integrating Eq. (11) considering the effect of both the water-plane area and the Center
over the wetted surface of the floater, the total Morison force is [32] Of Buoyancy (COB) according to Eq. (8) [36], in which, due to
symmetry, the only non-zero terms are [37]
FMor ¼ BMor h_ (12)
Hydrostatic
C33 ¼ rgA0

1154
G. Ferri, E. Marino, N. Bruschi et al. Renewable Energy 182 (2022) 1152e1170

Hydrostatic
C44 ¼ rg∬ y2 dA þ rgV0 zCOB   
A0 AðuÞ þ MPlatform þ MTurbine þ h € ðtÞ þ BðuÞ þ BTurbine
  
Hydrostatic
C55 ¼ rg∬ x2 dA þ rgV0 zCOB (17) þ BMor ðu; b; a; sÞ h_ ðtÞ þ CHydro þ CMooring þ CTurbine hðtÞ
A0  
¼ b F D;K ðu; bÞ eiut
where A0 is the water-plane area, namely the surface resulting by (18)
the intersection of the platform with the Sea Water Level (SWL). V0
is the buoyancy volume, resulting by the Archimedes principle, zCOB b ðu; bÞeiut
hðtÞ ¼ h (19)
is the body-fixed vertical location of the COB of the supporting
platform. Platform Mass matrix, MPlatform , is calculated, according
where hb ðu; bÞ is the complex amplitude displacement vector of the
to Ref. [38], considering mass and inertia of the ballast with respect
platform. AðuÞ is the added mass matrix, BðuÞ is the radiation
to the Centre of Mass CM. Both hydrostatic and the diagonal terms
damping matrix, CHydro is the hydrostatic stiffness matrix, CMooring
of MPlatform are adopted as input for FAST. The mooring lines are
modelled adopting the quasi-static formulation developed and is the mooring lines stiffness matrix, bF D;K is the diffraction and
implemented by Jonkman [37]. This approach is very efficient and Froude-Krilov force vector. Considering Eq. (14), the equation of
sufficiently accurate for dynamic analyses as long as the mass of the motion can be rewritten as
mooring lines is small with respect to the overall mass of the sys-  
tem (around 8% [37]). By considering static assumption, the  u2 MTOT ðuÞ þ iuBTOT ðu; b; a; sÞ þ CTOT hb ðu; bÞ ¼ b
F D;K ðu; bÞ
nonlinear system of equations is solved iteratively for the tension
(20)
exerted by the catenary cable at the fairlead. This value of the
tension is used to iteratively update the platform position satisfying By solving the above linear system for h b ðu; bÞ, the Response
the equilibrium condition. Details on the formulation are available Amplitude Operator (RAO), i.e., the linear transfer function of the
in Ref. [37]. dynamic system, is calculated iteratively as follows:
Firstly, a FAST v7 linearized simulation around the un-displaced  
configuration of the system is performed with no wave and no RAOðu; b; a; sÞ ¼  u2 MTOT ðuÞ þ iuBTOT ðu; b; a; sÞ þ CTOT 1
wind, which means that rotor speed is set to zero. Also, added mass,  
b
F D;K ðu; bÞ
radiation damping, Diffraction and Froude-Krilov forces are set to
zero. Once mass, damping and stiffness matrices due to platform, (21)
turbine and moorings are evaluated, a FASTv7 time-domain simu-
In Fig. 1, the workflow of the developed model is summarized:
lation is performed at the rated wind speed in order to check the
stability of the system. Collective blades Pitch and rotor speed are
evaluated according to the WT steady state performance curves. 2.4. Code verification
The steady state operating point, which is reached by applying the
rated wind loads, is then adopted to perform a linearization and to To verify the proposed FD model, the 5 MW NREL OC4 DeepC-
obtain turbines and mooring contribution. wind semisubmersible FOWT [30] (see Fig. 2) has been analysed. In
Table 1 the main platform specifications are reported.
Fig. 2a shows the 5 MW semisubmersible hull surface, which
has been modelled in ANSYS AQWA for the calculation of Added
2.3. Equation of motion Mass, Radiation Damping, Diffraction and Froude-Krilov Forces. To
do so, a Hydrodynamic Diffraction analysis has been performed in a
The overall 6 DoFs equation of motion of the floating platform range of wave frequencies from 0 Hz to 0.2 Hz and wave heading
can be expressed as direction of 0 . The code solves the Radiation-Diffraction problem

Fig. 1. Frequency domain algorithm workflow.

1155
G. Ferri, E. Marino, N. Bruschi et al. Renewable Energy 182 (2022) 1152e1170

Fig. 2. ANSYS AQWA hydrodynamic model of the 5 MW wind turbine (a); 5 MW NREL semisubmersible platform [30] (b).

Table 1 From the RAOs of the system motions in Surge, Heave and Pitch,
5 MW National Renewable Energy Laboratory (NREL) DeepCwind semisubmersible the Fairlead tension RAO of the cable aligned with the 0 wave
platform specifications, SWL indicates Sea Water Level and CM indicates Centre of
Mass.
heading direction is calculated. Since the platform is a rigid body
and the hypothesis of small rotations holds, the displacements of
5 MW NREL DeepCwind Platform the cable Fairlead is obtained as superimposition of a pure trans-
Depth of platform base below SWL 20 m lation (Surge-Heave) and a pure rotation (Pitch). Therefore, the
Elevation of main column above SWL 10 m corresponding RAO can be evaluated as
Elevation of offset columns above SWL 12 m
Length of upper columns 26 m
Length of Heave plates 6m  
RAO1 ðuÞ RAO5 ðuÞzFair
Depth to top base columns below SWL 14 m RAOFairDisp ðuÞ ¼ þ (23)
Diameter of main columns 6.5 m RAO3 ðuÞ RAO5 ðuÞxFair
Diameter of offset columns 12 m
Diameter of Heave plates 24 m where RAO1 ðuÞ, RAO3 ðuÞ and RAO5 ðuÞ are respectively platform
Diameter of pontoon and cross-braces 1.6 m
Platform CM location below SWL 13.46 m
Surge, Heave, and Pitch RAOs. ðxFair ; zFair Þ are the Fairlead co-
Water depth 200 m ordinates with respect to the platform reference frame (see Fig. 4).
Since in the quasi-static approach the cable dynamics is neglected,
the horizontal and vertical components of Fairlead tension are
found by multiplying the Fairlead displacement by the stiffness
matrix of the single cable (evaluated at the steady state equilibrium
presented in section 2.1. by discretizing the wetted surface of the
point of the system)
floating platform with quadrilateral elements.
The results of the hydrodynamic model are verified with those

calculated by NREL in Ref. [30] adopting WAMIT. Fig. 3 shows the RAOFairH ðuÞ
RAOFairF ðuÞ ¼ ¼ KP RAOFairDisp ðuÞ (24)
comparison between the modules of the Hydrodynamic forces for RAOFairV ðuÞ
0 wave heading. ANSYS AQWA results are plotted with solid lines,
while the WAMIT ones provided in Ref. [30] are represented with where RAOFairH ðuÞ and RAOFairV ðuÞ are Horizontal and Vertical
circle markers. Fig. 3a shows Surge (blue) and Heave forces (red),
Fairlead tension RAOs. KP is the 22 stiffness matrix of the selected
while Fig. 3b presents the Pitch force. As it is possible to observe, a
cable of the mooring systems. In the case of slack catenary mooring
very good agreement is achieved.
lines, it can be calculated as follows[39]:
RAOs of Surge, Heave, Pitch and cable Fairlead Tension obtained
from the present FD model are compared with the ones obtained
directly in the time domain with FASTv8. Being RAOs the linearized 2 31
2 3 vl vl
transfer functions of the dynamic system, their amplitude is P
K11 P
K12 6 vH vV 7  
calculated in the time domain as follows: KP ¼ 4 P P
5¼6
4
7 ¼ DP 1
5 (25)
K21 K22 vh vh
  vH vV
 
Sh ðuÞ ¼ Swave ðuÞRAOðuÞ2  (22)
where DP is the flexibility matrix which is made of the partial de-
where Swave ðuÞ is the input wave spectrum and Sh ðuÞ is the rivatives of the cable profile l and h (see Fig. 4), with respect to
response spectrum in one of the six DoFs of the system. horizontal and vertical tensions [39]
1156
G. Ferri, E. Marino, N. Bruschi et al. Renewable Energy 182 (2022) 1152e1170

Fig. 3. ANSYS AQWA hydrodynamic forces comparisons: Surge, Heave forces (a) and pitch moment (b).

clearly identified at around 0.0085 Hz, 0.055 Hz and 0.0395 Hz for

Surge, Heave, and Pitch motions, respectively. The coupling of Surge
and Pitch DoFs is well caught in Fig. 5a and c. Also Heave eigen-
frequency peak is well predicted by the FD model (see Fig. 5b).
Fairlead tension RAO (Fig. 5d) is characterized by the dominant
excitation at Surge eigenfrequency, but peaks at Pitch and Heave
eigenfrequencies can be noticed.

3. Platform optimization

The present FD model is adopted to perform the optimization of

Fig. 4. Catenary mooring system loads at the fairlead. a semisubmersible-like platform suitable for the installation of a
10 MW WT. The 10 MW DTU reference WT is implemented in FAST
for this purpose [41]. The investigated platform is shown in Fig. 6.
This concept achieves stability by waterplane area and centre of
" # gravity draft, therefore it is widely investigated in literature for
vl L 1 V 1 V
¼ þ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi þ sinh large WTs [21,42,43]. Firstly, three platform-related geometrical
vH EA u H2 þ V 2 H variables are considered, which are side column diameter d, plat-
" # form radius r, and draft drf; then two mooring system-related
vl vh 1 H variables are added, leading to a five-variable optimization prob-
¼ ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  1
vV vH u H2 þ V 2 lem. Cross braces layout (i.e. the members connecting the side
" # columns) and number of columns are kept unchanged with respect
vh V 1 1 to the OC4 Deepwind 5 MW NREL platform [30]. Due to their
¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi þ (26) complex behaviour and the uncertainties in the definition of the
vV u H2 þ V 2 EA
drag coefficients which, as reported in Ref. [44], usually requires
Here L is the unstretched length of the cable, u is the cable mass experimental tests and CFD calculations, Heave plates characteris-
in water per unit length, EA is the axial stiffness of the cable V and H tics are not considered as design variables. Heave plates diameter
are the horizontal and vertical tensions at the fairlead. H and V are and thickness are set, according to the OC4 Deepwind 5 MW NREL
calculated by means of quasi-static cable model [36], which solves platform, to two times side column diameter and to 6 m, respec-
the catenary nonlinear system of equations with NewtoneRaphson tively. A symmetric 3-catenary-line mooring system is adopted to
iterative procedure. anchor the system. The nominal diameter and the mechanical
Ramachandran et al. presented a procedure to directly evaluate properties of the cables are chosen accordingly to the ones
RAOs for a FOWT in FAST [40]. Herein the same simulation set up is designed for the 10 MW DTU reference FOWTs [42].
adopted. Six computations of 8000s length are performed dis-
carding the first 2000s. A white noise sea spectrum in the frequency 3.1. Hydrodynamic coefficients considerations for the optimization
band between 0.005 Hz and 0.2 Hz of 1 m/Hz2 is applied, changing analysis
the wave seeds. Power Spectral Densities (PSDs) of significant
quantities, such as platform displacements and cable tension, are Even though FD model increases the efficiency of the optimi-
evaluated for each of the six simulation and averaged out. In order zation procedure, the calculation of the hydrodynamic coefficients
to reproduce FASTv7 mooring lines model, the module MAPþþ of by solving the potential flow problem still requires a high compu-
FASTv8 is adopted to simulate cable dynamics. Out of diagonal tational cost. Running this calculation inside the optimization loop
terms of the viscous damping matrix, BMorison i;j , are neglected ac- slows down the process. Fig. 7a upper plot presents Surge-Surge
cording to Ref. [30]. Results are presented in Fig. 5. As it is possible Added mass coefficient (A11) surfaces at a constant side columns
to observe, the results of the present FD model (red lines) are diameter equal to 14 m, but different platform radius and draft. x
generally in excellent agreement with those obtained from the time and y axes are labelled with the frequency and the platform radius,
domain simulations in FAST (blue lines). Eigenfrequency peaks are respectively. The value of A11 is shown on the vertical axis. Different

1157
G. Ferri, E. Marino, N. Bruschi et al. Renewable Energy 182 (2022) 1152e1170

Fig. 5. OC4 DeepCwind 5 MW semisubmersible platform results: Surge RAO (a), Heave RAO (b), Pitch RAO (c), Fairlead Tension RAO of the cable aligned with the 0 wave heading
direction (d).

plots reveal that when the shape of the platform is fixed and only
few geometrical quantities vary (such as d, r and drf), Added Mass,
Radiation Damping, Diffraction and Froude Krilov forces are relative
smooth functions of such platform dimensions. Therefore, they can
be evaluated off-line over a grid of values and then be interpolated
on-line. This procedure is adopted herein and multidimensional
fitted functions for the hydrodynamic parameters are evaluated
over a grid of values of the design space variables, see Fig. 7b. This
procedure allows not to call the potential flow solver (ANSYS
AQWA) at every run in the optimization algorithm, leading to a
drastic reduction of the computational time.

3.2. Optimization procedure

In this section, the optimization procedures are described in

detail. The GA available in MATLAB is herein adopted. It is a wide-
spread method for the solution of highly nonlinear optimizations
problems. The method generates at each time step a population of
individuals. Following the process of natural selection [45], GA
repeatedly modifies the generated population until it evolves to-
wards an optimal solution. The optimizations aim to identify the
platforms which minimize the system response in terms of Surge,
Heave and Pitch RAOs under some specific constraints as discussed
in the following sections.
Fig. 6. Semisubmersible platform concept: first group of three design variables d, r, drf.
3.2.1. Objective functions
The Surge, Heave and Pitch RAO amplitudes at the respective
eigenfrequencies are chosen as objective functions to be mini-
surfaces refer to different drafts, spanning from 20 m (blue surface) mized. The three RAOs considered are not only indicators of the
to 32 m (green surface). Similarly, the bottom of Fig. 7a illustrates platform dynamic response, but also of the turbine (and mooring
the Pitch-Pitch Added mass coefficient (A55) surfaces. These two system, as discussed later in Section 2.4) performances. Indeed, by
1158
G. Ferri, E. Marino, N. Bruschi et al. Renewable Energy 182 (2022) 1152e1170

Fig. 7. Added mass coefficients for a semisubmersible-like platform of 14 m side column diameter, varying platform radius (y axis), frequency (x axis) and platform draft from 20 m
to 32 m (a); grid of platforms adopted for the fitting of the hydrodynamic coefficients (b).

combining Surge and Pitch RAOs, assuming a rigid displacement, Additional constraints are related to the total length of the ca-
the RAO of the nacelle acceleration can be estimated as bles and on the length of the cable portion resting on the seabed
(see Fig. 8). The first one, ensure that the triangle side length cri-
€ ðuÞ ¼  u ðRAO1 ðuÞ þ RAO5 ðuÞhnac Þ
2
RAOdnac (27) terion is satisfied:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
where RAO1 ðuÞ; RAO5 ðuÞ are respectively Surge and Pitch RAOs L> ðxanch  xFair Þ2 þ ðzanch  zFair Þ2 (30)
and hnac is the height of the nacelle above SWL. This quantity has
been widely adopted in literature (e.g., Refs. [7,13]) to assess the WT
where L is the cable unstretched length, xanch , xfair are respectively
performances in the frequency domain. Considering Eqs. (23), (24)
the horizontal distances from platform centreline to the anchor
and (27), it is possible to state that optimizing for Surge, Heave and
point to the seabed and to the fairlead; zdepth  zFair is the vertical
Pitch means to indirectly optimize the turbine performances.
For each platform generated by GA, the RAOs are calculated by distance between the fairlead and the seabed (see Fig. 8). A
means of the FD model described in Section 2. Following the load constraint on a minimum portion of the cable resting on the seabed,
case adopted for the validation, a white noise sea spectrum in the LB , prevent the anchor to be stressed by uplift forces that can
frequency band between 0.005 Hz and 0.2 Hz of 2 m/Hz2 is applied. provoke failures. As suggested by Zhou et al. [32], at least one tenth
Wind speed is set to zero and the turbine is placed in parked of the mooring line is required to rest on the seabed. Assuming a
condition. quasi-static cable model [47], LB can be evaluated as

jVFair j L
LB ¼ L   (31)
3.2.2. Constraints umoor 10
A first optimization constraint regards the static displacements
of the system under the maximum turbine thrust configuration. where VFair is the vertical tension at the fairlead and umoor is the
Such condition is expected when a wind speed of about 11.4 m/s is cable weight in water per unit length.
acting at hub height. Since the static displacements of a FOWT are Finally, also one mass constraint has been considered. It is
mostly affected by aerodynamic loads, in this case the wave load is requested that the total mass of the substructure
neglected. Constraints are enforced both on Pitch and Surge DOFs. A (platform þ mooring system), Mtot , should not be higher than the
maximum allowable mean Pitch angle of 5 is adopted in order to mass of an upscaled semisubmersible platform, Mupscaled :
avoid any effect on the turbine functioning even at the rated wind
speed. Namely, the Pitch constraint reads as Mtot  Mupscaled (32)

h5  5 with Vhub ¼ 11:4 m=s (28)

It is noted that the mean Pitch value is mainly affected by the
hydrostatic stiffness of the platform CHydrostatic ð5; 5Þ. On the con-
trary, Surge static displacement of the system is completely
determined by mooring lines stiffness since the hydrostatic plat-
form stiffness in Surge is zero. An uncontrolled translational
displacement of the system may lead to the failure of the dynamic
electric cables connecting the wind farm to land. For this reason,
according to Ref. [46], a platform admissible offset to water depth
ratio of 0.15 is employed as a constraint:

Dxoffset
 0:15: (29) Fig. 8. Catenary cable quantities related to mooring system constraints.
zdepth

1159
G. Ferri, E. Marino, N. Bruschi et al. Renewable Energy 182 (2022) 1152e1170

Upscaling is a design procedure which allows to obtain a plat- adopted. In particular, fairlead position and cable length are
form suitable for a particular turbine by scaling a platform designed adjusted according to the 3 geometrical variables in order to keep
for a different turbine. This approach has been adopted in [18,42]. In the overall mooring stiffness matrix constant for all the possible
the present study, the 5 MW Deepwind semisubmersible platform platform geometries. This means that the catenary shape of the
is upscaled in order to obtain a platform suitable for the 10 MW mooring in the un-displaced position remains almost unaltered
DTU reference turbine. This procedure is carried on by scaling all when d,r, and drf vary.
the geometry by the square root of the ratio between the powers of A second set of 3 optimizations is then performed adding 2
the two turbines: design variables related to the mooring lines, with the intent of
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi improving the results obtained by the first set. In particular, the
10MW pffiffiffi anchor distance from platform centreline, xanch , and the
Sf ¼ ¼ 2 (33)
5MW unstretched mooring line length, L are added to the set of variables
(See Fig. 8 and Table 4).
The main geometry dimension of the new platform obtained by
upscaling each length by Sf are reported in Table 2. The diameter of
the central column was scaled with a different scaling factor to 3.2.4. GA optimizations settings
match the turbine tower base diameter. MATALB GA optimization set up is here presented. Each set in-
cludes 3 optimizations aiming to find the floating system which
minimizes the response in a single degree of freedom at a time,
3.2.3. Design variables namely Surge, Heave and Pitch. For the first set of optimizations, a
As mentioned above, a total of five design variables is consid- maximum number of 5 generations, each of them composed by 200
ered (see Figs. 6 and 8). For each variable, lower and upper bounds, individuals, are considered. After the first generation, the next
and the considered spacing are listed in Table 3. A first set of 3 populations are chosen accordingly to the fraction of crossover
optimizations is performed considering only the first three design children, set to be the 80%, and the number of elite children, 2. For
variables, namely side column diameter, d, platform radius, r, and the second set, when 5 design variables are adopted, the set up
draft, drf. In this first part of the study, the mooring lines are not slightly changes. The population size is raised to 500 individuals in
considered as design variables. To focus specifically on the platform order to deal with the enrichment of the design space, while the
optimization (first set), the same procedure described in Ref. [24] is number of generations is reduced to 4. Crossover fraction is set to
50% and elite children count to 1. Table 4 summarizes the optimi-
zation set up.
Table 2
10 MW DTU upscaled semisubmersible specifications.
4. 3-Variable optimization results
10 MW DTU upscaled platform

Depth of platform base below SWL (draft) 28.30 m In this section, the results of the optimizations with 3 design
Elevation of main column above SWL 14.14 m variables are presented. Results of the 5-variable optimizations are
Elevation of offset columns above SWL 16.97 m
Length of upper columns 36.77 m
discussed in the following section. For both cases, platforms cor-
Length of Heave plates 8.48 m responding to minimum Surge, Heave and Pitch RAOs are
Depth to top base columns below SWL 19.80 m compared with the upscaled platform.
Diameter of main columns 8.30 m
Diameter of offset columns 16.97 m
Diameter of Heave plates 33.94 m
4.1. Surge RAO optimization (3 design variables)
Diameter of pontoon and cross-braces 1.6 m
Platform CM location below SWL 21.27 m Fig. 9 presents the result of the Surge RAO optimization. Fig. 9a
Water depth 200 m shows the objective function, i.e., the Surge RAO peak, at each
Total mass 3.846eþ7 kg
generation compared with corresponding value associated with the
upscaled platform (solid black line). Fig. 9b and c illustrate the

Table 3
Design variables space.

Design variable space Optimizations

Lower bound [m] Upper bound [m] Spacing [m] First Set Second Set

d 14 20 0.5  
r 30 45 0.5  
drf 20 32 0.5  
xanch 570 950 10 
L xanch 20 xanch þ20 10 

Table 4
Optimization procedures set up.

First Set Second Set

Surge optimization Heave optimization Pitch optimization Surge optimization Heave optimization Pitch optimization

Population [¡] 200 200 200 500 500 500

N of generations [¡] 5 5 5 4 4 4
Cross-over fraction [%] 80 80 80 50 50 50
Elite children [¡] 2 2 2 1 1 1

1160
G. Ferri, E. Marino, N. Bruschi et al. Renewable Energy 182 (2022) 1152e1170

values of the Heave (9b) and Pitch RAO peaks (9c) for the optimal platform radius and side column diameter (blue line Fig. 10d), leads
platform, again compared with the upscaled one. It is observed that to the increase of the Surge peak in 1st and 2nd generations with
the minimum Surge peak (Fig. 9a) remains almost constant and respect to the upscaled result. On the contrary, such dimensions
significantly below the upscaled peak value for all the generations. affect positively the Pitch peak, which reaches its minimum where
A different trend can be seen for the Heave peak (Fig. 9b), which the Surge is maximum. Overall, the Heave optimized solution is
after the 1st generation drops down below the upscaled result, achieved with side column diameter of 17 m, platform radius of
scoring its minimum equal to 2.48 m/m (Gens 2 and 3). From Gen 4, 33 m and draft equal to 20 m. Heave optimization leads to smaller
an increase can be noticed, leading to a final value of 2.76 m/m. The platform radius, minimum draft and larger columns with respect to
Pitch peak shows a similar trend to the Heave one. Fig. 9d shows the Surge optimization.
evolution of the geometry, in terms of the design variables d, r, and
drf. The Heave peak trend shown in Fig. 9b appears to follow the
draft pattern (magenta line Fig. 9d) from generation 1 to 5. When
the draft reduces from 30 m to 21 m, the Heave peak drops 4.3. Pitch RAO optimization (3 design variables)
dramatically. As far as it regards Pitch peak, its variation appears to
be related to platform draft and radius (red line Fig. 9d). The min- Pitch optimization results are presented in Fig. 11. Surge peak
imum Surge peak is achieved with a relatively slender platform (Fig. 11a) presents a significant increase, reaching a final value of
with 14 m column diameter, 39 m platform radius and 22.5 m draft. 40 m/m which is 30% larger than the upscaled result. Heave peak
In conclusion, compared with the upscaled platform, the Surge (Fig. 11b) shows a reduction to 2.90 m/m. As regards the Pitch peak
optimized one presents a noticeable reduction of Surge, Heave, and (Fig. 11c), a more regular trend can be seen. Moreover, the final
Pitch RAO peaks. value of 2.15 deg/m reduces the upscale result of 1 deg/m. This is
achieved by maximizing both side column diameter (20 m) and
4.2. Heave RAO optimization (3 design variables) platform radius (45 m) (Fig. 11d). Such a heavy platform assures the
best performance in Pitch, but the worst in Surge among the
Fig. 10 shows results of the Heave RAO peak optimization. It is optimized ones. This is a clear effect of the coupling between these
possible to observe from Fig. 10a that the optimized platform pre- two DoFs. Comparing the side column diameter pattern (blue line
sents a smaller decrease of the Surge peaks if compared with Fig. 11d) with Surge peak one (Fig. 11a), an inverse relationship
Fig. 9a. Both Heave (Fig. 10b) and Pitch (Fig. 10c) peaks are notice- between these two quantities can be noticed. As for the previous
ably below the upscaled values and do not exhibit significant var- results, the draft remains equal to 20 m. Overall, the Pitch opti-
iations, attaining a minimum value of 2.30 m/m and 2.64 m/m, mized platform shows a reduction of Pitch and Heave peaks respect
respectively. As can be seen in Fig. 10d, the combination of a larger to the upscaled one, but a dramatic increase of the Surge RAO peak.

Fig. 9. 3-variable Surge optimization results; evolution of the Surge (a), Heave (b), Pitch (c) peaks during the Surge peak optimization; platform geometry evolution (d).

1161
G. Ferri, E. Marino, N. Bruschi et al. Renewable Energy 182 (2022) 1152e1170

Fig. 10. 3-variable Heave optimization results; evolution of the Surge (a), Heave (b), Pitch (c) peaks during the Heave peak optimization; platform geometry evolution (d).

4.4. Comparison of the results (3 design variables) have been reduced, allowing the algorithm to span the design space
with less restriction than in the previous optimizations. In the
Results of the first set of optimizations are summarized in following, the results are discussed and compared with the ones
Table 5. As it is possible to observe, a reduced draft has a positive obtained with three design variables and the upscaled platform.
effect in all the investigated DoFs. Surge and Pitch optimal plat- Fig. 13 shows the 5-variable Surge optimization. To facilitate a
forms show opposite value of column diameter, which can be direct comparison, the optimal value (at the 5th generation) ob-
explained by the coupled behaviour between these two DoFs. tained with the 3-variable optimizations is also reported in the
Moreover, the minimization of the Pitch motion leads to a signifi- figures (solid blue line). A reduction of the Surge and Heave peaks
cant increase of the Surge response (Fig. 11a). The three optimized (Fig. 13a and b) can be seen with respect to the 3-variable case
platforms are presented in Fig. 12. (compare with Fig. 9a and b). Also, the Pitch peak (Fig. 13c) de-
Following the consideration of section 3.2.1, the mitigation of creases with respect to Fig. 9c. As it is possible to observe from
the Pitch motions is considered of primary importance however, Fig. 13d, these results are achieved with a smaller radius, 35 m, and
the Surge optimized solution shows a better behaviour than the draft, 20.5 m. Such platform is very similar to the Heave optimized
upscaled one in all the three investigate DoFs. This case is deemed one, however, thanks to the optimized mooring system, Surge
to be the best geometrical arrangement when 3 design variables are peaks remain low if compared to Fig. 9b. Fig. 13e presents the
considered. evolution of the cable design variables. The distance between the
However, to investigate the role of mooring lines in the opti- anchor and the fairlead (Lhyp) (the square root term in Eq. (30)) is
mization process and to check whether even better results can be plotted instead of xanch, this allows to qualitatively estimates how
achieved, 2 mooring-related design variables presented in section slack/taut are the cables. As can be seen, the reduction of the Pitch
3.2.3 are added and a new set of optimizations are performed and peak results in an expensive cable, which lies in the upper bound of
discussed in the following section. the domain (xanch ¼ 970 m and L ¼ 950 m). This solution might be
unfeasible due to the high costs; however, it is important to high-
light that a possible trade-off between moorings length and system
5. 5-Varabile optimization results
behaviour is achieved also in generations 1 and 2, with a very
reduced cable system (xanch ¼ 730 m, L ¼ 710 m).
The 5-variables optimizations are performed adopting the set-
Also for the 5-variable Heave optimization (Fig. 14), a significant
tings presented in Table 4. A larger number of individuals per
gain with respect to the 3-variables case (Fig. 10) is achieved in the
generation is adopted. Also, in order to increase the flexibility of the
Surge peak (Fig. 14a), which drops to 20 m/m. Heave (Fig. 14b) and
analysis, the number of elite children and the crossover fraction
1162
G. Ferri, E. Marino, N. Bruschi et al. Renewable Energy 182 (2022) 1152e1170

Fig. 11. 3-variable Pitch optimization results; evolution of the Surge (a), Heave (b), Pitch (c) peaks during the Pitch peak optimization; platform geometry evolution (d).

Table 5
3 variables Optimization results.

Surge optimization Heave optimization Pitch optimization Upscaled platform

d [m] 14 16.5 20 17
14 - 20
r [m] 39 33 45 40
30 - 45
drf [m] 22.5 20 20 28
20 - 32
Surge peak [m/m] 20.53 27.87 40.71 29.90
Heave peak [m/m] 2.76 2.30 2.90 3.94
Pitch peak [deg/m] 2.76 2.64 2.15 3.20

Pitch (Fig. 14c) peaks shows a slight reduction. As it is possible to increase of the Surge and Heave peak can be noticed (Fig. 15a and b)
observe from Fig. 14d, such results are obtained with a slender and basically no reduction in Pitch (Fig. 15c). Again, the platform
platform compared to the Heave optimized in Fig. 12 and Table 5. remains almost unchanged (d ¼ 19.5 m r ¼ 45 m drf ¼ 20 m). Such a
Side column diameter decreases from 17 m to 14.5 m, and platform large and heavy structure reduces any effects of mooring lines in
radius from 33 m to 31 m. As it is possible to observe in Fig. 14e, a these two degrees of freedom. As can be seen from Fig. 15e, the
larger distance between L and Lhyp curves can be noticed compared optimized cable arrangement lies in the lower bound of the design
to Fig. 13e for 1st to 3rd generation. Qualitatively, it means that this variable domain. As expected, this leads to an increase of the Surge
mooring arrangement is slacker and tends to increase its vertical peak in Fig. 15a for generation 4 with respect to the 3-variable Pitch
stiffness with respect to the horizontal ones. This leads to larger optimized.
Surge oscillations and to smaller Heave and Pitch oscillations. In The 5-variable optimized platforms are plotted in Fig. 16.
generation 4, a tauter and stiffer mooring system is achieved,
leading to the reduction of the Surge peak value.
Results of Pitch optimization are reported in Fig. 15 where, an

1163
G. Ferri, E. Marino, N. Bruschi et al. Renewable Energy 182 (2022) 1152e1170

Fig. 12. 3-variable optimized platforms plane view and lateral view.

Fig. 13. 5-variable Surge optimization results; evolution of the Surge (a), Heave (b), Pitch (c) peaks during the Surge peak optimization; platform (d) and mooring system (e)
geometry evolution.

1164
G. Ferri, E. Marino, N. Bruschi et al. Renewable Energy 182 (2022) 1152e1170

6. Comparisons of the results beneficial effect of a tauter and wider mooring system (Figs. 13e and
14e), Surge 5v and Heave 5v optimized platforms perform better
All the optimizations performed are focused on the mitigation of that the corresponding 3-variable ones in all the three DoFs. On the
the system response at the floating platform eigenfrequencies, by contrary, the 5-variable Pitch optimized, due to a slacker mooring
optimizing the substructure geometry when the turbine is in a system (Fig. 15e), exhibits an increase in Surge and Heave peaks.
parked state. Fig. 17 summarizes the results obtained for Surge,
Heave, and Pitch DoFs. Blue gradient bars refer to 3-variable opti-
mizations, while red gradient ones to 5-variables optimizations. In 6.1. Effects of the turbine in power production
Fig. 17aec, vivid colours (blue and red) refer to the optimized DoF,
whereas lighter colours indicate what happens in the considered As already mentioned, the results presented so far are all
DoF when RAO of other DoFs are being minimized. For example, in referred to the case with the turbine in a parked state. The dynamic
Fig. 17a Surge 3v and Surge 5v indicate that the Surge RAO peak is response of coupled system with the turbine in power production
minimized with 3 and 5 variables, respectively. Heave 3v and Heave must also be accounted. To this end, a third set of 5-variable opti-
5v indicate what happens to the Surge RAO peak when Heave RAO mizations is performed aiming at minimizing RAO peaks in power
peak is minimized under 3 and 5 variables, respectively. No sig- production condition. A constant wind speed of 11.4 m/s is applied
nificant improvement of the results can be noticed moving from 3 at the hub height. As explained in section 3.2.2, this is the most
to 5 variables in the corresponding DoF of the optimization (vivid critical condition for static displacement of the system and there-
blue and red bars in Fig. 17aec are basically equal). Focusing the fore for tower and cables stresses. In the present FD model, aero-
attention on the Surge peak (Fig. 17a), Pitch optimized platforms dynamic and rotating rotor effects result in an added damping,
are found to behave worse than the upscaled platform (taller col- which is expected to reduce mostly the Pitch DoF RAO peak with
umn bars). Such an increase may lead to fatigue problems in respect to the parked condition. Values obtained from this third set
mooring lines in harsh sea states. On the contrary, both in Heave are compared with those corresponding to the power production
(Fig. 17b) and Pitch (Fig. 17c) peaks, all the 6 optimized platforms case with an upscaled system, and to the 3 and 5-variable cases
behave significantly better than the upscaled one. Due to the (optimized in parked condition), see Figs. 12 and 16. Results are
shown in Fig. 18 where green gradients bars refer to the 5-variable

Fig. 14. 5-variable Heave optimization results; evolution of the Surge (a), Heave (b), Pitch (c) peaks during the Heave peak optimization; platform (d) and mooring system (e)
geometry evolution.

1165
G. Ferri, E. Marino, N. Bruschi et al. Renewable Energy 182 (2022) 1152e1170

Fig. 15. 5-variable Pitch optimization results; evolution of the Surge (a), Heave (b), Pitch (c) peaks during the Pitch peak optimization; platform (d) and mooring system (e) ge-
ometry evolution.

Fig. 16. 5-variable Pitch optimization results; evolution of the Surge (a), Heave (b), Pitch (c) peaks during the Pitch peak optimization; platform (d) and mooring system (e) ge-
ometry evolution.

1166
G. Ferri, E. Marino, N. Bruschi et al. Renewable Energy 182 (2022) 1152e1170

Fig. 17. Optimization procedures comparison: Surge (a), Heave (b) and Pitch (c) peaks values among 3-variables (blue gradient colours), 5-variables (red gradient colours) opti-
mizations and upscaled (grey) values.

power production optimization, whereas blue and red gradient opposite, their dynamic behaviour in power production worsens.
bars refer to the 3 and 5-variabe optimized for the parked state, but Indeed, power production Pitch optimized systems (vivid green
set here in power production for the sake of comparison. The added bar) presents a slender platform.
damping leads to a significant reduction of the peaks in the Surge
(Fig. 18a) and Pitch (Fig. 18c) DoFs among all the optimized systems. 6.2. Mean value of the systems responses in power production
On the contrary, due to the large damping caused by the Heave
plates, Heave DoF peaks (Fig. 18b) remain almost constant with Mean values of displacements and stresses have not been
respect to Fig. 17b. As it is possible to observe in Fig. 18a, Surge considered as objective functions therefore some of the optimized
power production optimized system (vivid green bar) achieves a configurations could be unfeasible due to excessively high steady
better behaviour with respect to the parked optimized ones (vivid responses. These quantities are mainly affected by mean aero-
blue and red bars), the peak is reduced from 18.4 m/m to 16.2 m/m dynamic loads and they are maximum at the WT rated wind speed
as far as it regards Fig. 18c, an interesting trend is noticeable. The [32]. For this reason, steady Surge, fairlead tension, Pitch and tower
Pitch optimized in parked condition appears to behave worse than base bending moment, are calculated by means of a time domain
all the other optimized systems. This is probably caused by the simulation in FAST with no incoming wave and steady rated wind
difference in platform sizes. As can be seen in Figs. 12 and 16, speed (11.4 m/s) at hub height. Results are presented in Figs. 19e21.
parked Pitch optimized platforms are characterized by the largest 3-variable optimizations show a similar mean Surge displacement
column diameters and radii, which result in the highest stiffnesses since the cable properties are updated according to the platform
in rotational DoFs and in the lowest mean rotations. On the geometry to keep the mooring stiffness almost constant (Fig. 19a). A

Fig. 18. Power production RAO peaks comparisons: Surge (a), Heave (b) and Pitch (c) peaks values among 3-variables parked (blue gradient colours), 5-variables parked (red
gradient colours) optimizations, 5-variable power production optimizations (green gradient colours) and upscaled (grey) values.

1167
G. Ferri, E. Marino, N. Bruschi et al. Renewable Energy 182 (2022) 1152e1170

Fig. 19. Steady response comparison: Surge (a) and Fairlead tension (b) under rated aerodynamic loads (11.4 m/s at hub height), 3-variables (blue gradient colours), 5-variables (red
gradient colours) optimizations and upscaled (grey) values.

significant decrease of the Surge drift can be seen for the 5-variable leading to a huge reduction of the dimension of the whole system
Surge and Heave optimized due to the tauter mooring system. The (platform þ moorings). Moreover, by comparing Fig. 15a and b, it
maximum Surge displacement is reached by the power production results that steady Surge and steady fairlead tension are inversely
Pitch optimized. Overall, the minimum Surge displacement is related.
achieved for the 5-variables Heave optimized, which is 2.4 times Fig. 20 presents the steady (mean) Pitch angle and the mean
lower that the upscaled one. Despite its taut cables, 5-variable Pitch tower base bending moment values at rated wind speed. In contrast
optimized steady Surge increases with respect to the other 5- with Fig. 19, steady Pitch (Fig. 20a) and tower base bending
variables optimized platform. This is caused by the smaller foot- moment (Fig. 20b) are in direct proportion. Larger Pitch angles lead
print of the mooring system (xanch ¼ 610 m, see Fig. 15e), which to the increase of the eccentricity of the Rotor-Nacelle (RN) mass
leads to a reduced stiffness in Surge DoF. As can be seen from with respect to the reference frame, which results in a larger mean
Fig. 19b, despite the smaller footprint of the mooring system, the tower base bending moment with respect to the upscaled geometry
cable fairlead tension of the 5-variable Pitch optimized remains (see Fig. 19b). As expected, this is not the case when Pitch is
almost constant with respect to the 3-variable optimizations, minimized in parked condition (see bright blue and red bars in

Fig. 20. Steady response comparison: Pitch (a) and Tower base bending moment (b) under rated aerodynamic loads (11.4 m/s at hub height), 3-variables (blue gradient colours), 5-
variables (red gradient colours) optimizations and upscaled (grey) values.

1168
G. Ferri, E. Marino, N. Bruschi et al. Renewable Energy 182 (2022) 1152e1170

Fig. 20a and b). Especially in Fig. 20a, it is possible to observe that platform, supporting a 10 MW WT [41] anchored by a catenary
small and light platforms, such as Surge (d ¼ 14 m, r ¼ 35 m, mooring system, is studied. A first set of optimizations with 3
drf ¼ 20.5 m) and Heave (d ¼ 14.5 m, r ¼ 31 m, drf ¼ 20.5 m) 5- design variables related to platform geometry reveal that lightness
variables parked optimized, Heave and Pitch power production and reduced platform footprint can limit Surge and Heave ampli-
optimized (d ¼ 14 m, r ¼ 33.5 m, drf ¼ 20.5 m), exhibit larger pitch fications at the eigenfrequencies. On the contrary, Pitch oscillations
rotations. minimization requires the maximization of platform column
Finally, in Fig. 21, a comparison in terms of platform mass is diameter and waterplane area. A minimum draft seems to be ideal
shown. Even if in the parked Pitch optimized case the columns for the peaks mitigation of all the DoFs. A second set of optimiza-
diameter and radius are larger than the upscaled platform, the tions is performed adding two design variables related to mooring
reduced draft always ensures an overall mass smaller than the lines. Results show that the improvement in the optimization of the
upscaled case. specific DoF is negligible. However, benefits can be achieved on the
Considering Figs. 19e20, the reduction attained by the power other DoFs. Indeed, the optimization of the cable geometry for the
production Pitch optimized system (see vivid green bar in Fig. 18c), minimization of the Surge peak leads to the reduction also of the
which is about 0.5 deg/m with respect to the corresponding parked Pitch dynamic response. The study is completed with a third set of
optimized, cannot justify the high Surge (Fig. 19a), Pitch (Fig. 20a) optimizations in power production condition and with the com-
displacements and tower stresses (Fig. 20b). Therefore, this plat- parison of the mean values of some significant quantities, such as
form is considered unfeasible. Pitch and tower base bending moment. Among these sets, three
optimized system are identified. The Surge optimized platforms (3-
variable parked, 5-variable parked and 5-variable power produc-
7. Conclusion
tion) seem to be the best solution for the reduction of the 3 DoFs
dynamic amplification at the natural frequencies, controlling the
In the present paper, a GA-based optimization procedure is
steady displacements and stresses on tower and cables. This is
performed on a 10 MW FOWT in order to find the optimal sizes of a
achieved with very similar platform dimensions: minimized col-
semi-submersible supporting platform. The response amplitude
umns diameter (14 m), draft (20.5 me22.5 m) and radii from 39.5 m
operators (RAOs), namely the linearized transfer functions of the
(3v parked and 5v power production) to 35 m (5v parked).
system, are adopted as objective functions. Attention has been
Considering the mooring lines, the minimum cable length is ach-
focused on the response in resonant conditions, in order to identify
ieved by the Surge power production optimized (770 m), while the
the floating substructure properties such that the dynamic system
parked ones reach a length of 828.4 m for the 3-variable, and 950 m
amplification at natural frequencies of Surge, Heave and Pitch DoFs
for the 5-variable. Despite the very high Surge peak, also the Pitch
(peaks of RAOs) is the lowest. To this end, a linearized frequency
optimized in 5-variable seems to be a competitive solution for the
domain simulation model has been developed. The mooring system
steady response, gaining low Surge, Pitch displacement and tower
is modelled through a quasi-static approach. The wind turbine
stresses. This is achieved with a wide platform (d ¼ 20 m, r ¼ 45 m,
contributions are modelled as mass and stiffness matrices, which
drf ¼ 20 m) and a reduced mooring system (L ¼ 590 m,
are obtained from FAST linearized simulations. The present FD
xanch ¼ 610 m) which may compensate the higher costs of the
model has been verified using results obtained from FAST time-
floater. Overall, for each optimizations set, an optimal solution can
domain analysis on the 5 MW NREL Deepwind semisubmersible
be identified, that is the Surge 3-variable parked and 5-variable
FOWT [30]. In the optimization procedures, a semisubmersible-like
power production system, and the Pitch 5-variable parked sys-
tem. Future development will be focused on considering different
objective functions, related to the dynamic amplification outside
the eigenfrequencies. Also, the development of efficient reliability-
based optimization procedure will be studied in order to identify
floating configurations which minimizes the probability of failure
of the system given specific met-ocean conditions.

CRediT authorship contribution statement

Giulio Ferri: Conceptualization, Methodology, Software, Vali-

dation, Resources, Data curation, Visualization, Writing e original
draft, Writing e review & editing. Enzo Marino: Conceptualization,
Methodology, Validation, Resources, Visualization, Writing e re-
view & editing, Supervision. Niccolo
Bruschi: Validation, Visuali-
zation, Writing e review & editing. Claudio Borri: Supervision,
Project administration.

Declaration of competing interest

The authors declare that they have no known competing

financial interests or personal relationships that could have
appeared to influence the work reported in this paper.

References

[1] Z. Chuang, S. Liu, Y. Lu, Influence of second order wave excitation loads on
Fig. 21. Platform mass comparison: 3-variables (blue gradient colours), 5-variables
coupled response of an offshore floating wind turbine, Int. J. Nav. Archit.
(red gradient colours) optimizations and upscaled (grey) values.
Ocean Eng. (2020) 1e9, no. February.

1169
G. Ferri, E. Marino, N. Bruschi et al. Renewable Energy 182 (2022) 1152e1170

[2] K. Xu, M. Zhang, Y. Shao, Z. Gao, T. Moan, Effect of wave nonlinearity on fa- 182e199.
tigue damage and extreme responses of a semi-submersible floating wind [24] G. Ferri, E. Marino, C. Borri, Optimal dimensions of a semisubmersible floating
turbine, Appl. Ocean Res. (2019). platform for a 10 MW wind turbine, Energies (2020).
[3] K. Xu, Y. Shao, Z. Gao, T. Moan, A study on fully nonlinear wave load effects on [25] A. Mockute, E. Marino, C. Lugni, C. Borri, Comparison of nonlinear wave-
floating wind turbine, J. Fluid Struct. (2019). loading models on rigid cylinders in regular waves, Energies 12 (21) (2019)
[4] M.I. Kvittem, T. Moan, Time domain analysis procedures for fatigue assess- 1e22.
ment of a semi-submersible wind turbine, Mar. Struct. 40 (2015) 38e59. [26] E. Marino, C. Lugni, C. Borri, The role of the nonlinear wave kinematics on the
[5] E.E. Bachynski, T. Moan, Ringing loads on tension leg platform wind turbines, global responses of an OWT in parked and operating conditions, J. Wind Eng.
Ocean Eng. 84 (2014) 237e248. Ind. Aerod. 123 (2013) 363e376.
[6] E.E. Bachynski, M.I. Kvittem, C. Luan, T. Moan, Wind-wave misalignment ef- [27] E. Marino, A. Giusti, L. Manuel, Offshore wind turbine fatigue loads: the in-
fects on floating wind turbines: motions and tower load effects, J. Offshore fluence of alternative wave modeling for different turbulent and mean winds,
Mech. Arctic Eng. 136 (4) (2014) 1e12. Renew. Energy 102 (2017) 157e169.
[7] P. Sclavounos, C. Tracy, S. Lee, Floating offshore wind turbines: responses in a [28] AQWA User Manual 15317, 2012, pp. 724e746. October.
seastate Pareto optimal designs and economic assessment, in: Proc. Int. Conf. [29] J.R. Morison, J.W. Johnson, S.A. Schaaf, The force exerted by surface waves on
Offshore Mech. Arct. Eng. - OMAE 6, 2008, pp. 31e41. piles, J. Petrol. Technol. 2 (1950) 149e154, 05.
[8] E.N. Wayman, P.D. Sclavounos, Coupled dynamic modeling of floating wind [30] A. Robertson, J. Jonkman, Definition of the Semisubmersible Floating System
turbine systems preprint, in: Offshore Technol. Conf. 1e4 May 2006, Houston, for Phase II of OC4, 2014 no. September.
TX, 2006. [31] F. Savenije, J. Peeringa, Aero-elastic Simulation of Offshore Wind Turbines in
[9] J. Jonkman, M.L. Buhl, FAST User’s Guide 123, 2007, pp. 407e408, 6. the Frequency Domain, 2009.
[10] C.-H. Lee, WAMIT Theory Manual, WAMIT Theory Man, 1995, pp. 1e19. [32] S. Zhou, K. Müller, C. Li, Y. Xiao, P.W. Cheng, Global sensitivity study on the
[11] M. Philippe, A. Babarit, P. Ferrant, Comparison of time and frequency domain semisubmersible substructure of a floating wind turbine: manufacturing cost,
simulations of an offshore floating wind turbine, in: Proc. Int. Conf. Offshore structural properties and hydrodynamics, Ocean Eng. 221 (July 2020) (2021)
Mech. Arct. Eng. - OMAE 5, 2011, pp. 589e598. 108585.
[12] M. Karimi, B. Buckham, C. Crawford, A fully coupled frequency domain model [33] A. Pegalajar-Jurado, M. Borg, H. Bredmose, An efficient frequency-domain
for floating offshore wind turbines, J. Ocean Eng. Mar. Energy 5 (2) (2019) model for quick load analysis of floating offshore wind turbines, Wind En-
135e158. ergy Sci. 3 (2) (2018) 693e712.
[13] M. Karimi, M. Hall, B. Buckham, C. Crawford, A multi-objective design opti- [34] J.M. Hegseth, E.E. Bachynski, A semi-analytical frequency domain model for
mization approach for floating offshore wind turbine support structures, efficient design evaluation of spar floating wind turbines, Mar. Struct. 64
J. Ocean Eng. Mar. Energy 3 (1) (2017) 69e87. (November 2018) (2019) 186e210.
[14] S. Dou, A. Pegalajar-Jurado, S. Wang, H. Bredmose, M. Stolpe, Optimization of [35] L. Tao, D. Dray, Hydrodynamic performance of solid and porous heave plates,
floating wind turbine support structures using frequency-domain analysis Ocean Eng. 35 (10) (2008) 1006e1014.
and analytical gradients, J. Phys. Conf. Ser. 1618 (4) (2020). [36] J.M. Jonkman, Dynamics of offshore floating wind turbines-model develop-
[15] M. Brommundt, L. Krause, K. Merz, M. Muskulus, Mooring system optimiza- ment and verification, Wind Energy 12 (5) (2009) 459e492.
tion for floating wind turbines using frequency domain analysis, Energy [37] J. Jonkman, Dynamics modeling and loads analysis of an offshore floating
Procedia 24 (January) (2012) 289e296. wind turbine 8 (11) (2012) 1595e1606.
[16] N. Bruschi, G. Ferri, E. Marino, Influence of Clumps-Weighted Moorings on a [38] O.M. Faltinsen, Sea Loads on Ships and Offshore Structures, 1990.
Spar Buoy O Ff Shore Wind Turbine,, 2020, pp. 15e18, no. October 2017. [39] M.K. Al-Solihat, M. Nahon, Stiffness of slack and taut moorings, Ships Offshore
[17] G. Barbanti, E. Marino, C. Borri, Mooring system optimization for a spar-buoy Struct. 11 (8) (2016) 890e904.
wind turbine in rough wind and sea conditions, Lect. Notes Civ. Eng. 27 (2011) [40] G.K. V Ramachandran, A. Robertson, J.M. Jonkman, M.D. Masciola, Investiga-
(2019) 87e98. tion of Response Amplitude Operators for Floating Offshore Wind Turbines
[18] M. Leimeister, E.E. Bachynski, M. Muskulus, P. Thomas, Rational upscaling of a Preprint, 2013 no. July.
semi-submersible floating platform supporting a wind turbine, Energy Pro- [41] C. Bak, et al., Design and performance of a 10 MW wind turbine, J. Wind
cedia (2016). Energy (2013) 1e138. July.
[19] J. Liu, E. Thomas, A. Goyal, L. Manuel, Design loads for a large wind turbine [42] W. Yu, et al., Qualification of Innovative Floating Substructures for 10MW
supported by a semi-submersible fl oating platform, Renew. Energy 138 Wind Turbines and Water Depths Greater than 50m, 2017.
(2019) 923e936. [43] C. Allen, et al., Definition of the UMaine VolturnUS-S Reference Platform
[20] J. Liu, L. Manuel, Alternative mooring systems for a very large offshore wind Developed for the IEA Wind 15-Megawatt Offshore Reference Wind Turbine,
turbine supported by a semisubmersible floating platform, J. Sol. Energy Eng. 2020, p. 41.
Trans. ASME 140 (5) (2018) 1e9. [44] F. Lemmer, W. Yu, P.W. Cheng, Iterative frequency-domain response of
[21] J. Azcona, F. Vittori, U. Schmidt, F. Savenije, Design Solutions for 10MW floating offshorewind turbines with Parametric drag, J. Mar. Sci. Eng. 6 (4)
Floating Offshore Wind Turbines Document Information, vol. 308974, 2017, (2018).
p. 308974. [45] MATLAB, Optimization Toolbox TM User ’ S Guide R 2014 B, 2014.
[22] H. Bredmose, et al., The Triple Spar campaign: model tests of a 10MW floating [46] G. Benassai, A. Campanile, V. Piscopo, A. Scamardella, Optimization of Mooring
wind turbine with waves, wind and pitch control, Energy Procedia 137 (2017) Systems for Floating Offshore Wind Turbines 57 (4) (2015) 1e19.
58e76. [47] M. Masciola, J. Jonkman, A. Robertson, Implementation of a multisegmented,
[23] W. ting Hsu, K.P. Thiagarajan, L. Manuel, Extreme mooring tensions due to quasi-static cable model, in: Proc. Int. Offshore Polar Eng. Conf., no. March,
snap loads on a floating offshore wind turbine system, Mar. Struct. 55 (2017) 2013, pp. 315e322.

1170