A parametric optimization approach for the initial design

of FOWT’s substructure and moorings in Brazilian

deep-water fields
Jordi Mas-Soler1 ,Giovanni A. do Amaral2 , Luccas Z. M. da Silva1 , Edgard B.
Malta3 , Lucas Henrique S. Carmo1 , Felipe Ruggeri4 and Alexandre N. Simos1
1
Naval Arch. & Ocean Eng. Department, Escola Politécnica. University of São Paulo (USP), Brazil
2
Offshore Mechanics Laboratory - LMO, Escola Politécnica. University of São Paulo (USP), Brazil
3
Technomar—Engenharia Oceânica, São Paulo, Brazil
4
Argonáutica—Engenharia e Pesquisa, São Paulo, Brazil

Abstract. As Floating Offshore Wind Turbines (FOWT) continue to demonstrate their technical
feasibility in different projects, the conceivable scenarios for the application of this technology rapidly
expand. Outside the mainstream perspective of floating farms connected to onshore grids, several projects
currently study the technical and economic feasibility of adopting FOWTs to supply part of the electrical
power required by the offshore production of oil and gas. Having this goal as a main target, the present
work addresses initial results obtained in a comprehensive research project that aims to prospect potential
uses for FOWT technology in the context of offshore oil and gas production in the Brazilian oil fields. In
this context, a parametrical optimization procedure has been adopted to provide suitable candidates for the
floating substructures and their mooring systems. The optimization framework was designed for automatic
generation of geometrical concepts already considering the equilibrium involving hull weight, ballast and
the tensions applied by the mooring lines. The latter are modeled in hybrid configuration, composed of
sections of steel chains and polyester rope, considering both catenary and taut-leg configurations. The main
goal is to minimize CAPEX costs but, knowing that the acceleration levels in the rotor hub are inevitably
linked to indirect costs that are hardly predictable at this stage, a multi-objective approach is preferred,
taking the minimization of the nacelle acceleration as one of the drives. Moreover, for a better resolution of
the site-specific design, the optimization is done not only for a few pre-selected metocean conditions, but
considering a long-term series of the simultaneous action of seas, swell waves, currents and winds. In this
paper, two illustrative concepts with different floater geometries are presented, both of them optimized for
a water depth of 600m. A discussion on the mooring configurations is made and the rationale involved in
the optimization procedure that leads to the final floater dimensions is investigated by connecting the floater
motion responses to the wave characteristics in the chosen field.

Keywords— Floating Offshore Wind Turbine FOWT, Semisubmersible, Hull and Mooring Optimization,
Deep-water conditions

1. Introduction
There is a rise in interest from the Oil and Gas (O&G) companies in Floating Offshore Wind Turbines (FOWT)
stemming from reducing CO2 emissions and maximizing oil recovery (e.g. [1], [2] and [3]). In the Brazilian
O&G scenario, [4] conclude that electrification of the offshore operations in Brazil using offshore wind may
help companies to reduce operating expenses (OPEX), meet the environmental regulations and maximize their
production in the country’s offshore fields. However, the Brazilian offshore fields present some challenges, such as
water depths ranging between 500m and 2000m, as well as the long distances from shore. In this case, the use of
an optimization framework for the FOWT’s hull and mooring systems under the specific environmental conditions
and water depth specifications of the installation site may be helpful to prospect reliable designs.
One of the particularities of this research work is that the dynamics of each FOWT concept is evaluated using
a long-term environmental data with a total duration of 12 years. This procedure is adopted in order to guarantee
a better appraisal of the correlations between wind, waves and current, something that is particularly important in
a zone where swell waves are known to be a rather important aspect. Along with the assessment of the systems’
dynamics, the optimization process relies on the automatic generation of different hull geometries and mooring
arrangements, achieved using the Edtools R software [5]. The mooring system stiffness is evaluated analytically
using the formulation proposed in [6] and [7], taking into the catenary equations and the geometric relations
between the mooring lines and the floating body. It should be noted that the dynamic assessment also evaluates if
the systems meet some of the fundamental requirements set by the DNV’s technical rules [8].
The next section provides a detailed review of the recent developments regarding the optimization process. This
section is followed by the description of the main features of the optimization framework. The following sections
focus on case-study of site-specific optimization of the FOWT. A summary of the environmental conditions and
the wind turbine adopted (namely, the DTU10MW, [9]) precede the characteristics of the optimal FOWT concepts.
A discussion of some interesting features of the optimal concepts obtained is also presented. The final section of
this work presents the main conclusions drawn and the next steps that are envisaged.

2. State-of-the-art and Methodology

Optimization frameworks have been adopted during the modelling and design process of FOWTs in order to
overcome the technical challenges arising from the specific interactions of these systems with both wind and
waves. In particular, a series of works published over the last years (see, among others, [10, 11, 12, 13]) have
significantly contributed to this aim.
Regarding the development of optimization processes for FOWT, in [14] the authors used a genetic algorithm
to develop a small three-legged semisubmersible platform, referred to as MiniFloat. The resulting hull was
latter adapted to support a multi-megawatt windturbine, leading to the WindFloat technology. [15] provides an
assessment of different design parameters for five concepts of Tension Leg Platforms (TLPs) using coupled time-
domain simulations along with spreadsheet calculations. In [16] the authors optimized the hull and mooring system
for different FOWT concepts by means of single and multi-objective optimization processes. The results indicated
that the optimal concepts that populate the Pareto Frontier (PF) are featured with complex geometries. The
optimal solutions of this work were later reviewed in [17], where the authors updated the model and optimization
framework. The resulting approach algorithm sought to optimize the systems costs, which depended on the
structure masses and mooring system’s features. Both of these works are based on the Multi-Niche Crowding
Genetic Algorithm strategy (see [18]), which guarantees the development of a diversified population of concepts
over the generations, avoiding early convergence to local minima. An optimization framework for spar type
platforms is provided in [19]. The methodology proposed considered the tower design and structural properties
along with the hull main dimensions and mooring features. Moreover, the optimization strategy adopted in this
work relied on gradient-based optimization, which stand as a desirable methodology for large design spaces.
A limitation of previous studies on FOWT design optimization concerns the assessment of the coupled hull-
mooring’s dynamics. Since turbines are getting larger and large water depths characterize regions suitable for
offshore wind energy generation, the equivalent mooring behavior should be modelled taking into account the
correct offset position of the platform for each sea condition [6]. In addition, the aforementioned studies generally
consider a reduced set of selected environmental conditions during the optimization. The adoption of long term
environmental series allows to properly assess the hull-mooring coupled dynamics and the fulfillment of the safety
class requirements with a better resolution of the metocean particularities of the installation site.
To address these issues, the present paper will investigate the use of multi-niche crowding genetic algorithm
strategy combined with the frequency-domain modelling of the system dynamics and the explicit formulation for
the stiffness matrix around a generic offset position. The heuristic optimization approach proposed provides a
computationally efficient framework, allowing the evaluation of the system dynamics and mooring performance
by means of long term series of environmental conditions in the global optimization loop.
The methodology proposed relies on the discussion of the optimization rationale leading to the FOWT features.
To accomplish this goal a case study has been conducted, considering two different geometrical arrangements of
semisubmersible platforms intended to support a state-of-the-art 10MW wind turbine. Selected concepts from the
Pareto Frontier stand as the most promising semisubmersible platforms for the site-specific conditions assessed.
Floater motion responses related to the wave characteristics in the chosen field are investigated by means of the
RAO of nacelle acceleration at each sea state condition. The outputs obtained also allow the exploration of the
cost implications on the FOWT platform features and performance, which stands as an important driver of the
optimization framework

3. Optimization Framework
The optimization of FOWT systems has been tackled by means of both mathematical programming methods
and (meta)heuristic methods. However, the adoption of mathematical programming strategies during the FOWT
optimization may lead to inefficient approaches. In this research, a Genetic Algorithm has been adopted to obtain
high-quality solutions of the FOWT optimization problem.

3.1. General Overview of the Optimization Strategy

The optimization approach relies on the Multi-Niche Crowding Genetic Algorithm strategy as proposed in [18].
Figure 1 provides a simple flow diagram of the approach adopted. The rationale relies on three basic steps, namely:
selection, recombination and replacement. The individuals that will produce an offspring are selected during the
first step, which grants each individual of the population the same chance for mating in every generation. The pairs
of individuals resulting from this stage exploit the similarity 1 between individuals and, at the same time, allows
some amount of explorations through the mating of few individuals from different niches.
The crossover and mutation operations stochastically generate new solutions by means of the recombination
of the genetic information of the existing population. For this purpose the well known k-point crossover operator
was adopted, recombining the genetic information from the parents’ chromosomes at randomly selected k-points.
During the recombination step, the diversity of the population is assured through the mutation operator, with a given
probability of occurrence (pm ) for each individual, by means of varying the value of a randomly selected number
of alleles within a specific range (see [20]). During the replacement step, the rationale uses a rule called Worst
Among Most Similar, which was initially proposed in [21] and stands as a variation of the Restricted Torunament
Selection method. The goal of this operator is to reinsert in the population the candidates generated during the
recombination step. In this method the population is split randomly into a pool of f subgroups, each one with
g solutions (individuals). Then, f individuals are selected from the subgroups based on their similarity with the
candidate resulting from the recombination step, which will replace the individual with the lowest fitness among
the f individuals selected, even if it has lower fitness than other individuals. This replacement strategy, along with
the selection approach, improve the diversity of the individuals crowding the population. This feature avoids early
convergence to local minima, mimicking the evolutionary process that results in the non-survival of the systems
of each sub-group that are not-fit-enough. The reader interested in a more detailed description of the Multi-Niche
Crowding Genetic Algorithm is advised, once again, to refer to [18] .
The following subsections provide the specifics of the optimization approach and the parametrization scheme
adopted for defining the geometries of the design space of the semisubmersible FOWT platforms.

3.2. Objective Function

Cost reduction is a major goal during the design process in any project and it can be achieved through innovative
and creative designs. As for the cost estimation of FOWTs, relevant data available in the literature is limited. [22]
provides a systematic analysis of the hull and mooring CAPEX cost considering five different types of FOWTs. In
[17] the authors estimated anchors acquisition costs in terms of their maximum steady-state load and the technology
type, which is selected based on the angle of mooring line at the anchor connection. As regards the installation
costs, the main driving factors are explored in some works (e.g. [23]), but the data is scarce and it is limited to few
components of the FOWT. Since the current paper is focused in the preliminary design of a FOWT, the estimation
and optimization of the installation cost has been left for future studies.
Obviously, cost-limited projects may conflict with high performance standards. The performance assessment of
FOWTs involves several aspects, such as dynamic responses, energy and structural fatigue. However, considering
all these aspects explicitly in a multi-objective optimization approach is not practical. Many authors try to
overcome this limitation by selecting the nacelle horizontal acceleration as the main performance parameter to
be minimized, since it is known to be directly correlated to the turbine operational downtime and also to the
1
Similarity metric between the individuals of the population has been evaluated according to the genotypic features of the hull
and mooring system.
 Figure 1. Flow diagram of the optimization framework.

fatigue of the structures. The goal of the optimization problem here is, therefore, to exploit the cost possibilities of
the FOWT platform and its performance by minimizing the multi-objective function provided in Eq.(1).
 N ! 

X ai mooring 

min {F} = min  , CostCAPEX , CostCAPEX 
hull
.
 
wi · (1)


0.2g

 

i=1
ai
where stands for the maximum horizontal dimensionless acceleration at the nacelle’s center of mass for the ith
0.2g
sea condition. The selected scaling factor, 0.2g, is typically understood as the threshold for safety operation of the
wind turbine [24]. Accelerations values, ai , were computed for each sea state using the frequency domain RAOs,
which were estimated with the software WAMIT R assuming a damping equal to 5% of the critical damping for all
the motions. These predetermined linear damping values are in accordance with the estimations provided in the
literature for similar floater configurations (e.g. [25]). wi is a weighting function2 for the N metocean conditions
wi = 1. The second and third terms in Eq.(1) regards the estimated hull and mooring costs
PN
evaluated, with i=1
of the system, which are computed using the data provided in [22] and [17]. The main features of the cost model
adopted are summarized in Table 1. Due to the fact that the optimization uses a single reference wind turbine, its
cost does not impact the final output and, therefore, it is not included in the objective function.
2
The analysis of the weighting function has been included for the interested reader in a dedicated appendix (Appendix A).
Table 1. Cost model features. Three different types of anchor are used, named: Drag Embebed Anchor
(DEA), Vertically Loaded Anchor (VLA) and Suction Plate Anchor (SPA).
CAPEX Description value Units
Hull Acquisition & Manufacturing 2.100,00 [US D/ton]
Mooring Chain Acquisition 2.000,00 [US D/ton]
Mooring Synthetic Acquisition 2.800,00 [US D/MBL/ton · m]
Anchors Acquisition DEA: 100; VLA: 120; SPA: 150 [US D/kN/un]

3.3. Design Variables and Automatic Generation of Geometrical Concepts

Two different hull arrangements are explored in this work: a semsubmersible type platform with a central
column (henceforth named FOWTC) and a semisubmersible platform without a central column (henceforth named
FOWTO). All hull geometries are composed by a given number of equally spaced outer columns with a common
center, whose diameters and distance from this center are variable. For the FOWTC, the central column diameter
is also a variable, with a minimum value equal to the wind turbine tower-base diameter. The platform hull is also
featured with a variable draft and a freeboard equal to 15m as a constant parameter for all the platform designs.
The pontoons are modelled as rectangular prisms with variable height, length given by the distance of the outer
columns to the center and width equal to the diameter of the outer columns.
The mooring system configuration for each design is determined by a variable number of mooring lines per
outer column, with three sections. The central section is a polyester line, while both ends are modelled as R4 stud
link type chains, whose diameters are variables. The total length of each line is the result of the variable mooring
radius, the variable length ratio between the central and anchor connection segments and a variable λ, which stands
as a similarity ratio between the line geometry and the fairlead and anchor radius. Thus, the mooring lines shape
may change between a (non-vertical) taut configuration (λ = 0) and a slack catenary line configuration (0 < λ ≤ 1).
Details regarding the design parameters are provided in Table 2, including the constrains applied to these
parameters. The 11 geometric design variables include: (a) the number of outer columns (Nc ), (b) diameter of the
outer columns (Do ), (c) diameter of the central column (Dc ), (d) draft of the floater (T ), (e) distance of the outer
columns to the center (Rh ), (f) pontoon height (H p ), (g) number of lines per column (Nl ), (h) diameters of mooring
segments (D poly and Dchain ), (i) anchor to depth radius (Rd ), (j) geometry ratio of the line (λ), and (k) the ratio
between segment lengths (delta).

Table 2. Geometric variable design of the system. The relations provided in this table have been obtained
heuristically and aim at generating physically feasible systems.
Parameter Min. Max. Units
Nc 3 6 [-]
Do 0.6Dc 25 [m]
Do (i f, Dc = 0) 10 25 [m]
Dc 10 25 [m]
T 15 30 [m]
Rh 0.55 ∗ (Dc + Do ) 2.05 ∗ (Dc + Do )] [m]
T
Hp 2 2 [m]
Nl 2 3 [-]
D poly 1 40 [cm]
Dchain 1 59 [cm]
Rd 0.5 2 [-]
lambda 0.2 1 [-]
delta 0.05 0.9 [-]
3.4. Constraints and Criteria
The phenotypic features of the FOWTs concepts obtained through the parametrization process are constrained by
a set of criteria, which guarantees that the resulting systems meet the design requirements and, at the same time,
keep the design space physically feasible.
First, the intact stability of the platform is assessed through the ratio between the area under the righting
moment and the heeling moment, which shall be larger than 1.3. Furthermore, the intact stability of the platform
is also ensured by imposing that the maximum static heeling angle due to the wind loads shall be less than eight
degrees.
Secondly, the optimization process will naturally result in heave motions featured with high natural frequencies
(due to the reduction of the structure mass, related to the cost reduction goal of the optimization process).
Therefore, it has been set that the natural period of the heave motion shall be larger than 15 seconds. The feasibility
of the pontoon dimensions is assessed through the maximum bending stress, estimated by applying the static loads
acting on the structure (i.e. structural mass, ballast mass, mooring loads at the fairleads, wind turbine and tower
mass). The resulting bending stress is compared with the yield strength of the steel.
In the third place, the behaviour of the mooring system is constrained by the maximum admissible mean offset,
which is equal to 30% of the water depth. The safety factor of the line, estimated as ratio between the resulting
maximum static tension to the maximum breaking strength (MBS) of the respective segment, shall be larger than
2.75. This value was based the design tension from [26] for Consequence Class 2 structures, and considering a
Dynamic Amplification Factor (DAF) of 1.5. As the quasi-static catenary model is used, the maximum tension is
determined by the environmental condition leading to the maximum distance between the anchor connection and
the fairlead of one of the lines of the system. The steady-state mean offset value is estimated using the equivalent
mooring stiffness matrix and the mean forces exerted on the FOWT, namely: the rotor thrust, mean wave drift force
and current forces on the hull.
Finally, it is important to highlight that the model also allows (non-vertical) taut configurations, resulting
from the variation of the tension angle at the anchor connection and the adoption of different anchor types (also
depending on the tension angle).

4. Case Study
The case study was based on the optimization of a semisubmersible type FOWT with a 10 MW Turbine. The
upwind three-bladed reference turbine was designed by DTU (Technical University of Denmark) as a 10 MW unit
[9], that results from a revised up-scaling of NREL’s 5 MW reference turbine [27]. According to the technical
specifications of the wind turbine, the hub is located at 129m from the tower base. The main particulars of the
wind turbine are reported in Table 3.

Table 3. Specifications of the DTU 10 MW turbine.

Parameter Value Units Parameter Value Units
Cut-in wind speed 4 [m/s] Rotor diameter 178.3 [m]
Cut-out wind speed 25 [m/s] Hub diameter 5.6 [m]
Rated wind speed 11.4 [m/s] Hub height 129 [m]
Rated power 10 [MW] Rotor mass 227962 [kg]
Rotor speeds 6 – 9.6 [rpm] Nacelle mass 446036 [kg]
Number of blades 3 [-] - - -

The installation site was supposed to be placed in deep-water conditions (about 600-meter depth), situated
approximately 180km offshore Rio de Janeiro, Brazil. The site-specific environmental information adopted for
this case-study was obtained from the ERA5 public database [28]. Out of the data provided in the referred data
base, 35,065 environmental conditions were selected, each one standing for a 3h-long environmental condition and,
therefore, spanning through more than 12 years. The use of long term environmental data allows the optimization
of the FOWT concepts taking into account the correlations between wind, waves and currents. Furthermore, the
optimization framework has been developed aiming at speeding-up the processing the large sets of environmental
conditions. These features allow the assessment of long-time series during the optimization without significantly
increasing the computational time, if compared with the framework performance for few environmental conditions.
 The set of environmental conditions selected range from mild waves with high probability of occurrence to
waves with significant wave heights (H s ) up to 5.1m and 16.4s of peak period (T p ). Each sea spectrum was
obtained as a directional bimodal spectrum through the superposition of two JONSWAP wave spectra from the
wind sea and swell components. The selected installation site shows a consistent trend with the swell seas featured
with larger amounts of energy than the wind sea components. The environmental conditions are combined with
wind-induced currents, with speed values up to 0.3m/s and a predominant S/SW direction.
Regarding the wind features, they oscillate seasonally with a predominant blowing direction from NE and
speeds up to 21.66m/s (for a reference height of 100m). The heading of the platform was set equal to 25.7 deg.,
which stands as the direction with largest amount of wind energy accumulated through the 35,064 environmental
conditions selected.
The assessment of the dynamics of the platform and mooring responses for each individual is based on a static
approach of the environmental loads. The seakeeping problem has been addressed in the frequency domain using
the acceleration RAOs of the platform referred to nacelle’s center of mass. Regarding the wind loads acting on the
structure, they are estimated using the rotor thrust, according to the wind speed and the thrust coefficients provided
in [9]. The drag forces induced by the currents are computed through a Morison model, which uses a transversal
drag coefficient equal to 0.61 and 2.5 for the columns and pontoons, respectively.
As described in section 3.3, this work aims at optimizing two different FOWT topologies, named FOWTC
and FOWTO. For the FOWTC concepts, the wind turbine is always assumed to be placed on the central column,
while for the FOWTO platforms it is supported by the bow column (i.e. outer column positioned in the heading
direction). As a consequence of the differences between these concepts, the optimization of the full design space
may create several niches, resulting in the suppression of the framework’s ability to fully explore the Pareto front
in a single run. For this reason, the FOWTC and FOWTO concepts were optimized in two different runs. In both
cases the results were generated by an identical framework using the parameters described previously, apart from
the differences regarding the wind turbine position and columns.

5. Results
The optimization results are presented in this section, aiming at demonstrating the rationale behind the framework
and identifying the most promising systems from the design space described in sections 3.3 and 3.4. The
optimization runs for the FOWTC and FOWTO were extended to 200 generations, each one with a population of
200 FOWT concepts. The evaluation steps of each individual are highly time consuming resulting in computational
time required to process a single generation (using a conventional desktop PC) equal to 45 minutes3 .

5.1. FOWTC multi-objective optimization

The objective function goals exploration resulted in a Pareto set composed by 25 concepts, with costs varying from
30.9 to 33.3 MUSD and weighted dimensionless accelerations varying between 0.14 and 0.16. As an illustration,
a concept has been selected based on its estimated cost, equal to 30.9 MUSD, which includes the hull acquisition
and manufacturing cost as well as the mooring acquisition cost.
The selected concept is a semisubmersible platform with a catenary mooring system, draft of 21.7m, outer
columns’ diameter of 12.5 and central column’s diameter equal to 20.8m, and radius of the outer cylinder array
equal to 37.7m. The resulting displacement is 25, 850 tonnes, being 18, 042 ballast tonnes and 324 tonnes of
mooring load. The mooring system is featured with anchor to depth radius of 2.4 and six lines, each one with
1, 564.6m of total length (of which 555.1m correspond to the first segment of the line in contact with sea floor,
931.3m of the polyester segment and 78.2m regarding the top chain segment). The maximum offset estimated is
35.1m, for which none of the anchors presents vertical loads. Figure 2 displays the main views of the FOWTC
geometry and mooring layout.
The minimization of the weighted horizontal dimensionless acceleration of the nacelle (see Eq.1) aims at
reducing the responses of the platform within the frequency range with the largest expected wave energy. The
results obtained illustrate the reduction of the acceleration responses along the generations. For instance, the
selected concept helps to exemplify how optimization framework leads to hull geometries that favor the attenuation
of accelerations for the metocean characteristics inherent to the long-term environmental series. For this case, Fig.
3 provides the 12 yr average wind sea and swell energy spectra, together with the nacelle’s horizontal acceleration
3
The estimation of the RAOs of each individual was carried out using 64CPUs from the CH-TPN’s cluster of the University
of São Paulo [29]
 Figure 2. FOWTC-I concept.

RAOs for the directions of interest, i.e. the ones with largest amount of energy: dashed red line for θ = 170 deg.
(regarding the wind sea) and blue line for θ = 345 deg. (for the swell sea). The trend in the acceleration RAOs
indicate minima within the period range where both (wind and swell seas) present their respective peaks of energy.

0.5 0.1

Swell Sea ( = 170deg.)

0.45 0.09
Wind Sea ( = 345deg.)
= 345deg.
0.4 = 170deg. 0.08
12yr-Average Sea Spectrum [m2s]

0.35 0.07
[(m/s2)/m]

0.3 0.06
acc.-htal.

0.25 0.05
RAOnacelle

0.2 0.04

0.15 0.03

0.1 0.02

0.05 0.01

0 0
5 10 15 20 25 30
Period [s]

Figure 3. FOWTC accelerations RAOS for the directions with largest amount of energy corresponding
to the wind sea (θ = 345 deg.) and swell sea(θ = 170 deg.). For graphical purposes, RAOs and (wind
and swell) sea directions are both referred to the bow of the platform.

5.2. FOWTO multi-objective optimization

The resulting PF is composed by 23 concepts, which show no significant differences with the acceleration levels
obtained for the FOWTC topology, and cluster around a hull cost of 27 MUSD and mooring costs close to 2
MUSD.
The concept selected from the PF, namely FOWTO, is featured with a catenary mooring system, with a
displacement of 18, 919 tonnes, being 16, 288 ballast tonnes and 218 tonnes of mooring load. The floater draft
is 23.7m, the radius of the outer cylinder array of 28.8m and column’s diameters of 16.7m. Figure 4 displays the
main views of the FOWTC geometry and mooring layout, with an anchor to depth radius equal to 3.1 and a total
of six lines, each one with 1, 934.7m of total length (662m correspond to the first segment of the line in contact
with sea floor, 1176m of the polyester segment and 96.7m regarding the top chain segment). The estimated cost
resulting from these features is 28.33 MUSD.

Figure 4. FOWTO-I concept.

The evaluation of the horizontal accelerations of the nacelle for the FOWTO concept drawn similar conclusions
to the ones obtained with the assessment of the dynamics of the FOWTC concept. Thus, the minima of the RAOs,
once again, are located within the wave period range featured with the largest expected amount of energy.

0.5 0.1

Swell Sea ( = 170deg.)

0.45 0.09
Wind Sea ( = 345deg.)
= 345deg.
0.4 = 170deg. 0.08
12yr-Average Sea Spectrum [m2s]

0.35 0.07
[(m/s2)/m]

0.3 0.06
acc.-htal.

0.25 0.05
RAOnacelle

0.2 0.04

0.15 0.03

0.1 0.02

0.05 0.01

0 0
5 10 15 20 25 30 35
Period [s]

Figure 5. FOWTO accelerations RAOS for the directions with largest amount of energy corresponding
to the wind sea (θ = 345 deg.) and swell sea(θ = 170 deg.). Being the RAOs and (wind and swell) sea
directions both referred to the bow of the platform.

The assessment of the accelerations obtained for each sea condition for the concepts populating the PF shows
that the maximum acceleration expected for the nacele never exceeds the threshold of 0.2g considering the
conditions in the 12 yr environmental series. This is certainly a consequence of the relatively mild sea conditions
that characterize the region.
6. Conclusions
This work presented the FOWT optimization results for two topologies of semisubmersible platforms, using a
10MW wind turbine. The optimization framework was based on a multi-objective genetic algorithm, which aimed
at minimizing the mooring and hull systems costs, and the maximum horizontal acceleration at the nacelle. Cost
estimations were based on the combination of hull and mooring cost models available in the literature. A quasi-
static mooring system model was adopted along with a linearized hydrodynamic model to define the dynamic
performance of the FOWTs and evaluate if they meet the criteria adopted.
The response of the Pareto front for both topologies evaluated was tracked by means of reapplying the
optimization framework. The two resulting optimal sets clustered around a hull cost of 28 MUSD and 2 MUSD
mooring costs and accelerations levels below 0.2g. It is important to highlight that the concepts obtained from the
optimization stand as a preliminary design for a FOWT and they shall be carefully evaluated by means of non-
linear dynamic simulations, and if it is the case, improved following technical requirements from the classification
societies. The selection of a final preliminary design, when comparing different topologies, shall be based on the
features estimated by the optimization framework (i.e. maximum accelerations and costs) as well as the specific
requirements specification of the project (e.g.. construction facilities, installation equipment and hardware).
The metocean data considered included 35,065 environmental conditions, encompassing a period of 12 years
of data for the selected installation site. Overall, the concepts in the Pareto fronts show a clear trend to minimize
the nacelle’s acceleration responses within the wave frequency range that contains the largest (wind and swell) sea
energy. Even though the use of a linear hydrodynamic model is reasonable for the sake of comparing different
floater geometries, the maximum acceleration values are non-conservative (i.e. they are likely underpredicted) and
thus more robust non-linear time-domain simulations would be desirable to validate the results obtained.
Regarding the rationale of the optimization, the study of the best-fit concepts shows that FOWTs with largest
hulls are characterized with robust mooring systems and small accelerations. The robustness of the mooring
system is conditioned to the diameter of the lines and the total number of lines per column and, therefore, different
combinations of these parameters may result in mooring systems with similar features. However, the optimization
framework adopted is indifferent to the variations in the number of lines, for example, since the cost estimation
model adopted does not include the installation expenses.
Some future work threads arise naturally from the present work. First, the lack of a goal that properly tracks
the installation costs may result in the non-fit enough concepts crowding the Pareto front. Also, adding the wind
turbine cost to the estimated cost of each concepts may help to assess the trade-off between the cost improvement
and its associated risks. Addressing these issues, however, is conditioned to the availability data concerning these
costs. Attention can also be paid to the ballast distribution, which can be optimized aiming to attenuate the energy
losses due to the inclination angle of the platform induced by the mean rotor thrust.

Acknowledgments
This work was developed as part of the R&D project conducted by Petrobras and the University of São Paulo
entitled ”Research and Development on Deep Water Floating Offshore Wind Turbines” (agreement Petrobras
#5900.0112605.19.9). Authors wish to thank Petrobras for the funding of this project and the Brazilian National
Petroleum Agency (Agência Nacional do Petróleo, ANP) for providing the regulatory framework under which this
funding takes place. Alexandre Simos also thanks the Brazilian National Council for Scientific and Technological
Development (CNPq) for his research grant (#06342/2020-0).

References
[1] Feller F 2017 Offshore Mediterranean Conference and Exhibition
[2] WIND-ENERGY-DENMARK 2019 Decarbonisation of O&G production by cost effective floating wind technolo-
gies https://my.eventbuizz.com/assets/editorImages/1570623775-Innovation_Zone_1400-1515_-_
Sarah_Thomas.pdf online (accessed December 2, 2021)
[3] Tenggren E, Olsen L E and Johansen D Aand Bracchi T 2020 Journal of Physics: Conference Series 1669 012026
[4] Shadman M, Estefen S F, Nunes K, Maali Amiri M, Tavares L F, Rangel P and Assad L P 2020 International Conference
on Offshore Mechanics and Arctic Engineering (American Society of Mechanical Engineers)
[5] Technomar 2021 Edtools https://www.technomar.com.br/engenharia-naval online (accessed December 3, 2021)
[6] Pesce C P, Amaral G A and Franzini G R 2018 International Conference on Offshore Mechanics and Arctic Engineering
(American Society of Mechanical Engineers)
[7] Amaral G A, Pesce C P and Franzini G R 2022 Marine Structures 84 103189
[8] DNV 2014 Design of offshore wind turbine structures, DNV-OS-J101
 [9] Bak C, Zahle F, Bitsche R, Taeseong K, Yde A, Henriksen L C, Natarajan A and Hansen M H 2013 Description of the
DTU 10 MW Reference Wind Turbine Tech. Rep. Report-I-0092 DTU Wind Energy Roskilde, Denmark
[10] Tracy C C H 2007 Parametric design of floating wind turbines Ph.D. thesis Massachusetts Institute of Technology
[11] Birk L 2009 Journal of Offshore Mechanics and Arctic Engineering 131
[12] Gilloteaux J C and Bozonnet P 2014 The Twenty-fourth International Ocean and Polar Engineering Conference
[13] Uzunoglu E and Soares C G 2019 Ocean Engineering 171 78–92
[14] Aubault A, Cermelli C and Roddier D 2007 International Conference on Offshore Mechanics and Arctic Engineering
[15] Bachynski E E and Moan T 2012 Marine Structures 29 89–114
[16] Hall M, Buckham B and Crawford C 2013 2013 MTS/IEEE OCEANS-Bergen (IEEE)
[17] Karimi M, Hall M, Buckham B and Crawford C 2017 Journal of Ocean Engineering and Marine Energy 3 69–87
[18] Cedeño W 1995 The multi-niche crowding genetic algorithm: analysis and applications Ph.D. thesis University of
California
[19] Hegseth J M, Bachynski E E and Martins J R 2020 Marine Structures 72 102771
[20] Mühlenbein H and Schlierkamp-Voosen D 1993 EVOLUTIONARY COMPUTATION 1 25–49
[21] Cedeño W and Vemuri V R 1999 Theoretical Computer Science 229 177–197
[22] Bjerkseter C and Ågotnes A 2013 Levelised costs of energy for offshore floating wind turbine concepts Master’s thesis
Norwegian University of Life Sciences
[23] Garcia-Teruel A, Rinaldi G, Thies P R, Johanning L and Jeffrey H 2021 Applied Energy 307 118067
[24] Nejad A R and Torsvik J 2021 Forschung im Ingenieurwesen 85 335–343
[25] Allen C, Viscelli A, Dagher H, Goupee A, Gaertner E, Abbas N, Hall M and Barter G 2020 Definition of the umaine
volturnus-s reference platform developed for the iea wind 15-megawatt offshore reference wind turbine Tech. rep.
National Renewable Energy Lab.(NREL), Golden, CO (United States)
[26] DNV 2021 Floating wind turbine structures, DNV-ST-0119
[27] Jonkman J, Butterfield S, Musial W and Scott G 2009 Definition of a 5MW reference wind turbine for offshore system
development Tech. Rep. NREL/TP-500-38060 National Renewable Energy Laboratory (NREL) CO, USA
[28] Hersbach H, Bell B, Berrisford P, Hirahara S, Horányi A, Muñoz-Sabater J, Nicolas J, Peubey C, Radu R, Schepers D
et al. 2020 Quarterly Journal of the Royal Meteorological Society 146 1999–2049
[29] TPN 2020 High-performance computing cluster https://tpn.usp.br/simulador/Facilities.html online
(accessed December 3, 2021)

Appendix A. Weighting Function

The amount of literature regarding the weighting function adopted in FOWT optimization problems is scarce.
For example, in [16], the authors adopted a weighting sum method to combine the RMS of the accelerations at
the nacelle of the turbine for all the N sea conditions assessed. However, no further details are given about the
methodology followed to determine the weight values.
It has been shown in the bulk of the paper that the horizontal accelerations at the nacelle stand as a wind
performance metric, since their minimization results in the minimization of the most critical loads and may increase
the total energy production over the time. In this work, the weighting sum method has been adopted to combine
the maximum acceleration horizontal values at the nacelle estimated for each one of the N sea states evaluated.
The rationale followed to determine the weight values presumes that the performance of the wind turbine is more
sensitive to the largest accelerations than the smallest ones. Thus, the model will result in the overweighting of
large accelerations and underweighting the smallest accelerations. Nonetheless, it is important to bear in mind
that the highest values of acceleration are featured by few occurrences, while the smallest accelerations are the
most common ones. The proposed approach shall be able to dynamically discriminate the tails of the acceleration
distribution and hence avoid concepts with optimal behaviour under these conditions. Furthermore, the values of
wi = 1. Therefore, each weight value can be
PN
the weights are restricted such that their sum equals to one, i.e. i=1
tuned by means of an analogue probability density function (P), i.e.:
N
X
wi = 1 → wi = P(ai )∆ ai (A.1)
i=1

The function P is, therefore, estimated as the normalized product of the empirical probability function of
occurrence of the accelerations (which is different for each individual), p(ai ), and a prior distribution that will
discriminate the lowest values of the acceleration, L(ai ), see eq. A.2. In other words, Pi stands for the probability
that an acceleration ai is relevant for the performance of a FOWT given its occurrence in the N sea conditions
evaluated during the optimization.
 L(ai )p(ai )
Pi = PN (A.2)
i=1 L(ai )p(ai )

In order to better understand the result of the weigh values, Fig. A1 provides an illustration of the results
obtained for an arbitrary FOWT concept from the optimization. In this case the results have been normalized
by the respective maximum values, which allows the display of the three distribution in the same figure. The
acceleration distribution is provided by the red line and shows that the maximum values obtained for the time
series assessed are small if compared with the reference threshold value, equal to 0.2g. In our study, this is a
consistent behaviour among the concepts generated during the optimization and the largest estimated acceleration
values are associated to sea states with returns periods up to 10 years. The black points represent the weigh values
for each sea conditions assessed. These weight values show that the approach adopted allows to discriminate the
lowest and highest values of the accelerations evaluated, resulting in FOWT concepts fitted for the most significant
sea states.

0.9

p(a)
0.8
L(a)
p/max(p), L/max(L), w/max(w)

w(a)
0.7

0.6

0.5

0.4

0.3

0.2

0.1

0
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07
Acc./0.2g [-]

Figure A1. Illustration of the distributions obtained for an arbitrary FOWT concept.