Renewable Energy 71 (2014) 729e741

Contents lists available at ScienceDirect

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

Reliability-based design optimization of monopile transition piece for

offshore wind turbine system
Yeon-Seung Lee a, Byung-Lyul Choi b, Ji Hyun Lee c, Soo Young Kim d, Soonhung Han a, *
a
Division of Ocean Systems Engineering, Korea Advanced Institute of Science and Technology, 291 Daehak-ro, Yuseong-gu,
Daejeon 305-701, Republic of Korea
b
PIDOTECH Inc., #312, HIT, Hanyang University, 17 Haengdang-Dong, Seongdong-Gu, Seoul 133-791, Republic of Korea
c
Samwon Millennia Inc., #724, Migum Park Building, 150 Gumgok-dong, Bundang-gu, Seongnam-City, Gyeonggi-do 463-806, Republic of Korea
d
Department of Naval Architecture and Ocean Engineering, Pusan National University, Pusan 609-735, Republic of Korea

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

Article history: This paper presents a reliability-based design optimization (RBDO) method for a monopile transition
Received 13 November 2013 piece in an offshore wind turbine system. Two design approaches are investigated for the cost-effective
Accepted 12 June 2014 and reliable design of the monopile transition piece: deterministic optimization (DO) and RBDO. First,
Available online 17 July 2014
dynamic response analysis of a reference offshore wind turbine is conducted to estimate design loads
considering site conditions off the southwest coast of Korea. Second, DO for minimizing the mass of a
Keywords:
conical monopile connection is carried out, including an assessment of the reliability of the DO design.
Offshore wind turbine
Next, RBDO is performed to achieve a design with the desired reliability while concurrently minimizing
Monopile support structure
Grout
the mass of the monopile transition piece. The present study shows that the structural design of the
Transition piece monopile connection is mostly dictated by the fatigue limit state and that DO does not guarantee
Deterministic optimization (DO) structural reliability even though the design satisfies all limit state function conditions. The proposed
Reliability-based design optimization RBDO process is shown to speed up the design cycle and enhance the reliability of the grouted
(RBDO) connection for offshore wind turbine support structures.
© 2014 Elsevier Ltd. All rights reserved.

1. Introduction uncertainties in loads and uncertainties in strength and stiffness of

the structure. Uncertainties in loads originate owing to (1) the
The monopile support structure of offshore wind turbines needs natural randomness of environmental factors such as wind, waves,
to have high reliability for the turbine systems to withstand the and currents; (2) statistical uncertainty in estimating and modeling
complex extreme nonlinear loads of the harsh environmental the environmental parameters; (3) physical model uncertainties
conditions in the ocean, which cause fatigue, structural failure, and, due to the aero-hydro-servo flexible structural simulation model of
occasionally, even critical accidents. Furthermore, its the integrated turbine; and (4) the analytical and numerical
manufacturing, installation, maintenance, and operation costs need structural analysis model. Uncertainties in strength and stiffness
to be feasible. Because a turbine system's capital cost is determined originate owing to the natural randomness of the material strength
by its target reliability, it is important to achieve a tradeoff among and stiffness and model uncertainties due to the resistance model.
the failure consequences, material consumption, and failure prob- Studies have extensively focused on the effect of extreme waves
ability at the design stage. such as nonlinear and breaking waves on the estimation of support
Many studies have investigated the feasibility of different fixed structure loads by the physical modeling of complex phenomena
support structures such as monopile, tripod, jacket, triple, and [6e9] and their uncertainty propagation to extreme wave loads. For
gravity-based foundation [1e4]. However, support structure design accurate and effective load estimation, stochastic procedures based
in consideration of system reliability has been limited to very on the first-order reliability method (FORM) and Monte Carlo
simple ways of implementing safety or load factors [5]. There are simulation (MCS) have been proposed [10e12]. The dynamic re-
two types of uncertainties in designing a support structure: sponses of an integrated offshore wind turbine and design loads are
obtained by time-domain simulations of a coupled aero-serve-
hydro-elastic model. A dynamic simulation model contains high
* Corresponding author. Tel.: þ82 (0)42 350 3040; fax: þ82 (0)42 350 1510.
levels of uncertainty. NREL's (National Renewable Energy Labora-
E-mail address: [email protected] (S. Han). tory) OC4 project, Europe Community's Upwind project, and others

http://dx.doi.org/10.1016/j.renene.2014.06.017
0960-1481/© 2014 Elsevier Ltd. All rights reserved.
730 Y.-S. Lee et al. / Renewable Energy 71 (2014) 729e741

have undertaken long-term code validations [13e15]. These codes Table 1

have been validated both in terms of the accuracy of the dynamic NREL 5 MW baseline wind turbine [2].

simulation and the predictability of a wide range of design spaces Rating 5 MW

for different support structures [1,5,16]. Currently, although the Rotor orientation Upwind
simulation accuracy has been validated for the design of an offshore Control Variable speed, collective pitch
Drivetrain High speed, multiple stage, gearbox
wind turbine, site-specific conditions need to be considered care- Rotor diameter 126 m
fully [17]. Another uncertainty is caused by the simplified analytical Hub height 90 m
models and characteristic parameters in design standards for the Cut-in, rated, cut-out wind speed 3 m/s, 11.4 m/s, 25 m/s
support structure of an offshore wind turbine [19e21]. It is origi- Cut-in, rated rotor speed 6.9 rpm, 12.1 rpm
Rated tip speed 80 m/s
nally based on those of offshore oil and gas production platforms
Overhang, shaft tilt, precone 5 m, 5 , 2.5
that are mainly subjected to axial loads instead of bending mo-
ments [22]. In contrast, the large-scale wind blades of offshore
wind turbines cause bending moments on the support structure
owing to the complicated aerodynamic loads and the nonlinear transition piece connecting the monopile and the tower. The ad-
dynamic response and fatigue of the structure [23,24]. vantages of a monopile support structure, when compared with the
Despite such diverse uncertainties, support structures are gener- other bottom fixed support structures, include minimal seabed
ally designed based on a deterministic limit state design for the esti- preparation requirements, most competitive manufacturing costs
mated design loads and partial safety factors [25e27]. The structural owing to the simple structure, and the most experienced support
systems' performance is optimized in consideration of their sizes, structure with offshore wind turbines. The disadvantages include
shapes, and material characteristics. Because a deterministic design, structure flexibility at large water depths, time-consuming instal-
especially an optimum one, is likely to meet the boundary of the lation owing to the grout setting time, and manufacturing con-
feasible region that consists of design constraints, it is usually associ- straints for large diameters and thickness that makes it very
ated with a high failure probability [28e31]. A reliability-based design difficult to go beyond 30 m water depth.
optimization (RBDO) methodology that simultaneously achieves the The state-of-the-art in monopile connection is to implement a
target reliability and minimum cost is developed for the limit state conical pile with or without shear keys and transition piece con-
design of the monopile connection of a 5 MW offshore wind turbine. nected by high-strength grout that can withstand the large dy-
Finally, we verified that RBDO of large-scale offshore wind turbines is namic bending moments [20]. The grouted connection serves to
clearly necessary for achieving a reliable and cost-effective design. This install the transition piece on the pile and to fix the offshore wind
study proposes an optimum design methodology to ensure the reli- turbine at the right position safely. Fig. 1 shows the initial reference
ability of offshore wind turbines during the design stage. design implemented in this study. Table 2 lists the grouted con-
The remainder of this paper is organized as follows. Section 2 nections' geometric design parameters and material properties.
introduces the characteristics of the 5 MW monopile reference This design achieves the structural safety requirements based on
turbine model adopted in this study. In Section 3, the design loads Det Norske Veritas (DNV) design guidelines [20].
of the monopile connection of the reference turbine are analyzed
for the support structure considering site-specific conditions off the 3. Design load analysis
southwest coast of the Republic of Korea. Section 4 describes the
analytical and numerical structural analysis of the monopile The structural design of offshore wind turbines is based on the
connection for the reference turbine based on the design loads verification of the structural integrity of each component under
predicted in Section 3. Section 5 presents the deterministic design external conditions such as wind, waves, and currents as well as
problem with a partial safety factor based on the ultimate and fa- other environmental and turbine conditions. The loads induced by
tigue limit state designs for a monopile connection by describing external conditions are divided into three categories: initial and
the limit state functions; furthermore, sensitivity analysis and DO gravitational loads, aerodynamic loads, and hydrodynamic loads.
design of the monopile connection are conducted. In Section 6, the Design loads are obtained by the time domain simulation of
reliability of the DO design is evaluated by reliability analysis. In coupled aero-servo-hydrodynamic-elastic models. The ultimate
Section 7, RBDO is undertaken for the monopile connection design
to achieve the target reliability with minimum mass. The cost-
effectiveness and reliability of the initial, optimum, and RBDO de-
signs are compared. GH bladed [32], a commercial code, is used to
calculate the ultimate design loads. PIAnO [33], developed by the
author, is used for stochastic design and analysis.

2. Characteristics of reference turbine model

2.1. NREL 5 MW baseline offshore wind turbine

NREL's offshore 5 MW baseline wind turbine is used as a refer-

ence in this study. This turbine is developed based on Senvion
5 MW prototype wind turbines and is considered representative of
typical utility-scale land- and sea-based multi-megawatt turbines.
Table 1 lists the parameters of this turbine.

2.2. Monopile support structure with grouted transition piece

The monopile support structure of an offshore wind turbine

consists of two parts: the monopile penetrating the seabed and the Fig. 1. Configuration of monopile connection of reference design.
 Y.-S. Lee et al. / Renewable Energy 71 (2014) 729e741 731

Table 2 are calculated as shown in Table 4. Most severe load cases are found
Geometry parameters and initial reference design. as DLC 6.2b as shown in Table 5. The ultimate design load is pre-
Geometry Symbol Value Unit dicted from the load case of DLC 6.2b, which states that the turbine
Size Cone angle a 1.00  is parked in idling or fault conditions and is subjected to extreme
Length of grouted section Lg 7.50 m wind and wave conditions.
Pile outer radius Rp 2.50 m While the ultimate design load is based on the extreme single
Thickness of transition piece ttp 0.17 m load case which is selected as extreme among the time domain
Wall thickness of pile tp 0.13 m
simulated load cases, fatigue limit state (FLS) check has to be based
Thickness of grout tg 0.16 m
Material Grout to steel interface friction coefficient m 0.4 e on the entire fatigue loads during the design life. In the present
Young's modulus for grout Eg 53.0 GPa optimization process, however, fatigue design load case DLC 1.2 is
Young's modulus for steel E 210.0 GPa utilized by assuming that it could represent the entire fatigue states
because DLC 1.2 generally covers around 80% of the total fatigue
loads. DLC 1.2 states that the turbine is operating for power pro-
strength of each load-carrying component under design loads duction and is subjected to normal turbulence model and irregular
should be verified to determine its structural integrity. sea state [25].
The uncertainties of external conditions affect the prediction
accuracy of design loads and, in turn, the reliability of the wind 4. Structural analysis
turbine system. Statistical data of the environmental conditions
and modeling of each load and load case were verified and vali- A monopile continuously resists the axial loads and shear forces
dated for a specific site to determine the design loads of the turbine caused by a rotor and nacelle assembly and tower and the bending
system [17]. In particular, individual data were compared with the moment induced by wind and wave load during the design life. The
measurements and the effects on loads were estimated accordingly. axial capacity depends on the surface roughness and irregularity in
In this study, because the estimation of the reliability of the the tubular sections of the grout connection [19e21]. Alternating
grouted connection is of interest, we instead focus on the structural bending moment causes grout wear on the sliding surfaces be-
reliability propagated by the uncertainty of design loads. The ulti- tween the pile and the transition piece, which reduces the bending
mate limit design loads of a reference 5 MW monopile-type moment capacity. The fatigue capacity of a grouted connection in
offshore wind turbine are calculated for the Yellow Sea off the air and seawater environments should be carefully assessed for
southwest coast of the Republic of Korea [34]. The monopile is fixed guaranteeing the design life.
to the seabed at a depth of 20 m. Load cases based on IEC61400-3
[25] are analyzed by using GH Bladed, a commercial code. There
are three representative models which can describe the monopile 4.1. Analytical capacities of grouted connections
with flexible foundation; apparent fixity model, coupled springs
model, distributed springs model. Since different foundation Grouted connections in wind turbine support structures are
models do not mainly affect the design loads for the grouted designed for the ultimate limit state and the fatigue limit state load
connection, the monopile with flexible foundation is modeled by an combinations. The ultimate limit state design is categorized by axial
apparent fixity model because of simplicity [16]. and torsional capacities and bending moment and shear capacities.
The fatigue limit state design is categorized by the fatigue capacity
of the grouted connection mainly subjected to bending moment.
3.1. Site-specific environmental condition

The reference area is located at 35 300 15.8600 latitude and 126 4.1.1. Ultimate limit state design capacities
040 08.9900 longitude off the southwest coast of Korea; this site is 4.1.1.1. Axial capacity. The characteristic ultimate capacity of axially
considered to have excellent economic feasibility. Kim et al. eval- loaded grouted tubular connections can be analytically calculated
uated the feasibility of offshore wind farm development in the [19,20]. A good design of conical grouted connections subjected to
South Korean Peninsula and selected appropriate sites with depths axial loading is believed to limit settlements due to axial loading. A
below 20 m and over 10 km from the coast in the Yellow Sea around design with a small cone angle could also denote a grouted
[35]. The area has been selected for installing around 2.5 GW connection with the defined minimum fabrication tolerances.
offshore wind power generators by 2030. Table 3 lists wind turbine However, a conical grouted connection is considered to be a more
class II with the site-specific marine data [25]. practical notation that explains the connection's physical behavior
with respect to axial capacity. A cone angle of 1e3 can be rec-
3.2. Design loads ommended. It is assumed that the grout material will not transfer
significant tensile stress in the hoop direction and that the grout
The dynamic responses of the NREL 5 MW reference turbine mainly serves to set up a pressure between the transition piece and
with the reference monopile support structure are analyzed and the pile if the former tries to slide downward relative to the latter.
the design loads for the grouted connection of the reference turbine The shear stress in an axially and torsionally loaded conical
connection satisfies

Table 3
Wind turbine class and marine conditions for reference site [25,34].
Table 4
Wind conditions Marine conditions Ultimate design loads of initial reference design.

Wind turbine class II Site-specific conditions Loads Symbol Value Unit

Vref [m/s] 42.5 Wave Hs [m] 3.33 Maximum Fz Fz 11,173.0 kN
Vave [m/s] 8.5 Ts [s] 8.46 Maximum torque MT 11,783.0 kN
Turbulence intensity (B) I15 () 0.16 Current [m/s] 1.646 Maximum Mxy Mmax 183,634.0 kN m
a () 3 Ave. water depth [m] 20 Minimum Mxy Mmin 429.0 kN m
732 Y.-S. Lee et al. / Renewable Energy 71 (2014) 729e741

Table 5
Design load case of ultimate design load for support structure [25].

Design situation Parked (standing still or idling)

Design load case (DLC) DLC 6.2b
Wind condition Extreme wind speed model (EWM),
V (Zhub) ¼ Ve50
Waves Reduced wave height (RWH), H ¼ Hr50
Wind and wave directionality Misaligned (MIS), multi-directional (MUL)
Sea currents Extreme current model (ECM)
Water level Extreme water level range (EWLR)
Other Conditions Loss of electrical network
Type of analysis Ultimate strength (U)
Partial safety factor Abnormal (A)

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  
t
t2sa þ t2st  k ; tk ¼ min tkf ; tkg (1)
gm

where tsa and tst are shear stresses in an axially and torsionally
Fig. 2. Loads and reaction force of grouted connection [19].
loaded connection, respectively. tk is the characteristic shear
strength of the connection, and it is defined as the minimum of the
characteristic interface shear strength due to surface irregularities
and friction, tkf , and the capacity of the grout matrix, tkg . gm is a 3pMtot
material factor for the ultimate limit state, and it is set to 3.0. The PNominal ¼ (6)
Rp L2g ðp þ 3mÞ þ 3p mR2p Lg
interface shear strength due to friction tkf is defined as
    The maximum horizontal opening at the top of the monopile
m$E d Rp Rtp tg E and the vertical deformation between the steel and the grout are
tkf ¼ $ K¼ þ þ (2)
K Rp tp ttp Eg Rp derived as

where m is the coefficient of friction of the grout to steel interface; E   2Rp

and Eg are Young's elastic modulus of steel and grout, respectively; dH ¼ 3 dp þ dtp ; dV ¼ dH (7)
Lg
K is the stiffness factor; and delta is the height of surface irregu-
larities. These equations are only valid in the following ranges: where

Rp Rtp
5  30 9  70 (3) pR2p pR2tp
tp ttp dp ¼ dtp ¼
Etp Ettp
The capacity of the grout matrix, tkg , may be taken as
The local contact pressure accounting for the discontinuity at
0 1 the ends of the grout can be presented in terms of a stress increase
0:7 B
2L C by a stress concentration factor such that the local pressure stress
tkg ¼ k$fck :@1  e =Rp A (4) becomes related to the nominal contact pressure as

pLocal ¼ SCF PNominal (8)

where k is an early age cycling reduction factor. fck is a characteristic
compressive cube strength of the grout and fcck, that for cylinder where SCF is the stress concentration factor. On the top of the pile,
strength. The cube strength can be taken as fck ¼ fcck þ 11MPa [19]. SCF is a function of radius and thickness of the transition piece and
defined by equation (9). On the lowest end of the transition piece,
SCF is a function of the radius and thickness of the pile.
4.1.1.2. Bending moment capacity. The bending moment is trans-
ferred from the transition piece to the pile through horizontal 3=
2
RTP
contact forces as shown in Fig. 2. Vertical friction forces due to SCF ¼ 1 þ 0:025 (9)
tTP
contact pressure also contribute to the moment capacity of the
grouted connection. The bending moment on the grouted con- Alternatively this local design stress can be derived directly from
nections in an offshore wind turbine is relatively much larger than linear finite element analyses (FEA). Based on the local design
the vertical static loading above the transition piece. When stress, the maximum principal stress ðs1 Þ and minimum principal
comparing the axial stress in the transition piece at the grouted stress ðs3 Þ in the grout are derived as
section, the maximum axial stress due to bending moments is
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
observed to be 30 times that of the axial stress due to self-weight pLocal
s1 ¼ 1þ 1 þ 4m2
for the wind turbine structures with the largest bending mo- 2
ments in the ultimate limit state [19,20]. The total moment of
transition piece, Mtot, is given by the summation of the moment qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
pLocal
actions due to contact pressure Mp, horizontal friction force Mmh, s3 ¼ 1 1 þ 4m2 (10)
2
and vertical friction force Mmv.

Mtot ¼ Mp þ Mmh þ Mmv (5) where m is a local friction coefficient. In the ultimate limit state, the
grout capacity to avoid progressive grout crushing is assessed by
An estimate of the maximum nominal pressure due to the total checking whether the stresses in the grout, expressed by the Tresca
moment is then derived from equation (5). stress ðfs Þ, satisfies the following requirement:
 Y.-S. Lee et al. / Renewable Energy 71 (2014) 729e741 733

fcck conditioned by the assumed simplifications. Through the dis-

fs  (11) cretization process, the FE method sets up an algebraic system of
gm
equations for unknown nodal values that merely approximate the
where Tresca stress fs ¼ s1  s3 and gm is the material factor for continuous solution. The FEA results are also compared with those
the fatigue limit state and is set to 3.0. from the analytical solution.
FEA of the grouted connection is modeled with double contact
4.1.2. Fatigue capacity interfaces between the grout and the steel tubes (both sides of the
Because offshore wind turbines of a tower and a monopile with grout member). FEA is conducted with contact friction on surfaces
large diameter are very slender and flexible, they are continuously without shear keys. The effect of slip is included in the contact
susceptible to wind and wave loads during service life. This makes it formulation when friction is present. The mesh size on the contact
difficult to avoid compressive to tensile stresses during a loading cycle surfaces accounts for nonlinear stress singularities at the surface
in a grouted connection of a monopile subjected to a dynamic bending edges. The mesh size therefore ensures that contact occurs on a
moment. The fatigue loads of the grouted transition piece are assessed minimum of 3 elements in the slip direction. Further, the element
only by bending moment and shear loading for the fatigue limit state, edge aspect ratio on the contact surfaces does not exceed 1:5. The
excluding axial loading and torque from the analysis. The accumulated grout elements have linear 8-node solid element with 3 translation
damage for the long-term stress cycle history is calculated using the degrees of freedom. The grout member has two first-order ele-
PalmgreneMiner summation, and it should not exceed 1.0. The ca- ments through its thickness. A conservative friction coefficient of
pacity is assumed to be adequate when the calculated design life for 0.4 is applied between steel and grout according to DNV-OS-J101
the largest acting amplitude corresponds to at least 2.0  106 cycles if [19].
the fatigue loading is subjected to wind and waves according to DNV The extreme design conditions for the FE model based on the
offshore standard 2012 [20]. The number of cycles to failure N can then NREL 5 MW offshore wind turbine are obtained from the load
be calculated according to the following SeN curve: analysis of GH-bladed, and FEA of the boundary condition, which
( does not affect the analysis results far away from the hot spot
log10 NI for log10 NI < X contact area, is fixed at a seabed depth of 20 m. Fig. 3 shows the FE
log10 N ¼
log10 NI ,ð1 þ 0:2ðlog10 NI  XÞÞ for log10 NI > X model and 3D CAD model. Table 6 lists basic information about
typical steel and grout materials. The grout material should have
where;
high strength and stiffness, low shrinkage, high early-age strength,
log10 NI ¼ C1 ð1  Smax =C5 Þð1  Smin =C5 Þ; high inner cohesion, i.e., no mixing with seawater, and good fatigue
C1 performance for offshore applications.
X¼ The axial and moment capacity obtained by the analytical DNV
ð1  Smin =C5 Þ þ 0:1,C1
design equations are compared with the FEM results. The DNV
(12) design equations give the calculation results for deformation and
shear stresses, and therefore, the evaluation for the analysis focuses
where the factor C1 is recommended as 8.0 for structures in water
on the grout instead of the transition piece and pile. This analysis
for those stress-blocks having stress variation in the compression-
does not include shear keys.
tension range. C5 is the fatigue strength parameter and is set to 0.8
for grout. Smax is the maximum relative stress and is defined as
Smax ¼ Smax,f/fcck,f. Smax,f is the maximum compressive stress in a 4.2.2. Axial capacity
cycle and fcck,f is the design fatigue strength derived as fcck,f ¼ fcck/ When axial forces and torsional moments are applied on the
gm. Smin is the minimum relative stress defined as Smin ¼ Smin,f/fcck,f. inclined surfaces of the cone, the axial capacity must satisfy the
gm is the material factor for the fatigue limit state and is set to 2.6. interface shear strength against the combined shear stresses by the
An endurance limit is defined for stress ranges such as [20] axial force and torsional moment. Therefore, the FE results are
compared with the DNV equations of axial capacity satisfying the
DS < 0:30  0:375Smin ; where DS ¼ Smax  Smin : (13) service life reliability.
It is an endurance limit check for the largest stress range
DS ¼ Smax  Smin ðMPaÞ derived by the time domain simulation of
the fatigue load DLC 1.2. If the equation (13) is satisfied, an infinite
number of cycles to failure applies and overrides the value of N
resulting from the equation (12).

4.2. Finite element analysis of support structure

This section targets to support and verify the analytical results

obtained in the previous section. The complexity of the structural
behavior of the grouted connection can be summarized as three
aspects: (1) Contact behavior of interface between grout and steel,
(2) Nonlinear material behavior of grout, (3) Choice of element
types and discretization. This section aims to analyze the grouted
connection to confirm the DNV results obtained in the previous
section. The FE model to be analyzed is constructed in accordance
with DNV guideline. NX/Nastran is used for this analysis.

4.2.1. Analysis model and boundary conditions

The numerical methods dealing with FE models are generally
presumed to be more accurate than the analytical procedures, Fig. 3. FE model and 3D CAD of reference monopile connection.
734 Y.-S. Lee et al. / Renewable Energy 71 (2014) 729e741

Table 6 that are mostly converged on the boundaries of the feasible region
Material properties [19]. that satisfies the failure modes stated by limit state functions.
Material Steel Grout Unit Reliability analysis is required to ensure system reliability by
Young's modulus 210.0 53.0 GPa
defining the probability distribution of each variable and its prop-
Poisson's ratio 0.3 0.19 e agation into the limit state functions. RBDO results in an optimum
Steel-grout friction coefficient 0.4 e design with a required reliability. In Sections 5 and 6, these pro-
Density 7850 2512 kg/m3 cedures and results are discussed.
Compressive strength e 80 MPa
Tensile fracture strength e 6.0 MPa
Tensile fracture energy e 500 N/M 5.1. Optimization problem formulation
Material safety factor (ULS) 1.15 3.0

The failure modes of a monopile connection consist of exceed-

The structural analysis shown in Fig. 4 represents the global ances of the axial and torque capacity and the bending moment and
shear stress of the entire analysis model, grout shear stress, and shear capacity. The two failure modes are described by the ultimate
horizontal displacement result for the grout and the pile. Table 7 and the fatigue limit state. The optimization problem to minimize
shows shear stresses and displacements subjected to axial loads mass has been formulated to satisfy the limit state functions for
by DNV and FE analysis, where tsa and tst are shear stresses in an axial capacity and bending moment capacity as design constraints.
axially and torsionally loaded connection, dH, the maximum hori- The limit state functions for the limit state design of a monopile
zontal displacement; and dV, the maximum vertical displacement. grout connection are as follows:
The safety factors are calculated by dividing the total shear stress by
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
giL  gi ðxÞ  giU ; i ¼ 1; 2
the shear strength such as t2sa þ t2st =ðtk =gm Þ. This is defined in
gi ðxÞ  0; i ¼ 3; :::; 8
equation (1) of Section 4.1. The shear strength of the initial design is
0.289. Based on the safety factors by DNV and FE analysis, it can be where, for ultimate limit states:
concluded that the initial design satisfies the characteristic shear
strength and DNV and FE analysis are both similar prediction Rp
g1 ¼ ; 5  g1  30
regarding the shear stresses and displacement subjected to axial tp
loads.
RTP
g2 ¼ ; 9  g2  70
4.2.3. Bending moment capacity tTP
The bending moment capacity is not considered for conven-
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
tional fixed offshore oil and gas platforms, but it is more important t
g3 ¼ t2sa þ t2st  k < 0
for offshore wind turbines because of their slenderness. The shear gm
stresses in the transition piece and monopile due to moment
loading on the grouted connections are usually much larger in a stest
g4 ¼ s1  <0
wind turbine structure than those from the vertical permanent gm
loading from the structure above the connection. The moment
loading is transferred from the transition piece to the monopile stest
g5 ¼ s3  <0
through horizontal contact forces. Vertical friction forces also occur gm
owing to the contact pressure that contributes to the moment ca-
pacity of the grouted connection [19]. Fig. 5 shows the overall shear fcck
stress of the entire FE model, maximum grout Tresca stress fs , g6 ¼ fs  <0
gm
maximum horizontal opening dH at the top of the monopile be-
tween the steel and the grout. Table 8 shows a comparison of the FE
for fatiguelimit states
 :
results and the DNV results. Although the results are both satisfies
the DNV guidelines, the difference between DNV and FE analysis in g7 ¼ log10 2  106  log10 N < 0
the maximum Tresca stress and maximum displacements sub- g8 ¼ DS  ð0:30  0:375Smin Þ < 0
jected to bending moments are relative bigger than those subjected
The limit state functions gi are described in equations (1)e(13)
to axial loads. DNV analysis is less conservative for the shear stress
in Section 4.1. g1eg6 represent the ultimate limit state functions
subjected to bending moment while it is more conservative or
and g7 and g8, the fatigue limit state functions. g1 and g2 limit the
similar for shear stresses in an axially and torsionally loaded
valid application range of linear axial capacities. g3 is the axial
connection.
safety function in the ultimate limit state design; it means that the
shear stresses in an axially and torsionally loaded connection has to
5. Deterministic optimum design be less than the axial capacity. g4eg6 are related to the bending
moment capacity. g4 and g5 represent the maximum and minimum
Structural optimization in the early development stage offers principal stress in a local position in the grouted connection. These
new technical potential and cost effectiveness in the design process contact pressures have to be less than the contact pressure capacity.
considering environmental load conditions [36]. The development g6 represents the bending moment safety and is derived such that
process becomes faster and more efficient by using optimization. the Tresca stress is less than the compressive cylinder capacity.
This results in structures that are lighter, stronger, and more du- According to the new edition of DNV J101, 2013 [20], the validity
rable, providing a competitive advantage for developing the tran- range of stress concentration factor (SCF) to determine the
sition piece. A support structure is exposed to severe levels of maximum and minimum principal stress in g4 and g5 is bounded in
uncertainties that are potential causes of failure of the stochastic terms of ttp and Rtp . This is shown in Fig. 6. The solid lines are SCFs
system. This is mostly due to uncertainties in environmental con- of the maximum (2750 mm) and minimum (2250 mm) radius of TP
ditions and the characteristics of the structural optimum design as a function of TP thickness, respectively. The point represents SCF
 Y.-S. Lee et al. / Renewable Energy 71 (2014) 729e741 735

Fig. 4. Shear stress and displacement subjected to axial loads by FEM.

of the initial design which is out of validity range. However, author generated in the entire design range by orthogonal array method
would like to use the current design formulation and initial design and sensitivities are analyzed synthetically by using ANOVA
due to the following reasons: SCF is insensitive to thickness in the (ANalysis Of VAriance) and regression analysis. One of its advan-
range over 100 mm and SCF of design is similar to those of the valid tages is the global design range search.
SCF for the 150 mm thickness of TP. According to DNV report No. In Table 10, sensitivities of design parameters to each limit state
95-3204 [18], the uncertainty of SCF for grout is defined as 20% of function are expressed in terms of % sensitivities respectively, by
coefficient of variation (COV). assuming the summation of % sensitivities of all investigate design
The fatigue limit states are expressed in terms of the limit state parameters to each individual limit state function is 100%.
function, g7 and g8. The constraints are very conservative e being The six design variables a; Lg ; Rp ; tp ; ttp ; tg are of importance for
based on the largest stress range, and requiring both a large allowed the design of monopile connection. The design loads are relatively
number of these cycles (constraint g7) and a range below an less sensitive to the limit state function in comparison to the ge-
endurance limit (constraint g8). For the practical design point of ometry design variables. In current limit state design, although the
view, these constraints could be alleviated in some cases. Table 9 partial safety factor 1.25 is applied to the design loads and the
shows the limit state function values of the initial reference material factors 3.0 for ULS and 2.6 for FLS, the initial design is safe
design of the monopile connection. enough against axial and bending moment loads.
The thickness of the pile tp and the transition piece ttp are
5.2. Sensitivity analysis related mostly to limit state functions, whereas that of the grout tg
is less sensitive. This means that the thickness of the transition
Sensitivity analysis helps to understand the characteristics of piece and that of the pile is sensitive design variables from the
the design system and develop a more effective approach to the viewpoint of both the ultimate limit state and the fatigue limit
optimization formulation for the problems. In this study, the sen- state. The inclination of the conical shape a is mostly related to the
sitivities of the limit state functions are investigated for two groups fatigue limit state. The outer radius of the pile has the greatest
of parameters: geometry parameters of the monopile connection influence on the total mass.
xgeometry ¼ a; Lg ; Rp ; tp ; ttp ; tg ; m; Eg ; E and predicted ultimate
design loadsxload ¼ fFz; MT ; Mmax ; Mmin g. The sensitivities of design 5.3. Optimum design based on DO
loads are additionally investigated because design loads are un-
certain owing to their highly varying characteristics. The sensitivity analysis indicates that the six design variables
Sensitivity analysis has been performed numerically by means are important for structure optimization and stability. The design
of design of experiment (DOE) and ANOVA. Candidate designs are formulations are aimed at minimizing the mass of the grouted
connection using the governing design variables. The limit state
Table 7
functions described in Table 9 are applied as constraints. The
Comparison of DNV and FEM results subjected to axial loads.
optimization problem is formulated as follows:
Axial DNV FEM

Grout tsa (MPa) 0.096 0.083 Find xgeometry ¼ a; Lg ; Rp ; tp ; ttp ; tg

tst (MPa) 0.045 0.039 minimize Mtotal ðxÞ
dH (mm) 0.098 0.11 subject to
dV (mm)
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
5.64 6.59 giL  gi ðxÞ  giU ; i ¼ 1; 2
0.366 0.317
Safety factor by equation (1) t2sa þ t2st =ðtk =gm Þ gi ðxÞ  0; i ¼ 3; :::; 8
736 Y.-S. Lee et al. / Renewable Energy 71 (2014) 729e741

Fig. 5. Shear stress and displacement subjected to bending moments by FEM.

The Progressive Quadratic Response Surface Method (PQRSM) is Table 13 shows a comparison of the initial design and DO design.
adopted in the present study [10]. PQRSM is a sequential approxi- The DO design with 32% reduction in total mass is achieved by
mate optimization method that repeats the design of the experi- reducing the thickness of the transition piece and grout and the
ment and the approximation. PQRSM has global convergence radius of the pile while slightly increasing the angle and length of
characteristics by modifying design ranges at each iteration even the transition piece. Both designs satisfy all the design re-
though it is a local optimization technique. Fig. 7 shows optimiza- quirements. However, the stress range of the fatigue limit state, g8,
tion convergence histories. The convergence history of the ultimate is very close to the boundary of state function. In this case, DO
limit state function of axial (g3) and bending safety (g6), fatigue design can easily violate the feasible design range even with a small
limit state function of design life (g7), and total mass are repre- change in the design. This phenomenon is closely related to the
sented. DO design was achieved after 15 iterations. occurrence of the structural failure. In general, since the DO design
is prone to reach the boundary of the feasible range, reliability may
Table 8 become an issue in this case. The DO design therefore needs to be
Comparison of DNV and FEM results subjected to bending moments.
assessed from a reliability perspective.
Bending DNV FEM

Local contact pressure, ðpLocal Þ 6.33 8.37 6. Reliability-based design optimization

Grout fs (MPa) 7.98 11.99
dH (mm) 11.43 5.16
dV (mm) 7.22 3.42
While structural systems are generally designed based on a
fcck 0.35 0.52 deterministic method, uncertainties exist in a real system.
Safety factor by equation (11) fs = gm Depending on the design problems, the system reliability has to be
assessed by evaluating the exceedance probability of the load car-
rying capacities due to uncertainties of design variables. If needed,
the design reliability has to be increased to the desired level. This
section describes the reliability assessment and RBDO.

6.1. Reliability assessment of DO design

Reliability analysis of DO design is carried out by two steps. The

first step is to quantify the uncertainty of variables. The second step

Table 9
Limit state function values of initial design.

Constraints Symbol Value Unit

Ratio of Rp =tp g1 18.224 e

Ratio of Rtp =ttp g2 15.877 e
Axial safety in ULS g3 0.190 N/m2
Min. principal stress in grout top in ULS g4 1.150 N/m2
Min. principal stress in grout bottom in ULS g5 1.030 N/m2
Bending safety in ULS g6 15.239 N/m2
Design life (log10N) g7 0.649
Fatigue stress range in FLS g8 0.003
Fig. 6. The validity range of stress concentration factor (SCF).
 Y.-S. Lee et al. / Renewable Energy 71 (2014) 729e741 737

Table 10 reliability. Table 11 shows the coefficients of variation (COV) and

% Sensitivities of design variables to each limit state function gi. distribution types of parameters. The uncertainties of the design
a Lg Rp ttp tp tg m Eg E Fz MT Mmax Mmin loads are modeled by a Gumbel distribution and extreme value
g1 0 0 27 0 72 0 0 0 0 0 0 0 0
distribution (EVD) according to an offshore standard [20]. The
g2 0 0 25 74 0 0 0 0 0 0 0 0 0 reliability of the monopile connection has been assessed by eval-
g3 0 2 2 37 33 0 11 0 12 1 0 0 0 uating the probability of satisfaction of the limit state functions
g4 0 7 0 82 0 0 9 1 0 0 0 1 0 described as follows:
g5 0 7 0 0 82 0 9 0 0 0 0 1 0
g6 0 8 0 91 0 0 0 0 0 0 0 1 0 h i
g7 3 10 1 46 35 0 0 1 1 1 1 1 1 Gi ðxÞ ¼ P giL  gi ðxÞ  giU ; i ¼ 1; 2
g8 5 14 0 52 26 0 0 0 0 0 0 3 0
Obj. 0 4 46 25 24 1 0 0 0 0 0 0 0 Gi ðxÞ ¼ P ½gi ðxÞ  0; i ¼ 3; :::; 8
where; x ¼ a; Lg ; Rp ; tp ; ttp ; tg ; m; Eg ; E; Fz ; MT ; Mbmax ; Mbmin

The basic concept of reliability analysis is to predict the failure

is to apply a probabilistic approach to evaluate the level of satis-
probability by a multi-dimensional integral of the failure region
faction of the design requirements and the target reliability.
ðgðxÞ  0Þ of the joint probability density function ðfx ðxÞÞ using the
According to DNV guidelines, the target annual failure proba-
limit state functions ðgðxÞ ¼ 0Þ.
bility of designs is under 1E-4. Since the present design method is
based on the DNV design standards established by the load and Z Z
resistance factor design (LRFD), DO design already includes certain Pf ¼ P½gðxÞ  0 ¼ / fX ðxÞdx
reliability by partial safety factors. In order to keep the conventional
gðxÞ0
design standards for wind turbine systems, the current design
problem is stated based on LRFD method by DNV guidelines. And Approximation integration methods have been widely studied
then uncertainties are additionally distributed on design variables owing to their complexity to evaluate the joint probability density
and the target reliability of the design is set to 0.9 (the failure function and multi-integration of the joint probability density
probability as 0.1). It is based on an assumption that the partial function. Reliability analysis falls into two categories: sampling
safety factors in LRFD would cover the target failure probability method and fast probability integration method.
under 1E-3. Therefore, it can be approximately assumed that the Monte Carlo sampling [11], a sampling method, does not
final failure probability of current design could be achieved under transform the limit state function and can calculate the failure
1E-4. This is a rough assumption for adapting the general design probabilities heuristically. To calculate the randomness of an input
standards of wind turbine systems in the present study. Otherwise, variable, it randomly simulates the samples depending on the
it would be too risky to use the LRFD based design guidelines by probability density functions of input variables. The accuracy of the
removing the partial safety factors only. failure probability depends on the iterating sampling size. In
The physical and statistical uncertainties of the design variables contrast, Latin hypercube sampling [12] is a representative variance
are assumed to have a normal distribution. Additionally, the design reduction method that can be used to reduce millions of iterations
loads are included to investigate their influences on the system of Monte Carlo samplings. The basic concept of Latin hypercube

Fig. 7. Optimization convergence history of limit state functions.

738 Y.-S. Lee et al. / Renewable Energy 71 (2014) 729e741

Table 11 distribution of G8 due to uncertainties of the design parameters.

Uncertainty distributions and COV of parameters. The shaded area is the region of violation for the deterministic state
Parameters COV Distribution function g8, namely, the region in which g8 > 0. The reliabilities of
Geometry a, Lg, Rp, ttp, tp, tg 1% Normal
the initial design and DO design are under 0.5. This indicates that
Material m, Eg, E 1% Normal the DO design suffers from system reliability problems although it
Load Fz, MT, Mmax, Mmin 10% Gumbel satisfies all the design requirements. RBDO is needed to ensure the
10% EVD desired level of reliability of the optimum design.

Table 12
6.2. Formulation of RBDO problem
Reliability of initial and DO designs.

Constraint Reliability of initial design Reliability of DO design The RBDO problem is formulated so as to derive an optimum
G1eG6 1.000 1.000 design with the desired reliability. The design variables are trans-
G7 0.906 0.907 formed as means of the random variables, and the constraints are
G8 0.473 0.465 transformed into probabilistic constraints. RBDO is formulated by
three processes. First, the design variables are defined as random
variables according to their characteristics; second, the design
sampling is to use a sectional approach of the sampling range variables are transformed as means of the random variables in the
evenly for the input random variables to describe real probable optimization process; and third, the constraints are defined as the
situations. probability of the limit state functions and target reliability of the
Latin hypercube sampling is used for the present reliability constraints. RBDO includes reliability analysis within the optimi-
analysis. When using sampling technique, the sample size is zation process to calculate the probabilistic constraints. The target
determined from both the accuracy and the effectiveness point of probability Ptarget of the limit states functions is set to 0.9. RBDO is
view. In the present design, the sample size of 1000 was deter- formulated as follows:
mined by considering the target reliability of 0.9. The computer

time for RBDO took around 5 h with 3.3 GHz dual CPU. Table 12
shows the reliabilities of the initial design and DO design. In the
Find mX ¼ ma ; mL ; mR ; mtp ; mttp ; mtg
g pb
table, reliability means the probability (G1,G2) that the limit state minimize Mtotal ðmX Þ
function value, g1 and g2 are between lower and upper bounds, or subject to h i
the probability (G3eG8) that the limit state functions of g3eg8 are P giL  gi ðxÞ  giU  Ptarget ; i ¼ 1; 2
less than upper bounds. The designs are in the safe region far from
P ½gi ðxÞ  0  Ptarget ; i ¼ 3; :::; 8
limit state functions with results of 1.0, except the fatigue limit
states of g7 and g8. The reliability of the fatigue limit state function, In addition to minimizing the total mass, RBDO derives a design
g8 is under 0.5 for both the initial design and the DO design. to satisfy the target reliability by moving the means of the design
The histogram in Fig. 8 shows the probability distribution and variables to a safer feasible region at each iteration of the optimi-
cumulative distribution of the limit state function g8, namely, the zation. The concept of RBDO is shown in Fig. 9. RBDO has a double-

Fig. 8. Probability distribution of limit state function g8.

 Y.-S. Lee et al. / Renewable Energy 71 (2014) 729e741 739

loop solution: the reliability analysis in the inner loop of the opti- Table 13
mization process. It requires many simulations as against the DO. Comparison of three designs: initial, DO, and RBDO.

In the present process, Latin hypercube sampling repeats Variable name Lower Initial DO design RBDO Upper
structural analysis in the inner loop and the optimization process bound design design bound
using Micro Genetic Algorithm (MGA) repeatedly runs reliability Design variables a 1.000 1.000 1.530 1.980 2.000
analysis in the outer loop [37]. Latin hypercube sampling is used Lg 7.000 7.500 7.805 7.754 8.000
to reduce the number of simulations and MGA, for a stable global Rp 2.000 2.500 2.000 2.034 3.000
ttp 0.050 0.170 0.146 0.175 0.200
search. MGA is a stable optimization method for numerical noise
tp 0.050 0.130 0.131 0.152 0.200
because it does not use gradient information or an approximate tg 0.050 0.160 0.050 0.051 0.200
model. MGA have been developed and modified in detail based Objective function Mass 316.3 213.8 249.5
on genetic algorithms. MGA uses a small population and no Constraints G1eG6 0.900 1.000 1.000 1.000 e
G7 0.900 0.906 0.907 1.000
mutation efficiently to reduce the number of iterations and e
G8 0.900 0.473 0.465 0.905 e
achieve fast convergence in the inner loop. The population is
reorganized when a local optimum is reached with small popu-
lation size.
the monopile connection are varied to minimize the total mass
6.3. Optimum design based on RBDO within the desired reliability.
Table 13shows three designs: initial design, DO design, and
The convergence history of RBDO is shown for the reliability RBDO design. The RBDO design satisfies all the limit state functions
variation of the fatigue stress range g8 and the variation of the total with the target reliability of 0.9, although the total mass is slightly
mass in Fig. 10. The reliability of the fatigue stress range g8 rapidly increased in comparison to the DO design, with the reduction rate
increases and converges near 0.9. The other limit state functions decreasing to 21% from 32%. The reliability of the fatigue limit state
have reliabilities under 1.0. During RBDO, the design candidates of function g8 reaches up to 0.905 compared to 0.465 after RBDO. The

Fig. 9. Concept of reliability-based design optimization.

Fig. 10. Convergence history of RBDO at limit state functions.

740 Y.-S. Lee et al. / Renewable Energy 71 (2014) 729e741

Fig. 11. Probability distributions of g7 and g8 for DO and RBDO designs.

reliability of g7 is improved to 1.0. RBDO improves the reliability of reliability up to 0.9 and therefore ensure system reliability. Fig. 12
the existing design to the desired level. shows a comparison of the design variations of the monopile
Fig. 11 shows the probabilistic distributions and cumulative connection for the initial design, DO design, and RBDO design.
probabilities of the limit state functions g7 and g8 for the DO design The most sensitive design variables for the total mass are the
and RBDO design. The probabilistic distribution of the RBDO design diameter of the pile Rp and the thickness of grout tg. Rp and tg are
(red in the web version) is confirmed to shift toward the safer region reduced by 20% and 68%, respectively, through deterministic mass
over 90% in comparison to that of the DO design (blue in the web minimization. These size reductions are achieved for the lower
version). The limit state function g7 is distributed completely outside bounds of the variables. While the thickness of the transition piece
the shaded area. The total mass of the initial design is decreased by ttp is decreased by 14% for mass reduction, it is again increased by
32% through the DO design and 21% through the RBDO design. The 17% to ensure the reliability of the fatigue state after RBDO. Simi-
slight increase in mass through RBDO is inevitable to achieve a target larly, the thickness of the pile is increased by 17%. The conical shape

Fig. 12. Design changes of conical monopile connection.

 Y.-S. Lee et al. / Renewable Energy 71 (2014) 729e741 741

of the connection helps to reduce the mass and increase the fatigue [5] Fischer T, de Vries W, Schmidt B. Upwind design basis (WP4: offshore foun-
dations and support structures). Upwind deliverable 4e10; 2010.
strength of the design. The inclination angle of the connection is
[6] Agarwal P, Manuel L. Simulation of offshore wind turbine response for long-
increased by up to 2 . By comparing the DO design and RBDO term extreme load prediction. Eng Struct 2009;31(10):2236e46.
design, we found that the thickness of the transition piece and pile, [7] Marino E, Borri C, Peil U. A fully nonlinear wave model to account for breaking
ttp and tp, are critical design variables for the fatigue strength, G8. A wave impact loads on offshore wind turbines. J Wind Eng Ind Aerodyn
2011;99(4):483e90.
conical shape is preferable for the fatigue strength of the monopile [8] Marino E, Lugni C, Borri C. A novel numerical strategy for the simulation of
connection as well as the ultimate strength. irregular nonlinear waves and their effects on the dynamic response of
offshore wind turbines. Comput Methods Appl Mech Eng 2013;255:275e88.
[9] Tarp-Johansen NJ, Agyriadis K, Carstens H, Tempel JVD, Camp T, Cheng PW.
7. Conclusions Review of modeling approaches for irregular, non-linear wave loading on
offshore wind turbines and their relevance for future designs. UpWind
Two categories of designs are accomplished in this study for the Deliverable Report D4.3.1; 2008.
[10] Hong KJ, Choi DH, Kim MS. Progressive quadratic approximation method for
cost-effective and reliable design of a monopile transition piece: effective constructing the second-order response surface models in the large
deterministic structural optimization and reliability-based design scaled system design. Trans Korean Soc Mech Eng A 2000;24(12):3040e52.
optimization. [11] Sailby E. Descriptive sampling: a better approach to Monte Carlo simulation.
J Opl Res Soc 1990;41(12):1133e42.
Design loads are predicted by the integrated dynamic analysis of [12] Saliby E. Descriptive sampling: an improvement over Latin hypercube sam-
the NREL 5 MW reference turbine with a reference monopile support pling. In: Proceedings of the 29th Conference on Winter Simulation. IEEE
structure. The analytical structural capacities are analyzed based on Computer Society; 1997. pp. 230e3.
[13] Jonkman J, Musial W. Offshore Code Comparison Collaboration (OC3) for IEA task
the design standards. Finite element analysis is conducted for vali-
23 offshore wind technology and deployment. Contract; 2010. 303-275-3000.
dating the analytical structural capacities of the monopile connections. [14] Passon P, Kühn M, Butterfield S, Jonkman J, Camp T, Larsen TJ. OC3dbench-
Deterministic design (DO) for a conical monopile connection has mark exercise of aero-elastic offshore wind turbine codes. J Phys 2007;75(1):
been undertaken, resulting in 32% reduction of the total mass. The 012071.
[15] Schepers JG, Heijdra JJ, Thomsen K, Larsen T, Foussekis D, Rawlinson Smith R,
most sensitive design variables for total mass are identified as the et al. Verification of European wind turbine design codes. EU-JOULE Project
diameter of the pile Rp and the thickness of grout tg. The reliability Final Report; 2002.
assessment showed that the DO design converges close to the fa- [16] Jonkman J, Butterfield S, Passon P, Larsen T, Camp T, Nichols J, et al. Offshore
Code Comparison Collaboration within IEA wind annex XXIII: phase II results
tigue limit state and its reliability measure is below 0.5. Hence, one regarding monopile foundation modeling. In: European Offshore Wind Con-
concludes that the structural design of the monopile connection is ference & Exhibition, 4e6 December 2007, Berlin, Germany; 2007.
mostly associated with the fatigue strength and that DO does not [17] Camp T. Design methods for offshore wind turbines at exposed sites. Final
Report, EU Joule III JOR3-CT95-0284; 2003.
guarantee the reliability of the structure although the design sat- [18] Det Norske Veritas (DNV), Joint Industry Project. Guideline for offshore structural
isfies all limit state function conditions. reliability analysis: examples for jacket platforms. Report No. 95-3204; 1996.
RBDO has targeted to ensure an acceptable level of reliability as [19] Det Norske Veritas (DNV). Design of offshore wind turbine structures.
Offshore standard DNV-OS-J101; 2010.
well as mass minimization. RBDO for the present design consists of [20] Det Norske Veritas (DNV). Offshore concrete structures. Offshore standard
two algorithmsdLatin hypercube sampling and MGAdfor higher DNV-OS-C502; September 2012.
convergence speed and global optimization. [21] Det Norske Veritas (DNV). Design of offshore wind turbine structures.
Offshore standard DNV-OS-J101; 2013.
By comparing the DO design and RBDO design, it is shown that
[22] API. Recommended practice for planning, designing and constructing fixed
the thickness of the transition piece and pile, ttp and tp, are critical offshore platforms-working stress design: API Recommended Practice 2A-
design variables for the fatigue strength. Therefore, the thicknesses WSD (RP 2A-WSD). American Petroleum Institute; 1993.
of the pile and transition piece are designed from a fatigue strength [23] Jensen JJ, Olsen AS, Mansour AE. Extreme wave and wind response pre-
dictions. Ocean Eng 2011;38(17):2244e53.
viewpoint to withstand fatigue loads over a 20-year lifetime. In [24] Sorokin SV. Analysis of grouted connection in monopile wind turbine foun-
addition, a conical shape is found to be preferable for the fatigue dations subjected to horizontal load transfer. Diss, Aalborg: Aalborg Univer-
strength of the monopile connection. sity; 2009.
[25] IEC-61400-3. Wind turbinesdpart 3: design requirements for offshore wind
The necessity of RBDO is evident, especially for a large-scale turbines. Edition 1.0. International Electrotechnical Commission; 2009.
offshore wind turbine in harsh and rapidly varying environmental [26] Jing Z, Li Y, Kang H. Application of DSI techniques to monopile foundations of
conditions, for reliable and cost-effective design purposes. The offshore wind turbines reliability problems. EJGE 15 Bund. A.; 2010.
[27] Lozano-Minguez E, Kolios AJ, Brennan FP. Multi-criteria assessment of offshore
design process of RBDO as proposed in this study has been proven wind turbine support structures. Renew Energy 2011;36(11):2831e7.
for speeding up the design cycle and increasing the reliability of the [28] Dong W, Moan T, Gao Z. Fatigue reliability analysis of the jacket support
grouted connection for offshore wind turbine support structures. structure for offshore wind turbine considering the effect of corrosion and
inspection. Reliab Eng Syst Saf 2012;106:11e27.
[29] Dong W, Moan T, Gao Z. Long-term fatigue analysis of multi-planar tubular
Acknowledgment joints for jacket-type offshore wind turbine in time domain. Eng Struct
2011;33(6):2002e14.
[30] Pinar Pe rez JM, García Ma rquez FP, Tobias A, Papaelias M. Wind turbine
This work was supported by the New & Renewable Energy of the
reliability analysis. Renew Sustain Energy Rev 2013;23:463e72.
Korea Institute of Energy Technology Evaluation and Planning [31] Taflanidis AA, Loukogeorgaki E, Angelides DC. Offshore wind turbine risk
(KETEP) grant (SUBJID 0000000014874) funded by the Korea quantification/evaluation under extreme environmental conditions. Reliab
Government Ministry of Knowledge Economy. Eng Syst Saf 2013;115:19e32.
[32] Garrad Hassan GL. Bladed theory manual version 4.1. Garrad Hassan & Part-
ners Ltd.; 2010.
References [33] PIAnO (Process Integration, Automation and Optimization). User's manual.
PIDOTECH Inc.; 2012.
[1] de Vries W, Vemula NK, Passon P, Fischer T, Kaufer D, Matha D, et al. Final [34] Lee DY, Kim JH, Lee YS, Lee JH. Sensitivity analysis and optimization of
report WP4. 2: support structure concepts for deep water sites. Tech. Report, monopile transition piece in 5 MW offshore wind turbines. In: Proceedings of
Project UpWind; 2011. the KWEA Spring Conference, KWEA; 2012.
[2] Jonkman J, Butterfield S, Musial W, Scott G. Definition of a 5-MW reference [35] Kim JY, Oh KY, Kang KS, Lee JS. Site selection of offshore wind farms around
wind turbine for offshore system development. NREL/TP-500-38060. Golden, the Korean Peninsula through economic evaluation. Renew Energy 2013;54:
CO: NREL; 2009. 189e95.
[3] Shi W, Park H, Chung C, Baek J, Kim Y, Kim C. Load analysis and comparison of [36] Muskulus M, Schafhirt S. Design optimization of offshore wind turbine sup-
different jacket foundations. Renew Energy 2013;54:201e10. port structures e a review. ISOPE J Ocean Wind Energy 2014;1:12e22.
[4] Shi W, Park H, Han J, Na S, Kim C. A study on the effect of different modeling [37] Krishnakumar K. Micro-genetic algorithms for stationary and non-stationary
parameters on the dynamic response of a jacket-type offshore wind turbine in function optimization. Intell Cont Adapt Syst 1989;1196:289.
the Korean Southwest Sea. Renew Energy 2013;58:50e9.