Soil Dynamics and Earthquake Engineering 105 (2018) 171–183

Contents lists available at ScienceDirect

Soil Dynamics and Earthquake Engineering

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

Probabilistic analysis of monopile-supported offshore wind turbine in clay T

a,⁎ b c
Sumanta Haldar , Jitendra Sharma , Dipanjan Basu
a
Department of Civil Engineering, School of Infrastructure, Indian Institute of Technology Bhubaneswar, Jatni 752050, India
b
Department of Civil Engineering, Lassonde School of Engineering, York University, 4700 Keele St., Toronto, ON, Canada M3J 1P3
c
Department of Civil and Environmental Engineering, University of Waterloo, 200 University Avenue W, Waterloo, ON, Canada N2L 3G1

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

Keywords: The dynamic behaviour of monopile supported offshore wind turbines (OWTs) is considerably influenced by
Clay inherent variabilities of soil properties and loading. This study investigates the effect of uncertainties in soil
Dynamics shear strength properties, and wind and wave loads on the reliability of OWT structures founded in clay. The
Offshore wind turbine OWT system is modeled as an Euler-Bernoulli beam and soil-structure interaction is incorporated using the
Soil-structure interaction
American Petroleum Institute based cyclic p-y relationships. The uncertainties in soil undrained shear strength,
Uncertainty
in the parameters and equations of the p-y method, and in wind velocity are considered. Random wave loads are
estimated from the code specified power spectral density function of the vertical sea surface elevation.
Uncertainties in OWT responses are quantified using Monte Carlo simulations. The effects of length and diameter
of the monopile, vertical sea surface displacement spectrum, and the probability distribution of wind speed on
the dynamic responses of OWT are investigated. The study shows that the various power spectral density
functions of wave surface displacement and various probability density functions of wind speed have a marginal
effect on the response and fatigue life of OWTs. The mean fundamental frequency of the OWT system is sig-
nificantly affected by the variability of the undrained shear strength of clayey soil.

1. Introduction [25,34]. Several studies in the past focused on probabilistic aspects of

Design of foundations plays an important role in the overall design
of offshore wind turbines (OWTs) because the foundation accounts for
about 30–40% of the total project cost [4–7]. Monopiles are economical
foundation options for OWTs installed at shallow water depths [1–5],
and are subjected to complex aerodynamic and hydrodynamic forces
arising from wind and ocean waves [16,17]. The tolerance in terms of
maximum allowable rotations of the monopile head in a vertical plane
at seabed (i.e. serviceability limit state (SLS) criterion) is considered as
design basis of OWT [18]. Likewise, fatigue limit state (FLS) of
minimum 107– 108 load cycles over 20–25 years to be sustained under
long term wind and wave loads [18–22]. In addition to SLS and FLS
criteria, the other design criterion is that the fundamental frequency of
OWT should not interfere with the rotor frequency (1P), tower sha-
dowing frequency (3P for 3 bladed OWT), and wave frequency
[8–15,19,21,22,27,28,78].
Estimation of OWT response against applied loads and its funda-
mental frequency are associated with significant uncertainties because
of inherent soil variability, uncertainty in applied loads, and un-
certainty in the analysis models [23–25]. Uncertainty in soil properties
arises from the heterogeneity of seabed and limited number of sampling
OWT design [24,32–43]. Out of the probabilistic studies performed on
OWT, most focused on reliability based design optimization of OWT
considering uncertainties in loading and material properties of struc-
ture [32–35,37–43], and the uncertainty in soil properties were not
considered. A few of these studies focused on the response of OWT
considering uncertainties in loading and soil shear strength parameters
[24,36]; `however, the dynamic soil-monopile-structure interaction was
not considered. In all these studies, significant variation in the re-
sponses of OWT was observed because of the incorporation of varia-
bility in loading and soil parameters. Various design codes related to
OWT design, such as Det Norske Veritas (DNV) [18,26], Germanischer
Lloyd (GL) [29], International Electrochemical Commission (IEC) [30]
and American Bureau of Shipping (ABS) [31], advocate the use of
probabilistic design models to calibrate partial factors used in design
and special design requirements [24]; however, no explicit metho-
dology on reliability based OWT design incorporating the effects of
dynamic soil-structure interaction is yet available in these codes. In
order to get further insights into the effect of different uncertainties on
OWT performance, a comprehensive assessment of the dynamic beha-
viour of OWT considering the variability of wind and wave loading and
soil parameters is required [16,44].

⁎
Corresponding author.
E-mail addresses: sumanta@iitbbs.ac.in (S. Haldar), jit.sharma@lassonde.yorku.ca (J. Sharma), dipanjan.basu@uwaterloo.ca (D. Basu).

https://doi.org/10.1016/j.soildyn.2017.11.028
Received 17 May 2017; Received in revised form 25 November 2017; Accepted 26 November 2017
0267-7261/ © 2017 Elsevier Ltd. All rights reserved.
S. Haldar et al.
In this paper, the effects of uncertainties in soil properties and ap-
plied loads on the response of monopile-supported OWTs are in-
vestigated using a nonlinear soil-structure interaction analysis per-
formed within a probabilistic framework. Dynamic finite element (FE)
analysis is performed to obtain the OWT responses in the time domain.
The monopiles are assumed to be embedded in clayey soil, and are
analyzed using the beam-on-nonlinear-foundation approach. The
American Petroleum Institute (API) [45] recommended cyclic p-y
curves for soft clays are used to model the lateral soil resistance. This
study primarily examines the lateral behaviour of OWT; therefore, the
monopile is assumed to rest on a firm ground so that the vertical pile
displacement is considered negligible compared with the corresponding
horizontal pile displacement. The uncertainties in the soil-structure
interaction model, soil properties, wave load, and wind load are con-
sidered, and combined through Monte-Carlo (M-C) simulations to ob-
tain the probability distributions of the maximum mudline rotation,
maximum rotation at tower top, and fatigue life. The statistics of OWT
responses are examined as functions of embedded length and diameter
of monopiles, soil stiffness, wind velocity, spectral density of waves,
and probability density function of wind speed. The probabilities of
failure based on SLS criterion for OWT design are also obtained.
2. Deterministic model
2.1. Problem definition
An OWT supported by a monopile embedded in a clay layer is
considered in this study (Fig. 1). The tower is assumed to behave like an
Euler-Bernoulli beam with flexural rigidity EpIp,. The monopile also
behaves like an elastic beam following Euler-Bernoulli theory, with
flexural rigidity EpIp, and resists lateral loads. Uniform cross-sections for
the monopile and the tower are assumed for simplicity. The monopile is
assumed to rest in a bed of nonlinear Winkler springs representing the
resistance of the clay layer against lateral pile movement. The hor-
izontal Winkler soil springs are assumed to be attached with each beam
element of the monopile and generate the lateral resistance against
monopile movement following the p-y model recommended by API
[45]. The tower and monopile are discretized into beam elements with
172
Soil Dynamics and Earthquake Engineering 105 (2018) 171–183
three degrees of freedom (lateral and vertical displacements, and ro-
tation) at each node.
The wind thrust acting on the rotating blades eventually gets
transmitted to the tower at the hub in the form of the force Fb. Fb is
considered dynamic with a frequency equal to the rotor frequency, and
is designated as 1P load (i.e., once per revolution). Two additional
forces F1tower and F2tower act on the tower — one over the part of tower
obstructed during the movement of the blades (F1tower) and the other on
the tower unobstructed by the blades (F2tower). The obstruction from
blade rotation occurs three times per revolution which makes F1tower a
dynamic force on the tower with a frequency three times of 1P (de-
signated as 3P load). The load F2tower on tower is static. The shaded
region in wind velocity profile (Fig. 1) shows the region need to be
considered for the estimation of Fb, F1tower and F2tower. The forces F1tower
and F2tower act at the centroid of the shaded area representing wind
velocity profile. The wave force Fw is assumed to be dynamic and acts at
the mean sea level (MSL) with a wave frequency.
2.2. Numerical model
Finite element analysis is carried out using OpenSees [52] to esti-
mate the responses of the OWT system. Two-noded beam elements (in
which the beam deflection and slope are the nodal degrees of freedom,
which are interpolated using the cubic Hermitian shape functions) are
used to model the monopile and tower. A roller support is introduced at
the base of monopile to prevent vertical displacement. The rotor blades,
nacelle and hub are combined into a single mass MRNA placed at the top
node of the tower with rotary inertia JRNA (c.f. Fig. 1). The soil re-
sistance against lateral pile movement is modeled as nonlinear springs
that follow the API p-y curves. The overall damping is modeled as a
series of viscous dashpots connected to the monopole such that each
dashpot is placed in parallel with each soil spring. A convergence study
was carried out and it was observed that 1 m length of beam elements
with soil springs attached at 1 m interval produced satisfactory results.
The dynamic equation of motion of the system is solved by using
Newmark’s average acceleration method (the ‘transient analysis’ solver
in OpenSees [52] was used). The nonlinearity in p-y curves is accounted
for by using a simple step-load controlled incremental algorithm
Fig. 1. Schematic diagram of OWT system in clay and the corre-
sponding finite element model.
S. Haldar et al.
following the Newton-Raphson method in which the tangent modulus is
updated at each step. This results in a decrease in the tangent stiffness
modulus because of increase in the load amplitude. Fb is applied at top
node of the tower. The points of applications of F1tower, F2tower and Fw
are shown in Fig. 1. In this study, wind and wave loads are considered
random the details of which are discussed in the following sections. The
fundamental frequency of the OWT system is determined using Eigen-
frequency analysis considering the initial slope of p-y curves [26,52].
2.3. p-y model
The cyclic p-y model for soft clay, recommended by API [45], is used
to characterize the Winkler soil springs attached to the pile. According
to the model, the nonlinear cyclic p-y equations for soft undrained clay
(with undrained shear strength su ≤ 100 kPa) can be generated based
on (i) undrained shear strength of clay (su), (ii) the strain (ε50) corre-
sponding to one-half the maximum stress on laboratory undrained
compression tests of undisturbed soil, and (iii) an empirical di-
mensionless constant (J).
Ashour et al. [47] presented a relationship between ε50 and su gra-
phically based on which a fitted equation (with the coefficient of de-
termination r2 = 0.98) is obtained that gives the average value of ε50
(i.e., value at the middle of the range proposed by Ashour et al. [47]) as
a function of su:
ε50 = 0.0025 su + 0.5121/ su (1)
Matlock [46] observed that J = 0.25 and 0.5 fitted the field test
results well at two different sites and that the value of J = 0.25 may be
applicable for stiffer clays, and proposed a value of J = 0.5 for Gulf of
Mexico clay. A constant value of J = 0.5 is used in this study for all
cases.
It is to be noted that the applicability of API [45] based cyclic p-y
curve is strictly limited to small diameter, slender piles and were pro-
posed based on field tests conducted for less than 200 cycles [48]. In
general, it overestimates soil reaction at great depths and under-
estimates it at the shallow depths for large diameter monopiles [1,12].
In addition, the behaviour of clay at low to medium strain level is not
accurately taken into account in the API [45] based cyclic p-y curves,
which may result in inaccurate monopile responses at low load levels
[18,49]. Notwithstanding the limitations of API [45] based p-y curves,
these are extensively used for monopile design in the offshore industries
[15,19,25] and also recommended in several design guidelines
[18,29,45].
2.4. Damping
In this study, aerodynamic damping (ηaero), hydrodynamic damping
(ηhyd), material damping (ηm), and radiation damping (ηrad) are con-
sidered. An aerodynamic damping of 3.5% is considered based on Kühn
[50]. The hydrodynamic damping in OWT results from the drag be-
tween water and structure and is about 0.15% [29], which is adopted in
this study. A material damping of 8% is used based on [51]. The ra-
diation damping is taken as 0.22% [29]. Therefore, an overall damping
of 12% (i.e., ηaero + ηhyd + ηm + ηrad = 8% + 3.5% + 0.15% + 0.22%
= 11.87% ≈ 12%) is used in the analysis.
2.5. Wind and wave loads
The total wind load acting on the tower is divided into two com-
ponents, the load acting on the turbine blades and the load acting on
the turbine tower. Total wind thrust Fb acting on turbine blades is given
by [53,62]
Fb = 0.5ρa πRT2 Vhub
2
CT (λs ) (2)
where Vhub is the wind speed at the hub height (m/s), RT is the rotor
radius (m), ρa is the air density (= 1.23 kg/m3 at 15 °C and 1 atm),
173
Soil Dynamics and Earthquake Engineering 105 (2018) 171–183
CT(λs) is the thrust coefficient and is a function of the tip speed ratio λs,
which is given by,
λs = Vr RT / Vhub (3)
in which Vr is the rotor speed in rad/s. Villalobos Jara [53] graphically
showed a relationship between CT and λs which is adopted for analysis.
Because of rotation of blades, the thrust Fb is considered to be acting as
a sinusoidal force with a rotational frequency of the generator. There-
fore, the dynamic force in the horizontal direction at the hub level
isFb sin(2πfr t ) where Fb is the amplitude of load, fr is rotor frequency
and t is the dynamic time.
The wind velocity profile is shown in Fig. 1, which results in a
distributed force Ftower on tower, and is given by [31],
Ftower (Z ) = 0.61 V 2 (Z ) Cs D (4)
where Ftower(Z) is the wind load acting on the tower in N/m, V(Z) is the
wind velocity profile in m/s, Z is the height from MSL to hub height in
m, Cs is the shape coefficient (0.5 for a cylindrical structure), and D is
the tower diameter which is assumed to be constant for the tower
height and the same as monopole diameter for simplicity.
The wind load acting on the tower is calculated based on the ve-
locity distribution over the tower height, and is expressed as [58]
V (Z ) = (u*/ ka) ln(Z / z 0) (5)
where ka is the Von Karman’s constant (= 0.4), Z is from hw (i.e., MSL)
to hub height (hw + Ht) in m (see Fig. 1), z0 is a terrain roughness
parameter and u* is the wind friction velocity calculated from 10 min
average wind speed (U10) measured at 10 m height from MSL. The
roughness parameter z0 has uncertainty associated with it because of
inadequate knowledge [59,60], and is treated as a random normal
variable with the mean equal to 0.0001 m in the open sea without
waves [58], and the coefficient of variation equal to 10% [61]. The
wind friction velocity u* is estimated from
u* = κ U10 (6)
where the surface friction coefficient κ is defined as
κ = {ka/ln(10/ z 0)} 2 (7)
The 3P force F1tower is estimated by integrating Ftower(Z), i.e., Eq. (4)
over the height from the tip of the blade covering the tower (i.e., Z = hw
+ Ht − RT) to hub height (i.e., Z = hw + Ht). This force is dynamic and
acts as a sinusoidal horizontal forceF1tower sin(6πfr t ) , whereF1tower is the
amplitude and t is the dynamic time. The F2tower is static and estimated
by integrating Eq. (4) over the height from MSL (i.e., Z = hw) to the tip
of the blade covering the tower (i.e., Z = hw + Ht − RT).
The deterministic wave force Fwave(t) on the structure is estimated
using Morison’s equation. The wave load is applied as lateral point load
at the mean sea level. Detailed descriptions of wind and wave loads are
given in Table 1.
3. Probabilistic model
3.1. Uncertainties in wind loads
The uncertainty in wind speed is modeled using the 10-min average
wind speed U10, standard deviation σU of U10, and probability density
function of U10 [54,55]. The variability in wave height is generally
considered using the power spectral density (PSD) functions, in which
the contribution of each frequency of the total power in ocean wave can
be described [16,56]. Variability in the wind speed is modeled by
treating U10 as a random variable with σU as the standard deviation
[58]. The fluctuation of U10 (referred to as turbulence) is characterised
by the turbulence intensity I = σU/U10. Effects of wind load parameters
are studied considering three U10 values, namely 7 m/s, 15 m/s, and
25 m/s, which cover the general range of wind speed in ocean en-
vironments. The turbulence intensity varies with mean wind speed, site
S. Haldar et al. Soil Dynamics and Earthquake Engineering 105 (2018) 171–183

Table 1
Deterministic wind and wave load on OWT.

Load Equation Remarks

Wind load at hub Fb Fb = 0.5ρa πRT2 Vhub

2
CT (λ s ) Villalobos Jara (2006) showed a relationship between CT and λs graphically which is
[53,62] λ s = Vr RT / Vhub ‵ adopted for analysis.
Wind load on tower F1tower = ∫h
hw + Ht
0.61 V 2 (z ) Cs Ddz V (z ) = U10
ln(z / z 0)
.
[31,58] w + Ht + RT ln(10 / z 0)
hw + Ht − RT It is treated as dynamic and acting as sinusoidal load (horizontal), F1tower sin(6πfr t ) , where
F2tower = ∫h 0.61 V 2 (z ) Cs Ddz
w
F1tower is the amplitude, t is the dynamic time and fr is the rotor frequency in Hz. F2tower is a
static load.
Wave load [27,28] η (t ) η (t )
Fw (t ) = CM ρπD 2 /4 ∫−d ü (t ) dz + CD ρD/2 ∫−d u̇ (t ) u̇ (t ) dz where η (t ) = 0.5hw cos(ωt )
w w
hw π cosh(k (z1 + dw ))
u̇ (t ) = cos(kx − ωt )
Tw sinh(kdw )
2hw π 2 cosh(k (z1 + dw ))
ü (t ) = 2 sin(kx − ωt )
Tw sinh(kdw )

Vhub = wind speed at the hub height (m/s), RT = rotor radius (m), ρa = air density (= 1.23 kg/m3 at 15 °C and 1 atm), CT(λs) = the thrust coefficient and is a function of the tip speed
ratio λs, Vr = rotor speed in rad/s, V(z) = wind velocity profile in m/s, z = height varies from MSL to hub height in m, Cs = shape coefficient (0.5 for a cylindrical structure), D = tower
and monopile diameter, z 0 = terrain roughness parameter = 0.0001 m in the open sea without waves, U10 = 10-min average wind speed at 10 m height from the MSL, dw = water depth
in m, CD = drag coefficient (0.7 for a smooth tubular section), CM = mass coefficient (2 for a smooth tubular section), ρ = mass density of the sea water (1030 kg/m3 approximately), u̇
(t) and ü(t) are the wave induced velocity and acceleration of water, respectively in the horizontal direction, and η (t ) = surface wave profile, hw = wave height in m, k = wave number in
m−1, x is the distance from the origin of the structure which is considered as zero, ω = wave frequency in rad/s, t = time in sec, λ = wave length in m.

location and surface roughness [16,30]. A constant value of I = 0.2 is
assumed in this study for simplicity. The uncertainties in wind load
estimation also arise from the choice of probability distribution of wind
speed [54]. The probability distribution for U10 is generally derived
from the available wind speed data; however, in the absence of site
specific data, analytical probability distributions may be assumed
[58,63,64]. Previous studies have indicated that wind speed distribu-
tion can be represented using Weibull, Rayleigh, Lognormal and
Gamma distributions [55,58,65,66]. These studies showed that the
probability distribution of wind speed cannot be universally determined
because it primarily depends on the local site conditions. In this study,
three probability distribution functions of U10, Weibull, Lognormal and
Gamma, are considered for probabilistic analysis.
3.2. Uncertainties in wave loads
The irregular and random nature of ocean waves is represented
using a random wave model. The random wave load Fwave (t ) can be
modeled as a superposition of many linear waves each with different
amplitude, frequency and random phase angles:
N
Fwave (t ) = ∑ Ai sin(2πfi t + φi)
i=1
where t is the time, φi is the phase of the ith wave assumed to be uni-
formly distributed over the interval (0, 2π), Ai is estimated from the
power spectral density (PSD) of wave force as
resulting PSD may not coincide with the initial PSD of wave force [69].
Hence, PSD of time history is further obtained using Fast Fourier
transform and updated as
SF0 (fi )
SF(j + 1) (fi ) = SF(j) (fi )⋅ (j )
SFB (fi ) (10)
where SF0 (fi )
is the prescribed wave force PSD function (derived from
(j )
PM or JONSWAP spectrum), SFB (fi ) is the derived PSD from the gen-
erated time history at the jth iteration, and SF(j) (fi ) is the PSD function at
the jth iteration. Note that, in the first iteration, SF(j) (fi ) is equal to SF0 (fi ) .
A convergence check is performed by estimating the root-mean-square
error between the target PSD and generated PSD from each iteration.
The convergence of PSD was generally achieved after 6 iterations for
the calculations performed in this research. A comparison of prescribed
PSD with computed PSD from the sample function obtained at the end
of the iteration process is presented in Fig. 2.
Arany et al. [27] suggested five important load cases and indicated
50 year extreme return wind and wave load model could be critical for
foundation design. In this study, the wind and wave loads are evaluated
using the probability distribution functions for wind speed and wave
spectra, respectively for short term stationary condition. This wind and
wave climate representation is not intended to cover extreme events,
(8)
which is future scope the study.
3.3. Uncertainties in soil properties and model

Ai = 2SF (fi ) Δf
(9)
in which SF (fi ) is the PSD function of wave force corresponding to the
frequency fi and Δf is the frequency interval used for calculating the
above mentioned function. The wave force spectrum can be obtained
from the vertical sea surface displacement of short term stationary ir-
regular sea state. The DNV code [58] suggests the use of Pierson-Mos-
kowitz (PM) and JONSWAP spectra as vertical sea surface displacement
spectrum, which are given in Table 2.
The time history of wave force is derived from SF (f ) , as given in
Table 2, for f = 0.04 – 0.5 Hz.
An iterative method is adopted to generate the time series of
Fwave (t ) . Initially, the time history of Fwave (t ) is generated from wave
force spectrum developed from either JONSWAP or PM spectrum.
Ideally this wave force spectrum and PSD of the generated Fwave (t )
should be identical. However, because of nonlinear transformation, the
The cyclic p-y curve requires inputs of two soil properties, namely,
the undrained shear strength of clay su and the soil unit weight γ. The
variability in the estimation of su from different types of tests can be
indirectly captured by the coefficient of variation COVsu of su. Previous
studies based on field and laboratory test results suggest that COVsu
varies within 20–40% and that the probability density function of su
follows a lognormal distribution [70–72]. It is reasonable to assume
that the representative value of COVsu is 20% for relatively uniform
deposits and for good quality tests, and that the representative value is
40% for deposits with greater degree of variability and/or for tests with
low quality control. Consequently, in this study, su is assumed to follow
alog-normally distributed random variable with three representative
values: COVsu = 20%, COVsu = 30% and COVsu = 40%. The soil unit
weight γ is assumed to follow a normal distribution with a COV of 5%
based on [71–73].
The p-y curves were developed empirically based on back-analysing

174
S. Haldar et al. Soil Dynamics and Earthquake Engineering 105 (2018) 171–183

Table 2
Probabilistic wind and wave load on OWT.

Description Remarks

Wind Load
Fb, F1tower and F2tower are estimated from the equations presented in Table 1. The roughness parameter z0 and U10 The z0 is uncertain because of inadequate knowledge
are treated as random variable. [59,60]. The variability of U10 is described by the
turbulence intensity I [16,30].
Wave Load
Spectral density function for wave force: SF (f ) =
8
(0.5CD ρD)2σu2̇ Su̇ (f ) + (CM ρπd 2/4)2Sü (f ) The wave force spectrum is estimated using Morison’s
π
equation [28,56,68]. CD is the drag coefficient (0.7 for a
cosh2 (kz ) cosh2 (kz ) ∝
where: Su̇ (f ) = (2πf )2 Sη (f ) ; Sü (f ) = (2πf ) 4 Sη (f ) and σu2̇ = ∫0 Su̇ (f ) df tubular section), CM is the mass coefficient (2 for a tubular
sinh2 (kdw ) sinh2 (kdw )
section), ρ is the mass density of the sea water (1030 kg/m3
−4
5 ⎛ 5 f ⎞ approximately), u̇ and ü are the wave induced horizontal
Sη (f ) = SPM (f ) = H 2 f 4 f −5 exp ⎜− ⎛ ⎞ ⎟; PM Spectra
32π s p
⎜ ⎟
4 fp velocity and acceleration of water, respectively, k is the
⎝ ⎝ ⎠ ⎠
or
wave number in m−1, dw is the water depth in m.Sη (f ) is the
⎛ f − f p ⎞2⎞ PSD of vertical sea-surface displacement which can be
−4 exp ⎜−0.5 ⎛⎜ represented as PM or JONSWAP spectrum [58]. Hs = 4 m
5 ⎛ 5 f ⎞ ⎜ σ . f p ⎟ ⎟⎟
Sη (f ) = SJ (f ) = H 2 f 4 f −5 (1 − 0.287 ln(λ ))exp ⎜− ⎛ ⎞ ⎟ λ ⎝ ⎝ ⎠ ⎠;
and fp = 0.1 Hz is considered in this study, because in real
32π s p
⎜ ⎟
4 fp
⎝ ⎝ ⎠ ⎠ ocean, short crested waves (with wave period less than 20 s)
JONSWAP Spectra are very likely [67]. λ = 3.3, σ = 0.07 when ω ≤ ωp and σ
= 0.09 when ω > ωp [58]

f is the wave frequency in Hz, fp is the spectral peak frequency in Hz, Hs is the significant wave height in m, σ is the spectral width parameter, λ is the peak enhancement factor.

3.4. Monte-Carlo simulations

A summary of the probabilistic data for the parametric study is

presented in Table 4. Monte Carlo simulations were performed to
generate the samples of selected random variables using the data. A
total of 5000 Monte-Carlo (M-C) simulations were performed for each
problem analyzed in this study. Similarly, 5000 samples of wave and
wind loads were generated. If the number of simulations is Nsim, then
the standard errors in the estimated mean and variance of OWT re-
sponses are respectively equal to the sample standard deviation times
1/ Nsim and sample standard deviation times 2/(Nsim − 1) [57]. For
Nsim = 5000, 1/ 5000 = 0.014 and 2/(5000 − 1) = 0.02; this implies
that the standard errors of the estimated mean and variance of OWT
responses are respectively 1.4% and 2% of the sample standard de-
viation. Thus, the estimated OWT responses in this study would gen-
erally be around 2% of the true quantities.

4. Wind turbine properties

Fig. 2. A comparison of prescribed power spectral density (PSD) with computed PSD from
the sample function obtained at the end of the iteration process (D = 6.5 m, dw = 20 m,
Hs = 4 m, JONSWAP PSD with γ = 3.3 and fp = 0.1 Hz).
limited number of field-test results [46] and, therefore, carries an in-
herent uncertainty. The uncertainty in the p-y equations can be attrib-
uted to uncertainties in the parameter ε50 and the empirical coefficient
J. In order to estimate the uncertainties in ε50 and J, p-y analyses were
performed to simulate 40 field pile-load tests subjected to static loads
(Table 3) reported in the literature, and the simulated and test results
were compared (by systematically matching the results through a trial
and error procedure) to obtain the mean and variance of ε50 and J. It
was found that J has a range of 0.15–1.0 with an estimated mean ( J ) of
0.57 and COV of 38%. J was assumed to follow lognormal PDF, as
determined from chi-square goodness of fit test (Fig. 3a). The parameter
ε50 has a range of 0.001–0.025 with an estimated mean (ε50 ) of 0.01 and
COV of 63%, and also follows a lognormal probability distribution
(Fig. 3b). Three mean values undrained shear strength (su ), namely 25
kPa, 50 kPa and 100 kPa are considered to investigate the effect of soil
strength. The selected values of undrained shear strength represent soft,
medium, and stiff clayey soils [74–76].
A three-bladed 5 MW variable speed OWT is considered for the
analysis. The pertinent geometric parameters for 5 MW OWT are col-
lected from the literature [12,77], which are summarized in Table 4.
The operational frequency (fr) of the OWT is 0.12–0.22 Hz. The OWT is
assumed to be installed with 20 m depth of water hw. The height of
tower Ht is taken as 90 m from the mean sea level. Three different
diameters of monopile and tower D = 6 m, 7 m and 8 m, are selected to
investigate the effect of pile diameter on the probabilistic response
statistics of OWT. In all these cases, the wall thickness of monopole and
tower is assumed to be 60 mm. The effect of embedded length of
monopole Lp is also investigated considering three different values of Lp,
40 m, 50 m and 60 m.
5. Estimation of probability of failure
From a probabilistic design point of view, OWTs may be considered
to be unserviceable if the mean maximum rotation of monopile at
mudline exceeds the specified limit. The probability of failure of OWT is
estimated in this study for the SLS criterion assuming maximum al-
lowable monopile rotation equal to ± 0.5⁰ at the seabed level. The First
Order Reliability Method (FORM) is used to determine the probabilities
of failure [24]. The limit state function for SLS can be mathematically
expressed as,

175
S. Haldar et al. Soil Dynamics and Earthquake Engineering 105 (2018) 171–183

Table 3
The database of field pile load test results.

Casea L (m) D (m) e (m) su (kPa) EpIp (kN-m2) Pile Typeb Reference J ε50

1 12.81 0.324 0.31 14.40 3.13 × 104 Steel pipe pile (Sabine) Matlock [46] 0.60 0.007
2 12.81 0.324 0.06 38.30 3.13 × 104 Steel pipe pile (Single pile) Matlock [46] 0.25 0.012
3 5.54 0.114 0.81 37.50 6.23 × 102 Open ended pipe pile (P1) Gill (1969) 0.40 0.010
4 6.22 0.219 0.81 34.30 5.45 × 103 Open ended pipe pile (P2) Gill (1969) 0.60 0.008
5 5.08 0.324 0.81 34.30 3.13 × 104 Open ended pipe pile (P3) Gill (1969) 0.30 0.010
6 8.13 0.406 0.81 34.30 4.85 × 104 Open ended pipe pile (P4) Gill (1969) 0.30 0.010
7 35.05 0.254 0.31 28.70 1.09 × 104 Cased concrete pile (St. Gabriel) Cappozoli (1968) 0.50 0.010
8 5.71 0.200 0.29 40.00 5.62 × 103 Steel pipe pile (P13) Wu et al. (1998) 0.30 0.009
9 14.00 0.500 0.72 40.00 9.79 × 104 Steel pipe pile (P17) Wu et al. (1998) 0.60 0.007
10 22.39 1.500 10.00 26.00 3.46 × 106 Steel pipe pile (Pile B) Nakai and Kishida (1982) 1.00 0.007
11 40.00 2.000 6.77 41.60 1.34 × 107 Steel pipe pile (Pile C) Nakai and Kishida (1982) 0.50 0.020
12 5.54 0.114 0.81 82.50 6.23×102 Open ended pipe pile Gill (1969) 0.70 0.008
13 5.08 0.324 0.81 64.50 3.13 × 104 Open ended pipe pile Gill (1969) 0.15 0.025
14 8.13 0.406 0.81 64.50 4.85 × 104 Open ended pipe pile Gill (1969) 0.50 0.015
15 15.20 0.244 0.38 43.50 1.41 × 104 Cased concrete pile (P1) Long et al. (2004) 0.60 0.010
16 15.20 0.244 0.27 43.50 1.41 × 104 Cased concrete pile (P2a) Long et al. (2004) 0.30 0.015
17 15.20 0.244 0.34 43.50 1.41 × 104 Cased concrete pile (P3) Long et al. (2004) 0.60 0.008
18 15.20 0.244 0.36 43.50 1.41 × 104 Cased concrete pile (P4) Long et al. (2004) 0.60 0.008
19 15.20 0.244 0.36 43.50 1.41 × 104 Cased concrete pile (P2b) Long et al. (2004) 0.70 0.007
20 15.20 0.244 0.38 43.50 1.41 × 104 Cased concrete pile (P6) Long et al. (2004) 0.60 0.007
21 15.20 0.244 0.25 43.50 1.41 × 104 Cased concrete pile (P10) Long et al. (2004) 0.50 0.007
22 15.20 0.244 0.41 43.50 1.41 × 104 Cased concrete pile (P11) Long et al. (2004) 0.30 0.008
23 13.12 0.273 0.31 73.40 1.40 × 104 Closed ended pile (Single pile) Brown et al. (1988) 0.60 0.004
24 16.50 0.406 1.00 44.10 5.14 × 105 Steel pipe pile (Single pile) Price and Wardle (1981) 0.50 0.004
25 9.10 0.305 0.00 50.00 1.93 × 104 Closed ended steel pile (Single pile) Rollins et al. (1998) 0.90 0.005
26 6.00 0.900 0.76 95.80 4.33 × 105 Concrete Drilled shaft Briaud et al. (1983) 0.90 0.004
27 4.50 0.900 0.76 95.80 4.33 × 105 Concrete Drilled shaft Briaud et al. (1983) 0.90 0.003
28 3.66 0.115 0.81 84.00 6.49 × 102 Open ended pipe pile (P1) Gill et al. (1970) 0.50 0.025
29 5.18 0.220 0.81 84.00 4.96 × 103 Open ended pipe pile (P2) Gill et al. (1970) 0.50 0.015
30 6.71 0.325 0.81 84.00 2.40 × 104 Open ended pipe pile (P3) Gill et al. (1970) 0.60 0.015
31 8.23 0.405 0.81 84.00 4.84 × 104 Open ended pipe pile (P4) Gill et al. (1970) 0.60 0.015
32 11.60 0.152 0.00 14.00 3.13 ×104 Steel pipe pile Matlock et al. (1980) 0.60 0.020
33 22.00 0.400 0.00 28.89 3.88 ×104 Reinforced Concrete pile (Pyramid building) Reuss et al. (1992) 0.60 0.020
34 22.00 0.400 0.00 48.00 3.88 × 104 Reinforced Concrete pile (Pyramid building) Reuss et al. (1992) 0.20 0.002
35 3.35 0.430 0.70 100.00 2.55×104 Steel pile (Bagnolet) Ruigrok (2010) 0.90 0.003
36 5.05 0.430 0.90 100.00 2.55×104 Steel pile (Bagnolet) Ruigrok (2010) 0.90 0.001
37 5.18 0.305 0.20 27.30 6.87×103 Steel pipe pile (Japan) Ruigrok (2010) 0.50 0.002
38 1.52 0.510 0.00 80.00 8.72×104 Concrete Drilled shaft Lutenegger and Miller (1993) 0.70 0.007
39 2.44 0.510 0.00 80.00 8.72×104 Concrete Drilled shaft Lutenegger and Miller (1993) 0.95 0.003
40 1.52 0.610 0.00 80.00 1.79×105 Concrete Drilled shaft Lutenegger and Miller (1993) 0.45 0.009

a
Case 1- 24, collected from Guo [81].
b
Pile location or pile number reported in literature is in parenthesis.

Fig. 3. Cumulative frequency from field tests and

fitted log-normal distribution of (a) J and (b) ε50.

gSLS (θpile, max ) = 0.5° − θpile, max
where θpile, max is the absolute maximum value of monopile rotation at
the sea bed (mudline). Failure of OWT performance occurs when g
(.) < 0 in Eq. (11). The probability of failure is estimated from the
(11) probability of limit state violation and is based on the second moment
properties (i.e., mean and standard deviation) and probability dis-
tribution of each of the random variables. The mean and standard de-
viation of θpile, max are estimated from the M-C simulations. If θpile, max is
assumed to be a log-normal random variable, then the probability of

176
S. Haldar et al. Soil Dynamics and Earthquake Engineering 105 (2018) 171–183

Table 4 Table 5
Summary of OWT parameters and input probabilistic data for parametric study. The fundamental frquency of OWT from deterministic analysis with Lp = 40 m and tw =
60 mm.
Property Value
su (kPa) D (m) fn (deterministic) (Hz)
Wind turbine
Rotor speed (fr) 0.12–0.22 Hz 20 6.5 0.270
Rotor diameter (Dr) 116 m 7.5 0.276
Rotor-Nacelle Assembly (RNA) 350 t 8.5 0.277
mass (MRNA) 50 6.5 0.297
RNA mass moment of inertia 4 × 107 kg-m2 7.5 0.326
(JRNA) 8.5 0.347
Water depth (dw) 20 m 100 6.5 0.311
Tower height (Ht) 90 m 7.5 0.355
Monopile length (Lp) 40, 50 and 60 m 8.5 0.392
Diameter of monopile and tower 6.5, 7.5 and 8.5 m
(D)
Thickness of monopole and tower 60 mm
(tw)
6.1. Deterministic analysis
Density of tower and monopile 7850 kg/m3
material Deterministic analyses were carried out for a 5 MW OWT structure
Young’s modulus of tower and 2.1×105 MPa considering su, ε50, J and γ′, U10, z0 as deterministic variables with their
monopile (Ep)
values the same as their corresponding mean values used in the prob-
Yield strength of steel 235 MPa
Soil abilistic analysis. The deterministic wave load is estimated using
Mean undrained shear strength 25, 50 and 100 kPa Morison’s equation in the time domain [58]. The loading frequencies
(su ) used for wind and wave loads are 0.22 Hz and 0.1 Hz, respectively. The
COVsu 20, 30% and 40% fundamental frequencies of OWT, maximum values of rotation at the
Probability distribution of su Lognormal
mudline (θpile, max ) and fatigue life are estimated for three different D
ε50 0.02 (su = 25 kPa), 0.008 (su = 50 kPa)
and 0.006 (su = 100 kPa) and undrained shear strength of clay (su) for U10 = 7 m/s, 15 m/s and
COVε50 63% 25 m/s. The corresponding results are presented in Tables 5, 6.
Probability distribution of ε50 Lognormal The fundamental frequencies of OWT are about 0.27 Hz, 0.276 Hz
J 0.5 and 0.277 Hz for D = 6.5 m, 7.5 m and 8.5 m, respectively, when Lp =
COVJ 38% 40 m and su = 20 kPa (Table 5). The fundamental frequencies are
Probability distribution of J Lognormal
10 kN/m3
greater than the highest operational frequency (1P) of OWT and less
Mean submerged unit weight (γ ′)
COVγ ′ 5% than lowest value of blade passing frequency (3P). Hence, the design of
Probability distribution of γ ′ Normal the turbine follows the soft-stiff approach [1,16,21,22]. A similar ob-
Wind load servation is made for su = 50 kPa and 100 kPa as well. Typical results of
U10 7, 15 and 25 m/s θpile, max and fatigue life for the deterministic soil properties and loads
Turbulence intensity (I) 0.2 are reported in Table 6 for su = 20 kPa and Lp = 40 m. The OWT
Probability distribution of U10 Weibull, Lognormal and Gamma
Mean terrain roughness ( z 0 ) 0.0001 m
responses for other sets of deterministic parameters are not presented
COVz 0 10% for the sake of brevity. An overall decrease in θpile, max is observed with
Probability distribution of z0 Normal an increase in diameter. This happens because the flexural rigidity of
Wave load the pile section increases with the diameter and correspondingly the
Significant wave height, Hs (m) 4m fundamental frequency of the OWT system increases and moves away
Wave frequency (f) 0.04–0.5 Hz for JONSWAP and PM
spectrum
from the 1P frequency. The fatigue life of OWT improves with monopile
diameter because the maximum bending stress in the monopile section
decreases. The responses from deterministic analysis are also compared
with probabilistic analysis in the subsequent sections.
failure is expresses as

pf , SLS = P(θpile, max ≥ 0.5°) = 1 − P(θpile, max ≤ 0.5°) 6.2. Impact of variability in soil

⎛ ln 0.5° − θln pile, max ⎞ The mean values of the fundamental frequency fn , maximum
= 1 − Φ⎜ ⎟
σln θpile, max (12) monopile rotation at seabed (θpile, max ) and fatigue life (log10 N ), and the
⎝ ⎠
coefficients of variation of fn (COVfn ) andθpile, max (COVθpile, max ), and fa-
where Φ(•) is the cumulative standard normal distribution. In Eq. (12), tigue life (COVlog10N ) and their probability density functions are
ln(θpile, max ) − 0.5σln2 θpile, max , σln θpile, max = ln(1 + COVθ2pile, max ) .
θln pile, max =
Table 6
OWT responses for deterministic soil and loading parameters with su = 20 kPa, Lp =
40 m, tw = 60 mm.
6. Results
U10 (m/sec) D (m) θpile, max(deterministic) (deg.) log10 N(deterministic)
The functional performance of an OWT is investigated for the se-
lected range of parameters considering the uncertainties in soil and 7 6.5 0.083 9.900
7.5 0.066 10.581
loading parameters. The fatigue life of OWT structure is estimated
8.5 0.055 11.178
based on the maximum bending stress in pile section caused by the 15 6.5 0.143 8.461
applied loads [79], as described in Appendix A. The effects of Lp, PSD of 7.5 0.110 9.129
vertical sea surface elevation, and probability distribution of U10 on the 8.5 0.094 9.739
fundamental frequency of OWT fn, θpile, max , fatigue life, and probability 25 6.5 0.325 6.648
7.5 0.233 7.315
of failure based on SLS are also examined. The statistics of these re-
8.5 0.192 7.930
sponses were obtained by sample averaging.

177
S. Haldar et al. Soil Dynamics and Earthquake Engineering 105 (2018) 171–183

less for stiff soil (i.e., for su = 100 kPa) than for soft soil (i.e., for su =
20 kPa) (cf. Fig. 5(c)-(f)). COVfn increases with an increase in D/tw for
all the cases shown. The influence of COVsu on design approaches of
OWT is evident from Fig. 6. The operating frequency (1P) of 5 MW
OWT is considered to be 0.12–0.22 Hz, and the corresponding blade
passing frequency (3P) is 0.36–0.66 Hz, as shown in Fig. 6. Based on the
deterministic analysis, the fundamental frequency of OWT system lies
in soft-stiff zone for su = 20 kPa, D = 6.5 m, Lp = 40 m (cf. Table 7).
Because of the variability in undrained shear strength, the fundamental
frequency moves toward the 1P range and the risk of accidental re-
sonance increases as COVsu increases.

6.2.2. Impact on the monopile response and fatigue life of OWT

Realizations of the time history of rotation of monopile θpile at
mudline obtained from the Monte Carlo simulations are shown in Fig. 7
for COVsu = 40%, su = 20 kPa, D = 8.5 m, Lp = 40 m, U10 = 7 m/s,
Fig. 4. Probability density of the fundamental frequency of OWT from MC simulations,
turbulence intensity = 10%, PDF for U10 as lognormal distribution, and
and fitted normal distribution for D = 7.5 m, COVsu = 30%, and su = 50 kPa. PM power spectral density for vertical sea surface elevation. In all these
cases, the rotor frequency is assumed to be 0.22 Hz. In addition, the
corresponding mean and deterministic responses are also shown. When
the soil properties vary randomly, monopile response also vary sig-
Table 7
The mean fundamental frquency of OWT from probabilistic analyses.
nificantly. The probability densities of θpile, max and log10 N are evaluated
and it is found, using chi-square goodness of fit tests, that there is a
su (kPa) D (m) fn (Hz) good agreement between the simulated data and assumed log-normal
log (%) distribution for θpile, max , and between the simulated data and assumed
normal distribution for log10 N . Fig. 8 shows a typical histogram and
20 30 40 corresponding fitted distribution of θpile, max for D = 8.5 m, U10 = 7 m/
s, COVsu = 40% and su = 20 kPa. Similar trends were found for the
20 6.5 0.269 0.256 0.239
7.5 0.271 0.256 0.243 other cases as well. Fig. 9 shows a typical histogram and the corre-
8.5 0.271 0.257 0.244 sponding fitted distribution of log10 N for D = 8.5 m, U10 = 7 m/s,
50 6.5 0.294 0.291 0.288 COVsu = 40% and su = 20 kPa. Similar observations were found for the
7.5 0.321 0.315 0.310
other cases investigated in this study. The sample mean values of
8.5 0.338 0.329 0.321
100 6.5 0.309 0.307 0.306
maximum rotation of monopile at mudline (θpile, max ) and fatigue life
7.5 0.352 0.349 0.347 (log10 N ) were obtained for the selected range (Table 4) of D, su and
8.5 0.388 0.384 0.381 COVsu . The values of θpile, max and log10 N are normalized with respect to
the corresponding deterministic values (θpile, max /θpile, max (deterministic) and
log10 N /log10 Ndeterministic respectively) given in Table 6 and plotted as
functions of normalized D (i.e., D/tw) for U10 = 7 m/s, 15 m/s and
reported in the following sections. 25 m/s, COVsu = 40% and forsu = 20 kPa (cf. Fig. 10(a)-(b)). θpile, max
increases from their corresponding deterministic values with an in-
6.2.1. Impact on the fundamental frequency of OWT crease in U10 and D/tw. This increase is as high as 67% for θpile, max from
The probability density of the fundamental frequency of OWT was the corresponding deterministic values, when D = 8.5 m, U10 = 7 m/s,
estimated and it was found, using chi-square goodness of fit tests, that su = 20 kPa and COVsu = 40%. This increase in OWT rotation occurs
there was a good agreement between the simulated results and assumed because the fundamental frequency of OWT in the probabilistic case is
normal distribution. Fig. 4 shows a typical histogram and the corre- more close to the operational frequency of turbine (1P) than for the
sponding fitted distribution of the fundamental frequency. Fig. 4 is deterministic case. The mean fatigue life for the probabilistic cases is
plotted for D = 6.5 m, COVsu = 20% and su = 20 kPa. Similar trends found to be marginally less than the corresponding deterministic cases.
were found for the other values of D, su and COVsu . The sample mean of An overall reduction in mean fatigue life is observed with increase in
the fundamental frequency fn of OWT are presented in Table 7 for the U10 value because of increased stress in the monopile-tower sections.
selected ranges of D, su and COVsu . The fundamental frequency is nor- Similar trends were observed for other COVsu and su values. Fig. 11(a)-(b)
malized with respect to the corresponding deterministic result show the COVs of θpile, max (COVθpile, max ) and log10 N (COVlog10N ) as func-
( fn / fn (determinisitc) ) and plotted as functions of normalized D (i.e., D/tw) tions of D/tw ratios for U10 = 7 m/s, 15 m/s and 25 m/s, COVsu = 40%
for COVsu = 20%, 30% and 40% and for su = 20 kPa, 50 kPa and and forsu = 20 kPa. COVθpile, max and COVlog10N increase with an increase
100 kPa (cf. Fig. 5(a)–(c)). It is found that fn / fn (determinisitc) decreases in U10 and D/tw. The effect of COVsu and su on θpile, max , log10 N ,
with an increase in COVsu (Fig. 5(a)-(c)). Increase in COVsu increases the COVθpile, max and COVlog10N are presented in Table 8 for U10 = 7 m/s and
weak pockets in the soil which reduces the overall soil stiffness because D = 6.5 m. The values of θpile, max decreases with increase in su , as
of which the fundamental frequency of the OWT system decreases from expected, while log10 N increases because of increase in su . COVθpile, max
the deterministic results. This effect is more pronounced in the case of and COVlog10N decrease with increase in su and increases marginally with
soft soil (i.e., for su = 20 kPa). It was observed that the reduction in fn increase in COVsu (Table 8).
with respect to the corresponding deterministic values is more for large The effect of U10 and D/tw on the probabilities of failure for SLS is
diameter monopiles (e.g., D = 8.5 m) than for small diameter mono- apparent from Fig. 12. The SLS based failure probability increases with
piles (e.g., D = 6.5 m) for the selected values of Su and COVsu (Fig. 5(a)- increase in U10 because of increase in θpile, max and COVθpile, max (Fig. 12).
(c)). Fig. 5(d)-(f) show the COVfn as functions of D/tw for different va- The effect of D/tw is also evident from Fig. 12. The probabilities of
lues of su and COVsu . COVfn increases with increase in COVsu for all the su failure for both the failure modes decrease as D/tw increases – this is
values. However, the increase in COVfn for a particular value of COVsu is because of a decrease in θpile, max .

178
S. Haldar et al.
Fig. 5. Normalized mean fundamental frequency of OWT versus normalized diameter of monopile for Lp = 40 m, tw = 60 mm: (a) su = 20 kPa, (b) su = 50 kPa, (c) su = 100 kPa; and COV
of fundamental frequency of OWT versus normalized diameter of monopile for: (d) su = 20 kPa, (e) su = 50 kPa, and (f) su = 100 kPa.
Fig. 6. Deterministic and probabilistic design approaches of 5 MW OWT for D = 6.5 m,
su = 20 kPa.
6.3. Impact of monopile embedment depth on the probabilistic responses
In order to investigate the effect of monopile length on θpile, max ,
log10 N , COVθpile, max , and COVlog10N the responses are estimated for Lp =
40 m, 50 m and 60 m, for su = 20 kPa, COVsu = 40%, and D = 6.5 m.
The wind load is estimated considering U10 = 7 m/s with lognormal
probability distribution. The wave load is estimated assuming the PM
spectra for vertical sea surface displacement. The results are presented
in Table 9. θpile, max for Lp = 60 m decreases by about 13% from that of
the case with Lp = 40 m. The effect of Lp on the mean fatigue life is
found to be rather negligible. COVθpile, max and COVlog10N are also pre-
sented in Table 9, which shows that the case with Lp = 40 m results in
179
Soil Dynamics and Earthquake Engineering 105 (2018) 171–183
Fig. 7. Monopile rotation at mudline from Monte Carlo simulation along with the cor-
responding mean and deterministic responses for COVsu = 40%, su = 20 kPa, D = 8.5 m,
U10 = 7 m/s with 10% COV and lognormal distribution and PM spectra.
greater COVθpile, max compared with the case with Lp = 60 m. This is
because the longer embedment depth increases the averaging effect and
reduces the uncertainty in OWT responses, which is also reported by
Depina et al. [80]. Further, it was observed that Lp has a marginal
impact on COVlog10N (cf. Table 9). The effect of Lp on the probability of
failure for SLS is also presented in Table 9 — the failure probability
decreases with increase in Lp.
6.4. Impact of PSD of vertical sea surface displacement on the probabilistic
responses
To investigate the effect of various power spectral density functions
on the probabilistic responses and fatigue life of OWTs, the wave load is
estimated considering Pierson-Moskowitz (PM) and JONSWAP spectral
S. Haldar et al.
Fig. 8. Probability density from M-C simulation and fitted lognormal distribution for
θpile, max when D = 8.5 m, U10 = 7 m/s, COVsu = 40% and su = 20 kPa.
Fig. 9. Probability density of fatigue life from M-C simulation and fitted normal dis-
tribution when U10 = 7 m/s, D = 8.5 m, COVsu = 40% and su = 20 kPa.
density functions (PSDs). For these simulations, the wind load is esti-
mated considering U10 = 7 m/s with lognormal probability distribu-
tion. The mean responses and fatigue life of OWT are estimated for su =
20 kPa, COVsu = 40%, D = 6.5 m and Lp = 40 m. The values of θpile, max
and log10 N are presented in Table 9. It is observed that different PSDs of
vertical sea surface displacement have marginal impact on the θpile, max
180
Soil Dynamics and Earthquake Engineering 105 (2018) 171–183
and log10 N . COVs of θpile, max and log10 N were also estimated for the two
different PSD functions, which are presented in Table 10. Table 10
shows that the different PSD functions have marginal impact on
COVθpile, max and COVlog10N . The probability of failure for SLS is also
presented in Table 10 and it is observed that the PSD of vertical sea
surface displacement has marginal impact on the failure probability.
6.5. Impact of probability distribution of U10 on the probabilistic responses
To examine the effects of the probability distribution of U10 on
θpile, max and log10 N , and the COVs of θpile, max and log10 N , three prob-
ability density functions (PDFs) of U10, namely, Gamma, Log-normal,
and Weibull distributions, are selected for U10 = 7 m/s, su = 20 kPa,
COVsu = 40%, D = 6.5 m, and Lp = 40 m. The wave load is estimated
assuming the PM spectra for vertical sea surface displacement. θpile, max
and log10 N are presented in Table 11 for the wind load estimated based
on the three PDFs. It is found that the different PDFs of U10 have
marginal impact on the mean responses and fatigue life of OWT. Dif-
ferent PDFs of U10 have a marginal impact on COVθpile, max and COVlog10N .
The probability of failure for SLS for different PDFs of U10 is also pre-
sented in Table 11, and it is observed that different PDFs have marginal
impact on the failure probability.
7. Conclusions
The effects of uncertainty and variability of soil properties, and of
the uncertainties in wind and wave load estimations on the responses
and fatigue life of monopile supported offshore wind turbines (OWTs)
are investigated. The performance of the OWTs was examined based on
the failure probability in terms of serviceability and fatigue limit states.
The soil was modeled using nonlinear cyclic p-y relationships. The
tower and monopile were assumed to behave as Euler-Bernoulli beams.
A parametric study was performed using Monte Carlo simulations in
which the statistics of the maximum monopile rotation at mudline,
maximum tower rotation at top, and fatigue life were investigated for
different mean undrained shear strengths of soil and their coefficients of
variation. Based on the study, the following conclusions can be made:
1. The variability in soil properties creates weak zones within the soil
because of which the soil stiffness reduces and the mean maximum
rotation of monopile at the mudline and tower top increases from
their corresponding deterministic values. The mean fatigue life of
OWT decreases because of soil variability from the corresponding
deterministic value.
2. The maximum rotation of monopile at the mudline and tower top
were found to follow log-normal distributions, and the fatigue life
followed a normal distribution. The fundamental frequency of OWT
Fig. 10. Non-dimensional mean responses for
COVsu = 40% and su = 20 kPa for different D/tw ra-
tios: (a) maximum monopile rotation at mudline and
(b) fatigue life.
S. Haldar et al. Soil Dynamics and Earthquake Engineering 105 (2018) 171–183

Fig. 11. COVs of (a) maximum monopile rotation at

mudline and (b) fatigue life for different D/tw ratios,
when COVsu = 40% and su = 20 kPa.

Table 8 Table 10
Mean and COV of responses of OWT for various Su and COVsu values. Impact of power spectral density functions (PSDs) of vertical sea surface displacement on
the mean and COV of responses of OWT (D = 6.5 m, tw = 60 mm, Su = 20 kPa, Lp =
log (%) su (kPa) θpile, max (deg.) COVθ pile, max (%) log10 N COVlog10N (%) 40 m, COVsu = 40%, and U10 = 7 m/s).

20 0.105 30.038 9.497 2.129 PSD θpile, max COVθ pile, max (%) pf ,SLS log10 N COVlog10N (%)
20 50 0.084 28.623 10.622 2.146 (deg.)
100 0.084 22.222 11.182 1.427
20 0.106 31.737 9.522 2.203 PM 0.107 32.46 2.34 × 10−7 9.54 2.28
30 50 0.085 29.689 10.624 2.031 JONSWAP 0.121 32.91 2.33 × 10−6 9.24 2.16
100 0.083 27.166 11.141 1.543
20 0.107 32.028 9.536 2.280
40 50 0.085 31.368 10.597 2.064
100 0.084 27.357 11.146 1.520
Table 11
Impact of probability distribution functions (PDFs) of U10 on the mean and COV of re-
sponses of OWT (D = 6.5 m, tw = 60 mm, Su = 20 kPa, COVsu = 40%, and U10 = 7 m/s).

PDF θpile, max COVθ pile, max (%) pf ,SLS log10 N COVlog10N (%)
(deg.)

Gamma 0.107 32.03 1.81 × 10−7 9.51 2.28

Lognormal 0.107 32.46 2.34 × 10−7 9.54 2.28
Weibull 0.107 33.05 3.78 × 10−7 9.53 2.28

undrained shear strength of soil. This effect was more pronounced in

the case of soft soils. The monopile diameter greatly affected the
mean fundamental frequency of OWT. The reduction in mean fun-
damental frequency from the corresponding deterministic value
Fig. 12. Probability of failure of OWT for serviceability limit state criterion increased with an increase in diameter of monopile.
(θpile, max ≤ ± 0.5°) when COVsu = 40%, Lp = 40 m, tw = 60 mm and su = 20 kPa. 4. The coefficient of variation of maximum rotation of monopile at the
mudline and tower top increases with an increase in the wind speed
value and monopile diameter. However, the coefficient of variation
Table 9
of fatigue life changes marginally because of an increase in the wind
Impact of Lp on the mean and COV of responses of OWT (D = 6.5 m, tw = 60 mm, Su = 20 speed and monopile diameter.
kPa,COVsu = 40%, and U10 = 7 m/s). 5. The mean maximum rotation of monopile at the mudline and tower
top was found to decrease because of an increase in the embedment
Lp (m) θpile, max COVθ pile, max (%) pf ,SLS log10 N COVlog10N (%)
length of monopile. The mean fatigue life was found to be margin-
(deg.)
ally influenced by the embedment length of monopile. The coeffi-
40 0.107 32.46 2.34 × 10−7 9.536 2.28 cients of variation of maximum rotation of monopile at the mudline
50 0.096 26.85 7.98 × 10−11 9.518 2.26 and tower top decreased with increase in the embedment length of
60 0.093 23.24 4.24 × 10−14 9.501 2.24 monopile. This aspect is important for designing the minimum
embedment length of monopile, which is used to ensure the stability
of the OWT structure.
6. The effects of various power spectral density functions of the wave
was found to follow a normal distribution. surface displacement on the OWT means and coefficients of varia-
3. The mean fundamental frequency of the OWT system was found to tion of the different responses and fatigue life were found to be
decrease from its deterministic value because of the variability of marginal. Various probability density functions of wind speed have

181
S. Haldar et al.
a marginal impact on the means and coefficients of variations of
responses and fatigue life.
7. The probabilities of failure based on the SLS and FLS are sig-
nificantly influenced by the wind speed. It was observed that, at low
wind speed, the SLS criterion governs the design, while, at higher
wind speed, the FLS criterion governs the design.
Appendix A. The fatigue in tubular welded joints in offshore wind turbine is assessed using the following [79]
log10 N = log10 a − m log10 (Δσ (tw / tref ) χ )
where and N is the predicted number of cycles to failure (or fatigue life) for stress range Δσ in MPa (i.e., the difference between maximum and
minimum bending stress in a stress cycle), m is the negative slope of fatigue resistance curve of the monopile material available in DNV [21], log10 a is
the intercept of the log10N axis, tref is the reference thickness of welded connection, tw is the thickness through which a crack growth is likely — this is
the wall thickness of the tower and monopile, and χ is the thickness exponent which is considered to be 0.25. A reference thickness tref = 32 mm is
considered for weld in tubular joint in this study, as suggested in DNV- RP-C203 [79]. The value of tw = tref is used, when tw < tref [79]. Further,
log10 a = 11.7 and m = 3 for N < 106 and log10 a = 15.6 and m = 5 for N > 106 [79] is considered in this study according to DNV [79].
References
[1] Lombardi D, Bhattacharya S, Wood DM. Dynamic soil-structure interaction of
monopile supported wind turbines in cohesive soil. Soil Dyn Earthq Eng
2013;49:165–80.
[2] Zaaijer M.B., Henderson A.R. Review of current activities in offshore wind energy,
Paper No. 2004-ARH-02; 2004.
[3] Madsen PH, Natarajan A. Challenges and prospects for wind energy to attain 20%
grid penetration by 2020 in India. Curr Sci 2011;101:35–42.
[4] Westgate Z.J., Dejong J.T. Geotechnical considerations for offshore wind turbines;
2005.
[5] Ram B. Energy from offshore wind, Offshore Technology Conference, Houston, TX,
U.S.A; 2006.
[6] Junginger M. Learning in renewable energy technology development [Ph.D.
Thesis]. The Netherlands: Utrecht University; 2005.
[7] Yun TS, Lee JS, Lee SC, Kim YJ, Yoon HK. Geotechnical issues related to renewable
energy. KSCE J Civil Eng 2011;15:635–42.
[8] Harte M, Basu B, Nielsen SRK. Dynamic analysis of wind turbines including soil-
structure interaction. Eng Struct 2012;45:509–18.
[9] Zaaijer MB. Foundation modelling to assess dynamic behaviour of offshore wind
turbines. Appl Ocean Res 2006;28:45–57.
[10] Moskowitz L. Estimates of the power spectrums for fully developed seas for wind
speeds of 20 to 40 knots. J Geophys Res 1964;69:5161–79.
[11] LeBlanc C. Design of offshore wind turbine support structure structures [Ph.D.
Thesis]. Denmark: Aalborg University; 2009.
[12] Tempel J, Molenaar DP. Wind turbine structural dynamics – a review of the prin-
ciples for modern power generation, onshore and offshore. 26. The Netherlands:
Wind Engineering, Delft University of Technology; 2002. p. 211–20.
[13] van Kuik GAM. Are wind turbines growing too fast? In: Proceedings of the European
wind energy conference and exhibition, Copenhagen, Denmark; 2001.
[14] NREL. National Renewable Energy Laboratory, Washington; 2004.
[15] Damgaard M, Ibsen LB, Andersen LV, Andersen JKF. Cross-wind modal properties of
offshore wind turbines identified by full scale testing. J Wind Eng Ind Aerodyn
2013;116:94–108.
[16] Arany L, Bhattacharya S, Macdonald J, Hogan SJ. Simplified critical mudline
bending moment spectra of offshore wind turbine support structures. Wind Energy
2014. http://dx.doi.org/10.1002/we.1812.
[17] Byrne B., Houlsby G.T. Assessing novel foundation options for offshore wind tur-
bines. World Maritime Technology Conference, London, U.K; 2006.
[18] DNVGL-ST-0126. Support structures for wind turbines. Det Norske Veritas AS;
2016.
[19] Lesny K., Wiemann J., Richwien W. Evaluation of pile diameter effects on soil pile
stiffness. In: Documentation der 7th German wind energy conference DEWEK;
2004.
[20] TAPS. Provisional type certification scheme for wind turbine generator systems in
India. New Delhi: Ministry of New and Renewable Energy; 2000.
[21] Bhattacharya S. Challenges in design of foundations for offshore wind turbines.
Engineering & Technology Reference; 2014. http://dx.doi.org/10.1049/etr.2014.
0041. [ISSN 2056-4007].
[22] Bhattacharya S, Nikitas N, Garnsey J, Alexander NA, Cox JA, Lombardi D, Wood
DM, Nash DFT. Observed dynamic soil-structure interaction in scale testing of
offshore wind turbine foundation. Soil Dyn Earthq Eng 2013;54:47–60.
[23] Fenton GA, Griffiths DV. Three-dimensional probabilistic foundation settlement. J
Geotech Geoenviron Eng-ASCE 2005;131(2):232–9.
[24] Carswell W, Arwade SR, DeGroot DJ, Lackner MA. Soil-structure reliability of off-
shore wind turbine monopile foundations. Wind Energy 2015;18(3):483–98.
[25] Andersen LV, Vahdatirad MJ, Sichani MT, Sorensen JD. Natural frequencies of wind
turbines on monopile foundations in clayey soils-a probabilistic approach. Comput
Geotech 2012;43:1–11.
182
Soil Dynamics and Earthquake Engineering 105 (2018) 171–183
Acknowledgements
This work was supported by a grant from the NSERC and York
University to the second author. The source of support is greatly ap-
preciated. The authors are also thankful to the anonymous reviewers for
their critical comments which have been very useful in improving the
work.
(A.1)
[26] DNV Cn-30.6. Structural reliability analysis of marine structures. Norway: DET
NORSKE VERITAS Classification; 1992.
[27] Arany L, Bhattacharya S, Macdonald J, Hogan SJ. Design of monopiles for offshore
wind turbines in 10 steps. Soil Dyn Earthq Eng 2017;92:126–52.
[28] Arany L, Bhattacharya S, Macdonald J, Hogan SJ. Simplified critical mudline
bending moment spectra of offshore wind turbine support structures. Wind Energy
2015;18:2171–97.
[29] Germanischer Lloyd (GL). Guideline for the certification of offshore wind turbines;
Hamburg, Germany; 2005.
[30] IEC 61400-1. Wind turbines- Part 1: design requirements. Geneva: International
Electrotechnical Commission; 2005.
[31] ABS. American Bureau of Shipping: Rules for Building and Classing Offshore
Installation; 1997.
[32] Yang H, Zhu Y, Lu Q, Zhang J. Dynamic reliability based design optimization of the
tripod sub-structure of offshore wind turbines. Renew Energy 2015;78:16–25.
[33] Cheng PW, van Bussel GJW, van Kuik GAM, Vugts JH. Reliability-based design
methods to determine the extreme response distribution of offshore wind turbines.
Wind Energy 2003;6:1–22.
[34] Negro V, Lopez-Gutierrez J-S, Esteban MD, Matutano C. Uncertainties in design of
support structures and foundations for offshore wind turbines. Renew Energy
2004;63:125–32.
[35] Mardfekri M, Gardoni P. Multi-hazard reliability assessment of offshore wind tur-
bines. Wind Energy 2014. http://dx.doi.org/10.1002/we.1768.
[36] Vahdatirad M.J., Andersen L.V., Ibsen L.B., Clausen J, Soresen J.D. Probabilistic
three-dimensional model of an offshore monopile foundation: Reliability based
approach, In proceedings of international conference on case histories in geo-
technical engineering, Chicago; 1 – 8; 2013.
[37] De RS. Offshore structural system reliability: wave-load modeling, system behavior
and analysis [M.Sc. Thesis]. USA: Stanford University; 1990.
[38] Søresen JD, Toft HS. Probabilistic design of wind turbines. Energies 2010;3:241–57.
[39] Tavner PJ, Xiang J, Spinato F. Reliability analysis for wind turbines. Wind Energy
2007;10:1–18.
[40] Hsu Y, Wu W-F, Chang Y-C. Reliability analysis of wind turbine towers. Procedia
Eng 2014;79:218–24.
[41] Agarwal P. Structural reliability of offshore wind turbines [Ph.D. Thesis]. USA: The
University of Texas at Austin; 2008.
[42] Arwade SR, Lackner MA, Gtigoriu MD. Probabilistic models for wind turbine and
wind farm performance. J Sol Energy Eng ASME 2011;133:1–9.
[43] Sørensen JD, Tarp-Johansen NJ. Reliability-based optimization and optimal relia-
bility level of offshore wind turbines. J Offshore Polar Eng 2005;15:1–6.
[44] Lee Y-S, Choi B-L, Lee JH, Kim SY, Han S. Reliability-based design optimization of
monopile transition piece for offshore wind turbine system. Renew Energy
2014;71:729–41.
[45] American Petroleum Institute (API). Petroleum and natural gas industries-specific
requirements for offshore structures. Part 4 – geotechnical and foundation design
considerations; 2011.
[46] Matlock H. Correlations for design of laterally loaded piles in soft clay, Offshore
Technology Conference. Dallas, Texas; 1970.
[47] Ashour M, Norris G, Pilling P. Lateral loading of a pile in layered soil using the stain
wedge model. J Geotech Geoenviron Eng ASCE 1998;124:303–15.
[48] Achmus M, Kuo Y, Rahman KA. Behavior of monopile foundations under cyclic
lateral load. Comput Geotech 2009;36:725–35.
[49] Bouzid DJA, Bhattacharya S, Dash SR. Winkler springs (p-y curves) for pile design
from stress-strain of soils: fe assessment of scaling coefficients using the mobilized
strength design concept. Geomech Eng 2013;5:379–99. http://dx.doi.org/10.
12989/gae.2013.5.5.379.
[50] Kuhn M. Dynamics and design optimisation of offshore wind energy conversion
systems [Ph.D. Thesis]. Delft University of Technology; 2001.
[51] DNV. Guidelines for design of wind turbines-DNV/Riso. U.S.A: Code of practice,
DNV; 2002.
S. Haldar et al.
[52] OpenSEES. Open System for Earthquake Engineering Simulation; 2007. 〈http://
opensees.berkeley.edu〉.
[53] Villalobos Jara FA. Model testing of foundations for offshore wind turbines [Ph.D.
Thesis]. University of Oxford; 2006.
[54] Morgan EC, Lackner M, Vogel RM, Baise LG. Probability distributions for offshore
wind speeds. Energy Convers Manag 2011;52:15–26.
[55] Masseran N, Razali AM, Ibrahim K, Zaharim A, Sopian K. The probability dis-
tribution model of wind speed over east Malaysia. Res J Appl Sci, Eng Technol
2013;6(10):1774–9.
[56] Deo M.C. Waves and structures, 〈http://www.civil.iitb.ac.in/~mcdeo/waves.
html〉.
[57] Fenton G.A., Paice G.M., Griffiths D.V. Probabilistic analysis of foundation settle-
ment. in: Proceedings of the ASCE uncertainity ’96 conference, uncertainty in the
geological environment: from theory to practice, Madison, Wisconsin, 651 – 665;
1996.
[58] DNV-RP-C205. Environmental conditions and environmental loads. DET NORSKE
VERITAS; 2010.
[59] Minciarelli F, Gioffre M, Grigoriu M, Simiu E. Estimates of extreme wind effects and
wind load factors: influence of knowledge uncertianities. Probabilistic Eng Mech
2001;16:331–40.
[60] Diniz SMC, Sadek F., Simiu E. Wind speed estimation uncertainities: effects of cli-
matological and micrometeorological parameters. In: Deodatis and P Spanos (eds.).
Proceeding of 4th stochastic mechanics conference; 2003. p. 169–78.
[61] Hart GC, Raggett JD, Huang S, Dow S. Structural design using a wind tunnel test
program and risk analysis In: Proceedings of the sixth international conference on
wind engineering]. The Netherlands: Elsevier Sc.; 1983. p. 27–33.
[62] Lombardi D. Dynamics of offshore wind turbines [M.Sc. Thesis]. University of
Bristol; 2010.
[63] Jamdade SG, Jamdade PG. Analysis of wind speed data for four locations in Ireland
based on Weibull distribution’s linear regression model. Int J Renew Energy Res
2012;2(3):451–5.
[64] Soukissian T., Karathanasi F., Falcieri F. Wind speed distributions in the Italian
coasts, International Conference on Renewable Energies and Power Quality,
Cordoba, Spain; 2014.
[65] Carta JA, Ramlrez P, Velazquez S. A review of wind speed probability distributions
183
Soil Dynamics and Earthquake Engineering 105 (2018) 171–183
used in wind energy analysis Case studies in the Canary Islands. Renew Sustain
Energy Rev 2009;13:933–55.
[66] LoBrano V, Orioli A, Ciulla G, Culotta S. Quality of wind speed fitting distributions
for the urban area of Palermo, Italy. Renew Energy 2011;36(3):1026–39.
[67] Li M, Zhang H, Guan H. Study of offshore monopile behavior due to ocean waves.
Ocean Eng 2011;38:1946–56.
[68] Borgman LE. A statistical theory for hydrodynamic forces on objects: wave Res. Rep.
HEL 9 - 6. California, USA: Hydraulic Eng. Lab., University of California Berkeley;
1965.
[69] Popeseu R, Deodatis G, Prevost JH. Simulation of homogeneous nonGaussian sto-
chastic vector fields. Prob Eng Mech 1998;13(1):1–13.
[70] Lumb P. Application of statistics in soil mechanics. In: Lee IK, editor. Soil me-
chanics—new horizons. Elsevier; 1974.
[71] Lacasse S, Nadim F. Uncertainties in characterising soil properties. Proc Uncertain
Geol Environ 1996;96:49–73.
[72] Phoon K-K, Kulhawy FH. Characterization of geotechnical variability. Can Geotech
J 1999;36(4):612–24.
[73] Baecher GB, Christian JT. Reliability and statistics in geotechnical engineering.
London and New York: John Wiley and Sons; 2003.
[74] Das BM. Advance soil mechanics. 3rd eds. Taylor and Francis, Taylor and Francis e-
Library; 1983.
[75] Terzaghi K, Peck RB, Mesri G. Soil mechanics in engineering practice. 3rd ed. John
Wiley and Sons; 2010. [ISBN: 978-81-265-2381-8].
[76] Stewart DP. User manual PYGMY. University of Western Australia; 2000.
[77] NREL. National Renewable Energy Laboratory, Washington; 2004.
[78] D. Pappusetty, M. Pando. Finite Element Analyses of Offshore Monopile Deflection
Accumulation under Harmonic Loading. Proceedings from Soil Behavior
Fundamentals to Innovations in Geotechnical Engineering; 585-596; 2014.
[79] DNV-RP-C203. Fatigue design of offshore steel structures. DET NORSKE VERITAS;
2010.
[80] Depina I, Le TMH, Eiksund G, Benz T. Behavior of cyclically loaded monopile
foundation for offshore wind turbines in heterogeneous sands. Comput Geotech
2015;65:266–77.
[81] Guo WD. Simple model for nonlinear response of 52 laterally loaded piles. J
Geotech Geoenviron Eng, ASCE 2013;139(2):234–52.