Ocean Engineering 287 (2023) 115845

Contents lists available at ScienceDirect

Ocean Engineering
journal homepage: www.elsevier.com/locate/oceaneng

Study on bearing performance and reliability of modified monopiles for

OWTs using deterministic and probabilistic analysis methods
Qilin Yin , Wei Wang , Jinjin Zhai *
School of Naval Architecture & Ocean Engineering, Jiangsu University of Science and Technology, Zhenjiang, 212003, China

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

Handling Editor: Prof. A.I. Incecik Existing studies mainly focus on the bearing capacity of monopiles, with less emphasis on reliability analysis,
which is an important aspect of the performance of offshore wind turbine (OWT) foundations. This paper in­
Keywords: vestigates the bearing capacity and reliability of traditional and hybrid monopiles using deterministic and
Hybrid monopile probabilistic analyses. The three-dimensional finite element model (FEM) considering the structure-soil inter­
Bearing performance and reliability
action is established and its accuracy is verified through comparisons with existing research findings. The re­
Environmental parameter correlation
sponses of monopiles under wind and wave loads are compared in detail, including the rotation angle at the
FEM-BPNN-RSM coupled method
Numerical analysis mudline, structural stress, and soil displacement. The environmental design parameters are obtained based on
the 30-year annual extreme data, and the correlation between these environmental parameters is considered to
obtain a large amount of random environmental parameter data. Then, the back propagation neural network
(BPNN) and the response surface method (RSM) are combined to construct the failure response surface. Research
shows that the reliability index of the plated monopile does not improve significantly, while the skirted-
monopiles perform much better than the traditional monopile. The proposed FEM-BPNN-RSM coupled meth­
odology provides a more comprehensive understanding of the monopile bearing capacity, and can be used for the
future monopile structure design of OWTs.

1. Introduction 2018). Therefore, the study of wind turbine foundation structures is

critical to the overall economics of OWTs.
Wind energy is one of the world’s most promising and rapidly In certain offshore regions, the bedrock lies shallowly beneath the
growing renewable energy sources. Due to the more stable and stronger seabed. Conventional monopile foundations installed in such areas
wind conditions at sea, coupled with less visual impact, a large number require rock sockets at the top of the bedrock, significantly escalating
of offshore wind farms have been constructed and are currently under construction complexity and costs, consequently restraining wind power
construction (Ahmed and Hawlader, 2016). In recent years, China has development in these locations. One effective method of increasing the
vigorously pursued offshore wind power as part of its commitment to bearing capacity of monopile foundations is to enhance their design and
achieving “dual carbon” goals, resulting in a rapid surge in installed increase their diameter (Wang et al., 2018). Enhanced monopile foun­
capacity. The monopile foundation remains one of the most commonly dations often feature skirts, wing-like structures, and other additions
used foundation types in offshore wind power engineering, preferred for around the traditional circular pile foundations (Wang et al., 2020; Ma
its simplicity and ease of construction. Globally, approximately 75% of and Yang, 2020). This foundation design demonstrates excellent per­
the offshore wind turbines (OWTs) are equipped with monopile foun­ formance in wind resistance, wave resistance, and tilting resistance,
dations. These foundations typically consist of steel components with thereby enhancing the overall stability of OWTs. Current research on the
large diameters ranging from 3 to 8 m and are typically installed at bearing performance of enhanced monopile foundations mainly centers
shallow water depths of up to 40 m (Wu et al., 2019). As the foundation on optimization design, structural analysis, and experimental in­
structure accounts for a significant portion of the offshore wind farm vestigations. Regarding optimization design, researchers explore ways
construction costs, the expenses rise proportionally with increasing to improve the bearing capacity of enhanced monopile foundations and
water depth, at times reaching up to 30% (Carswell, 2015; Kim and Kim, continually study and design new types of enhanced monopiles. For

* Corresponding author.
E-mail address: [email protected] (J. Zhai).

https://doi.org/10.1016/j.oceaneng.2023.115845
Received 31 May 2023; Received in revised form 28 August 2023; Accepted 10 September 2023
Available online 16 September 2023
0029-8018/© 2023 Elsevier Ltd. All rights reserved.
Q. Yin et al. Ocean Engineering 287 (2023) 115845

example, Yang et al. (2018) employed finite element analysis to inves­ Table 1
tigate the influence of solid and crushed stone wheels on foundation Detailed parameters of NREL-5MW.
performance, providing a benchmark for preliminary design of Parameters Values
enhanced monopile friction wheel foundations. Chen et al. (2020) pro­
Power/MW 5
posed an improved monopile foundation consisting of a traditional Blade projection area/m2 551
monopile and a broad, shallow bucket structure. Dynamic response re­ Rotor diameter/m 126
sults show that this foundation is most effective under the combined Cut-in, cut-out, rated wind speed/m⋅s− 1
3, 25, 11.4
loads of waves and water flow, effectively suppressing rotation and Hub mass/t 350
Hub height/m 87.6
lateral displacement of the base. Arany et al. (2017) proposed a method Monopile diameter/m, thickness/m 7, 0.06
to simplify the design of monopiles and demonstrated its scientific Tower thickness/m 0.019–0.027
soundness of the method through examples. From a structural analysis
perspective, researchers primarily evaluate the performance of
enhanced monopile foundations through numerical simulations and
Table 2
experimental studies. Several studies (Haiderali and Madabhushi, 2016; Material properties of Q345 steel.
Zhai and Yin, 2022) have shown that enhanced monopile foundations
Type Density (kg/ Yielding strength Modulus of Poisson’s
with skirt and wing structures effectively improve lateral bearing ca­
m3) (MPa) elasticity (GPa) ratio
pacity, with the enhancement effect increasing as the length of addi­
Q345 7850 345 210 0.3
tional components increases. From an experimental research
perspective, Wang et al. (2017) examined the lateral bearing capacity of
three suction bucket foundations with different aspect ratios through practices.
centrifugal testing, finding that the bucket-shaped foundation out­
performs the original monopile foundation. Additionally, in 2018, Wang 2. Structural design and analysis method
et al. (2018) proposed an analysis method to quantitatively estimate the
cyclic lateral displacement of a new type of enhanced foundation under 2.1. Wind turbine parameters
service conditions, deriving the degradation coefficient based on
centrifuge test results. In this paper, we focus on the NREL-5MW wind turbine, which is
However, previous studies have primarily focused on the determin­ assumed to be installed in the East China Sea with a water depth of 20 m.
istic analysis of the bearing performance of monopile foundations, with Table 1 shows the detailed wind turbine parameters, while Table 2
less emphasis on reliability analysis. Currently, research on wind turbine shows the material properties of Q345 steel used by the wind turbine.
foundation reliability mainly centers on large monopile and suction
bucket foundations, such as the reliability analysis of cylindrical foun­
dations in layered soils based on numerical simulations (Zhao et al., 2.2. Monopile foundation structural design
2022), reliability analysis of suction bucket foundations for offshore
wind turbines in silty sand (Yoon et al., 2019), and fatigue reliability A preliminary monopile design has been proposed with a diameter of
assessment of monopile structures based on cumulative fatigue damage D = 7 m, a wall thickness of 0.06 m, and a depth of 35 m into the mud. To
(Horn and Leira, 2019). Due to the involvement of complex environ­ enhance the stability and bearing capacity of the traditional monopile
mental and structural parameters, reliability analysis is essential for foundation while reducing construction costs, current research activities
accurately predicting the performance of wind turbine foundations. A are focused on the utilization of hybrid foundations, which involve a
promising approach to address this challenge is to perform combination of shallow and deep foundations (Bhattacharya, 2019; Zhai
probability-based reliability calculations. For instance, Jiang (2016) and Yin, 2022). Depending on the specific design requirements and
investigated the reliability of monopile foundations of nearshore wind geological conditions, different types of hybrid monopile foundations
turbines under wind and wave loads based on probability theory from can be selected and applied. In this paper, we propose two types of
statistical mathematics. Reliability analysis proves to be more hybrid monopile foundations: one featuring a circular plate and the
cost-effective and provides a more comprehensive understanding of the other incorporating a hollow skirt added to the outside of the cylindrical
possibility of structural failure, compared to the conservatism of deter­ pipe of the traditional monopile foundation. Fig. 1 illustrates the sche­
ministic analysis, making it useful in practical applications to support matic diagram of the traditional and hybrid monopile foundations,
monopile foundations. while Table 3 provides their respective geometric dimensions. It is
In this study, we utilize the NREL-5MW wind turbine (Fan, 2016) as necessary to note that owing to the existence of enhanced structures, the
an example to illustrate the structural design of four fixed foundation installation of the hybrid monopiles may require the suction-assisted
types, including the addition of a circular plate and skirt to the tradi­ installation method (Tran et al., 2007).
tional monopile foundation. The bearing capacity reliability analysis is The Type I monopile, with a diameter of D, an embedment depth of
conducted using the FEM-BPNN-RSM coupling method for the designed 5D, and an aspect ratio of 5, is considered as the reference configuration.
configurations. The results of the reliability analysis for different con­ The objective of this study is to investigate the influence of additional
figurations are investigated. In this study, we take into consideration the structures on the bearing capacity and reliability of a monopile foun­
correlation of wind and wave environmental parameters (Dong et al., dation in comparison to the traditional monopile design.
2018a,b) and use the structure-soil coupled finite element model (FEM)
to evaluate the rotation response of the wind turbine monopile foun­ 2.3. Reliability analysis methods
dation at the mudline under wind and wave loads. We use the back
propagation neural network – response surface method (BPNN-RSM) 2.3.1. Reliability analysis process
coupled method to calculate the reliability of hybrid monopile founda­ The analytic expression for the performance function forms the basis
tions. The findings demonstrate that the hybrid monopile foundations of most existing reliability calculation methods (Zhang, 2009). How­
significantly enhance the reliability of the monopile to a certain extent ever, for complex structural systems such as wind turbines, the reli­
compared with the traditional monopile foundations. The proposed ability analysis model cannot be pre-determined, making it challenging
FEM-BPNN-RSM combined approach provides a more comprehensive to use methods like the JC method. Therefore, this paper proposes a
understanding of the performance and reliability of the modified novel approach that combines the BPNN-based RSM with the FEM to
monopiles, offering valuable insights for design and engineering more accurately predict the bearing capacity reliability of wind turbine

2
Q. Yin et al. Ocean Engineering 287 (2023) 115845

Fig. 1. Schematic diagram of traditional and hybrid monopile structures.

function or surface, referred to as the response surface. RSMs fit an

Table 3
approximate expression for the explicit function Z = g(X1 , X2 , ⋯, Xn )
Geometric dimensions of traditional and hybrid monopile structures.
using a finite experimental regression, where X1 , X2 , ⋯, Xn is a vector of
Monopile Monopile Embedment Skirt Skirt n random variables. Generally, the response surface function is repre­
type Diameter depth Diameter Length
sented by a quadratic polynomial with no cross terms. However, for
I D 5D reliability analysis of complex engineering structures like wind turbine
II D 5D 3D
foundations, the accuracy requirements cannot be satisfied by a
III D 5D 3D 0.5D
IV D 5D 3D D quadratic polynomial without cross terms. Consequently, a neural
network with robust nonlinear mapping capability can be employed as a
response surface to effectively address the accuracy issue from a theo­
foundations. A detailed introduction to this method will be presented in retical perspective.
the next section, and the specific analysis process is illustrated in Fig. 2. The BPNN (Qiu et al., 2020) is a feedforward multilayer neural
network trained by the backpropagation algorithm for error minimiza­
2.3.2. BPNN-RSM coupled method tion. From a theoretical perspective, the BP network can simulate any
The RSM (Dong et al., 2017; Ji et al., 2018) is a widely employed complex nonlinear mapping from the input to the output, making it
approach for reliability analysis based on experimental design. Its highly valuable in the context of heap foundation engineering. The
fundamental concept involves replacing a hidden or time-consuming learning process of the BP network consists of two main stages: a for­
true functional or bound-state surface with an easily manipulable ward computation process and an error backpropagation process. This

Fig. 2. Reliability analysis process used in this study.

3
Q. Yin et al. Ocean Engineering 287 (2023) 115845

Monte Carlo method (Deng et al., 2005; Park et al., 2004) combined
with artificial neural networks to determine the reliability of structural
systems. The process involves the following steps.

(1) To produce data for {xi , g(xi )}(i = 1, 2, ⋯, S) training the network,
which can be obtained by numerical analysis or experiments to
obtain the response y(xi ) of the structure under different values of
the basic random variables xi .
(2) Drawing up the structure of the neural network.
(3) Training the network with data {xi , g(xi )}(i = 1, 2, ⋯, S)
(4) Obtained xi (i = 1, 2, ⋯, N) via random sampling
(5) Using the trained network to simulate and calculate g(xi )(i = 1, 2,
⋯, N)
(6) To calculate the estimated probability of failure pf , and to use the
∧

n
formula pf = Nf to calculate the probability of failure, where nf
∧

represents the number of failure samples, and N represents the

Fig. 3. Structure of the BPNN.
total number of samples.

allows the network to automatically establishes a highly nonlinear

3. Finite element analysis
mapping relationship between the input and output by learning of a
large number of samples. The network architecture typically comprises
3.1. Finite element model establishment
an input layer, a hidden layer, and an output layer, as shown in Fig. 3.
The forward propagation process involves processing the input infor­
3.1.1. Soil parameters
mation layer-by-layer through the hidden layer and passing it to the
The study is carried out under typical geological conditions of the
output layer. The state of each layer of neurons only affects the state of
East China Sea, with a single layer of silty sand (Wu, 2020) as the soil.
the neurons in the following layer. When the expected result is not
The Mohr-Coulomb constitutive model, primarily suitable for granular
achieved at the output layer, the backpropagation of the error is used to
materials under monotonic loading is introduced here since soil defor­
return the error signal along the original path of the connection, and the
mation is involved in the FEMs.
weights of each layer of neurons are adjusted to minimize the error
The yield surface function of the Mohr-Coulomb model is:
signal. Assuming a 3-layer neural network, the output signal from the
neurons in the input layer is Xi , the output signal of the hidden layer F = Rmc q − p tan φ − c = 0 (4)
neurons is Zj , the output signal of the output layer neurons is Yk , the
connection weights between each layer are ωji and ωkj , and the weight of where φ is the inclination angle of the Mohr-Coulomb yield surface on
the hidden layer is bj , the actual output is gk . the q-p stress surface, which is called the friction angle of the material,
The output signal from the hidden layer is and 0◦ ≤ φ ≤90◦ ; c is the cohesion of the material; Rmc = (Θ, φ) is
( calculated by the following equation, which controls the shape of the
∑ ) yield surface in the π plane.
Zj = f ωji Xi − bj (1)
i 1 ( π) 1 ( π)
Rmc = √̅̅̅ sin Θ + + cos Θ + tan φ (5)
The error between the output of the network and the true output is 3cos φ 3 3 3
( 3
∑ ∑ ∑ ) where Θ is the polar declination angle, defined as cos(3Θ) = qr 3 , r is the
Yk = ωkj Zj = ωkj f ωji Xi − bj (2)
j j i third bias stress invariant J3. The soil mechanical parameters can be
found in Table 4.
The error between the output of the network output and the true
output is 3.1.2. Mesh and boundary conditions
1∑ The finite element model comprises the monopile, soil and tower
E= (gk − Yk )2 (3) structure. To simplify the gravity of the cabin, hub, and blades, a
2
reference point with a point mass at the tower top is introduced, coupled
The error is back-propagated through the BP algorithm to adjust the
with each other. The tower using shell elements of varying cross-section
weights and thresholds, and ultimately a set of optimal weights and
dimension, which account for changes in diameter and thickness along
thresholds are obtained to minimize the error. At this point the training
the tower’s height direction. The thickness variation is defined using a
of the neural network is complete, and the nonlinear mapping process is
linear expression field, specifically:
performed.
T = aH + b (6)
2.3.3. Monte Carlo method
Monte Carlo simulation (MCS) is a numerical computation method where H is the height coordinate and T is the corresponding thickness of
based on the statistical sampling theory, utilizing random computer the tower at that height, which varies as a function of the Cartesian
simulations to study objective phenomena. In this paper, we use the coordinate value H. The values of a and b are based on the actual

Table 4
Geological parameters of the East China Sea used in this study (Wu, 2020).
Soil type Young’s modulus, Poisson’s ratio, μ Friction angle, φ/◦ Dilation angle, ψ /◦ Cohesion, c/kPa Effective unit weight,
E/MPa γ/kN⋅m− 3

Silty sand 39 0.3 32 15 4 9

4
Q. Yin et al. Ocean Engineering 287 (2023) 115845

Fig. 4. Finite element models considering the structure-soil interaction.

Fig. 5. Comparison of moment-rotation curves with previous literature.

thickness of the tower. 3.2. Model validation

In this model, the horizontal dimension of the soil is set to 20 times
the diameter of the monopile (20D), and the vertical dimension is 2.5 This paper aims at the actual geology conditions in the East China
times the monopile diameter plus the monopile length (2.5D + L). To Sea. The seafloor soils under investigation mainly consist of silty sand,
ensure appropriate refinement, the soil mesh is refined within a range of which is a simplification based on actual geological survey results (Wu,
3 times the monopile diameter in the horizontal direction, with a min­ 2020). To validate the accuracy of the structure-soil interaction model,
imum mesh size of approximately 0.5 m near the monopile and a we have reconstructed a structure-soil coupled model based on the
maximum mesh size of no more than 15 m. The top section above the approach of Chen et al. (2020), as described in our published paper (Zhai
monopile base is split in single precision along the axis of the monopile and Yin, 2022), and the results are shown in Fig. 5. The horizontal load is
with a spacing of 0.5 m–2 m, resulting in a total of approximately 61,000 applied at two heights on the tower, corresponding to two moment arms
to 64,000 grids. At the vertical boundary, the soil is constrained with a relative to the mudline, simulating wind load and wave load conditions.
horizontal displacement, and all translational degrees of freedom are The moment and rotation curves of the two models are highly compa­
constrained at the bottom boundary. rable, indicating the effectiveness of the finite element method
When considering the pile-soil interaction, the normal force between employed in this research in the structure-soil coupled analyses. It is
the monopile foundation and the soil is treated as a “hard contact”, important to note that Chen et al. (2020) have validated their numerical
meaning that only pressure transfer during contact is considered. The models by comparing them with previous experimental and finite
tangential coefficient of friction is set to 0.5, allowing for separation element results (Li et al., 2015).
between the monopile and the soil after contact. Fig. 4 illustrates the
finite element models of Type I and Type IV monopiles, which are the 3.3. Load calculation
final models used in this study.
3.3.1. Wind load
Due to the variation in wind velocities at different wind turbine

5
Q. Yin et al. Ocean Engineering 287 (2023) 115845

( )β
z
Vhub = V10 (7)
z10

In this formula, V10 represents the average wind speed at a height of 10

m above sea level, whereas β = 0.2 represents the index of the wind
speed profile.

(1) When the wind speed at the hub height is between the cut-in wind
speed and the cut-out wind speed (Arany et al., 2017), the thrust
force can be calculated by the following equation:
1 2
w = ρ0 CT AR Vhub (8)
2
where Fw represents the thrust force; ρ0 = 1.025 kg/m3 is the air
density; AR is the swept area of the blades; Vhub is the average wind speed
Fig. 6. Diagram of horizontal wave loads acting on monopiles. at the hub height; CT is the thrust coefficient.

heights, it is necessary to convert the wind speed at 10 m above sea level (2) If the wind speed at the hub height is greater than the cut-out
to the wind speed at the hub height of the wind turbine in order to wind speed (Ma and Chen, 2021), then the thrust force can be
calculate wind loads. The formula for calculating wind speed conversion calculated using the following equation:
is as follows: 1
Fw = ρ0 Cs AB Vhub 2 (9)
2

Fig. 7. Response cloud diagrams of support structures and upper tower.

6
Q. Yin et al. Ocean Engineering 287 (2023) 115845

Fig. 8. The vector graph of soil displacement under the same typical load condition.

7
Q. Yin et al. Ocean Engineering 287 (2023) 115845

where Cs is the drag coefficient in the equation, taken to be 1.5, and AB is

the total projected area of the blades, with the other parameters being
the same as in Equation (8).
Equation (9) is also utilized to calculate the wind load on the tower,
considering a drag coefficient of Cs = 0.5. As the wind speed varies at
different tower heights, the tower is divided into sections to approxi­
mate the change in wind speed.

3.3.2. Wave load

The wave loads on the monopile foundation are calculated using
Morison’s equation. Airy wave theory is employed for the waves, and
the wave loads on the monopile foundation are distributed on the side
surface of the monopile foundation below the water surface and above
the mud surface. The horizontal load received over a length dz can be
calculated using Equation (10), and the schematic diagram of the hori­
zontal loading is presented in Fig. 6.

1 π D2 ∂ux
dFH = fH dz = CD ρDUX |ux |dz + CM ρ dz (10)
2 4 ∂t Fig. 9. Location of rotation centers.
The total force of the wave loads acting on the monopile foundation
can be obtained by integration:
Table 5
γDH 2 ′ γπ D2 H ′ Response values of rotation at the mudline.
FH = C D K1 cos|cos θ| + CM K2 sin θ (11)
2 8 Monopile type I II III IV

2π 2π Rotation angle at the mudline, θ 0.00653 0.00632 0.00436 0.00379

θ = kx − ωt, k = ,ω = (12) Difference – 3.2% 33.23% 41.96%
L T
( ) ( )
2k d + H2 + sh2k d + H2
K′1 = (13) monopiles under wind and wave loads. The maximum soil displace­
8sh2kd ments are measured as 0.091 m, 0.090 m, 0.063 m and 0.045 m for Type
shkd I, II, III, IV monopiles, respectively. Notably, the soil displacement de­
K′2 = (14) creases by 50% after adding a long skirt with a length of D to the Type I
chkd
monopile. This reduction can be attributed to the additional structure,
where FH represents the total wave force on the monopile within the which increases the contact area and frictional resistance between the
length range above the mudline, γ represents the density of water, D monopile and the soil, thereby reducing the settlement and deformation
represents the diameter of the monopile, L represents the wave length, T of the soil around the monopile. The position of the rotation center is
represents the wave period, H represents the wave height, d represents marked in the displacement vector diagram in Fig. 8, and the rotation
the water depth above the mudline, CD and CM represent the drag force center points of the four monopile types sequentially increase.
coefficient and inertia force coefficient related to the cross-sectional The distance between the rotation centers and the mudline is shown
shape, respectively, with reference to DNV codes (DNV, 2014). In this in Fig. 9. The normalized rotation center depths for Type I, Type II, Type
paper, CD is taken as 1.2, and CM is taken as 2.0. III, and Type IV monopiles are d/L = 0.66, 0.6, 0.53, and 0.48, respec­
tively. Compared to the Type I monopile, the depth has been reduced by
4. Results and analyses 9%, 19.7%, and 27.3%, respectively. The elevation increase of the
rotation center corresponds to an enhancement in the bearing perfor­
4.1. Finite element analysis results mance of the monopile structures.
Table 5 compares the rotation angle at the mudline for the four
4.1.1. Stresses and displacements under typical load condition monopile types under the typical load condition considered in this
According to DNV codes, the rotation angle of OWT monopiles study. The addition of both a circular plate and a skirt effectively reduces
caused by environmental loads must not exceed 0.25◦ (DNV, 2014). This the rotation angle and improves the bearing capacity of the monopile.
section will introduce the deterministic analysis based on the finite The numerical differences indicate that the effect of adding a circular
element model, conducted using the software ABAQUS. A typical load plate to enhance the bearing capacity of the monopile is not significant,
condition of a wind speed of 48 m/s, a wave height of 10.36 m, and a with the rotation angle decreasing by only 3.2%. On the other hand, the
wave period of 7.36s was calculated to show the structural responses. addition of a skirt has a more pronounced enhancement effect, resulting
Under this load condition, the rotation angle at the mudline for Type III in a 41.96% decrease in the rotation angle. Additionally, the use of a
monopile exactly reaches 0.25◦ . The stress and displacement distribu­ skirt increases the contact area and frictional resistance between the
tions of Type I and Type III monopile structures under the typical monopile and the soil, reducing soil settlement and deformation around
wind-wave load condition are presented in Fig. 7. Results show that the the monopile, thereby leading to reduced displacement of the soil.
maximum stress of the wind turbine structures occurs on the windward The soil stress distribution diagrams are presented in Fig. 10. The
side of the tower bottom, and the maximum displacement occurs at the addition of a skirt to the monopile results in a significant change in the
tower top, which is in agreement with the actual engineering scenario. stress distribution of the soil around the monopile, leading to a reduction
The maximum stress of Type I was found to be 229 MPa, which is below in the soil stress around the skirt. This observation indicates that there is
the yield strength of Q345 steel, and the structure’s safety factor was still a considerable bearing capacity margin for the hybrid foundation
calculated to be 1.5. Similarly, the maximum stress of Type III was 275 within the surrounding soil.
MPa, which is also below the yield strength of Q345 steel, and the Table 6 presents the rotation angle values at the mudline obtained by
structure’s safety factor was calculated to be 1.25. applying wind load, wave load, and combined wind-wave load sepa­
Fig. 8 shows the soil displacement vector diagrams of four types of rately on the conventional monopile model. The results indicate that the

8
Q. Yin et al. Ocean Engineering 287 (2023) 115845

Fig. 10. The cloud graph of soil stress.

Table 6 Table 7
Response values of rotation angle at the mudline under different load conditions. Parameter estimation and goodness-of-fit test results for W, H, and T.
Load conditions Rotation angle at the mudline, θ (◦ ) μ α γ (0.05)

Wind load 0.00517 W 7.447 21.459 2.443 0.1596 0.2417

Wave load 0.0014 H 0.312 6.613 2.676 0.1150
Wind and wave loads 0.00653 T 3.003 3.136 2.542 0.1160

rotation angle under combined wind-wave loading is slightly greater In this paper, the Weibull distribution is utilized to fit the marginal
than that under wind loading alone, while both are greater than the distribution of wind speed(W), wave height(H) and wave period(T),
rotation angle under wave loading alone. These findings highlight the respectively. In this paper, the least square method is applied to fit the
dominant role of wind loading in the rotational behavior of the parameters of Weibull distribution. The parameter estimation results of
monopile. W, H, and T are listed in Table 7.
In this study, the K–S test is employed to assess the goodness-of-fit of
4.2. Environmental parameters the marginal distributions of W, H, and T. Assume the K–S statistics
Dn = sup |Fn (x) − F0 (x)| (F(x) is the actual distribution for the sample
− ∞<x+∞
This section provides a brief overview of the process for handling data and F0(x) is the theoretical distribution), and the K–S critical value
environmental parameters. The Weibull distribution was first proposed is Dn(α) for different sample sizes n (α is the significance level). If the K–S
by Swedish physicists Weibull and has been applied to various fields test statistics, the hypothetical theoretical distribution is considered
such as component life and fatigue degree analysis (Weibull, 1951). It acceptable for fitting the sample data; otherwise, hypothetical theoret­
has been utilized by researchers like Justus, Stewart and Essenwanger to ical distribution is not suitable for fitting the sample data.
fit the wind speed data, which showed the effectiveness in modeling Table 7 presents the results of the K–S test. The statistic values for W,
maximum wind speed (Justus and Mikhail, 1976; Stewart and Essen­ H, and T are smaller than the test statistic D30 (0.05) = 0.2417, indi­
wanger, 1978; Justus et al., 1978). In recent years, the Weibull distri­ cating that these distributions for W, H, and T successfully pass the
bution has been widely used in hydrologic calculation and frequency statistical test. Therefore, the Weibull distribution is adopted as the
analysis for coastal and offshore engineering (Morgan et al., 2011; Dong marginal distribution for W, H, and T and it is used to construct the joint
et al., 2018a,b). The probability density function and probability func­ probability model. The fitting curves of W, H, and T are shown in Fig. 11,
tion of the three-parameter Weibull distribution are expressed as (Wais, respectively.
2017):
γ (x − μ)γ− 1 [ (x − μ)γ ] 4.2.1. Joint probability analysis
f (x) = exp − x≥μ (15) In this paper, a copula function is used to construct the joint prob­
α α α
⎧ ability distribution model of W, H, and T. Sklar (1959) proposed the
[ ( )γ ]
⎨ 1 − exp − x − μ , x≥μ concept of copulas and researched transformations between bivariate
F(x) = α (16) joint distributions and marginal distributions, enabling accurate and
⎩
0， x<μ reliable description of the relevant margins.
If FXYZ (x, y, z) is the joint probability distribution of marginal dis­
where, μ is the location parameter, α is the scale parameter, γ is the tributions FX(x), FY(y) and FZ(z), then a trivariate copula C (u, v, w) must
shape parameter.

9
Q. Yin et al. Ocean Engineering 287 (2023) 115845

Fig. 11. Marginal fitting curves of wind speed, wave height and wave period.

exist, so that for arbitrary (x, y, z) ∈ R3, the link between a trivariate
distribution FXYZ (x, y, z) and the associated three-dimensional copula C (2) Frank copula:
is given by the functional identity stated by Sklar’s Theorem (Sklar, { }
1959; Nelsen, 2006): 1
C(u, v, w; θ) = − ln 1 +
[exp(− θu) − 1][exp(− θv) − 1][exp(− θw) − 1]
,θ
θ [exp(− θ) − 1]2
F(x, y, z) = CUVW (u, v, w) = CFX ,FY ,FZ (FX (x), FY (y), FZ (z)) (17)
∈R
There are four kinds of commonly used Archimedean copulas: the
(19)
Clayton copula, Frank copula, Gumbel-Hougaard copula, and Ali-
Mikhail-Haq copula, all of which play an important role in costal engi­
neering. The probability functions of these four copula functions are (3) Gumbel-Hougaard (GH) copula:
listed below. { [ ]1θ }
C(u, v, w; θ) = exp − (− ln u)θ + (− ln v)θ + (− ln w)θ , θ ∈ [1, ∞)
(1) Clayton copula:
(20)
[( )− 1 ]
C(u, v, w; θ) = max u− θ + v− θ + w− θ − 2 θ , 0 , θ ∈ [ − 1, ∞)\{0} (18)
(4) Ali-Mikhail-Haq (AMH) copula:

10
Q. Yin et al. Ocean Engineering 287 (2023) 115845

Fig. 12. Goodness-of-fit test.

Table 8
Three-dimensional goodness-of-fit test results.
Goodness-of-fit Clayton Frank GH AMH

RMSE 0.0529 0.2006 0.0663 0.1666

AIC − 174.41 − 94.38 − 160.83 − 105.54

uvw
C(u, v, w; θ) = , θ ∈ [ − 1, 1) (21)
[1 − θ(1 − u)(1 − v)(1 − w)]
Based on the marginal distributions of W, H, and T, four kinds of
three multivariate copulas are used to establish the joint probability
models. Fig. 12 compares the empirical frequency and theoretical fre­
quency. According to Fig. 12, Clayton copula and GH copula show a
robust fit between empirical and theoretical frequencies. But the best
goodness-of-fit of joint distribution for W, H, and T cannot be obtained
solely based on visual inspection. Therefore, the RMSE and AIC criteria
(Zhai et al., 2021) are calculated to obtain the optimal copula to
construct the joint distribution model (Table 8). Based on Table 8,
three-dimensional Clayton copula is the optimal copula for constructing
the joint distributions of W, H, and T.

4.2.2. Random data generation

Fig. 13. 3D scatter plot of the randomly generated sample data.
In this paper, Marshal-Olkin method through Laplace transform is
adopted to generate the random numbers for three-dimensional Clayton 1
1 (1 + s)− θ , which is the inverse function of Clayton copula; then, assume
copula C(u, v, w; θ) = (u− θ + v− θ + w− θ − 2)− θ (Tang et al., 2023). Firstly,
we create independent stochastic numbers vu, vv, vw, which obey iden­ that su = − lnYvu ,sv = − lnYvv ,sw = − lnYvw . Finally, the random numbers of
tically distribution and uniform distribution. Secondly, we create sto­ Clayton copula can be obtained, which are expressed as u = φ− 1 (su ),v =
( )
chastic numbers Y = Γ 1θ , 1 , which obey G function. Here, G function φ− 1 (sv ), w = φ− 1 (sw ), respectively. The 3D scatter plot of the randomly
through the Laplace transform can be transformed as φ− 1 (s) = generated sample data is shown in Fig. 13.

11
Q. Yin et al. Ocean Engineering 287 (2023) 115845

Table 9 randomly assigned 70%, 15%, and 15% of the samples, respectively. The
Selection of test points for the BPNN-based RSM. hidden layer contained a fixed number of 5 neurons. The fitted curve
Environmental Parameters Values and the calculated curve of the resulting best-fit value are shown in
Fig. 14, where the Levenberg-Marquardt algorithm was used. The curve
Wind speed, V10 (m/s) 20, 40, 49, 50, 51, 52, 53, 54, 60, 67
Wave Height, H (m) 3.5, 5.5, 7.5, 8.5, 10.5 almost passes through the origin, and the coefficient of the best-fit curve
Wave Period, T (s) 4, 5, 6, 7, 8 is 0.999, which is close to 1. Similarly, correlation curve between the
fitting value and calculated value of Type IV monopile is shown in
Fig. 15, and the coefficient of the best-fit curve is 0.989. These results
4.3. Establishment and validation of neural network response surface indicate good agreement between the data. Detailed analysis reveals
that the trained backpropagation neural network exhibits high fitting
As demonstrated by the study conducted by Tang et al., in 2023, accuracy, enabling its utilization for the prediction and evaluation of the
employing neural networks to construct the response surface can result rotation angle at the mudline of wind turbines.
in improved computational accuracy. The research reveals that at a wind
speed of approximately 52 m/s, the rotation angle at the mudline ex­
ceeds 0.25◦ , leading to the wind turbine failure. To further enhance the 4.4. Reliability analysis
predictive accuracy of the neural network, additional training points are
incorporated around the wind speed of 52 m/s. The selected experi­ In practical engineering, statistical characteristics of environmental
mental points are listed in Table 9. parameters and other factors such as the upper structure, monopile, and
The combination of wind speed, wave height and wave period listed soil geological parameters have influences on monopile reliabilities.
in Table 9 resulted in 250 training points. Therefore, a total of 250 × 4 = This study mainly focuses on the uncertainty of the environmental pa­
1000 finite element calculations were performed to obtain the rotation rameters and their influence on the reliability analysis of the monopile
angle at the mudline for the four monopile types. For illustrative pur­ structure. The neural network is utilized to predict the response values
poses, Type I monopile was used as an example, and a neural network corresponding to a large number of environmental parameters extrac­
model was trained using the traingd training function. The input data ted. According to the DNV regulations, the rotation angle at the mudline
consisted of the 250 sets of wind speed, wave height, and period, while of a wind turbine’s monopile due to environmental loads must not
the corresponding output was the angle of rotation at the mudline of the exceed 0.25◦ . The results of the failure probability and reliability index
wind turbine. The training data, validation data, and test data were are presented in Table 10. According to the DNV guideline for offshore

Fig. 14. Correlation curve between fitting value and calculated value of Type I monopile.

12
Q. Yin et al. Ocean Engineering 287 (2023) 115845

Fig. 15. Correlation curve between fitting value and calculated value of Type IV monopile.

Table 10 Table 11
The results of the failure probability and reliability index. Correlation coefficients between environmental parameters and rotation
Type I II III IV
angle.
Environmental parameters Correlation coefficients
Pf 6.26% 4.3% 0.8705% 0.265%
β 1.53 1.72 2.37 2.78 Wind speed, V (m/s) 0.941
Wave height, H (m) 0.622
Wave period, T (s) 0.587
structural reliability analysis – general, the allowed failure probabilities
are different in terms of the failure consequences, namely not serious,
serious and very serious. The failure consequence of “not serious” cor­ the rotation angle at the mudline. Subsequently, the resulting correla­
responds to a target failure probability is β = 3.09. Despite the fact that tion coefficient was analyzed and presented in Table 11.
the failure consequence of an excessive rotation is not supposed to be The correlation coefficient serves as a measure of the strength and
more severe than the failure consequences provided in the DNV code, direction of the relationship between two variables. A coefficient close
the bearing capacity reliability of Type I monopile still appears to fail the to 1.0 indicates a strong positive correlation, while a coefficient close to
design requirements. The addition of a circular plate and skirt increases − 1.0 indicates a strong negative correlation. The closer the coefficient is
the reliability of Type II, Type III, and Type IV monopiles, and the to 0, the weaker the relationship between the two variables. The
reliability rises with increasing buried skirt depth. Type IV monopile has strength of the correlation coefficient can be classified as follows: an
a failure probability of 0.265% and a reliability index of 2.78, which is extremely strong correlation is indicated by a coefficient of 0.8–1.0, a
81.7% higher than the traditional monopile. The computed results strong correlation is indicated by 0.6–0.8, a moderate correlation is
demonstrate that the additions of a circular plate and skirt effectively indicated by 0.4–0.6, a weak correlation is indicated by 0.2–0.4, and a
improve the bearing capacity of the monopile foundation and reduces very weak or absent correlation is indicated by 0.0–0.2.
the probability of failure. Based on the results presented in Tables 11 and it can be concluded
To examine the relationship between wind turbine failure and that wind speed has the strongest and most consistent relationship with
environmental parameters such as wind speed, wave height, and period, the rotation angle, exhibiting an extremely high correlation. The second
Kendall’s rank correlation coefficient was utilized in this study. The strongest relationship is between wave period and wave height,
initial step involved conducting a statistical test to determine the exis­ demonstrating a moderate correlation with a consistent direction of
tence of a significant relationship between the relevant parameters and change. These findings suggest that wind speed is the most influential
factor in determining the rotation angle of the wind turbine.

13
Q. Yin et al. Ocean Engineering 287 (2023) 115845

The FEM-BPNN-RSM method employed in this study demonstrates Declaration of competing interest
strong applicability in the reliability analysis of marine engineering
structures involving multiple design parameters. This paper focuses The authors declare that they have no known competing financial
solely on the influence of environmental parameters in the reliability interests or personal relationships that could have appeared to influence
analysis. Further exploration is needed to investigate the impact of wind the work reported in this paper.
turbine structural parameters and soil properties on the monopile’s
bearing capacity reliability. Additionally, using neural networks require Data availability
a large amount of training data to obtain accurate prediction models.
Including more parameters would necessitate a larger sample size and Data will be made available on request.
increase computational complexity. Therefore, in the following study
with more statistical variables, efficient design is required for grid se­ Acknowledgements
lection schemes.
This work has received funding from the Natural Science Foundation
5. Conclusions of Jiangsu Province of China [Grant numbers BK20190970,
BK20190974]. This work has also received funding from the Post­
This paper introduces three hybrid monopiles based on the tradi­ graduate Research & Practice Innovation Program of Jiangsu Province
tional monopile and investigates their bearing capacity performance in [SJCX23_2211].
real ocean environments through deterministic finite element analysis
(FEA). Furthermore, this study examines the bearing capacity reliability References
using the probabilistic FEM-BPNN-RSM method. A three-dimensional
monopile-soil coupled FEA model is established, and its accuracy is Ahmed, S.S., Hawlader, B., 2016. Numerical analysis of large-diameter monopiles in
validated. The BPNN is used to construct the wind turbine failure dense sand supporting offshore wind turbines. Int. J. GeoMech. 16 (5), 04016018.
Arany, L., Bhattacharya, S., Macdonald, J., Hogan, S.J., 2017. Design of monopiles for
response surface, and the bearing capacity reliability of the four
offshore wind turbines in 10 steps. Soil Dynam. Earthq. Eng. 92, 126–152.
monopile types is analyzed using the Monte Carlo method. This research Bhattacharya, S., 2019. Design of Foundations for Offshore Wind Turbines. Wiley,
provides valuable insights for the structural design and construction of Hoboken. https://doi.org/10.1002/9781119128137.
Carswell, W., 2015. Soil-structure Modeling and Design Considerations for Offshore
offshore wind turbine foundations in the future. The main findings are as
Wind Turbine Monopile Foundations. University of Massachusetts Amherst,
follows. Amherst.
Chen, D., Gao, P., Huang, S., Li, C., Yu, X., 2020. Static and dynamic loading behavior of
(1) This study takes into account the correlation of wind and wave a hybrid foundation for offshore wind turbines. Mar. Struct. 71, 102727.
Deng, J., Gu, D., Li, X., Yue, Z.Q., 2005. Structural reliability analysis for implicit
environmental parameters and uses the Weibull distribution to fit performance functions using artificial neural network. Struct. Saf. 27 (1), 25–48.
the marginal distributions of wind speed, wave height, and period DNV, 2014. DNV-OS-J101 Design of Offshore Wind Turbine Structures. Det Norske
in the East China Sea. Among the four types of three multivariate Veritas, Norway.
Dong, S., Sheng, J.L., Zhai, M.Y., Yang, M.C., Huang, T., 2017. Reliability analysis of 3D
copula probability models, the three-dimensional Clayton copula earth slope based on NN-RSM. Ind. Miner. Process. 46 (6), 53–55.
is the optimal copula for constructing the joint distribution of W, Dong, S., Liao, Z., Feng, L., Tao, S.S., 2018a. Environmental parameter estimation with
H, and T in the East China Sea. The parameters of the Weibull nested logistic distribution for ocean platform in the northern area of South China
Sea. J. Ship Mech. 22, 1181–1188.
distribution are fitted using the least squares method. Dong, S., Chen, C., Tao, S., Gao, J., 2018b. Stochastic model for estimating extreme water
(2) Three modified monopiles are proposed by incorporating a cir­ level in port and coastal engineering design. J. Ocean Univ. China 17, 744–752.
cular plate and skirt into the traditional monopile. These modi­ Fan, H.Y., 2016. Dynamics and Fatigue Analysis on Monopile Foundations of Offshore
Wind Turbines under Wind and Wave Loads. Harbin Institute of Technology, Harbin.
fications significantly enhance the reliability of the offshore wind
Haiderali, A., Madabhushi, G., 2016. Improving the lateral capacity of monopiles in
turbine monopile foundations. Among the additions, the skirt has submarine clay. Proc. Inst. Civ. Eng-gr. 169 (4), 239–252.
a more pronounced effect on the bearing capacity and reliability Horn, J.T., Leira, B.J., 2019. Fatigue reliability assessment of offshore wind turbines with
stochastic availability. Reliab. Eng. Syst. Saf. 191, 106550.
due to its ability to utilize a larger area of soil.
Ji, X.Q., Wang, J., Gao, H.B., 2018. Plastic reliability analysis of offshore wind turbine
(3) Compared to Type I monopiles, Type II, III, and IV monopiles structure throughout its life cycle. Wind Energy (12), 72–77.
have rotation centers at shallower depths, with reductions of 9%, Jiang, B.W., 2016. Reliability Analysis of Offshore Wind Turbine Foundation under Wind
19.7%, and 27.3%, respectively. The rotation angle at the mud­ Loads and Wave Loads. Chongqing Jiaotong University, Chongqing.
Justus, C.G., Mikhail, A., 1976. Height variation of wind speed and wind distributions
line of the monopile foundation is significantly affected by wind statistics. Geophys. Res. Lett. 3 (5), 261–264.
speed, while the influence of waves on structural response is Justus, C.G., Hargraves, W.R., Mikhail, A., Graber, D., 1978. Methods for estimating
much smaller in comparison. wind speed frequency distributions. J. Appl. Meteorol. 350–353.
Kim, H.G., Kim, B.J., 2018. Feasibility study of new hybrid piled concrete foundation for
(4) The paper aims to go beyond deterministic analysis by incorpo­ offshore wind turbine. Appl. Ocean Res. 76, 11–21.
rating probabilistic analysis, which takes into account un­ Li, D., Zhang, Y., Feng, L., Gao, Y., 2015. Capacity of modified suction caissons in marine
certainties and variations in environmental parameters that affect sand under static horizontal loading. Ocean. Eng. 102, 1–16.
Ma, H., Chen, C., 2021. Scour protection assessment of monopile foundation design for
the bearing capacity reliability of the modified monopiles. This offshore wind turbines. Ocean. Eng. 231, 109083.
combined approach provides a more comprehensive under­ Ma, H., Yang, J., 2020. A novel hybrid monopile foundation for offshore wind turbines.
standing of the performance and reliability of the modified Ocean. Eng. 198, 106963, 2.
Morgan, E.C., Lackner, M., Vogel, R.M., Baise, L.G., 2011. Probability distributions for
monopiles, offering valuable insights for design and engineering
offshore wind speeds. Energy Convers. Manag. 52 (1), 15–26.
practices in the field. Nelsen, R.B., 2006. An Introduction to Copulas. Springer, New York, pp. 109–155.
Park, S., Choi, S., Sikorsky, C., 2004. Efficient method for calculation of system reliability
of a complex structure. Int. J. Solid Struct. 41 (18–19), 5035–5050.
CRediT authorship contribution statement
Qiu, B., Lu, Y., Sun, L., Qu, X., Xue, Y., Tong, F., 2020. Research on the damage
prediction method of offshore wind turbine tower structure based on improved
Qilin Yin: Methodology, Validation, Funding acquisition, Writing – neural network. Measurement 151, 107141.
review & editing. Wei Wang: Data curation, Writing – original draft. Sklar, M., 1959. Fonctions de repartition an dimensions et leurs marges. Publ. Inst.
Statist. Univ. Paris. 8, 229–231.
Jinjin Zhai: Funding acquisition, Conceptualization, Writing – original Stewart, D.A., Essenwanger, O.M., 1978. Frequency distribution of wind speed near the
draft, Writing – review & editing. surface. J. Appl. Metor. 1633–1642.
Tang, S.X., Zhai, J.J., Yin, Q.L., 2023. Bearing capacity reliability analysis using the
coupled FEM-BPNN-RSM for monopile-supported OWTs. Ocean. Eng. 276, 114228.
Tran, M.N., Randolph, M.F., Airey, D.W., 2007. Installation of suction caissons in sand
with silt layers. J. Geotech. Geoenviron. 133 (10), 1183–1191.

14
Q. Yin et al. Ocean Engineering 287 (2023) 115845

Wais, P., 2017. Two and three-parameter Weibull distribution in available wind power Wu, X., Hu, Y., Li, Y., Yang, J., Duan, L., Wang, T., 2019. Foundations of offshore wind
analysis. Renew. Energy 103, 15–29. turbines: a review. Renew. Sustain. Energy Rev. 104, 379–393.
Wang, X., Yang, X., Zeng, X., 2017. Centrifuge modeling of lateral bearing behavior of Yang, X., Zeng, X., Wang, X., 2018. Performance of monopile-friction wheel foundations
offshore wind turbine with suction bucket foundation in sand. Ocean. Eng. 139, under lateral loading for offshore wind turbines. Appl. Ocean Res. 78, 14–24.
140–151. Yoon, G.L., Yi, J.H., Bae, K.T., 2019. Reliability analysis of suction bucket foundation for
Wang, X., Zeng, X., Yang, X., Li, J., 2018. Feasibility study of offshore wind turbines with offshore wind turbine in silty sand. J. Korean GEO-environ. Soc. 20 (12), 41–47.
hybrid monopile foundation based on centrifuge modeling. Appl. Energy 209, Zhai, J.J., Yin, Q.L., 2022. Finite element analysis of newly designed monopiles for
127–139. offshore wind turbines on seabed with shallowly buried batholith. Ships Offshore
Wang, X., Zeng, X., Yang, X., Li, J., 2020. Liquefaction characteristics of offshore wind Struct. 1–13.
turbine with hybrid monopile foundation via centrifuge modelling. Renew. Energy Zhai, J.J., Yin, Q.L., Dong, S., 2021. Co-occurrence probability of typhoon surges
145, 2358–2372. affecting multiple estuaries in the northern coastal region of Taiwan. Reg. Stud. Mar.
Weibull, W., 1951. A statistical distribution function of wide applicability. J. Appl. Mech. Sci. 42, 101582.
18, 293–297. Zhang, M.Z., 2009. Structural Reliability Analysis-Methods and Procedure. Science Press,
Wu, Y.C., 2020. Study on Bearing Capacity and Optimal Design of Super Large Diameter Beijing.
Monopile Foundation for Offshore Wind Turbines. Shandong University of Science Zhao, H., Jiang, Q., Weng, P., 2022. Reliability analysis of thin-and thick-walled bucket
and Technology, Qingdao. foundations in layered soil based on numerical simulation. Arabian J. Geosci. 15
(17), 1424.

15