Marine Structures 58 (2018) 154–171

Contents lists available at ScienceDirect

Marine Structures
journal homepage: www.elsevier.com/locate/marstruc

Long-term extreme response analysis of a long-span pontoon bridge

T
a,b,∗ c a c
Finn-Idar Grøtta Giske , Knut Andreas Kvåle , Bernt Johan Leira , Ole Øiseth
a
Department of Marine Technology, NTNU, 7491 Trondheim, Norway
b
Multiconsult, Nedre Skøyen vei 2, 0213 Oslo, Norway
c
Department of Structural Engineering, NTNU, 7491 Trondheim, Norway

AR TI CLE I NF O AB S T R A CT

Keywords: For the assessment of extreme load effects needed in design of marine structures, a full long-term
Marine structures analysis is recognized as the most accurate approach. However, due to the very large number of
Pontoon structural response analyses traditionally needed for this approach, the computational effort is
Floating bridge usually considered to increase above acceptable levels for complex structures such as floating
Extreme response
bridges. In this paper, a framework for full long-term extreme response analysis is demonstrated
Long-term response
IFORM
for a long-span pontoon bridge subjected to wave loads. This framework utilizes some recently
ISORM developed approaches which are based on the inverse first- and second-order reliability methods
(IFORM and ISORM). Using the IFORM and ISORM approaches, characteristic values of the long-
term extreme response are calculated in an efficient manner. By comparing with results obtained
by full numerical integration, the accuracy of the methods is investigated. Particularly the ISORM
method is seen to provide high accuracy. The full long-term analysis is also compared with the
environmental contour method.

1. Introduction

Fjord crossing technology is currently a research topic of high interest in Norway. Due to the width and depth of the fjords
considered, floating bridges represent very relevant concepts as they utilize bouyancy for vertical support. The design of more
extreme yet reliable fjord crossing structures motivates development of the methods for long-term stochastic extreme response
analysis.
For the evaluation of extreme response of marine structures due to environmental loads, a full long-term analysis is recognized as
the most accurate design approach [1]. In principle, the full long-term approach takes into account all possible combinations of
environmental parameters. This means that for straightforward methods such as full numerical integration and crude Monte Carlo
simulation, a very large number of short-term response calculations have to be conducted. For complex structures like floating
bridges, each short-term calculation is usually very time consuming, and the full long-term approach is often considered infeasible.
As an alternative to performing full long-term analyses, the environmental contour method [2,3] is a widely used simplified
approach. First, environmental contours corresponding to specified annual exceedance probabilities are determined without any
consideration of the structural response. Traditionally the contours are determined using an inverse first-order reliability method
(IFORM) approach [4], but alternative methods do exist [5–7]. The most critical point along the contour is then determined, and an
estimate for the long-term extreme response is finally obtained. Only a few short-term response calculations are used, making the
environmental contour method highly efficient. However, some sort of calibration against full long-term analysis is required [1].
Also, due to simplified modelling of response variability, the environmental contour method may perform poorly for certain types of

∗
Corresponding author. Multiconsult, Nedre Skøyen vei 2, 0213 Oslo, Norway.
E-mail address: fi[email protected] (F.-I.G. Giske).

https://doi.org/10.1016/j.marstruc.2017.11.010
Received 28 August 2017; Received in revised form 9 November 2017; Accepted 20 November 2017
0951-8339/ © 2017 Elsevier Ltd. All rights reserved.
F.-I.G. Giske et al. Marine Structures 58 (2018) 154–171

structures [8,9].
In recent years, efforts have been made to reduce the number of short-term calculations required for full long-term extreme
response evaluation. These are based on the observation that many combinations of environmental parameters contribute little or
nothing to the long-term extreme response. A copula based environmental modelling approach is proposed in Ref. [10], and the
copula concept is further utilized in an adaptive refinement algorithm for more efficient long-term integration. In Ref. [11] an IFORM
approach [4] is used to provide an estimate of the long-term extreme response. The IFORM method also indicates where the largest
contribution to the long-term response is located, and this information is used in an importance sampling Monte Carlo simulation
approach, improving the accuracy of the extreme response estimate. Further investigation of IFORM as a method for long-term
extreme response estimation is carried out in Ref. [12], and in Ref. [13] an inverse second-order reliability method (ISORM) approach
is proposed. These developments provide methods for carrying out full long-term analyses with a limited amount of short-term
response calculations.
IFORM and ISORM are efficient and easily implemented methods, which is important for their practical application to long-term
extreme response analysis. Still, it should be noted that having the long-term extreme response analysis formulated in terms of a
reliability problem, as described in Ref. [12], a variety of methods from the field of structural reliability can also be applied. In
particular, efficient simulation methods such as importance sampling [14,15] and subset simulation [16,17] could be used iteratively
to calculate characteristic response values. Alternatively, efficient methods for reliability-based design optimization (RBDO), e.g. Ref.
[18], could be used for direct calculation.
The long-term analysis can also be made more efficient by improving the efficiency of each short-term response calculation. One
example is the method described in Ref. [19], which is demonstrated in Ref. [20] for pontoon bridges.
In the present paper it is shown how these recent developments can be used to perform full long-term extreme response analyses
for a pontoon floating bridge subjected to first-order wave loads. Specifically, the inverse reliability approaches IFORM and ISORM
[12,13] are applied. It is demonstrated that the efficiency of these methods make full long-term extreme response analyses feasible,
also for complex structures such as floating bridges. Furthermore, the framework proposed in this paper can be used for calibration of
the environmental contour method.

2. Modelling the stochastic dynamic response of pontoon bridges

2.1. Stochastic modelling of waves

∼
For a short-term period of duration T , the sea elevation is modelled as a homogeneous and stationary stochastic process with zero
mean. The sea elevation process is denoted η(x , y, t ) , where x , y are the spatial variables and t is the time variable. Assuming linear
wave theory, the wave number κ (ω) is a function of angular frequency defined by the dispersion relation ω2 = κg tanh(κd ) , and the
cross-spectral density between the wave elevation at two points (xm , ym ) and (xn , yn ) can be expressed in terms of a one-dimensional
wave spectrum Sη(ω) and a spreading function Ψ(θ , ω) as
π
Smn (ω) = Sη(ω) ∫ −π Ψ(θ, ω)e−iκ(ω)(Δx cosθ+Δysinθ)dθ.
Here Δx = xm − xn and Δy = ym − yn are the spatial separations of the points. For details we refer to [19].
The sea elevation is further assumed to be a Gaussian process which means that the cross-spectral densities provide a complete
description of the process. Hence the wave situation is completely described by the wave spectrum Sη(ω) and the spreading function
Ψ(θ , ω) . Various theoretical models given in terms of environmental parameters exist in the literature [21,22]. In this paper we use
the generalized Pierson-Moskowitz spectrum [22] given by
−5 −4
Hs2Tz ⎛ ωTz ⎞⎧ 1 ωTz ⎞ ⎫
Sη(ω) = 2 ⎜ exp − ⎛⎜
⎟ ⎟ ,
8π ⎝ 2π ⎠ ⎨ π ⎝ 2π ⎠ ⎬
⎩ ⎭
where Hs is the significant wave height and Tz is the zero-crossing period. The spreading function is of the cos-2s type, defined by a
mean wave direction Θ relative to the x-axis and an ω-dependent spreading parameter s (ω) as
s (ω)
22s(ω)Γ 2(s (ω) + 1) ⎛ 2 θ − Θ ⎞
Ψ(θ , ω) = ⎜cos ⎟ ,
2π Γ(2s (ω) + 1) ⎝ 2 ⎠

where Γ(⋅) is the gamma function. Fig. 1 shows the wave spectrum Sη(ω) plotted in the nondimensional scale ωTz /2π , and the
spreading function is shown for different values of s (ω) . In this paper we have used a constant spreading s (ω) = 10 , but it could
equally well be defined as ω-dependent.

2.2. Stochastic modelling of first-order wave excitation loads

For pontoon floating bridges the structure will experience wave loads only where the pontoons are located. Considering the
pontoons as rigid bodies, the bridge will experience loads in six degrees of freedom (DOFs) from each pontoon, three translational
DOFs and three rotational DOFs. Thus, for a bridge with N pontoons we have loading in 6N DOFs, and we can define a wave
excitation load vector q(t ) = [q1(t ), q2(t ), … , qN (t )], where qm(t ) denotes the 6-element load vector of pontoon number m. The

155
F.-I.G. Giske et al. Marine Structures 58 (2018) 154–171

Fig. 1. Definition of the directional spectrum.

individual components of the load vector q(t ) can be denoted by qμ (t ) , assigning a global index μ ∈ {1,2, … , 6N } to each DOF.
Consider pontoon m with a local coordinate system (x͠ , y͠ ) , which is located with its origin at the point (xm , ym ) and rotated
counterclockwise with an angle αm relative to the global coordinate system (x , y ) as shown in Fig. 2. The wave excitation loads due to
a regular wave with angular frequency ω in the direction β͠ relative to the x͠ -axis of the pontoon can be computed using linear
potential theory software such as Wadam [23]. The loads are then reported in terms of the 6-element complex transfer function vector
fm (β͠ , ω) . Considering only first-order wave loads, the wave excitation load process qm(t ) corresponding to the wave elevation process
η(x , y, t ) can be obtained by superposition of loads from regular waves. This results in a stationary Gaussian load process q(t ) with
zero mean and a 6N -by-6N cross-spectral density matrix Sq(ω) whose elements are given by
π
Sq q (ω) = Sη(ω)
μ ν
∫ −π Ψ(θ, ω)fμ (θ − αm, ω)fν (θ − αn, ω) e−iκ(ω)(Δx cosθ+Δysinθ)dθ, (1)

where the overline denotes complex conjugation. Here fμ (β͠ , ω) is the μ-th component of the total transfer function vector
f (β͠ , ω) = [f1 (β͠ , ω), f2 (β͠ , ω), ⋯, fN (β͠ , ω)], i.e. the complex transfer function of the DOF μ.
A method for efficient calculation of the cross-spectral density matrix Sq(ω) based on the expression Eqn. (1) is given in Refs.
[19,20]. In Ref. [19] the derivation of the cross-spectral densities is also explained in more detail.

Fig. 2. The local coordinate system of a pontoon.

156
F.-I.G. Giske et al. Marine Structures 58 (2018) 154–171

2.3. Equations of motion

By employing the framework of the finite element method (FEM), the equations of motion describing the linear behaviour of a
floating bridge can be written as
Msu¨(t ) + Csu˙ (t ) + Ksu (t ) = qh(t ),

where Ms , Cs and Ks are the structural system matrices, excluding all fluid-structure interaction contributions; u (t ) is the displacement
vector; qh(t ) is the total hydrodynamic action, including both wave excitation and fluid-structure interaction contributions; t is the
∂u
time variable; and u̇ ≡ ∂t
. The total hydrodynamic action may be decomposed as follows:
∞ ∞
qh(t ) = − ⎛⎜ ∫ −∞ Mh(t − τ )u¨(t )dτ + ∫ −∞ Ch(t − τ )u˙ (t )dτ + Khu(t )⎞ + q(t ).
⎟

⎝ ⎠
Here the first term represents the fluid-structure interaction, with Mh(t ) and Ch(t ) being the time-domain representations of added
mass and added damping respectively, and Kh being the hydrostatic stiffness. The second term, q(t ) , is the wave excitation load.
Convolution integrals in the time domain are equivalent to multiplication in the frequency domain, such that the total hydrodynamic
action may be written as follows by enforcing frequency domain notation:

qˆh(ω) = − (−ω2M
ˆ h(ω) + iωCˆh(ω) + Kh)uˆ (ω) + qˆ(ω).

Here hats denote the frequency domain counterparts of the different quantities. Finally, the equation of motion of the system may
be written on the following compact form, in the frequency domain:
(−ω2M (ω) + iωC (ω) + K )uˆ (ω) = qˆ(ω)
where M (ω) = Ms + M ˆ h(ω) , C (ω) = Cs + Cˆh(ω) and K = Ks + Kh .
The second-order probabilistic properties of zero-mean response and wave excitation processes are fully described by cross-
spectral densities. The stochastic frequency domain problem is easily solved by applying the power spectral density method. The
cross-spectral density matrix of the response is then calculated as

Su (ω) = H (ω)Sq(ω)H (ω) H , (2)

2 −1 H
where H (ω) = (−ω M (ω) + iωC (ω) + K ) and [⋅] denotes the conjugate transpose. The cross-spectral density matrix Sq(ω) of the
wave excitation load is found as explained in Section 2.2. More details on the subject may be found in e.g. Refs. [1,24–27].

3. Short-term response model for the case study bridge

The case study bridge consists of an S-shaped continuous girder box, which is supported on 20 pontoons. Fig. 3 depicts the most
important geometrical properties of the bridge. 16 symmetrically positioned cables provide side-support by fixation to the sea bed, cf.
Fig. 4. The cross section of the girder is illustrated in Fig. 5. It is highlighted that the modelled bridge is considered merely as a useful
example for the application of the methodology, and does not necessarily represent a feasible design.

3.1. Numerical response model set-up

The study carried out is performed using the approach presented in Ref. [27], and the reader is referred to that paper for a detailed
description of the methodology for the numerical model set-up. The most important details are repeated here, for the convenience of
the reader.
Two different sub-structures are used to create the full bridge model:

(I) A structural sub-structure, based on an Abaqus model incorporating all structural components and also including pontoon inertia
and buoyancy.
(II) A hydrodynamic sub-structure, based on a Wadam model providing fluid-structure interaction terms, but excluding the buoyancy
and pontoon inertia which are both included in (I).

To combine the two sub-structures, a modal decomposition is carried out in Abaqus [28]. The resulting mode shapes are referred
to as the dry mode shapes, and are used as a new coordinate basis. The mode shapes are defined by the DOFs characterizing the rigid
body motion of all pontoons. The frequency-dependent mass and damping contributions originating from the hydrodynamic model
(II) are transformed to the coordinate basis defined by the dry mode shapes, before they are added to the modal system matrices from
the structural model (I). It is noted that the results from the single pontoon analysis is duplicated and used for all pontoons, but
necessary transformations and matrix book keeping are applied such that the orientation and additions are correct. The wave ex-
citation cross-spectral density matrix Sq(ω) , given by Eqn. (1), is transformed to the coordinate basis given by the dry mode shapes,
before the power spectral density method, cf. Eqn (2), is applied to calculate the spectral density of the response. In the final step, the
response spectral density is transformed back to the physical DOFs of the pontoons. The main reason for carrying out this basis
transformation is to avoid the extraction of all the free DOFs of the finite element model, as static condensation is not appropriate for

157
F.-I.G. Giske et al. Marine Structures 58 (2018) 154–171

Fig. 3. Geometry of the bridge model.

Fig. 4. Position of cables. L refers to the horizontal distance, as defined in Fig. 3.

158
F.-I.G. Giske et al. Marine Structures 58 (2018) 154–171

Fig. 5. Main dimensions of cross section.

dynamic problems. It should be noted that although a reduced order model is obtained, this is not a mode by mode approach because
the modes will be coupled due to the hydrodynamic contributions.

3.2. Structural model

The continuous girder box is modelled in Abaqus with beam elements, with a generalized cross section characterized by the
parameters shown in Table 1. The cables are modelled as beam elements, with circular cross sections.
In an initial static step, pre-tensioning of cables, cable bouyancy, global gravity, and static uplift forces are applied to the
structure. The cables are pre-tensioned by assuming a constant negative temperature, which corresponds to a pre-tension of ap-
proximately 5000 kN. Fluid inertial effects are included for the cables; however, no drag damping is considered.

3.3. Hydrodynamic pontoon model

A single hydrodynamic analysis, carried out in Wadam, is used to establish all system matrix contributions from the fluid-structure
interaction. The geometry of the pontoon is depicted in Fig. 6. In the model set-up, buoyancy and inertia of the pontoon itself were
added as local contributions to the bridge at the locations of the pontoons. The added hydrodynamic mass and damping coefficients,
referring to the local coordinate system of the pontoon, are plotted in Figs. 8 and 9.

3.4. Modal parameters and shapes

Due to the frequency dependency of the hydrodynamic contributions, the eigenvalue problem is solved by iteration, as described
in Ref. [27]. The resulting 10 first undamped natural frequencies and critical damping ratios are shown in Table 2, and the real part of
the corresponding mode shapes are depicted in Fig. 7. Fig. 7 reveals that the first 10 modes all have lateral motion patterns. From
frequencies above the natural frequency of mode 10 and up, numerous cable modes are present.

4. Long-term extreme response

For the modelling of long-term extreme response of marine structures, the long-term situation can be considered as a collection of
∼ ∼
N short-term states, each of duration T . During each short-term state the environmental processes are assumed stationary and
defined by a set of n environmental parameters W = [W1, W2, … , Wn]. In this paper, we only consider the sea elevation, which is
defined in terms of the environmental parameters W = [Hs , Tz , Θ], cf. Section 2.1. We will assume that the joint probability density
function (PDF) of the environmental parameters, denoted fW (w ) , is given. This PDF can be estimated by fitting a probabilistic model
to a scatter diagram of recorded sea states [1].
The methodology presented in this paper for the calculation of extreme response is illustrated for a single response process.
Specifically, we consider the horizontal transverse displacement of pontoon number five from the left in Fig. 3. This is the dis-
placement along the local x͠ -axis of this pontoon (see Fig. 2), and will henceforth simply be referred to as the response process,
denoted R(t ) . Being the response of a linear and time-invariant dynamical system, R(t ) will be a stationary Gaussian process with zero
mean because the load process is. Hence, the response R(t ) is fully characterized by its spectral density SR(ω) , which is obtained as a
diagonal element of the cross-spectral density matrix Su (ω) given by Eqn. (2). Fig. 10 shows an example of the response spectrum
SR(ω) for a short-term situation where the environmental variables are given by W = [Hs , Tz , Θ] = [1 m, 6 s, − π /2].

Table 1
Parameters used for the generalized cross section.

Parameter Value Description

A 1.026 m2 Cross-sectional area

Iy 10.79 m4 Second moment of area about axis y
Iz 29.34 m4 Second moment of area about axis z
J 24.92 m 4 Polar moment of area
zc 3736 mm Distance from bottom to neutral axis
yc 0 mm Distance from center to neutral axis

159
F.-I.G. Giske et al. Marine Structures 58 (2018) 154–171

Fig. 6. Main dimensions of pontoon and local coordinate system.

4.1. Short-term extreme value distribution

The maximal value of the response process R(t ) during a short-term period with given environmental variables W will be a
∼
random variable. This short-term extreme response is denoted by R W and its cumulative distribution function (CDF) is
∼ ∼
FR W (r w ) = Prob[R ≤ r W = w] = Prob[R ≤ r Hs = hs , Tz = tz , Θ = θ ]. As explained in detail in Ref. [1], the short-term extreme
∼

value distribution FR∼ W (r w ) can be found by assuming independent upcrossings of high levels r as
∼
⎧ T m2(w ) r 2 ⎫⎫
FR∼ W (r w ) = exp − exp⎧ − ,
⎨ 2π m0(w ) ⎨
⎩ 2m ⎭⎬
0 (w ) ⎬ (3)
⎩ ⎭
which holds for reasonably large values of r. Here the i-th moment mi (w ) of the response spectrum SR(ω) is defined as
∞
mi (w ) = ∫0 ωiSR(ω)dω.

Note that SR(ω) is dependent on the environmental parameters w , though not written explicitly.
It should be noted that although Eqn. (1) and thereby Eqn. (3) are based on the assumption of homogeneity, which may be
questioned for floating bridge applications, the general method presented in this paper is readily used along with other ways of
calculating the short-term CDF FR∼ W (r w ) . The only required assumption is that the response process can be approximated as sta-
∼
tionary for some short-term period T .

4.2. Long-term extreme response models

∼
The long-term CDF of the short-term extreme value is denoted FR∼(r ) , and gives the distribution of the largest response value R
during an arbitrarily chosen short-term condition. This can be obtained as an average of all short-term CDFs FR∼ W (r w ) weighted by
the distribution fW (w ) of the environmental parameters. In order to estimate fW (w ) in the first place, an ergodicity assumption is
required for the environmental parameters [29], and hence FR∼(r ) should be expressed as an ergodic average [1,29]. This yields the
long-term extreme response formulation

⎧ ⎫
FR∼(r ) = exp
⎨
∫ (lnFR∼ W (r w ))fW (w )dw .
⎬
⎩w ⎭ (4)
A very common approximate formulation, is given by the population mean

FR∼(r ) ≈ ∫ FR∼ W (r w)fW (w)dw.

w (5)
The formulations Eqns. (4) and (5) are discussed in more detail in Refs. [11,12].
The long-term CDF FR∼(r ) can be evaluated by solving the integrals in Eqns. (4) and (5) numerically. Unfortunately, full numerical
integration requires a very large amount of short-term response calculations, since the short-term CDF FR∼ W (r w ) must be calculated
for a very large number of environmental conditions. This motivates the use of inverse reliability methods for calculation of long-term
extreme response.

4.3. Writing the long-term CDF in terms of a reliability problem

In order to use reliability methods for evaluating the long-term CDF FR∼(r ) , it must be rewritten in terms of a reliability problem. A
reliability problem in the general sense [30] is an integral of the form

160
F.-I.G. Giske et al. Marine Structures 58 (2018) 154–171

Fig. 7. Mode shapes from numerical eigenvalue solution, corresponding to natural frequencies and damping ratios presented in Table 2. Note that the eigenvectors are
complex, and their mode shape representation is therefore a snapshot. (a) Mode 1; (b) mode 2; (c) mode 3; (d) mode 4; (e) mode 5; (f) mode 6; (g) mode 7; (h) mode 8;
(i) mode 9; (j) mode 10.

∫ fV (v )dv,
G(v ) ≤ 0

where V is a random vector with joint PDF fV (v ) and G(v ) is a function referred to as the limit state function.
For the approximate formulation Eqn. (5), it is well known that the long-term CDF can be expressed in terms of a reliability
problem by rewriting

∫ FR∼ W (r w)fW (w)dw = ∫ ∫ fR∼ W (r͠ w)drf͠ W (w)dw.

w w r͠ ≤ r

∼
Introducing the random vector V = [W , R ], whose joint PDF is given by fV (v ) = fR∼ W (r͠ w )fW (w ) , Eqn. (5) yields

161
F.-I.G. Giske et al. Marine Structures 58 (2018) 154–171

Fig. 8. Translational (a) and rotational (b) damping coefficients of a single pontoon. Dots indicate the original data from Wadam, whilst lines represent interpolated
data. The coordinates refer to the local coordinate system of the pontoon, as shown in Fig. 6.

FR∼(r ) ≈ ∫ fV (v )dv = 1 − ∫ fV (v )dv .

r͠ ≤ r r ≤ r͠

Finally, we obtain

FR∼(r ) ≈ 1 − ∫ fV (v )dv ,
Gr (v ) ≤ 0 (6)

where Gr (v ) = r − r͠ = r − vn + 1, with vn + 1 being the (n + 1) -th component of the vector v .

The exact formulation Eqn. (4) can be used directly to obtain a better approximation for the long-term CDF in terms of a reliability
problem [12,13,31]. Equation (4) is rewritten by multiplying and dividing the integral by some freely chosen constant C ≥ 1. Then
unity is added and subtracted, keeping in mind that ∫w fW (w )dw = 1. Specifically, we obtain

⎧ ⎛
FR∼(r ) = exp −C ⎜1 −
⎨
∫w ⎛1 + C1 lnFR∼ W (r w)⎞fW (w)dw⎞⎟⎫⎬.
⎜ ⎟

⎩ ⎝ ⎝ ⎠ ⎠⎭
1 1
Introducing the random variable Y defined by the CDF FY W (y w ) = max{1 + C
lnFR∼ W (y w ), 0} , the factor 1 + C
lnFR∼ W (r w ) in the
above integral can be replaced by FY W (r w ) . This yields the approximation

⎧
FR∼(r ) ≈ exp −C ⎜⎛1 −
⎨ ⎝
∫w FY W (r w)fW (w)dw⎞⎫⎬.
⎟

⎩ ⎠⎭ (7)
1
Here the domain where 1 + w ) < 0 is disregarded. This is a very good approximation for large values of r, since
C
lnFR∼ W (r
FR∼ W (r w ) will be close to unity. Furthermore, by increasing the value of C, the approximation will improve. Now the approximation

162
F.-I.G. Giske et al. Marine Structures 58 (2018) 154–171

Fig. 9. Translational (a) and rotational (b) mass coefficients of a single pontoon. Dots indicate the original data from Wadam, whilst lines represent interpolated data.
The coordinates refer to the local coordinate system of the pontoon, as shown in Fig. 6.

Table 2
Modal parameters from the numerical eigenvalue solution, corresponding to mode shapes illu-
strated in Fig. 7. The undamped natural frequency is denoted ωn , and ξ is the corresponding
critical damping ratio.

Mode number ωn [rad/s] ξ [%]

Mode 1 (Fig. 7a) 0.11 1.53

Mode 2 (Fig. 7b) 0.15 1.28
Mode 3 (Fig. 7c) 0.16 1.18
Mode 4 (Fig. 7d) 0.18 1.02
Mode 5 (Fig. 7e) 0.20 0.88
Mode 6 (Fig. 7f) 0.24 0.79
Mode 7 (Fig. 7g) 0.28 0.75
Mode 8 (Fig. 7h) 0.33 0.81
Mode 9 (Fig. 7i) 0.38 1.06
Mode 10 (Fig. 7j) 0.45 1.87

Eqn. (7) obtained directly from the exact formulation Eqn. (4) can be written in terms of a reliability problem using the same
approach as for the approximate formulation. Finally, the long-term CDF is expressed as

⎧ ⎫
FR∼(r ) ≈ exp −C
⎨
∫ fV (v )dv ,
⎬
⎩ Gr (v ) ≤ 0 ⎭ (8)

163
F.-I.G. Giske et al. Marine Structures 58 (2018) 154–171

Fig. 10. The response spectrum SR(ω) when W = [Hs , Tz , Θ] = [1 m, 6 s, − π /2].

where V = [W , Y ] and Gr (v ) = r − y = r − vn + 1.

4.4. Calculation of extreme response using inverse reliability methods

When long-term extreme responses are calculated for design purposes, we usually seek the characteristic response value rq which
has a specified annual exceedance probability q. This may also be referred to as the response value with a return period of 1/ q years,
or simply the 1/ q -year response. The characteristic response rq is found by requiring
q
1 − FR∼(rq) = ∼ ,
N
∼ ∼ ∼
where N = 1 yr/ T is the number of short-term periods in one year. In this paper we have used T = 3 h, which gives
∼
N = 365⋅8 = 2920 . If we denote by r͠q the long-term extreme response obtained when using the approximate formulation Eqn. (5) for
the long-term CDF, we have from Eqn. (6) that r͠q must satisfy
q
∫ fV (v )dv = ∼ .
N
Gr͠q (v ) ≤ 0 (9)

Similarly, using Eqn. (8), which corresponds to the exact formulation Eqn. (4), yields the following equation for rq :

1 ⎛ q
∫ fV (v )dv = − ln⎜1 − ∼ ⎟⎞.
Grq(v ) ≤ 0
C ⎝ N⎠
(10)

Now the problem of finding r͠q and rq that satisfies Eqns. (9) and (10) can be solved in an approximate manner using inverse
reliability methods. Taking Eqn. (9) as an example, the random vector V is transformed into a vector U of independent standard
normal random variables by the Rosenblatt transformation U = T(V ) , and Eqn. (9) becomes
q
∫ fU (u )du = ∼ ,
N
gr͠ (u ) ≤ 0
q (11)
−1
where gr͠ (u ) = Gr͠q(T (u )) = r͠q − r͠ (u ) is the transformed limit state function and fU (u ) is the multivariate standard normal PDF.
q
Using the first-order reliability method (FORM) to approximate the integral in Eqn. (11), the inverse FORM (IFORM) problem can be
derived as

r͠ qF = maxr͠ (u );  subject to  u = β, (12)

∼
where β = − Φ (q/ N ) with Φ(⋅) being the standard normal CDF. Thus, solving the IFORM problem Eqn. (12) provides an estimate r͠ qF
−1

for the characteristic extreme response value r͠q . In this work we have used the solution algorithm proposed in Ref. [12]. For details on
the transformation to standard normal variables and the derivation of the IFORM problem the reader is referred to [4,12,30].
If, on the other hand, the second-order reliability method (SORM) is used to approximate the integral in Eqn. (11), an inverse
SORM (ISORM) method can be derived. In Ref. [13] an ISORM approach is proposed where the IFORM problem Eqn. (12) is solved
repeatedly, updating the value of β which is unknown in this case. The characteristic extreme response estimate provided by the
ISORM method is denoted r͠ qS .
The inverse reliability methods IFORM and ISORM can be applied to Eqn. (10) using the same approach as described above,
providing long-term extreme response estimates that approximate rq . We denote these estimates by rqF and rqS respectively. The only

164
F.-I.G. Giske et al. Marine Structures 58 (2018) 154–171

⎛ 1 q ⎞
differences will be that V is transformed instead of V , and that β = − Φ−1⎜− C ln⎛⎜1 − ∼ ⎞⎟⎟.
N
⎝ ⎝ ⎠⎠
It is reported in Refs. [12,13] that the use of reliability methods appears to give good accuracy for the calculated long-term
extreme response while keeping the number of required short-term response calculations within reasonable levels.

4.5. Environmental contour method

Even though IFORM and ISORM represent efficient methods for extreme response evaluation, some cases may still call for a more
simplified approach. The environmental contour method has been proposed as such a simplified approach for estimating char-
acteristic long-term extreme response values [3]. It is developed in Ref. [4] based on the approximate formulation Eqn. (5) and the
IFORM approximation. In fact, the method can be considered as a special case of the IFORM problem Eqn. (12) where the short-term
extreme response is regarded as deterministic [4].
The environmental contour corresponding to a given annual exceedance probability q is found from the joint environmental PDF
fW (w ) without any consideration of the structural response. Then, the most unfavourable combination of environmental parameters
along this q-probability contour, referred to as the design point, is identified. In this paper the design point, denoted by ŵ , is taken as
the point along the contour where the median, i.e. the 0.5-fractile, of the short-term distribution FR∼ W (r w ) attains its maximal value.
This maximization problem is the same as the IFORM problem Eqn. (12), but since the extreme response is regarded deterministic the
dimension is reduced by one. Nevertheless, the same solution algorithm can be applied to obtain the design point ŵ . In order to
account for the randomness of the short-term extreme value, the characteristic response value is chosen as the p-fractile, p > 0.5, of
the short-term extreme value distribution FR∼ W (r wˆ ) at the design point. The appropriate value for p must be validated by a full long-
term analysis [1].
It is worth mentioning that it is possible to derive an environmental contour method based on the IFORM solution of the exact
formulation Eqn. (4). In that case, we would use the short-term distribution FY W (y w ) instead of FR∼ W (r w ) , and the q-probability
⎛ 1 q ⎞ ∼
contour would be defined in the standard normal space by a radius β = − Φ−1⎜− C ln⎛⎜1 − ∼ ⎞⎟⎟ instead of β = − Φ−1(q/ N ) . This would,
N
⎝ ⎝ ⎠⎠
however, introduce contours dependent on the parameter C, and the appealing simplicity of the contour method would be under-
mined.

5. Numerical results

5.1. Environmental models

The environmental parameters defining the short-term wave situation according to Section 2.1 are the significant wave height Hs ,
the zero-crossing period Tz and the mean wave direction Θ. Using the conditional modelling approach described in Refs. [32,33], the
CDF of the significant wave height Hs is given by a 2-parameter Weibull distribution
β
⎧ h ⎫
FHs (h) = 1 − exp −⎛⎜ ⎞⎟ ,
⎨ ⎝α⎠ ⎬
⎩ ⎭ (13)

and the zero-crossing period Tz has a conditional lognormal distribution

lnt − μ(h) ⎞
FTz Hs (t h) = Φ⎛⎜ ⎟,

⎝ σ (h) ⎠ (14)

where μ(h) = a0 + a1ha2 and σ (h) = b0 + b1e b2h . Here α, β and a0 , a1, a2 , b0 , b1, b2 are the parameters of the distributions. For the
mean wave direction Θ, we use a distribution independent of Hs and Tz , given by the CDF

⎧ 0, for θ < − π ,
⎪ θ 2 π
⎪ 2(1 + π ) , for  − π ≤ θ < − 2 ,
FΘ(θ) =
⎨1 − 2( θ )2 , for  − π
≤ θ < 0,
⎪ π 2
⎪1, for θ ≥ 0. (15)
⎩
This means that the PDF fΘ (θ) , obtained by differentiating Eqn. (15) with respect to θ, is piecewise linear between −π and 0 with a
π
peak at − 2 . Similarly, the PDFs fH (h) and fT H (t h) can be obtained by differentiating Eqns. (13) and (14) with respect to h and t
s z s
respectively, and the joint PDF of the environmental parameters is given as
fW (w ) = fH , T ,Θ (h, t , θ) = fH (h)fT (t h)fΘ (θ). (16)
s z s z Hs

The environmental model Eqn. (16) where all three environmental parameters are random variables will be referred to as EM1.
Different environmental models can be obtained by considering some of the environmental parameters as deterministic. If for in-
stance the zero-crossing period is taken as the conditional median obtained from the CDF Eqn. (14), i.e. Tz = exp{μ(Hs )} , we obtain the

165
F.-I.G. Giske et al. Marine Structures 58 (2018) 154–171

Table 3
Overview of the considered environmental models.

Joint PDF Hs Tz Θ
Name
Value/CDF α β Value/CDF a0 a1 a2 b0 b1 b2 Value/CDF

EM1 Eqn. (16) Eqn. (13) 0.587 1.59 Eqn. (14) 0.151 0.339 0.167 0.07 0.3449 −0.6219 Eqn. (15)
EM2 Eqn. (17) Eqn. (13) ” ” exp{μ(Hs )} ” ” ” n/a n/a n/a Eqn. (15)
EM3 Eqn. (18) Eqn. (13) ” ” Eqn. (14) ” ” ” 0.07 0.3449 −0.6219 −π/2
EM4 Eqn. (18) Eqn. (13) 0.550 1.53 Eqn. (14) −0.120 1.439 0.150 0.07 0.0978 −0.0382 −π/2

environmental model

fW (w ) = fH ,Θ (h, θ) = fH (h)fΘ (θ). (17)

s s

This will be referred to as EM2. We also consider an environmental model where the mean wave direction is given as Θ = − π /2 .
This yields

fW (w ) = fH , T (h, t ) = fH (h)fT (t h), (18)

s z s z Hs

which will be referred to as EM3. The environmental models EM1, EM2 and EM3 will all have the same values for the distribution
parameters. We also consider a model EM4, which is given by Eqn. (18) with different parameter values. An overview of the
environmental models and their distribution parameters is provided in Table 3
The environmental models are illustrated in Fig. 11 by displaying the environmental contours corresponding to annual ex-
ceedance probabilities q = 10−2 and q = 10−4 , i.e. the 100-year and 10 000-year contours. For the two-dimensional models EM2, EM3
and EM4, the isoprobability contours obtained from the PDFs Eqns. (17) and (18) are also shown. Note that EM2 and EM3 are
obtained from EM1 by regarding as deterministic Tz and Θ respectively. EM4 represent a different model entirely. However, for all the
models considered, the significant wave heights Hs with return periods of 100 and 10 000 years are approximately 2.9 m and 3.5 m
respectively.

Fig. 11. The contours corresponding to annual exceedance probabilities q = 10−2 and q = 10−4 for the different environmental models. For the two-dimensional
models EM2, EM3 and EM4 the PDFs are illustrated by displaying the isoprobability contours. EM2 and EM3 are obtained from EM1 by regarding as deterministic Tz
and Θ respectively.

166
F.-I.G. Giske et al. Marine Structures 58 (2018) 154–171

Table 4
The characteristic extreme response values as calculated by the different methods for an annual exceedance probability q = 10−2 .

Approximate formulation Exact formulation

r͠ qF [cm] r͠ qS [cm] r͠q [cm] rqF [cm] rqS [cm] rq [cm]

C=1 C = 104 C = 106

EM1 56.4 48.7 54.4 56.8 49.6 58.7 58.5 61.8

EM2 2.52 2.39 2.42 2.54 2.42 2.55 2.55 2.58
EM3 53.3 49.7 51.6 53.7 50.6 61.0 60.8 62.1
EM4 246.8 243.4 243.4 249.2 248.3 284.9 284.6 284.6

5.2. Characteristic extreme response values

5.2.1. Inverse reliability methods

Estimates for the characteristic response value rq were calculated using the methods described in Section 4.4. The value r͠q was
obtained by numerically solving the integral in the approximate formulation Eqn. (5), and r͠ qF and r͠ qS denote the IFORM and ISORM
approximations of r͠q . Similarly, rq was found by applying numerical integration to the formulation Eqn. (4) and the reliability method
approximations are denoted rqF and rqS . For rqF and rqS , different values of the constant C in Eqns. (8) and (10) could be used. In this
paper, C = 1 is used for rqF , while the values C = 1, C = 104 and C = 106 are used for rqS .
For the calculation of r͠q and rq by numerical integration, the ranges of the integration variables were Hs ∈ [0,10] m, Tz ∈ [0.4, 20] s,
and Θ ∈ [ − π , 0]. The applied bin sizes were ΔHs = 0.1 m, ΔTz = 0.2 s, ΔΘ = π/39 for EM1, ΔHs = 0.05 m, ΔΘ = π/39 for EM2 and
ΔHs = 0.05 m, ΔTz = 0.05 s for EM3 and EM4. It should be noted that these ranges and bin sizes are chosen such that r͠q and rq can be
regarded as exact values, and the number of integration points may therefore be excessive.
The obtained values for the characteristic extreme response estimates are presented in Tables 4 and 5 for annual exceedance
probabilities q = 10−2 and q = 10−4 , respectively. When compared to the values r͠q and rq , it is seen that the reliability method ap-
proximations provide reasonable estimates for the characteristic response value. Especially the ISORM method with C chosen as 104
or 106 yields very good estimates.
In Tables 4 and 5, the characteristic response values are seen to vary quite a lot between the different environmental models. This
is a result of the response being very sensitive to the zero-crossing period Tz . In Fig. 11 it is seen that large values of Tz have a larger
probability of occurrence for EM4 than for EM3, resulting in a significantly larger extreme response. For EM2, Tz is fixed at its median
value, disregarding large values of Tz . This results in a smaller extreme response for EM2. EM1 and EM3, on the other hand, have the
same model for Tz and give quite similar results.
For each of the extreme response estimates in Tables 4 and 5, the corresponding number of executed short-term response cal-
culations, denoted by nst or similar, is reported in Tables 6 and 7 It is clear that IFORM and ISORM represent efficient methods for full
long-term extreme response analysis. ISORM roughly doubles the computational effort compared to IFORM.

5.2.2. Environmental contour method

Using some common choices for the fractile level p, characteristic extreme response estimates denoted rqp were obtained for the
environmental contour method. These estimates are presented in Tables 8 and 9 for annual exceedance probabilities q = 10−2 and
q = 10−4 respectively. Comparing these results to the exact long-term extreme response rq in Tables 4 and 5, we observe that all the
considered choices of p give reasonable rough estimates for the long-term response.
The exact fractile levels corresponding to the full long-term estimates can also be calculated. For the exact extreme response value
rq , the corresponding fractile level is given as
pq = FR∼ W (rq wˆ ).

Table 5
The characteristic extreme response values as calculated by the different methods for an annual exceedance probability q = 10−4 .

Approximate formulation Exact formulation

r͠ qF [cm] r͠ qS [cm] r͠q [cm] rqF [cm] rqS [cm] rq [cm]

C=1 C = 104 C = 106

EM1 97.0 83.5 85.2 97.7 85.2 92.8 92.4 92.6

EM2 3.47 3.30 3.33 3.47 3.33 3.39 3.39 3.41
EM3 97.0 87.5 87.8 97.8 89.1 96.8 96.7 97.2
EM4 406.3 398.5 398.0 408.2 403.9 420.3 420.1 420.1

167
F.-I.G. Giske et al. Marine Structures 58 (2018) 154–171

Table 6
The number of short-term response calculations performed for each of the long-term extreme response estimates in Table 4.

Approximate formulation Exact formulation

n͠ stF n͠ stS n͠ st nstF nstS nst

C=1 C = 104 C = 106

EM1 19 87 399960 23 111 111 113 399960

EM2 43 81 8040 43 81 71 67 8040
EM3 61 113 78993 60 117 225 147 78993
EM4 33 67 78993 33 67 81 177 78993

Table 7
The number of short-term response calculations performed for each of the long-term extreme response estimates in Table 5.

Approximate formulation Exact formulation

n͠ stF n͠ stS n͠ st nstF nstS nst

C=1 C = 104 C = 106

EM1 13 104 399960 18 110 109 126 399960

EM2 57 95 8040 57 94 89 85 8040
EM3 160 276 78993 256 384 167 155 78993
EM4 101 159 78993 105 151 82 190 78993

Table 8
The characteristic extreme response values as calculated by the environmental contour method using different fractile levels p for an annual exceedance prob-
ability q = 10−2 . The corresponding design points are illustrated by diamond markers in Figs. 12 and 13.

rqp [cm]

p = 0.80 p = 0.85 p = 0.90 p = 0.95

EM1 58.6 59.6 61.0 63.3

EM2 2.60 2.64 2.69 2.79
EM3 55.3 56.2 57.6 59.7
EM4 258.5 263.3 269.7 280.1

Table 9
The characteristic extreme response values as calculated by the environmental contour method using different fractile levels p for an annual exceedance prob-
ability q = 10−4 . The corresponding design points are illustrated by diamond markers in Figs. 12 and 13.

rqp [cm]

p = 0.80 p = 0.85 p = 0.90 p = 0.95

EM1 101.0 102.9 105.4 109.5

EM2 3.50 3.56 3.63 3.76
EM3 100.9 102.8 105.4 109.5
EM4 418.1 425.9 436.5 453.4

Table 10
The fractile levels pq corresponding to the exact extreme response values rq in Tables 4 and 5

EM1 EM2 EM3 EM4

pq ,.q=10−2 0.92 0.77 0.98 0.96

pq , q = 10 −4 0.43 0.69 0.66 0.81

168
F.-I.G. Giske et al. Marine Structures 58 (2018) 154–171

Fig. 12. The design points corresponding to the characteristic extreme response value rqS (ISORM, C = 106 ) and the environmental contour method for annual
exceedance probabilities q = 10−2 (left) and q = 10−4 (right). The contribution g(w ) to the long-term integral is also illustrated by displaying isosurfaces for the values
0.9, 0.5 and 0.1.

Table 10 shows the fractile levels corresponding to the exact extreme response values rq in Tables 4 and 5 We see that there is a
large variation in the obtained fractiles, indicating that one single fractile level does not give accurate estimates for all the considered
cases. However, as seen in Tables 8 and 9, rough estimates can still be obtained. When regarded as rough approximations, Tables 8
and 9 show that the extreme response estimates are generally not overly sensitive to changing fractile levels. Still, if the fractile level
should be much larger than 0.9, which is the case for EM3 and EM4 when q = 10−2 , the environmental contour method may un-
derestimate the extreme response quite severely.
Considering Tables 8–10, reasonable choices for the fractile values are perhaps p = 0.95 for q = 10−2 and p = 0.80 for q = 10−4 .
Thus, p has a larger value for the highest annual exceedance probability. This is in contrast to the choices of p = 0.90 for q = 10−2 and
p = 0.95 for q = 10−4 , which are common for offshore structures [3]. It should also be noted that instead of using rq , which is obtained
by full numerical integration, the IFORM and ISORM estimates can be used to determine appropriate values for the fractile levels.

5.3. Design points

In addition to giving an estimate for the characteristic extreme response, the inverse reliability methods will produce a design
point which represents the most critical combination of environmental parameters for the specified annual exceedance probability q.
The design point corresponding to the ISORM estimate rqS (C = 106 ) is shown in Fig. 12 for EM1. In Fig. 13 the design points are shown
for the two-dimensional environmental models EM2, EM3 and EM4, also including the IFORM design points corresponding to the
estimates rqF (C = 1). In addition, the contour method design points are shown in Figs. 12 and 13. As explained in Section 4.5, these
have been obtained by maximizing the median value of the short-term CDF FR∼ W (r w ) on the respective contours.
The relative contribution of different sea states to the long-term integral in Eqn. (4) is illustrated in Figs. 12 and 13 by the function
g(w ) . This function is defined as a normalized version of the integrand in Eqn. (4) for r = rq . Specifically,

1
g (w ) = − ln(FR∼ W (rq w ))fW (w ),
M

where M is chosen such that the maximal value of g(w ) equals unity.
By considering Figs. 12 and 13, we observe that the main contribution to the long-term integral is located within a rather
concentrated region. Furthermore, the design points quite successfully locate this region. The ISORM design point (C = 106 ) almost
exactly pinpoints the location of the largest contribution. However, local maxima other than the main contribution might occur. This
can be observed in the left part of Fig. 12, corresponding to q = 102 for EM1. If such a local maximum represent a significant
contribution, this may result in an underestimation of the long-term extreme response as seen in the first row of Table 4. This is a
known shortcoming of the inverse reliability methods, and they should therefore be used with some caution.

6. Concluding remarks

A framework for full long-term extreme response analysis has been demonstrated for a long-span case study bridge. Using recently
developed IFORM and ISORM approaches, the extreme response was calculated in an efficient manner. Comparison with full nu-
merical integration revealed that especially the ISORM method gives high accuracy. It has thus been shown that the proposed
framework can be applied successfully for complex structures. Still, limitations do exist, e.g. in the presence of multiple local maxima
for the contribution to the long-term integral. Therefore, future work should focus on comparison with alternative approaches and
further verification of the IFORM and ISORM methods, especially for nonlinear response.
The full long-term analysis was also compared with the environmental contour method. The results show that the contour method
can be used to obtain rough estimates of the long-term extreme response. Furthermore, a proper fractile level p could be determined
by comparison with the IFORM and ISORM results.

169
F.-I.G. Giske et al. Marine Structures 58 (2018) 154–171

Fig. 13. The design points corresponding to the characteristic extreme response values rqF (IFORM, C = 1), rqS (ISORM, C = 106 ) and the environmental contour method
for annual exceedance probabilities q = 10−2 (left) and q = 10−4 (right). The contribution g(w ) to the long-term integral is also illustrated in each case.

Acknowledgements

The authors are grateful for grants which are provided by Multiconsult ASA and the Research Council of Norway.

References

[1] Naess A, Moan T. Stochastic dynamics of marine structures. Cambridge: Cambridge University Press9781139021364; 2012. http://dx.doi.org/10.1017/
CBO9781139021364http://ebooks.cambridge.org/ebook.jsf?bid=CBO9781139021364.
[2] Haver S, Kleiven G. Environmental contour lines for design purposes: why and when? ASME Conf Proc 2004;2004(37432):337–45. http://dx.doi.org/10.1115/
OMAE2004-51157http://proceedings.asmedigitalcollection.asme.org/proceeding.aspx?articleid=1628932.
[3] Haver S, Winterstein S. Environmental contour lines: a method for estimating long term extremes by a short term analysis. Trans Soc Nav Archit Mar Eng
2010;116:116–27. (October). http://www.sname.org/HigherLogic/System/DownloadDocumentFile.ashx?DocumentFileKey=ccb5aac7-6b77-420b-b233-
836fc6e13597.
[4] Winterstein SR, Ude TC, Cornell CA, Bjerager P, Haver S. Environmental parameters for extreme response: inverse FORM with omission factors. Proc. 6th int.
Conf. Struct. Saf. Reliab. 1993. Innsbruck, Austria http://www.rms-group.org/RMS_Papers/pdf/Todd/innsbruck.pdf.
[5] Huseby AB, Vanem E, Natvig B. A new approach to environmental contours for ocean engineering applications based on direct Monte Carlo simulations. Ocean
Eng 2013;60:124–35. http://dx.doi.org/10.1016/j.oceaneng.2012.12.034http://www.sciencedirect.com/science/article/pii/S0029801812004532.
[6] Huseby AB, Vanem E, Natvig B. Alternative environmental contours for structural reliability analysis. Struct Saf 2015;54:32–45. http://dx.doi.org/10.1016/j.
strusafe.2014.12.003http://linkinghub.elsevier.com/retrieve/pii/S0167473014001143.

170
F.-I.G. Giske et al. Marine Structures 58 (2018) 154–171

[7] Montes-Iturrizaga R, Heredia-Zavoni E. Environmental contours using copulas. Appl Ocean Res 2015;52:125–39. http://dx.doi.org/10.1016/j.apor.2015.05.
007http://www.sciencedirect.com/science/article/pii/S0141118715000747.
[8] Agarwal P, Manuel L. Simulation of offshore wind turbine response for long-term extreme load prediction. Eng Struct 2009;31(10):2236–46. http://dx.doi.org/
10.1016/j.engstruct.2009.04.002https://doi.org/10.1016/j.engstruct.2009.04.002.
[9] Li Q, Gao Z, Moan T. Modified environmental contour method for predicting long-term extreme responses of bottom-fixed offshore wind turbines. Mar Struct
2016;48:15–32. http://dx.doi.org/10.1016/j.marstruc.2016.03.003http://www.sciencedirect.com/science/article/pii/S0951833916300181.
[10] Zhang Y, Beer M, Quek ST. Long-term performance assessment and design of offshore structures. Comput Struct 2015;154:101–15. http://dx.doi.org/10.1016/j.
compstruc.2015.02.029http://www.sciencedirect.com/science/article/pii/S004579491500067X.
[11] Sagrilo L, Naess A, Doria A. On the long-term response of marine structures. Appl Ocean Res 2011;33(3):208–14. http://dx.doi.org/10.1016/j.apor.2011.02.
005http://www.sciencedirect.com/science/article/pii/S0141118711000204.
[12] Giske FIG, Leira BJ, Øiseth O. Full long-term extreme response analysis of marine structures using inverse FORM. Probabilistic Eng Mech 2017;50:1–8. http://dx.
doi.org/10.1016/j.probengmech.2017.10.007.
[13] Giske FIG, Leira BJ, Øiseth O. Long-term extreme response analysis of marine structures using inverse SORM. ASME 2017 36th int. Conf. Ocean. Offshore arct.
Eng 3A. ASME978-0-7918-5765-6; 2017. http://dx.doi.org/10.1115/OMAE2017-61409http://proceedings.asmedigitalcollection.asme.org/proceeding.aspx?
doi=10.1115/OMAE2017-61409.
[14] Au SK, Beck JL. First excursion probabilities for linear systems by very efficient importance sampling. Probabilistic Eng Mech 2001;16(3):193–207. http://dx.
doi.org/10.1016/S0266-8920(01)00002-9.
[15] Papaioannou I, Papadimitriou C, Straub D. Sequential importance sampling for structural reliability analysis. Struct Saf 2016;62:66–75. http://dx.doi.org/10.
1016/j.strusafe.2016.06.002http://www.sciencedirect.com/science/article/pii/S0167473016300169.
[16] Au SK, Beck JL. Estimation of small failure probabilities in high dimensions by subset simulation. Probabilistic Eng Mech 2001;16(4):263–77. http://dx.doi.org/
10.1016/S0266-8920(01)00019-4.
[17] Ching J, Beck JL, Au SK. Hybrid Subset Simulation method for reliability estimation of dynamical systems subject to stochastic excitation. Probabilistic Eng Mech
2005;20(3):199–214. http://dx.doi.org/10.1016/j.probengmech.2004.09.001.
[18] Dubourg V, Sudret B, Bourinet JM. Reliability-based design optimization using kriging surrogates and subset simulation. Struct Multidiscip Optim
2011;44(5):673–90. http://dx.doi.org/10.1007/s00158-011-0653-8http://link.springer.com/10.1007/s00158-011-0653-8.
[19] Giske FIG, Leira BJ, Øiseth O. Efficient computation of cross-spectral densities in the stochastic modelling of waves and wave loads. Appl Ocean Res
2017;62:70–88. http://dx.doi.org/10.1016/j.apor.2016.11.007.
[20] Giske FIG, Leira BJ, Øiseth O. Stochastic Modelling of Wave Loads on Floating Bridges : Efficient Calculation of Cross-Spectral Densities. 19th congr. IABSE,
challenges des. Constr. An innov. Sustain. Built environ. 2016978-3-85748-144-4. p. 48–56. September.
[21] Hauser D, Kahma K, Krogstad H. Measuring and analysing the directional spectra of ocean waves. Luxembourg: Publications Office of the European Union92-
898-0003-8; 2005.
[22] Stansberg CT, Contento G, Hong SW, Irani M, Ishida S, Mercier R, et al. The specialist committee on waves final report and recommendations to the 23rd ITTC.
Proceedings of the 23rd ITTC. 2002. p. 505–736http://ittc.info/media/1469/waves.pdf.
[23] DNV, Wadam 2014. Høvik, Norway.
[24] Sigbjörnsson R. Stochastic theory of wave loading processes. Eng Struct 1979;1(2):58–64. http://dx.doi.org/10.1016/0141-0296(79)90014-2http://www.
sciencedirect.com/science/article/pii/0141029679900142.
[25] Langen I, Sigbjörnsson R. On stochastic dynamics of floating bridges. Eng Struct 1980;2(4):209–16. http://dx.doi.org/10.1016/0141-0296(80)90002-4http://
www.sciencedirect.com/science/article/pii/0141029680900024.
[26] Ochi MK. Applied probability and stochastic processes in engineering and physical sciences. New York: Wiley0471857424; 1990http://books.google.com/
books?ei%3d;JFOBU7TtFYHcObC4gLgK%7b%26%7did%3dID3xAAAAMAAJ%7b%26%7dpgis%3d1.
[27] Kvåle KA, Sigbjörnsson R, Øiseth O. Modelling the stochastic dynamic behaviour of a pontoon bridge: A case study. Comput Struct 2016;165:123–35. http://dx.
doi.org/10.1016/j.compstruc.2015.12.009http://www.sciencedirect.com/science/article/pii/S004579491500334X.
[28] dassault systèmes simulia corp.,. Abaqus. Providence: Rhode Island, USA; 2014.
[29] Naess A. Technical note: On the long-term statistics of extremes. Appl Ocean Res 1984;6(4):227–8. http://dx.doi.org/10.1016/0141-1187(84)90061-0http://
www.sciencedirect.com/science/article/pii/0141118784900610.
[30] Melchers RE. Structural reliability analysis and prediction. second ed. John Wiley & Sons9780471987710; 1999.
[31] Giske FIG, Leira BJ, Øiseth O. Long-term stochastic extreme response analysis of floating bridges. Procedia Eng 2017;199:1175–80. http://dx.doi.org/10.1016/j.
proeng.2017.09.305http://www.sciencedirect.com/science/article/pii/S1877705817337888.
[32] DNV. DNV-RP-C205 Environmental conditions and environmental loads Tech. Rep. October; DNV; Høvik, Norway 2010https://rules.dnvgl.com/docs/pdf/DNV/
codes/docs/2010-10/RP-C205.pdf.
[33] Bitner-Gregersen EM, Haver S. Joint environmental model for reliability calculations. Proc. First int. Offshore polar eng. Conf. Edinburgh, United Kingdom:
International Society of Offshore and Polar Engineers; 1991. p. 246–53https://www.onepetro.org/conference-paper/ISOPE-I-91-031.

171