Marine Structures 59 (2018) 60–79

Contents lists available at ScienceDirect

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

An EWA framework for the probabilistic-based structural integrity

T
assessment of offshore platforms
M. Zeinoddinia, Y. Yaghubi Namina, H. Matin Nikoob,∗, H. Estekanchic, M. Kimiaeid
a
Department of Civil Engineering, K.N.Toosi University of Technology, Tehran, Iran
b
Centre for Infrastructure Monitoring and Protection (CIMP), Curtin University, Perth, Australia
c
Department of Civil Engineering, Sharif University of Technology, Tehran, Iran
d
Centre for Offshore Foundation Systems (COFS), University of Western Australia, Perth, Australia

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

Keywords: Endurance Wave Analysis (EWA) is a novel, simple, economical but yet reliable method for the
Offshore structures integrity assessment of offshore platforms against extreme irregular waves. The current paper
Extreme sea waves presents a framework for the probabilistic assessment of offshore platforms based on EWA. The
Endurance Wave Analysis (EWA) stochastic nature and the uncertainties in the ocean waves were taken into account by in-
Performance-based design
troducing a set of artificial, random and gradually intensifying wave trains. To simulate other
uncertainties in the loading and the strength of the structure a Latin Hypercube Sampling (LHS)
type of Monte Carlo Simulation (MCS) was employed. The EWA methodology and different
structural performance criteria were considered. The EWA results were then used to obtain the
hazard functions and fragility curves for a number of Engineering Demand Parameters (EDPs).
Similar to onshore buildings, and for the sake of comparison only, different structural perfor-
mance criteria were considered. The mean annual probabilities of structural failure (Pf) at dif-
ferent structural performance limit states were also competently estimated. The results proved
that EWA is an effective and robust tool for the probabilistic-based assessment of offshore plat-
forms. It yet requires very short simulation time, as compared to the costly conventional 3-h time-
domain analyses. In contrast to the probabilistic seismic assessments, EWA does not requires
separate Multi Stripe Analyses (MSAs), because EWA is intrinsically multi stripe.

1. Introduction

A reliable structural integrity assessment of an offshore structure needs proper modelling of the sea loads, structure and its
strength [1–5]. Inherent randomness in the wave profiles, variability in the sea parameters, uncertainties in the wave force para-
meters and in the structural modelling approach would have direct impacts on the structural assessment results [6]. Fixed offshore
platforms are commonly designed to maintain a factor of safety under a load corresponding to a metocean event with a probability of
exceedance of 1 in 100 years. With this approach, often referred to as the “environmental overload methodology”, normal design of
an offshore platform will deliver a structure with an ultimate capacity in the range of approximately 1.8–2.5 times the loads from the
100-year event [1]. The ratio between the ultimate lateral capacity and the lateral load from the 100-year event is usually referred to
as “Residual Strength Ratio (RSR)” [7]. The RSR represents an explicit factor of safety incorporated in the design codes as well as an
implicit margin associated with uncertainties e.g. in steel strength, conservatism in foundation capacity, presence of additional steels
for temporary load conditions (construction, transportation and installation) and the increased capacity associated with system

∗
Corresponding author.
E-mail address: [email protected] (H. Matin Nikoo).

https://doi.org/10.1016/j.marstruc.2018.01.002
Received 1 July 2017; Received in revised form 3 December 2017; Accepted 5 January 2018
Available online 06 February 2018
0951-8339/ © 2018 Elsevier Ltd. All rights reserved.
M. Zeinoddini et al. Marine Structures 59 (2018) 60–79

versus component failure of a three-dimensional space-frame structure. The “environmental overload methodology” criteria, how-
ever, requires proper definitions for regional acceptance criteria which accordingly needs to take account of the slope of the regional
environmental hazard curve.
A more consistent assessment, as compared to the “environmental overload methodology”, can be obtained by directly calculating
the probability of failure of the structure. This is applicable to offshore structures in all regions, by taking into account the un-
certainties associated to the system. This requires i) proper modelling tools for predicting the failure response of the structure and ii) a
suitable scheme for incorporating the uncertainties into the analytical model.
Various approaches have already been adopted to predict the failure response of offshore structures under wave loading [8–13].
However, the proposed approaches can be debated from the accuracy, reliability and computational cost viewpoints. For example, in
the framework proposed by Manuel et al. [8] two harmonic wave heights namely design-level and ultimate-level heights are used in
order to design the structural elements according to the target reliability. This method involves carrying out a series of push-over
analyses so inherently neglects the stochastic nature of the ocean waves and requires iterative design cycles. Incremental Wave
Analysis (IWA) and Probabilistic IWA (PIWA) have also been introduced as substitutes to the push-over approach [10,14]. In these
methods, dynamic responses of offshore platforms at different limit states are estimated under extreme regular waves. PIWA provides
a probabilistic framework which can consider potential uncertainties associated in the structural response, structural capacity and the
wave forces. However, the IWA and the PIWA concepts are based on regular waves which can lead to inaccurate assessments, in
particular in dynamically sensitive structures.
A novel, simple but yet reliable method for assessment of the structural performance of offshore platforms against extreme
irregular waves is Endurance Wave Analysis (EWA). EWA can provide an economical integrity assessment framework for offshore
platforms and has already been presented and discussed in a number of research papers [15–19]. Previous published studies on EWA,
however, deal with deterministic modelling of offshore structures. Main aim of this study is to adapt and employ EWA methodology
as an efficient, robust and useful tool for the metocean probabilistic-based structural integrity assessment of an example fixed offshore
platform. Performance-based engineering is well established for structural assessment of high rise buildings but it is not an accepted
practice yet for offshore platforms, though it has been used in few previous research studies [20]. It should be noted that definition of
performance based criteria or accurate estimation of the ultimate strength of the platform are not parts of the current study. Therefore
in order to only lay the ground for future studies, simplified assumptions in structural modelling or the performance-based criteria for
onshore buildings have been used in this paper wherever needed. EWA approach can consider different performance criteria, in-
corporate the associated uncertainties in the system and give a deep insight into the nonlinear behaviour of the offshore structure.

2. Methods and procedures

2.1. Review of the EWA concept

Here the EWA approach is briefly reviewed, but more information can be found in Refs. [15,19]. The concept of EWA is analogous
to the “Endurance Time Analysis (ETA)”. ETA was proposed for the nonlinear seismic time-domain analysis of onshore structure,
using artificial time series of step-wisely intensifying ground accelerations [21–24]. ETA has found wide applications for the seismic
evaluation of the structures.
ETA and EWA are both inspired by the standard protocol (exercise) test which is broadly used by cardiologists in order to evaluate
the physical conditions of the human cardiovascular system [25,26] [27]. Similarly, EWA employs a time-history of gradually in-
tensifying waves for the collapse analyses of offshore structures. This time-history is called the Intensifying Wave Train Function
(IWTF). Fig. 1 demonstrates some of the similarities between a cardiopulmonary exercise test and the EWA approach. As it can be
noticed, similar to a cardiopulmonary exercise test, the nonlinear dynamic response of an offshore structure can be evaluated when it
is simulated under an IWTF. An arbitrary damage index, for example a limit in the deck drift, can be defined as a monitoring
parameter in the temporal response events. The sea-state which causes the structure to reach the predefined damage target can be
easily identified by a single dynamic run.
EWA can be considered as a dynamic time-domain random type of push-over analysis of the structure. EWA advantages are
documented in previous relevant studies [e.g. Ref. [15]]. It allows for modelling the progressively deteriorating trend in the storm
events and takes into account the so called “loading history effect” or “loading memory effect”. It employs Constrained NewWaves
(CNW) which duly represents the stochastic nature of the ocean waves and the variability of the wave height and frequency. EWA, in
fact, is a dynamic time domain type of analysis in which randomness in the loading and nonlinearities in the structural behaviour are
taken into account. It also requires relatively short-time simulations. It can be efficiently employed both in the design of new
platforms and for integrity assessment of the existing platforms.

2.2. Intensifying Wave Train Function (IWTF)

Calibrated IWTFs are integral parts of EWA. To produce a single IWTF, a number of step-wisely intensifying time series of
Constrained NewWaves (CNWs) [28] are joined together. Some previous studies have demonstrated that short-duration CNWs are
proper alternatives to common long-duration random time history simulations in predicting the maximum structural response [e.g.
Refs. [15,29–31]] of offshore platforms under randomly generated waves. The maximum dynamic nonlinear structural responses
delivered by CNW models are, reportedly, pretty comparable to those obtained from long 3-h simulations under extreme waves.
Incorporating CNWs in the IWTF allows for delivering a desirable maximum wave crest height in each sea-state as well as capturing

61
M. Zeinoddini et al. Marine Structures 59 (2018) 60–79

Fig. 1. Comparison between the cardiopulmonary exercise test and Endurance Wave Analysis (EWA).

the randomness of the wave loading. Fig. 2 (left) shows a sample profile of a NewWave. Fig. 2 (right) displays the same NewWave
along with a sample Random Surface Elevation (RSE). RSE is in fact an arbitrary short length (here 100s) of a random 3 h wave. The
random 3 h wave is produced using the relevant sea-state spectrum. The NewWave is then constrained to the RSE to produce the CNW
as shown in the figure. Fig. 3 gives a sample IWTF consisting short-duration irregular CNWs. The maximum wave height in an IWTF
well goes gradually beyond the design sea-state. Zeinoddini et al. [15] used the EWA approach based on intensifying CNWs for a
deterministic integrity assessment of a jacket type offshore structure. They showed that the EWA estimates for the ultimate strength of
the jacket platform were very close to those obtained from large number of very time consuming dynamic analyses of the structure
against 3 h irregular wave.
In evaluation of the ultimate resistance of offshore platforms, temporal evolution of the extraordinary storm should be taken into
account. This should cover the growth and decay parts of the storm [32]. The rare storm which is due to cause a total collapse in the
structure may take deca-hours to grow. The structure, therefor, encounters many large waves before facing the highest wave. In fact,
the devastating wave acts on a structure which had already damaged progressively by previous high waves.
A single CNW model overlooks the memory effects, or the bearings of the previous high waves on the structure. An IWTF is a
sequence of CNWs with growing maximum crest heights. It takes into account the temporal evolution of the storm or the so called

62
M. Zeinoddini et al. Marine Structures 59 (2018) 60–79

Fig. 2. Typical profiles for a NewWave (left) and a CNW (right), in which the NewWave is constrained by a Random Surface Elevation (RSE) (adapted in part from Ref.
[15]).

Fig. 3. A sample of IWTF profile.

memory effects. In addition, the IWTF identifies the wave leading to the total collapse of the structure in a single run. Numerous runs
of isolated CNWs will be required to pinpoint the wave leading to the total collapse but it yet lacks the memory effects.
The CNW's formulation used in the present study does not include the effects of second order random waves and short-crested
waves. Second order waves contribute to the global structural responses of structures with relatively low natural frequencies, such as
floating or jack-up platforms. For example [33] modelled a jack-up rig as a linear SDOF system and predicted its extreme response in
second order stochastic waves by the FORM. Similar to that used by Ref. [34], the second order unidirectional components were
introduced to provide corrections to linear waves. They showed that the increased wave crests, seen in second order waves, had
meaningful roles on the horizontal sway of the jack-up deck. Nevertheless, the second order waves and short-crested waves could be
easily incorporated in the EWA, if needed (see also [35]). It is noted that in this study, the first generation of IWTF was utilised which
compromises a series of linear CNW wave profiles as well as a linear growth function. Short-crested waves and the second-order
effects [33,35] can also be included in the IWTF. This may be considered in future studies on EWA, which will make the proposed
method more applicable and realistic for a wider range of marine structures.

2.3. Wave load model

ABAQUS/AQUA Finite Element program [36] was used for the nonlinear dynamic time-domain structural modelling of the
platform. It is able to calculate the buoyancy, drag and inertia forces acting on the submerged structural elements. The program
supports a number of regular wave theories (e.g. Airy's and Stoke's) but not the irregular waves. In the current study, a user-defined
subroutine was developed and integrated in to the ABAQUS/AQUA calculation process, as an add-in program. This makes it possible
to model an offshore structure against any irregular wave profile. As a specific case, it was used to build IWTFs based on Constrained
NewWaves and to analyse the structure against this type of step-wisely intensifying waves. The subroutine generates any desired
irregular and random wave profile, calculates the wave/current kinematics in each time increment, performs wave stretching cal-
culations and introduces the results to the integration points in all structural elements. The wave kinematics at the integration points
along with the Morison equation are used to compute the wave/current forces acting on the members.
In this study, to avoid a too high computational time, 10 IWTFs are used in the EWA to model the randomness and uncertainties in
the sea wave elevation. Each IWTF includes 34 random CNWs so, in total, the jacket is simulated against 340 random short duration
wave time-series. It is, however, acknowledged that a higher number of IWTFs can increase the accuracy of the results.

63
M. Zeinoddini et al. Marine Structures 59 (2018) 60–79

Table 1
Statistical characteristic of the random variables [10,37–39].

Variable Type Mean COV

Above splash zone Below splash zone Above splash zone Below splash zone

Drag coefficient, Cd Lognormal 0.65 1.10 0.25 0.25

Inertia coefficient, Cm Lognormal 1.60 1.27 0.1 0.1
Marine growth thickness (mm) Lognormal 50 0.3
Yield stress, Fy Lognormal 335 0.07

2.4. Uncertainty model

Uncertainties involved in the metocean structural assessment of offshore platforms include: i) inherent randomness of the ocean
waves, ii) uncertainties in parameters influencing the wave/current loads, iii) uncertainties in structural strength, iv) uncertainties in
mechanical characteristic of soil substrate, and v) uncertainties due to inadequate knowledge about the present conditions of the
structure for the existing platforms [10,37–39].
In the current study, sample uncertainties in the categories i to iii above were considered in a probabilistic-based structural
assessment of an existing offshore platform. For this, 10 different irregular IWTFs were considered for the structural assessment. Each
IWTF was an assemblages of 34 random CNWs (Fig. 3). The IWTFs' profiles, therefore, explicitly embodied the randomness of the
ocean waves. The marine growth thickness (MG), drag force coefficient (Cd) and inertia force coefficient (Cm) were treated as random
loading variables. The yield stress of the steel materials (Fy) was also considered as a sample random strength variable.
It is acknowledged that the variability of the wave characteristics on the structural response is far more important than sample
load and strength uncertainties simulated by the LHS. The variability in the wave characteristics was taken separately into con-
sideration using a stratified random sampling method. The wave population was first stratified in 34 subpopulations. A subpopulation
was characterised by two wave parameters (Hs and Tp) as well as a frequency of occurrence based on the wave hazard data. Each wave
subpopulation comprised 10 equiprobable random CNWs with the same Hs and Tp but different profiles and frequency contents. In
overall, each structural model was examined under 340 stratified irregular CNWs. The stratified random sampling with such a
population appeared to duly model the variability of the critical wave characteristics.
Table 1 provides the statistical characteristics of the random variables considered in the study. It is noteworthy that for the
hydrodynamic coefficients (Cd and Cm) two different sets of probability distributions were considered based on whether the member
was located above or below the upper elevation of the splash zone.
The Latin Hypercube Sampling (LHS) [40], which is a special type of Monte Carlo Simulation (MCS), was used to generate input
modelling data for the aforementioned load/strength variables. LHS is a stratified sampling method that has shown to outperform the
MCS. It is ideally suited to reducing the dispersion of the MCS on nonlinear structures. With the standard LHS approach the prob-
ability distribution function of each random variable is stratified into N equiprobable intervals. If K is the number of random
variables, the sampling space will then be K-dimensional. An N × K matrix P, in which each of the K columns is a random per-
mutation of 1,…, N, plus an N × K matrix R of independent random numbers from the uniform (0, 1) distribution are then estab-
lished. These matrices form the basic sampling plan which is represented by the matrix S as [41]:
1
S= (P − R)
N (1)

Each element of S (sij) is then mapped according to its target marginal distribution as:

xˆij = Fx−j 1 (sij ) (2)

where, Fx−j 1
represents the inverse of the target Cumulative Distribution Function (CDF) for variable j. The vector below now contains
input data for a single deterministic structural model:
x̂i = [xˆi1 xˆi2 … xˆi3] (3)

The probability of failure in the standard LHS can be defined as:

m
1
Pf =
m
∑ I [g (uj) ]
j=1 (4)

where, m is the number of simulations and uj is the sample number j from the standard random vector and I [g (uj )] is defined as
below [41]:

0, if g (uj ) > 0 (Safe)

I [g (uj )] = ⎧
⎨ 1, if g (uj ) < 0 (Failure) (5)
⎩
Considering four random variables of Cd, Cm, MG and Fy (say K = 4) plus five realisations for each random variable (say N = 5),
20 vectors (x̂i ) were generated by the Latin Hypercube Sampling. Each vector was representing stratified samples of the desired

64
M. Zeinoddini et al. Marine Structures 59 (2018) 60–79

Fig. 4. Schematic representation of the nonlinear equilibrium path and the IO, LS and CP performance limit states.

random variables. Accordingly, 20 equiprobable structural models, each containing sample input data for one deterministic analysis
were defined. It was also assumed that all random variables are uncorrelated. This does not seem a far-fetched assumption because
very low correlations among random variables such as the yield stress of steel materials, marine growth thickness, drag coefficient
and inertia coefficient were reported by other researchers [42].
The accuracy of the LHS method can be improved by increase in the number of layers of the random parameters but the com-
putation cost also considerably increases. A test was designed to reach a trade-off between the computation costs and the accuracy.
The maximum base shear under a CNW representing the design sea state was considered as a target random realization. LHS si-
mulation with different layer numbers of N=10.20 and 50 were adopted. The Cumulative Distribution Function (CDF) for N=20
stayed reasonably close to that for N=50 but the CDF of the base shear for N=10 stayed apart from the two. To avoid excessive
computational costs N=20 was used in the current study.

2.5. Performance limit states

A probabilistic-based structural assessment is expected to deliver the probability that different structural performance levels are
exceeded under loading conditions. Performance levels or limit states are, therefore, vital parts of the assessment practice. In
earthquake engineering, performance limit states such as “Immediate Occupancy (IO)”, “Life Safety (LS)” and “Collapse Prevention
(CP)” are well quantified and documented based on available experimental data and post-earthquake measurements. A schematic
presentation of the IO, LS and CP limit states for a typical static pushover analysis, adapted from Ref. [43], is given in Fig. 4.
For offshore structures, however, performance-based limit states under wave loading are not still developed. This is a potential
area of future research. In offshore engineering practice, ultimate capacity is used as the reference for deterministic structural
integrity assessment of platforms (e.g. Refs. [7,44,45]). In the current study, inspired by the codes' practice, an “Ultimate Level” (UL)
limit state was considered for the probabilistic-based structural assessment of the fixed offshore platform. This performance limit
state is marked as point C in Fig. 5. Fig. 5 schematically illustrates the global nonlinear equilibrium path of an offshore platform in

Fig. 5. A schematic nonlinear equilibrium path in a fixed steel offshore platform showing the CP and the UL structural performance targets.

65
M. Zeinoddini et al. Marine Structures 59 (2018) 60–79

Fig. 6. A global view (left) and a side view of the case study platform model including the soil-pile interaction provisions (right).

terms of the base shear against the lateral deck displacement. Initial yielding occurs when the base shear surpasses the design load
level plus a predefined safety margin. Afterward, material and geometrical nonlinearities arise in the structure, so the slope of the
equilibrium path decreases. This trend continues until the base shear arrives at a peak value. The UL limit, so, denotes the overall
peak in the equilibrium path. Beyond the UL limit, despite the increase in the deck displacement, the base shear would not reach the
maximum previously experienced.
In addition to the UL limit state, a “Collapse Prevention” limit state was also considered in the current study. Analogous to [10],
the CP limit was defined as a point before the ultimate capacity where the slope of the equilibrium path was significantly falling. As
shown in Fig. 5, the CP limit showed wherever the tangent to the equilibrium path reduced to 15% of the initial slope.

3. The study case and its modelling

3.1. General information

An EWA probabilistic-based structural assessment was implemented on a sample four-legged offshore platform located in the
Persian Gulf which has been installed in 1971 and the operator is now intending to extend its service life for another 25 years. The
mean water depth is around 67 m. The legs are battered (slope of 1:8) and laterally supported by knee type braces. The jacket
horizontal dimensions are 15.6 m × 15.6 m at working point and 33.2 m × 33.2 m at mud line. Four through-leg grouted piles
anchor the jacket to the seabed. The topside weight is about 2400t with two main decks. The main natural period of the platform is
2.8s. Fig. 6 shows the three and two dimensional views of the platform model.

3.2. Structural model

As the wave height progressively increases during the EWA, the platform model would essentially experience geometrical and
material nonlinearities, overstressing, local or overall member buckling and local failure prior to global collapse of the platform.
Nonlinear methods are, therefore, required to demonstrate that a platform has adequate residual strength and stability to withstand
extreme metocean loading.
The advanced finite element program ABAQUS [36] was utilised for the nonlinear time-domain dynamic, as well as the quasi-
static analyses of the platform. The Timoshenko three dimensional Beam31 elements were adopted for modelling the tubular jacket
members and piles. Up to 10 elements were used to model a single tubular member. Geometrical nonlinearity due to large

66
M. Zeinoddini et al. Marine Structures 59 (2018) 60–79

Fig. 7. Photos of severe corrosions and holes on spider deck members (EL. (+) 3.05) [46].

displacements/strains as well as the material nonlinearity were considered in the modelling. The von Mises yielding criterion with a
combined isotropic/kinematic hardening model was used to describe the plastic cyclic behaviour of the steel materials. In the
modelling of the platform, instead of nominal yield strength, the mean yield strength of the steel material was used [7]. This resulted
in around 17% increase in the nominal material yield strength.
Very small initial imperfections were introduced at mid length of the jacket members to enable capturing the overall buckling and
post buckling behaviour of the members in compression. Primary members on the topside framing were included in the structural
model to represent the correct stiffness of the topsides. The deck girders and other primary structural components were modelled as
beam elements. An elastic behaviour was assumed for the topside members. The topside gravitational forces and masses were
introduced as equivalent point forces/lumped masses to the main topside joints. The deck appurtenances were not explicitly mod-
elled, although their mass and relevant wind/wave loads were included.
The structural damages, anomalies degradations, corrosions etc. as reported in a recent underwater/above water survey were
rationally incorporated in the platform structural model [46]. Fig. 7 shows sample photos of damaged members at spider deck of the
platform. Two considerations were taken in order to model anomalies such as holes in Fig. 7. The member was assumed fully flooded,
say no buoyancy was assigned to perforated members. Similar to that in Ref. [47], a reduced cross section was assumed for a member
length around the hole. A corrosion allowance, covering the post survey service life period of the platform, was considered for the
jacket members located in the splash zone.
Each pile was modelled using a set of short length beam-column elements. The nonlinear pile-soil interactions were characterised
by t-z (for axial shaft skin friction), p-y (for lateral load bearing) and q-z (for the pile end bearing) springs as introduced in Ref. [7].
Sample locations of the soil springs are shown in Fig. 6. The p-y springs were modelled with cyclic strength degradation. The
contributions of other soil parameters, such as loading rate effects, cyclic loading hysteresis, reconsolidation (or time effects) and
aging for cohesive soils were not considered in these analyses.
An implicit, direct integration, dynamic approach based on Newark's constant average acceleration method was used to solve the
nonlinear equations of the motion against the wave actions. A Rayleigh structural damping [48,49], based on 2.5% for the first
natural mode and 5% for the second natural mode, was considered for the models. With the quasi-static (push-over) analyses a
modified RIKS algorithm (arc-length method) was used to allow tracing the unstable paths through the analysis.

3.3. Metocean data

The hindcast study for the jacket location [5] indicated that the long term statistics of the significant wave height (Hs) for the 3 h
sea states can be represented as the tail of an exponential distribution with its probability of exceedance given by:
Hs
P (Hs ) = Ae− λ (6)

where, P(Hs) is the mean annual frequency of exceedance of Hs, A is the annual number of the (3 h) sea states in this wave direction
and λ is the standard deviation. P(Hs) was then related to return period (RP) by assuming that exceedance events follow a Poisson
process:
1 1
P (Hs ) = 1 − e− RP ≈ : for Large RPs
RP (7)

Fig. 8 gives the mean annual exceedance frequency of the significant wave height (λ Hs (h) or the wave height hazard curve) for
waves approaching the platform in True North-West direction. The ordinate in the figure gives the mean annual frequency of ex-
ceedance of the significant wave height in True North-West direction. The aim of the current paper was to illustrate the EWA
potentials for the probabilistic-based structural integrity assessment of offshore platforms. For the sake of brevity, the calculations
were only presented for the sample True North-West wave direction. This was the most critical wave direction for the present
platform. Nevertheless, to obtain the overall probability of failure, it is necessary to repeat the calculations for the lesser important
waves in remaining directions and sum up their contributions.
The mean annual frequency of occurrence of the storms or the likelihood that the significant wave height will equal a specified

67
M. Zeinoddini et al. Marine Structures 59 (2018) 60–79

Fig. 8. The mean annual exceedance frequency of the significant wave height (wave height hazard curve) for waves approaching the platform in True North-West (the
most critical) direction.

value, h, was extracted directly from the probabilistic wave height hazard analysis of the site. As it can be noticed in a half-loga-
rithmic scale the mean annual exceedance curve can be approximated by a straight line.
The waves in the current study were assumed to be long-crested. Short-crested waves could be easily incorporated in the EWA but
their contributions seems to be small during very extreme storms. The directional spreading, however, was taken into consideration
using a wave kinematics factor as per [7].

3.4. Deck inundation modelling

Wave-in-deck loading (deck inundation) has recently gained interest as some offshore codes require evaluations of the platforms
against the 10,000 year wave as an accidental load. The potential risks from the wave-in-deck events have been proved on numerous
occasions for example, in the aftermath of hurricanes in the Gulf of Mexico and with subsiding fixed offshore platforms in the Ekofisk
oilfields in the North Sea.
A variety of models have been put forward to calculate the horizontal wave-in-deck loads but there exist no general consensus
within the offshore industry on their merits [41]. The analytical solutions which have been proposed can be grouped into two broad
categories:

• Silhouette approach: This is included in the API [7] and ISO [45] offshore codes and the models are based on definition of a
projected area in the deck structure. The silhouette approach includes drag and momentum formulations.
• Detailed Component approach: It requires examining separately the wave loading on each individual component on the topside
structure.

The significant wave height in an IWTF gradually increases and goes well beyond the 100 year design significant wave height.
Therefore, there would be a good possibility for the deck inundation. To take into account the wave-in-deck loading, hydrodynamic
coefficients (Cd and Cm) of the topside elements were tuned to produce a maximum wave-in-deck force close to that suggested by API
RP 2A [7] silhouette approach. The wave-in-deck loads were then automatically calculated by the ABAQUS/AQUA program when the
wave crest exceeded the cellar deck elevation.
The API RP 2A model is widely accepted in industry practice [50]. It was calibrated against Gulf of Mexico related’ experiments
and provides a rough but reasonable estimate for the maximum horizontal wave-in-deck load [51]. The model, however, does not
explicitly include the inertia, slamming or pressure gradient effects [52]. The wave phasing and the load vertical components are also
neglected [53]. It also provides no data on the load time history which is required for the dynamic and time domain analysis of the
structure. Some researchers believe that the silhouette or the global approach may underestimate the impact loads in extreme
conditions [54].
The procedure used in the current study appears to overcome some of the aforementioned shortcomings. The hydrodynamic
coefficients (Cd and Cm) of the main topside elements were tuned to produce a maximum horizontal wave-in-deck force which was
reasonably close to that suggested by API RP 2A silhouette approach. The model was then able to take into account the inertia effects,
the wave phasing effects, temporal variation of the wave load, movements of the structure and vertical component of the deck
inundation load during analysis of the structure. It is, however, acknowledged that the proposed approach needs further validation
against test data.

68
M. Zeinoddini et al. Marine Structures 59 (2018) 60–79

3.5. Simulation scheme

As aforementioned, 20 equiprobable structural models were considered. They were accounting for the uncertainties in the
loading/strength modelling parameters. Each structural model was independently analysed against 10 IWTFs. The IWTFs all followed
similar intensifying trends but had different random contents. The IWTFs, so, were treated as equiprobable random metocean events.
Each IWTF was itself built from a sequence of 34 gradually intensifying CNWs.
The significant wave height (Hs) increment in two following CNWs was equal to 1 m up to Hs = 7 m. Smaller increments were
chosen afterward to capture more accurately the wave height at which the target performance limit state was violated. The significant
wave height increments were, so, equal to 0.5 m up to Hs = 8 m, 0.25 m up to Hs = 14 m and 1 m later on up to Hs = 15 m. It is noted
that the 100 year design wave in the most critical incident direction (True North-West) was Hs = 6.6 m (or Hmax = 12.2 m) and the
associated spectral peak period was 11s. Assuming that the wave climate in the platform location can be expressed by
Pierson–Moskowitz spectra the following relationship was considered between the spectral peak period for a given significant wave
height [55]:

Hs
Tp = 2π
0.16g (8)

Eq. (8) presents a one-to-one relation between Hs and Tp where, Tp is the spectral peak period and g is the gravitational accel-
eration. It is noteworthy that a scatter diagram was available for the fatigue analysis of the present jacket structure but it seemed
unhelpful for constructing the IWFTs. This was because the number of Hs bins in the scatter diagram was 8 which appeared too small
(each IWFT had 34 intervals for Hs). More importantly, the scatter diagram was only covering the design wave ranges. The EWA,
should inevitably include very unlikely freak waves, well beyond the design wave, to model the structure collapse. 24 out of 34 CNWs
used in the current study had Hs larger than the design Hs; for them no scatter data was available.
The Pierson–Moskowitz spectrum was utilised for the current study based on the design documents of the platform. It is noted that
the IWTF profiles and subsequently the structural response would noticeably change if other wave spectra such as the JOWNSWAP
were used.
For each simulation, all gravitational loads were introduced to the model in an initial static step. In a second dynamic load step
the structure was analysed under a single IWTF, such as that given in Fig. 3, plus the corresponding wind and current loads. In
overall, N = 20 × 10 = 200 time-domain nonlinear simulations of the structure against irregular wave loads (each one comprising
34 consecutive CNWs as shown in Fig. 3) were performed. Fig. 9 gives a presentation of the simulations scheme.

4. Results and discussions

The results of those 6800 time-domain simulations of the platform against irregular waves were providing an unprecedented
amount of data for establishing a probabilistic assessment of the structural response. The data were used for the statistical analysis of
the platform long-term response to the storm events. They were also employed to generate hazard curves for a set of Engineering

Fig. 9. Simulations arrangement for the EWA probabilistic-based structural assessment of the jacket platform.

69
M. Zeinoddini et al. Marine Structures 59 (2018) 60–79

Fig. 10. A sample EWA response for a structural model subjected to one IWTF, as time series of the deck horizontal displacement.

Demand Parameters (EDPs). The fragility curves, as plots of the probability of failure of the structure as a function of a particular
demand variable, were also established. Finally, the probability of the platform failure at different CP or UL performance limit states
were calculated using the corresponding fragility and hazard curves.

4.1. Sample EWA response time series

Fig. 10 and Fig. 11 give typical EWA responses obtained from simulating one of the 20 structural models to one of the IWTFs. The
figures depict the time series of the deck horizontal displacement and the overall base shear, respectively. It can be noticed that the
maximum deck displacement and the maximum base shear gradually increased as the height of the successive CNWs step-wisely
increased (see also Fig. 3). With this case at a CNW with Hs = 11.75 m (for the sea-state corresponding to 2,300s ≤ t ≤ 2,400s in the
IWTF), the deck demonstrated unbounded displacements. Unbounded responses in a dynamic time domain analysis essentially in-
dicates on occurrence of a global dynamic instability in the structural system [56]. Fig. 10 very interestingly shows that the memory
effects, or the bearings of the previous high waves on the structure response, were modelled by the IWTFs. Although the wave crest in
following CNWs incorporated in an IWTF was increasing incrementally, the deck maximum horizontal displacement did not ne-
cessarily followed the suit. This was because the structural response to a certain CNW could be exacerbated or mitigated by the
actions from earlier high waves. For example earlier waves might have left behind permanent structural/foundation deformations
which may make the structure a softer or a stiffer target for the next coming wave. These type of memory effects would be overlooked
if, instead of IWFTs, the structure was examined against isolated CNWs.
It is noted that the failure modes in different EWA runs were not essentially similar. Failure in the foundation system was the
governing mode but buckling in the bracing system was also noticed in some cases to trigger a global dynamic instability.
A sample equilibrium path (plot of the maximum base shear against the maximum deck horizontal displacement) from another
EWA is shown in Fig. 12. In this figure, the maximum deck displacement observed within each sea-state in the EWA (34 in total) was
plotted against the maximum base shear experienced in the same sea-state. The figure does not cover the data for the sea-states
beyond the structural failure.

4.2. Hazard curves

4.2.1. Definitions
One product of the probabilistic EWA is the hazard curve for a given Engineering Demand Parameters (EDP) or λEDP (x ) . A hazard
curve for a given EDP provides the mean annual frequency (MAF) of the EDP exceeding a specified value of x. Possible EDPs for a
jacket type offshore platform can be the maximum deck drift, maximum inter-story drift, maximum or permanent pile displacement,
maximum conductor drift ratio, etc. A EDP hazard curve can be obtained by convolving the complementary conditional cumulative

Fig. 11. A sample EWA response for a structural model subjected to one IWTF, as time series of the base shear.

70
M. Zeinoddini et al. Marine Structures 59 (2018) 60–79

Fig. 12. Equilibrium path (plot of the base shear against the deck horizontal displacement) from a sample EWA.

distribution function of the EDP given wave height (h) with the wave hazard curve. The wave hazard curve, λ Hs (h) represents the
mean annual frequency (MAF) of Hs exceeding a specified value of h. It is normally made available by a site specific wave climatology
study. As an example, Fig. 8 was giving the wave hazard curve, λEDP (x ) in True North-West direction for the platform location.

4.2.2. Convolution integrals

Calculation of the EDP hazard curve will be significantly simplified by decoupling the wave hazard curve and the conditional
cumulative distribution function of the EDP:

λEDP (x ) = ∫
h
( dλ Hs (h)
dh )
× G EDP Hs (x h) dh

λ Hs (h) = ν . αi. P [Hs > h]

GEDP Hs (x h). =P [EDP> x Hs = h] (9)
In Eq. (9):

• ν is the total annual number of storms. It can be taken as the number of 3-h sea-states over a one year period in all incident
directions, when no storm threshold is considered, so, ν = 365.25 × 8 = 2922 [32],
• α gives the fraction of waves propagating (over a long-term period) in ith direction,
i

• H is the significant wave height,

s

• G (x h) is the conditional probability of exceeding a specified EDP value of x for H = h; and as will be explained, it can be
EDP Hs s
calculated from the EWA results,
• P [H > h] is the cumulative probability of exceedance for H in the ith incident direction.
s s

If the values of λ Hs (h) and G EDP Hs (x h) are available as discrete data points, the hazard curve for a given EDP can be obtained
from:

p [Hs = h1] × G EDP Hs (x h1)+

⎧ ⎫
p [Hs = h2] × G EDP Hs (x h2)+
λEDP (x ) = ν . αi
⎨… +⎬
⎩ p [Hs = hn] × GEDP Hs (x hn) ⎭ (10)
where, p [Hs = h] is the probability density of Hs which can be obtained from differentiating the wave hazard curve.

4.2.3. Multiple stripe analysis

It is noted that λ Hs (h) for the platform location in 8 directions were available from the wave climatology study. To obtain the EWA
hazard curve from Eq. (9) (or Eq. (10)) only G EDP Hs (x h) , or the conditional probability of exceeding the specified EDP value of x, for
Hs = h, was needed. This was achieved by re-compiling the EWA results based on a Multiple Stripe Analysis (MSA) methodology [57].
The MSA was already employed for the probability-based seismic assessment of onshore structures. In summary, MSA is a col-
lection of single-stripe analyses. A single stripe refers to a number of independent nonlinear dynamic seismic analyses of the structure
against a selected set of ground acceleration records, all scaled to an identical Intensity Measure (IM) value. The procedure is then
repeated at multiple levels of the IM to provide statistical information about the EDP over a wide range of IM values.
For the seismic application of MSA, however, a huge number of separate dynamic analysis of the structure are required. On the

71
M. Zeinoddini et al. Marine Structures 59 (2018) 60–79

contrary, the number of simulations required with EWA are considerably small. This is because, the EWA methodology is intrinsically
a multi stripe type of simulation. For example, an IWTF in the current study was an assemblage of 34 gradually intensifying sea states
(see Fig. 3). Each sea state was characterised by a certain Hs and modelled by an irregular NewWave which took 100s. From the MSA
standpoint, each sea state can be treated as a single stripe. By analysing the structure to a single IWTF, the time domain dynamic
response of the structure for 34 multiple wave stripes were made available. As aforementioned, the total number of IWTFs was 10 and
the total number of the structural models was 20. As a result, up to 200 independent nonlinear dynamic analyses of the structure were
easily available at each wave stripe (sea-state).
The steps taken to obtain G EDP Hs (x h) and consequently the hazard curve for the desired EDP can be generally articulated as
follows:

• The desired EDP (e.g. maximum lateral deck displacement) is selected,

• The time series of the EDP in kth EWA (k = 1 to 200) at ith stripe (ith CNW representing a sea state with H = h ) is extracted, s i

• The maximum value of the EDP in the time series is identified,

• Above steps are repeated over the total number of EWAs (k = 1 to 200) at ith stripe (ith NewWave) to generate 200 data points for
the G (x h) at the wave stripe of H = h ,
EDP Hs s i

• Above steps are repeated for all wave stripes (i = 1 to 34). This generates multiple sets of 200 data points for G (x h) at EDP Hs
multiple levels of the wave stripes which provides statistical information about the EDP over a wide range H values, s

• At a single stripe G (x h) is no longer conditional to the significant wave height and can therefore be presented by a customary
EDP Hs
probabilistic function. The statistical properties of G (x h) , however, varies at different stripes (significant wave heights). It
EDP Hs
will be now more convenient to fit a proper parametric probability distribution function to the data points at every single wave
stripe of hi (200 data points at each hi),
• Using the probability distribution function for G EDP Hs (x h) along with the wave hazard curve λ Hs (h) , the EDP hazard curve λEDP (x )
can now be obtained from Eq. (9) or Eq. (10).

4.2.4. Sample MSA outputs

As an example, Fig. 13 presents the data points for G EDP Hs (x h) obtained from implementation of the MSA on the EWA outputs.
The EDP selected is the base shear, so the figure gives the data for the maximum base shear experienced by the platform in 34 Hs
stripes. For clarity, only results from 4 (out of 10) IWTFs are presented thus, this figure presents 80 (instead of 200) data points in
each wave stripe. The lines connecting the 16th, 50th and 84th percentiles of the data points in individual stripes are also shown in
Fig. 13 (left). To generate a continuous presentation of G EDP Hs (x h) , a set of standard probability distribution functions were ex-
amined for each wave stripe. For example, the best fit for wave stripe (Hs=7 m) was a lognormal function with μ = 1.7 MN and
σ = 0.43 MN. The practice was repeated for other of Hs = hi (i = 1 to 34) but, the results are not presented in Fig. 13 to avoid
overloading the image.
Fig. 13 (right) gives the same MSA results as in Fig. 13 (left), but with highlighting the failure cases. Within a single wave stripe in
Fig. 13 (right), the black diamond markers show the maximum base shear for non-failure cases. The red triangles mark the maximum
base shears in the failure cases. In this figure, a CP performance limit state was used to describe the failure (collapse) condition. As it
can be noticed, a single wave stripe may contain more than one collapse case, essentially due to uncertainties involved in the loads,
structural systems and ocean wave profiles.
Fig. 13 shows a gradual decrease in the total number of data points in wave stripes higher than Hs = 7 m. This is because a number
of EWAs stopped running as they had violated the target performance limit in a previous wave stripe. For example, there remained

Fig. 13. Multiple Stripe Analysis (MSA) on the base shear response out of 80 EWA simulations with 16th, 50th and 84th percentiles of the base shear response (left),
Marking the collapse and non-collapse cases in different wave stripes (right).

72
M. Zeinoddini et al. Marine Structures 59 (2018) 60–79

Fig. 14. Base shear (left) and RSR (right) hazard curves for waves approaching the platform from True North-West direction.

only one data point in the wave stripe of Hs = 12.75 m. This reduction in the number of data points was causing a difficulty in using
Eq. (9) or Eq. (10) and for defining a parametric probability distribution function for G EDP Hs (x h) . To overcome this difficulty, the
missing data points in wave stripes of high intensity were estimated by the nonlinear extrapolating the non-failure data from the
previous wave stripes in the same EWA run. By substituting the missing data with the data obtained from the extrapolation, all wave
stripes owned the full (200) data points and Eqs. (7) and (8) could then be used to extract the desired hazard curve.

4.2.5. Sample hazard curves

Fig. 14 (left), as an example, depicts the base shear hazard curve obtained from the probabilistic EWAs of the structure. The figure
also gives a parametric presentation of the hazard curve. This is derived by fitting an exponential distribution to the data. Fig. 14
(right) gives the RSR hazard curve obtained from the probabilistic EWAs plus an exponential distribution fitted to the data. It is noted
that these figures give the results for waves approaching the platform in True North-West direction.

4.3. Fragility curves

Collapse fragility functions have become increasingly popular in structural integrity assessments. A fragility function specifies the
probability of collapse of a structure, or violating a limit state of interest, as a function of a specific value of Intensity Measure (IM). A
counted fragility curve can be obtained by treating the collapse capacity data as random samples, i.e., equally likely outcomes. A
probability of 1/N (N = total number of realisations) can be assigned to a single collapse case. The probability of collapse increases
by 1/N after each individual collapse occurs. This approach, for producing the collapse fragility curve, is called the “Relative
Frequency” method.
In the current study, two performance limit states of CP and UL were considered. Different Intensity Measures (IMs) or EDPs such
as the deck lateral displacement, significant wave height, base shear, RSR, etc. were considered in obtaining the fragility function. As
aforementioned, the total number of EWAs was 200. This means that for an IM (or EDP) of interest, two sets of 200 collapse data
points were delivered by 200 EWAs. Each data set (200 realisations), could be used to generate a Cumulative Distribution Function
(CDF) for the desired IM (or EDP) and the performance limit state. The outcome CDF is also called the fragility function (curve).
Fig. 15 presents a sample collapse fragility curve for the platform studied. The figure depicts the CDF of the violating the CP and
UL performance limit states against the base shear values. The irregular curves represent the counted fragility curves, obtained from
the ordered data points. The smooth curves are the parametric fragility curve derived by fitting lognormal distributions to the data. It
can be noticed that uncertainty bands in the base shear fragility functions for the CP and UL limit states remained almost the same.
The fragility curve for the CP limit state showed a shift to the left hand of the UL curve, indicating on a much more restrict
performance criteria.
Table 2 provides a summary of statistical data for the parametric fragility curves in the CP and UL limit states delivered by the
probabilistic EWA method. In order to identify the appropriateness of the parametric distribution functions, the Kolmogorov-Smirnov
tests (K–S tests) were employed. The lognormal distribution function was found to provide the best fit to the empirical results. It is
acknowledged that the appropriate distribution function might change by structure type, wave model, water depth and ocean lo-
cation.
The empirical fragility curves presented in Fig. 15 incorporate the variability in bellow modelling parameters:

- In the wave characteristics (Hs and Tp),

- In sample wave load parameters (the marine growth thickness, drag force coefficient and inertia force coefficient),
- In sample strength parameter (Fy),
- In the wave-in-deck loading (through variability in the drag force and inertia force coefficients).

73
M. Zeinoddini et al. Marine Structures 59 (2018) 60–79

Fig. 15. Empirical and analytical base shear fragility curves obtained from EWA approach in the CP and UL limit states.

Table 2
Summary statistical data for the parametric fragility curves in the CP and UL limit states obtained from the probabilistic EWA approach.

Mean COV Distribution Statistics Parameters

CP Limit State
Base Shear (MN) 2.53 0.081 Lognormal μ = 0.925, σ = 0.083
Normalised Base Shear (MN) 2.354 0.237 Lognormal μ = 0.828, σ = 0.233
Hs at Collapse (m) 9.73 0.114 Lognormal μ = 2.269, σ = 0.113
Deck Displacement (m) 0.265 0.162 Lognormal μ = −1.340, σ = 0.15
UL Limit State
Base Shear (MN) 2.95 0.087 Lognormal μ = 1.081, σ = 0.072
Normalised Base Shear (MN) 2.737 0.218 Lognormal μ = 0.983, σ = 0.222
Hs at Collapse (m) 10.87 0.115 Lognormal μ = 2.379, σ = 0.115
Deck Displacement (m) 0.829 0.337 Lognormal μ = −0.278, σ = 0.44

Probabilistic push-over analyses were also conducted using 20 equiprobable platform models considered in the current study.
Table 3 gives the statistical characteristics of the parametric fragility curves in the UL limit states obtained from the probabilistic
push-over analyses. Once again, the best fits for the parametric fragility curves are found to be lognormal distributions.
From Tables 2 and 3, it can be noticed that the mean values for the base shear fragility functions from the push-over analysis and
EWA were relatively close. The Coefficient of Variance (COV) of these two type of the analysis were, however, remarkably different.
In comparison to conventional push-over analysis, dispersion of the results from EWA was considerably high. Fig. 16 provides
comparisons between the fragility curves resulted from different approaches used in the current study and for different limit states.
The fragility curves for the base shear, horizontal deck displacement and RSR are given in this figure.
As it can be seen the medians for the base shear and deck displacement fragility curves from the push-over analysis were relatively
close to those from the UL limit state from EWA. The uncertainty ranges, however, were lower in the fragility curves from the push-
over analysis as compared to those from EWA. The deck displacement fragility curves from EWA fell far apart in the CP and UL limit
states. The uncertainty range in the CP limit state was considerably small as compared to that for the UL state. This is essentially
because close to the dynamic instability (the UL limit state) the structure undergoes excessive and unbounded displacements.
Interestingly, the RSR fragility curves from the push-over and EWA are very close in their medians and in their uncertainty ranges.
In general, the fragility curves from the push-over approach demonstrated considerably lower uncertainty ranges as compared to
those from EWA. It appears that a probabilistic push-over approach falls short to properly portrait the full scope of the uncertainties
involved in the modelling of an offshore structure. This is most probably because the quasi-static push-over method ignores the

Table 3
Summary statistical data for the parametric fragility curves obtained from the probabilistic push-over analyses.

Mean COV Distribution Statistics Parameters

Base shear (MN) 3.02 0.021 Lognormal μ = 1.061, σ = 0.021

Normalised Base shear 2.809 0.217 Lognormal μ = 1.010, σ = 0.220
Deck Dis.(m) 3.206 0.10 Lognormal μ = 1.161, σ = 0.104

74
M. Zeinoddini et al. Marine Structures 59 (2018) 60–79

Fig. 16. Comparisons between the fragility curves resulted from different approaches used in the current study and for different limit states. a) Base shear fragility
curves; b) Deck horizontal displacement fragility curves; c) RSR fragility curves.

frequency contents of the ocean waves, the randomness of the sea elevation, the dynamic effects and the wave-in-deck.
By comparing the uncertainty ranges in the fragility curves from the push-over and EWA methods it can also be concluded that the
randomness of the ocean waves had higher impacts on the performance of the offshore structure as compared to other uncertainties
considered in the loading and the strength of the jacket platform. The RSR fragility curve, however, can be regarded as an exception
because almost identical fragility curves were delivered by the push-over and EWA methods.

4.4. Probability of failure

The mean annual probability of structural failure is straightforwardly derived from the EWA results. The probability of failure is a
consistent measure for the integrity assessment of a structure. It can be applied to all offshore structures in national and international
regions. For this, the estimates for the mean annual probability of failure are compared against a target probability of failure that
applies globally. In the current study, the convolution integral and the direct counting methods were separately implemented to
derive the mean annual probability of structural failure.

4.4.1. Convolution integral method

It is assumed that the hazard function λEDP (x ) and the fragility function FEDP (x ) are stochastically independent and already
available (see Sections 4.2 and 4.3). Mathematically, the Cumulative Distribution Function (CDF) of the probability of failure, Pf(x),
can be calculated from each of the below convolution integrals:

x ⎡ ⎛ dλEDP (x ) ⎟⎞ ⎤ dx = x
⎡ ⎛ dFEDP (x ) ⎞. λEDP x ⎤ dx = x
Pf (x ) = ∫0 ⎢FEDP (x ). ⎜ dx
⎝ ⎠⎦
⎥ ∫0 ⎢
⎣⎝ dx ⎠
() ⎥
⎦
∫0 fEDP (x ). λEDP (x )dx
(11)
⎣
Accordingly, the mean annual probability of failure (Pf), can be calculated as follows:

75
M. Zeinoddini et al. Marine Structures 59 (2018) 60–79

∞
Pf = ∫0 fEDP (x ). λEDP (x )dx (12)

It is clear that different probabilities of failure will be obtained for a single structure under different performance limit states. With
a certain performance limit state, the mean annual probability of failure, Pf, obtained from a pair of hazard and the fragility functions
also depends on the EDP (or IM) employed.
It is noted that Eq. (11) and Eq. (12) can only be used when the hazard function λEDP (x ) and the fragility function FEDP (x ) have
continuous expressions. Accordingly, the probabilities of failure obtained from the above equations are called as parametric. When
the hazard and the fragility functions are available as discrete data points, the convolution integrals can be substituted by sum-
mations. The probabilities of failure obtained from the latter approach are called as empirical.

4.4.2. Direct counting method

As aforementioned, the probabilistic EWA performed in the current study included N = 200 × 34 = 6800 time-domain nonlinear
simulations of the structure against irregular waves of different intensity. The EWA approach distinctly discriminated the collapse
and non-collapse cases over all 6800 cases analysed. By knowing the annual frequency in each of the 34 sea states considered in EWA
from the site wave climatology study, the mean annual probability of failure can be directly calculated. Provided that the wave
hazard curve λ Hs (h) and the MSA results are available as sets of discrete data points, the probability of failure can be directly
calculated as follows:
n n
(NC H s=hi )
Pf = ∑ (Pc H s=hi). ν. αi. p (H s=hi) = ∑ Nsim
Δλ Hs (hi )
i=1 i=1 (13)

where, λ Hs (h) presents the wave hazard curve, Pc H s=hi is the probability of the collapse (violating the limit state) in ith wave stripe,
NC H s=hi is the number of collapse cases counted in ith wave stripe, Nsim is the total number of EWA simulations (200 in the current
study) and n is the total number of wave stripes (34 in the current study). The remaining parameters were described earlier. This is,
accordingly, called a counting method.
In fact, the convolution integral method determines an annual probability of failure that is notional. The, counting method is
rather actuarial, therefore, provides a more accurate estimate for the mean annual probability of failure. With the counting method
the Pf at a certain performance limit state will be independent to the EDP (or IM) considered. It, however, varies if the target
performance limit state changes.

4.4.3. Target probability of failure

An API JIP on regional acceptance criteria [58] suggests values listed in Table 4 for the maximum acceptable annual probability of
failure (Pf). The values in the table correspond to a UL performance limit state and appear to be analogous to those in ISO 19,902
[45]. In fact, ISO considers a target of Pf = 3 × 10−5 for a new fixed offshore platform against environmental loads. The ISO target,
however, covers additional risks, which are difficult to mitigate. Therefore, ISO gives the risk for the collapse of a manned and non-
evacuated platform in extreme storms a fraction of 1 × 10−4. For an existing offshore platform it is not reasonably practical to
increase the capacity of the platform and the target Pf = 1 × 10−4 seems justifiable.

4.4.4. Analytical, empirical and direct counting results

Examples of the density and the cumulative distribution functions (PDF and CDF) for the probability of structural failure obtained
from EWA results are presented in Figs. 17 and 18. The figures depict the functions delivered by both the analytical and the empirical
methods. Fig. 17 gives the probability of failure in the UL limit state as a function of the base shear. Fig. 18 gives the same in the CP
limit state as a function of the significant wave height. As it can be noticed, the analytical and empirical results were not the same,
essentially due to the approximations involved in the curve fittings used for defining the hazard and fragility functions in the
analytical approach.
Table 5 provides a summary on the platform annual probability of failure in the CP and UL limit states from EWA and push-over
methods. It gives the results obtained from direct counting of the failure cases, which are considered to be more accurate, as well as
those obtained from the analytical and the empirical coevolution integrals. The latter, as it can be seen in the table, varied depending
on the EDP used for the hazard and the fragility curves. The table also provides the target annual probability of failure for the
platform. It is noted that the wave climatology study indicated on True North-West direction as the dominant and the most critical
wave incident direction. The hazard and fragility curves in Sections 4.2 and 4.3 were presented for the critical direction. They were
used in the current section for the probability of failure analyses. The mean annual probabilities of failure in this section were,
therefore, slightly underestimated because they did not include lesser contributions from the smaller waves in remaining directions.

Table 4
Maximum acceptable annual probability of failure in a UL performance limit state (Pf) [58].

Manned and non-evacuated (high life safety consequence) Pf ≤ 1 × 10−4

Unmanned with medium or low business or environmental consequence Pf ≤ 1 × 10−3
Unmanned but with high business or environmental consequence Pf ≤ 5 × 10−3

76
M. Zeinoddini et al. Marine Structures 59 (2018) 60–79

Fig. 17. The density and the cumulative distribution functions (PDF and CDF) for the annual probability of structural failure in the UL limit state in terms of the base
shear obtained by implementing the convolution integrals to the EWA results.

Fig. 18. The density and the cumulative distribution functions (PDF and CDF) for the annual probability of structural failure in the CP limit state in terms of the
significant wave height obtained by implementing the convolution integrals to the EWA results.

Table 5
Summary on the platform annual probability of failure in the CP and UL limit states from the EWA and push-over methods.

Parameter Mean Annual Probability of Failure Acceptable Annual Probability of Failure

Convolution Integral Counting

Name Unit Analytical Empirical

EWA(UL) BS MN 6.28E-05 6.86E-05 8.52E-05 10.0E-05

Deck Displacement m 4.39E-05 1.93E-05 8.52E-05
RSR – 7.34E-05 1.93E-05 8.52E-05
Hs m 6.56E-05 8.51E-05 8.52E-05
EWA(CP) BS MN 6.01E-04 5.75E-04 4.21E-04 –
Deck Displacement m 6.91E-04 2.96E-04 4.21E-04
RSR – 4.91E-04 5.95E-04 4.21E-04
Hs m 2.76E-04 1.38E-04 4.21E-04
Push-over(UL) BS MN 5.21E-05 6.86E-05 8.52E-05 10.0E-05
Deck Displacement m 2.57E-05 1.93E-05 8.52E-05
RSR – 4.57E-05 1.93E-05 8.52E-05

5. Closing remarks

The current paper discusses a framework for probabilistic structural integrity assessment of offshore platforms based on the
Endurance Wave Analysis (EWA). The stochastic nature and the uncertainties in the ocean waves were taken into account by in-
troducing a set of random gradually intensifying wave trains. To simulate other uncertainties in the loading and the strength of the
structure a Latin Hypercube Sampling Monte Carlo Simulation technique was employed. The EWA probabilistic assessment was
implemented on an existing aging platform. It included N = 20 × 10 × 34 = 6800 dynamic time-domain nonlinear simulations of

77
M. Zeinoddini et al. Marine Structures 59 (2018) 60–79

the structure against irregular waves. This number of simulations provided unprecedented amount of data for the probabilistic
assessment of the offshore structure. The results were used to obtain the hazard functions and fragility curves for a number of
Engineering Demand Parameters (EDPs) or Intensity Measures (IMs). The mean annual probability of structural failure (Pf) at dif-
ferent structural performance limit states was also estimated.
The results proved that EWA can be effectively implemented for the probabilistic-based design and assessment of offshore
platforms and for generation of the hazard and fragility curves, the mean annual frequency of exceeding functions various perfor-
mance criteria for a desired EDP (or IM) and calculating the annual probability of structural failure (Pf).
This was owing to the EWA abilities to consider the progressively deteriorating nature of the sea conditions during an extreme
storm event, to incorporate waves of different heights and frequencies in a single dynamic analysis, to take into account the irre-
gularity and randomness of the ocean waves, to incorporate variability/uncertainties in the modelling and to require relatively short
simulation times. Another EWA advantage in the probabilistic-based assessment of offshore platforms was that the EWA methodology
is intrinsically a multi stripe type of simulation. As a result, and in contrary to the seismic application, separate MSAs were no longer
required.
The most accurate practice for the probabilistic-based assessment of an offshore platform is to conduct full MCSs based on large
numbers of time consuming 3-h dynamic time-domain analysis. It was already proved in previous studies that the EWA predictions
for the maximum metocean response of fixed offshore structures are reasonably well comparable to those from the 3-h dynamic time-
domain analysis. The EWA estimates for the probabilistic-based response of the offshore structures, therefore, were going to yield to
the same level of accuracy, as that from the 3-h dynamic time-domain analysis, by using comparatively short duration time series of
the irregular water surface elevations. EWA, therefore, can be considered as an economical and proper substitute to the costly 3-h
dynamic time-domain analysis. The LHS-MCS approach considered in the current study provided further and incredible reduction to
the simulation cost. EWA was also providing a good insight into the nonlinear behaviour of the offshore structure. It was also possible
to consider different damage indices or limit states for evaluating the structural performance.
The probabilistic EWA assessment results were also compared to those from a probabilistic push-over approach. It was found that
the fragility curves from the push-over approach were very narrow-banded as compared to those from EWA. It appeared that the
probabilistic push-over approach fell short to properly portrait the full scope of the uncertainties involved in the modelling of an
offshore structure. This was partly because, in contrast to EWA, the probabilistic push-over approach ignored the frequency contents
of the ocean waves, the randomness of the sea elevation, the dynamic effects and the wave in-deck loads nonlinear dynamic type of
analysis.

Acknowledgements

The authors wish to thank TWI Persia for providing the data required for modelling of the platform.

References

[1] Zeinoddini M. Design and performance of fixed offshore platforms. Iranian National Oceanology Institute; 2006.
[2] Najafian G. Comparison of three different methods of moments for derivation of probability distribution parameters. Appl Ocean Res 2010;32:298–307.
[3] Golpour H, Zeinoddini M, Khalili H, Golbaz A, Yaghubi Y, Adl MR, Ranjbar P, Saberi R, Matin Nikoo H. Structural integrity assessment of aging fixed steel
offshore jacket platforms: a Persian Gulf case study. ASME 2013 32nd international conference on ocean. Offshore and Arctic Engineering: American Society of
Mechanical Engineers; 2013. V02ATA055–V02AT02A.
[4] Hirdaris SE, Bai W, Dessi D, Ergin A, Gu X, Hermundstad OA, Huijsmans R, Iijima K, Nielsen UD, Parunov J, Fonseca N, Papanikolaou A, Argyriadis K, Incecik A.
Loads for use in the design of ships and offshore structures. Ocean Eng 2014;78:131–74.
[5] Zeinoddini M, Ranjbar P, Khalili H, Ranaei A, Golpour H, Fakheri J. Remaining fatigue life assessment of aging fixed steel offshore jacket platforms. Structure
Infrastructure Eng 2016;12:223–38.
[6] Salvadori G, Durante F, Tomasicchio G, D'Alessandro F. Practical guidelines for the multivariate assessment of the structural risk in coastal and off-shore
engineering. Coast Eng 2015;95:77–83.
[7] API. Recommended practice for planning, designing and constructing fixed offshore platforms, working stress design (WSD). Washington, DC: American
Petroleum Institute; 2010.
[8] Manuel L, Schmucker D, Cornell C, Carballo J. A reliability-based design format for jacket platforms under wave loads. Mar Struct 1998;11:413–28.
[9] Agarwal P, Manuel L. Simulation of offshore wind turbine response for long-term extreme load prediction. Eng Struct 2009;31:2236–46.
[10] Golafshani AA, Ebrahimian H, Bagheri V, Holmas T. Assessment of offshore platforms under extreme waves by probabilistic incremental wave analysis. J Constr
Steel Res 2011;67:759–69.
[11] Mardfekri M, Gardoni P. Probabilistic demand models and fragility estimates for offshore wind turbine support structures. Eng Struct 2013;52:478–87.
[12] Matin Nikoo H, Zeinoddini M, Estekanchi H. Probabilistic estimation of offshore jacket type platforms capacity using incremental random wave analysis (irwa).
Int J Maritime Technol 2013;8:1–18.
[13] Mirzadeh J, Kimiaei M, Cassidy MJ. A framework to efficiently calculate the probability of failure of dynamically sensitive structures in a random sea. Ocean Eng
2015;110:215–26.
[14] Golafshani A, Bagheri V, Ebrahimian H, Holmas T. Incremental wave analysis and its application to performance-based assessment of jacket platforms. J Constr
Steel Res 2011;67:1649–57.
[15] Zeinoddini M, Matin Nikoo H, Estekanchi H. Endurance Wave Analysis (EWA) and its application for assessment of offshore structures under extreme waves.
Appl Ocean Res 2012;37:98–110.
[16] Matin Nikoo H, Zeinoddini M, Estekanchi H, Golestani M. Integrating constrained random waves in endurance time analysis of offshore structures subjected to
sea waves. ASME 2012 31st international conference on ocean. Offshore and Arctic Engineering: American Society of Mechanical Engineers; 2012. p. 677–84.
[17] Diznab MD, Mohajernassab S, Seif M, Tabeshpour M, Mehdigholi H. Assessment of offshore structures under extreme wave conditions by Modified Endurance
Wave Analysis. Mar Struct 2014;39:50–69.
[18] Jahanmard V, Diznab MAD, Mehdigholi H, Tabeshpour MR, Seif MS. Performance-based assessment of steel jacket platforms by wave endurance time method.
Ships Offshore Struct 2017;12:32–42.
[19] Mohajernassab S, Diznab MAD, Mehdigholi H, Seif MS, Tabeshpour MR. Modification of endurance wave analysis based on New-wave theory. Ships Offshore

78
M. Zeinoddini et al. Marine Structures 59 (2018) 60–79

Struct 2017;12:330–40.
[20] Augusti G, Ciampoli M. Performance-Based Design in risk assessment and reduction. Probabilist Eng Mech 2008;23:496–508.
[21] Estekanchi H, Valamanesh V, Vafai A. Application of endurance time method in linear seismic analysis. Eng Struct 2007;29:2551–62.
[22] Hariri-Ardebili M, Sattar S, Estekanchi H. Performance-based seismic assessment of steel frames using endurance time analysis. Eng Struct 2014;69:216–34.
[23] Foyouzat M, Estekanchi H. Application of rigid-perfectly plastic spectra in improved seismic response assessment by Endurance Time method. Eng Struct
2016;111:24–35.
[24] Guo A, Shen Y, Bai J, Li H. Application of the endurance time method to the seismic analysis and evaluation of highway bridges considering pounding effects.
Eng Struct 2017;131:220–30.
[25] Stewart RA, Kittelson J, Kay IP. Statistical methods to improve the precision of the treadmill exercise test. J Am Coll Cardiol 2000;36:1274–9.
[26] Vanhees L, Lefevre J, Philippaerts R, Martens M, Huygens W, Troosters T, et al. How to assess physical activity? How to assess physical fitness? Eur J Cardiovasc
Prev Rehabil 2005;12:102–14.
[27] Fletcher GF, Balady GJ, Amsterdam EA, Chaitman B, Eckel R, Fleg J, et al. Exercise standards for testing and training. Circulation 2001;104:1694–740.
[28] Taylor P, Jonathan P, Harland L. Time domain simulation of jack-up dynamics witli the extremes of a Gaussian process. 1997.
[29] Cassidy M, Taylor RE, Houlsby G. Analysis of jack-up units using a constrained NewWave methodology. Appl Ocean Res 2001;23:221–34.
[30] Mirzadeh J, Kimiaei M, Cassidy MJ. Performance of an example jack-up platform under directional random ocean waves. Appl Ocean Res 2016;54:87–100.
[31] Mirzadeh J, Kimiaei M, Cassidy MJ. Effects of irregular nonlinear ocean waves on the dynamic performance of an example jack-up structure during an extreme
event. Mar Struct 2016;49:148–62.
[32] Veritas N. Environmental conditions and environmental loads. Det Norske Veritas; 2017.
[33] Jensen JJ, Capul J. Extreme response predictions for jack-up units in second order stochastic waves by FORM. Probabilist Eng Mech 2006;21:330–7.
[34] Tromans PS, Vanderschuren L. A spectral response surface method for calculating crest elevation statistics. J Offshore Mech Arctic Eng 2004;126:51–3.
[35] Jensen JJ. Conditional second-order short-crested water waves applied to extreme wave episodes. J Fluid Mech 2005;545:29–40.
[36] SIMULIA. ABAQUS theory manuals 6.9 ed: Dassault Systèmes, Realistic Simulation.
[37] Dalane JI. System reliability in design and maintenance of fixed offshore structures. Norway: Department of Marine Structures, Norwegian Institute of
Technology, University of Trondheim; 1993.
[38] Ellingwood B. Development of a probability based load criterion for American National Standard A58: building code requirements for minimum design loads in
buildings and other structures. US Department of Commerce, National Bureau of Standards; 1980.
[39] Baecher GB, Christian JT. Reliability and statistics in geotechnical engineering. John Wiley & Sons; 2005.
[40] Helton JC, Davis FJ. Latin hypercube sampling and the propagation of uncertainty in analyses of complex systems. Reliab Eng Syst Saf 2003;81:23–69.
[41] Olsson A, Sandberg G, Dahlblom O. On Latin hypercube sampling for structural reliability analysis. Struct Saf 2003;25:47–68.
[42] Hezarjaribi M, Bahaari M, Bagheri V, Ebrahimian H. Sensitivity analysis of jacket-type offshore platforms under extreme waves. J Constr Steel Res
2013;83:147–55.
[43] Venture SJ, Committee GD. Recommended seismic design criteria for new steel moment-frame buildings. Federal Emergency Management Agency; 2000.
[44] API R. 2SIM, structural integrity management of fixed offshore structures. American Petroleum Institute; 2011.
[45] ISO P. Natural gas industries-Fixed steel offshore structures. ISO; 2007.
[46] DULAM. Final report jacket inspection for platform R 2011. Reshadat Field.
[47] (ABS) ABoS. Commentary on the guide for buckling and ultimate strength assessment for offshore structures. Houston, TX, USA: ABS Plaza; 2014.
[48] Houlsby G, Taylor RE. Probabilistic models applicable to the short-term extreme response analysis of jack-up platforms. 2003.
[49] Clough RW, Penzien J. Dynamics of structures. Mcgraw Hill; 1975. p. 234–300.
[50] Pákozdi C, Östeman A, Stansberg CT, Peric M, Lu H, Baarholm R. Estimation of wave in deck load using CFD validated against model test data. The twenty-fifth
international ocean and polar engineering conference. International Society of Offshore and Polar Engineers; 2015.
[51] van Raaij K, Gudmestad OT. Wave-in-deck loading on fixed steel jacket decks. Mar Struct 2007;20:164–84.
[52] Howarth M, Dier A, Jones W, Hunt R. Jack-up response to wave-in-deck loads during extreme storms. Mar Struct 2004;17:275–90.
[53] Scharnke J, Hennig J. Vertical wave impact loading on a fixed platform deck. Proceedings of the 34th international conference on ocean, offshore and arctic
engineering (OMAE2015-41131). New York (NY): ASME St. John's, Canada; 2015.
[54] Suyuthi A, Haver SK. Extreme loads due to wave breaking against platform column. The nineteenth international offshore and polar engineering conference.
International Society of Offshore and Polar Engineers; 2009.
[55] Sorensen RM. Basic coastal engineering. Springer Science & Business Media; 2005.
[56] Zeinoddini M, Parke G. Elastic shakedown and adaptation of the response in laterally impacted steel tubes. Int J Damage Mech 2011;20:400–22.
[57] Jalayer F, Cornell C. Alternative non-linear demand estimation methods for probability-based seismic assessments. Earthquake Eng Struct Dyn 2009;38:951–72.
[58] Li L, Li P, Liu Y. How we determine the design environmental conditions and how they impact the structural reliabilities? ASME 2014 33rd international
conference on ocean, offshore and arctic engineering. American Society of Mechanical Engineers; 2014. V04AT2A017–V04AT02A.

79