Marine Structures 36 (2014) 88–104

Contents lists available at ScienceDirect

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

A simplified method for reliability- and

integrity-based design of engineering systems
and its application to offshore mooring systems
Mir Emad Mousavi a, *, Paolo Gardoni b
a
Texas A&M University, TAMU-3136 College Station, TX 77843, USA
b
University of Illinois at Urbana-Champaign, Urbana, IL, USA

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

Article history:
This paper presents a simplified method for the reliability- and the
Received 15 February 2013
integrity-based optimal design of engineering systems and its
Received in revised form 26 December 2013
Accepted 4 February 2014 application to offshore mooring systems. The design of structural
systems is transitioning from the conventional methods, which are
Keywords:
based on factors of safety, to more advanced methods, which
Reliability require calculation of the failure probability of the designed system
Integrity for each project. Using factors of safety to account for the un-
Design certainties in the capacity (strength) or demands can lead to sys-
System tems with different reliabilities. This is because the number and
Structure arrangement of components in each system and the correlation of
Offshore their responses could be different, which could affect the system
Mooring
reliability. The generic factors of safety that are specified at the
Optimization
component level do not account for such differences. Still, using
Probability
factors of safety, as a measure of system safety, is preferred by
many engineers because of the simplicity in their application. The
aim of this paper is to provide a simplified method for design of
engineering systems that directly involves the system annual
failure probability as a measure of system safety, concerning sys-
tem strength limit state. In this method, using results of conven-
tional deterministic analysis, the optimality factors for an
integrity-based optimal design are used instead of generic safety
factors to assure the system safety. The optimality factors, which
estimate the necessary change in average component capacities,
are computed especially for each component and a target system
annual probability of system failure using regression models that

* Corresponding author. Present address: Aker Solutions Inc., 3010 Briarpark Dr., Suite 500, Houston, TX 77042, USA.
Tel.: þ1 713 981 2047.
E-mail address: emad.mousavi@akersolutions.com (M.E. Mousavi).

0951-8339/$ – see front matter Ó 2014 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.marstruc.2014.02.001
 M.E. Mousavi, P. Gardoni / Marine Structures 36 (2014) 88–104 89

estimate the effect of short and long term extreme events on

structural response. Because in practice, it is convenient to use the
return period as a measure to quantify the likelihood of extreme
events, the regression model in this paper is a relationship be-
tween the component demands and the annual probability density
function corresponding to every return period. This method ac-
counts for the uncertainties in the environmental loads and
structural capacities, and identifies the target mean capacity of
each component for maximizing its integrity and meeting the
reliability requirement. In addition, because various failure modes
in a structural system can lead to different consequences
(including damage costs), a method is introduced to compute
optimality factors for designated failure modes. By calculating the
probability of system failure, this method can be used for risk-
based decision-making that considers the failure costs and con-
sequences. The proposed method can also be used on existing
structures to identify the riskiest components as part of inspection
and improvement planning. The proposed method is discussed
and illustrated considering offshore mooring systems. However,
the method is general and applicable also to other engineering
systems. In the case study of this paper, the method is first used to
quantify the reliability of a mooring system, then this design is
revised to meet the DNV recommended annual probability of
failure and for maximizing system integrity as well as for a
designated failure mode in which the anchor chains are the first
components to fail in the system.
Ó 2014 Elsevier Ltd. All rights reserved.

1. Introduction

The extensive costs and damages induced by the 2010 oil spill in the Gulf of Mexico as well as other
offshore failures during recent years have intensified the efforts to improve the reliability and integrity
of offshore systems. The relatively high rates of offshore failure are more than what is considered
acceptable, which implies a need for improved methods for better safety of current and future offshore
structures. For example, more than twenty-three permanent mooring systems have failed since 2000;
1500 mooring lines were either repaired or replaced. The damage cost of a single mooring failure event
was approximately $1.8 billion [1]. The costs of loss of lives or damages to the environment cannot be
quantified. As offshore drilling and production sites move to deeper and more challenging environ-
ments, the safety of offshore systems becomes even more important, demanding technology devel-
opment that accounts for their inherent uncertainties. In response, the design methods seem to be in a
transition from the conventional methods, which use factors of safety (FoS), to more advanced
methods, which include the system reliability and risk assessments.
The simplicity of using FoS to account for the uncertainties has made their application favorable
to many engineers; however, as highlighted by the API RP 2SK [2], several studies have shown that
the current design practice results in systems with inconsistent failure probabilities, and thus,
reliabilities. It is because the safety of a system does not solely depend on the FoS of its compo-
nents. In turn, structural reliability methods can be used to design an offshore system for a target
probability of failure. The DNV OS E301 [3] already facilitate the application of structural reliability
methods for design of mooring systems by providing a target annual probability of failure, as an
alternative to its FoS-based method; however, such application requires practical methods that are
feasible using the available data and decent amount of engineering and computing resources.
Developing simplified methods that can quantify and target the offshore system reliability is an
important step toward the successful advancements in offshore engineering, including the design
of offshore mooring systems.
90 M.E. Mousavi, P. Gardoni / Marine Structures 36 (2014) 88–104

The aim of this paper is to propose a simplified method that can be used for the design of reliable
mooring systems, which is also applicable to other engineering systems. The proposed method is based
on the definition of integrity-based optimal design [4]. In assessing the system reliability in this paper,
any failure in the intact system is assumed undesirable, so this method targets the optimal design for
maximum system integrity given a target annual probability of any component failure in the intact
system. This definition of system failure complies with the ultimate limit-state (ULS) criterion provided
for the target annual probability of failure by DNV OS E301. The computation of the annual failure
probability of such a system and its components and determining the optimal average strength of its
components under extreme environmental conditions to meet the target reliability and achieve a
maximum integrity or designated failure mode are discussed in the paper. The scope of this paper is
limited to the considerations related to an ultimate-strength failure of the structural components.
Therefore, other design considerations (e.g. the fatigue failure, geometrical limitations) are beyond the
scope of this work and should be checked separately.
After this introduction, the paper is organized into two main sections. The next section describes the
proposed method. Next, the proposed method is illustrated in a case study about mooring systems.

2. Methodology

This section first discusses the calculation of the probability of failure of a series system. Then, a
simplified method is introduced to calculate the annual probability of failure for structural systems.
Next, the proposed method is used to derive simplified equations for an integrity-based optimal design.
Finally, the adjustments in the method for favoring a designated failure mode are presented.
A closed-form solution for the probability of failure of a mooring system requires a relationship
between the probability of failure of the system and its components. In this study, by assuming the
independency of the component failures under conditioned demand, a more realistic approach is used.
Thus, for a given series system that is subject to a system demand (environmental load), dE, assuming
that the component failures are conditionally independent, the conditional probability of system
failure can be written as [7]

N n
Y   o Y
N    

Pðgs  0jdE Þ ¼ 1  1  P gj  0dE ¼ 1 1  P cj  dj dE (1)
j¼1 j¼1

where gs is the system limit-state function, N is the total number of components, gs  0 indicates the
system failure [8], P($) denotes the probability, gj is the component limit-state function,

gj ¼ cj  dj (2)

where cj is the component capacity (strength) and a random variable, dj is the component demand
(maximum internal force) that is determined through a deterministic analysis (e.g. Finite Element)
under each dE, and therefore gj  0 indicates the component failure. Then, using the total probability
rule [9], the probability of the system failure can be calculated as
2 3
Z Z Y
N    
Pðgs  0Þ ¼ Pðgs  0jdE ÞfðdE ÞddE ¼ 41  
1  P cj  dj dE 5  fðdE ÞddE (3)
j¼1
B B

where f(dE) is the probability density function (PDF) and B is the domain of dE. Similarly, the probability
of failure of each component can be computed as:
  Z
  
P gj  0 ¼ P cj  dj dE fðdE ÞddE (4)
B

Eqs. (3) and (4) suggest that

 calculating the probability of failure of a component or the system
requires computing Pðcj  dj dE Þ, which depends on the cumulative distribution function of cj, CDFcj,
and on f(dE). Please note that depending on the availability of statistical data, the length of each
 M.E. Mousavi, P. Gardoni / Marine Structures 36 (2014) 88–104 91

component in this analysis could equal the full length of the actual mooring components or the lengthy
components could be divided into several shorter components for this reliability analysis. A finer
component size would improve the accuracy of this reliability assessment provided that the statistical
data are also based on similar size test samples. A change in the length of structural components would
not alter the equations in this paper because the system is treated as a series system and thus using the
same formulation with relevant number of components and capacity statistical data and component
demand information are applicable.
The determination of CDFcj and f(dE) are discussed in the following sections. Please note that the
above formulations are general; however, in the following sections, f(dE) will be determined and
discussed for the annual failure probabilities and therefore will be the annual probability density
function of the environmental demands.

2.1. Estimating CDFcj

The capacities of the mooring system components approximately follow a lognormal distribution
[5,10]. Therefore, we can write
2 pffiffiffiffiffiffiffiffiffiffiffiffiffiffi2  3 2 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 3
"   # 1þCOVj
  ln dj  mj 6 ln
½ =d 7 6ln 1 þ COVj2 =lj 7
6 E c 7 6 rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 7
CDFcj dj ¼ F ¼ F6rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
j j

 5ffi 7 ¼ F 4  ffi 5 (5)
sj 4
ln 1 þ COVj 2 ln 1 þ COVj 2

where mj is the mean of the natural logarithm of cj

 !
   1 Var c
mj ¼ ln E cj  ln 1 þ  2j (6)
2 E cj

and sj is the standard deviation of the natural logarithm of cj

vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u (  )
u Var c
sj ¼ tln 1 þ  2j (7)
E cj

and E[cj] and Var(cj) are the mean and the variance of cj respectively. E[cj] can be estimated by the
sample mean of the component capacities, b c j , and Var(cj) can be estimated by the sample variance of
the component capacities, Varðbc j Þ. In Eq. (5), COVj is the coefficient of variation of cj defined as
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 ffi
Var cj
COVj ¼  (8)
E cj

and lj is the mean safety factor of component j, which we define as


E cj
lj ¼ (9)
dj

Eq. (5) calculates the conditional probability of failure of component j given dj. This equation is
visualized in Fig. 1. Because the coefficient of variation for a class of components is usually similar and if
the component capacity follows a lognormal probability distribution, given such information, this figure
can be used to estimate this conditional probability of failure for a target mean safety factor. Note that, as
expected, as lj approaches 1 the probability of failure approaches 0.5 irrespective of the value of COVj.

2.2. Estimating f(dE) and calculating the component and system annual failure probabilities

It is common to use the mean recurrence time, usually known as the (mean) return period of
an extreme event, T, as a measure to quantify its likelihood. T is the average time interval
92 M.E. Mousavi, P. Gardoni / Marine Structures 36 (2014) 88–104

Fig. 1. A visualization of Eq. (5).

between two consecutive occurrences of such an event and 1=T is the average rate of its occur-
rence in one year [9]. It is common to compute the likelihood that an event with a return
period of T will not be exceeded within t ¼ 1 year by assuming it as a Poisson Process and using
an exponential function as et=T , in which t is the reference duration (i.e. 1 year) [9]. Thus, if dj is
the internal force in component j in response to the environmental condition dE that
corresponds to T dE , then f(dE) is the derivative of the cumulative distribution function of T dE with
respect to dE

d  1=T d  dT dE 1 1=T d dT dE
f ðdE Þ ¼ e E ¼ 2
e E (10)
dT dE dd E T dE ddE

Note that T dE > 0 (years). Eq. (10) is the first step for a change of variable from dE to T dE in the
integral in Eqs. (3) and (4). An advanced reliability analysis of an offshore structural system involves the
development of the combined probability distribution of environmental events and uses them to
define the probability corresponding to any structural response. However, in this paper, by decoupling
the analysis of environmental conditions with various return periods from the structural response
analysis, which are usually performed by different specialists in the industry, f(dE) is expressed in terms
of T dE as a simplified practical approach that allows engineers who use the metocean data to analyze or
design the structure to complete a reliability-based design using the same analysis results that they
typically produce.
The next step before the integrals in Eqs. (3) and (4) can be used is developing a method for esti-
mating dj given T dE . For any structural component j, dj is usually measured or computed using physical
or numerical models under a limited number of environmental conditions corresponding to a few
return periods. However, in reality, dj is a continuous random variable. A regression model can estimate
dj as a function of T dE based on the available data of dj versus T dE . Review of available data from dynamic
(frequency domain) analysis of mooring systems under environmental conditions with return periods
of 1, 10, 20, 50 and 100 years suggests that for all the mooring components, dj has a highly linear
relationship with the logarithm (in base 10) of T dE . Based on this observation, following the general
formulation for probability demand models [11–13], we suggest the following regression model to
estimate dj as a function of T dE :
 M.E. Mousavi, P. Gardoni / Marine Structures 36 (2014) 88–104 93

 
dj ¼ qj1 þ qj2 log T dE þ sdj 3 dj (11)

where Qj ¼ ðqj ; s2dj Þ indicates the unknown model parameters in which qj ¼ ðqj1 ; qj2 Þ, 3 dj is a
random variable with zero mean and unit variance, and sdj is the standard deviation of the model error
sdj3 dj. Please note that under very large T dE , the proposed linear regression model in Eq. (11) may no
longer be accurate; however, as implied by Eq. (10), the contribution of such large T dE is not significant.
Still, estimating the model parameters using results of deterministic analysis with larger return periods
or possible use of nonlinear regression models could improve the accuracy of the model. Eq. (11) can be
written in a matrix form as

dj ¼ Hqj þ sdj 3 (12)

where dj is the vector of component demands, H is the matrix of regressors (1 and the logarithm of T dE ),
and 3 is a vector standard normal random variable. For example, if k data points are available for dj
versus T dE , we can write
2 3 2  3 2 3
dj1 1 log T dE 1   3 dj1
6 dj2 7 6  7
6 7 ¼ 61 log T dE 2 7 qj1 þ s 6 7
6 3 dj2 7 (13)
4 « 5 4« « 5 qj2 dj 4 « 5
 
djk 1 log T dE k 3 djk

The posterior distribution of Qj as suggested by Box and Tiao [14] is

       b    
p Qj dj fp s2j s2dj p q j qj ; sdj p Qj
2
(14)

where pð:Þ is the probability of the event inside the parenthesis and

qb j ¼ ðH0 HÞ1 H0 dj
 0  
b
s2j ¼ 1y dj  d b
dj  d
j j
(15)
y ¼ k2
b
b ¼ Hq
d j j

Assuming independency of qj and s2dj we can write [14]

     
p Qj ¼ p qj p s2dj fs2
dj (16)

and thus rewrite Eq. (14) as

        
  b ; s2
p qj ; s2dj dj fp s2dj s2j p qj  q j dj (17)

where the marginal posterior distribution of s2dj is the inverse chi-square distribution ys2j c2
y with mean
equal to ys2j =ðy  2Þ and the variance equal to 2y2 s4j =½ðy  2Þ2 ðy  4Þ. The marginal posterior distri-
b ; s2 ðH0 HÞ1 ; y, which is written as
bution of qj is a multivariate t distribution, tk ½ q j j

    ðyþ2Þ=2
   Gðyþ2
2 Þ
jH0 Hj1=2 s2 qb
q j H0 H qb
qj
p qj dj ¼ 2  1 þ ys 2
½Gð12Þ Gð2yÞy (18)
N < qjq < N; q ¼ 1; 2

where qb is the mode (and the mean) of q with covariance matrix ys2 ðH0 HÞ1 =ðy  2Þ.
j j j
b in Eq. (11), we can compute b
Thus, using q d j , which is the median value of dj for a given T dE . Then, by
j
substituting Eqs. (5) and (10) into Eqs. (3) and (4), and by changing the integral variable from dE to T dE ,
we can estimate the probability of failure of each component as
94 M.E. Mousavi, P. Gardoni / Marine Structures 36 (2014) 88–104

8 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  9
>  >
  Z >
<ln Tk 1 þ COVj2 =b l j  > = e1=T dE
P gj  0 z F rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 ffi dE dT dE
>
>  > T 2
>
0 : ln 1 þ COVj 2
 ; d E

8 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  9
 >
Z N >>
<ln 1 þ COVj2 =b l j  >
= e1=T dE
F rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 
þ
>  ffi dE > 2
dT dE (19)
>
: ln 1 þ COV 2  > ; T dE
Tk
j

The two integrals in Eq. (19) correspond to the interpolation and extrapolation regions of the
domain of environmental loads in this analysis model because b
l j in both the regions is estimated based
on the regression model based on response data from 1 to T k years return periods:

E cj b
cj
b
lj ¼ z (20)
b b  
dj q 1j þ bq 2j log T dE
The two integrals in Eq. (19) can be used for assessing the contribution of each region in the
computed annual probability of system failure. For such calculation, larger T k, which is the largest
return period that is used as input in Eq. (13), will improve the accuracy of the analysis. Therefore, we
summarize Eq. (19) as:
8 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  9
>  >
  Z N >
<ln 1 þ COVj2 =b l j  > = e1=T dE
P gj  0 z   ffi dE
F rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi dT dE (21)
> > T 2
>
: ln 1 þ COV 2  > ;
0
dE
j

Similarly, Pðgs  0Þ can be computed as

2 2 8 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  933
>  >
Z N6 YN 6 >
<ln 1 þ COVj2 =b l j  >=77 e1=T dE
Pðgs  0Þz 61 
4
61  F
4   ffi dE 7
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 7
55 dT dE (22)
>
>  > 2
: ln 1 þ COV 2  > ; T dE
0 j¼1
j

In the derivation of Eqs. (21) and (22), it is assumed that b

q 1j and bq 2j and thus bl j are estimated using
the results of a model in which the environmental conditions corresponding to each T dE are simulated
in all directions, but some analysis models only simulate unidirectional environmental loads. In turn,
hurricane conditions are usually associated with various possible loading directions on a floating
structure because of their cyclic nature and because the offshore platforms are usually located far from
any major barriers. If the hurricane model already applies the realistic hurricane loading corresponding
to each return period, then Eqs. (21) and (22) can be used directly for computing the probability of
failure of components or the system. However, if a simplified modeling in which all the possible loading
directions are applied in unidirectional loading models are used, the conditional annual probability of
failure can be computed for each direction, a. For the special case where under each return period, the
occurrence of the dominant environmental load is equally possible in all the directions, we can write
the joint PDF of dE and a as

1 dT dE da
fðdE ; aÞ ¼ e1=T dE (23)
T dE
2 ddE 2p

Thus, we can modify Eqs. (21) and (22) as

8 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi dir  9
> 
  1 Z 2p Z N <>ln 1 þ COVj2 =b l j  > >
= 1=T dE
P gj  0 z F rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi d E e dT dE da (24)
2p 0 >    > 2
>
:  >
 ; T dE
0
ln 1 þ COV 2 j
 M.E. Mousavi, P. Gardoni / Marine Structures 36 (2014) 88–104 95

and
2 2 8 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi dir  933
> 
Z 2p Z N 6 YN 6 >ln
< 1 þ COVj2 =b l j  > >
=77 e1=T dE
1 6 6 77
Pðgs  0Þz 61  61  F   ffi dE 77
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi dT dE da (25)
2p 0 0 4 4 >
>  >
> 55 T 2
j¼1 : ln 1 þ COV 2
 ; dE
j

where

dir bc j
b
l z (26)
j
b dir
b dir  
q j1 þ q j2 log T dE
In practice, the waves are unlikely to be completely in one direction, which could lead to some
dissipation of their energy and lower loading magnitude on the offshore system compared to results of
unidirectional wave loading models; therefore, Eqs. (23)–(25) are associated with some conservatism.
If the likelihoods of the occurrence of the dominant environmental loads are not equal in all the di-
rections, then statistical data about such likelihoods are necessary to update Eqs. (23), (23) and (24)
before the proposed simplified reliability-based design method is used.
Finally, for the special conditions where a component is under no demand (internal forces)
until environmental loads are large (e.g. anchors of a catenary mooring system could have zero
loads unless vessel motion exceeds a minimum offset), a bilinear model can be used instead of Eq.
(11) as
(
0 if 1  T dE < T 0j
dj ¼   (27)
qj1 þ qj2 log T dE þ sdj 3 dj if T 0j  T dE  T k

where T 0j is the largest return period with a zero dj and can be estimated as

   
b b
0zb
q 1j þ bq 2j log T 0j 0T 0j z10  q 1j = q 2j (28)

2.3. Optimal design for maximum integrity

This section discusses a method that computes the optimal average capacity of each component
so that the system meets a target reliability requirement but also has a maximum integrity.
Mousavi and Gardoni [4] discussed the optimal design of engineering systems in terms of their
integrity, where integrity was defined as the balanced contribution of system components to its
reliability. They suggested that to achieve a maximum integrity, the conditional probabilities of
failure of all the system components given system failure should be the same. Then, they proposed
the Integrity Index as a measure to quantify the system integrity, which equals 1 minus the
maximum difference of the conditional probabilities of failure of system components given system
failure. They also showed that in series system, such optimal design leads to equal probabilities of
failure for all the system components, which they called the optimal failure probability (OFP) [4].
They suggested that OFP of series systems with small component probabilities of failure could be
estimated as:
 
     P gsoptimal  0
 optimal optimal
OFP ¼ P gj  0gs  0  P gs 0 z (29)
N
We can use Eq. (29) for an estimation of the annual optimal failure probability of the components
for a target annual probability of system failure. Mousavi and Gardoni also defined the Equivalent
Demand, dej , as a representative of all the demands on a component as:
96 M.E. Mousavi, P. Gardoni / Marine Structures 36 (2014) 88–104

h  i
dej ¼ CDF1
cj P gj  0 (30)

Therefore, using Eq. (5), we can write

rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 ffi n  o
exp ln 1 þ COVj2  F1 P gj  0

dej ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi E cj (31)
1 þ COVj 2

By replacing the component (annual) failure probability in Eq. (31) with the (annual) OFP, we es-
timate the optimal mean capacity of each component as:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
h i dej 1 þ COVj2
E coptimal
j
¼ rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 ffi  (32)
exp ln 1 þ COVj2  F1 fOFPg

Mousavi and Gardoni [4] suggested that an integrity-based optimal design of each component can
be assessed using optimality factor of each component for a given target (annual) probability of system
failure, where the optimality factor, gj, is defined as the ratio of the optimal to the current mean ca-
pacity of each component. Using Eqs. (31) and (32), we can derive an equation for the optimality factor
of each component as
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 ffi n n  o o
gj ¼ exp ln 1 þ COVj2  F1 P gj  0  F1 fOFPg (33)

where gj ¼ E½coptimal
j
=E½cj . Eq. (33) can be used directly to find the Optimality Factor of a component
given its OFP and current (annual) probability of failure. We can also rewrite Eq. (33) in terms of
reliability indices as
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 ffi  
optimal
gj ¼ exp ln 1 þ COVj2  bj  bj (34)

where

boptimal
j ¼ F1 fOFPg (35)

and
n  o
bj ¼ F1 P cj  dej (36)

are the reliability indices that correspond to OFP and the current probability of failure of component j.
An important observation from Eq. (34) is that the Optimality Factor depends on the difference be-
tween the optimal and current reliability indices of a component (with lognormal capacity distribu-
tion) and not their absolute reliability indices.
Fig. 2 plots Eq. (34). This figure can be used to find the optimality factors corresponding to the
typical differences of component reliability indices and the coefficients of variations of the capacities
for a design with maximum integrity. In this plot, the optimality factors corresponding to each positive
difference of reliability indices equal 1 divided by the negative value of the same differences.
If updating the system components based on the optimality factors changes the component de-
mands (e.g. if the weight of the current and optimal components are different), the computation of the
optimality factors should be determined iteratively. Iterative computation can also be used to minimize
the error due to the approximation in Eq. (29). By updating the component capacity probability dis-
tributions using the updated average capacities based on Eq. (34), and calculating the (annual) prob-
ability of failure for the updated system using Eq. (3), the new (annual) probability of system failure can
be compared with that of the target (annual) probability of failure. OFP or bj
optimal
are the same for all
 M.E. Mousavi, P. Gardoni / Marine Structures 36 (2014) 88–104 97

Fig. 2. A visualization of Eq. (32).

the components in the series system and can be numerically determined to match the computed with
the target (annual) probability of system failure.

2.4. Optimal design for a designated failure scenario

It is not possible to design a system with no chance of failure; thus, imposing a preferred failure
scenario to the system while maintaining its target safety level, is favorable to many engineering
systems. Designating a preferred failure scenario can potentially reduce the costs of damages and
consequences of a system failure. This can be particularly important to offshore structures for which
some of their possible failure scenarios are significantly more costly than other ones. For example, the
failure of an anchor can directly damage the subsea equipment but the anchor chain failures are usually
less important.
We suggest designing a failure scenario by adjusting the average capacities of the system compo-
nents so that selected components have a higher chance of failure compared to other components but
they are set so that the overall system (annual) probability of failure still meets a target reliability
requirement. To do so, we divide the components into two groups: fuse and non-fuse, based on the
preferred failure scenario. Fuse components are those that are preferred to fail first, protecting the non-
fuse components. The OFP for the two groups of components could be updated as

R  
optimal
OFPfuse z  P gs 0 (37)
Nf

and

1R  
optimal
OFPnonfuse z  P gs 0 (38)
N  Nf

where Nf is the number of fuse components, R is the conditional probability of the preferred failure
scenario given system failure (e.g. 0.99), and OFBfuse and OFBnon-fuse are the optimal failure probabilities
of the fuse and non-fuse components. They can be used instead of the OFP in Eqs. (33) and (35) to find
the optimality factor of each component.
98 M.E. Mousavi, P. Gardoni / Marine Structures 36 (2014) 88–104

In the derivation of Eqs. (37) and (38), the assumptions of independency of the failure events of the
components and that these probabilities are small are made; therefore, the probability of failure of two
subseries systems become the summation of the probabilities of failure of their components (the
higher order terms in probabilities of the intersection of the failure events are neglected.) For more
complex problems or for increasing the accuracy of the estimated probabilities from these two
equations, the OFPs can be determined iteratively by adjusting the capacity distribution parameters
(e.g. mean capacity) and calculating the probability of failure of series fuse or non-fuse subsystems.

3. Case study: a mooring system

This section illustrates the proposed simplified method for the reliability- and integrity-based design
of engineering systems considering a mooring system. Mousavi and Gardoni [4] assessed the integrity-
based optimal design of a mooring system in a case study by estimating the (annual) PDF of the
environmental loads using an annual probability mass function (PMF). They first estimated the (annual)
probability of failure and the Integrity Index of the mooring system; then, they calculated the optimality
factors of the system components to maximize the Integrity Index but maintain the same (annual)
probability of failure. In this example, we will assess the same mooring system based on the new
proposed method. This method is expected to produce more accurate results because of a key difference
between the two studies, which is that the (annual) PDF of the demand in this study is estimated
throughout the domain of the component demands using a regression model rather than using a PMF at
a few return periods between 1 and 100 years. Moreover, the failure events of the components are not
assumed independent in this study, but only conditionally are assumed independent. We have also used
smaller size components with coefficient of variations of component capacities that directly correspond
to typical test results, which is expected to produce more realistic results.
The MODU of this case study is designed based on API RP 2SK [2] requirements, but because API
does not directly suggest a target annual probability of failure, we used the suggestion by DNV OS E301
[3] for the target annual probability of system failure in calculation of the optimality factors. Three
target annual probabilities of system failure are discussed by DNV OS E301 for the following three limit
states: the ultimate limit state (ULS), the accidental limit state (ALS), and the fatigue limit state (FLS).
Among them, the ULS is relevant to the calculated annual probability of failure in this case study. Two
ULS target annual probabilities of failure are suggested in DNV OS E301: 104 for Class 1 mooring
systems, where the consequences of a failure are unlikely to lead to unacceptable consequences (loss of
life or collision with an adjacent platform, uncontrolled outflow of oil and gas, capsize or shrinking),
and 105 for Class 2 mooring systems, where unacceptable consequences are expected in case of a
failure. For this illustration, we use 104 as the target annual probability of system failure.
We considered two design strategies: Strategy A corresponds to a maximum system integrity (I ¼ 1)
and Strategy B corresponds to a design where the anchor chains serve as fuse components with a 99%
chance that they will fail before other mooring components fail. The same dynamic (frequency domain)
analyses produced by InterMoor Inc. [15] and used by Mousavi and Gardoni [4] are used in this assessment,
except that because we used a finer component size in our reliability assessment, we used linear inter-
polation to estimate the demands within discretized components using available demand data at their two
ends (the error associated with this approach can be eliminated by using data from a finer mesh in the FE
model). Similar to the assessment of Mousavi and Gardoni and for the purpose of comparability with the
original design, we did not add any additional assumptions concerning the effects of corrosion or marine
growth compared to the original design in the analysis. In anyways, such phenomena seem irrelevant to
this temporary mooring system. As suggested by Mousavi and Gardoni, such phenomena could be
included in the probabilistic analysis of long term projects by considering the worst conditions in the
analysis, by conducting sensitivity analyses, or by analyzing the system under different levels of likelihood
of marine growth and corrosion rates and applying the total probability rule to summarize the results.

3.1. Mooring system properties

This Mobile Offshore Drilling Unit (MODU) is a semisubmersible drilling rig located at 4320 ft
(1317 m) water depth in the Gulf of Mexico. It includes 12 mooring lines (8 conventional and 4 preset
 M.E. Mousavi, P. Gardoni / Marine Structures 36 (2014) 88–104 99

Table 1
Basic properties of the mooring system components.

Mooring lines Line components Anchor

a capacity (kipa)
Type Stiffness Wet weight Friction Min break Length (ft )
(103 kipa) (lb/fta) factor strength (kipa)
1, 5, & 8 Anchor chain 135 87.3 1.0 1450 4138–4145 733
Rig wire 137 18.8 0.6 1452 6107–6144
2, 4, & 6 Anchor chain 135 87.3 1.0 1450 4072–4293 1729
Rig wire 137 18.8 0.6 1452 5980–6208
3&7 Anchor chain 135 87.3 1.0 1450 3962–4253 1064
Rig wire 137 18.8 0.6 1452 6037–6248
9, 10, 11, & 12 Anchor chain 135 87.3 1.0 1570 3200 1385
Insert wire 168 23.1 0.6 1825 6500
Rig chain 135 87.3 1.0 1850 457–600
a
1 kip is about 4.45 kN, 1 ft is about 0.30 m and 1 lb/ft is about 1.49 kg/m.

hurricane lines) and anchor foundations. Basic properties of the mooring components are given in
Table 1. The lines are arranged as four groups of three lines with about 15 angle between the lines. The
payout (horizontal distance between the two ends) of each lines in about 8500 ft (2591 m) and the
pretensions of the conventional and hurricane lines are about 200 kips (890 kN) and 250 kips
(1112 kN), respectively. Because this system has a total of 40 components, for Strategy B, 12 components
will be considered as fuse and the remaining components as non-fuse.

3.2. Environmental loads

Table 2 shows the wave, wind, and current data that were used for environmental conditions in this
study. 72 load directions (every 5 ) were considered for each return period (therefore, a total of 288
load cases were used for five return periods of 1-, 10-, 20-, 50-, and 100-year hurricane.) The wind
direction was taken to be 15 different from the wave and current directions in all the load cases. The
NPD (Norwegian Petroleum Directorate) and the JONSWAP (Joint North Sea Wave Analysis Program)
energy spectrums were used for winds and waves respectively. The peakedness parameter of 2 and
spreading parameters of 2.5 were assumed for the waves. No kinematic reduction factor was applied.

3.3. Random variables

The uncertainty in the component loads (demand) and ultimate-strengths (capacity) are consid-
ered. The statistical properties of component capacities are given in Table 3, which are based on
suggestions from Refs. [7,10,16]. In this case study we descritized the segments so that every two chain
components (links) have a length of six times its diameter and each cable component has a length of
32.5 times its diameter [17].
The regression parameters for the component demands are determined using dynamic analysis
(accounting for mean, wave frequency, and low frequency motions using a decoupled frequency-
domain modeling) under environmental conditions corresponding to 10-, 20-, 50-, and 100-year

Table 2
Metocean data for the project site.

Environment Hurricane return period (years)

10 20 50 100
Wind 1-min Speed, knots (m/s) 73.8 (38.0) 89.5 (46.1) 101.0 (52.0) 108.6 (55.9)
1-h Speed, knots (m/s) 59.8 (30.8) 71.1 (36.6) 79.1 (40.7) 84.3 (43.4)
Wave Significant height, ft (m) 36.1 (11.0) 42.7 (13.0) 47.2 (14.4) 50.2 (15.3)
Peak period, sec 12.06 12.78 13.32 13.59
Surface current speed, knots (m/s) 1.9 (1.0) 2.1 (1.1) 2.9 (1.5) 3.3 (1.7)
100 M.E. Mousavi, P. Gardoni / Marine Structures 36 (2014) 88–104

Table 3
Statistical properties of the capacity random variables.

Random variable PDF Mean/nominal C.O.V.

Anchor capacity Lognormal 1.1 0.15
Chain strength Lognormal 1.1 0.05
Wire strength Lognormal 1.1 0.05

return periods. The review of the results confirm that the mean of 3 dj is about zero and its coefficient of
variation (C.O.V.) is about 0.02, which indicates that the regression model error is small.

3.4. Deterministic results

Deterministic results indicated that both the intact and damaged (1 broken line) conditions of the
mooring system of this study meet the API RP 2SK requirements. The minimum safety factor of the
mooring lines under the 10-year peak wave (hurricane) environment ranged from 2.3 to 3.3 in different
direction in the intact system and from 1.8 to 2.3 in the damaged system. The anchors also met the
requirement with safety factors ranging from 1.81 to 4.86. Under this 10-year extreme environment,
the vessel offset remained close to 10% of water depth in all the directions at the intact system and
between 13% and 14% of the depth in the damaged condition.

3.5. Probabilistic results and discussion

Table 4 presents the annual probability of failure and the Integrity Index of the mooring system
using the method proposed in this paper as well as the result by Mousavi and Gardoni [4]. The results
indicate that the mooring system Integrity Index of 0.78 is smaller than that of Mousavi and Gardoni. By
further evaluating the effect of each of the differences between the two studies, we observed that by
maintaining the size of the reliability analysis components same as that of Mousavi and Gardoni, the
Integrity Index was increased to 0.89, which shows that by reducing the size of the elements in reli-
ability analysis, we are capturing more detailed information about the mooring system that cannot be
captured using larger size elements. In turn, the annual failure probability in this study is computed to
be 1.71  103, which is slightly smaller but in a same order of magnitude of the estimation of Mousavi
and Gardoni, which was 5.05  103. Mousavi and Gardoni estimated the PDF of the demands using a
four-point PMF but in this study, a more advance model is used to estimate the PDF of the environ-
mental loads. For example, they assumed an annual probability of 0.9 for the maximum demands with
the return periods corresponding to 10-year extreme event response. Such method is likely to over-
estimate the annual probability of failure of the system for that return period. However, their method is
likely to underestimate the contribution of the environmental loads with greater than 100-year return
periods as their analysis was limited to the 100-year demands. In turn, the discretization of the
mooring segments to smaller components in this study has reduced the probability of failure because
under each environmental condition, the maximum demand is smaller in a major part of each long
segment, which is more realistic and unbiased compared to when lengthy components are considered
as individual elements. Computation also shows that if complete independency was assumed for the
failure of the mooring components, the annual probability of system failure in this study would in-
crease to 5.72  103, which is almost four times the computed annual probability of failure using Eq.
(3). This result highlights the significance of the conditional independency assumption compared to

Table 4
Summary of the integrity and reliability analysis.

Method Probability of failure Integrity Index

Mousavi and Gardoni [4] 5.05  103 0.89
This study 1.71  103 0.78
 M.E. Mousavi, P. Gardoni / Marine Structures 36 (2014) 88–104 101

the non-conditional independency assumption in this reliability analysis. Please note that the calcu-
lated annual probabilities of failure in both studies are larger than the suggested target by DNV OS
E301; however, the design still meets the API RP 2SK requirements (the system could still possibly meet
DNV’s deterministic-based design requirements). The method of estimating the environmental loads
in this project is likely to involve some conservatism, which is not accounted for in any of the two
studies. Also, the DNV recommended annual probability of failure may not be necessary for this
temporary project. However, to illustrate the application of the proposed methodology and the in-
formation that can be obtained with this analysis we apply the proposed quantitative integrity analysis
to this system.
The OFP corresponding to a probability of system failure of 104 for the Strategy A was first esti-
mated using Eq. (29) and used in Eq. (33) to find the optimality factors of all the components. Using Eq.
(22), the updated system annual probability of failure considering such estimated optimality factors
was computed to be 4.85  104. Through an iterative computation, the OFP was adjusted to confirm an
annual probability of system failure of 104 for this mooring system. Similarly, Eqs. (35) and (36) were
used through an iterative process to find the optimality factors for Strategy B. Results are given in Table
5 and shown in Fig. 3.
Review of the results in Fig. 3 suggests that some anchors of this system have the smallest opti-
mality factors among the components whilst the anchor chains and some other anchors have the
highest. The optimality factors of most of the components in this system are greater than 1, which is
expected because the target annual probability of system failure is smaller than the current annual
probability of failure. The imbalance between the contributions of system components to its overall
safety is identified and quantified using these results. For example, the higher relative capacity of some
of the anchors in this mooring system could avoid any anchor failure that could possibly damage the
subsea equipment and pipelines; however, this analysis highlights that some other anchors would
need an increased capacity to achieve a similar level of safety. Most of the anchor chains and the rig
chains of this system should also be stronger under both design strategies; the optimality factors of the
anchor chains and rig chains vary from 1.36 to 1.53 and from 1.20 to 1.22 under Strategy A and from 1.21
to 1.37 and from 1.66 to 1.69 under Strategy B, respectively. The rig and insert wires also require
increased average capacity to meet the reliability requirement; the rig wire optimality factors range
from 1.42 to 1.57 under Strategy A and from 1.97 to 2.18 under Strategy B, the insert wire capacities also

Table 5
Probability of failure and optimality factors for the mooring system components.

Component Line Failure probability Optimality factor, gi Line Failure probability Optimality factor, gi

Strategy A Strategy B Strategy A Strategy B

Anchor 1 1.91  1004 1.60 2.43 7 1.21  1005 1.08 1.63
2 1.61  1008 0.58 0.88 8 3.00  1004 1.72 2.59
3 1.45  1005 1.10 1.67 9 6.23  1006 1.00 1.50
4 8.23  1008 0.66 0.99 10 3.32  1006 0.93 1.40
5 2.40  1004 1.66 2.52 11 4.30  1006 0.96 1.44
6 1.95  1007 0.71 1.07 12 2.50  1006 0.90 1.37
Anchor chain 1 2.17  1004 1.48 1.32 7 2.28  1004 1.49 1.33
2 1.36  1004 1.39 1.24 8 2.84  1004 1.53 1.37
3 2.01  1004 1.46 1.31 9 1.39  1004 1.40 1.25
4 1.93  1004 1.46 1.29 10 1.17  1004 1.37 1.22
5 2.55  1004 1.51 1.35 11 1.22  1004 1.38 1.23
6 2.65  1004 1.52 1.36 12 1.07  1004 1.36 1.21
Rig wire 1 2.83  1004 1.51 2.11 5 3.32  1004 1.54 2.15
2 1.81  1004 1.42 1.97 6 3.47  1004 1.55 2.16
3 2.64  1004 1.50 2.09 7 2.98  1004 1.52 2.11
4 2.52  1004 1.49 2.07 8 3.68  1004 1.57 2.18
Insert wire 9 5.38  1005 1.24 1.72 11 4.75  1005 1.22 1.70
10 4.47  1005 1.21 1.69 12 4.18  1005 1.20 1.68
Rig chain 9 5.25  1005 1.22 1.69 11 4.83  1005 1.21 1.68
10 4.36  1005 1.20 1.66 12 4.31  1005 1.20 1.66
102 M.E. Mousavi, P. Gardoni / Marine Structures 36 (2014) 88–104

Fig. 3. Component optimality factors for the target probability of system failure; top: Strategy A, bottom: Strategy B.

should increase by factors between 1.20 and 1.24 under Strategy A and between 1.68 and 1.72 under
Strategy B. These results clearly suggest that the rig wires in this mooring system are the riskiest
components and the likely failure mode of the intact mooring system. They also compare the safety of
various components in the system and identify the safest or riskiest components and the necessary
change to meet the target level of safety.
A benefit of using various directions and return periods in this analysis is that the variability of the
component demands from individual analysis results does not affect the results significantly. For
example, sensitivity analyses were performed to evaluate the effect of reducing the number of assessed
directions or number of return periods on the results. It is observed that the Integrity Index and the
component optimality factors have insignificant variations unless the number of analyzed directions is
reduced by a factor of 9 (less than 8 directions compared to the original 72 directions), which is mainly
because of the almost symmetric loading and structural geometry in this example. The annual prob-
ability of failure of the system also remained similar (less than 1% of change) after reducing similarly
 M.E. Mousavi, P. Gardoni / Marine Structures 36 (2014) 88–104 103

the number of directions that were included in the analysis. It is also observed that performing analysis
over less number of return periods, as long as the maximum and minimum return periods remain
consistent, causes negligible changes in the results, which confirms the linear correlation between the
component demands and the logarithm of the return periods, but also that small variations of
component demands have minor effect on the system integrity and reliability results. These results
show the flexibility and consistency of the proposed method for a typical mooring system in both
reliability (calculating the annual probability of failure) and integrity (calculating the Integrity Index
and the optimality factors) analysis.

4. Conclusions

This paper presents a simplified method for the reliability- and integrity-based design of engi-
neering systems with emphasis on mooring systems. The method is based on the assumption that the
system component capacities follow a lognormal distribution and the internal forces in the compo-
nents can be predicted using a linear regression model as a function of the logarithm of their return
periods. It is assumed that the conditional probabilities of failure of the system components given the
environmental load are independent from each other. Then, using the rules of probability, a simplified
approach is proposed for quantitative reliability and integrity analysis of the mooring system. First, a
method is proposed to develop a regression model and compute the annual probability of failure of the
mooring system or its components. Then closed form solutions are derived to estimate the optimality
factors of the mooring components to achieve the target reliability and maximum system integrity. A
method is also discussed to design the system failure scenario and meet a target annual probability of
system failure at the same time. These methods identify the over- and under-designed components
and provide a basis for changing their design to achieve a target level of safety and integrity. The
method is then applied to a mooring system that was designed based on API RP 2SK guidelines for
station-keeping systems. The annual probability of failure of this system was computed, and compared
with the DNV OS E301 recommendation for a reliability-based design. Results show that the mooring
system will not meet the 1004 recommended annual probability of failure unless most of its com-
ponents have higher average strengths. The optimality factors (ratio of optimal to current average
capacities) of all the components to meet the DNV safety requirement were computed for two design
strategy of maximizing the system integrity and designing the system failure scenario so that with 0.99
probability, the anchor chains are the first components that will fail if a failure occurs, while main-
taining the 1004 annual probability of system failure.

Acknowledgements

We greatly appreciate InterMoor Inc. and particularly Mr. Kent Longridge for supporting this study
and providing detailed information about the conventional design and analysis of a typical mooring
system.

References

[1] Maslin E. Mooring line failures attract industry attention. In: Palmer Brion, editor. Offshore engineer. Pontiac, Illilnois, USA:
RR Donnelley & Sons; July 2013. pp. 46–7.
[2] API. Design and analysis of station keeping systems for floating structures. Recommended Practice 2SK. 3rd ed. Wash-
ington, D.C.: American Petroleum Institute; 2008.
[3] DNV. Position mooring (DNV-OS-E301). Det Norske Veritas; 2010. p. 57.
[4] Mousavi ME, Gardoni P. Integrity Index and Integrity-based Optimal Design of structural systems. Eng Struct 2014;60:206–
13.
[5] Choi YJ. Reliability assessment of foundations for offshore mooring systems under extreme environments. Austin, Texas:
The University of Texas at Austin; 2007.
[7] Mousavi ME, Gardoni P, Madouliat M. Progressive reliability method and its application to offshore mooring systems. Eng
Struct 2013;56:2131–8.
[8] Ditlevsen O, Madsen HO. Structural reliability methods. New York: Wiley; 1996.
[9] Ang AH, Tang WH. Probability concepts in engineering: emphasis on applications in civil & environmental engineering. 2,
illustrated ed. Denvers, MA: Wiley; 2007.
104 M.E. Mousavi, P. Gardoni / Marine Structures 36 (2014) 88–104

[10] Choi YJ, Gilbert R, Ding Y, Zhang J. Reliability of mooring systems for floating production systems. Final Project Report
Prepared for Minerals Management Service. MMS Project Number 423 and OTRC Industry Consortium; 2006.
[11] Gardoni P, Nemati KM, Noguchi T. Bayesian statistical framework to construct probabilistic models for the elastic modulus
of concrete. Mater Civ Eng 2007;19:898–905.
[12] Gardoni P, Der Kiureghian A, Mosalam KM. Probabilistic capacity models and fragility estimates for reinforced concrete
columns based on experimental observations. Eng Mech 2002;128:1024–38.
[13] Gardoni P, Mosalam KM, Der Kiureghian A. Probabilistic seismic demand models and fragility estimates for RC bridges.
J Earthq Eng 2003;7:79–106.
[14] Box GEP, Tiao GC. Bayesian inference in statistical analysis: mass; 1992.
[15] InterMoor. MODU mooring analysis report. Houston, TX: InterMoor Incorporated; 2001.
[16] Skjong R, Gregersen EB, Cramer E, Croker A, Hagen Ø, Korneliussen G, et al. Guideline for offshore structural reliability
analysis – general. DNV: 95-2018. DNV; 1995.
[17] Vazquez-Hernandez AO, Ellwanger GB, Sagrilo LVS. Reliability-based comparative study for mooring lines design criteria.
Appl Ocean Res 2006;28(6):398–406.