You are on page 1of 8

Applied Ocean Research 30 (2008) 5461

Contents lists available at ScienceDirect

Applied Ocean Research


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

Reliability-based design criterion for TLP tendons


F. Barranco-Cicilia a, , E.C.P. Lima b , L.V.S. Sagrilo b
a

Instituto Mexicano del Petroleo, Deep Waters Exploitation Department, Eje Central Lzaro Crdenas 152, Gustavo A Madero 07730, D.F., Mexico

COPPE-Federal University of Rio de Janeiro, Civil Engineering Department, Cidade Universitria, Centro de Tecnologia, Bloco I-2000, Sala I-116, Ilha de Fundo 21945-970,
Rio de Janeiro, Brazil

article

info

Article history:
Received 16 June 2007
Received in revised form
23 February 2008
Accepted 12 May 2008
Available online 1 July 2008
Keywords:
TLP
Tendons
Hurricanes
Reliability-based design
Floating systems

a b s t r a c t
This paper presents a methodology to perform a Load and Resistance Factor Design (LRFD) criterion
for the design of Tension Leg Platform (TLP) tendons in their intact condition. The proposed design
criterion considers the Ultimate Limit State (ULS) for the tendon sections, expressed in terms of the
expected value of the extreme Interaction Ratio (IR), considering long-term storm sea states, and takes
into account the dynamic load effects interaction and the statistics of its associated extreme response.
The partial safety factors are calibrated through a long-term reliability-based methodology for the storm
environmental conditions in deep waters of the Campeche Bay, Mexico. Different target reliability values
are considered in order to evaluate the effect of this key parameter on safety factors. The results show that
the partial safety factors reflect both the uncertainty content and the importance of the random variables
in structural reliability analysis. When tendons are designed according to the developed LRFD criterion,
a less scattered variation of reliability indexes is obtained for different tendon sections across a single or
a variety of TLP designs. Finally, it is observed that the target reliability value has a strong influence over
the safety factor values and thus over the final size of the structural elements.
2008 Elsevier Ltd. All rights reserved.

1. Introduction
TLPs are floating structures anchored to the sea bottom through
vertical tensioned tendons. The taut tendons hold the platform in
place and they remain tensioned due to the excess of buoyancy
caused by the floating hull. The tendon system is a critical
component for the TLPs, since its failure may lead to the collapse
of the whole structure involving human lives, economic losses
and damages to the environment. Thus, the tendon system has
to be designed to withstand the possible occurrence of different
limit states such as maximum and minimum tension, hydrostatic
collapse and fatigue for intact and damage conditions [1].
On the other hand, the structural design commonly follows
technical rules, practice recommendations or codes that have
been developed based on Bayesian evolution taking into account
successful projects as well as structural failures [2]. However, these
standard practices present some difficulties to be directly applied
to the novel structural concepts, with few design projects available
and little operational experience. In these cases, procedures based
on structural reliability become powerful design tools, because
they are able to take into account random uncertainties associated
with load and resistance variables and to consider the safety

Corresponding author. Tel.: +52 55 91757403,


E-mail address: fbarran@imp.mx (F. Barranco-Cicilia).

0141-1187/$ see front matter 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.apor.2008.05.001

margins required by the structures owners and governmental


agencies.
Currently, there are two well known methodologies for
structural design: the Working Stress Design (WSD) and the LRFD
criterion. In the WSD method, the sum of maximum stresses
from static and dynamic external loads is limited to the nominal
resistance of the structural element divided by a global safety
factor. However, this approach has some obvious drawbacks, for
instance, despite the fact that environmental loading is usually
more uncertain than the permanent loads, both types of loading are
covered by the same safety factor. In addition, the LRFD criterion
produces structural elements with more uniform reliability levels
in comparison with the design obtained following the WSD, since
LRDF considers a set of partial safety factors instead of an overall
safety factor. These partial safety factors are a function of the
variables randomness levels and they are calibrated through
reliability analyses in order to reach specific safety targets [3].
In this work, an LRFD format is adopted to develop a ULS design
criterion for a TLP tendon under intact condition of its mooring
system. Due to the common marine structural design practice
of using a sea state with 100 year return period, the proposed
design check equation corresponds to the expected value of the
100 year tendon interaction ratio extreme response. The tendon
extreme response is generated using long-term storm sea states.
The partial safety factors are calibrated by means of a reliabilitybased methodology using three different TLP models under the

F. Barranco-Cicilia et al. / Applied Ocean Research 30 (2008) 5461

action of hurricanes and winter storms observed during the past


century in the deep waters of the Campeche Bay offshore Mexico.
2. Calibration of the design criterion
Nowadays a well supported reliability-based methodology for
code calibration [3] is available. The first calibration procedures
and applications were developed by Allen [4], CIRIA [5], Ravindra
and Galambos [6], and Ellingwood [7]. In the case of TLPs, the
DNV-OS-C105 [8] is one of the first offshore standards that have
established a LRFD criterion where partial safety factors have been
calibrated based on reliability methods.
In this study, the calibration process consists of defining
the scope, selecting and sizing the calibration points along TLP
tendons, establishing the target reliability value, applying the
reliability analysis to the calibration points, defining the design
criterion format and finally, calibrating numerically the partial
safety factors. This calibration process is schematically shown
in Fig. 1.
2.1. Scope of the design criterion
The proposed design criterion can be applied to any section
of the TLP steel tendons considering the whole mooring system
in the intact condition. The design check equation represents
the ULS for a tendon section and is expressed in terms of the
expected value of its long-term extreme Interaction Ratio (IR).
The IR is a non-dimensional ratio between the load effects and
the tendon section strength to be defined in Section 3. The shortterm (that corresponding to 3 h of duration) IR takes into account
the interaction between the external hydrostatic pressure and
the dynamic tension force and bending moment. These dynamic
loading effects are evaluated in this work by means of TLP fully
coupled (hull/tendons/risers) structural analyses [9] considering
permanent, variable and environmental loads. Wind, current, tide
variation, and first and second order wave load effects (springing
and slow drift) are the environmental loads considered in the
analysis. However, it should be pointed out that the ringing effect
was not included in this study due to the difficulty of properly
quantifying these force components at present [10]. Finally, the
partial safety factors are calibrated for extreme tendon response
taking as reference a 100-yr return period.
2.2. Calibration test cases
A group of realistic designs of TLPs has to be employed to
perform the reliability-based new design criterion calibration. Due
to the fact that the proposed design check equation is related to
the ULS condition, only the most loaded tendon for each TLP is
considered in the calibration process. The most loaded tendon is
identified as the one having the largest failure probability.
Generally, the largest axial tension for a tendon occurs at the
top of the tendon and conversely, the highest value of hydrostatic
pressure occurs at the bottom of the tendon. On the other hand,
the largest bending moment is usually located somewhere in the
middle of the tendon length. Therefore, the tendon sections should
be reviewed at different locations. In this work, tendon sections at
the top, in the middle and at the bottom of the most loaded tendon
are selected for calibration purposes. An initial sizing of the TLP
tendons has been carried out through design recommendations
which are found in API RP-2T [1].
2.3. Target reliability
The target reliability, which can be expressed through either
reliability index or failure probability [3], is a measure of the

Fig. 1. Main steps of the calibration process.

55

56

F. Barranco-Cicilia et al. / Applied Ocean Research 30 (2008) 5461

structural safety level. In order to obtain a suitable target reliability


level, besides technical requirements, it is also necessary to
take into account economical, social, political, and environmental
aspects by considering the consequences of the structure failure
such as: human lives, direct and indirect economic losses and
possible damage to the environment. A risk-based methodology to
obtain target reliabilities for novel offshore structures is presented
in [11]. Several methods can be applied in order to establish target
reliabilities, presented as follows:
(a) Identify the implicit or explicit reliability levels in design codes,
considered as safe enough by the technical community;
(b) Develop a risk study to obtain an optimum reliability value
based on a cost-benefit criterion; and
(c) Evaluate the reliability of failure of many structures with a
successful service life history.
It is out of the scope of this work to review or look further into
any of these methodologies. Instead, three different values for the
annual target reliability levels (104 , 105 , and 106 ), have been
used in order to evaluate the effect of this key parameter on the
safety factors and, of course, in the final dimensions of the TLP
tendons.
3. Structural reliability analysis
Reliability analyses are used here for re-designing the tendons
in order to achieve the target probability of failure for all the
sections selected as calibration test cases. In this phase of the
calibrating criterion process a limit state function and probabilistic
characterization of the basic random variables involved in problem
must be defined.
In this work, the reliability analysis for the sections of a TLP
tendon was developed using the methodology, limit state function
and probabilistic description of random variables described in
Ref. [12]. The quoted methodology is based on tendon extreme
response statistics generated by applying several short-term sea
states (storms) which have been observed in a long time period.
First, for each short-term sea state a random nonlinear dynamic
analysis for the TLPs model is carried out and the obtained
results consist of a time-series of the tension forces and bending
moments for the tendon sections. The dynamic interaction ratios
(IRs), defined below, are calculated using those internal forces
time series, the hydrostatic external pressure and the geometric
and mechanical properties of the tendons. It should be noted that
the purpose of the IR is for it to be used in the tendon design.
The IR is defined in order to prevent the possibility of the TLP
tendons going into the plastic range, when the TLP experience
extreme responses. Therefore, only geometrical nonlinearities are
taken into account in the short-term sea state nonlinear dynamic
random analysis of the TLPs tendons. Using the IR time-series,
the cumulative probability functions (CPFs) are fitted to the shortterm maxima and extreme IR values. The contributions from
all short-term extreme CPFs are integrated in order to obtain
an unconditional long-term distribution of the tendon extreme
response. This latter distribution is conditional upon the vector
E which contains the time independent random variables used
X
to include the randomness in load effects, structural strengths
of tendons, and numerical models in the structural reliability
E is defined as:
analysis. Thus, the vector X

E = XCA , XCE , XIRT , XIRB , XIRp , XFy


X


(1)

where XCA is the model uncertainty random variable associated


with environmental load effects, XCE is the model uncertainty
random variable associated with static load effects, X IRT is the
random variable associated with the uncertainty in the axial
strength given by the IR numerical model, XIRB is the random

variable associated with the uncertainty in the bending strength


given by the IR numerical model, XIRp is the random variable
associated with the uncertainty in the hydrostatic collapse
strength given by the IR numerical model, and XFy is the random
variable associated with the uncertainty in the material yield
strength (Fy).
The failure probability of a tendon section is evaluated by means
of using the integrated CPF of the tendon extreme interaction ratio
(IR) and,

Z
pf =
E
allX

pf (E
x)fXE (E
x)dE
x

where pf (E
x) = P
F TA

E
IRext |X

(2)



E x ,
= 1 F TA
E 1| X = E
IRext |X


E = Ex for the extreme
|
ir
X
|XE

E = Ex
IRext 1| X

E = Ex is the CPF F TA
1| X

IRext

IR in a reference period TA used in reliability analysis (usually 1


E = Ex and IR = 1 (by the definition, see
year) evaluated for X
E ) is
below, the limit state is violated when IR > 1), and fXE (X

E
the joint distribution of the time independent random variables X
described in Eq. (1). The mentioned CPF is obtained considering a
ULS equation for the tendon section and the long-term storm sea
states distribution as described in what follows.
The ULS equation employed in this work is based on a section
interaction ratio (IR) presented in Ref. [13]. The ULS is reached
whenever IR > 1. For a generic short-term sea state the interaction
ratio IR in time-domain is given by:


E
IR t , YE X


=

fT (t )
FT

fB (t )

FB

p
pC

(3)

with
fT (t ) = XCE fT,CE + XCA fT,CA (t )
fB (t ) = XCE fB,CE + XCA fB,CA (t )
FT = XIRT XFy
FB = XIRB XFy [4 (1 + t0 /D) / 0.006 D/t0 ]

= 2 + 0.023 D/t0
p = am hv s


q
1
1
2
poel
poel 4 po pel
pC = XIRp
2

(4)

poel = po + pel (1 + 3 0 D/t0 )


pel = 2

E (t0 /D)3
1 2

po = 2 XFy t0 /D

where
fT,CE
is the tension stress due to static loads,
fT,CA (t ) is the dynamic tension stress due to environmental
forces,
fB,CE
is the bending stress due to static loads,
fB,CA (t ) is the dynamic bending stress due to environmental
forces,
FT
is the tendon axial strength,
FB
is the tendon bending strength,
p
is the acting external pressure,
pC
is the collapse pressure,
pel
is the linear buckling pressure,
po
is the yield pressure,
D
is the tendon pipe external diameter,
t0
is the tendon pipe thickness,
0
is the ovality parameter of the pipe,
E
is the Youngs modulus of the material,

is the Poissons coefficient of the material.

F. Barranco-Cicilia et al. / Applied Ocean Research 30 (2008) 5461

In Eq. (3) YE is a vector with short-term environmental


parameters associated with wave, wind and current, such as
significant wave height, mean wave upcrossing period, wind
velocity and so on [14]. These parameters are represented by
their joint probability model, i.e., fYE (E
y). According to Ref. [10], the
geometrical properties, Youngs modulus and Poissons coefficient
for the material have been considered in the IR model as
deterministic variables due to the high quality tendon fabrication
process. Based on the simulated time-history of IR for a specific
sea state and a set of known values for the time independent
E and XE = Ex, a short-term extreme
parameters, i.e., YE = y
E can be
x, y
probability distribution for IR, i.e., F st E E ir| E
IRext |X ,Y

obtained by means of a probability distribution F st

E ,YE
IRmax |X

E
x, y
ir| E

fitted to the maxima sample using Weibull-tail or Hermite model


approaches [14]. For a reference period TA the IR long-term
E = Ex is given by [15]
extreme value distribution conditional on X
F TA
E
IRext |X

i

E = Ex = exp F TA 1 F st E ir| Ex
ir| X
IRExt |X

(5)

where F is the mean rate of occurrence of tropical storms and


hurricanes in the plataform location and
F st
E
IRExt |X


x =
ir| E

E fYE (Ey)dyE.
F st
x, y
E ,YE ir| E
IRExt |X


E
all y

(6)

The tendon section failure probability defined by Eq. (2) can


then be evaluated by means of the First Order Reliability Method
(FORM) [13]. A summary of the main steps of the reliability analysis
is shown in Fig. 2
4. Format of the design criterion
The essence of the new design criterion is expressed through
the design check equation to verify any tendon section. The
proposed criterion establishes that any tendon section along its
whole length has to attain the following condition [14]:

(TA)

E IRext

i
h

1.0
i

(7)

(TA)

where E IRext

is the expected value of the extreme tendon IR

considering a long-term period TA which here, in order to be in


accordance with most design standards, is arbitrarily set equal to
100 years. The expected value in Eq. (7) is given by
E IRTA
ext =

(TA)

ir fIRext (ir) dir

(8)

(TA)

where fIRext (ir) is the probability density function (PDF) of the IR


extreme value for a long-period TA. This distribution is specifically
defined in the design criterion as follows:
(TA)

(TA)

fIRext (ir) = F TA FIRext (ir) fIRstext (ir)

(9)

where
fIRstext (ir) =

st
dFIR
(ir)
ext

dir

(TA)

st
FIRext (ir) = exp F TA 1.0 FIR
(ir)
ext
st
FIR
(ir) =
ext



Z
E
all y

st
FIR
E =Ey
ext Y

(10)


E f E (Ey)dyE.
ir y
Y

These equations correspond to those presented in the previous


section but now considering a set of pre-defined values for
E and a set of safety factors E in the ULS equation
variables X

Fig. 2. Methodology for reliability analysis of TLP tendons.

57

58

F. Barranco-Cicilia et al. / Applied Ocean Research 30 (2008) 5461

as described in what follows. In the design criterion each shortterm realization of the tendon interaction ratio IR(t , YE ) has to be
evaluated with the following expression:

Table 1
Characteristics of the TLP hull



fB (t )
p
fT (t )

E
+
+ (p )
IR(t , Y ) = (F )

Columns:

FT

FB

Parameter

(11)

pC

Value
Number
Diameter
High
Distance between axes
Number
Transversal section

where

Pontoons:

fT (t ) = CE fT,CE + CA fT,CA (t )

Draft
Displacement
Vertical position of C. G.
Gyration ratio in pitch and roll
Gyration ratio in yaw
Area exposed to wind
Vertical position

fB (t ) = CE fB,CE + CA fB,CA (t )

(12)

where CE ,CA ,F and p are the partial safety factors associated


with stress due to static loading effects, stress due to environmental loading effects, interaction ratio of tension and bending stresses,
and the ratio of external hydrostatic pressure between tendon collapse strength, respectively. The other variables keep the same definitions as in Eq. (3). In this way, the vector of partial safety factors
is defined as:

T

E = CE , CA , F , p .

(13)

Static stresses at a tendon section are calculated considering


dead weight, buoyancy, nominal pretension and quasi-static
effects, for instance, tide variations, foundation mispositioning,
platform centre of gravity mispositioning, and hydrostatic pressure
at the upper end of the tendon [16]. Dynamic stresses, such as the
axial force and bending moment, caused by environmental loading
are evaluated using time-domain random nonlinear (geometric)
dynamic analysis results minus their corresponding initial static
values. The geometrically nonlinear random dynamic analysis is
performed considering: current, mean hourly wind speed and
its associated random fluctuation (spectrum) and irregular first
(wave frequency) and second order (slow drift and springing)
wave forces. Nominal values for axial, bending and hydrostatic
collapse strengths for tendon sections are computed in Eq. (11)
using the minimum value of the yield strength (Fy) of the material
as informed by the manufacturer.
The safety factors E must be calibrated in order to obtain a
failure probability of the tendon section against the ultimate state
as close as possible to a target value when the design equation (see
Eq. (7)) is employed. It must be noticed that this criterion belongs
to class of response-based design criteria.
4.1. Numerical method for calibrating partial safety factors
The set of partial safety factors (E ) that best approximates the
reliability of a set of calibration test cases to a target value can be
obtained by minimizing the following mean square error function:
ErCal (E ) =

ntlp X
ns 
X
i=1 j=1

(TA)

1.0 E IRext

i 2
ij

wij

(14)

where ntlp is the number of TLP models considered in the


calibration process, ns is the number of sections of the most loaded
tendon of each TLP model, and wij is a weighting factor. The sum

Pntlp Pns

of the weighting factors has to satisfy the condition i=1


j=1 wij
= 1.0.
In this study, the optimization problem in Eq. (14) was
solved through a simple version of the Evolutionary Strategies
(ES) algorithm [17]. The ES algorithm is based on a population
E ),
consisting of a unique individual (the safety factors vector
E during its life. The mutation
submitted to a mutation operator
operator components are represented by normal independent
variables having zero mean and standard deviation . The
optimization procedure consists of the following steps:

4
16.20 m
43.52 m
51.45 m
4
8.4 m 7.35 m
27.3 m
32070 ton
5.95 m
30.44 m
31.74 m
3000 m2
52.5 m

1. The initial individual of the population Ej , j = 0 is


established. This initial searching point can be made equal to E0 =
{1.0, 1.0, 1.0, 1.0}T ;

2. The objective function in Eq. (14), ErCal Ej , is evaluated for
the individual Ej .
E j is generated by applying the
3. The offspring of the individual
follow mutation operator:

h i n
o
Ej+1 = Ej + NE (0, )

(15)

E (0, ) is the mutation vector of statistically independent


where N
random numbers generated from a normal distribution with zero

mean and standard deviation , and is a diagonal matrix of


importance coefficients for the safety factors. The coefficients
ii can be established from the importance factors for random
variables (FORM results, see [3]) associated with each partial safety
factor. Due to the fact that many combinations of safety factors

might be a solution for the optimization problem, the matrix is


introduced to guarantee convergence of the ES algorithm to values
that express the randomness level of their
 related variables.
4. The objective function ErCal Ej+1 is evaluated for the new
member Ej+1 of the population. This offspring member will only
be accepted in replacement of its progenitor, if it satisfies all the
problem constraints and
 produces a better result for the objective
function (ErCal Ej+1 < ErCal Ej ). If these conditions are not
satisfied the offspring is eliminated and its progenitor continues
as member of the population.
5. The ES algorithm continues through steps 3 and 4 until a given
tolerance margin is achieved.
5. Numerical applications
In this section, as an application example of the proposed
methodology for the TLP tendons ULS design criterion, the partial
safety factors are calibrated considering the Campeche Bay storm
conditions [12,14]. Three platforms, identified as TLP-01, TLP-02,
and TLP-03, are considered in the calibration process. All three
TLPs have the same deck, floating hull and riser system. The
platforms differ only by the water depth and tendon system. The
main characteristics of the floating hull are presented in Fig. 3 and
Table 1.
TLP-01 has a mooring system composed of 12 tendons (3
tendons per corner) and it is located in a site with 1000 m water
depth, TLP-02 has 8 tendons (at positions 1, 3, 4, 6, 7, 9, 10 and
12 in Fig. 3) and it is located in 1000 m water depth while TLP03 has 12 tendons and it is in a site with 500 m water depth.
Characteristics of the tendons are listed in Table 2. All the tendon
systems have been initially sized using the procedure presented
in API-RP-2T [1] and considering the requirements for maximum
stress at net section, hydrostatic collapse, and minimum effective

F. Barranco-Cicilia et al. / Applied Ocean Research 30 (2008) 5461

59

Table 2
Characteristics of tendons for the three TLPs
Parameter

TLP-01

TLP-02

TLP-03

Number of moorings
Tendon length (m)
External diameter (m)
Internal diameter (m)
Dry weight (kN/m)
Axial stiffness (EA) (kN)
Bending stiffness (EI) (kN m2 )
Pretension (kN)

12
973.9
0.6604
0.5906
5.30
14 249 407
699 002
8826

8
973.9
0.8128
0.7239
8.29
22 278 261
1649 531
13 239

12
473.9
0.6604
0.60325
4.39
11 780 000
588 900
7500

Table 3
Probability description of the random variables [11]
Random variable

XCE
XCA
XIRT
XIRB
XIRp
XFy

Fig. 3. Floating hull of TLP case study.

tension. The structural design of the tendons has been performed


by searching for tube sizes with API interaction ratios close to 1.0.
It is important to point out that the API design criteria are based
on 100-yr extreme environmental conditions instead of the 100-yr
extreme response approach employed in this work.
Campeche Bay is located in the Gulf of Mexico, and this marine
region is often subjected to storms that generate extreme sea
states. These sea states have to be considered in the design of
offshore platforms. A summary of short-term metocean data used
in this work, based on the main hurricanes and winter storms
which occurred in the past century in deep waters of Campeche

E for
Bay, are found in Ref. [12,14]. Since the joint PDF fYE y
storm sea state parameters is not available for local conditions
at the Campeche Bay, in this study the evaluation of the longterm extreme response in Eq. (6) has been approximated by the
following equation:
FIRext |XE ir| E
x
=

1
NF

NF
X
i=1

st ,i

Ei
ir E
x, y

E , YEi
IRext X

(16)

where NF is the number of storms included in the analysis (in this


case study NF is equal to 31 [12,14]). A similar approach has been
employed to compute Eq. (10).
Details about the geometrically non-linear fully-coupled random dynamic structural analysis of the platform, tendons and risers system, under each short term sea state, and the fitting process

PDF type

Parameters

Normal
Normal
Normal
Normal
Normal
Extreme
minimum
value

CoV (%)

1.00
1.00
1.20
1.00
0.95
Parent PDF
(Lognormal)
482.5 MPa

5
25
8
7
13

of extreme value CPFs are presented in [14]. The probabilistic deE is given in Table 3. All reliability
scription of the random variables X
calculations have been made with the FORM technique [3]. Moreover, reliability analysis has been used to identify the most loaded
tendon for each TLP and to size the calibration points in order to
satisfy the target failure probability. Sections at the top, middle,
and base of each tendon have been analyzed. Tendon 4 has been
identified as the most loaded one for the three TLPs. One explanation for tendon 4 overloading is that the storm sea states used
correspond to the dominant incidence direction at the Campeche
Bay [12,14] as depicted in Fig. 3.
A total of 9 calibration points (test cases) have been considered
in the calibration process: sections at the base, middle, and top
of the most loaded tendon of every TLP (ntlp = 3, ns = 3).
The annual probability of failure and their respective reliability
indexes of the 9 calibration points are presented in Table 4. For each
tendon considered, the largest failure probability of the sections is
located at the base. Table 4 also shows both external diameter (D)
and thickness (t0 ) of the calibration points which satisfy the target
annual failure probabilities Pftrg = 1 104 , 1 105 , 1 106 .
It is important to mention that only the tendon thicknesses have
been modified, assuming that small variations in tube thickness do
not modify the tendon internal forces obtained from the dynamic
analyses.
The nine re-sized tendon sections satisfying each target
reliability value have then been used to calibrate the partial safety

factors E = CE , CA , F , p in Eq. (11). The minimum value


specified for the material (415 MPa) has been considered as the
characteristic value of the yield strength. In the calibration process,
a reference period TA (see Eq. (4)) equal to 100 years has been
used. This time period is a common reference value employed in
the engineering practice for designing marine structures. Also, an
initial vector E = {1.0, 1.0, 1.0, 1.0}T , a standard deviation =
0.01, and an importance coefficients matrix with the elements
11 = 0.35, 22 = 1.0, 33 = 0.75, and 44 = 0.35 have been
considered for minimizing the error function in Eq. (14) through

the ES algorithm. The elements of were established taking into


account the importance factors of the random variables, which are
reported in [13], starting with a unit value for the most significant
variable in reliability analysis. In order to satisfy the convergence
criterion of the ES algorithm, a tolerance margin equal to 1 103

60

F. Barranco-Cicilia et al. / Applied Ocean Research 30 (2008) 5461

Table 4
Tendon sections satisfying reliability targets
Calibration point

TLP-01
Base
Middle
Top
TLP-02
Base
Middle
Top
TLP-03
Base
Middle
Top

D (m)

Initial sections

Re-sized sections t0 (m)

t0 (m)

Pf

Pf = 1 104

Pf = 1 105

Pf = 1 106

0.6604
0.6604
0.6604

0.0349
0.0349
0.0349

1.50537 104
7.84507 105
2.91542 105

3.614
3.780
4.020

0.03566
0.03451
0.03280

0.03946
0.03849
0.03690

0.043352
0.042473
0.041076

0.8128
0.8128
0.8128

0.0445
0.0445
0.0445

8.30646 105
6.61093 105
2.12800 105

3.766
3.822
4.093

0.04405
0.04342
0.04091

0.04882
0.04842
0.04610

0.053654
0.053468
0.051286

0.6604
0.6604
0.6604

0.0286
0.0286
0.0286

3.25639 105
2.11482 105
1.47631 105

3.993
4.095
4.177

0.02716
0.02644
0.02584

0.03009
0.02953
0.02911

0.033188
0.032697
0.032474

Fig. 5. Reliability indexes of tendon sections before and after the calibrating
process for Pftrg = 1 104 .

Fig. 4. Convergence of the objective function ErCal (E


) to the tolerance margin
(1 103 ) through the ES algorithm, for the case of Pftrg = 1 104 .

has been established. In addition, the same weighting factor wij


equal to 1/9 was considered for all calibration points.
Fig. 4 shows how the objective function ErCal (E ) converges to
the tolerance margin through the ES algorithm. The case where
Pftrg = 1 104 is considered. The ES algorithm needed 78 iterations to solve the optimization problem. In this case, 39 of the
78 generated vectors of partial safety factors satisfied the restrictions of the optimization problem and
 gave a better
 result for
the objective function, i.e., ErCal Ej+1 < ErCal Ej . For iteration number 1 ErCal (E ) = 0.10855, and for iteration number 78
ErCal (E ) = 0.00098 reaching the tolerance margin to convergence (1 103 ). The vector of partial safety factors obtained is

T
CE = 1.003, CA = 1.223, F = 1.081, p = 1.043 . In
 h
i 
(TA)
the same figure, the variation of the mean value E IRext
E =

ij

of the long-term extreme IR expected values for the nine tendons section is shown. For the
 initial values
 of the safety factors

h
i
TA)
E = {1.0, 1.0, 1.0, 1.0}T E IR(ext

is equal to 0.673, and

ij

when ES converges this function is equal to 1.002. It worth mentioning that the values of the expected extreme IR can be equal
to, above or below 1.0, depending on the set of safety partial factors generated by the random ES algorithm. However, the selected
safety factors vector must satisfy the constrains of the optimization
problem and attain the tolerance margin in the mean square error
function ErCal (E ).
The partial safety factors obtained for each target failure
probability considered in this study are shown in Table 5. It can
be observed that the safety factor associated with stresses due
to dynamic actions presents the greatest value, and conversely,
the factor associated with static effects shows the smallest
value. Partial safety factors obtained from the calibration process
reflect the uncertainty level and the importance of the random

Table 5
Partial safety factors for different target failure probabilities (TA = 100 years)
Pftrg

CE

CA

1 104
1 105
1 106

1.003
1.015
1.013

1.223
1.256
1.302

1.081
1.175
1.203

1.043
1.037
1.035

Table 6
Dispersion of the calibration points reliability indexes for Pftrg = 1 104
Statistic parameters

Mean
Standard deviation
CoV (%)

Reliability index
Before calibration

After calibration

3.929
0.190
4.827

3.748
0.050
1.336

variables in the reliability analysis. The calibration results suggest


that stresses associated with static loads can be considered as
deterministic variables. The variation of the safety factors through
the target reliability values can also be observed in Table 5. As seen,
the values of CE and p present a minimum variation when the
target failure probability is changed. On the other hand, the factors
CA and F significantly increase when the target reliability has a
smaller value. These results show that the safety factors related
to the random variables with high levels of uncertainty present
significant changes for different target failure probabilities.
Using the safety factors presented in Table 5 for the case where
Pftrg = 1 104 , each calibration point was re-sized in order
to attain the proposed design criterion given by Eq. (4). Then, the
failure probability and associated reliability index for each tendon
section was re-evaluated. In Fig. 5, the new reliability indexes
are compared with those obtained before the calibration process
using the initial designs based on API-RP-2T. Table 6 shows that
the tendon sections before the calibrating process have reliability
indexes with a standard deviation four times greater than the resized ones. All these results lead us to conclude that the tendons
designed according to the proposed criterion have less reliability

F. Barranco-Cicilia et al. / Applied Ocean Research 30 (2008) 5461

dispersion around the target value. This can be expected because


the LRFD criterion was developed in order to take into account
the variation of the structural response characteristics from design
to design. This is achieved by a set of partial safety factors
applied independently to the static loads, the environmental loads,
the interaction ratio of tension and bending stresses, and the
ratio between external hydrostatic pressure and tendon collapse
strength. Therefore, the LRFD criterion produces very uniform
values for the calibrated reliability indexes of tendon sections for
the three different TLP designs. For this reason, the authors believe
that the proposed methodology to perform a LRFD criterion for the
design of TLP tendons seems to be suitable. Of course, a greater
number of TLP designs should be investigated in order to confirm
this assumption and the value of the safety factors.

61

to the developed LRFD criterion show a less scattered variation


of reliability indexes, this result is obtained for different tendon
sections across a single or a range of TLP designs. On the other hand,
the designs based on a long-term response approach involve a
larger computational cost in comparison with the standard criteria
based on extreme environmental conditions.
Finally, it worth mentioning that the figures presented in this
work cannot be considered as final safety factors. These factors
have to be adjusted by taking into account a greater number of
TLP models, representing variations in geometric and material
characteristics, hull size, etc. Furthermore, updated environmental
data and their joint probabilistic description should also be
considered.
References

6. Conclusions
In this paper an LRFD criterion for structural design of TLP
tendons considering the mooring system in the intact condition
has been presented. The design criterion considers the ULS
condition for tendon sections, taking into account both dynamic
interactions of load effects and the statistics of its associated
extreme response. The proposed design check equation consists
in a long-term response approach and establishes that the
expected value for the 100-yr extreme IR of tendon sections
shall be less than or equal to one. The partial safety factors
have been calibrated through a long-term reliability-based method
considering the storm environmental conditions in deep waters
from the Campeche Bay offshore Mexico. The randomness of loads,
resistance of tendons and analytic limit state models for calculation
of tendon strength has been included in the reliability analysis.
Also, an Evolutionary Strategy-based optimization algorithm has
been developed to find the set of partial safety factors. Different
target reliability values have been considered in order to evaluate
the effect of this key parameter on the partial safety factors.
The results showed that partial safety factors reflect both
uncertainty content and the importance of the random variables
in structural reliability analysis. The calibrated values suggest
that the stresses generated by static loads can be considered
as deterministic variables. Safety factors related to the random
variables with high levels of uncertainty present significant
changes for different target failure probabilities. When the
proposed design criterion is applied to TLP tendons using the
calibrated safety factors associated with different target failure
probabilities, it has been found that the tendon thickness increases
by 10% when Pftrg value changes from 1 10 4 to 1 105 , and by
20% when Pftrg value changes from 1 104 to 1 106 . Another
important aspect to highlight is that tendons designed according

[1] API. Recommended practice for planning, designing and constructing tension
leg platforms, API-RP-2T, 2nd ed. Washington (USA). 1997.
[2] Stahl B. Reliability engineering and risk analysis. In: The Text book design of
fixed offshore structures. USA: Van Nostrand Reinhold Co.; 1985 [Chapter 5].
[3] Melchers RE. Structural reliability analysis and prediction. 2nd ed. England:
Wiley; 2001.
[4] Allen DE. Limit states designA probabilistic study. Canadian Journal of Civil
Engineering 1975;2(1):3649.
[5] CIRIA. Rationalization of safety and serviceability factors in structural codes.
Report no. 63. London: Construction Industry Research and Information
Association; 1977.
[6] Ravindra MK, Galambos TV. Load and resistance factor design for steel. Journal
of Structural Division, ASCE 1978;104(ST9):133753.
[7] Ellingwood B, Galambos TV, MacGregor JC, Cornell CA. Development of a
probability based load criteria for American national standard A58. NBS Special
publication no. 577. US Department of Commerce; 1980.
[8] DNV, Structural design of TLPs (LRFD method). DNV-OS-C105. Norway; 2005.
[9] DNV, DeepC Program, Deep water coupled floater motion analysis, Hovik
(Norway); 2002.
[10] DNV. Guideline for offshore structural reliability analysisexamples for
tension leg platforms. Joint industry project report no. 95-3198. Revision 02.
Hovik (Norway); 1995.
[11] Bhattacharya B, Wang S, Basu R, Ma K, Menon B. Reliability-based combination
of environmental parameters for the design of novel floating structures.
In: Proceedings of the 18th international offshore mechanic and arctic
engineering (OMAE) symposium, vol. I. Saint Johns, Newfoundland (Canada):
ASME; 1999. p. 16.
[12] Cicilia FB, Lima ECP, Sagrilo LVS. Reliability of TLP tendons under storm sea
states. In: Proceedings of the 24th international offshore mechanic and arctic
engineering (OMAE), symposium, Halkidiki (Greece): ASME; 2005. paper
OMAE-67170.
[13] Moan T, Estefen SF, Saevik S, Zimmer RA. Limit states for the ultimate
strength of tubulars subjected to pressure, bending and tension loads. Marine
Structures 1994;7(25):32344.
[14] Cicilia FB. Reliability-based design criterion for a tlp tendon system. D.Sc.
thesis. Brazil: COPPE-UFRJ; 2004 [in Portuguese].
[15] Guenard YV. Application of system reliability analysis to offshore structures.
Report no 71. John A. Blume Earthquake Engineering Center, Stanford
University; 1984.
[16] Alves LHM. Dynamic analysis of tension leg platform tendons. M.Sc. thesis.
Brazil: COPPE-UFRJ; 1996 [in Portuguese].
[17] Lagaros ND, Papadrakakis M, Kokossalakis G. Structural optimization using
evolutionary algorithms. Computers and Structures 2002;80(7):57189.