Vous êtes sur la page 1sur 15

Reliability Analysis of FPSO Components and Calibration of Safety

Factors Using Response Surfaces


R. Montes-Iturrizaga (V), G. Inda-Sarmiento (V), A. Morandi (M), F. Silva-Gonzlez (V), E. Heredia-
Zavoni (V), D. Straub (V)

In this paper a procedure is presented for the reliability analysis and calibration of safety factors for ultimate limit states of
the hull and mooring system of an FPSO system. The formulation of the limit state equations is given first. The method of
response surface (RS) is used to relate extreme environmental variables to bending moment on the hull and to tensions on the
mooring lines. Second order polynomials are used for the RS, and its coefficients are estimated by regression using data from
structural analyses. For the purpose of assessing the ultimate capacity of mooring lines, these are modeled as series system
composed of many elements. The moorings capacity is then estimated as a function of the number of elements and their
individual tension resistances using extreme value theory. For the calibration of safety factors, a simplified method, which
decouples reliability analysis, is used. A hull model and three mooring system models at different water depths were used as
an application example for calibration of safety factors considering metocean conditions of the Mexican region of the Gulf of
Mexico.

KEYWORDS: Structural Reliability, Calibration, Safety Factors, Mooring Lines, Hull, FPSO.

INTRODUCTION
Most modern design codes are reliability-based and use the Load and Resistance Factor Design (LRFD) format. They are based on
design equations which compare resistances and loads (or load effects). Loads and resistances are subject to uncertainty and are
introduced in the design equations by design values. Design values for resistances are characteristic values divided by a partial safety
factor (typically greater than one), and design values for load effects are characteristic values multiplied by a partial safety factor
(typically greater than one).

The process of selecting target reliability values and determining the corresponding safety factors for a given limit state function is
known as code calibration (see e.g. [1]). The use of structural reliability methods to determine the partial safety factors guarantees
that the level of safety of any structure designed according to the codes is homogeneous and independent of the choice of material and
the operational and environmental conditions. Calibration of safety factors for mooring lines has been described in [2], [3] and [4]. For
FPSO hull girders reliability assessment and calibration of safety factors, see e.g. [5]. In this paper, we use the nested reliability
formulation introduced by Montes et al. [6] using RS and present results of the calibration carried out for environmental conditions of
the Mexican portion of the Gulf of Mexico. Safety factors are calibrated for the ultimate limit states describing tension failure in
mooring lines and bending moment on the hull under extreme meteorological and oceanographic conditions.

The paper starts out by introducing the corresponding limit state functions as a function of resistances and the environmental variables.
In the case of moorings, to estimate the mooring line capacity each line is modeled as composed of a large number of elements in a
series system; the system resistance is estimated as a function of the elements resistance by means of extreme value theory. The hull is
modeled considering it as a beam on an elastic foundation. The tension load acting on the mooring lines and the bending moment in the
hull are expressed as functions of the extreme environmental variables based on dynamic analyses in the frequency domain.

In order to decouple the reliability analyses from the dynamic analyses, a response surface (RS) approach is applied, which defines an
approximate parametric relation between the environmental variables and the response. The calculation of the probability of failure
involves the solution of a nested reliability problem [6] since the dynamic contribution of the extreme response is a stochastic process.
It is proposed to solve this problem by characterizing the maximum line tension or hull bending moment conditional on an extreme sea
state, and then solve the nested reliability problem.

Numerical results are presented for the environmental conditions of the Mexican region of the Gulf of Mexico. Partial safety factors are
determined considering design cases at three locations with different water depths, taking into account influence of input variables in
RS and reliability analyses.

Reliability Analysis of FPSO Components and Calibration of Safety Factors Using Response Surfaces
LIMIT STATE FUNCTIONS AND DESIGN EQUATIONS

Tension in Mooring Lines


The limit state function for ultimate tension failure in a mooring line is written as:

g (, r ) r t mean () Tdyn () 0 (1)

where r is the mooring line resistance, t mean is the quasi-static line tension, due to pretension and mean environmental loads
(including the effects of wind, current and mean wave drift), Tdyn is the maximum dynamic line tension of the random process during
an extreme sea-state due to wave induced motions in a time span t , and is the set of extreme environmental variables describing
the sea state. Note that for the random variable Tmean lower case is used in Eq.1 because this tension is known given the extreme
environmental variables . Tdyn includes 6-degree of freedom low-frequency and wave-frequency components of the mooring
line dynamic tensions, where the wave-frequency components are dominant as will be explained further in the paper. Response surfaces
(see below) are used to model uncertainties (standard deviation and mean average crossing rate) of the wave frequency dynamic tension
and the low frequency vessel motions. Tensions are assumed to be constant along the line. As noted in Eq.1, the maximum dynamic
tension is written in capital letters to emphasize that its value is uncertain for a given sea state .

The design equation corresponding to Eq.1, is:

rC
Tmeantmean,C Tdyn tdyn,C 0 (2)
R

where rC is the characteristic resistance of the mooring line, t mean,C is the characteristic value of quasi-static tension Tmean , t dyn,C is
the characteristic value of Tdyn , Tmean and Tdyn are the relevant partial safety factors. The characteristic values of the load are obtained
by calculating Tmean and the expected value of Tdyn for different environmental vectors 100 associated to 100 year return periods
(annual exceeding probability equal to 0.01). These vectors are taken from environmental contours that characterize the metocean
hazard at the selected locations. The characteristic values t mean,C and t dyn,C are then selected among all values calculated for the
different 100 as those that give the highest total response (the sum of Tmean and the expected value of Tdyn ).

Hull Bending
The limit state function for ultimate (yielding) bending failure in the hull can be expressed by:

g (, f y , s, xS , xW ) f y s ( xS mS xW MW ()) 0 (3)

where f y is the material yielding strength , s is the hull girder section modulus, mS is the maximum still water bending moment
associated to given static load condition, M W is the maximum wave-induced bending moment for the given static load condition.
Factors x S and xW represent model uncertainties on mS and mW respectively. Note that mS is deterministic given a static load
condition and that is the reason we use lower case in Eq. 3. As in the case of the maximum dynamic tension, M W is written in capital
letters because its value is uncertain for a given sea state . Although hull girder yield is discussed in this paper, buckling of hull
stiffened panel was addressed by separate component reliability calculations, which are not described here.

Maximum still water bending moment, M S , (here we use upper case because is a random variable in general) is usually modeled in the
literature by using a Rayleigh distribution for sagging and an exponential distribution for hogging, given a load condition. In order to
take into account frequent changes of load conditions M S is supposed to be a rectangular Poisson process, so that its probability
distribution is approximate by a Type I extreme variable distribution. On the other hand, maximum wave-induced bending moment,
M W , is considered to have a Rayleigh distribution for short-term and a Weibull distribution for long-term, see e.g. [5],[7], and [8].

Reliability Analysis of FPSO Components and Calibration of Safety Factors Using Response Surfaces
However, the approach we use in this paper is different. Maximum static bending M S is modeled in the case study as having only
three static load conditions (ballast, medium loaded and fully loaded) with their corresponding probability of occurrence during the
FPSO service lifetime. As to M W , we do not model explicitly its probability distribution, but we fit a Response Surface (RS) to relate
M W to the environmental variables associated to extreme sea states using results of frequency domain response analyses. As stated,
this implies the solution of a nested reliability problem. Once the probability of failure under an extreme event is obtained, the
corresponding annual probability is evaluated using the mean rate of occurrence of extreme sea states, as shown later.

Furthermore, it is important to mention that the RS method has already been used to establish simplified and explicit limit state
functions for time variant reliability assessment of FPSO hull girders, including deterioration mechanisms as fatigue and corrosion
[5,7,8].

The design equation corresponding to Eq.3, is:

f y ,C sC
( M S xS ,C mS ,C MW xW ,C mW ,C ) 0 (4)
R

where f y ,C is the characteristic material strength taken as the design nominal steel yield strength (a mean bias factor of 1.15 is used to
represent safe side bias in design yield strength), sC is the characteristic hull section modulus (mean value in this case), mS ,C is the
characteristic value of the maximum still water bending moment, and mW ,C is the characteristic value of the maximum wave-induced
bending moment.
The characteristic value of M S is the weighted average of the values associated to all static load conditions. In the case of M W the
characteristic value is calculated from environmental contours for a 100-year return period. As in the case of tension in mooring lines,
the characteristic value mW ,C is selected among all values calculated for the different 100 as that which gives the highest total
response E[M W ] .

ENVIRONMENTAL LOADING

Environmental Variables
Seven environmental variables are considered for describing sea-states: significant wave height H S (m), peak period TP (s), 10-m
hourly wind speed U10 (m/s), surface current speed U C (m/s), incident wave angle (relative to bow) wave (rad), relative angle
between wave and wind directions WW (rad), and relative angle between wave and current directions WC (rad). The three directional
parameters are taken as deterministic variables for the reliability analysis, and only the first four variables are modeled as random
variables; the random vector of environmental variables is thus [ H S , TP ,U10 ,U C ] .

Tension Loads

Maximum Dynamic Tension


It is generally considered that sea waves can be modeled as Gaussian processes. Provided the dynamic tension in mooring lines is
primarily governed by the first-order response (Sarkar and Eatock-Taylor [9]), it seems reasonable to assume that the dynamic tension
is also a Gaussian process. Furthermore, it has been shown by Choi et al. [10] that the mean values and standard deviations calculated
using time-domain response analysis correspond well with those calculated based on a Gaussian approximation. In this work, it is
assumed that conditional on a given extreme sea state described by [ H S , TP ,U10 ,U C ] , the dynamic tension is a Gaussian process,
Tdyn
so that the probability distribution of the maximum dynamic tension is given as [11]:

1.2 x
1 exp q
2
x 2 T ,
FTdyn ( x ) 1 exp exp t ,x0
2 T , x2
2
exp 2 1 (5)
2
T ,

Reliability Analysis of FPSO Components and Calibration of Safety Factors Using Response Surfaces
where the standard deviation and mean crossing rate of the dynamic tension (which include contributions from both low frequency and
wave frequency response), T , () , () , as well as the shape parameter of the dynamic tension power spectrum,
q q() , are all functions of the uncertain environmental variables describing the extreme sea states. The shape parameter q
is defined in terms of the first three moments of the dynamic tension power spectrum, i i=1,2,3, which also depend on ,

12
q 1 (6)
0 2

In Eq.6 i is the i -th moment of the dynamic tension power spectrum. It should be kept in mind that parameters in Eq.5 are dependent
on vector of environmental variables. The annual occurrence of extreme sea states and the parameters describing the environmental
variables in [ H S , TP ,U10 ,U C ] during an extreme sea state, are modeled probabilistically based on extreme value analysis of
metocean hindcast data, as described below. Parameters T , and in the distribution of Tdyn were obtained by means of linear
dynamic response analyses of mooring models and expressed as functions of by means of response functions. Here, a value
q 0.5 has been used after observing that the influence of this variable on the resulting reliability index is low for the entire domain
of possible q between 0 and 1. In order to simplify reliability calculations, response surfaces for parameters T , and were
developed as a function of .

The expected value and the standard deviation of Tdyn can be approximated by [12]:

E[Tdyn ] 2 ln( t / 2) T ,
T , (7)
T dyn ,

6 2 ln( t / 2)

Quasi-static Tension
The quasi-static tension Tmean is computed for different combinations of environmental variables and include the effects of wind,
current and mean wave drift. The results are used to establish response surfaces of Tmean , as described later. These are then employed
in the reliability analyses.

Wave-induced Bending Moment


Here we assumed that the wave-induced bending moment is Gaussian during an extreme event. The mean value MW () and standard
deviation MW () of M W were approximated as a function of using response surfaces.

DISTRIBUTION FUNCTION OF MOORING LINE RESISTANCE

The mooring line is composed of M segments (or material types) and each segment i has N i (i 1,2...M ) components (or elements)
(Fig.1). The distribution function of the resistance at each component in segment i is denoted by FRi (r ) . Since all of the components
in a given segment constitute a series system, the resistance of that segment is equal to the resistance of the weakest element. We can
use extreme value theory (for minima) in order to define the corresponding asymptotic ( N ) distribution function as follows [13,
14]:

Reliability Analysis of FPSO Components and Calibration of Safety Factors Using Response Surfaces
Segment I

Segment II

Component of segment II
Segment III

Figure 1: Definition of components and segments of a mooring line (three segments)


FRi (r ) 1 exp e i ( r ui ) (8)

where
u i is the most likely (mode) value of the distribution [14].

Finally, if we consider that the segments in a single mooring line also constitute a series system, the mooring line resistance will equal
the resistance of the weakest segment. Hence, the resistance for the mooring line has the following distribution function [15]:


M
FR (r ) 1 1 FRi (r ) (9)
i 1

In this paper we model the resistance for the whole mooring line by a Type I extreme distribution for minima. The parameters of this
distribution are selected so that its mean value and standard deviation are equal to the corresponding values of the function in Eq.9. The
characteristic value rC of the mooring line resistance is equal to the characteristic value of the Type I extreme distribution, rC u .

RESPONSE SURFACES
Dynamic analyses were carried out in order to compute parameters describing the mooring line tensions and hull bending moments as a
function of the sea states as described by environmental vector . On the basis of these results, response surfaces were developed to
reduce the number of the (computationally demanding) structural analyses and to decouple the reliability analyses from the structural
analysis. RS are parametric functions of the response quantities of interest as a function of the basic random variables, here , which
approximate the true relationship between these variables and response [16].

Second order polynomial expansions were used for the RS, as indicated in Eq.10. Choosing second order polynomials is common in
engineering modeling, often using the argument that higher order polynomial expansions can be difficult to control in the sense that
they may behave highly non-linearly and erratically.

T () c aT T b (10)

Here, T () is the response (quasi-static tension, maximum dynamic tension or wave-induced moment parameters), is the vector of
random variables (environmental) ( r x1) and c , a ( r x1) and b ( r x r ) are unknown coefficients to be determined. If we have
evaluated the response for n combinations of values of the environmental variables, then:


T (11)

where T is a vector ( n x1) of computed responses, is a matrix containing all linear, quadratic and cross combinations of the
components of , is a vector containing all the polynomial coefficients ( c , a and b ) and is an error vector, unbiased and zero
mean. As an example, in the case r 2 :

Reliability Analysis of FPSO Components and Calibration of Safety Factors Using Response Surfaces
[c, a1 , a 2 , b11, b22 , b12 ]T (12)

1 11 21 2
11 221 11 21

(13)
1 12n 22 n 1n 2 n
1n 2n

An estimator of vector can be obtained by regression analysis from:

(T ) 1 T T
(14)

Uncertainties in the estimator can be quantified [17]. However, it has been found that the influence of such uncertainty on the results
of reliability analyses is small.

RELIABILITY ANALYSIS
In this section a brief description is given of the procedure recently developed by Montes et al. [6] for solving the nested reliability
problem in mooring lines. The same procedure is extended here to hull bending (for further details see [6]).

Given a limit state function g (x) , where x is a vector of basic random variables, the probability of failure is calculated as follows:

PF Pg ( X) 0 f X (x)dx
(15)
g ( x )0

In the case of mooring systems, the dynamic and quasi-static tensions are functions of the variables in the environmental vector .
Furthermore, the maximum dynamic tension Tdyn is uncertain for given values of the environmental variables, as discussed before.
Therefore, it is beneficial to reformulate the reliability problem in Eq.15 to

PF Pr ( F x) f X (x)dx (16)
x

where Pr ( F x) is the probability of failure conditional on the environmental random variables x . It can be shown that the above form
reduces to Eq.15 with limit state function given as follows:

g (x) u (x) (17)

wherein u is an auxiliary standard Normal variable, (x) is the reliability index as a function of x , and 1 is the inverse standard
normal distribution. The corresponding limit state function for mooring line failure is then written as

g (, r , u) u 1[1 FTdyn (r t mean ())] 0 (18)

In the case of hull bending moment, assuming M W is Gaussian, the limit state function for hull bending, following Eq.17, is

f y s mS xS MW ()
g (, f y , s, xS , xW , u ) u 0 (19)
xS MW ()

Finally, an approximate relationship between the probability of failure during a given extreme event PF and the probability of failure
during any event in a given year, PF a , is:

Reliability Analysis of FPSO Components and Calibration of Safety Factors Using Response Surfaces
PF a PF (20)

where is the mean rate of extreme events per year.

CALIBRATION OF SAFETY FACTORS


Hereafter, we adopt the procedure from [1], which allows decoupling the safety factor calibration from the reliability analysis:
2
m
min j 0 ( j ) 0 (21)
j 1

where 0 is the target reliability index, j is the direction cosines vector in the standard Normal space associated with the most
probable point of failure of design case j ( j 1,2,...,m) , j is a weight factor for design case j , m is the number of design cases,
and is a vector that is obtained from the solution of the optimization problem in Eq. 21. The representative design point for all cases
in standard Normal space, is calculated by multiplying 0 times vector .

The next step is to transform this design point into the physical space, x D . However, the joint probability density of the basic variables
is in general different for each design case and x D , and consequently the final set of safety factors, will be different for each design
case, due to the simplified method we use here (in other more elaborated methods a unique set of partial safety factors is obtained
independently of the design cases [1]). Hence, in order to get a single set of safety factors, we will consider two options: (a) to take the
weighted average or (b) to take the maximum of all design cases considered. From x D , we then compute the values of rD , t mean, D
and t dyn, D , for mooring lines tension, and the values of f y , D , s D , x S , D , xW ,D , mS ,D and mW , D .for hull bending. Safety factors are
then obtained using the corresponding characteristic values of these variables and accounting properly for the contribution of random
variable u in limit state equations 18 and 19 to the dynamic component of the mooring lines tension and the hull bending moment.

CASE STUDY
Three FPSO systems are considered at different water depths in the Mexican side of the Gulf of Mexico (see Table 1). The FPSOs have
an internal turret and single-point mooring systems consisting of twelve lines divided in four groups of three lines each, spaced 90
around the hull. The properties of the mooring line systems are shown in Tables 2 to 4. The length of each component in a cable
segment is taken equal to 32.5 diameters [15]. In the case of chain segments each link is equivalent to one component, i.e. there are two
elements every six diameters in length, considering overlapping. The number of elements per line is also shown in Tables 2 to 4. The
resistance mean value is obtained by multiplying the minimum breaking load and resistance factor K , which depends on the line
material .

The basic variables for hull bending are modeled as shown in Table 5. Three static load conditions were considered for the hull: ballast,
medium loaded and fully loaded. The corresponding static bending moments are (in MN-m): 5518, 5479 and 2903. The frequencies of
occurrence of these load conditions are considered to be 0.1, 0.8 and 0.1, respectively.

Environmental Conditions
Environmental conditions were characterized following the work in Silva et al. [19]. Extreme sea states were selected using the Peaks
over Threshold method and a hindcast database for hurricanes and winter storms [18]. In total, 54 storm events were selected at
locations 1 and 2 and 58 events at location 3. The average number of extreme events per year, (Eq.20) is 54/41=1.32 year-1 for
locations 1 and 2, and 58/41=1.41 year-1 for location 3. From the selected extreme sea states the statistics of the environmental variables
was determined. The parameters of the marginal probability distributions are given in Tables 6 to 8. The joint probability distributions
of environmental variables were obtained using the Nataf model as formulated in Silva et al. [19].

Response Surfaces and Reliability Analyses


Tension in Mooring Lines
The resistance for the whole mooring line was modeled using a Type I extreme distribution for minima with equivalent parameters. For
the mooring line systems considered here, the resulting probability distribution parameters are shown in Table 9. In order to fit
response surfaces for the tension loads, dynamic structural analyses were carried out. The considered environmental combinations

Reliability Analysis of FPSO Components and Calibration of Safety Factors Using Response Surfaces
were H S (m)={11, 12 ,13 ,14}; TP (s)={13, 15, 17}; U10 (m/s)={22, 25, 28}; U C (m/s)={1.5, 1.75, 2}; wave ()={0, 22.5, 45}; wind
()={-45 to 90 @ 22.5}; current ()={-45 to 90 @ 22.5}. The RS coefficients were estimated for the first four environmental variables.

Reliability indexes were obtained using the distribution of Tdyn in Eq.18. In order to evaluate the influence of the number of
environmental variables for the RS on the reliability indices, two additional cases were analyzed using only two ( H S , TP ) and three
environmental variables ( H S , TP , U10 ). Results in Table 10 show that differences between all cases are very small (less than 2%),
which therefore indicates that H S and TP are of major relevance for the structural response in comparison with the other
environmental variables.

To illustrate, we show in Figs. 2 to 4 RS corresponding to location 1 for the case of two environmental variables ( H S , TP ).

Hull Bending
Values of environmental variables for building the response surfaces were H S (m)={11, 12 ,13 ,14}; TP (s)={13, 15, 17}; wave
()={0, 30}. Reliability indexes were obtained accounting for the probability distribution of M W () in Eq.19. Results are shown in
Table 11 for reliability indexes (conditional on a given static load condition) and in Table 12 for total reliability indexes.

360 Location 1 200 Location 1


350 180
160
340
140
330
120
320 [T dyn] (KN) 100
T mean (KN)
310 80
300 60
40
290
20 13.4
280 13.4 0 12.2
270 12.2 13 13.6 Hs (m)
13 13.6 Hs (m) 14.2 14.8
15.4 11
14.2 14.8 16 16.6
15.4 11
16 16.6 Tp (s)
Tp (s)
Figure 3: Response surface for Tdyn using two environmental
Figure 2: Response surface for Tmean using two environmental
variables (location 1)
variables (location 1)

Calibration of Safety Factors


In the calibration of safety factors, weight factors j (Eq.21) were considered equal for all design cases. In the case of moorings there
are three design cases: one for each location. In the case of hull bending there are six design cases considering the three locations and
two wave directions. A target annual reliability index 0 equal to 4.4 was selected based on proposed values by the Joint Committee
on Structural Safety for systems with large consequences of failure and normal relative cost of safety measure [20]. Combinations of
environmental variables associated with the maximum design responses for a 100-year return period for mooring lines and hull bending
are shown in Table 13.
Location 1
180
160 Tension in mooring lines
140
Table 14 lists the safety factors for each mooring line design case.
120
100
Results show that safety factors for resistance, R , and for quasi-
N cycles =t /2
80
60
static tension, Tmean , are close to 1.0 with no significant
40 variations for cases of two, three or four environmental variables.
20 13.4 Values of partial safety factor for the maximum dynamic tension,
0
13 13.6
14.2 14.8
12.2
Hs (m) Tdyn , are between 3.5 and 4.2.
15.4 11
16 16.6
Tp (s)
Figure 4: Response surface for the number of stress cycles of The safety factor for the dynamic component of the mooring line
dynamic tension using two environmental variables (location 1) tension, Tdyn , is greater than those found in earlier studies [3,4].
On one hand, this can be explained by the target reliability index;

Reliability Analysis of FPSO Components and Calibration of Safety Factors Using Response Surfaces
e.g. in [4] a target reliability of 3.71 was used, whereas the value used here is 4.4. Also it is believed that these greater values of the
safety factor are due to the environmental characteristics of the site considered. Because the number of extreme events in the
considered region is low, there is greater uncertainty on the extreme environmental conditions. This has a strong influence on the tails
of the distributions of the environmental variables, in particular H S , which is believed to be responsible for the differences of the
results here compared to earlier studies.

In a similar manner than for RS and reliability analyses, here a comparison were carried out when using only two ( H S , TP ) or three
environmental variables ( H S , TP , U10 ). Results are shown in Table 14. Differences between all cases are in general small (less than
10%). Again, the influence of H S and TP is dominant.

Hull Bending
In Table 15 the resulting safety factors for the hull at the different locations are presented. It can be observed that factors R and M S
are equal for each design case i (i 1,2,...,6) . This is understandable if we consider firstly that the probability density of Fy is
constant, then the relationship between its characteristic value and its design value is also constant for each design case. In the case of
the hull section modulus, S , the probability density is different for each design case, but S is a normal variable with constant
coefficient of variation ( i ). Then, taking into consideration that the characteristic value sC ,i equals the mean i for each design case,
the relationship between sC ,i and the corresponding design value s D,i will be constant. The partial safety factor for the static moment
is also constant since M S is deterministic given a static load condition and factor X S is normally distributed with the same parameters
for all design cases.

As to MW , we can see that this safety factor varies from 2.27 to 2.51. These values are higher than other used for tankers. We think
that this can be explained by the target reliability index used (e.g. 4.4) and also due to the environmental characteristics of the site
considered.

It should be kept in mind, as explained in section Calibration of Safety Factors, that due to the simplified method for calibration used
in this paper, the final set of partial safety factors is different for each design case. For this reason, we considered two options: (1) to
take the weighted average; and (2) to take the maximum of all design cases considered, in order to get a unique set of safety factors not
related to any of the three locations. However, note that this unique set of partial safety factor depends on the environmental conditions
(i.e., if we have another group of design cases and keep the target reliability as a constant, we will get in general another set of partial
safety factors if the environmental conditions are different to those considered in this work).

CONCLUSION
In this paper, a procedure utilizing response surfaces (RS) is presented to perform reliability assessment and calibration of safety factors
for the ultimate limit states of bending and tension for the hull and mooring lines of FPSO systems, respectively. The maximum
structural responses are related to extreme environmental (input) variables by using a RS approach, which decouples the reliability
analyses from the dynamic analyses. As shown in this paper, by using RS the influence of input variables on reliability analyses can be
easily evaluated.

The procedure includes a formulation from a previous study for solving the nested reliability analysis. For the calibration of safety
factors, a simplified method which decouples reliability analysis is used. In the application, three FPSO concepts were selected and
partial safety factors were calculated for a given target reliability index of 4.4. The resulting safety factors are similar for all considered
design cases. This is explained by the similarities in the design and the environmental conditions for the different cases. In the case of
moorings, the resulting safety factor for the dynamic contribution to the mooring line tension is significantly higher than those found in
earlier studies. It is likely that this is due to the specific characteristics of the environmental conditions in the considered area, where
extreme events are rarer than at the locations considered in previous studies. This points to the importance of utilizing safety factors
that are site specific (i.e. other numerical results could be found for other environmental sites or design cases).

ACKNOWLEDGEMENTS
We recognized Ibrahim Abu-Odeh from Global Maritime and Omar Vazquez from IMP, who worked on the FPSO response analyses
to produce data for the RS.

Reliability Analysis of FPSO Components and Calibration of Safety Factors Using Response Surfaces
REFERENCES
[1] GAYTON N., Mohamed A., Sorensen J.D., Pendola M., Lemaire M. (2004). Calibration methods for reliability-based design codes.
Structural Safety, 26, pp. 91121.
[2] HRTE T., Lie H., Mathisen J. (1998). Calibration of an ultimate limit state for mooring lines. Proc. 17th International
Conference on offshore Mechanics and Arctic Engineering, OMAE 98, Lisbon, Portugal, N1457.
[3] MATHISEN J, Larsen K., Hrte T., Halvor L. (1998). Calibration of a progressive collapse limit sate for mooring lines. Proc. 17th
International Conference on offshore Mechanics and Arctic Engineering, OMAE 98, Lisbon, Portugal, N1458.
[4] VAZQUEZ-HERNANDEZ A.O., Ellwanger G.B., Sagrilo L.V.S. (2006). Reliability-based comparative study for mooring lines
design criteria. Applied Ocean Research, 28 (6), pp. 398406.
[5] SUN H.-H., Bai Y. (2003). Time-variant reliability assessment of FPSO hull girders. Marine Structures, 16, pp. 219-253.
[6] MONTES-ITURRIZAGA R., Heredia-Zavoni E., Silva-Gonzlez F. and Straub D. (2010); Nested reliability analysis and
calibration of safety factors for mooring lines, Applied Ocean Research (submitted for revision).
[7] SUN H.-H., Guedes Soares C. (2003). Reliability-based structural design of ship-type FPSO units. Journal of Offshore Mechanics
and Arctic Engineering, 125, pp. 108-113, May.
[8] SUN H.-H., Guedes Soares C. (2006). Reliability-based inspection of corroded ship type FPSO hulls. Journal of Ship Research,
50(2), pp. 171-180, June.
[9] SARKAR A., Eatock Taylor R. (2000). Effects of mooring line drag damping on response statistics of vessels excited by first and
second-order wave forces. Ocean Engineering, 27, pp. 667 686.
[10] CHOI Y.J., Gilbert R.B., Ding Y., Zhang J. (2006). Reliability of Mooring Systems for Floating Production Systems. Report,
Offshore Technology Research Center, Texas A&M University.
[11] VANMARCKE E.H. (1975). On the distribution of the first passage time for normal stationary random processes. Journal of
Applied Mechanics. ASME, March, pp. 215-220.
[12] VANMARCKE E.H. (1983). Random fields: Analysis and synthesis. The Massachusetts Institute of Technology.
[13] ANG A. H.-S., Tang W.H. (1984). Probability concepts in engineering planning and design, Volume II: Decision risk and
reliability. John Wiley and Sons.
[14] MONTES-ITURRIZAGA R., Heredia-Zavoni E., Silva-Gonzlez F. (2007). On the estimation of mooring line characteristic
resistance for reliability analysis. Applied Ocean Research, 29 (4), pp. 239-241.
[15] VAZQUEZ-HERNANDEZ A.O. (2004). Reliability-based calibration methodology for mooring line design. D.Sc. thesis,
COPPE/UFRJ, Brazil (in portuguese).
[16] BUCHER, C.G., U. Bourgound. (1990). A fast and efficient response surface approach for structural reliability problems.
Structural Safety ,7 (1), pp. 57-66.
[17] RAIFFA H., Schlaifer R. (1961), Applied Statistical Decision Theory. Harward University Press. Cambridge Uni-versity
Press.Cambridge, Mass.
[18] OWI Oceanweather Inc. (2006). Update of metocean design data for Zona Norte and Sonda de Campeche. Final Report for IMP
(Revised), July 11.
[19] SILVA-GONZALEZ F., Heredia-Zavoni E. and Montes-Iturrizaga R. (2011), Multivariate characterization of extreme sea-states
using Nataf transformation, sent for possible publication to Ocean Engineering.
[20] JCSS (2006). Probabilistic Model Code, Joint Committee on Structural Safety, (www.jcss.ethz.ch).

Table 1: Water depths of the selected FPSOs


Location Depth (m)
1 500
2 1000
3 1500

Table 2: Properties of mooring lines in Location 1


Minimum Resistance Resistance Coefficient of Cable or link Number of
breaking load, Length,
Segment Material factor, diameter, elements,
(kN) K
variation, R L (m)
d (m) N
1 Chain 1423 1.2 0.05 130 0.122 3552
2 Polyester 1560 1.1 0.05 950 0.232 126
3 Chain 1423 1.2 0.05 285 0.122 779

Reliability Analysis of FPSO Components and Calibration of Safety Factors Using Response Surfaces
Table 3: Properties of mooring lines in Location 2
Minimum Resistance Resistance Coefficient of Cable or link Number of
breaking load, variation, Length, elements,
Segment Material factor, diameter,
(kN) K R L (m)
d (m) N

1 Chain 1213 1.2 0.05 150 0.111 450


2 Polyester 1296 1.1 0.05 1500 0.150 308
3 Chain 1213 1.2 0.05 150 0.111 450

Table 4: Properties of mooring lines in Location 3


Minimum Resistance Resistance Coefficient of Cable or link Number of
breaking load, variation, Length, elements,
Segment Material factor, diameter,
(kN) K R L (m)
d (m) N

1 Chain 2018 1.2 0.05 150 0.150 333


2 Polyester 2052 1.1 0.05 2000 0.248 248
3 Chain 2018 1.2 0.05 150 0.150 333

Table 5: Modeling of basic variables for hull bending


Standard
Variable Type Mean
Deviation

Fy (MPa) lognormal 362.25 28.98


3
S (m ) normal 47 1.88

XS normal 1 0.10
XW normal 1 0.25

Table 6: Statistical models of environmental variables: Location 1


Parameters
Standard
Variable Type Mean
deviation
Scale* Shape Location**

HS (m) Weibull 3p. 2.46 1.47 4.00 6.23 1.54


TP (s) Lognormal
0.12 2.43 11.48 1.37
U10 (m/s) Lognormal 0.16 2.80 16.71 2.74
UC (m/s) lognormal 0.78 -1.24 0.39 0.36

Table 7: Statistical models of environmental variables: Location 2


Parameters
Standard
Variable Type Mean
deviation
Scale* Shape Location**

HS (m) Weibull 3p. 2.36 1.42 4.00 6.15 1.53


TP (s) Lognormal
0.12 2.44 11.50 1.33
U10 (m/s) Lognormal 0.17 2.79 16.60 2.85
UC (m/s) lognormal 0.85 -1.41 0.35 0.36

Reliability Analysis of FPSO Components and Calibration of Safety Factors Using Response Surfaces
Table 8: Statistical models of environmental variables: Location 3
Parameters
Standard
Variable Type Mean
deviation
Scale* Shape Location**

HS (m) Weibull 3p. 2.08 1.21 4.00 5.95 1.63


TP (s) Lognormal
0.12 2.41 11.20 1.32
U10 (m/s) Lognormal 0.16 2.79 16.42 2.64
UC (m/s) lognormal 0.68 -1.03 0.45 0.35
* For lognormal variables standard deviation of the variable natural logarithm
** For lognormal variables the expected value of the variable natural logarithm

Table 9: Statistical model of mooring line resistance (kN)


Parameters Standard
Location Mean
Scale Location deviation
1 0.0432 1407.3 1394.0 29.69
2 0.0474 1216.2 1204.0 27.09
3 0.0290 1953.0 1933.1 44.20

Table 10: Annual reliability index for mooring lines

Annual reliability index, a


Location
Four environmental variables Three environmental variables Two environmental variables
( H S , TP , U 10 , U C ) ( H S , TP , U 10 ) ( H S , TP )
1 4.25 4.24 (-0.2) 4.27 (0.5)
2 3.64 3.62 (-0.5) 3.62 (-0.5)
3 4.54 4.63 (2.0) 4.59 (1.1)
Note.- In parenthesis: differences with respect to case of four environmental variables in percentage (%)

Table 11: Annual reliability index for hull bending (conditional on a static load condition)
Annual reliability index, a
Location
Ballast Medium Loaded Full Loaded
wave =0 wave =30 wave =0 wave =30 wave =0 wave =30
1 3.22 3.26 3.11 3.18 3.20 3.26
2 3.22 3.27 3.12 3.18 3.21 3.27
3 3.11 3.14 3.02 3.06 3.14 3.14

Table 12: Annual reliability index for hull bending (total)


Annual reliability index, a
Location
wave =0 wave =30
1 3.13 3.20
2 3.14 3.20
3 3.04 3.08

Table 13: Combinations of design values of the environmental variables associated with
maximum response (mooring tension and hull bending)
Location HS (m) TP (m) U10 (m/s) UC (m/s)

1 10.77 14.45 22.79 2.0

2 10.56 14.46 23.11 2.0

3 11.03 14.41 22.88 2.0

Reliability Analysis of FPSO Components and Calibration of Safety Factors Using Response Surfaces
Table 14: Partial safety factors for mooring lines
Four environmental variables Three environmental variables Two environmental variables
Safety
( H S , TP , U 10 , U C ) ( H S , TP , U 10 ) ( H S , TP )
Factors
Loc.1 Loc.2 Loc.3 Mean Max Loc.1 Loc.2 Loc.3 Mean Max Loc.1 Loc.2 Loc.3 Mean Max
1.01 1.01 1.02 1.01 1.02 1.01 1.01 1.01 1.01 1.01
R 1.02 1.01 1.02 1.02 1.02
(-1.0) (0.0) (0.0) (-1.0) (0.0) (-1.0) (0.0) (-1.0) (-1.0) (-1.0)
Tmean 1.17 1.05 1.09 1.10 1.17
1.15
(-1.7)
1.06
(1.0)
1.10
(0.9)
1.10
(0.0)
1.15
(-1.7)
1.13
(-3.4)
1.03
(-1.9)
1.08
(-1.0)
1.08
(-1.8)
1.13
(-3.4)
Tdyn 3.84 3.76 4.13 3.91 4.13 3.74 3.74 4.21 3.90 4.21
3.53 3.79 4.21 3.84 4.21
(8.8) (-0.8) (-1.9) (1.8 (-1.9) (5.9) (-1.3) (0.0) (1.6) (0.0)
Note.- Between parenthesis differences (%) respect to case of four environmental variables

Table 15: Partial safety factors for hull bending


Safety Design case
Factors 1 2 3 4 5 6 Mean Max.
R 0.99 0.99 0.99 0.99 0.99 0.99 0.99 0.99
MS 1.01 1.01 1.01 1.01 1.01 1.01 1.01 1.01

MW 2.36 2.28 2.44 2.34 2.51 2.27 2.37 2.51

Reliability Analysis of FPSO Components and Calibration of Safety Factors Using Response Surfaces
Discussion
Lyuben D. Ivanov (Visitor)
In this paper, the authors have used a reliability-based procedure utilizing response surfaces to calculate the safety factors for the hull
girder bending and mooring lines tension of FPSO systems. The level of knowledge and corresponding uncertainties in these three
areas (reliability methods, hull girder bending, and tension of mooring lines) are different; therefore, it is natural that questions may
arise when these three areas are combined. Although it is not possible to solve all problems in those areas in one paper, I would like to
ask the authors for their opinion on several comments I have:

It is a correct statement that the reliability of a mooring line depends on the reliability of the weakest link. One can observe in
Tables 2, 3, and 4 that the number of elements is very largefrom several hundred to several thousand. How can one find which
one is the weakest? This weakest element can be anywhere and its location may cause a different path of failure propagation.
It is interesting to know how Turkstras rule was applied for the combination of the quasi-static and dynamic load on mooring
lines.
The limit function for ultimate hull girder bending in Eq. 3 can be used as a first approximation if the ultimate hull girder bending
moment (BM) is greater than the first-yield BM. Unfortunately this is not always the case. Due to buckling, the stresses in the
farthest fiber from the Neutral Axis cannot reach the yield stress. As a result, the ultimate hull girder BM is smaller than the first-
yield BM
The reference to literature that the probabilistic distribution of the sagging still water BM follows Rayleigh distribution and the
hogging still water BM follows the exponential distribution is arguable. Data for FPSOs loading cases with corresponding still
water BM is available [1] that suggests other types of its probabilistic distribution. Change of the probabilistic distribution will
certainly change the numerical results for the calculated safety factors.
It is assumed that the hull girder wave-induced BM is Gaussian. What is the basis of this assumption? In shipbuilding, Weibull
distribution for the wave-induced BM is used for its long-term distribution derived by individual amplitude statistics.
It is stated that the probability density function of the hull girder section modulus is Gaussian with constant coefficient of variance;
however, it has been proven [2] that the mean value of the section modulus decreases with hull girder aging while the standard
deviation increases over time. So, neither of these two parameters is constant over time.
In all calculations, the effect of time on safety indexes is not shown. The reason is that the time-variant reliability approach is not
applied. Therefore, it is desirable to have the comments of the authors on this topic which is very important for Risk-based
Inspection planning of FPSOs.

Finally, I would like to say that the authors have prepared an interesting and useful paper.

[1] IVANOV, Lyuben D., Albert Ku, Bei-Qing Huang, Vivianne C. S. Krzonkala Probabilistic presentation of the total bending
moments of FPSOs, Ships and Offshore Structures, vol. 6, Nos. 1-2, 2011, 45-58 or in Proceedings of MARSTRUCT2009
Conference, Lisbon, Portugal, 16-18 March 2009
[2] IVANOV, Lyuben D. A Probabilistic Presentation of All Hull Girder Geometric Properties at any Ships Age, RINA
Transactions, International Journal of Maritime Engineering, vol. 149, part A3, 2007, pp. 45-92)

Authors Closure
The authors appreciate Dr. Ivanovs comments and thank him for his valuable and fruitful discussion.

Mooring lines are modeled as series systems of segments. Each line segment is also a series system composed of a large number of
elements with resistances that are considered to be statistically independent and identically distributed. Under such assumptions any
element along a line segment is equally likely to be the weakest link. Determining which one is the weakest link is not really relevant
for characterizing capacity of a line segment, or, for that matter, of the whole mooring line, since failure of the weakest element
(whatever its location) entails the failure of the system. This fact is taken into account by means of extreme value theory in the paper
and has been analyzed in previous works. In Montes-Iturrizaga et al. (2007) it is shown that depending on the type of distribution for
the resistance of the components, type I or type II extreme distributions for minima can be obtained for the resistance of a line
segment.

We agree with the discusser, and due consideration must be given to the characterization of the bending capacity. The work reported
in the paper rather focuses on characterizing the maximum still water bending moment in limit state in Eq. 3. It is expressed in terms

Reliability Analysis of FPSO Components and Calibration of Safety Factors Using Response Surfaces
of metocean variables by means of response surfaces, and the nested reliability problem that results from considering that such
variables are random is then addressed. Buckling of hull-stiffened panels was considered in separate component reliability calculations
which, as indicated, are not included in the paper.

On the modeling of the still water bending moment (SWBM) data available suggests different possible probability distributions
models. Based on statistical analyses of operational data for an FPSO, Wang and Moan (1996) proposed exponential and Rayleigh
distributions, which have been used in previous reliability studies as referenced in the paper. Huang and Moan (2005) developed a
more versatile distribution model considering that SWBMs are subject to operational control and that some abnormal operations and
exceptions may result in an allowable value being exceeded. For the short-term variability, they developed a Weibull model which
accounts for load control during operations. More recently Ivanov et al. (2009) developed a bimodal normal distribution model for
hogging and sagging considering that the bending phenomenon is one and should have one probabilistic distribution. Their statistical
analysis of records of an FPSO in a four-year period showed bimodal normal distributions to be well-suited to model the data.
Assessing the effect of using different probability distributions for the SWBM on the calibrated safety factors is a task that remains to
be addressed by future studies.

The nested reliability formulation used in the paper is general and does not require that a particular probability distribution be
assumed. It may be applied using a Weibull distribution for the wave-induced bending moment. Under the Gaussian assumption an
analytical expression is obtained for the limit state function in terms of the mean and standard deviation of the maximum wave-
induced bending moment, which are both functions of the vector of random environmental variables, as given by the response
surfaces. Such an assumption is introduced for simplicity of the solution for the purpose of the application example.

In the paper the safety factors are calibrated based on annual target reliability indexes without considering deteriorating processes (i.e.
these indexes are supposed to be constant). This assumption is a usual practice in the case of the design of new structures and it is also
the philosophy behind safety factors established in modern standards and regulations for offshore platforms. However, we recognize
that for risk-based inspection planning, time-variant reliability analyses are necessary. The authors have carried out a number of
studies on optimal risk-based inspection planning of fixed marine platforms and are aware that proper consideration must be given to
damaged and deterioration mechanisms, as well as to their influence on the variation of the probability of failure along time.

IVANOV, L.D., Ku, A., HUANG, B.Q., KRZONKALA V.C.S. (2009). Probabilistic presentation of the total bending moments of
FPSOs. 2nd Intl. Conf. on Analysis and Design of Marine Structures, MARSTRUCT2009, Lisbon, Portugal.
HUANG, W., MOAN, T. (2005). Combination of global still water and wave load effects for reliability-based design of floating
production, storage and offloading (FPSO) vessels. Applied Ocean Research, 27, 127141.
MONTES-ITURRIZAGA, R., HEREDIA-ZAVONI, E., SILVA-GONZLEZ, F. (2007). On the estimation of mooring line
characteristic resistance for reliability analysis. Applied Ocean Research, 29 (4), pp. 239-241.
WANG, X., MOAN, T. (1996). Stochastic and deterministic combinations of still water and wave bending moments in ships. Marine
Structures, 9, 787-810.

Reliability Analysis of FPSO Components and Calibration of Safety Factors Using Response Surfaces

Vous aimerez peut-être aussi