Ocean Engineering
journal homepage: www.elsevier.com/locate/oceaneng
a r t i c l e i n f o
abstract
Article history:
Received 18 January 2011
Accepted 28 January 2012
Editor-in-Chief: A.I. Incecik
Available online 19 March 2012
Structural systems will normally fail as a consequence of a chain of different components failure es.
In this paper, fatigue reliability of xed offshore platforms is investigated by analyzing different failure
scenarios. In order to evaluate the occurrence probability of a special scenario, it is divided into a nite
number of sub-scenarios. All combinations of time sequences are generated for a given sequence
of failures, using a specially developed program.
In order to calculate the occurrence probability of each scenario, a massive reliability analysis
should be done for each of corresponding sub-scenarios. A large number of sub-scenarios should be
analyzed, therefore implementing time consuming traditional methods for evaluating fatigue reliability
may be unrealistic, and a simplied and accurate method is required. Herein, an Articial transfer
function is used to calculate the cumulative fatigue degradation in components. The preciseness of the
proposed method is evaluated using a numerical model, and then, it is used to calculate the occurrence
probability of each sub-scenario. Based on the calculated values, probability of occurrence is obtained
for each scenario, and nally, the failure probability of entire system is calculated. The proposed
method can be used in inspection planning and evaluating the life extension of existing offshore
platforms.
& 2012 Elsevier Ltd. All rights reserved.
Keywords:
Fatigue damage
Structural reliability
Failure scenario
Articial transfer function
1. Introduction
Fatigue deteriorating phenomenon is a major failure mode
accounted for in design codes of offshore structures. The awareness of fatigue degradation in metallic structures started in the
mid 19th century with the occurrence of fatigue failures in the
railway industry (Gordon, 1978). In later years, fatigue failure
of pentagon-type semi-submersible Alexander L. Kielland platform, caused an increased focus on fatigue problems in the
offshore industry (Moan, 1981). A detailed theoretical background
for fatigue analysis is given by Martindale and Wirshing (1983),
Stahl and Geyer (1984), Almar-Naess (1985), Gurney (1979),
Kumar and Karsan (1990), Maddox (1992), Suresh (1991), Dover
and Madhava Rao (1996). More recently, fatigue crack growth has
been studied by among others Lassen (1997).
Considering the high costs of installation and repair of xed
offshore platforms and their environmental disasters in the case
of failure, it is necessary to create more detailed procedures for
Corresponding author.
E-mail addresses: Gholizad@uma.ac.ir (A. Gholizad),
Golafshani@sharif.edu (A.A. Golafshani).
0029-8018/$ - see front matter & 2012 Elsevier Ltd. All rights reserved.
doi:10.1016/j.oceaneng.2012.01.033
Nomenclature
G( )
gi
H(f)
SY(f)
SX(f)
g(z)
DF
A
O
B
TL
m
Transfer function
Response spectrum
Excitation spectrum
Limit state function for fatigue
Value of the PalmgrenMiner damage index at failure
Fatigue strength coefcient
Dened stress parameter
Inaccuracies in estimating the fatigue stresses
Design life in seconds
Negative inverse slope of the logarithmic SN Curve
N A S m
li
si
f0i
f
a, C
Ts
TZ
F( )
li
Mi
DF A
T L
Bm O
O 2 2m G 1
n
m X
g f l sm
2 i1 i 0i i i
2. Approximate method
In this section, an approximate method is developed for
calculating fatigue reliability of components in an offshore platform. This method will be used later to calculate occurrence
probability of each sub-scenario. Since there are a large number of
sub-scenarios to be analyzed, the proposed method should have
the possible highest accuracy while doing the least amount of
calculations.
2.1. Assumptions
As stated in API (2000), A transfer function denes the ratio of
the range of a structural response quantity to the wave height as a
function of frequency. In a linear system, having an excitation
spectrum SX(f), and a transfer function H(f) at a specic point,
stress process SY(f) can be achieved by:
SY f H2f SX f
Table 1
Denition of random variables Madhavan Pillai and Meher Prasad (2000).
Variable
Distribution
Mean/median
COV
A
B
Lognormal
Lognormal
Lognormal
0.63
0.20
0.30
DF
a
si CH
s
a2
T 5s
0:0038 4
16
T z 1 1=T 4s T 4z 9=4
In Eq. (3), m is the fatigue exponent; n is the total number of seastate in the wave scatter diagram; gi is the fraction of time spent
in the ith sea state; li is the wide band correction factor; G( ) is
representative of gamma distribution function and f0i, si are the
zero mean crossing frequency and standard deviation of stress
process for ith sea state, given by:
q
R1 2
0 f SY f df
4
f 0i
si
Z 1
si
SY f df
li am 1am1ei bm
10
where
am 0:9260:033m
11
bm 1:587m2:323
12
LnmT =T L 0:5s2LnT
sLnT
14
mT
mDF mA
mB O
15
q
2
sLnT Ln1 C A 1 C D 1 C B m
16
Converting the unit of T from seconds into years and substituting a value of 0.827 for wide band correction factor, the
equation can be rewritten as:
!
n
X
b 111:365Ln
gi f 0i sm
1:365LnT
18
i
i1
HS
TZ
(ft)
(m)
2.5
3.5
1.0
2.0
3.0
4.0
5.0
6.0
7.0
8.0
9.0
10.0
11.0
12.0
13.0
14.0
0.30
0.61
0.91
1.22
1.52
1.83
2.13
2.44
2.74
3.05
3.35
3.96
4.27
4.57
23
17
17
53
16
2
1
3
52
127
35
4
4
4.5
5.5
17
74
61
17
4
1
32
11
20
16
9
4
1
14
6
14
7
3
14
7
2
3
1
6
10
2
3
8
4
2
1
6.5
2
2
1
2
1
4
2
2
1
1
1
Table 2
Sea-scatter data Vughts and Kinra (1976).
20
Pf sys
K
X
P sce k
21
k1
Fig. 3. Evaluating preciseness of the two-parameter ATF method: (a) numerical model, (b) member 1, (c) member 2 and (d) member 3.
scenario have failed and all other members are in the safe margin.
Hence, Psubk (o) can be calculated as:
O
X
P subk o
22
o1
Psubk o
m=
2Y
kth sce
Rmo
m AY
kth sce
P f mo
23
Fig. 4. All possible occurrence times for scenario with three events.
Fig. 5. Simplied model of the platform, members numbering and some failure scenarios.
where, F{ } is representative of the normal distribution function. In order to use this equation, the entire jacket should be
analyzed SC times. In this equation, j 0 denotes intact state of
the platform where no failure has occurred in members. Subsequently, j 1 denotes that the rst component of the corresponding scenario has failed and this member should be removed from
numerical model of platform. In the above expression, Tj 1 is the
(j1)th component of the oth sub-scenario which denotes the
time between jth and (j 1)th failures.
In Eq. (23), Pf(m)o is the failure probability of mth member
(mAkth scenario), at the time determined by the oth sub-scenario.
Let m be the qth component of the corresponding scenario.
n
X
!)
m
0,i i q1 T q 1
f 0,i s g
i1
8
0
19
q1 X
n
<
=
X
m
@
f 0,i s0,i gi j T j 1 A
F:111:365Ln
;
qa 1
j 0i 1
25
(
Pf mo F 111:365Ln
n
X
i1
!)
f 0,i sm
0,i gi q1 T q 1
(
F 111:365Ln
n
X
!)
f 0,i sm
0,i gi q1 T q
q1
26
i1
4. Numerical example
In this section, an example jacket is analyzed using the proposed
method and its failure probability is obtained by the means of a
specially developed computer code. The numerical model of jacket is
shown in Fig. 5 and its members properties are listed in Table 3.
In this example, it is assumed that the failure of the entire jacket
will occur in the form of one of the 18 scenarios which are shown in
Fig. 5. All of the possible sub-scenarios are generated for these
scenarios and the numerical model is analyzed using the prescribed
method.
In order to nd natural period of the jacket in each analysis
(which is required when using approximate method), the jacket is
modeled using a cantilever beam and three lumped masses
(shown in Fig. 5). It must be noted that the amount of these
lumped masses are constant during all stages of the analysis,
while the stiffness of each part varies based on the collapsed
members in each phase of analysis.
If it is intended to plot reliability of the entire jacket (b) versus
its life time (T), then the numerical model should be analyzed
using different values for T. Fig. 6 shows the corresponding results
for two scenarios which are analyzed using a life time of 25 years.
Each of these diagrams contains the resultant probability of
occurrence for 2925 sub-scenarios. These values are calculated
using Eq. (23) in conjunction with Eqs. (24), (25) and (26).
Based on the discussions of previous section, each of the subscenarios is an independent random process, and hence, they can
be added up to give the occurrence probability of the corresponding scenarios. For example, one can add up the occurrence
probabilities of the Fig. 6(a) using Eq. (22) to achieve only a one
Table 3
Members properties of example jacket (cylindrical sections).
Members number
Thickness (cm)
1,
2,
3,
5,
120
80
80
100
2.25
1.25
1.5
2
4
6, 10
7, 11
8, 9, 12
point on the Fig. 7(a). This should be done for all other scenarios
using different values for in-service life of the jacket (5, 10, 15, 20
and 25 years) to obtain all of the curves shown in Fig. 7(a).
As it is apparent from Fig. 7(a), the relative vulnerability of
different failure paths may be determined from this gure. Based
on the results, one can nd the critical scenarios (most probable
failure paths) from Fig. 7(a) and strengthen the corresponding
weak members involved in these scenarios.
Using Eq. (21), one can add up the curves of Fig. 7(a), to obtain
failure probability of the entire jacket in different years of the
jackets life. This is shown in Fig. 7(b), which represents the failure
probability of the entire jacket among its constitutive components (failure probability of the scenarios).
Failure probability of the entire jacket under fatigue degradation is re-plotted in Fig. 8(a). This curve may be converted to show
the reliability index of the entire jacket under fatigue loads:
27
Fig. 6. Results of analysis for two sample scenarios of failure: (a) 12-10-9, (b) 8-6-5.
Fig. 7. Results of analyses for different scenarios: (a) Failure probability of each scenario, (b) adding failure probability os scenarios.
5
4.5
1.2
Reliability index
1.4
1
0.8
0.6
4
3.5
3
0.4
2.5
0.2
10
15
Years in service
20
25
1.5
10
15
Years in service
20
25
Fig. 8. Fatigue parameters of the entire jacket: (a) Failure probability of entire jacket, (b) fatigue reliability index of entire jacket.