System Failure Probability of Offshore Jack Up Platforms in The Combination of Fatigue and Fracture (Pub Year)

System failure probability of offshore jack-up platforms
in the combination of fatigue and fracture

Naser Shabakhty
Department of Civil Engineering, Faculty of Engineering, University of Sistan and Baluchestan, PO Box 98164-161, Zahedan, Iran
a r t i c l e i n f o
Article history:
Received 10 April 2010
Received in revised form 2 September 2010
Accepted 5 September 2010
Available online 25 September 2010
Keywords:
Offshore
Jack-ups
System failure probability
Fatigue
Fracture
a b s t r a c t
As it is known, fatigue is the process of damage accumulation in material due to the stress
uctuation. The failure occurs when accumulated damage exceed critical level. In practice,
uncertainties are incorporated in fatigue estimation especially in the loads and in the
capacity of structures. Reasonably these uncertainties can be incorporated within reliabil-
ity framework or probability context. It is possible to estimate the fatigue reliability based
on crack propagation in fracture mechanics and also using reliability context. However the
rst failure of structural element under fatigue degradation may not lead to the collapse of
system, an approach to combine in the system failures is needed.
This paper presents a new approach to estimate structural system failure probability of
jack-up platforms in combination of fatigue and fracture. The probability of failure for each
component is initially estimated using typical fatigue and fracture failure scenario. But for
the second element failure in combination of fatigue and fracture, a new approach is pre-
sented. This method is based on the Monte Carlo Simulation of crack size in accordance
with the crack growth approach in fracture mechanics. Important sequences leading to
the structural collapse are identied within the branch and bound technique. Finally the
system failure probability through combination of important paths leading to the struc-
tural collapse is obtained. Advantage of this method in addition to its simplicity in appli-
cation is possibility of using the FORM technique to estimate each failure path
individually and nally combine them together to determine system reliability. It is shown
the system failure under combination of fatigue and fractures has higher probability of
occurrence than the purely rst or in combination with the second fatigue failures only.
2010 Elsevier Ltd. All rights reserved.
1. Introduction
The objective of the structural design is to build proper structures meeting functionality, safety and economical aspects.
They are closely connected to each other and an iterative procedure is necessary to achieve an optimal design. The conse-
quences of failure concern the safety of humans, pollution, and the cost of structures and equipment. Therefore, the assess-
ment of the safety of offshore platforms including jack-up becomes essential.
Time-varying nature of the environmental loads such as wave and wind will cause uctuation of stresses in the structural
components of platforms. The fatigue damage is a deterioration process, which arises from uctuation of stress, and should
be evaluated before reaching a critical level. Traditionally, fatigue has not been considered as an important problem in the
jack-up platforms. The main reason was that the most platforms were designed in the area with low or moderate environment
1350-6307/$ - see front matter 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.engfailanal.2010.09.002
Tel.: +98 21 33050192.

E-mail address: shabakhty@eng.usb.ac.ir
Engineering Failure Analysis 18 (2011) 223243
Contents lists available at ScienceDirect
Engineering Failure Analysis
j our nal homepage: www. el sevi er . com/ l ocat e/ engf ai l anal
conditions. Recently, by using jack-up platforms in the region with more severe sea states or as permanent platforms, the
fatigue damage becomes an important factor in evaluating the integrity of the structures [1].
Design practice today is mainly based on the component level. However, the safety of the overall structure (the system)
due to degradation like fatigue may be of greater interest when platforms are in operation for great parts of their original
design life and intention is to extend operation beyond the predicted lifetime. In addition, the whole platform consists of
several individual elements which the failure of one element may not lead to the system collapse. This effect is known as
the redundancy in platform and Jack-up structures such as other platforms, the failure of its individual element does not lead
to the structural collapse. Hence, a procedure is required to estimate system reliability of jack-up platforms cause by fatigue
damage
In the last decade several investigations are carried out on the system reliability of jacket platforms due to sequence fail-
ure in fatigue and under extreme environmental loads or combination of these two failure modes, (Dalane [2]). However,
researches in jack-up platforms are more restricted to the component level e.g. Jensen [3] and Shabakhty et al. [25] or
the system effect under failure of extreme environmental loads excluding fatigue failure mode, e.g. Karunakaran [4]. How-
ever, Shabakhty et al. [25] attempt to investigate fatigue reliability of jack-up platform in the component level but they do
not incorporate the fracture failure effect not only in component level but also in the more complex combination of system
level. Therefore, in the following sections a methodology for system reliability of jack-up platform in combination of failure
in fatigue and fracture are developed and investigated.
2. Fatigue damage model
The main parameters governing the fatigue crack propagation are geometry of joint, location and dimension of crack ini-
tiation, fatigue material characteristics and environmental loads. They are random in nature, analysis and design based on
probabilistic method is consequently most appropriated. The design criteria are generally a limiting value of the probability
that the fatigue damage exceed the critical value during the design life.
The fatigue damage prediction may be performed principally into two techniques. The rst is SN approach, which relates
a constant stress range, S, to the number of stress cycles, N leading to the failure. The SN curve is mainly obtained from
empirical fatigue experiments. To extend this approach to the variable stress amplitude, the PalmgrenMiner model is ap-
plied [5]. The second approach is mainly established on the fracture mechanics techniques, which describe the propagation
of crack and its growth under developments of stress eld at the crack tip. We applied this technique in our research and
more detail is given in the following sections.
2.1. Nucleation and propagation of crack
Fatigue is the process of damage accumulation initiated from yielding in the material by sliding of atomic layers. This slid-
ing is caused by a combination of dislocations and local stress concentrations and each slip, however small, related to the
small deteriorations in the structural material [5]. The dislocations are increasing under cyclic stresses and combining to
each other to cause the nal plastic deformations. Microscopic cracks are thus created and jointed to each other to produce
nal major cracks. The total combination time of crack initiation and growth constitutes the complete lifetime of fatigue
damage accumulation as shown in Fig. 1.
The fatigue crack typically occurs on the free surface of the body at places of high stress concentrations e.g. weld toes,
surface imperfections, grinding boundaries, etc. Based on material properties and loading type, the nucleation phase can
be of different importance in estimating fatigue life. Experimental observation indicates that at high-cycle fatigue (low stress
amplitudes) signicant proportion of fatigue life may consume by crack initiation period whereas at low-cycle fatigue (high
stress amplitude) crack starts to develop in the early cycles. In addition, in some structural details such as joint connections
in offshore platforms where defects are practically unavoidable due to the fabrication processes, crack propagation may be of
considerable in beginning of the rst load application [2].
2.2. Fatigue crack growth model
The best expression to predict the growth of a surface crack per stress cycle at any points along the crack front is given
by Paris and Erdogan [6]. The shape of fatigue crack is assumed to be semi-elliptical and remains semi-elliptic during
Micro-crack
growth
Nucleation Macro-crack
growth
m 10 m 100 m 1 mm
Complete lifetime of fatigue crack
Final failure
Nucleation Period Crack growth period
Complete lifetime of fatigue crack
Fig. 1. Nucleation and propagation of crack [5].
224 N. Shabakhty / Engineering Failure Analysis 18 (2011) 223243
propagation of crack until failure [7]. Fig. 2 shows the crack depths (a) and crack length (2c), which are two main factors to
describe the crack front. Based on ParisErdogan expression, the increment of crack size dr(u), during application of load
cycle dN at a specic point along the crack front is related to the range of stress intensity factor DK
r
(u) described by
dr(/)
dN
= C
r
(/)(DK
r
(/))
m
; DK
r
(/) > 0 (1)
C
r
(u) and m are the fatigue material characteristics, and u is the angle of location. This differential equation satises the
crack front at all points during crack propagation. The following pair of differential equations can be derived extending this
expression in two directions of depth and surface,
da
dN
= C
A
(DK
A
)
m
; a(N
0
) = a
0
(2)
dc
dN
= C
C
(DK
C
)
m
; c(N
0
) = c
0
(3)
Which two subscripts C and A refer to the end point and deepest point of the crack at the surface, respectively. Shang-Xian
[7] shows the material property m depends on the fatigue crack propagation. Therefore, it is reasonable to assume m inde-
pendent of crack size in both directions of depth and surface. Investigation of Raju and Newman [8] show the material
parameters C
A
and C
C
may slightly differ due to the general three-dimensional stress eld. Therefore, they propose to relate
C
A
and C
C
with the following expression.
C
A
= 1:1
m
C
C
(4)
The failure criteria in fatigue actually refers to a critical value of the crack depth a or crack length c, thus it is convenient to
rewritten the last two equations with,
dc
da
=
C
C
C
A
DK
C
DK
A
_ _
m
; c(a
0
) = c
0
(5)
da
dN
= C
A
(DK
A
)
m
; N(a
0
) = N
0
(6)
The general expression for the stress intensity factor is K = Yr

pa
_
. The geometry function Y accounts for the effect of all
boundaries such as width, thickness, crack front curvature and etc. Raju and Newman [8] proposed empirical expression to
specify the stress intensity factor K(u) for the surface crack in a nite plate subjected to the remote tension and bending
loads. This expression is tted on the available nite element results for two types of remote tension and bending loads ap-
plied to the surface cracked-plate. The inferred stress intensity factor from this research is given by
DK
plate
(/) = D(r
t
r
b
)
pa
_
(1=Q)
1=2
1 HS
b
=S
t
1 S
b
=S
t
F
a
t
;
a
c
;
c
b
; /
_ _
= Dr

pa
_
Y
plate
(7)
where r
t
is the remote uniform tension stress and r
b
is the remote outer-ber bending stress. Q is the shape factor and
parameters of F and H are the boundary-correction terms. This equation is derived for the cases of 0 < a/c 6 1.0, 0 6
a/t < 1.0, c/b < 0.5 and 0 6 u 6 p. In later investigation of Raju and Newman [9] for determination of stress intensity factor
in rods and pipes, they found differences between stress intensity factor in semi-elliptical crack of plates and pipes, when c/b
approach to zero are not signicant and the stress factor at two points of depth A (u = p/2) and surface C (u = 0) can
therefore express with
P
M
t
t
X
Y
2c
2b
r
C
A
a
2c
2b
r
C
A
Fig. 2. Semi-elliptical surface crack [7].

N. Shabakhty / Engineering Failure Analysis 18 (2011) 223243 225
DK
A
= Dr
total
pa
_
Y
A
(8)
DK
C
= Dr
total
pc
_
Y
C
(9)
where r
total
and two terms of Y
A
and Y
C
are given by
r
total
= r
b
r
t
; a =
r
b
r
t
(10)
Y
A
= (1=Q)
1=2
1 H
A
a
1 a
F
a
t
;
a
c
; 0;
p
2
_ _
(11)
Y
C
= (1=Q)
1=2
1 H
C
a
1 a
F
a
t
;
a
c
; 0; 0
_ _
(12)
H
A
and H
C
are the H-value taken at two points u = p/2 and u = 0, respectively. More details on F, Q and H functions are given
in Raju and Newman [9]. Either xing the ratio of a/c or expressing crack length as function of crack depth, Shang-Xian [7]
shows Eq. (5) reduce to a constant value and Eq. (6) can solve independently. This model refers to one-dimensional crack
growth and is applied in this research. For offshore tubular elements, Kirkemo [10] recommends RajuNewmans expression
with a weld toe correction factor given by Smith and Hurtworth [11] to take into account effect of discontinuity in the weld
toe due to the welding processes. The geometry function is therefore multiplication of two terms Y
unwelded
and M
k
, where
Y
unwelded
is the geometry function of a semi-elliptical crack using linear stress eld (Raju and Newman equation) with a
xed aspect ratio, and M
k
is weld toe correction factor. According to this approach, the geometry function for constant
rations of a/c and a with the value 0.15 and 5 respectively, is given through the following expressions.
Y
welded
= M
k
(a)Y
unwelded
Y
unwelded
= 1:08 0:7(a=t)
M
k
(a) = 1:0 1:24 exp(22:1(a=t)) 3:17exp(357(a=t))
(13)
Inserting Eq. (8) into expression (2) and separating the variables, the following relation is derived.
da
(Y
A
pa
_
)
m
= C
A
(Dr
total
)
m
dN (14)
Under variable amplitude loads, the integration of this equation through the thickness of element from an initial defect
size a
0
to a crack size a(t) after time t gives,
_
a(t)
a
0
da
C
A
(Y
A
pa
_
)
m
=
N(t)
i=1
S
m
i
(15)
where Dr
i,total
in the stress intensity substituted by the term S
i
and shows the stress-range magnitude of ith stress range.
Moreover, N(t) is the number of stress cycles likely to occur during the service time. The right hand side of this expression
only depends on the load factors and the left hand side relies on the crack size. Therefore we can names to the fatigue loading
function, w
L
(t) and the fatigue strength function, w
R
(t) respectively [10]
3. Long-term stress distribution
A direct procedure to calculate distribution of loading function is to simulate the stress time history and estimating all the
stress cycles by using one of existence counting methods. When the stress process is narrow-banded, we can simply use Ray-
leigh distribution to specify the stress ranges. But in the case of wide-banded stress ranges as usually occur in offshore state,
more sophisticated cyclic counting methods inevitably need to be applied. The widely applied counting method is rain-ow
counting because it gives good results on the fatigue damage prediction. However, this counting method is computationally
very expensive, much simpler approach such as peak counting should be applied in offshore and marine structures [3]. In the
peak counting, the local maximums of the stress process are of more concern and each local maximum is paired with the
local minima of the same amplitude. This assumption is conceptually validated because in a large stress process it is possible
to nd for each local maximuma local minimumof the same magnitude. Due to non-Gausian nature of wave loads and in the
following stress process, however the see-wave excitation behaves as stationary Gaussian process; it is difcult to predict
the probability density of these maximumpeaks. A methodology based on the prediction of joint probability density function
of the structural stress response and its rst and second derivatives are presented in the Gupta et al. [26]. However, the pre-
sented method shows good agreement with the more complex Monte Carlo Simulation technique, but it is more time con-
suming and expensive in fatigue reliability estimation. Hence, in the following section a simple method has been presented.
The stress cycle (S
i
) in Eq. (15) can be suppose as random variable and for a large number of stress cycles N(t), its sum-
mation approximated with its expected values as
w
L
(t) ~ E

N(t)
i=1
S
m
i
_ _
= E[N(t)[E[S
m
[ (16)
If the probability density function of the stress range is known, the expected number of stress ranges E[N(t)] and mth
value of the stress range E[S
m
] can simply be specied. In the case of peak counting method, the expected number of stress
ranges may be taken as the expected number of peaks in the stress time history, hence, E [N(t)] = m
p
t and the mth expected
stress range becomes,
E[S
m
[ =
_

0
S
m
f
S
(S) dS (17)
in which f
S
(S) is the probability density function of stress range (S) and m
p
is the expected frequency of the peaks.
An offshore platform such as jack-up rig is subjected to the stress uctuation mainly in action of loads arises from waves
and their variations during lifetime of platform. Hence, the long-term distribution of the stress ranges is of more concern. To
specify this long term, we can combine as weighted average of all short-term combinations in which each weighting term is
the ratio between the numbers of stress cycles in the actual short-term period to the long-term mean number of stress
cycles. Therefore, we can estimate the long-term distribution with
F
S
(S) =
_
H
S
_
T
Z
w(H
S
; T
Z
)F
S
(s[h
S
; t
Z
)f
H
S
;T
Z
(h
S
; t
Z
)dh
S
dt
Z
(18)
where f
Hs,Tz
(h
S
,t
Z
) is the joint density function of signicant wave height and zero-uncrossing period, F
S
(s|h
S
,t
Z
) is the short-
term stress range distribution function and w(H
S
,T
Z
) is the weighting factor [2]. This long-term distribution function is of a
somewhat complicated form and tting a simpler function to this equation is of interest. A two-parameter Weibull distribu-
tion in the following form is of more simple and easy in generally application [1],
F
S
(S) = 1 exp
S
A
_ _
B
_ _
(19)
where A and B are scale and shape parameters of distribution, respectively. Hence, using this probability distribution, the
loading function becomes,
w
l
L
(t) = m
l
tA
m
C 1
m
B
_ _
(20)
m
l
is the long-term average frequency of the stress range and estimated from
m
l
=
k
i=1
m
p
i
P
i
(21)
P
i
is the relative occurrence frequency of ith sea state and the superscript l in this formulation indicates the long-term sum-
mation [13]. In the probabilistic analysis the actual stress range could be never known exactly due to uncertainty involved in
the stress analysis. In this research to account this uncertainty in the crack growth formulation, the stress range is multiplied
by a random variable (d
S
) which is known as stress model uncertainty. Further discussion on this uncertainty will be given at
the next section.
4. Fatigue limit state function and basic random variables
To take into account uncertainties in the hydrodynamic loads, the stress calculations and the stress concentration factors,
the nal distribution of stress range are multiplied with three respectively random correction factors d
F
, d
S
and d
SCF.
Uncer-
tainty in the force calculation (d
F
) compose of contributions from wave load calculations, which partially arise from inertia
(C
M
) and drag terms (C
D
) in the Morison equation, use of simplication in the wave theory (Airy), and nally from external
loads (live and dead). There is little information about uncertainty in the force calculation therefore the lognormal distribu-
tion with a mean 1.0 and coefcient of variation 0.1 is supposed here as recommended by Dalane [2] for offshore platform
The stress model uncertainty (d
S
) represents uncertainty in the stress calculations due to the simplication of the actual
model in FEM, the soil-structure interaction, and any other aspect in the structural analysis. Sometimes this uncertainty
is combining with the force model uncertainty.
The lognormal distribution is usually applied to describe the probability distribution of this random variable with the
mean value of 1.0 as recommended by Dalane [2]. Due to lack of information about coefcient of variation of this random
variable in jack-up platforms, a value of 0.15 is taken in this research. This is higher than the value recommended by Dalane
with amount of 0.1 for jacket platform. The higher value of 0.15 is selected because of simplication in the other two legs
model applied in FEM analysis, in the soil-structure interaction as simple spring model and in other structural analysis as-
pects. Furthermore, only this variable is supposed to have the spatial correlation with other stress uncertainty in the jack-up
platform and its correlation is estimated in accordance with the time history of stress process derived in FEM analysis.
The uncertainty in the stress concentration factor (d
SCF
) arises from discrepancy between parametric formula and mea-
sured stress concentration factor. The lognormal distribution is applied to model this uncertainty with the mean value equal
1.0 but there is some scatters in the literature about its coefcient of variation, ranging from 0.05 to 0.25 [2,10,12]. However,
there is not sophisticated information about this variable for jack-up platform, therefore the value of coefcient of variation
0.1 recommended by Dalane [2] is adapted in this research. Now, by applying these random modication factors in expres-
sion (20) and omitting the super scribe letter of l for long term, the fatigue loading function is given by,
w
L
(T) = m
l
d
m
F
d
m
S
d
m
SCF
A
m
C 1
m
B
_ _
..
g
T = gT (22)
and fatigue strength function with the following expression.
w
R
(a
th
) = d
tf
_
a
th
a
0
da
Cd
m
Y
[Y(a)(
pa
_
)[
m
(23)
The time-to-failure is linked to the propagation of a crack through the thickness of element (a
th
) and a random correction
model d
tf
is applied to consider time to a section failure. A random geometry correction factor d
Y
is also incorporate in this
expression to consider uncertainties arising from simplications of geometry function, crack coalescences and effect of weld
geometry, as recommended by Moan et al. [12]. The statistics of d
Y
is invariant to the crack size, which implies that the
geometry function Y(a) randomized by simply multiplying with d
Y
. This random variable is usually modeled as a normal var-
iable with a coefcient of variation ranging from 0.05 to 0.20 [2,10,16] In the present study, the geometry function is mod-
eled in accordance with the Raju and Newman [9] approach and adding the empirical correction term given by Smith and
Hurhworth [11]. d
Y
is therefore assumed as normal distribution with the coefcient of variation 0.1.
To carry out fatigue reliability analysis, we need an expression to dene safety margin or limit between failure and safe
domain, which also is know as limit state function. In crack growth approach, this expression can be estimated by deriving
the random time-to-failure of the joint e.g. J1.
T
J1
=
w
R
(a
th
)
g
=
d
tf
J1
m
l
J1
d
m
F
J1
d
m
S
J1
d
m
SCF
J1
C
J1
A
m
J1
C(1
m
B
J1
)
_
a
thJ1
a
0J1
da
d
m
Y
J1
[Y(a)(
pa
_
)[
m
(24)
If this random time-to-failure becomes less than the lifetime of the structure (T
life
), fatigue failure is expected to occur in
this joint and vice versa, if it is greater than the expected lifetime it means that the joint is functioning well. Therefore, the
limit state function, which we are looking for can be expressed as,
g
J1
= T
J1
T
life
(25)
and its failure probability could be estimated with
P
fJ1
= P[g
J1
6 0[ = P[T
J1
T
life
6 0[ = P[T
J1
6 T
life
[ (26)
This expression will be used for each joint in the structural element and eight hot-spot points distributed regularly
around intersection of each element to determine the failure probability. Joint with the highest failure probability is likely
to be the rst joint to fail. However, this may not be generally true and other possibilities should be taken into account. The
branch and bound technique can be applied to determine the most important failure sequences of the joints under fatigue
failure sequences. The failure probability P
fJ1
is related to the reliability index b with [14],
P
fJ1
= P[g
J1
6 0[ = U(b) (27)
where U(.) is the standard normal distribution function. One of methods of FORM, SORM or Monte Carlo can be applied to
determine the failure probability. More details about calculation of reliability index or failure probability can be found in
Thoft-Christensen and Murotsu [14].
5. Fracture limit state
To establish a fracture limit state for a reliability analysis, the fracture assessment diagram (FAD) can be used to extend
the probabilistic model treating the major source of uncertainty through a set of random variables. The limit state is there-
fore formulated as,
g
f1
= K
rf
K
r
= [1 0:14L
2
r
[[0:3 0:7exp(0:65L
6
r
)[ K
r
(28)
where K
r
is a fracture parameter, which is specied based on failure assessment diagram given in BS7910 [15].
K
r
= K
p
r
K
s
r
q
c
=
K
p
I
K
s
I
K
IC
q
c
=
Y(a)
pa
_
(r
p
r
s
)
K
IC
q
c
(29)
in which K
I
is the linear elastic stress intensity factor and K
IC
is the material fracture toughness, r
p
and r
s
are primary and
secondary stresses, respectively. Furthermore, q
c
is a correction factor to account for the plasticity interaction and is dened
by the following expression,
q
c
=
q L
r
6 0:8
4q(1:05 L
r
) 0:8 < L
r
6 1:05
0 1:05 < L
r
_
_
(30)
where
q = 0:1
r
s
L
r
r
p
_ _
0:714
0:007
r
s
L
r
r
p
_ _
2
0:00003
r
s
L
r
r
p
_ _
5
6 0:25 (31)
L
r
is the plastic collapse parameter and dened as the ratio between the reference section stress (r
ref
) and the yield strength
(r
y
),
L
r
=
r
ref
r
y
(32)
and K
rf
is the non-dimensional fracture parameter at failure described with expression,
K
rf
= [1 0:14L
2
r
[[0:3 0:7exp(0:65L
6
r
)[ for L
r
6 L
r;max
(33)
The failure assessment diagram of this level is shown in Fig. 3. The failure assessment diagram is a curve at K
rf
= 1 and
L
r
= 0 to K
rf
= 0 while L
r
vary from 0 to a very large value. At high values of L
r
the cracked component will be in a situation
of general yielding and may fail due to plastic collapse instead of fracture.
In fracture limit state, we need to specify the crack dimension at extreme loads. The crack dimension a(t) may be taken as
a random initial weld defect dimension a
0
used in the crack growth or more appropriately using the crack dimensions after a
service exposure of duration t when fatigue damage extended. The fracture failure occurs due to extreme primary loading.
The maximum stresses can be calculated from the maximum wave loads in a marine structure and related to the random
environmental parameters [16].
Dijkstra et al. [17] presented another method to determine the fracture limit state function in accordance with the frac-
ture results of 38 fracture experiments of tubular and wide plate elements as shown in Fig. 3. They recommend using the
circular limit state to represent the fracture failure appropriately, where the angle of spreading is independent and only
the radius of a circle would be a relevant parameter to represent the failure mode. The fracture limit state is therefore de-
scribed with,
g
f2
= R
f
R
act
(34)
where R
f
is the radius of the fracture failure. They also recommend to suppose distribution of this variable with the lognor-
mal type in such as way that its mean and standard deviation are 1.7 and 0.4 respectively [17]. R
act
is the acting fracture
radius and specied through the following expression depend on the plastic collapse parameter L
r
and the fracture param-
eter K
r
, with the following expression.
R
act
=
K
2
r
L
2
r
_
(35)
Kr
0
0
1 2
3
Lr
1
2
3
wide plate
tubulars
Rfracture
FAD BS 7910
R
f
r
a
c
t
u
r
e
Fig. 3. Dijkstra method to specify fracture limit state [17].

Joint Committee of Structural Safety (JCSS) recommends this approach in the draft of fracture probabilistic analysis of
metallic structures, however a lower standard deviation as well as 0.306 is supposed. In this research we use this method
in fracture limit state.
6. Sequences of joints failures in fatigue
The next step is to establish a formulation for the next joint to fail when the rst failure occurs. In this case, when the rst
joint, which fails due to fatigue degradation, is J1 and the relevant element is disconnected from the structure, the next joint
that fails subsequently might be a joint J2. The total random time to reach the sequence failure of joint J2 can therefore be
divided into the time when the rst joint is in an intact state and reaches to failure, T
J1
, and the time of failure of joint J2, i.e.
T
J2fJ1
T
J1
, see Fig. 4.
In terms of the linear damage accumulation model for fatigue damage, the joint J2 has a fatigue strength function like
expression 5.2, but the fatigue loading function is the combination of two terms. The rst one is the fatigue loading function
in joint J2 when joint J1 is in an intact state and reaches to failure and the next one from the failure of joint J1 up to failure J2.
The total fatigue loading function can therefore be estimated from the following expression,
w
LJ2
(T
J2
) = w
LJ2
(T
J1
) w
LJ2fJ1
(T
J2fJ1
T
J1
) (36)
where w
LJ2
(TJ2) is the total fatigue loading function of joint J2; w
LJ2
(TJ1) and w
LJ2fJ1
(T
J2fJ1
T
J1
) are the fatigue loading func-
tions of joint J2 before and after failure of joint J1 respectively.
By using equation (36) as a loading function and applying the same modication factors to consider the uncertainties in
the hydrodynamic load, the stress concentration factor and the time to section failure, the following expression can be gen-
erated to relate the fatigue strength function and loading function of joint J2.
d
tfJ2
_
a
thJ2
a
0J2
da
d
m
Y
J2
[Y(a)(
pa
_
)[
m
= m
l
J2
d
m
F
J2
d
m
S
J2
d
m
SCF
J2
C
J2
A
m
J2
C 1
m
B
J2
_ _
T
J1
m
l
J2fJ1
d
m
F
J2
d
m
S
J2fJ1
d
m
SCF
J2
C
J2
A
m
J2fJ1
C 1
m
B
J2fJ1
_ _
(T
J2fJ1
T
J1
)
(37)
The left hand side of this expression is the fatigue strength function and the right hand side shows the total fatigue load-
ing at the time of failure. The modication d
tfJ2
is to consider the correction of section failure. This modication is therefore
modeled independent of the loading function i.e. does not change when the fatigue loading function changes. The time-
to-failure of J2 followed by J1 can therefore be expressed by reformulating expression (37) as,
T
J1
T
J2 f J1
J1
J2
Fig. 4. Sequence of fatigue failure of joint J2 followed by J1.
Fig. 5. An overview of Neka (Iran Khazar) jack-up platform in the operation [13].
T
J2fJ1
=
d
tf
J2
_
a
thJ2
a
0J2
da
d
m
Y
J2
[Y(a)(
pa
_
)[
m m
l
J2
d
m
F
J2
d
m
S
J2
d
m
SCF
J2
C
J2
A
m
J2
C 1
m
B
J2
_ _
T
J1
m
l
J2fJ1
d
m
F
J2
d
m
S
J2fJ1
d
m
SCF
J2
C
J2
A
m
J2fJ1
C 1
m
B
J2fJ1
_ _ T
J1
(38)
and the failure probability can be calculated for the sequence failure of joint J2 followed by J1 with,
P
fJ2fJ1
= P[(T
J2fJ1
6 T
life
) (T
J1
6 T
life
) (T
J1
6 T
J2
)[ (39)
The last term in the right hand side of this expression is added to conrm the sequence of J2 followed by J1. This condi-
tion, however, can easily be released without introducing an important error. The only consequence is that the failure do-
main is enlarged with a part of another failure scenario, but this will automatically be accounted for in the system
analysis. So, for reason of simplicity, (39) replace by the following expression.
P
fJ2fJ1
~ P[(T
J2fJ1
6 T
life
) (T
J1
6 T
life
)[ (40)
To nd a formulation for the higher sequences of a joint failure, the same approach as for two-sequence failures can be
utilized. Assume that the failure of joint Jn is followed by the failure of joints J1, J2, . . . , Jn 1 respectively, the loading func-
tion can then be specied by the combination of all relevant loading functions in the sequence of the failures respectively,
i.e.,
w
L
(T
Jn
) = w
LJn
(T
J1
) w
LJnfJ1
(T
J2fJ1
T
J1
) w
LJnfJ1;J
2
;...;Jn1
(T
JnfJ1;J2;...;Jn1
T
Jn1fJ1;J2;...;Jn2
) (41)
Using this loading function and applying same modication factors to represent uncertainties in load, stress and concen-
tration factor, following general equation can be derived for the time-to-failure of joint Jn when failure of joints
J1, J2, . . . , Jn 1 have already occurred respectively,
T
JnfJ1;...;Jn1
=
d
tfJn
_
a
thJn
a
0Jn
da
d
m
Y
Jn
[Y(a)(
pa
_
)[
m

n1
i=1
w
LJ
n
fJ1;J
2
;...;Ji
(T
JifJ1;J2;...;Ji1
T
J
i1
fJ1;J2;...;Ji2
)
m
l
JnfJ1;...;Jn1
d
m
F
Jn
d
m
S
JnfJ1;...;Jn1
d
m
SCF
Jn
C
Jn
A
m
JnfJ1;...;Jn1
C 1
m
B
JnfJ1;...;Jn1
_ _ T
Jn1fJ1;...;Jn2
(42)
Fatigue loading function for this sequence of joint failure can be given with,
w
LJnfJ
1
;J
2
;...Ji
= m
l
JnfJ1;J2;...;Ji
d
m
FJn
d
m
S
JnfJ1;J2;...;Ji
d
m
SCF
Jn
C
Jn
A
m
JnfJ1;J2;...;J
i
C 1
m
B
JnfJ1;J2;...;Ji
_ _
(43)
Finally, failure probability of the sequences of k joints i.e. J1, J2, J3, . . . , Jk can be specied with intersection of all possible
component failures i.e.,
P
fJkfJ1;J2;...;Jk1
= P

k
i=1
(g
JifJ1;J2;...;Ji1
6 0)
_ _
= P

k
i=1
(T
JifJ1;J2;...;Ji1
6 T
life
)
_ _
(44)
The number of joint failures that should be taken into account in this scenario depends on the redundancy of the structure
and importance of joint failures with respect to the system collapse. In a redundant structure, numerous sequences might be
supposed for system collapse. Therefore, it makes calculation of system reliability a hard computational afford. However, it is
observed that all these sequences may not participate signicantly in the nal failure probability and only some of them con-
tribute signicantly. Hence, it is recommended to use search technique to identify the important sequences leading to struc-
tural collapse [14].
7. System reliability in combinations of fatigue and fracture failure modes
In this section, a formulation for combinations of fatigue failure at one and failure due to fracture in another joint is given.
Generally, four scenarios can be distinguished as,
v Fatigue failure of rst joint followed by a fracture failure of second joint
v Fracture failure of rst joint followed by a fracture failure of second joint
v Fracture failure of rst joint followed by a fatigue failure of second joint
v Fatigue failure of rst joint followed by a fatigue failure of second joint
Formulation for fatigue failure of rst joint followed by a fatigue failure of second joint is presented in last section. A
methodology for calculation of system reliability in combination of fatigue and fracture failure mode is presented in subse-
quent sections. Finally, all sequences will be combined to determine the total system reliability.
As recommended in Shetty [27], the ratio between secondary to yield stresses (r
S
/r
Y
) in fracture formulation is 1.0 and
crack size assumed to occur due to fatigue degradation. Therefore, fatigue-fracture interaction becomes a time-dependent
problem and time-variant methodology required to solve such problem. A research work to deal with such problem is
conducted by Marley [16]. In this work, the fatigue-fracture process assumed as a crossing problem in such a way that
the structural component fail when the strength, which deteriorates due to fatigue crack growth, is crossing by a load pro-
cess in random time. The result of this method has been compared with a simplied time-invariant approach, which assumes
the resistance is equal to the strength at the end of the service life. Finally Marely [16] concludes that the approximated
time-invariant method gives close failure probability value to the more complicated time-variant method and therefore is
possible to simplify with time-invariant approach. In our research, we use the simplied time-invariant approach given
by Marely and a development to include the fracture failure of second joint in the system reliability calculation.
7.1. Fatigue failure of rst joint followed by a fracture failure of second joint
In this failure scenario, the rst joint failure takes place through fatigue and is followed by fracture failure of the second
joint. Expression (24) will be used with some modications to specify the fatigue crack size in the fracture limit state func-
tion. In this research we recommend a new approach for crack size based on fatigue degradation. In fact, a crack in a struc-
tural joint occurs from fatigue degradation and propagates with an increasing service time of the structure. Therefore, the
crack size (a) will be a random parameter because it depends on several other random variables such as fatigue character-
istics, hydrodynamic loads, stress and stress concentration factors, geometry function and nally initial crack size. Using
the crack growth formulation in fatigue limit state (Eq. (24)), the random crack size a
t
for the rst joint can be determined
with,
_
a
t
a
0
da
(d
Y
Y
A
pa
_
)
m
= C
A
m
l
sd
m
F
d
m
S
d
m
SCF
A
m
C 1
m
B
_ _
(45)
without the correction term for the section failure (d
tfJ1
). Since some of parameters are random variables, the Monte Carlo
Simulation (MCS) technique is applied to simulate these random variables. Each simulated variables are placed in this
expression and the formulation is solved to determine the characteristic of random crack size.
Applying the fatigue crack growth formulation for the second joint failure without the correction term for the section fail-
ure (d
tfJ2
), the following expression can be derived [13],
_
a(t)
a
0J2
da
d
m
Y
J2
Y(a)
m
(
pa
_
)
m
= m
l
J2
d
m
F
J2
d
m
S
J2
d
m
SCF
J2
C
J2
A
m
J2
C 1
m
B
J2
_ _
T
J1
m
l
J2fJ1
d
m
F
J2
d
m
S
J2fJ1
d
m
SCF
J2
C
J2
A
m
J2fJ1
C 1
m
B
J2fJ1
_ _
(T
life
T
J1
) (46)
The MCS has been carried out for several elements and eight hot-spot points for each end of the member of the structure.
The characteristics of the tted lognormal distribution on crack size are determined using the rst and second moment of the
statistics resulting from the MSC technique for each hot spot. Since, in the simulation of a fatigue crack size only the cases
where T
J1
is less than the lifetime of a platform is taken into account, its probability will be conditional failure probability.
The probability of failure for the combination of fatigue and fracture can therefore be estimated with.
P
f;J1fJ2
= P[(g
fJ2
6 0) (T
J1
6 T
life
)
[P[T
J1
6 T
life
[ (47)
The second expression in this formulation is the fatigue failure event of the rst joint, [T
J1
6 T
life
] and the conditional
expression shows the subsequent fracture failure event [(g
fJ2
6 0)|(T
J1
6 T
life
)].
To specify the time of the rst element failure in fatigue, the formulation (23) is recalled here. The fracture limit state
function can be described by two expression (28) and (34). In the following sections the formulation (34) will be applied
because it gives a little more conservative failure probability than expression (28). However, as is shown in Shabakthy
[13], the maximum difference between results of these two methods is less than 17%.
7.2. Fracture failure of rst joint followed by a fracture failure of second joint
In this failure scenario the rst joint failure occurs in fracture and followed by fracture failure of the second joint. For the
rst joint failure a methodology described in Section 5 can be applied here. However, determination of crack size in the frac-
ture limit is essential. In this situation, we can use formulation (45) when time parameter is adjusted to lifetime or an indi-
vidual service time of structure under investigation.
After failure of the rst joint in fracture, the crack size for the second joint failure has been computed in accordance with
the fracture limit state function of the rst joint and the fatigue formulation of the second joint. It means rstly the charac-
teristic of random crack size is determined using MCS and fracture expression (28). Then, the simulated crack size from frac-
ture limit state can be substituted in the fatigue formulation to specify T
J1
(expression (24) without section failure model
uncertainty, d
tf
). However, only the values of random crack size which makes T
J1
less than the lifetime (T
life
) should be
collected in MCS technique. The reason is other joint should not fail before this joint. Now substituting the simulated T
J1
in expression (37), the random crack size for the second joint failure is determined. By repeating this simulation several
times, the statistics of the random crack size for the second joint failure are estimated. Now, we substitute the statistical
characteristics of random crack in the fracture limit state, it is possible to calculate the sequences of the second joint failure
in fracture. Since in this approach only the simulated values of the crack when fracture failure occurs are selected from MCS,
the calculated failure probability is a conditional failure probability. The nal failure probability of this sequence can be esti-
mated with the following formulation,
P
f;J1fJ2
= P[(g
fJ2
6 0)[(g
fJ1
6 0)[P[g
fJ1
6 0[ (48)
where the second expression is the probability of a fracture failure event of the rst joint, [g
fJ1
6 0] and the conditional
expression shows the subsequent fracture failure event, [(g
fJ2
6 0)|(g
fJ1
6 0)]. The FORM approach can be applied to estimate
the failure probability for the rst and second term of this expression.
7.3. Fracture failure of rst joint followed by a fatigue failure of second joint
In this failure scenario, rst joint failure occurs in fracture and followed by a fatigue failure of the second joint. The same
approach for rst joint failure in fracture carried out in the previous section is applied here. But, only difference is in occur-
rence of fatigue failure instead of fracture.
We use again the same expression of (38) for the second joint failure but the random time to rst joint failure is estimated
in accordance with the fracture limit state function and fatigue crack growth. It means Monte Carlo Simulation technique is
applied to simulate the crack size of the rst joint using the fracture formulation. Then, the simulated fracture crack size is
substituted in the crack growth expression for the rst joint to obtain T
J1
(expression 5.3 without section failure model
uncertainty, d
tf
). Since in this calculation, only the simulated values of the crack size when T
J1
is less than T
life
(fracture failure
occurs) are selected from MCS, the calculated failure probability is therefore the conditional failure probability. The failure
probability for this sequence can be obtained from the following expression.
P
f;J1fJ2
= P[(T
J2fJ1
6 T
life
) (g
fJ1
6 0)
[P[g
fJ1
6 0[ (49)
The second expression is the fracture failure event in the rst joint, [g
fJ1
6 0] and the conditional expression shows sub-
sequent fatigue failure event, [(T
J2fJ1
6 T
life
)|(g
fJ1
6 0)]. In this failure scenario, primary stress is supposed to be the only cor-
related random variable with the primary stress of other joints for the system calculation. Furthermore, the stress
uncertainty model in the fatigue limit state is also correlated with the primary stress in the fracture limit state.
8. Search technique and branch trees obtained for dominating failure sequences
In redundant offshore structures, innumerable failure sequences can be expected to occur but only some of them contrib-
ute signicantly to the collapse or system failure and others have a very low probability of occurring. Identication of these
important sequences is essential and the branch and bound technique have been developed in this research to distinguish
dominant failure sequences.
In order to investigate the system reliability in combination of fatigue and fracture failures, a Neka jack-up platform,
which is also known as Iran Khazar is considered here as a case study. Fig. 5 shows an overview of this platform in opera-
tional site. The fatigue damage originates in the joint intersections of elements therefore the stress distribution around each
joint should be determined. Due to memory restriction of the computer, all structural details of the platform are not modeled
and a detail of one leg shown in Fig. 6. However, the validation of the main concept of this research may not diverge signif-
icantly by this simplication. For the other two legs, the truss legs are idealized as string beam elements with the equivalent
stiffness and hydrodynamics properties described in the Site Specic Assessment of Mobile Jack-up Units, Bulletin 5-5A [28].
Adapted computer model is a three-dimensional space frame with totally 524 elements and 229 nodes. More detail about
structural properties can be found in Shabakhty [13] and Gupta et al. [26].
The direction of the dominant wave loading is longitudinal from the fore leg to the aft legs, perpendicular to the line con-
necting the two aft legs. This assumption is on the bases of the critical direction leading to structural failure [18]. The ref-
erence sea states (H
S
, T
Z
) are selected from the scatter diagram of the Hutton area [19] and the hydrodynamic loads
corresponding to each sea state are calculated with the NOSDA program [20]. However, to represent better the Gaussian nat-
ures of sea wave, a modication based on non-deterministic spectrum amplitude are applied in this program [13].
Fatigue failure usually occurs in the intersection of an element at a welding joint. For each member two potential fatigue
failures are expected at the end joints. An element fails and separates from the structure if a joint failure in any of these ends
occurs. From the nite element analysis of the structure, it is observed that structural collapse often occurs when any two
brace members of the structure in one bay or a chord member fails.
The hot spot stresses around the intersection (8 points) are computed using the time history of stress process described in
Section 3 and the long-term distribution functions of all the hot-spot points are calculated for each joint through the peak
counting method and by applying the inuence function described in Section 3. The bending to membrane stress ratios are
calculated for all hot-spot points according to the stress time history and the geometry function is modied by using this ratio
and a correction term to consider discontinuities arising from the welding toe prole, as recommended by Smith and
Hurworth [11]. The statistical characteristics of random variables are discussed in Section 4 and they are shown in Table 1.
For each member, two ends denoted as S and E for start and end is shown. Failure probability are calculated for eight hot-
spot points regularly distributed around intersection (varying from zero to 315) for each joint and the maximum one se-
lected as the critical one. For 10 highest values according to the 20-years design lifetime of platform is calculated and shown
in Table 2. In most cases the highest failure probability happens in the end of element (E) at hot-spot point position 270.
This can be attributing to the hot spot position that has a signicant combination of axial and in plane bending stresses.
Moreover, due to the symmetry shape of the brace elements in the leg, the calculated failure probability for each two
elements in one bay have the same value but this balance may not maintain after the rst joint failure. Changes in the state
of the structure due to the failure of one element redistribute stresses in other elements and in their stress uncertainty.
Therefore, uncertainty in the stress calculations is represented by a random variable for each damage state. However, the
uncertainty in the different states are assumed to be the same, e.g. in a damaged state where joints J1, J2,. . ., Jn 1 failed,
d
SJnfJ1,J2,. . .,Jn1
is still assumed lognormal with a mean value 1.0 and coefcient of variation 0.15
Fig. 6. Neka jack-up nite element model and a close up of elements below the lower guide in one leg [13].
Table 1
Characteristics of random variables in the fatigue limit state function.
Parameter Variable Distribution Mean value Cov
Initial crack size (mm) a
0
Exponential 0.11 1.00
Member thickness (mm) a
th
Normal vary 0.04
Fatigue parameter in air ln C
A
Normal 29.84 St. dev. = 0.55
Fatigue parameter in water ln C
A
Normal 31.01 St. dev. = 0.77
Section failure model d
tf
Normal 1.50 0.50
Geometry function model d
y
Normal 1.00 0.10
Load model d
F
Lognormal 1.00 0.10
Stress model d
S
Lognormal 1.00 0.15
Stress concentration model d
SCF
Lognormal 1.00 0.10
Fatigue parameter in air m Fixed 3.10
Table 2
Failure probability around the intersection of joints.
Element 0 45 90 135 180 225 270 315
284E, 286E 3.69E6 3.73E5 5.27E4 3.81E5 3.60E6 7.08E4 8.22E3 7.03E4
289E, 291E 3.31E5 1.90E4 2.07E3 2.02E4 3.84E5 2.38E3 1.97E2 2.32E3
290E, 291E 1.65E5 1.59E4 1.81E3 1.52E4 1.54E5 2.21E3 1.98E2 2.25E3
295E, 297E 6.58E5 3.16E4 3.11E3 3.11E4 6.64E5 3.59E3 2.79E2 3.57E3
296E, 298E 6.86E5 6.63E5 5.86E3 5.80E4 5.49E5 4.78E3 3.77E2 5.17E3
The correlations between random variables are an important factor, which can affect on the estimated failure probability
[14]. In this research, no correlation between two random variables of d
F
and d
SCF
either in each individual joint or between
two joints of elements in both intact and damage states are supposed. But, we put all attribute correlation in our model into
stress model uncertainty (d
S
) both at the different joints and states of the structure.
To specify this correlation, the time history of the stress process is used as an indication. Both for the intact and damaged
state of the structure, the correlation coefcient is estimated with,
q
d
SJi
;d
SJj
=
Cov[S
Ji
; S
Jj
[
Var[S
Ji
[ Var[S
Jj
[
_ (50)
where S
Ji
and S
Jj
are the hot spot stresses in two joints Ji and Jj under investigation respectively. The same expression can be
used to specify the correlation between other joints when the structure is in a damaged state. The correlation between the
stress model calculation in joint 296E270 (end of element 296 in hot spot 270) and joint 290E (correlation between d
SJ296
and
d
SJ290
) are calculated for several hot spots and sea states. The results are given in Table 3, which shows strong correlations
between them. The variation of correlation coefcient around the intersection and the sea states is less signicant. However,
a small difference is observed for higher sea states (signicant wave height 7.75 and 10.45) but since most fatigue damage
arises from lower sea states (the lower sea states have a higher occurrence probability in scatter diagrams), it can be stated
that the correlation does not vary signicantly with the wave height and therefore can sufciently be represented by only
one value. The high correlation arise from this fact that all elements in structure have close connection to each other, how-
ever, its correlation will decrease by increasing distance of elements to each other in structure. This conclusion is further-
more validated by comparing the estimated correlation coefcient for other elements in the platform in such a way the
same trend is observed. Hence, it may generally be concluded that only one value would be appropriate to represent the cor-
relation in the fatigue reliability analysis without considering any variation around the intersection or the sea state However,
for the reliability calculations in the following section, the variation of the correlation coefcient around the intersection has
been taken into account due to the possibility of changing the sign of a correlation coefcient, see Table 4. This table shows
correlation between hot spot stresses around the intersection of element 170, 298 and 52 when element 296 is in a damaged
state. We observe the correlations between stress models before and after failure of element 296 are low. Calculating the
correlation between stress uncertainties in two situations of intact and damaged state for other joints it is possible to esti-
mate the failure probability in expression (40) of the second joint when the rst joint has already failed.
Many sequences leading to the system collapse can be expected for jack-up, but the branch and bound technique iden-
ties the important sequences leading to the structural collapse. Fig. 7 shows branch tree obtained for the jack-up platform
in sequence of fatigue failure. Due to symmetric shape of K bracing elements, two important joint failures are recognized in
this branch tree. They are at the end of two elements 296 and 298 and in hot spot 270 (296E270 and 298E270) with a failure
probability of 0.037753. We should keep in mind each element has two end joints, with letter S and E standing for the start
and end of element respectively. Furthermore, the last three numbers indicate the critical position of the hot spot. Failure
sequences with the probabilities smaller than 2.0E5 are not shown in this gure. The most likely collapse to occur after
failure of element 296 is end joint of element 298 in which its probability of occurrence is 0.0037. The second most likely
failure sequence is the failure of joint at the end of element 298 followed by failure at the end of element 296. The eleven
highest failure sequences are ranked and shown in Fig. 7. There is an increase in the reliability from the rst to the second
fatigue failure and therefore there is a low probability of occurrence for the second consequent fatigue failure, even after the
rst fatigue failure occurs.
Table 3
Correlation between stress uncertainty in joint 296E270 and several hot-spot points in joint 290E for a number of sea states.
Position (h) 0 45 90 135 180 225 270 315
HS = 1.75 0.992 0.995 0.995 0.995 0.992 0.988 0.988 0.988
HS = 3.25 0.989 0.993 0.993 0.993 0.989 0.988 0.984 0.985
HS = 5.00 0.984 0.989 0.990 0.989 0.984 0.988 0.978 0.978
HS = 7.75 0.975 0.984 0.985 0.984 0.975 0.988 0.964 0.965
HS = 10.45 0.968 0.982 0.983 0.982 0.968 0.988 0.951 0.952
Table 4
Correlation between stress uncertainties in several hot-spot points around the intersection of the second joint failure when the rst joint failure is 296E270.
Position (h) 0 45 90 135 180 225 270 315
Ele170S 0.0384 0.0423 0.0422 0.0423 0.0414 0.0420 0.0421 0.0420
Ele170S 0.0414 0.0418 0.0418 0.0417 0.0387 0.0422 0.0420 0.0420
Ele298S 0.0437 0.0438 0.0436 0.0433 0.0428 0.0427 0.0429 0.0432
Ele298E 0.0433 0.0447 0.0449 0.0448 0.0432 0.0419 0.0418 0.0421
Ele52S 0.0575 0.0587 0.0609 0.0627 0.0632 0.0622 0.0602 0.0581
Ele52E 0.0637 0.0615 0.0591 0.0580 0.0583 0.0597 0.0619 0.0638
In the structural reliability, any sequences leading to the structural collapse are dened as the cut set, therefore the
event of failure of the end joint of element 296 followed by the end joint 298 constitutes the rst cut set, C
1
=
[T
J298E270fJ296E270
6 T
life
[ [T
J296E270
6 T
life
[. The most signicant cut set event can be identied through the branch tree.
We nally can combine the probability of all these cut set events. For very special cases, when all the cut sets are disjoints
i.e. no two cut sets can simultaneously occur, the system failure probability is the summation of the probabilities of all cut
set events. However, this is not the case for the system reliability calculation of the jack-up platform under investigation and
a more advanced method is required.
In terms of series and parallel systems, the sequence leading to structural collapse (each cut set) constitutes the parallel
system and the combination of all these sequences represents the series system. For the jack-up platform under investiga-
tion and based on eleven importance failure sequences leading to structural collapse, the series and parallel system model
can be shown as the conguration presented in Fig. 8, where the rst row shows the rst joint failure and the second one
belong to the second failure sequences. According to this model, the system failure event can be represented with
expression,
X
system
=
_
k
jC
k
g
j
(x) 6 0
_ _
(51)
0.001711
0.019884
0.019723
0.0082276
0.008829
0.001524
0.000754
0.037753
2
9
6
E
2
7
0
2
9
8
E
2
7
0
Intact
0.027930
0.037753
0.000492
2
9
5
E
2
7
0
290E270
289E270
2
8
4
E
2
7
0
2
8
3
E
2
7
0
2
9
5
S
0
1
7
0
E
0
56E180
55E0
0.027930
0.019884
0.019723
2
9
7
E
2
7
0
0.0082276
0.001527
0.000492
0.001711
2
9
2
E
2
7
0
2
9
1
E
2
7
0
2
8
6
E
2
7
0
2
8
5
E
2
7
0 2
9
7
S
0
1
7
2
E
0
5
7
E
1
8
0
0.008829
3.7512E-3
1.9314E-4
2
9
8
E
2
7
0
1
7
0
E
9
0
3.7803E-4
2.1552E-3
169S90
297E270
1.5638E-3
292E270
1.2500E-3
2
9
1
E
2
7
0
6.9303E-5
2.1496E-3
1
7
1
S
9
0
295E
270
3.7512E-3
1.9314E-4
296E270
172E90
[1]
[7]
[3]
[2]
[4]
1.1819E-4
164E270
2.9424E-4
1
6
3
S
90
[8]
7.0865E-5
295E270
5.8556E-5
2
9
7
E
2
7
0
2.5958E-5
2
9
8
S
2
7
0
7.0865E-5
5.8556E-5
2.5958E-5
297E270
295E
270
296S270
3.6183-5
2.4574E-5
6.9396E-5
1
7
2
E
0
2
9
8
E
2
7
0
296E270
7.0754E-5
4.6461E-5
7.0825E-5
2.4947E-5
298E270
2
9
6
E
2
7
0
1
7
2
E
0
1
7
0
E
0
1.2500E-3
2.9431E-4
165S90
2
8
9
E
2
7
0
1.5638E-3
2
9
0
E
2
7
0
1.1828E-4
166E270
[9]
[10]
[5]
[6]
[11]
0.001711
0.019884
0.019723
0.0082276
0.008829
0.001524
0.000754
0.037753
2
9
6
E
2
7
0
2
9
8
E
2
7
0
Intact
0.027930
0.037753
0.000492
2
9
5
E
2
7
0
290E270
289E270
2
8
4
E
2
7
0
2
8
3
E
2
7
0
2
9
5
S
0
1
7
0
E
0
56E180
55E0
0.027930
0.019884
0.019723
2
9
7
E
2
7
0
0.0082276
0.001527
0.000492
0.001711
2
9
2
E
2
7
0
2
9
1
E
2
7
0
2
8
6
E
2
7
0
2
8
5
E
2
7
0 2
9
7
S
0
1
7
2
E
0
5
7
E
1
8
0
0.008829
3.7512E-3
1.9314E-4
2
9
8
E
2
7
0
1
7
0
E
9
0
3.7803E-4
2.1552E-3
169S90
297E270
1.5638E-3
292E270
1.2500E-3
2
9
1
E
2
7
0
6.9303E-5
2.1496E-3
1
7
1
S
9
0
295E
270
3.7512E-3
1.9314E-4
296E270
172E90
[1]
[7]
[3]
[2]
[4]
1.1819E-4
164E270
2.9424E-4
1
6
3
S
90
[8]
7.0865E-5
295E270
5.8556E-5
2
9
7
E
2
7
0
2
9
8
S
2
7
0
7.0865E-5
5.8556E-5
2.5958E-5
297E270
295E
270
296S270
3.6183-5
2.4574E-5
6.9396E-5
1
7
2
E
0
2
9
8
E
2
7
0
296E270
7.0754E-5
4.6461E-5
7.0825E-5
2.4947E-5
298E270
2
9
6
E
2
7
0
1
7
2
E
0
1
7
0
E
0
1.2500E-3
2.9431E-4
165S90
2
8
9
E
2
7
0
1.5638E-3
2
9
0
E
2
7
0
1.1828E-4
166E270
[9]
[10]
[5]
[6]
[11]
65S90 9
Fig. 7. Branch tree for whole fatigue failure sequences.
where C
k
is the cut set number k. In this expression, due to simplication some intersections of surviving events of remaining
members are incorporated. The nal system failure probability is approximated with,
P
f;system
= P(X
sys
) = P
_
k
C
k
_ _
(52)
The nal system failure probability can be evaluated according to the following steps. First, calculating the failure prob-
ability of each parallel subsystem and then determining correlation between them. Finally evaluate the probability of series
system of parallel subsystem. The equivalent linear safety margin for parallel subsystems (cut sets) is estimated using the
method described in Gollwitzer and Rackwitz [23]. To calculate the multi-normal probability in a series system, two
approaches are exercised here. At rst, the rst-order approximation method proposed by Hohenbichler and Rackwitz
[21] is applied. In this method, the dimension of the multi-normal probability is reduced in each step by one using the con-
cept of conditional multiplication. Due to the linearization technique applied in this method, the rst-order error might be
quite signicant in the case of a high correlation between random variables, therefore Tang and Melchers [22] recommend
improving the rst-order multi-normal integral. They use the standard bi-variate normal integral to calculate the conditional
distribution. These approaches are called here Crude First Order Multi-Normal (Crude-FOMN) and Improved First Order
Multi-Normal (Improved-FOMN) respectively. The system reliability index or the failure probability is calculated based on
these two methods and given in Table 5. The results show that the improved method gives higher reliability Indies but
differences between them are not so signicant.
An alternative approach to the system reliability calculation of a series system is the bound technique. The bound tech-
nique is principally derived from the general inclusionexclusion rule of a probability calculation. The computation of the
probability of the intersection event is difcult for a large number of components. Because of this difculty, there has been
continues interest developing bounds on the systemprobability that employ the marginal component probability or the joint
probability i.e. the bi-component probability or tri-component probability. For series systems, the probability bound using
uni-component probability is often too wide for practical application; therefore, Ditlevsen [24] provides a better bound by
involving bi-component in addition to the uni-component. The upper and lower bound are depending on the ordering of the
cut sets and the order should be changed in such a way that the maximumof the lower bound and the minimumof the upper
bound are ensuing. The results of the uni-component bounds and the Ditlevsen bounds for the series system shown in Fig. 8
are calculated and given in Table 5. The uni-component bounds give wider ranges than the Ditlevsen bounds as already
expected. Moreover, the Ditlevsen bounds are close to the result of the crude-FOMN and improved-FOMN. Discrepancy
between results of the FOMN approaches and Ditlevsen bounds can apparently relate to the approximation applied in FOMN
approach for the linearization of non-linear limit state and correlation between two linearized limit stat functions, however,
the differences are not so signicant as can be observed.
If we assume the system failure probability of a jack-up structure as the result of improved-FOMN, it can state the system
failure probability of platforms is lower than the rst component failure in fatigue and the system effect is of more impor-
tance. This issue can relate to the redundancy of structures. After failure of the rst element due to fatigue, the environmen-
tal loads will transmit by frame action of the remaining elements and hence signicant increasing in stress state of surviving
elements is expected However, because of the rst failure may typically occur in a diagonal or a horizontal brace, increase in
the stress parameters is not too high and some extra time needs before additional member failure. Thus, to have a sequence
of two fatigue failures occurring subsequently, rst failure must occur early (earlier than the second one) and the second
failure must occur in the interval time between rst failure and the lifetime of platform. This by far has less probability
of occurrence than just a single element failure. Using the formulation given in Section 7.1 and estimating failure probability
in expression (47), we can establish branch tree for combination of rst failure in fatigue and second failure in fracture. The
[2]
298E270
296E270
[1]
296E270
298E270
[3]
295E270
297E270
[4]
297E270
295E270
[7]
290E270
292E270
[5]
56E180
[6]
57E180
[8]
292E270
290E270
[9]
289E270
291E270
[10]
291E270
289E270
[11]
55E0
[2]
298E270
296E270
[2]
298E270
296E270
[1]
296E270
298E270
[1]
296E270
298E270
[3]
295E270
297E270
[3]
295E270
297E270
[4]
297E270
295E270
[4]
297E270
295E270
[7]
290E270
292E270
[7]
290E270
292E270
[5]
56E180
[5]
56E180
[6]
57E180
[6]
57E180
[8]
292E270
290E270
[8]
292E270
290E270
[9]
289E270
291E270
[9]
289E270
291E270
[10]
291E270
289E270
[10]
291E270
289E270
[11]
55E0
[11]
55E0
Fig. 8. System failure events as series and parallel system through eleven important fatigue failure sequences.
Table 5
System reliability and failure probability of jack-up platform in entirely sequences of fatigue failures.
Crude Improved Uni-component bounds Ditlevsen bounds
FOMN FOMN Upper Lower Upper Lower
P
f
0.019484 0.016548 0.021406 0.003751 0.019516 0.014704
b 2.0645 2.1309 2.0255 2.6737 2.0638 2.1779
results obtained for such scenario is presented in Fig. 9. To estimate the failure probability of the rst and second term in
expression (47), FORM approach is employed but the crack size needed in the conditional expression is determined based
on Monte Carlo Simulation. A detail of procedure is explained in Shabakhty [13]. Table 6 shows characteristics of the random
variables applied in the fracture limit state function.
Several hot-spot points around the intersection of an element are susceptible to the fracture failure. We supposed eight
points regularly distributed around the intersection and the failure probability is individually calculated for each one. The
highest one therefore selected as the critical one. The results for some critical elements are shown in Table 7. Results derived
for the case that the rst failure occurs at the end of element 296 due to the fatigue damage and at the hot-spot point 270.
For both ends of element, failure probabilities are estimated and the highest one is selected as the critical one and is
3.8970E-3
1
7
0
E
9
0
(F
ra
c
)
[1]
0.019884
0.019723
0.022843
0.008829
0.010875
0.027210
0.037753
2
9
6
E
2
7
0
(
F
a
t
)
2
9
8
E
2
7
0
(
F
a
t
)
Intact
0.027930
0.037753
2
9
5
E
2
7
0
(F
a
t)
290E270 (Fat)
289E270 (Fat)
2
9
5
E
2
7
0

(
F
r
a
c
)
2
8
3
E
2
7
0
(
F
a
t
)
5
3
E
1
8
0
(F
rac)
289E270(Frac)
0.027930
0.019884
0.019723
2
9
7
E
2
7
0
(
F
a
t
)
0.001711
0.010673
0.001711
2
9
2
E
2
7
0

(
F
a
t
)
2
9
1
E
2
7
0

(
F
a
t
)
5
6
E
18
0
(F
at)
2
8
5
E
2
7
0

(
F
a
t
)
54E180 (Frac)
5
7
E
1
8
0

(
F
a
t
)
0.008829
2
8
4
E
2
7
0

(
F
a
t
)
0.0082276
2
9
7
E
2
7
0

(
F
r
a
c
)
0.022306
2
9
1
E
2
7
0
(
F
r
a
c
)
0.026589
3.5794E-3
1.3792E-3
7.5082E-4
2
95
E
27
0 (F
rac)
53E180 (Frac)
297E270 (Frac)
1.1745E-3
5.5947E-4
1.1279E-4
2
9
7
E
2
7
0
(
F
r
a
c
)
5
3
E
1
8
0
(
F
ra
c
)
54E135(Frac)
9.9181E-4
1.0728E-4
291E270(Frac)
50E180(Frac)
2.1605E-3
3.8567E-4
291E270(Frac)
50E180(Frac)
6.8178E-4
3.4029E-3
289E270(Frac)
164E90(Frac)
3.8297E-3
3.5161E-3
1.3492E-3
7.3281E-4
1
72
E
90
(F
rac)
297E270 (Frac)
54E0 (Frac)
295E270 (Frac)
9.7154E-4
1.0424E-4
289E270(Frac)
51E0(Frac)
1.1496E-3
5.4607E-4
1.0950E-4
295E270(Frac)
54E0(Frac)
53E45(Frac)
2.1237E-3
3.7638E-4
289E270(Frac)
51E0(Frac)
6.6681E-4
3.3526E-3
291E270(Frac)
166E90(Frac)
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[11]
[12]
[13]
[14]
[9]
[10]
3.8970E-3
1
7
0
E
9
0
(F
ra
c
)
[1]
0.019884
0.019723
0.022843
0.008829
0.010875
0.027210
0.037753
2
9
6
E
2
7
0
(
F
a
t
)
2
9
8
E
2
7
0
(
F
a
t
)
Intact
0.027930
0.037753
2
9
5
E
2
7
0
(F
a
t)
290E270 (Fat)
289E270 (Fat)
2
9
5
E
2
7
0

(
F
r
a
c
)
2
8
3
E
2
7
0
(
F
a
t
)
5
3
E
1
8
0
(F
rac)
289E270(Frac)
0.027930
0.019884
0.019723
2
9
7
E
2
7
0
(
F
a
t
)
0.001711
0.010673
0.001711
2
9
2
E
2
7
0

(
F
a
t
)
2
9
1
E
2
7
0

(
F
a
t
)
5
6
E
18
0
(F
at)
2
8
5
E
2
7
0

(
F
a
t
)
54E180 (Frac)
5
7
E
1
8
0

(
F
a
t
)
0.008829
2
8
4
E
2
7
0

(
F
a
t
)
0.0082276
2
9
7
E
2
7
0

(
F
r
a
c
)
0.022306
2
9
1
E
2
7
0
(
F
r
a
c
)
0.026589
3.5794E-3
1.3792E-3
7.5082E-4
2
95
E
27
0 (F
rac)
53E180 (Frac)
297E270 (Frac)
1.1745E-3
5.5947E-4
1.1279E-4
2
9
7
E
2
7
0
(
F
r
a
c
)
5
3
E
1
8
0
(
F
ra
c
)
54E135(Frac)
9.9181E-4
1.0728E-4
291E270(Frac)
50E180(Frac)
2.1605E-3
3.8567E-4
291E270(Frac)
50E180(Frac)
6.8178E-4
3.4029E-3
289E270(Frac)
164E90(Frac)
3.8297E-3
3.5161E-3
1.3492E-3
7.3281E-4
1
72
E
90
(F
rac)
297E270 (Frac)
54E0 (Frac)
295E270 (Frac)
9.7154E-4
1.0424E-4
289E270(Frac)
51E0(Frac)
1.1496E-3
5.4607E-4
1.0950E-4
295E270(Frac)
54E0(Frac)
53E45(Frac)
2.1237E-3
3.7638E-4
289E270(Frac)
51E0(Frac)
6.6681E-4
3.3526E-3
291E270(Frac)
166E90(Frac)
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[11]
[12]
[13]
[14]
[9]
[10]
Fig. 9. Branch tree for fatigue failure of the rst joint and fracture failure of the second joint.
Table 6
Characteristics of random variables in the fracture limit state.
Parameter Variable Distribution Mean value Cov
Primary stress (kN/m
2
) r
P
Gumbel Vary Vary
Fracture toughness (kN/m
2

mm
_
) K
IC
Lognormal 6.5E6 0.25
Fatigue crack size (mm) a Lognormal vary vary
Geometry function model d
y
Normal 1.00 0.10
Load model d
F
Lognormal 1.00 0.10
Yield stress (kN/m
2
) r
y
Lognormal 6.85E5 0.08
Stress concentration model d
SCF
Lognormal 1.00 0.10
attributed to that element. All random variables except the primary stress in the fracture limit state function are indepen-
dent. This correlated primary stress is calculated based on the extreme environmental wave load that may expect occur dur-
ing the return period, say here 50 years. But, the stress history in the fatigue limit state is through the long-term combination
of the several short-term stress histories according to the scatter diagram. The important sequences of this failure scenario
are found using branch tree method. The branch tree is established through the failure probability determined by expression
(47) for the combination of fatigue and fracture failure modes and results of failure probability higher than 1.0E4 is shown
in Fig. 9.
The most important failure sequence is fatigue failure of the joint at the end of element 296 (hot spot 270) followed by
fracture failure of end of element 170 (hot-spot point 90). The highest fourteen failure sequences identied through this
branch tree are ranked in Fig. 9 and the system reliability calculation is estimated according to these sequences. Table 8
shows results according to two methods of crude-FOMN and improved-FOMN. Also, the results of uni-bounds and Ditlevsen
bounds are presented in this table as well. The FOMN give results within uni-component bounds but crude-FOMN is outside
of the Ditlevsen bounds. However, it is inside the Uni-component bounds. Since the improved-FOMN is within the Ditlevsen
bounds, this failure probability can be supposed as the system failure probability of the jack-up platform for the combination
of fatigue and fracture failure modes with a value of 0.03178.
Another failure scenario taken into account in this research is wholly failures in fractures. To establish a branch tree for
combination of rst and second failure in fracture, formulation given in Section 7.3 are take here into account again and
branch tree result is shown in Fig. 10. Now, applying the FORM approach, the failure probability for the rst and second joint
failures in fracture is estimated with help of expression (48). For simulation of random crack size needed in fracture limit
state, we apply the well known Monte Carlo Simulation. For simulation of crack size in the rst joint failure, uses has done
by fatigue function but for the second joint failure, the combination of fracture and fatigue expressions are applied. More
detail about the method of simulation is given in Shabakhty [13] as well. The primary stress is the only random variable cor-
related with the primary stress of other joints The primary stress in the fracture limit state is therefore calculated according
to the extreme environmental wave loads that can be expected during the design return period of structure here 50-year.
The calculated failure probabilities for this failure scenario are presented in Fig. 10 for several important sequences leading
to the structural collapse. Failure probabilities less than 1.0E6 are not taken into account in this gure
The rst and second important failure sequences are the fracture failure of the chord elements 53 and 54 with the failure
probabilities 0.0108 and 0.0106 respectively. The highest six failure sequences identied through this branch tree are ranked
in Fig. 10 and the system reliability calculation is computed in accordance with these sequences and is given in Table 9. It is
clear from result that the reliability index for this failure scenario is higher than that provided by previous scenario.
The last failure scenario is the rst joint in fracture which will follow by fatigue failure of the second joint. The same ap-
proach for the rst joint failure in fracture given in last section will be carried out here again. Only difference is the second
failure, which is assumed to occur in fatigue instead of fracture. For the second joint failure, the same expression for fatigue
failure 37 is used here once more but the random time to the rst joint failure is estimated in accordance with the fracture
function and fatigue crack growth. It means MCS is applied to simulate the crack size of the rst joint using the fracture for-
mulation, then simulated fracture crack size is substituted in the crack growth expression for the rst joint to obtain T
J1
. The
stress uncertainty model in the fatigue limit state is also correlated with the primary stress in the fracture limit state. The
branch tree has been established and the result is shown in Fig. 11 for the failure probability higher than 1.0E4. The highest
fourteen failure sequences identied through this branch tree are ranked in this gure as well. The system reliability calcu-
lation is carried out in accordance with these sequences are given in Table 10. As is clear from Fig. 11, one of important fail-
ure sequences is fracture failure of bracing element 289 followed by fatigue failure of 290.
Comparing calculated system failure probabilities in Table 10 shows Ditlevsen bounds give closer results than the uni-
component bounds with the values varying between 0.0724 and 0.0820. The result of crude-FOMN is out of the Ditlevsen
bounds but the improved-FOMN is within these bounds and can be supposed as the nal system failure probability for this
Table 7
The failure probability of the second element failure in fracture when the rst failure is in fatigue at the hot-spot point of 270 in the end of element 296. The
term S stands for start and E for the end of the element.
Element 169 170 172 295 297 53
H.S. point 90 270 315 90 270 270
P
f
(S) 2.1624E6 9.8906E7 1.6628E9 3.6049E6 5.8564E5 3.5074E4
H.S. point 270 90 0 270 270 180
P
f
(E) 1.1016E7 3.8970E3 7.5210E5 7.5082E4 3.5794E3 1.3792E3
Table 8
System failure probability for the combination of sequences of fatigue failure of the rst joint followed by fracture failure of the second joint.
P
f
0.03306 0.03178 0.033802 0.003897 0.03258 0.03161
b 1.8376 1.8553 1.8276 2.6609 1.8442 1.8577
failure scenario. If we compare this failure probability with the values calculated for other failure scenarios, it can be con-
cluded that this failure scenario gives the highest failure probability and care should be given to this failure state. However,
it should be kept in mind that the result is probably a conservative approximation because of using the simplied time-
invariant reliability approach and applying the design storm with return period 50 years.
In this section we combine all failure sequences identied in the previous sections to calculate the nal system failure
probability of the Neka jack-up platform for the combination of fatigue and fracture. The forty-one important sequences
identied in the previous sections are shown in Fig. 12. The nal system failure probability is based on the failure probability
calculated for each of the failure sequence and its correlation, which is estimated between the randomvariables and the limit
state functions. The results are given in Table 11 in accordance with the FOMN approaches and Ditlevsen bound. The crude-
FOMN and improved-FOMN results are within the Ditlevsen bounds. The Ditlevsen bounds give narrower results than the
uni-component bound. The nal system failure probability for the jack-up structure can be supposed as the result of im-
proved-FOMN with the value 0.1143. It is clear that this value is higher than the rst joint failure in fatigue or fracture alone
and care should be given to this type of failure in the structural analysis of jack-up platforms, when ageing due to fatigue is
taken into account. Because of this high probability of failure occurrence, it shows why the institutes giving certicate to
marine operation companies enforce to inspection regularly during lifetime of the platform.
9. Concluding remarks
The system reliability of jack-up platform under several failure scenarios leading to the structural collapse has been esti-
mated. At rst, a methodology for fatigue sequences leading to structural collapse has been presented to take into account
0.022306
0.019884
0.019723
0.022843
0.008829
0.010875
0.027210
0.037753
2
9
6
E
2
7
0
(
F
a
t
)
2
9
8
E
2
7
0
(
F
a
t
)
Intact
0.027930
0.037753
2
9
5
E
2
7
0
(
F
a
t)
290E270
(Fat)
289E270 (Fat)
2
9
5
E
2
7
0
(
F
r
a
c
)
2
8
3
E
2
7
0
(
F
a
t
)
5
3
E
1
8
0
(F
rac)
289E
270(F
rac)
0.027930
0.019884
0.019723
2
9
7
E
2
7
0
(
F
a
t
)
0.0082276
0.010673
0.001711
2
9
2
E
2
7
0
(
F
a
t
)
2
9
1
E
2
7
0
(
F
a
t
)
2
8
6
E
2
7
0
(F
at)
2
8
5
E
2
7
0
(
F
a
t
)
54E180 (Frac)
5
7
E
1
8
0
(
F
a
t
)
0.008829
2
8
4
E
2
7
0
(
F
a
t
)
0.0082276
2
9
7
E
2
7
0
(
F
r
a
c
)
2
9
1
E
2
7
0
(
F
r
a
c
)
0.026589
3.8314E-5
5
3
E
1
8
0
(F
ra
c
)
2.1461E-5
54E180 (Frac)
1.4467E-4 297E270 (Frac)
2.4768E-5
1.9764E-4
291E270 (Frac)
5
0
E
1
8
0
(F
rac)
8.1114E-6
51S135
(Frac)
3.6226E-5
5
4
E
0
(F
ra
c
)
2.0359E-5
53E0 (Frac)
1.3734E-4 295E270 (Frac)
2.3394E-5
1.8784E-4
289E270 (Frac)
5
1
E
0
(F
rac)
7.6387E-6
50S45
(Frac)
[1]
[2]
[3]
[4]
[5]
[6]
0.022306
0.019884
0.019723
0.022843
0.008829
0.010875
0.027210
0.037753
2
9
6
E
2
7
0
(
F
a
t
)
2
9
8
E
2
7
0
(
F
a
t
)
Intact
0.027930
0.037753
2
9
5
E
2
7
0
(
F
a
t)
290E270
(Fat)
289E270 (Fat)
2
9
5
E
2
7
0
(
F
r
a
c
)
2
8
3
E
2
7
0
(
F
a
t
)
5
3
E
1
8
0
(F
rac)
289E
270(F
rac)
0.027930
0.019884
0.019723
2
9
7
E
2
7
0
(
F
a
t
)
0.0082276
0.010673
0.001711
2
9
2
E
2
7
0
(
F
a
t
)
2
9
1
E
2
7
0
(
F
a
t
)
2
8
6
E
2
7
0
(F
at)
2
8
5
E
2
7
0
(
F
a
t
)
54E180 (Frac)
5
7
E
1
8
0
(
F
a
t
)
0.008829
2
8
4
E
2
7
0
(
F
a
t
)
0.0082276
2
9
7
E
2
7
0
(
F
r
a
c
)
2
9
1
E
2
7
0
(
F
r
a
c
)
0.026589
3.8314E-5
5
3
E
1
8
0
(F
ra
c
)
2.1461E-5
54E180 (Frac)
1.4467E-4 297E270 (Frac)
2.4768E-5
1.9764E-4
291E270 (Frac)
5
0
E
1
8
0
(F
rac)
8.1114E-6
51S135
(Frac)
3.6226E-5
5
4
E
0
(F
ra
c
)
2.0359E-5
53E0 (Frac)
1.3734E-4 295E270 (Frac)
2.3394E-5
1.8784E-4
289E270 (Frac)
5
1
E
0
(F
rac)
7.6387E-6
50S45
(Frac)
[1]
[2]
[3]
[4]
[5]
[6]
Fig. 10. Branch tree for combination of rst and second joints failures in fracture.
Table 9
System failure probability for combination of rst and second failure sequences in fracture.
P
f
0.02208 0.02207 0.0221 0.01087 0.02208 0.02207
b 2.0126 2.0127 2.0125 2.2947 2.0126 2.0127
fatigue degradation up to the time of fatigue failure. Then, for fatigue and fracture failure sequences of the rst and second
joint respectively, an approach based on degradation up to the time of failure is presented. For fracture reliability analysis, a
simple but time-invariant approach based on extreme stresses estimated from a design storm is offered. The referenced time
applied for the fracture and fatigue reliability calculation is the lifetime of the structure. However, the provided method for
system reliability calculation can extend to other service time e.g. one or four years. Each failure sequences events leading to
the structural collapse in the branch tree are modeled as parallel subsystems and the nal structural failure events form the
series system, which combines all the parallel subsystems. Four scenarios for combination of fatigue and fracture leading to
0.019884
0.019723
0.022843
0.008829
0.010875
0.027210
0.037753
2
9
6
E
2
7
0
(
F
a
t
)
2
9
8
E
2
7
0
(
F
a
t
)
Intact
0.027930
2
9
5
E
2
7
0
(F
a
t)
290E270
(Fat)
289E270 (Fat)
2
9
5
E
2
7
0
(
F
r
a
c
)
2
8
3
E
2
7
0
(
F
a
t
)
5
3
E
1
8
0
(
F
rac
)
289E270(
Frac
)
0.027930
0.019884
0.019723
2
9
7
E
2
7
0
(
F
a
t
)
0.0082276
0.010673
0.001711
2
9
2
E
2
7
0
(
F
a
t
)
2
9
1
E
2
7
0
(
F
a
t
)
2
8
6
E
2
7
0
(F
at)
2
8
5
E
2
7
0
(
F
a
t
)
54E180 (Frac)
5
7
E
1
8
0
(
F
a
t
)
0.008829
2
8
4
E
2
7
0
(
F
a
t
)
0.0082276
2
9
7
E
2
7
0
(
F
r
a
c
)
0.022306
2
9
1
E
2
7
0
(
F
r
a
c
)
0.026589
5.7779E-3
7.6704E-3 290E270 (Fat)
2
9
2
E
2
7
0
(F
a
t)
5.6476E-3
2
9
1
E
2
7
0
(
F
a
t
)
2.0308E-3
1.1604E-3
165S135 (Fat)
1
6
3
S
1
8
0
(
F
a
t
)
4.2997E-3
297E270 (Fat)
5.4948E-3
2
9
8
E
2
7
0
(
F
a
t
)
7.5078E-3
296E270 (Fat)
1.7022E-3
1
7
0
E
0
(
F
a
t
)
1.1397E-3
1
6
9
S
1
8
0
(
F
a
t
)
7.9239E-4
1
7
2
E
0
(F
a
t)
6.9662E-4
1
7
1
S
1
3
5
(F
at)
3.5631E-4
5
2
E
4
5
(
F
a
t
)
3.0928E-4
54E180 (Fat)
8.0536E-4
6.3524E-4
2.9889E-4
1
6
4
E
0
(
F
a
t
)
1
6
6
E
1
8
0
(F
at)
5
0
S
1
8
0
(
F
a
t
)
.
.
4.1389E-3
295E270
(Fat)
5.2966E-3
2
9
6
E
2
7
0
(
F
a
t
)
7.2510E-3
298E270 (Fat)
1.6310E-3
1
7
2
E
1
8
0
(
F
a
t
)
1.0901E-3
1
7
1
S
0
(
F
a
t
)
7.5683E-4
1
7
0
E
1
8
0
(F
a
t)
6.65025E-4
169S45 (Fat)
3.3930E-4
5
2
E
1
3
5
(
F
a
t
)
2.9436E-4
5
3
E
0
(F
a
t)
5.5691E-3
7.4058E-3
292E270 (Fat)
290E270 (Fat)
5.4428E-3
2
8
9
E
2
7
0
(
F
a
t
)
1.9472E-3
1.1100E-3
1
6
5
S
0
(
F
a
t
)
7.6921E-4
6.062E-4
2.8446E-4
1
6
6
E
1
8
0
(
F
a
t
)
1
6
4
E
0
(
F
a
t
)
5
1
S
0
(
F
a
t
)
1
6
3
S
0
(F
a
t)
0.037753
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
0.019884
0.019723
0.022843
0.008829
0.010875
0.027210
0.037753
2
9
6
E
2
7
0
(
F
a
t
)
2
9
8
E
2
7
0
(
F
a
t
)
Intact
0.027930
2
9
5
E
2
7
0
(F
a
t)
290E270
(Fat)
289E270 (Fat)
2
9
5
E
2
7
0
(
F
r
a
c
)
2
8
3
E
2
7
0
(
F
a
t
)
5
3
E
1
8
0
(
F
rac
)
289E270(
Frac
)
0.027930
0.019884
0.019723
2
9
7
E
2
7
0
(
F
a
t
)
0.0082276
0.010673
0.001711
2
9
2
E
2
7
0
(
F
a
t
)
2
9
1
E
2
7
0
(
F
a
t
)
2
8
6
E
2
7
0
(F
at)
2
8
5
E
2
7
0
(
F
a
t
)
54E180 (Frac)
5
7
E
1
8
0
(
F
a
t
)
0.008829
2
8
4
E
2
7
0
(
F
a
t
)
0.0082276
2
9
7
E
2
7
0
(
F
r
a
c
)
0.022306
2
9
1
E
2
7
0
(
F
r
a
c
)
0.026589
5.7779E-3
7.6704E-3 290E270 (Fat)
2
9
2
E
2
7
0
(F
a
t)
5.6476E-3
2
9
1
E
2
7
0
(
F
a
t
)
2.0308E-3
1.1604E-3
165S135 (Fat)
1
6
3
S
1
8
0
(
F
a
t
)
4.2997E-3
297E270 (Fat)
5.4948E-3
2
9
8
E
2
7
0
(
F
a
t
)
7.5078E-3
296E270 (Fat)
1.7022E-3
1
7
0
E
0
(
F
a
t
)
1.1397E-3
1
6
9
S
1
8
0
(
F
a
t
)
7.9239E-4
1
7
2
E
0
(F
a
t)
6.9662E-4
1
7
1
S
1
3
5
(F
at)
3.5631E-4
5
2
E
4
5
(
F
a
t
)
3.0928E-4
54E180 (Fat)
8.0536E-4
6.3524E-4
2.9889E-4
1
6
4
E
0
(
F
a
t
)
1
6
6
E
1
8
0
(F
at)
5
0
S
1
8
0
(
F
a
t
)
.
.
4.1389E-3
295E270
(Fat)
5.2966E-3
2
9
6
E
2
7
0
(
F
a
t
)
7.2510E-3
298E270 (Fat)
1.6310E-3
1
7
2
E
1
8
0
(
F
a
t
)
1.0901E-3
1
7
1
S
0
(
F
a
t
)
7.5683E-4
1
7
0
E
1
8
0
(F
a
t)
6.65025E-4
169S45 (Fat)
3.3930E-4
5
2
E
1
3
5
(
F
a
t
)
2.9436E-4
5
3
E
0
(F
a
t)
5.5691E-3
7.4058E-3
292E270 (Fat)
290E270 (Fat)
5.4428E-3
2
8
9
E
2
7
0
(
F
a
t
)
1.9472E-3
1.1100E-3
1
6
5
S
0
(
F
a
t
)
7.6921E-4
6.062E-4
2.8446E-4
1
6
6
E
1
8
0
(
F
a
t
)
1
6
4
E
0
(
F
a
t
)
5
1
S
0
(
F
a
t
)
1
6
3
S
0
(F
a
t)
0.037753
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
Fig. 11. Branch tree obtained for the combination of the rst joint failure in fracture followed by the second joint failure in fatigue.
Table 10
System failure probability for combination of sequences of fracture failure of the rst joint followed by fatigue failure of the second joint.
P
f
0.0825 0.0760 0.08916 0.01087 0.08208 0.07242
b 1.3885 1.4320 1.3459 2.2947 1.3912 1.4580
the structure collapse considered in this research. At rst, wholly combination of fatigue failures with the probability of
occurrence of 0.016548 is estimated. Next, the rst joint in fatigue and second joint failure in fracture with probability of
0.03178 the rst joint in fracture and second joint in fatigue with probability 0.076 and nally wholly failure in fractures
with the value 0.02207 are obtained. The result shows rst joint failure in fracture followed by the second joint failure in
fatigue scenario has the highest value comparing to others and this scenario should be taken into account, when failure
of structure is of concern. The failure probability of the structural system in the combination of fatigue and fracture failure
modes calculated through the forty-one signicant failure sequences identied in the branch tree leading to the structural
collapse. The nal failure probability estimated using FOMN and Ditlevsen bounds. The results show signicant system
effects and Ditlevsen bounds give close failure probability for the lower and upper bound with values 0.10406 and
0.13037 respectively. The improved-FOMN result is within the Ditlevsen bounds and expects as the nal system failure prob-
ability for the whole combination of fatigue and fracture with a value of 0.1143. The calculated system failure probability in
0.019884
0.019723
0.022843
0.008829
0.010875
0.027210
0.037753
2
9
6
E
2
7
0
(
F
a
t
)
2
9
8
E
2
7
0
(
F
a
t
)
Intact
0.027930
2
9
5
E
2
7
0
(F
a
t)
290E270 (Fat)
289E270 (Fat)
2
9
5
E
2
7
0
(
F
r
a
c
)
2
8
3
E
2
7
0
(
F
a
t)
5
3
E
1
8
0
(F
rac)
289E270(Frac)
0.027930
0.019884
0.019723
2
9
7
E
2
7
0
(
F
a
t
)
0.0082276
0.010673
0.001711
2
9
2
E
2
7
0
(
F
a
t
)
2
9
1
E
2
7
0
(
F
a
t
)
2
86
E
2
70
(F
at)
2
8
5
E
2
7
0

(
F
a
t
)
54E180 (Frac)
5
7
E
1
8
0
(
F
a
t
)
0.008829
2
8
4
E
2
7
0
(
F
a
t
)
0.0082276
2
9
7
E
2
7
0
(
F
r
a
c
)
0.022306
2
9
1
E
2
7
0
(
F
r
a
c
)
0.026589
5.7779E-3
7.6704E-3
290E270 (Fat)
292E270
(Fat)
5.6476E-3
2
9
1
E
2
7
0
(F
a
t)
4.2997E-3
297E270 (Fat)
5.4948E-3
298E270 (Fat)
7.5078E-3
2
9
6
E
2
7
0
(F
at)
1.4467E-4
2
9
7
E
2
7
0
(F
ra
c
)
1.9764E-4
291E270
(Frac)
. 4.1389E-3
295E270 (Fat)
5.2966E-3
2
9
6
E
2
7
0
(F
a
t)
7.2510E-3
2
9
8
E
2
7
0
(F
a
t)
1.9764E-4
295E270 (Frac)
5.5691E-3
7.4058E-3
292E270
(Fat)
290E270 (Fat)
5.4428E-3
2
8
9
E
2
7
0
(F
a
t)
1.8784E-4
289E270 (Frac)
0.037753
3.8970E-3
3.7512E-3
.
2
9
8
E
2
7
0
(F
at)
1.3792E-3
3.5794E-3
170E90 (Frac)
53E180 (Frac)
2
9
7
E
2
7
0
(
F
r
a
c
) 2.1552E-3 297E270 (Fat)
1.1745E-3
297E270 (Frac)
2.1496E-3
1.1496E-3
2
9
5
E
2
7
0
(F
at)
295E270 (Frac)
.
1.2500E-3
291E270
(Fat)
1.5638E-3
292E270 (Fat)
1.2500E-3
289E270 (Fat)
3.4029E-3
164E90 (Frac)
2.1605E-3
2
9
1
E
2
7
0
(F
rac)
.
3.3526E-3
2.1237E-3
1.5638E-3
2
9
0
E
2
7
0
(F
at)
289E270 (Frac)
166E90 (Frac)
.
3.8297E-3
3.5161E-3
1.3492E-3
3.7512E-3
2
9
5
E
2
7
0
(F
ra
c
)
1
7
2
E
9
0
(
F
r
a
c
)
296E270
(Fat)
54E0 (Frac)
0.001711
5
6
E
1
8
0
(
F
a
t
)
0.000754
5
5
E
0
(
F
a
t
)
[17]
[15]
[32]
[19]
[39]
[3]
[7]
[8]
[6]
[40]
[14]
[12]
[5]
[9]
[41]
[11]
[18]
[33]
[20]
[16]
[22]
[25]
[29]
[26]
[37]
[35]
[28]
[30]
[21]
[23]
[36]
[27]
[1]
[2]
[24]
[34]
[31]
[4]
[13]
[10]
[38]
0.019884
0.019723
0.022843
0.008829
0.010875
0.027210
0.037753
2
9
6
E
2
7
0
(
F
a
t
)
2
9
8
E
2
7
0
(
F
a
t
)
Intact
0.027930
2
9
5
E
2
7
0
(F
a
t)
290E270 (Fat)
289E270 (Fat)
2
9
5
E
2
7
0
(
F
r
a
c
)
2
8
3
E
2
7
0
(
F
a
t)
5
3
E
1
8
0
(F
rac)
289E270(Frac)
0.027930
0.019884
0.019723
2
9
7
E
2
7
0
(
F
a
t
)
0.0082276
0.010673
0.001711
2
9
2
E
2
7
0
(
F
a
t
)
2
9
1
E
2
7
0
(
F
a
t
)
2
86
E
2
70
(F
at)
2
8
5
E
2
7
0

(
F
a
t
)
54E180 (Frac)
5
7
E
1
8
0
(
F
a
t
)
0.008829
2
8
4
E
2
7
0
(
F
a
t
)
0.0082276
2
9
7
E
2
7
0
(
F
r
a
c
)
0.022306
2
9
1
E
2
7
0
(
F
r
a
c
)
0.026589
5.7779E-3
7.6704E-3
290E270 (Fat)
292E270
(Fat)
5.6476E-3
2
9
1
E
2
7
0
(F
a
t)
4.2997E-3
297E270 (Fat)
5.4948E-3
298E270 (Fat)
7.5078E-3
2
9
6
E
2
7
0
(F
at)
1.4467E-4
2
9
7
E
2
7
0
(F
ra
c
)
1.9764E-4
291E270
(Frac)
. 4.1389E-3
295E270 (Fat)
5.2966E-3
2
9
6
E
2
7
0
(F
a
t)
7.2510E-3
2
9
8
E
2
7
0
(F
a
t)
1.9764E-4
295E270 (Frac)
5.5691E-3
7.4058E-3
292E270
(Fat)
290E270 (Fat)
5.4428E-3
2
8
9
E
2
7
0
(F
a
t)
1.8784E-4
289E270 (Frac)
0.037753
3.8970E-3
3.7512E-3
.
2
9
8
E
2
7
0
(F
at)
1.3792E-3
3.5794E-3
170E90 (Frac)
53E180 (Frac)
2
9
7
E
2
7
0
(
F
r
a
c
) 2.1552E-3 297E270 (Fat)
1.1745E-3
297E270 (Frac)
2.1496E-3
1.1496E-3
2
9
5
E
2
7
0
(F
at)
295E270 (Frac)
.
1.2500E-3
291E270
(Fat)
1.5638E-3
292E270 (Fat)
1.2500E-3
289E270 (Fat)
3.4029E-3
164E90 (Frac)
2.1605E-3
2
9
1
E
2
7
0
(F
rac)
.
3.3526E-3
2.1237E-3
1.5638E-3
2
9
0
E
2
7
0
(F
at)
289E270 (Frac)
166E90 (Frac)
.
3.8297E-3
3.5161E-3
1.3492E-3
3.7512E-3
2
9
5
E
2
7
0
(F
ra
c
)
1
7
2
E
9
0
(
F
r
a
c
)
296E270
(Fat)
54E0 (Frac)
0.001711
5
6
E
1
8
0
(
F
a
t
)
0.000754
5
5
E
0
(
F
a
t
)
[17]
[15]
[32]
[19]
[39]
[3]
[7]
[8]
[6]
[40]
[14]
[12]
[5]
[9]
[41]
[11]
[18]
[33]
[20]
[16]
[22]
[25]
[29]
[26]
[37]
[35]
[28]
[30]
[21]
[23]
[36]
[27]
[1]
[2]
[24]
[34]
[31]
[4]
[13]
[10]
[38]
Fig. 12. Branch tree obtained for the combination of the important sequences in fatigue and fracture failure of the rst and the second joints.
Table 11
System failure probability for all combination of fracture and fatigue in the rst and second failures.
P
f
0.12719 0.11443 0.13641 0.01087 0.13037 0.10406
b 1.1398 1.2033 1.0966 2.2947 1.1246 1.2588
combination of fatigue and fracture is higher than the sequences in combination of fully system failure only in fatigue or
fracture. This high probability of failure in fatigue and fracture shows why the regulation societies enforce the offshore com-
pany to inspect their platforms during the lifetime of structures.
Acknowledgment
Author of this paper appreciates the guidance of Prof. Hengst and Ir. Boonstra from marine engineering faculty of TU Delft.
Also, cooperation and advice of Associate Prof. P.V. Gelder and Prof. A.C.W.M. Vrouwenvelder both from civil engineering fac-
ulty of TU Delft during author stay in Netherlands are acknowledged.
References
[1] Det Norsk Veritas. Strength analysis of main structures of self elevating units, Classication note no. 31.5, Norway, 1992.
[2] Dalane JI. System reliability in design and maintenance of xed offshore structures. Ph.D. Thesis, Department of Marine Structures, The Norwegian
Institute of Technology, Trondheim, Norway, 1993.
[3] Jensen JJ. Fatigue damage due to non-gaussian response. J Eng Mech 1990;116(1):2405.
[4] Karunakaran D. Nonlinear dynamic response and reliability analysis of drag-dominated offshore platforms. Ph.D. Thesis, Department of Marine
Structures, The Norwegian institute of Technology, Trondheim, Norway, November, 1993.
[5] Sobczyk K, Spencer BFJ. Random fatigue, from data to theory. Academic press; 1992. ISBN: 0-12-654225-2.
[6] Paris PC, Erdogan F. A critical analysis of crack propagation laws. J Basic Eng ASME 1963;85:52834.
[7] Shang-Xian W. Shape change of surface crack during fatigue growth. Eng Fract Mech 1985;22(5):897913.
[8] Raju IS, Newman JC. An empirical stress intensity factor equation for the surface crack. Eng Fract Mech 1981;15:18592. J.C.
[9] Raju IS, Newman JC. Stress intensity factor for circumferential surface crack in pipes and rods under bending and tension loads. Fract Mech, ASTM
1986;17:789805.
[10] Kirkemo F. Application of probabilistic fracture mechanics to offshore structures. Appl Mech Rev, ASME 1988;41(2):6184.
[11] Smith IJ, Hurworth SJ. The effect of geometry changes upon the predicted fatigue strength of welded joints. Res. Report No. 244, Welding Inst,
Cambridge, England, 1984.
[12] Moan T, Hovde GO, Blanker AM. Reliability-based fatigue design criteria for offshore structures considering the effect of inspection and repair. In:
Proceedings of the 25th annual offshore technology conference (OTC), Paper no. 7189, Houston, USA, May 36, 1993.
[13] Shabakhty N. durable reliability of jack-up platforms, the impact of fatigue, fracture and effect of extreme environmental loads on the structural
reliability. Ph.D. Thesis, Faculty of Mechanical Engineering and Marine Technology, Delft University of Technology, Delft, The Netherlands, 2004. ISBN:
90-3700217-X.
[14] Thoft-Christensen P, Murotsu Y. Application of structural system reliability theory. Springer-Verlag; 1986.
[15] BS7910. Guidance on methods for assessing the acceptability of aws in fusion welded structures. British Standard Institute, BSI, 1999.
[16] Marley MJ. Time variant reliability under fatigue degradation. Ph.D. Thesis, Department of Marine Structures, The Norwegian Institute of Technology,
Trondheim, Norway, 1991.
[17] Dijkstra OD, Van Manen SE, Gijsbers FBJ. Probabilistic maintenance planning for the tubular joints in the steel gates in the Eastern Scheldt storm surge
barrier. International Journal of TNO Building and Construction Research and the School for Advanced Studies in Construction in The Netherlands
(HERON) 1994;39(2):3563.
[18] SNAME, Guidelines for site specic assessment of mobile jack-up units. Society of Naval Architects and Marine Engineers, Technical and Research
Bulletin 5-5A, New Jersey, 1997.
[19] Marex. Hutton Area, main study report and addendum. Marex House, Marine Exploration Limited, UK, 1979.
[20] Liu P, Massie WW. NOSDA nonlinear dynamic analysis program for offshore structures. Research Report, Faculty of Civil Engineering, Workgroup
Offshore Technology, Delft University of Technology, December, 1988.
[21] Hohenbichler M, Rackwitz R. First-order concepts in system reliability. Struct Saf 1983;1(3):17788.
[22] Tang LK, Melchers RE. Improved approximation for multinormal integral. Struct Saf 1987;4(4):8193.
[23] Gollwitzer S, Rackwitz R. Equivalent components in rst-order system reliability. Reliab Eng 1983;5:99115.
[24] Ditlevsen O. Narrow reliability bounds for structural systems. J Struct Mech 1979;7(4):45372.
[25] Shabakhty N, Van Gelder P, Boonstra H. Reliability analysis of jack-up platforms based on fatigue degradation. In: Proceeding of the 27th international
conference on offshore mechanics and arctic engineering, OMAE 2008-57700, June 1520, Estoril, Portugal, 2008.
[26] Gupta S, Shabakhty N, Van Gelder P. Fatigue damage in randomly vibrating jack-up platforms under non-Gaussian loads. Appl Ocean Res
2006;28(6):40719.
[27] Shetty N. System reliability of xed offshore structures under fatigue deterioration. Ph.D. thesis, Department of Civil Engineering, Imperial College of
Science, Technology and Medicine, London, England, 1992.
[28] Bennett WT, Site Specic Assessment of Mobile Jack-up Units. Society of Naval Architects and Marine Engineers, Technical and Research Bulletin 5-5A,
Jersey City, 1994.

System Failure Probability of Offshore Jack Up Platforms in The Combination of Fatigue and Fracture (Pub Year)

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

System Failure Probability of Offshore Jack Up Platforms in The Combination of Fatigue and Fracture (Pub Year)

Uploaded by

Copyright:

Available Formats

System failure probability of offshore jack-up platforms

in the combination of fatigue and fracture

Tel.: +98 21 33050192.

Fig. 2. Semi-elliptical surface crack [7].

Fig. 3. Dijkstra method to specify fracture limit state [17].

You might also like