Chapter 5 Reliability calculation and analysis of ship structure

5.1 General
As described in Chapter 1, the procedure of reliability calculation and analysis includes
the following main steps:
1) Identify all possible and significant failure modes of the structure or operation
under consideration.
2) Formulate a failure criterion and establish a relevant failure limit state function
for each of mode of failure.
3) Choose and identify stochastic variables and parameters for each failure mode of
the structure or operation under consideration.
4) Calculate the reliability or failure probability of the structure of each failure
mode of the structure or operation under consideration.
5) Assess the structure reliability against the given reliability target, as to whether
the calculated reliability is sufficient or not and modify the concept if necessary.
For the reliability calculation and analysis of ship structure, the following information is
required: loads, resistance and reliability methods. In chapter 4, loads have been
introduced, so this chapter will focus on resistance (strength) of ship structure,
establishment of failure mode, choice and determination of distribution types of basic
random variables, analysis methods and so on. Finally, fatigue reliability will
represented also.

5.2 Failure modes

A marine structure is a systemic and complex structure. The failure modes are various,
including yielding, buckling, fatigue, corrosion, wear, erosion, etc. There are three types
of behaviour usually considered in the analysis of ship structure, besides fatigue
behaviour.
The primary behaviour is associated with the ship as a whole. The ship is usually
considered as a beam subjected to its own weight (including cargo) and supported by
buoyancy distributed along its length. Acceleration effects and inertia loads are included
by applying the equations of motions of the ship.
The secondary behaviour is associated with a stiffened panel between bulkheads or
webframes. Stiffened plate theory is used to determine deflections and stresses in the
panel. The panel is usually subjected to in-plane loads resulting from the overall
bending of the hull.
The tertiary behaviour is associated with plates between stiffeners considered as
isotropic plates. These also are subjected to in-plane loads and, therefore, buckling
becomes an important consideration.
Fatigue of ship details is an important concern in ship design. Reliability analysis of
fatigue failure is discussed also in this chapter.
In general, if Z is a random variable function representing the total load (response) and
S is a random variable function representing the strength (ultimate moment capacity and
so on) of a ship, then the limit failure function is defined as
G(X)=S-Z

98
where S, Z refer to section 5.3 and Chapter 4, respective ly.
Failure occurs when the total applied bending moment Z exceeds the ultimate capacity
S, that is, when the G is negative. Therefore, the probability of failure is:
p f = P[G ≤ 0]

5.3 Ultimate strength of ship structure

Among the three types of behaviour in the analysis of ship structures, primary failure
modes, also called global or hull failure modes, consist of the fully plastic moment
mode, the initial yield moment mode, and the instability collapse moment mode. The
last includes buckling and post-buckling strength of the hull and is always the governing
mode of failure. The fully plastic mode gives an upper bound on the ultimate moment. It
is never attained in a hull of normal proportions. The initial yield mode assumes that
buckling does no t occur prior to yielding and is considered here only since it is a
function of the standard elastic section modulus of the ship and the yield strength of the
material, both normally used in current design practice. It should be noted, however,
that the initial yield moment is higher than the true instability collapse moment. The
secondary mode of failure relates to failure of a stiffened panel of the hull. Two main
modes of failure are possible, stiffened- induced or plate- induced failure. The tertiary
mode of failure is associated with failure of a plate between stiffeners. Each of the three
levels of structural analysis discussed above may lead to one or several failure modes.
The following is a guideline to consider all the above failure modes.

5.3.1 Overview of the calculation methods of ultimate strength of a ship structure

In order to get the ultimate strength of a ship structure, many calculation methods are
used. These methods can be classified into the following:
a) Explicit method
b) Empirical (experimental) method
c) Semi-FEM (Finite element method)
d) FEM
The above methods include the following methods:
a) Caldwell’s method
b) Smith’s method
c) Progressive collapse behaviour of hull girder under longitudinal bending
d) Finite element method: such as ASYSYS, ADINA, NASTRAN, etc.
e) Idealised structural unit method (ISUM)
The above methods each have their own advantages. Although FEM and ISUM
methods can be utilized by means of commercial computer codes, it is not convenient
for use in practical ship structure design and reliability analysis. It is also necessary to
derive a simple expression for calculation of the hull ultimate strength so that it can be
used as a design equation or failure function in reliability analysis. Classification
societies provide design criteria for structural scantlings, which are usually based on
first yielding and elastic buckling with a simple correction for plasticity. These
expressions may not be the ultimate limit state. To obtain an acceptable margin of safety
against overall hull collapse, the ultimate strength provides a more reasonable criterion
than the conventional elastic buckling or first yield criteria. The following will give
simple formulations of global hull strength, stiffened panel strength and unstiffened
panel strength, respectively.

99
5.3.2 Estimation of ultimate strength of the global hull
In practical engineering, the ultimate limit state of the hull girder for a marine structure
is usually considered as the typical failure mode of marine structures. These failure
modes of the hull girder may be described by a number of possible failure mechanisms,
including:
a) Overall stiffener induced failure
b) Plate induced failure and local buckling of the stiffener
c) Local buckling of the stiffer, etc.
Among these failure modes, the overall stiffener induced failure mode is usually
selected as the most important failure mode.
For this section, the limit state function is:
G( X ) = M u − ( M w + M sw )
where M u is ultimate hull girder moment capacity, determined by critical buckling
stress.
M w is wave bending moment.
M sw is still water bending moment.
In this section, calculation methods of M u will be introduced and a simple formulation
for estimating global hull strength will be represented.
In general, the ultimate moment can be determined by generating a moment-curvature
diagram for the hull section. In essence, this kind of method applies a known amount of
curvature to the section and then calculates the bending moment that would be required
to generate that curvature. This computation is repeated, increasing the curvature each
time, for a user-selected range of curvatures. The resulting moment-curvature data pairs
are then plotted. The ultimate moment of the section is then estimated by reading the
value off of the curve where it ‘levels out’. In other words, this is when a small increase
in applied moment yields a very large increase in curvature.
Previous studies on the development of a simple formula for the ultimate hull strength
prediction may be classified into three approaches:
1. An analytical approach based on assumed stress distribution over the hull section,
from which the moment of resistance of the hull is theoretically calculated taking
account into buckling in the compression flange and yielding in the tension flange.
2. An empirical approach where an expression is derived on basis of experimental or
numerical data from scaled full models.
3. A linear approach where the behaviour of the hull up to collapse of the
compression flange is assumed to be linear, and the ultimate moment capacity of
the hull is basically expressed as the ultimate strength of the compression flange
multiplied by the elastic section modulus with a simple correction for buckling
and yielding. The third method is quite simple, but its accuracy may not be good
because after buckling of the compression flange, the behaviour of the hull is no
longer linear and the neutral axis changes position. An empirical approach may
provide reasonable solutions for conventional hulls, but the user has to be careful
in using empirical formulas for new or general-type hulls since usually they are
derived on the basis of limited data. On the other hand, analytical formulations can
be applied to new or general hulls because they include section effects more
precisely.
The following formulation is based on the analytical approach.

100
The overall collapse of a ship’s hull under a vertical bending moment is governed by
collapse of the compression flange. Also, there is still some reserve strength beyond
collapse of the compression flange. The Figure 2.2.4 in the literature SSC398 shows a
credible distribution of longitudinal stresses of the hull cross-section at the overall
collapse stage. This is the basic assumption of formulation derivation in this part. It was
postulated that parts of the compressed side shells, as well as the compression flange,
will reach their ultimate limit state in compression. The tension flange will reach the
yield strength of the material, while compressed side shells in the immediate vicinity of
the final neutral axis, as well as all side shells under tension, are assumed to remain
elastic and the stress distribution there is assumed to be linear. The neutral axis location,
as well as the depth at which the stress distribution starts to become linear, can be
determined from two conditions: a) no axis force exists on the hull girder; b) the stress
distribution is linear near the neutral axis. The neutral axis has moved toward the
tension flange from its initial position in the intact hull section. The ultimate strength
moment of the hull was then calculated by integration of the assumed distribution with
respect to the neutral axis. This resulted in explicit ultimate moment expression for the
sagging and hogging conditions.
The stress distribution can be expressed by (the x- y coordinates are taken as shown in
Figure 2.2.4 of the literature SSC398):
In sagging condition [SSC-398]:
σ x = σ yB at y=0
1
=− [σ uS + σ yS ] y − H σ yS ] 0< y < H
H
= σ B' (5.3.1)
1
= − [σ uS + σ yS ] D B − Hσ yS ] at y = D B
H
= −σ uS H≤y< D
= −σ uD at y = D
where compressive stress is negative and tensile stress is positive. The relative symbols
refer to the nomenclature in the end of this section.
Similar, in hogging condition:
σ x = σ yD at y=0
1
= − [σ uS + σ yS ] y − H σ yS ] 0< y < H
H
= −σ uS H≤y< D (5.3.2)
= σ uB
'
at y = D - D B
= −σ uB at y = D
From the condition that no axial force acts on the hull girder, the depth of the collapsed
sides (D-H), can be calculated such that
∫ σ x dA = 0 (5.3.3)
Therefore, in the sagging condition
2A A'
AD ( −σ uD ) + S ( D − H )( −σ uS ) + B [ −(σ uS + σ yS ) DB + Hσ yS ] + ABσ yB
D H

101
 2 AS 1 H
D H ∫0
+ ⋅ [ −(σ uS + σ yS ) y + Hσ yS ]dy = 0 (5.3.4)

or, since H must take a positive value

C1 D + C12 D 2 + 4C 2 D
H = (5.3.5)
2
where
ADσ uD + 2 AS σ uS − ABσ yB − A'Bσ yS
C1 =
AS (σ uS + σ yS )
AB' DB
C1 =
AS
The position of the neutral axis, where the longitudinal stress is zero, can be determined
from the condition that the stress distribution is linear, namely
g = y σ x =0 (5.3.6)
Therefore, the location of the neutral axis above the base line in the sagging condition is
obtained by substituting equations. (5.3.1) and (5.3.5) into equation. (5.3.6) as
( C1 D + C12 D 2 + 4C2 D )σ yS
g= (5.3.7)
2(σ uS + σ yS )
Similarly, in the hogging condition, the depth of the collapsed sides under compression,
(D-H), can be obtained from
2A
AB ( −σ uB ) + S ( D − H )( −σ uS ) + AB' ( −σ uB ) + ADσ yD
D
2 AS 1 H
D H ∫0
+ ⋅ [ −(σ uS + σ yS )Y + Hσ yS ]dy = 0 (5.3.8)

or
ABσ uB + AB' σ uB
'
+ 2 ASσ uS − ADσ yD
H =D (5.3.9)
AS (σ uS + σ yS )
The neutral axis below the deck in the hogging condition can also be obtained by
substituting equation (5.3.9) into equation (5.3.6)
ABσ uBσ yS + AB' σ uB
'
σ yS + 2 AS σ uSσ yS − ADσ yD σ yS
g=D (5.3.10)
AS (σ uS + σ yS ) 2
The ultimate moment capacity of the hull under sagging bending moment is [SSC-398]:
2A D + H − 2g
M us = AD ( D − g )( −σ uD ) + S ( D − H ) ( −σ uS )
D 2
A'
+ AB ( − g )σ yB + B ( g − DB )[(σ uS + σ yS ) DB − Hσ yS ]
H
2 AS 1 H
D H ∫0
+ [ − (σ uS + σ yS ) y + Hσ yS ]( y − g ) dy (5.3.11)

or

102
 AS
M us = − AD ( D − g )σ uD − ( D − H )( D + H − 2 g )σ uS − AB gσ yB
D
A'B
+ ( g − DB )[( DBσ uS − ( H − DB )σ yS ]
H
A H
+ S [( 2H − 3g )σ uS − ( H − 3g )σ yS ] (5.3.12)
3D
with H and g defined by equation (5.3.4) and (5.3.7), respectively.
Similarly, in the hogging condition, the ultimate moment capacity of the hull is given by
2A D + H − 2g
M uh = AB ( D − g )σ uB + S ( D − H ) σ uS
D 2
+ AB' ( D − g − DB )σ uB + AD gσ yD ]
'

2 AS 1 H
D H ∫0
- [ − (σ uS + σ yS ) y + Hσ yS ]( y − g ) dy (5.3.13)

or
M uh = AD gσ yD + AB ( D − g )σ uB + AB' ( D − g − DB )σ uB
'

AS
+ ( D − H )( D + H − 2 g )σ uS
D
A H
+ S [( 2H − 3 g )σ uS − ( H − 3 g )σ yS ] (5.3.14)
3D
with H and g defined by equations (5.3.8) and (5.3.10), respectively.
To calculate the ultimate moment capacity of the hull using equation (5.3.12) or
(5.3.14), the ultimate strength of the compression flange and the sides in the vicinity of
the compression flange, which are usually stiffened panels, must be known.
Theoretically, the possible failure modes of a stiffened panel under compressive loads
can be divided into three classes (Smith, 1977):
a) Local collapse of plate between stiffeners
b) Overall collapse plate with longitudinal and transverse stiffeners
c) Torsional/ flexural buckling of stiffener with effective plating
The collapse of a stiffened panel will occur at the lowest value of the ultimate load
calculated from 2 and 3 of the above three collapse patterns (Mansour, 1980, 1986;
Ueda, 1995). Calculation of the ultimate strength considering all possible modes is not
an easy task. For practical purposes, therefore, a number of simple fo rmulas have been
suggested. One promising formula is recommended. On the basis of existing and new
collapse test results for a total number of 130 stiffened panels with appropriate values of
initial imperfections, Paik and Lee (1995) derived an empirical formula for the ultimate
compressive strength of a stiffened panel as a function of the plate slenderness ratio
β and the column (stiffened) slenderness ratio λ , namely
σu
= ( 0.995 + 0.936λ2 + 0.1710β 2 + 0.188λ2 β 2 − 0.067λ4 ) −0.5 (5.3.15)
σy
where the terms including λ should be removed, i.e., λ =0, for application to
unstiffened plates.
For calculation of the hull ultimate strength using equations (5.3.12) or (5.3.14), a
designer needs to know in adva nce the ultimate strengths of the compression flange, as
well as the sides in the vicinity of the compression flange.

103
The following are the relative nomenclature:
AB = Total sectional area of outer bottom
AB ' = Total sectional area of inner bottom
AD = Total sectional area of deck
AS = Half-sectional area of all sides (including longitudinal
bulkheads and inner sides)
as = Sectional area of a longitud inal stiffener plating
b = Breadth of plate between longitudinal stiffeners
D = Hull depth
DB = Height of double bottom
E = Young’s modulus
g = Neutral axis position above the base line in the sagging
condition or below the deck in the hogging condition
H = Depth of hull section in linear elastic state
Is = Moment of inertia of a longitudinal stiffer with effective plating
l = Length of a longitudinal stiffener between transverse beams
ME = Elastic bending moment
Mp = Fully plastic bending moment of hull section
Mu = Ultimate bending moment capacity of hull section
M uh , M us = Ultimate bending moment on hogging or sagging conditions
r = Radius of gyration of a longitudinal stiffer with effective
plating [ r = ( I s / as ) 0.5 ]
t = Plate thickness
Z = Elastic section modulus at the compression flange
ZB , ZD = Elastic section modulus at bottom or deck

β b σ
= Slenderness ratio of plate between stiffeners [ β = ⋅ ( y ) 0.5 ]
t E
Slenderness ratio of a longitudinal stiffener with effective
λ = l σ y 0.5
plating [ λ = ⋅( ) ]
πr E
σy = Yield strength of the material
σ yB , σ yB
' = Yield strength of outer bottom, inner bottom
σ yD , σ yS = Yield strength of deck, side
σu = Ultimate buckling strength of the compression flange
σ uB , σ uB
' = Ultimate buckling strength of outer bottom, inner bottom
σ uD , σ uD = Ultimate buckling strength of deck, side

5.3.3 Estimation for a stiffened panel ultimate strength

The stiffened panel is the most commonly used structural element in ship, appearing in
decks, bottoms, bulkheads, and side shells. It absorbs out of plane (lateral) loads and
distributes those loads to the ship’s primary structure. It also carries part of the

104
longitudinal bending stress because of the orientation of the stiffeners. The amount of
in-plane compression or tension experienced depends primarily on the location of the
panel. Deck panels tend to experience large in-plane loads and small lateral pressures, if
any. Bottom panels experience large in-plane loads, but usually with significant lateral
pressures.
The definition of a stiffened gross panel, for this work, is a panel of plating which has
stiffeners running in two orthogonal directions. This panel is bounded by other
structure, which has significantly greater stiffness in the planes of the loads when
compared to the panel and its stiffeners. These boundaries would be provided by
structure, such as transverse bulkheads, longitudinal bulkheads, side shell, or large
longitudinal girders.
The collapse of a stiffened panel can be prevented by choosing the size of the transverse
stiffeners so that they provide sufficient flexural rigidity to enforce nodes at the location
of the transverse stiffeners. If the transverse stiffeners act as nodes, which is usually the
case, then the collapse of the stiffened panel is controlled by the strength of the
longitudinally-stiffened panel. The strength of a longitudinally-stiffened panel is usually
governed by the strength of its stiffness together with the effective plating. The effective
plating is determined from buckling considerations if the plate is under edge
compressive stress, or from shear lag analysis, if the stiffened plate is subjected to
lateral load. Only ultimate strength limit state is considered since, when a column
buckles, it reaches immediately its ultimate strength, in most cases.
For this section, the limit state function is:
G( X ) = σ u − σ
where σ u is ultimate strength of a longitudinal stiffened panel.
σ is stress in practical structure under practical loads.
Under loads acting on the ship, the stress σ can be determined by:
Hogging:
M + k wM w
σ = sw
SM b
Sagging:
− M sw + k w ( M w + k d M d )
σ =
SM b
where SM b is the appropriate section modulus.
M d is a slamming- induced moment.
k w is a factor to combine the wave-dynamic composite moments with the
stillwater moment.
k d is a factor to combine the wave- induced and dynamic moments.
It should be noticed that since the slamming- induced moment is a sagging moment, it is
included only when the sagging loading condition is considered.

The effective plating under edge compression can be determined from:

σ 
be = b ul 
 σ0 

105
 σ
where  ul σ  is to be determined from:
 0

b. For a b ≥ 1.0
 σ  0.5
 cr  if β ≥ 3.5
 σ 0 
σ ul  2.25 1.25
= − 2 if 1.0 ≤ β < 3.5
σ0  β β


1.00 if β < 1.0
where
σ cr 4π 2 1 3.612
= ⋅ 2 =
σ 0 12(1 − υ ) β2
β2
c. For a b < 1.0
2
σu  1 
= αCu + 0.08(1 − α )1 + 2  ≤ 1.0
σ0  β 
where
 4π 2 1 
0. 5

 ⋅ 2  if β ≥ 3.5
 12(1 − υ ) β 
2


2.25 1.25
Cu =  − 2 if 1.0 ≤ β < 3.5
 β β


1.00 if β < 1.0
The relevant notations refer to the list of nomenclature at the end of this section.
For plates under lateral load, the effective plating can be determined from shear lag
design curves such as those presented by Schade (1951) and Mansour (1970).
The strength of a longitudinal stiffened panel is governed as follows:

1) Column Buckling
The strength in this mode is given by:
 π 2E
if σ cr ≤ σ P

( )
 lr
σ cr = 
2

 1
σ 0 − C if σ cr > σ P
 s

σs π 2E
Cs = σs =
σ p (σ 0 − σ p ) (l r )
2

2) Beam-Column buckling

106
If a stiffener is subjected to axial stresses and lateral load that include a moment M, the
following relationship provides the strength as a beam-column:
σ M
+C =1
σ cr Mu
where M u is the moment at which the flanges are fully plastic, and
Cm
C= ≥ 1.0
 σ
1−  σ  
 cr 

M
C m = 0.6 + 0.4 1 ≥ 0.4
M2
M1 M1
M 2 > 0 for single curvature bending, and M 2 < 0 for double curvature bending.

3) Torsional / Symmetric Sections

a. Doubly-symmetric sections
In this case, the shear and the centroid of the section coincide. Therefore, the torsional
and flexural modes are decoupled.
1  π 2 ECw 
   GJ + 2
 if σ ct ≤ σ P
 
I 0 l 
  σ 
σ ct =  σ p (1 − p σ ) 
σ 1 -
 0
0
 if σ ct > σ P
σ
  t 
  
where
1 π 2 ECw 

σ t =  GJ + 
I0  l 2 

b. Section with a single plane of symmetry

In this section, the shear and centroid of the section do not coincide and the ultimate
limit state is governed by a combination of torsion and flexural buckling.
(i) Elastic range
σ tfe ≤ σ p
σ tfe is the smallest root of the following equation:
Ic 2
σ − σ tfe (σ cr + σ t ) + σ cr σ t = 0
I 0 tfe
(ii) Plastic range
σ tfe > σ p
 σp 
 σ p (1 − σ 0 ) 
σ tfp = σ 0 1 − 
 σ tfe 
 

107
The following are the relative nomenclature of section 5.3.3 and 5.3.4:
Material
E = Modulus of elasticity
G = Shear modulus
υ = Poisson’s ratio
σ0 = Average yield stress in compression
σ0
τ = , Average yield stress in shear
3
σp = Proportional (linear elastic) limit stress in compression; may
be taken as 60% of σ 0
σp
yp = , Ratio indicating the start of non- linear behaviour
σ0
Plate between stiffeners
a = Plate length
b = Plate width
t = Plate thickness
α = a / b aspect ratio; can be smaller or larger than 1
β
b σ
= ; slenderness ratio
t E
Stiffeners
Ac = Cross-sectional area (including effective plating)
Aw = Web area
b = Spacing between stiffeners
be = Effective width of plating
bf = Flange width
Cw = Warping constant
d = Web depth
I = Moment of inertia (including effective plating)
Ic = Polar moment of inertia about centroid
I0 = Polar moment of inertia about shear centre
I fc = Moment of inertia of compression flange in lateral bending
I ft = Moment of inertia of tension flange in lateral bending
I1 = Moment of inertia in the plane of minimum bending
I2 = Moment of inertia in the plane of maximum bending
J = Torsional constant
l = Stiffener’s length between transverse girders
r = Radius of gyration
S = Section modulus
t = Plate thickness

108
 tf = Flange thickness
tw = Web thickness
Applied loads and stresses
p = Applied uniform pressure normal to plate
σ = Normal stress
σ x , σ y = Normal stress in the x and y directions
τ = Applied edge shear stress
q = Load per unit length
Critical and ultimate stresses
σ cr = Critical buckling compressive stress
σ ul = Ultimate compressive stress
τ ul = Ultimate shear stress (acting alone)
Bending moment and axial loads
M = Applied bending moment
M0 = Fully plastic bending mome nt
M1 = Smaller end moment in the plane of bending
M2 = Larger end moment in the plane of bending
M fy = Moment at which the flange are fully plastic
My = Moment at which yield first occurs
Mu = Ultimate limit state
P = Applied axial force
P0 = P0 = σ 0 Ac ; Fully plastic axial force

5.3.4 Estimation for an unstiffened panel’s ultimate strength

For this section, the limit state function is:
G( X ) = σ u − σ
where σ u is ultimate strength of a unstiffened panel.
σ is stress in practical structure under practical loads.
Under loads acting on the ship, the stress σ can be determined by:
Hogging:
M + k wM w
σ = sw
SM d
Sagging:
− M sw + k w ( M w + k d M d )
σ =
SM d
where SM b is the appropriate section modulus.
M d is a slamming- induced moment.
k w is a factor which combines the wave-dynamic composite moments with the
stillwater moment.

109
 k d is a factor which combines the wave- induced and dynamic moments.
It should be noticed that since the slamming- induced moment is a sagging moment, it is
included only when the sagging loading condition is considered.
The strength of an unstiffened plate subjected to an in-plane load acting on the ‘b’ edge
of the plate is given (refer to SSC-398):
a. For a / b ≥ 1.0
 σ cr
 if β ≥ 3.5
 σ 0

σ ul  2.25 1.25
= − 2 if 1.0 ≤ β < 3.5
σ0  β β


1.0 if β < 1.0
where
σ cr 4π 2 1 3.612
= ⋅ 2 =
σ 0 12(1 − υ ) β 2
β2

b. For a / b < 1.0

σu  1 
= αCu + 0.08(1 − α )1 + 2  ≤ 1.0
σ0  β 
where
 π
 if β ≥ 3.5
 β 3(1 − υ 2
)
 2.25 1.25
Cu =  − 2 if 1.0 ≤ β < 3.5
 β β


1.0 if β < 1.0
where the above nomenclatures refer to the nomenclatures at the end of section 5.3.3.

5.4 limit state equations of ship structure

According to the above situations of strength and loads, in general, there are eight limit
equations for each ship. The same limit state equations are used for both the short-term
and long-term time frames. Thus, for each failure mode, there is one equation for the
hogging loading condition and one for the sagging condition. There are four failure
modes – for a total of eight equations. If all situations need to be done for research that
includes short-term and long term, sagging and hogging, calculate 16 results of
probabilities of a ship structure. These equations are shown in section 5.3. Table 5.4.1 is
a summary of these equations.

5.5 Analysis of uncertainties and choice of basic variables and their distributions
In theory, all variables should be considered as stochastic variables, but in practical
engineering, designers and analysts always choose certain variables as stochastic
variables that affect the structure failure more significantly.

110
Reliability analysis requires a statistical description for each of the input variables. This
means that, for each variable, it is necessary to choose an appropriate distribution and
then determine the mean value and variability (coefficient of variation or standard
deviation). The following are main uncertainties associated with ship probabilistic
analysis.

Table 5.4.1 Limit state equations

Failure mode Hogging Sagging
G = M IY − [M sw + k w M w ] G = M IY − [− M sw + k w ( M w + k d M d ) ]
Primary
(initial yield)
Primary
(ultimate G = M U − [M sw + k w M w ] G = M U − [− M sw + k w ( M w + k d M d ) ]
strength)
M sw + k w M w − M sw + k w ( M w + k d M d )
Secondary G = σ u, 2 − G = σ u ,2 −
SM b SM b
M + k wM w
G = σ u, 3 − sw − M sw + k w ( M w + k d M d )
Tertiary SM d G = σ u ,3 −
SM d

5.5.1 Strengths
Uncertainties in the ship structure strengths are basically related to uncertainties in
material characteristics and to approximation in the adopted model.
About yield strength (Guido Casella, 1998), σ s , it is usually been supposed that
material yield strength follows a log-normal distribution with mean values ranging
around 1.10 times the nominal value (corresponding to the 1% fractile) and a standard
deviation of 20MPa.
Uncertainty of Young’s modulus, E , may be determined from tests in the laboratory.
Coupon tests and stub tests are two alternative procedures. Young’s modulus follows
normal distribution in engineering practice.
ISSC’91 suggests a normal distribution for the uncertainties related to the model. In that
case, an approximate model was adopted and a mean of 1.0 and a COV of 0.15 were
used.
In the present case, all strengths (i.e. M IY , M u , σ u, 2 , σ u ,3 ) of ship structure are assumed
to be lognormally distributed. The section moduli are assumed to follow a normal
distribution.
In order to correct for inherent conservatio n in the calculated strengths, the mean values
analysis of all the strengths used in the reliability analysis are taken to be 15 percent
larger than the calculated values (SSC-368). This bias is suggested because all of these
failure criteria have their basis in the minimum yield strength of the material
(Galambos, 1989, and SSC-368). The following is another method to determine the
uncertainty of the strengths. An overall uncertainty was assumed to include both the
above mentioned effects, having a log-normal distribution with a COV of 0.15 and a
mean value of the calculated strength values.
The calculated section modulus values are used, unchanged, as the mean values for
those variables. The relative variation coefficients are suggested about 4%.

111
5.5.2 Loads
Uncertainties in loads can be divided into the following two aspects:
a) The stillwater bending moment ( M sw )
b) The wave-induced and dynamic bending moments ( M w , M dw ).
Load uncertainties are basically related to approximation in the adopted model and
uncertainties in significant wave height, H s , Peak period, T p , and so on. About
uncertainty of significant wave height, H s , it is suggested that H s follows a
transformed 3 parameter Weibull distribution. For peak period, T p , its distribution is
related with a significant wave height and usually follows a transformed conditional
log- normal distribution for given H s .

5.5.2.1 Stillwater bending moment M sw

The still water bending moment M sw is the static contributions to the vertical bending
moment caused by buoyancy and the static load distribution of the ship. Since cargo and
ballast are being changed over time, the still water bending moment will also change
over time. The stillwater bending moment is assumed to follow a normal distribution
with either a coefficient of variation of 0.15 for the military ships or a coefficient of
variation of 0.25 for the commercial ships. The stillwater bending moment for each ship
is generally assumed to be the same for every case, but some cases are different such as
tankers etc. Reference values for stillwater moments are calculated for either full or
maximum allowable conditions, the mean values (for use in the reliability analysis)
have to be reduced from the reference values. For the military ship, the mean value is
assumed to be 80% of the full load calculated value. For the commercial ships, the mean
value is assumed to be 60% of the calculated maximum allowable value (see SSC398,
SSC373, Mansour 1993). Tanker is a special ship, because they spend about half of
their operating life in a fully loaded cond ition and the other half in a ballast condition.
When in the fully loaded condition, the tanker has a sagging stillwater bending moment.
For the ballast condition, the tanker’s stillwater bending moment is a hogging moment.

5.5.2.2 Extreme wave-induced be nding moment M w

The mean values for the extreme wave- induce moment, M w , are calculated using
second-order strip theory. In the ISSC’91 report, the extreme wave bending moment in
short/ long-term seastate, M w , is suggested to follow normal distribution normally and
an overall coefficient is proposed, which takes into account the following effects:
(1) uncertainties in the sea model;
(2) uncertainties in the ship linear response;
(3) correction with the full scale measurement under ‘linear condition’.
For the short-term time frame, an extreme sea condition is chosen based on a small
encounter probability. The ship is also assumed to be travelling in head seas. The 50
percent probability-of-exceedence value is taken as the mean for the reliability analysis.
The extreme wave moment is assumed to follow the extreme value distribution with a
coefficient of variation 0.10.
For the long-term time frame, a time-weighting method is used to calculate an
operational profile. From this profile and repeated short-term analysis, probability-of-
exceedence of extreme lifetime hogging and sagging bending moments is generated for

112
each ship. The value at the 50% exceedence probability is taken as the mean value for
the extreme wave bending moment distribution. To the extreme wave- induce moment
with long-term time frame, an extreme value distribution is used and a coefficient of
variation of 0.10 is used (SSC-368).

5.5.2.3 Dynamic slamming bending moment M d

The slamming- induced moment is a sagging moment and is included only when the
sagging loading condition is considered. The dynamic moment is taken to be a fraction
of the extreme wave moment. The extreme dynamic moment is taken to follow the
extreme value distribution. In the SSC398 report, the mean extreme dynamic moment is
assumed to be 40 percent of the mean extreme wave moment for the fine-hulled
warships. For the fuller-formed commercial ships, it is suggested to be 20 percent of
M w . A coefficient of variation of 0.3 is used.

5.5.3 Load combination factors

To account for the correlations between various loads, two load combination factors are
used in reliability analysis of ship structure. Two combination factors are needed: one to
combine the wave-induced and dynamic moments ( k d ), a second to combine the wave-
dynamic composite moments with the stillwater moment ( k w ). The normal distributions
are chosen to model these uncertainties. Coefficient of variations was selected for each
of the factors, with k w having a coefficient of variation of 0.05, and k d having a
coefficient of variation of 0.15. The mean values are suggested to be 1.0 for k w and 0.7
for k d .

5.5.4 Joint probability density function

To each limit state function, strength and load variables are considered to be
independent. The joint probability density function is the product of all PDFs of each
random variable.

5.6 Calculation and analysis

With all the variables defined and the limit state equations formulated, the next step is
to actually perform the reliability calculation by the methods in Chapter 2 as in the
following formula:
Pf = ∫ f X ( X )dX
G ( X )≤0

One of the principal purposes of a reliability calculation is to get the failure probability
and reliability index of ship structure being researched. These probabilities are used to
assess the safety of ship structure in question. Other important results are the
coordinates of the design point in both original and standard normal space. These
coordinates are necessary to determine the partial safety factors in design of marine
structures. The reliability analysis of ship structure can be divided into time- invariant
analysis and time-variant analysis.

5.6.1 Time -invariant analysis

Time- invariant analysis means that all stochastic variables associated with the reliability
analysis of ship structure are not variable in the reference period of, suc h as, one year.

113
This is a basic and important reliability analysis of ship structures.

5.6.2 Time -variant analysis

Time-variant analysis means that some or all of the stochastic variables are time-
dependent and is, in fact, a stochastic process. In reliability analysis of ship structures,
still- water bending moment M sw , wave- induced bending moment M w and ultimate
bending moment M u are, in principle, time-dependent, i.e., they vary with time in the
reference period of 1 year. Among these variables, M u may be considered as a
deterministic parameter, neglecting corrosion, fatigue and any other time-dependent
phenomenon. Furthermore, M sw will be considered as constant for a voyage (as in
Guedes Soares C, 1985). A jump process, with a stochastic amplitude characterised by a
Beta distribution function and by a deterministic duration, corresponding to the voyage
duration, represents the time dependence of M sw . M w and will be represented by
means of its extreme value in a given period T. No closed-form solution is available to
calculate the probability of failure of structures for which two or three components Si
of the state vector Z are time dependent [ S i = S i (t ) ].
The approach generally followed is of an asymptotic nature, based on the scheme of a
Poisson process which models the exits of the structural state function in the failure
domain [Casella, G. 1998 and Bryla P. etc. 1991]. The intensity of this Poisson process
corresponds to the out-crossing rate of the failure surface. The expected number of exits
can consequently be determined.
The presence of time- invariant components R in the state vector Z[ Z ( t ) = ( R, S ( t ))]
complicates the problem, because the Poissonian nature of the process is lost. In this
case, the expected number E[ N + ( t r )] of out-crossings within time t, conditional on a
given realisation r of the time-independent variables R, is first evaluated [Casella, G.
1998 and Bryla P. etc. 1991, Breitung K, etc. 1982]. The associated probability of
failure is then calculated and un-conditioned over all the possible realisations of R.
Computation procedures depend on the type of time dependence of the various state
variables: the variables are all time- invariant except two ( M sw and M w ) which can be
modelled with a step-wise dependence on time. The following is a general procedure of
time- variant analysis of ship structural reliability.

5.6.2.1 Rectangular wave renewal jump process

A rectangular wave renewal process is characterised in general by the amplitude X and
the mean jump rate λ . The mean number of jumps E[ N j ] in a given time interval
(t1 , t 2 ) is given by λ ( t2 − t1 ) and the amplitude of each jump is an independent random
variable, with its own distribution. Extending the model to two or more processes, each
with a stationary jump rate and with amplitudes serially independent and independent
from each other, given rise to a stationary, vector rectangular wave process. This kind of
time dependence can be handled in load combination problems [Breitung K, etc. 1982].
The expected number E[ N + (t1 , t 2 )] should be appropriately selected.

5.6.2.2 Time -variant calculations

114
The two processes M sw and M w will be modelled according to the above me ntioned
scheme of wave renewal processes with the additional, simplifying, hypothesis of equal
duration T j for each jump (i.e. constant jump rate = 1T ) for both processes.
j

M sw , as stated above, is modelled by a Beta distributed amplitude associated with the

(constant) time duration of a voyage.
As regards M w , the constant jump duration has been selected according to the
following schemes (amplitudes are distributed as the extreme values in the same period)
[Casella, G. 1998]:
a) a jump each hour
b) a jump for each sea state (duration: 3 h)
c) a jump every day
d) a jump every voyage (about 20 days, depending on the loading condition)
e) two jumps in a year
f) a single jump in one year (as the whole analysis is referred to one year, in this
case the process M w is practically converted into a time- independent variable).

5.7 Fatigue Reliability

The fatigue limit state is associated with the damage effect of repeated loading.
Assessment of the fatigue life of a given ship structural detail is obviously associated
with uncertainties. Some of these uncertainties are due to the stochastic nature of the
wave- induced loads and the scatter in the fatigue resistance of the detail. Other
uncertainties exist in the approaches for calculation of the accumulated damage. In
order to deal with these uncertainties properly, a reliability-based framework is the best
solution, as it gives the possibility of treating each source of uncertainty independently,
and then finally assessing their combined effect on the calculated damage.
The fatigue reliability analysis and application mainly include the following three tasks:

5.7.1 Determination of uncertainties

In order to do a fatigue reliability ana lysis, the first task is to assess and quantify the
major uncertainties that exist in the fatigue checking procedure adopted. In the present
work the major uncertainties considered are:
uncertainty in fatigue damage accumulation assumption,
uncertainty in SN curve,
uncertainty in load calculation,
uncertainty in nominal stress calculation,
uncertainty in hot-spot stress calculation,
uncertainty in detail quality.

Here detail quality refers to misalignments, weld quality, etc. Many other aspects such
as the influence of plate thickness, mean stress, corrosion, weld treatments or
improvements, two-slope SN curve, and material are taken into account when the
fatigue strength is evaluated, but this can be done through deterministic coefficients
which are assumed not to influence the model uncertainty.

5.7.2 Establishment of failure limit state

115
There are two approaches to the fatigue problem, the Palmgren-Miner approach based
on S-N curves, that will be used here, and the fracture mechanics approach.
The S-N curves are obtained by experiments and give the number of stress cycles to
failure. Such curves are of the form (Rasmus FolsO etc, 2002, SSC-368):
N ⋅ ∆S m = C (5.7.1)
where
N is the number of cycles to failure
∆S is the stress range
m is the inverse slope of the S-N curve
C is determined from the S-N curve by
log C = log a − 2σ log N (5.7.2)
where
a is a constant referring to the mean S-N curve
σ log N is the standard deviation of logN
The fatigue life calculation is determined based on the assumption of linear cumulative
damage (Palmgren-Miner rule). Application of this assumption implies that the long-
term distribution of stress range is replaced by a stress histogram consisting of an
equivalent set of constant amplitude stress range blocks.
The time to failure of a detail can be expressed as Wirsching, P.H., 1987:
~ ~
~ ∆ ⋅C
T = ~ Fm ~ (5.7.3)
B ⋅Ω
where
~
∆ F is the value of the Palmgren-Miner damage index at failure.
~
C and m are obtained from the S-N curves.
~
B is the ratio between actual and estimated stress range.
Ω is the stress parameter.
T , ∆ F , C , B are random variables. If the long-term distribution of the wave process is
assumed to be a series of short-term sea states that are stationary, zero- mean, Gausian
and narrow banded, and if, in addition, the structure is linear, the stress range will
follow a Rayleigh distribution and Ω is determined from [Wirsching, P. H., 1987 and
Ximenes, 1990]:

Ω=
(2 2 ) m
 m
Γ1 +  ⋅ ∑ p j λ(0mj−1) / 2 λ02.5j (5.7.4)
2π  2 j
where
p j is the probability of occurrence of the j-th sea state.
λ0 j , λ 2 j are the zero and second stress spectrum moments in the j th sea state,
1 λ2 j
respectively. Note that is the frequency of the stress process in the j th sea
2π λ0 j
state.
The fatigue limit state function is expressed as:
~ ~
∆F ⋅C
g( X ) = ~ m ~ −τ (5.7.5)
B ⋅Ω

116
where τ is the service life of the ship.

5.7.3 Calculation of fatigue reliability

The failure probability is calculated by following formula:
Pf = ∫ f X ( X ) dX
g ( X )≤ 0

5.7.4 Establishment of target reliability index

This is concerned with a selection of a target safety index (maximum acceptable
probability of failure), which can be used in the calculation process to define partial
safety factors. For this purpose the ‘actual’ safety indices for a large number of
structural details in a wide selection of existing ships are suggested to be determined
first. Selecting the target safety index on the basis of actual safety indices ensures that
the new guidelines will impose the same safety against fatigue cracking as the existing
rules.

5.7.5 Calibration of partial safety factors

When the modelling of the uncertainties and the determination of the target safety
indexes have been completed, the final task is to calibrate the partial safety factors. The
usual approach is based on FORM or SORM methods, but in this case this will reduce
the flexibility of the partial safety factors. Application of the FORM/ SORM methods
will in this case mean that a complete and unique set of partial safety factors will be
determined for each possible combination of calculation approaches. However, when
adopting the more simple lognormal reliability format, as previously done in the
offshore area [Wirsching PH. 1984], it is possible to obtain partial safety factors that
can be combined arbitrarily, thus allowing an easy selection of partial safety factors that
together with any combination of calculation approaches yield the required safety
against fatigue cracking. For the procedure and methods of calibration of partial safety
factors see the relevant section in Chapter 6.

5.8 Conclusions
In this chapter, the procedure of calculating reliability of ship structure is introduced in
detail, including identity of failure modes, calculation of loads and strength of ship
structures, establishment of limit state functions, etc.
Besides the above, in this chapter, time- variant computations are outlined. In the time
variant scheme, the reliability index, plotted as a function of the number of jumps for
the dynamic load component, shows an asymptotic behaviour. The asymptotic value can
be obtained by adoption of jumps corresponding to a single sea-state for the wave-
induced bending moment. Time- invariant computations can be effectively used to
approximate the results of the more complex time-variant approach.
Finally, fatigue reliability is introduced simply.

117