Professional Documents
Culture Documents
4.1 General
For performing reliability analyses of marine structures, certain specific load and
strength data are necessary. Prior to estimating the loads acting on ships or marine
structures, a statistical representation of the environment is necessary. This includes
waves, wind, ice, seismic effects and currents. The last four items are more important
for fixed offshore structure than for floating vessels. The environmental information can
then be used as input to determine the loads acting on the structure. Typically, an input/
output spectral analysis procedure is used to determine the short-term loads, as loads
in a specific sea condition (stationary condition). The required transfer function is
determined from first- or second-order strip theory using the equations of motion of the
vessel, or from a towing tank experiment. In offshore structures, Morisons equation is
usually used to determine the wave load transfer function.
Prediction of the loads in stationary sea conditions (spectral analysis) is not sufficient
for the reliability analysis. Extreme values and long-term (lifetime) prediction of loads
and their statistics are more valuable. For this purpose, order statistics and statistics of
extremes play a very important role. Gumbels theory of asymptotic distributions is
often used in this regard. In the long time prediction, the fatigue loads, i.e., the cyclic
repetitive loads which cause cumulative damage to the structure, must also be
considered. Methods of combining the loads, such as static and dynamic, including
high- and low- frequently loads, should be considered. In nature, many of these loads act
simultaneously, therefore their combination must be evaluated for a meaningful
reliability analysis.
In assessing the reliability of ship structures, two general loading situations may be
used: short-term or long-term analysis. At the design stage, if the route of the ship is
known and if that route is more or less permanent, then the probability of failure can be
predicted using long-term analysis. If, on the other hand, the ship is likely to take a
variety of routes during its lifetime, then short-term analysis can be used to obtain the
probability of failure under one or more conditions that are considered to be the severest
the ship may encounter during its lifetime.
The criterion usually used in the short-term analysis is to consider the single most
severe sea condition (a sea condition with a specified return period, or more
appropriately, a sea condition with a specified encounter probability) and subject the
vessel to this condition for a specified period of time.
These short- and long-term analyses will naturally produce different final results for the
safety margins. Therefore, care must be taken when comparing safety margins of
different ships, i.e., the method and criterion used in predicting the loads acting on the
ship will have a considerable impact on the resulting safety index. To further amplify
this point, the long-term distribution of the wave loads acting on a ship may be
determined by tracing the expected route/s of the ship during its lifetime. Based on
ocean wave statistics along the route/s, the long-term (lifetime) wave load probability
distribution for the entire history may be determined. In the short-term analysis, extreme
load distribution is predicted on the basis of criteria such as one extreme sea storm of a
specific encounter probability and duration, or a short-term operation in a specific
location under severe sea conditions. It should be noted that there is a fundamental
difference between computed results based on these two avenues. In the short-term
80
analysis, the computed probabilities of failure are conditional probabilities given the
occurrence of an extreme wave load per selected criterion. Care must be taken in this
case in determining the response of the ship to this extreme load since non- linearity will
play an important role. In the long-term analysis, the resulting probabilities of failure
are associated with the entire history of the expected loads acting on a ship during its
lifetime, and are dependent upon the selected route/s likely for the ship.
The procedure of ship structural reliability analysis for the short-term analysis is as
follows:
a) From ship route (if known), obtain ocean wave statistics, and a specified
encounter probability (or return period) determine the design storm condition
(see section 4.3).
b) Calculate the rms value of the wave bending moment in the design sea
condition, using either second order strip theory, or towing tank experiment.
Calculate also the Stillwater bending moment (see section 4.3).
c) Estimate the strength parameters for each failure mode (see Chapter 5)
d) Calculate the probability of failure or the safety index for each failure mode.
The resulting probabilities are conditional probabilities. They are conditioned on
encountering the design storm.
In general, the long-term procedure entails the determination of the probability
distribution of the maximum load during the lifetime of a ship taking into consideration
the wave statistics along the ship route, loading conditions, speed, and heading. The
procedure is particularly important for fatigue reliability analysis, where the entire
history of loading should be determined. In that fatigue case, the long-term distribution,
instead of the maximum load distribution, is required and is usually assumed to be
Weibull. Several procedures have been proposed in the literature for determination of
the lifetime maximum load distribution. Although their details may vary (sometimes
depending on the ship type), most of them have common characteristics as follows:
a) Define the mission profile of the ship which includes
1) ship route
2) expected total years of service
3) number of days per year the ship is expected to be at port and underway
4) nominal cruising speed and maximum operating speed in each sea state,
and the corresponding fraction of time during operation
5) distribution of ship headings
6) distribution of loading conditions
b) From the ship route and available wave statistics, obtain the frequency of
occurrence of different sea conditions the ship will encounter in each of the
geographic areas (zones).
c) On the basis of the above, determine the frequency of encountering different sea
conditions, loading conditions, speeds, and headings.
d) Determine the wave loads in each sea condition, loading condition, speed, and
heading using first- or second-order strip theory.
e) Use an extrapolation procedure to determine the distribution of the maximum
load in a lifetime.
4.2 Review of computational methods of loads on marine structures
This section will present an overview of load computational methods (ISSC,
COMMITTEE I.2, 2003).
81
82
for example, a floating platform with four columns. Computation of the real wave
elevation, including the effects of non-linearities associated with trapped waves,
remains a challenge.
The capability to predict second-order forces on large volume structures has been
available for several years. However, it seems to be a challenge to compute local
quantities as velocities and wave elevation accurately. This requires careful
discretization of the boundary value problem. Perturbation to 3rd-order does not seem to
be an accurate way to handle the ringing problem. Fully non- linear potential theory
based on computations indicates that the 3rd-order forces are over-predicted in the
perturbation scheme, while the fully non- linear formulation gives similar forces as
measured. It is assumed that viscous forces play a minor role in the ringing problem.
Most programs do not handle breaking waves.
It still seems there is a long way to go before the NWT (New wave theory) becomes a
practical tool for engineering. Realistic Reynolds numbers are not obtained. However,
use of CFD (Computational Fluid Dynamics) techniques is a possible option to estimate
viscous damping effects.
CFD methods are being used to address VIV (Vortex-induced Vibration)- induced
fatigue. However, while the use of such methods is supported by classification
societies, it is realized that the general application of CFD to the design process is still
aways off.
From the fact that several methods for computing the hydroelastic response of a VLFS
(Very Large Floating Structure) are used, linear analyses seem to work well. A
remaining challenge is obviously non- linear computations to obtain better response
estimates close to resonance.
The need to explore increasing water depths requires that designers now address issues
related to slender bodies and coupled response in the relevant environments. This has
led to innovative testing techniques such as hybrid model testing. Such methods
provide a means to assess the behavior of new design concepts in a realistic way, but
do not provide the designer with insight as to the physical causes of such behavior.
In addition to modeling difficulties, the need to operate in ever increasing water depths
requires a better understanding of possible areas of operations. For example, current
fluctuations may manifest themslves as both a VIV- induced extreme load as well as a
fatigue load. Although designers may use a current profile, such profiles typically
assume constant direction over depth. A more rigorous handling of such fluctuations,
where probabilities of curent speed as a function of depth would be preferable.
However, such a scatter diagram will require extensive in-situ measurements.
4.2.3 Hydrodynamic Impact Loads
The hydrodynamic impact problem for ships manifests itself in both a global (whipping)
and local response. In predicting the global whipping response, it is important that the
instantaneous free surface elevation be properly taken into account but local
hydroelastic effects can be neglected. However, hydroelasticity may be an issue for
local structure. Typically, the importance of hydroelasticity will be a function of
natural frequency of both the structure and the duration of the slam load. Thus, the
structural designer should take the time to estimate these quantities before suggesting a
design load. In addition, when performing model tests, the experiment should be set up
such that not only impact pressures but also structural strains are measured.
83
84
Since there exists a range of possible options for looking at short- and long-term load
statistics, the application of any one method can be complicated and strongly depends
on personal experience of an engineer or scientist, the method used for estimate of
distribution parameters and the sample size. The possibility of process standardization,
sensitivity of long-term predictions with respect to type of short-term peak probability
distribution, and uncertainty in log-term predictions should be further investigated.
4.2.5 Experimental Uncertainty, Verification and Validation of Numerical Codes
There has been some progress in the development and practical implementation of
experimental and numerical uncertainty procedures. Efforts of experimental and CFD
communities are reflected in fact that some sort of validation analysis accompanies
more and more publications of measured and computed data.
The verification and validations procedure and their practical applications for viscous
fluid flow codes (RANS) seem to be better organized than for codes based on potential
flow theory. The viscous fluid flows simulations are usually accompanied by detailed
quantitative verification and validation procedures, when validation of inviscid /
irrotational-based codes mostly means qualitative comparison to published measured
data or semi-empirical calculations.
Efforts to create a new generation of benchmark data for validation purposes are
required. Experiments designed by experimentalists and code developers, carried out
according to specified procedure, including measurement of all required parameters
with required accuracy are needed. The procedure should consider and include
statistical aspects of model handling with respect to course-keeping (manual and/or
autopilot) and power delivery (constant speed or revolutions).
Standardization of experimental uncertainty procedure: identification of all possible
elemental error sources, common methods for calculation of certain properties, and
standardization of experimental procedures are required for consistent interpretation of
experimental results and uncertainties associated with those experiments.
While the process of determining experimental and numerical uncertainty is all well and
good, there are some practical concerns. For example, in order to address errors
associated with experimental technique, multiple runs require to be performed. In an
ideal situation, tests performed within a linear regime should be repeated at least 3
times. These tests should be repeated on not only on different days but with different
test crews as well. For tests where non- linearities are expected, a minimum of 5 repeat
tests are necessary.
Once such data become available the next question is, what is a reasonable level of
verification and validation of numerical methods? In theory, a series of convergence
studies, followed by comparisons with regular and irregular wave data would suffice.
However, in practice this may not be practical. For instance, how many convergence
studies should be done? Is one set, performed for one wavelength sufficient, or are
more required? And what role should full scale trials play, if any, in the validation
process? Then comes the question of hull forms to be studied. Is a validation against
one set of model tests for a specific hull form represent validation or just a documented
solution? From a pragmatic standpoint, code validation will be limited to what the user
is willing to pay for. However, a minimum threshold and some basic guidelines need to
be developed.
4.3 Loads
85
and the average waiting period, i.e., the return period, is:
Ri = E[W i ] = [1 FYi ( x )] 1
Ri = [1 ( Fxi ( x )) ]
ki
(4.3.5)
(4.3.6)
The relationship between the non-encounter probability and the return period Ri can be
obtained as follows:
Pnei = (1 Ri 1 ) Li
(4.3.7)
and the probability of encounter Pei is:
Pei = 1 (1 Ri1 ) Li
(4.3.8)
The return period Ri of a wave height x , in any zone i can be estimated from:
n
Ri = i y 0
(4.3.9)
n0
and
n i = [1 Fxi ( x)] 1
(4.3.10)
where
n 0 is total number of wave data collected in zone i
y 0 is number of years of data collection in zone i
n i is expected number of waves in zone i necessary to exceed wave height x
n i can be calculated from equation (4.3.10) for any value of design wave x .
The procedure for determining the ship/ storm encounter probability in any zone i can
be summarized as follows:
a) Use wave data in zone i to determine the form and the parameters of the
distribution function of wave heights Fxi (x ) , using any method of parameter
estimation, e.g., methods of moments, or regression analysis.
b) For the prescribed design wave height (or sea state characterized by a significant
wave height) predict the number of waves necessary to exceed the design wave
height using equation (4.3.10).
c) Using equation (4.3.9) to estimate the return period associated with the design
wave height or significant wave height.
d) Determine the probability of encounter in zone i form equation (4.3.8) and
information on how long the ship operates in zone i , i.e., Li .
4.3.1.2 Ship routes and the associated encounter probabilities
In order to determine the probability of encountering a wave height (or a specified sea
state) along a ship route, the zones and the harbour are considered as members in a
series system. A series system is defined as a system in which a state of encounter
occurs if an encounter occurs in any of its members. Similarly, the system nonencounter probability Pne can be realized only if mutual non-counter takes place in all
zones, i.e.,
n
(4.3.11)
Pne = P Ai
i=1
where Ai is the event of no encounter in zone i , P[ Ai ] = Pnei .
87
Pe = 1 Pne
If Ai are assumed statistically independent, then:
n
i =1
i =1
(4.3.12)
Pne = P( Ai ) = Pnei
(4.3.13)
and
n
Pe = 1 Pnei
(4.3.14)
i =1
(4.3.15)
(4.3.16)
i =1
These bounds are tight if the non-encounter probability in any of the zones is
dominated.
If the members of a series system are equally corrected, then an extension of the work
by Stuart, summarized in Structural Reliability and Its Application (Thoft-Christensen
and Baker, 1982) leads to the following system probability of encounter:
n
+ t
Pe = 1 i
(4.3.17)
(t )dt
1
i =1
where i cab be calculated from:
i = 1 ( Pei ) = 1 (1 Pnei )
(4.3.18)
and is the standard Gaussian cumulative distribution and density function,
respectively.
is the correlation coefficient.
4.3.2 Wave loads and load combinations
Estimating wave- induced loads is one of the most important tasks in ship design.
Principles of Naval Architecture (Lewis, Ed., 1988-89) suggests that there are four
methods by which wave- induced loads can be determined:
a) approximate methods
b) strain and/ or pressure measurements of full scale ships
c) laboratory measures of loads on methods
d) direct computation of wave induced fluid loads.
The above methods have their own advantages and limitations, respectively. For marine
design and marine structural reliability analysis, a simple and relatively accurate method
is more important. To calculate wave loads, the Second Order Strip Theory (SOST) is
usually used.
The following section will describe a simple formulation for determining slightly nonlinear extreme wave loads and load combinations. Then non- linear hogging and sagging
88
mr =
ln v01T
ln v 02T
mc =
ln v0 cT
ln v02T
i ( 2 ln v 0iT 1)
i = k i 1 +
+ i ( 2 ln v0iT 3)
(4.3.23)
(5.8 + 2 i ) 2 ln v0i T 30
i = 1 + 1 .5( i 3) 1
(4.3.24)
2
i i2
+
k i = 1 + 0.5
(4.3.25)
+
3
54
i
The difference between sagging and hogging moments manifests itself in the sign of the
skewness , i.e., is positive for sagging and has the same value but with a negative
sign for hogging.
An extreme value of a load f associated with exceedence probability can be
determined by:
89
f = 2 ln v 0T
(4.3.26)
where
v0
(4.3.27)
ln( 1 ) 1
The most probable extreme value is associated with an exceedence probability
= 0.6321.
In order to estimate the most probable extreme value of a slightly non-linear load (see
equation (4.3.22)), the non-linearity parameter needs to be evaluated first. Evaluating
means determination of the skewness and kurtosis . Both of these moments are
shown (Mansour and Jensen, 1995) to depend on the significant wave height H S , the
zero up-crossing period of waves Tz , ship geometry, ship speed V , and heading angle
, i.e.,
= ( H S , Tz , shipgeomet ry ,V , )
(4.3.28)
= ( H S , Tz , shipgeomet ry ,V , )
(4.3.29)
v 0 =
m0
m2
m0 , m2 are the sea spectral moments.
Furthermore, the skewness and kurtosis have the following relations with H S :
= f (T z , shipgeomet ry, V , )
(4.3.30)
Hs
3
= g (T z , shipgeomet ry, V , )
(4.3.31)
H s2
The units used for the significant wave he ight H S are metres. The zero up-crossing
period, Tz , is expressed in seconds. Charts of the skewness and kurtosis as a function of
each parameter are shown in Figures 2.1.11 to 2.1.36 of the literature SSC-398. H S and
Tz are not independent and the relation between H S and Tz is given by:
where T z = 2
Hs
(4.3.32)
g
The charts for frequently occurring pairs of H S and Tz can be referred to Figures
2.1.38 to 2.1.41 of the literature SSC-398.
T z = 11.12
fi
fc
f
H i ( )
H i* ( )
Hs
k
K
K
L
mr , mc
m n ,i
M (t )
M 0 (t )
Ni
r
Re(.)
S x ( ), S c ( )
T
Tz
U (t )
V
zf
0i
ij
c
l
= Stress ratio 2
1
= Real part of a complex function
= Wave and combined response spectra, respectively
= Time of exposure
= Zero up-crossing period of waves
= Standard Gaussian process
= Ship speed
= Vertical distance between ADK and AW P
= skewness
= Kurtosis
= Refer to equation (4.3.24)
= Non-linearity parameter
= Band width parameter
= Standard deviation
= Mean value
= Rate of zero up-crossing of load process i
= Correlation coefficient between to response components i
and j
= Non-linear standard deviation
= Standard deviation of the combined response
= Linear standard deviation
= Ship heading angle
= Frequency
91
fc =
f w2 + f sl2
where f w and f sl are the individual (wave and slam related) extreme stresses, with the
two processes assumed uncorrelated (in terms of frequency, not intensity) because of
their typical frequency separation.
4.4 Probabilistic methods
The ship design process requires knowledge of extreme lifetime loads. The loads can
be obtained from a variety of processes. The non- linear predictions combined with the
extreme of random process theory and long-term statistics are the preferable methods.
The fully non- linear simulations, though available, are still in development /
improvement, verificatio n and validation stages. The partially non- linear and linear
codes are still the most popular tools for loads calculations. The 14th ISSC (2000c) VI.1
Extreme Hull Girder Loading Special Task Committee presents very comprehensive
overview of approaches for estimate of short-term extreme values and prediction of
lifetime design loads. Based on the complexity of the hydrodynamic wave process and
length of available time series, different statistical methods can be applied to estimate
short- and long-term extremes.
In the linear cases, the wave loads in random sea can be calculated from linear potential
theory in frequency domain. The loads process follows Gaussian distribution and its
peaks are Rayleigh distribution. The Hermite transformation of a standard Gaussian
process can be employed to model the slightly non- linear problem. The extreme values
can be calculated using formulas similar to linear Rayleigh distribution, which includes
parameter-representing non- linearity of the process (skewness and kurtosis). For strong
non- linear process, wave- induced loads must be calculated using time-domain methods
and statistical analyses are conducted on actual load peaks obtained from load~time
histories. Sagging and hogging moment peaks should be treated separately. The threeparameter Weibull distribution is the most successful in modelling the wave- induced
92
load statistics (Wang, 2001). The pdf of the three-parameter Weibull distribution is
given by:
f ( x ) = cc ( x )c 1 e [ ( x )]
Where x=0, c is the shape parameter, ? is scale parameter, and d is location parameter.
The advantages of the Weibull model for distribution of vertical bending moment are
summarized as follows:
The calculation of distribution parameters is relatively simple,
It is sufficient for representing non- linearities in sagging and hogging moments for
various types of ships and operating conditions,
It gives good distribution of peak values around the tail area for reliable estimate of
extremes, and
It allows for estimate of non- linearities by straightforward comparison to Raleigh
distribution.
Tawn and Heffernan (2001) analyzed statistical methods for short-term distribution of
impact pressure loads from seakeeping model experiments with bulk carriers. The
analysis involved vertical pressure loads inserted on the top of the most forward cargo
hold during operation in various environments. Two statistical models were considered:
the Weibull and generalized Pareto distributions. The quality of fit of both distributions
was assessed using P-P (fitted probability vs. sample probability) and Q-Q (fitted
quantile vs. sample quantile) diagnostic plots assuming threshold pressure of 5 kPa. In
the case of this experiment, the P-P plot shows slightly better fit of Pareto distribution,
and Q-Q plot demonstrate superior fit of Pareto distribution. Specifically, the Weibull
model overestimates probability of large impact occurrence, which is crucial when
extrapolating to obtain desired extremes. The generalized Pareto distribution fit the
pressure impact data better that the Weibull distribution for all examined experimental
cases.
Lu at el. (2002) examines four statistical models for prediction of the most probable
extreme value of jackup structure responses due to excitation of random wave motions.
The methods are compared in terms of calculated extremes and their sensitivity to
structural stiffness, simulation time step and random seed techniques. The analyses
were conducted for two typical jackup structures. The extreme value is defined as one
with exceedence probability of 10-3 , which reflects a storm of 3 hours persistence. The
four applied methods are: drag/inertia parameter method, three parameter Weibull
distribution, asymptotic Gumbel distribution and Hermit transformation. In the
discussion of results the authors indicate that from the accuracy point, the three last
methods provide reasonably close predictions. The Gumbel distribution is described as
theoretically most accurate, if enough simulations are generated, and the Hermit
transformation as the most robust and efficient. Results of sensitivity analysis are not
conclusive and authors indicate need for more research in those areas.
c
93
In order to estimate the lifetime extreme value of ship response, a long-term analysis
has to be performed. This approach requires that the long-term peak distribution of
maxima is obtained as a weighted sum of short term probabilities of exceedence in all
possible combinations of mean wave periods, significant wave heights, heading angles
and speed. When the linear process is considered the short-term responses can be
calculated using the superposition method. To obtain non- linear responses, timedomain simulation or experimental data are required. This leads to an expensive, time
consuming and unpractical process.
Sagli (2000), Videiro and Moan (1999), and Baarholm and Moan (2001) present the
contour line approach and demonstrate that the long-term load extremes for marine
structures can be estimated in efficient manner by considering only a few short-term sea
states, instead of determining responses for all sea states. The initial step in analysis is
to calculate coefficients of contribution CR(si) for linear responses.
CR ( si ) =
where:
QR(R>rD|si,0 ,u0 ) is the short term cumulative probability distribution,
fHs,Tp(his,tsi) is long-term joint probability distribution,
wsi is weight function,
? hi, ? ti are grid size,
QLT (rD) is long-term probability of exceedence.
The values of those coefficients are used, to establish the sea state with the maximum
value of CR(si), which is then used in an iterative process of more complicated nonlinear analysis. The sea state with the maximum CR(si) and all sea states with
significant coefficient of contribution are established based on non- linear runs. The
desired extreme design value can then be calculated.
Guedes Soares and Dogliani (2000) investigated the effects, during a double bottom
tanker voyage, of loading condition changes on distribution of still-water bending
moment at midships. Significant differences with respect to still water loads between
departure and arrival conditions were observed, even if the differences in the
displacement and mean draft were small. Assuming that the departure and arrival loads
can be described by a normal distribution and that the change between departure and
arrival is linear, the authors conclude that the still water loads at a random point in time
can be obtained from a Gaussian distribution. The mean and standard deviation of that
distribution can be calculated as the average of mean and standard deviation at
departure and arrival.
Wu and Hermundstad (2002) propose a new approach to estimate long-term probability
of exceedence of non- linear sagging and hogging moments at midships for the S-175
containership. They applied time-domain simulation code to calculate short-term non94
linear wave-induced loads. The generalized gamma distribution was used to fit shortterm peak distribution and a simplified long-term procedure was used to obtain
probability of exceedence of sagging and hogging values given by DNV rules. The
applied generalized gamma probability density distribution has the form:
c cr cr 1 ( y ) c
u y e
,0 < y <
( r )
? (r) is gamma function, and c, r and are distribution parameters.
The first step is to select an appropriate wave scatter diagram. In the traditional method,
the initial number of significant wave height and modal period combinations could be
large and require a number of time-consuming non-linear simulations. From the
practical point of view, it is desirable to conduct as few simulations as possible. So, the
first step of the new process includes simulations for head seas and the most probable
mean HS and TP values for each sea state. This process allows for identification of the
most critical sea states and then additional runs for variation of headings and TP for the
critical sea states. The long-term probability of exceedence for the non- linear process is
determined from:
c
( y1 ) c( r 1) e ( y1 ) n
P( y > y1 ) =
P( ) P( H S , T1 )
( r )
HS
T1
g ( y) =
96