Journal of Ship Research, Vol. 52, No. 1, March 2008, pp.

30–44

Abnormal Wave-Induced Load Effects in Ship Structures

C. Guedes Soares, N. Fonseca, and R. Pascoal

Center of Marine Engineering and Technology, Technical University of Lisbon, Instituto Superior Técnico, Lisbon, Portugal

The paper presents an approach to determine the global load effects induced on ship
structures by abnormal, freak, or episodic waves. It refers to the present procedure
of determining extreme values of wave-induced responses, including the recent ad-
vances of adopting time series of wave elevation as reference design conditions to
calculate the wave-induced structural loads on ships in heavy weather. It is shown
how this procedure can be extended to account for abnormal or episodic waves.
Reference is made to what is presently known about abnormal or freak waves,
showing that although it is possible to determine the loads induced by these waves
in floating and fixed structures, the present knowledge about the probability of oc-
currence of these waves is not enough to allow a wave design criterion to be defined
in a way consistent with the present probabilistic approaches. However, it is sug-
gested that at the present stage of knowledge it is possible to determine the loads
induced by abnormal waves similar to ones that have been measured at various
ocean locations and that are thus realistic; a method is described to perform such
calculations. Although this information cannot replace the wave-induced loads cal-
culated with the presently established procedures, it can serve as guidance for the
design. An application example is presented of a containership subjected to a wave
trace that includes an episodic wave that was measured during a severe storm in
Central North Sea. The measured wave time history is modified in order to investigate
the influence of the wave steepness on the induced vertical motions and loads. The
loads induced by the abnormal wave are compared for the first time with extreme
values from long-term distributions.

Keywords: abnormal waves; freak waves; nonlinear vertical bending moments; time
domain simulations; extreme loads

1. Introduction and thus the wave-induced load effects are dominated by nonlinear
waves and nonlinear ship responses.
NOWADAYS there is a tendency to move from empirical proce- To properly account for the nonlinear effects that develop in
dures to methods based on the first principles to define the criteria heavy weather, it is necessary to consider time domain codes, as
for reliability-based structural design of ships (Guedes Soares et otherwise the correlation between consecutive motion cycles that
al. 1996). The approach relies on a probabilistic model of struc- can involve large-amplitude motion with associated nonlinear
tural strength and on a long-term probability distribution of wave- forces and such phenomena as water on deck are not properly
induced loads, allowing the identification of the conditions that are taken into account in the frequency domain, which is based on
the largest contributors to the probability of failure. When design- linearity assumptions. Several approaches have been proposed re-
ing for ultimate strength, these conditions occur in heavy weather, cently to cope with this problem. This paper adopts the method of
Fonseca and Guedes Soares (1998a), which deals with the effects
of large-amplitude motions, including water on deck. Only rigid
Manuscript received at SNAME headquarters October 2004; revised manu- body motions and global structural loads are considered, and
script received April 2007. therefore, although the significant changes of buoyancy in the

30 MARCH 2008 0022-4502/08/5201-0030$00.55/0 JOURNAL OF SHIP RESEARCH

forward sections associated with slamming are taken into consid- extreme response is first determined by linear frequency domain
eration, the impact due to slamming and the resulting transient hull analysis. Then, using the theory of Gaussian processes near a
vibrations are not taken into account in the present study. maximum, the corresponding deterministic wave elevation is pro-
Not accounting for the vibratory response in this context is not duced and applied in the nonlinear simulation.
considered an important limitation because it normally occurs in Another possibility is to use the contour line approach (Sagli &
frequency ranges that are typically one order of magnitude higher Moan 2001). With a long-term analysis of the wave climatology it
than the wave encounter frequency. This shorter duration of the is possible to define a set of sea states corresponding to a specific
stress cycles limits its effect on the global failure of the hull return period. These sea states include the worst environmental
structure, which normally requires a larger duration to respond and conditions and lie in a closed contour on the scatter diagram. The
collapse. design extreme response is the most probable extreme value de-
A consistent and first principles–based methodology to estimate termined within all short-term sea states of the contour line.
the design wave structural loads on ships requires a correct physi- More attention has been given recently to the design for acci-
cal hydrodynamic model of the ship responses in large waves and dental loads. Thus, in addition to the traditional design for the
a proper stochastic characterization of the waves and of the ship most probable extreme situations, which would be covered by the
responses. To calculate the design wave loads, one has to consider approaches discussed above, one should also consider very un-
all possible sea states that the ship encounters during its opera- likely accidental situations in which the ship should be able to
tional lifetime. For the linear case, the procedure relies on linear survive, although with damage. Typical cases are the efforts of
long-term calculations and it is computationally efficient (Guedes design for damage stability and for crashworthiness. Another situ-
Soares & Moan 1991). ation, which is of concern in this paper, is the design for abnormal
For several reasons, the same is not true if the ship responses are waves.
nonlinear: First, the time domain simulations are much more time The occurrence and the damaging effects of abnormal waves
consuming than linear frequency domain calculations. Second, the were identified a while ago, but more recently researchers have
hydrodynamic model needs to be run for all possible sea states of shown a renewed interest (e.g., Faulkner & Buckley 1997) and
the scatter diagram, since the responses depend on the significant have made proposals for specific aspects of design, such as hatch
wave height. Finally, short-term statistics cannot be represented by covers. To develop a consistent approach for the design of the ship
the Rayleigh distribution, and more complicated distributions need hull girder to these kinds of loads, it is necessary to be able to
to be applied. Altogether this means that, presently, practical so- predict the likelihood of encountering such waves and also to
lutions are possible only with simplified stochastic procedures. predict the effects they induce on the hull. One could anticipate
Several such procedures have been proposed during recent years that if these questions are solved, there might be a set of design
to calculate the design wave loads on ships and offshore structures time series that would represent a set of possible abnormal waves
accounting for the nonlinear effects. The next paragraphs present and the relative performance of different ship structural designs
a short review of these methods. could be compared on the basis of their response to these time
Guedes Soares (1993) and Guedes Soares and Schellin (1998) series.
generalized the linear long-term predictions procedure to account This paper explores the feasibility of progressing in that direc-
for the nonlinear asymmetry of the vertical bending moment. The tion, by determining the loads induced by a time series that was
new method uses form functions that transform the linear transfer recorded in the North Sea, which includes one such abnormal
functions to nonlinear pseudo-transfer functions associated with wave with a height of about 26 meters. Since there is an idea that
different sea states. the wave-induced loads depend not only on wave height but on
Several of the proposed procedures are based on the assumption steepness also, the steepness of the abnormal wave was artificially
that the linear model is a good identifier of the conditions in which changed, and it was shown that the wave-induced load effects are
the extreme wave loads occur. Adegeest et al. (1998) propose the indeed sensitive to those changes.
design regular wave method, in which a regular wave is deter- Therefore, these results are encouraging as to the possibility of
mined based on linear long-term calculations, and then a nonlinear evolving in the direction of defining a criterion for accidental
simulation is run for that regular wave to determine the design design against abnormal waves, in the sense that available com-
structural loads. putation tools will be able to deal with specified design time
The coefficients of contribution method also uses linear long- series.
term calculations to identify the sea states of the scatter diagram Some preliminary results were presented by Fonseca et al.
that probably contribute most to the probability of exceedance of (2001) concerning the motions and structural loads on a contain-
the structural loads during the ship’s life. This way the nonlinear ership induced by the mentioned rogue wave. The results were
simulation program is applied to a selected small number of sea obtained by a nonlinear time domain seakeeping method. The
states only. Sagli and Moan (2001) studied the nonlinear vertical simulations were carried out in head waves, for the service speed
bending moment on a containership in the northern North Sea. and also zero speed. In the present paper a more realistic speed of
The critical wave episodes method is another possibility, ad- two-thirds the service speed is used, and a comprehensive inves-
equate for the most advanced and time-consuming hydrodynamic tigation of the influence of the steepness of the rogue wave on the
models. Torhaug et al. (1998) identified the random incident wave ship responses is carried out. The maximum vertical bending mo-
sequences that result on the ship extreme response by performing ments induced by the rogue waves are compared with rule values
linear time domain analysis for each relevant stationary sea state. and with design values obtained from linear and nonlinear long-
The wave sequences resulting in the largest linear responses are term distributions.
then applied to the nonlinear analysis. In the most likely response The same methodology was applied some years later by Clauss
method (Adegeest et al. 1998, Pastoor et al. 2003), the most likely et al. (2004) and Guedes Soares et al. (2006b) to investigate the

MARCH 2008 JOURNAL OF SHIP RESEARCH 31

structural wave loads on a floating production, storage, and off- one of the inequalities and thus that can be classified as abnormal
loading unit (FPSO) induced by deterministic rogue waves. This waves.
work included numerical and experimental investigations. Com- The parameters that identify the abnormal waves have been
parisons between experiments and simulations showed that the related to the sea state statistics that describe degree of nonlinear-
numerical model is able to represent remarkably well the wave- ity, namely, the coefficients of skewness and kurtosis, which
induced bending moment at midship in the highly nonlinear should be zero when the distribution of sea surface elevation can
waves, including the asymmetry of the sagging and hogging be modeled by a Gaussian distribution.
peaks, providing an experimental validation of the method used in The correlations between the maximum crest height Crmax D and
the present paper. several sea state characteristics have been determined by Guedes
Soares et al. (2003). The one found to be more significant is with
the coefficient of skewness ␥3. The regression equation between
2. Characteristics of large and abnormal waves the coefficient of skewness and the vertical asymmetry of the
maximum wave, Crmax D/Hmax D, in the full-scale data was:
Different authors have suggested that extremely high ocean
waves may be generated by different mechanisms, but all mecha- ␥3 = 1.9216 Crmax D Ⲑ Hmax D + 0.98831 (1)
nisms that can lead to these waves are not fully identified and The regression between the abnormality index, calculated with
understood. White and Fornberg (1998) and Lavrenov (1998) ex- down-crossing definitions AID, and the coefficient of kurtosis ␥4
plained the appearance of the abnormal waves by a wave ampli- for all those sea state records from the North Alwin and the
fication due to the interaction with current. Pelinovsky and Kharif Draupner is:
(2000) modeled the temporal and spatial focusing as a result of the
wind wave dispersion and of special distribution of their fre- AI = 0.467␥4 + 1.577 (2)
quency. Trulsen and Dysthe (1997) suggested that the Benjamin-
A relationship has been derived to approximate the statistics
Feir instability can cause breaking of a wave train into periodic
obtained from full-scale data:
groups and, further, within each group a focusing takes place,
producing very large and steep waves. Henderson et al. (1999) ␥4 = 3.764␥23 + 0.236 (3)
also suggested nonlinear instability as a cause of the big waves.
Whatever the nature of the generation process, the designers These relations allow the degree of nonlinearity of a sea state, as
and operators of marine structures require information about the defined by its coefficients of skewness and of kurtosis, to be used
representative shapes of these large waves so that they can be to determine the expected values of the abnormality index and the
appropriately considered in design. This is supported by the ex- vertical asymmetry of the maximum wave.
amples of heavy weather damages caused by giant waves that have Guedes Soares et al. (2004) studied the correlation between
been presented in the literature during the last 20 years, for ex- various sea state parameters and the parameters from the largest
ample, by Kjeldsen (1997) and Faulkner and Buckley (1997), individual waves in a sea state. In some cases these waves corre-
which contributed to a more widespread belief that the damage to sponded to the abnormal waves mentioned, but in other cases they
engineering facilities operated at sea for extended periods is often corresponded to waves to be expected in that population. One
determined by the few extreme sea states, not by the frequent and conclusion from that study was that in general the correlations
moderate ones. Also, in these sea states the damage will generally obtained for the larger waves in each sea state would also fit well
be induced by few extreme waves rather than by the large majority the ones identified as being abnormal, meaning that in this sense
of other waves. no change in the trend was apparent or, in other words, no abnor-
Therefore, in the engineering community there has been a major mal behavior was detected.
interest in properly understanding and describing the conditions The correlation between total ship losses in heavy weather and
associated with extreme waves so that they can be used in the the average steepness of sea states in those geographical areas has
design process of ships (Fonseca et al. 2001) and offshore struc- been revealed to be surprisingly high, leading to the hypothesis
tures (Clauss et al. 2002). that steepness may be a governing factor in the damages induced
The definition of abnormal or freak wave that seems to be more by bad weather in ships (Guedes Soares et al. 2001). It is therefore
generally accepted is the one that has a height larger than twice the of interest to assess the steepness of the large waves so that this
significant wave height, as reviewed in Guedes Soares et al. wave characteristic can be properly modeled in the criteria to be
(2003). This ratio was denoted the abnormality index (AI) and will used for design, in addition to the height and vertical asymmetry
be used as a reference. discussed before.
Some authors also refer to the ratio of the maximum crest height The significant wave height is the variable most widely used to
to significant wave height as a governing criterion to identify characterize the degree of severity of sea states, and on many
abnormal waves. In fact, Tomita and Kawamura (2000) define a occasions it is the only one used as a design parameter. Therefore,
genuine freak wave as one in which the AI > 2 and CrD/HsD > it is important to determine how the geometric parameters of the
1.30, where CrD is the crest of the down-crossing wave and HsD is highest waves in each sea state correlate with the significant wave
the significant wave height when using the down-crossing defini- height. This will allow conditional probability models to be used
tion of waves. to estimate the expected geometric parameters of the largest wave
Guedes Soares et al. (2003) analyzed nine time series of the based on the prediction of the expected significant wave height.
Draupner from a storm in the beginning of 1995 and the 421 Guedes Soares et al. (2004) studied a series of wave records
records of the North Alywn during the storm of November 1997, from the North Cormorant in the North Sea and found that the
from which 24 records were identified as having waves that satisfy correlation between the significant wave height and the crest of the

32 MARCH 2008 JOURNAL OF SHIP RESEARCH

maximum registered wave was 0.93. The correlation between sig- infinite number of harmonic waves. A wave frequency and ran-
nificant wave height and the coefficients of steepness of the indi- dom phase angle define each harmonic component. Consider an
vidual wave, characterizing the front of the maximum wave, were inertial reference system advancing with the ship mean forward
studied and regressions were derived. speed, with the origin at the mean waterline level at the vertical of
Linear sea states can have the probability density of free surface the center of gravity, the x axis pointing to the mean bow direction,
elevation modeled by a Gaussian distribution that has zero skew- the y axis pointing to port side, and the vertical axis pointing
ness. This suggests that one can use the skewness of the sea states upward. The wave elevation of an irregular and long crested sea
as a measure of the degree of nonlinearity, and thus it is very state represented in this reference system is:
interesting to know how the degree of nonlinearity relates to the N

兺共␨ 兲
steepness and asymmetry of the largest waves in the sea state.
The correlation between Hs and skewness ␥3 is 0.70, which ␨共x, y, t兲 = − a n cos关共␻e兲nt + knx cos共␤兲 + kny sin共␤兲 + ␧n兴
n=1
shows that there is a significant relation between these parameters; (5)
that is, the higher the sea state Hs, the more skewed or nonlinear
it is. The relationship between these parameters was also repre- where (␨a)n, kn and ␧n are, respectively, the amplitude, wave num-
sented by regressions in Guedes Soares et al. (2003). ber, and the random phase angle of the harmonic component n,
Kurtosis is the other statistic that may characterize the nonlin- and (␻e)n is the encounter frequency of the harmonic component
earity of sea states. It was concluded by Guedes Soares et al. n. The random phase angles are uniformly distributed between 0
(2004) that this parameter is not correlated with any other except and 2␲. For deep water, the dispersion relation that relates the
the abnormality index. The present data set can be approximated by: wave frequency and wave number is (␻0)n2 ⳱ kng.
The exciting forces due to the incident waves are decomposed
AI = 0.5325 ␥4 + 1.6252 (4) into a diffraction part and the Froude-Krylov part. The diffraction
part, which is related to the scattering of the incident wave field
The two studies of Guedes Soares et al. (2003, 2004) were based
due to the presence of the moving ship, is kept linear. It results
on limited data sets from the North Sea and produced results that
from the solution of the hydrodynamic problem of the ship ad-
are more important with respect to identifying the significance of
vancing with constant speed through the incident waves and re-
correlations among various parameters than for the values of the
strained at her mean position. Since this is a linear problem and the
derived regression coefficients. Although the degree of correlation
exciting waves are known a priori, it can be solved in the fre-
may be maintained in other geographical areas, the values of the
quency domain and the resulting transfer functions may be used to
regression coefficients will certainly vary.
generate a time history of the diffraction heave force and pitch
However, it was shown that it is possible to relate Hs with such
moment.
parameters as the coefficient of skewness and of kurtosis, which
In irregular waves the incident wave field pressure results from
describe the nonlinearity of sea states, and in turn with such pa-
the superposition of all the harmonic components used to represent
rameters as abnormality index, vertical asymmetry, and steepness,
that wave. Since it is assumed that the diffraction forces are linear,
which describe the shape of the largest waves in the sea state.
in irregular waves they are calculated by superimposing the dif-
Therefore, once the extreme sea states for design are specified, it
fraction contribution from every harmonic component defining the
is possible to determine the characteristics of individual large
irregular wave. The Froude-Krylov part is related to the incident
waves that would be expected to occur in those sea states. These
wave potential and results from the integration at each time step of
would be the waves considered in the analysis described hereafter.
the associated pressure over the wetted surface of the hull under
the undisturbed wave profile.
3. Prediction of global structural loads induced by Within linear theory of harmonic waves, the wave pressure is
deterministic wave traces defined up to the mean waterline (z ⳱ 0) and does not extend
upward. For this reason, an approximation is assumed to account
3.1. Time domain response method for the wave pressure above the mean waterline when the crests
pass through the hull. The pressure is zero at the free surface, and
A nonlinear time domain strip method is used to calculate the it is assumed to be hydrostatic between the free surface and z ⳱ 0.
wave-induced vertical bending moment on a containership ad- The hydrostatic force and moment are calculated at each time
vancing through a deterministic wave trace that includes a very step by integration of the hydrostatic pressure over the wetted hull
large abnormal wave. The response method assumes that the non- under the undisturbed wave profile. The radiation forces, which
linear contribution for the vertical bending moment is dominated are calculated using a strip method, are represented in the time
by hydrostatic and Froude-Krylov forces; thus, these components domain by infinite frequency added masses, radiation restoring
depend on the instantaneous hull wetted surface. Radiation and coefficients, and convolution integrals of memory functions.
diffraction forces are linear. Additionally, green water loads on the The radiation restoring forces, associated with the restoring co-
deck, which contribute to the calculation of motions and global efficients, represent a correction to the hydrodynamic steady
loads, are represented by the momentum method. A detailed pre- forces acting on the ship due to the steady flow. The convolution
sentation of the method is given by Fonseca and Guedes Soares integrals represent the effects of the whole past history of the
(1998a, 1998b), and its performance compared with measure- motion accounting for the memory effects due to the radiated
ments can be seen in Fonseca and Guedes Soares (2004a, 2004b), waves. The memory functions and the radiation restoring coeffi-
for example. cients are obtained by relating the radiation forces in the time
Following St. Denis and Pierson (1953), the wave elevation of domain and in the frequency domain by means of Fourier analysis.
an irregular and stationary sea state may be represented by an The vertical forces associated with the green water on deck,

MARCH 2008 JOURNAL OF SHIP RESEARCH 33

which is considered to occur when the relative motion is larger time domain method presented in the former section. Basically, a
than the freeboard, are calculated using the momentum method simulation is carried out with the ship advancing through a wave
(Buchner 1995). The mass of water on the deck is proportional to trace that includes the “abnormal wave” and has the same signif-
the height of water on the deck, which is given by the difference icant wave height and mean wave period as the one measured. If
between the relative motion and the freeboard of the ship. The the ship is not advancing (zero forward speed), then at the midship
effect of water on deck is accounted for on the global wave fre- position the simulated wave trace is similar to the measured one.
quency loads, but the transient vibratory responses that may be As a first step to calculate the wave exciting forces, it is nec-
associated with the impact of waves on the deck are not consid- essary to derive a representation in the time and space domains of
ered. the incident wave field that is consistent with the time history of
The method does not account either for the transient vibration the wave elevation defined at a particular point. The time record
response associated with slamming loads. However, this is not was collected from a stationary ultrasonic sensor measuring the
important in the context since their effect on the global maximum distance between its location in a platform and the sea surface.
vertical loads at midship are in general relatively small. They are This time signal was inverted in order to obtain relative wave
felt via elastic response when the bow reenters the free surface, elevation and was transformed from time to frequency domain by
whereas the maximum midship bending occurs approximately a fast Fourier transform (FFT) algorithm. The frequency domain
when the bow is fully submerged; thus, the maximum whipping representation of the time signal is then used to simulate the origi-
moment and maximum sagging moment are in general out of nal time history of the wave elevation. With the wave trace de-
phase. These vibratory responses, which can be captured by an composed into harmonics and assuming linear superposition, it is
elastic model of the ship, are in a frequency higher than those of possible calculate the linear diffraction forces and also the non-
interest in this study.
linear Froude-Krylov forces. Details of the procedure may be
The wave-induced dynamic structural loads at a cross section
found in Fonseca et al. (2001).
are given by the difference between the inertia forces and the sum
Figure 1 presents a comparison between the measured wave
of the hydrodynamic forces acting on the part of the hull forward
elevation (symbols) and the simulated wave elevation (line) for a
of that section. The formulation to calculate all hydrodynamic
history of 4 minutes, and 100 harmonics were used for the simu-
contributions for the loads is consistent with the formulation ap-
lation. If the objective is to assess the ship responses to the large
plied to solve the unsteady motion problem. The convention for
wave only, then a simulation of around 30 to 60 seconds is enough
the loads is such that the hogging bending moment is positive.
A comprehensive experimental program was carried out with a to represent the transient effects of the ship responses to the waves
model of a containership advancing in regular and irregular waves, encountered before the large wave. Using smaller duration time
with the objective of acquiring nonlinear experimental data to traces reduces the computational effort. The figure shows the
compare with results from the computer code applying the de- “freak” wave occurring at a time instant around t ⳱ 265 seconds.
scribed methodology (Fonseca & Guedes Soares 2004a, 2004b). It The presented time history of wave elevation was measured on
was concluded that the computer code is able to represent the a fixed point in space; it can be used to calculate the dynamic
nonlinear effects identified in the experiments. responses of stationary vessels, such as FPSO ships. It can be
Some other computer codes are available to perform these types assumed, for instance, that this wave elevation is felt at midship.
of calculations. The Committee VI.1 of the International Ship and It may also be interesting to use part of the wave record informa-
Offshore Structures Committee (Jensen et al. 2000) has reviewed tion to calculate the wave-induced loads on advancing ships sub-
the available methods and concluded that for practical applications jected to particular events. As an example, it is interesting to
the methods that were more appropriate are the ones based on an estimate the ship responses to the very large wave occurring at t ⳱
approach similar to the one sketched above. Examples of such 265 seconds. To do so, one has to start the simulation with the
methods are those proposed by Xia et al. (1998), Watanabe and advancing ship starting from a specific position in space and pass-
Sawada (1986), and Tao and Incecik (1998). A comparison of the ing through the position of the wave sensor at the exact instant
predictions of these types of codes can be found in Watanabe and when the wave crest occurs.
Guedes Soares (1999). The methodology to produce the wave field that the ship en-
counters as it approaches the point in space where the big wave
3.2. Simulation of wave time traces appears is presented in Fonseca et al. (2001). It is based on the
wave’s dispersion relation. In fact, the wave field is the same as
In this paper the ship responses, and in particular the structural the one corresponding to the simulation in Fig. 1. However, the
loads, are calculated applying a predefined time series of wave wave trace on the ship reference system is different.
elevation that consists of a 20-minute wave elevation record from This is illustrated in Fig. 2, where the simulated wave elevation
a severe storm that occurred in the Central North Sea and was is represented on a reference system advancing at 10.36 m/s (ship
reported for the first time by Haver and Karunakaran (1998). This speed), that crosses the measurement point exactly at the instant
interesting wave trace was measured by a laser sensor on the when the very large wave crest occurs. The symbols represent the
Draupner jacket platform on January 1, 1995. It includes a very wave elevation measured at the point fixed in space, and the line
large wave crest that can be considered an abnormal wave accord- stands for the simulation in the ship reference system.
ing to the presently accepted definitions. The objective of this When the ship is considered to be advancing at different speeds,
study is to assess how the ship behaves in such an extreme wave the simulated time trace does not reproduce exactly the time trace
and how the induced structural loads compare to the design wave before and after the largest wave, but it reproduces the largest
loads calculated from traditional methods. wave well. Since sea states are random in nature, it is not critical
The motions and loads are calculated using a code based on the that the time trace before and after is precisely reproduced as long

34 MARCH 2008 JOURNAL OF SHIP RESEARCH

Fig. 1 Comparison between measured (symbols) and simulated (line) wave elevations. A history of 4 minutes was Fourier analyzed, and 100
harmonics were used for simulation. From Fonseca et al. (2001)

as its statistical properties are well reproduced, as is indeed the where ␴2 is the variance of the process. The variance can be
case. obtained from the spectrum by integration:

兰 S共␻兲d␻
⬁
4. Long-term predictions ␴2 = (8)
0

The approach for calculating long-term cumulative probability which is applicable to both the input and the response spectrum.
distributions of ship wave-induced responses is a well-defined and If the predictions are made for periods of longer duration, then
accepted one. It was initially proposed by Fukuda (1967) and the sea state cannot be considered stationary and the solution is to
applied by Guedes Soares and Moan (1991) among others. There consider a succession of periods of stationary sea states during
are various procedures for determining the functions needed for which the process is defined by an uncertain value of the variance.
the calculation, but the usual basic assumptions are the linearity of Therefore, the Rayleigh distribution becomes conditioned on the
the ship responses with respect to the exciting waves and the value of the variance, and the marginal distribution must be ob-
Gaussian properties of the sea state. With these premises linear tained by integrating over all values of the variance, weighted by
spectral analysis may be applied to derive simple statistics of the their probability density function f(␴). Thus, the probability of
sea state and ship responses. exceeding an amplitude r in a long term is given by (Fukuda 1967,
The first step in the calculation procedure is to determine the Guedes Soares & Moan 1991):
response transfer function, or amplitude response as a function of
the wave frequency. Then the response spectrum SR(␻) is obtained
兰 Q 共r|␴兲 f 共␴兲d␴
⬁
from the input wave spectrum Sw(␻) and the transfer function QL共r兲 = S (9)
0
H(␻):
where QS is the short-term Rayleigh distribution given by (7).
SR共␻兲 = Sw共␻兲|H共␻兲|2 (6)
The kernel in (9) has been modified to account for the fact that,
The sea state is modeled as a stationary, zero mean, Gaussian when the time spent in each sea state is equal but the average
process, and because the responses are linear, the response process response period is different, a ship will get a different number of
will be described by the same model. This together with the as- cycles in different sea states. This is achieved by multiplying the
sumption of a narrow band process implies that a Rayleigh dis- original kernel with the following condition-dependent weighing
tribution describes the amplitudes or the peaks of the processes factor:
(Longuet-Higgins 1952), according to which the probability of
exceeding the level r is given by: T2
w= (10)

冉 冊
T2|␴
r2
QS共r兲 = exp − (7)
2␴2 where the global average response period, T2, is given by:

Fig. 2 Measured wave time trace (symbols) and simulated wave elevation represented on the ship reference system with constant forward speed
(line). From Fonseca et al. (2001)

MARCH 2008 JOURNAL OF SHIP RESEARCH 35

 Fig. 3 Body lines of the S-175 containership

T2 = 冉兰 ␴
1
T2|␴
f 共␴兲d␴ 冊 (11)
5. Calculation example

5.1. Structural loads induced by the abnormal wave

and the condition average response is

冑
This section presents results from time domain simulations of
m0 the motions and structural loads on a containership advancing in
T2|␴ = 2␲ . head waves. The excitation is given by deterministic wave traces
m2
that include abnormal waves. Calculations are performed for the
In order to introduce the nonlinear asymmetry of the vertical bend- well-known S-175 containership. Figure 3 shows the ship body
ing moment on the calculation procedure, Guedes Soares (1993) lines, and Table 1 lists the main particulars.
and Guedes Soares and Schellin (1996) proposed a generalization The first results correspond to the ship without advance speed in
of the former method. Their method uses form functions that long crested waves corresponding to the wave trace presented in
transform the linear transfer functions to “pseudo nonlinear trans- Fig. 1. The wave elevation from Fig. 1 is imposed at the longitu-
fer functions” associated with different sea states. The form func- dinal position of the ship center of gravity (nearly midship). Figure
tion depends on the wave amplitude and frequency and may be 4 presents simulations of the wave elevation (␨) at the center of
represented by (Guedes Soares & Schellin 1996): gravity (CG), heave (␰3), pitch (␰5), relative motion at the forward
␸NL共␻, Hs兲 = ␸1共␻兲 ⭈ ␸2共HS兲 (12) perpendicular (␰r), vertical shear force (V3), and vertical bending
moment (M5) at midship.
The corrected transfer function, or the “pseudo nonlinear transfer This simulation was performed for a long interval of time, 20
function,” becomes: minutes, and 300 harmonics were used to reconstruct the wave
trace and therefore to calculate the exciting forces. In fact, the 300
H共␻, HS兲 = ␸NLH共␻兲 (13) harmonics result from a Fourier analysis of 32 minutes that cor-
respond to 20 minutes of record plus 12 minutes of zeros, which
and the response spectrum accounting for the nonlinear effects is: results in the nearest 2n number of points necessary to apply the
FFT.
SR共␻,Hs兲 = Sw共␻兲 ⭈ H2共␻, Hs兲 (14) To better assess the importance of nonlinear effects, nonlinear
simulations (continuous lines) are plotted together with corre-
From this point on, the method to calculate short-term and long-
term distributions is the same as presented before.
In Guedes Soares and Schellin (1996) the “nonlinear transfer Table 1 Main particulars of the S-175 containership
functions” were calculated by correcting frequency domain results
by the nonlinear effects of changing restoring forces, but another Length between perpendiculars Lpp (m) 175.0
possibility is to calculate “nonlinear transfer functions” for a range Beam B (m) 25.40
of wave amplitudes using a time domain code and then calculate Depth D (m) 15.40
the response spectra for all sea states in the scatter diagram using Draft T (m) 9.50
(14). This procedure has been used by Guedes Soares et al. Displacement ⌬ (ton) 24,742
Longitudinal position of CG LCG (m) −2.43
(2006b) to calculate the design wave bending moment on an FPSO
Block coefficient Cb 0.572
unit and is applied here also to calculate the expected maximum
Pitch radius of gyr. Kyy/Lpp 0.24
bending moment on a containership during its operational lifetime.

36 MARCH 2008 JOURNAL OF SHIP RESEARCH

Fig. 4 Simulation of vertical responses in head long crested waves. Fn = 0, significant wave height of 11.4 m, and zero upcrossing mean wave
period of 11.3 seconds. From Fonseca et al. (2001)

MARCH 2008 JOURNAL OF SHIP RESEARCH 37

Fig. 5 Simulation of vertical responses in head long crested waves. Fn = 0.16, significant wave height of 11.4 m, and zero upcrossing mean wave
period of 11.3 seconds

38 MARCH 2008 JOURNAL OF SHIP RESEARCH

 Fig. 6 Sequence of ship position and wave profile (Fn = 0.16)

sponding linear simulations (dashed lines). The graphs show results from a simulation in a long crested wave field correspond-
clearly that the largest responses occur when the ship encounters ing to 4 minutes of the wave trace presented before (Fig. 1). The
the large wave (around t ⳱ 265 seconds). One observes also that 4 minute wave trace includes the abnormal wave. The number of
the nonlinear asymmetry of the vertical motions is relatively harmonics used to represent the wave field is 80, which results in
small, but in particular the vertical bending moment is highly a very accurate numerical wave elevation since the wave trace is
nonlinear since the nonlinear sagging peaks (negative) are much relatively short. The ship advances in head waves, with a reduced
larger than the corresponding linear ones. speed of 13 knots, and it is forced to pass through the point of
The interest of the methodology presented here is to calculate wave measurement in space at the instant when the very large
the ship’s responses and, in particular, the structural loads induced wave crest occurs. The symbols in Fig. 5 are the same as presented
by predefined wave traces that include abnormal waves. In this before. Additionally, the green water on deck vertical force per
case relatively short simulations are required. Figure 5 presents the meter at the forward perpendicular is included and represented by

MARCH 2008 JOURNAL OF SHIP RESEARCH 39

 Fig. 7 Simulated wave traces at one point fixed in space and with different wave slopes

Fig. 8 Simulation of wave elevation contours (spatial distribution). Three wave slopes of the large wave are considered.

fgw. Continuous lines represent nonlinear simulated results of the before the wave is extremely steep (t ⳱ 162 seconds). This is in
ship responses, and dashed lines stand for linear results. accordance with the reports from some captains who often say that
The graphs show that the ship motions are slightly nonlinear such abnormal waves look like big walls of water.
around the abnormal wave, since the nonlinear simulations present It is of interest to verify if the numerical model is sensitive to
larger peaks compared to the linear ones. The forces due to water different slopes of the large waves, making it possible to deter-
on deck at the forward perpendicular reach a very large value, mine if the vertical responses are significantly different for waves
which is higher than 2,000 kN/m. This occurs when the ship with the same height but different slopes. In order to do this, the
immerses the bow deeply into the water after the passing through measured wave record was modified around the large wave crest,
the large wave. The estimated height of water on deck is close to between t ⳱ 140 seconds and t ⳱180 seconds, shortening and
15 m at the forward perpendicular. also enlarging the wave period. This resulted in modified large
It is also observed that the vertical bending moment at midship wave crests, with respectively increased and decreased wave
is highly asymmetric, with the largest nonlinear sagging peak slopes, but the same wave height. Compared to the original wave
showing magnitudes approximately three times larger than the record, the slope was varied plus and minus 40%. The wave slopes
corresponding linear ones. This peak occurs when the ship im- are defined here as the ratio between the wave height (vertical
merses the bow after encountering the large crest. Finally, it is distance between the large crest and the previous trough) and the
interesting to note that the water on deck vertical forces tends to horizontal distance between the large crest and the previous
reduce the sagging moment, since these forces act downward to trough. The three wave slopes are 0.19, 0.32, and 0.44.
produce a hogging moment. However, in the case of the largest The three simulated wave elevations, at one fixed point in
sagging moment, the green water forces occur slightly later than space, are presented in Fig. 7, where the solid line represents the
the sagging peak; thus, they do not contribute to reducing the simulation of the original record and the dashed line with large
maximum sagging moment. segments stands for the simulation with decreased wave slope.
Figure 6 presents a sequence of the ship position and wave The dashed line with small segments represents the simulation
profile as the ship encounters the abnormal wave and passes with increased slope. It is clear that before t ⳱140 seconds all
through it. The sequence clearly shows that the largest relative simulations are coincident and after t ⳱ 180 seconds the curves
motions at the bow do not occur when the bow encounters the are similar but affected by phase shifts.
abnormal wave. Consequently, the largest structural loads do not The different wave slopes are more clearly observed in Fig. 8,
occur at this instant either, as can be observed in the graphs of Fig. which presents the simulated wave contours in space along the
5. The largest relative motions occur approximately at t ⳱ 169 direction of propagation of the assumed long crested waves, at the
seconds, when the large crest is at the stern. instant when the large wave crest occurs. The origin of the x axis
It is interesting to note that although the wave profile at the time is coincident with the measurement point.
instant when the maximum wave elevation was measured is not Figure 9 presents the vertical ship responses to these three
extremely steep (t ⳱ 164.1 seconds in Fig. 6), in the instants waves, where correspondence between different lines and wave

40 MARCH 2008 JOURNAL OF SHIP RESEARCH

Fig. 9 Simulation of vertical responses in head long crested waves. Comparison of results for three different wave slopes of the large wave. Fn
= 0.16, significant wave height of 11.4 m, and zero upcrossing mean wave period of 11.3 seconds

slopes is the same as described before for the wave elevations. The the design of ships. With this idea in mind, it is interesting to
results show that heave and pitch motions are sensitive to the wave compare the maximum structural loads induced by the abnormal
slope, as the ship tends to follow more the wave for longer wave wave with those resulting from long-term predictions and also
lengths. The relative motions at the bow also feel the differences with the values from classification society rules. Expected maxi-
between the wave slopes, and the consequence is different water mum wave bending moments were calculated using the linear and
on deck forces for different wave slopes. Finally, the vertical nonlinear long-term prediction methods presented in section 4.
bending moment at midship also presents large differences regard- The ship speed is the same as used for the time domain simula-
ing the sagging peaks for the three waves. The resulting values are tions, V ⳱ 13 knots. The wave statistics of the North Atlantic are
presented in Table 2. represented by a scatter diagram prepared by Guedes Soares
Besides large global structural loads, very steep waves, or (1996).
nearly vertical waves, may induce slamming type loads on the ship Figure 10 presents the probability of exceedance of the vertical
sides or on the bow above the waterline, or even on the ship bending moment at midship calculated by the long-term proce-
superstructure and equipment on the deck. Presently, the applied dures. The continuous and dashed lines represent, respectively,
numerical method is not able to represent such type of loads. linear and nonlinear long-term predictions. The nonlinear long-
term predictions correspond to pseudo transfer functions, defined
5.2. Comparisons with design wave bending moments by the sagging peaks, and depend on the wave amplitude. The
horizontal dashed line represents the probability level of 10−8,
It is of interest to assess the possibility of using the loads in- which corresponds approximately to an operational life of 20
duced by very rare abnormal waves as an additional criterion in years. The graph includes also the maximum sagging and hogging

MARCH 2008 JOURNAL OF SHIP RESEARCH 41

 Figure 9 (continued)

peaks obtained from the simulations in the measured abnormal linear long-term calculations for an exceedance probability of 10−8
wave (vertical lines with notation MAW), together with the mini- is 48% higher than the value given by the rules for sagging. The
mum rule wave bending moments as required by the International nonlinear long-term predictions give a sagging moment 2.7 times
Association of Classification Societies (IACS). The numerical val- larger than the rule value. Concerning the time domain simulation
ues of the maximum bending moments are presented in Table 3. in the measured abnormal wave, the maximum nonlinear sagging
The results show that the bending moment predicted by the moment is around 2.1 times larger than the rules one.

Table 2 Maximum bending moment sagging and hogging peaks 6. Conclusions

around the abnormal wave (Fn = 0.16)
The paper considers the possibility of using a predefined time
Sagging Hogging series of wave elevation to calculate global wave-induced struc-
Sagging Hogging Relative Relative tural loads on ships. The procedure can be used as an additional
(MNm) (MNm) Increase Increase design criteria aimed at considering the effects of accidental situ-
ations associated with abnormal waves.
−40% slope 1,591.3 622.77 −27% −7.9%
A method to calculate the ship responses to a deterministic and
Measured abnormal wave 2,179.6 675.84 0 0
+40% slope 1,759.9 577.79 −19.3% −14.5%
irregular wave time series is described. The method, together with
a time domain code, is applied to calculate the nonlinear responses

42 MARCH 2008 JOURNAL OF SHIP RESEARCH

 Fig. 10 Comparison of wave-induced bending moment at midship (Fn = 0.25)

of a containership to a 20-minute wave trace measured in the been obtained and a summary of the main conclusions from those
Central North Sea during a storm. This is an interesting wave trace results can be found in Guedes Soares et al. (2006a).
because it includes a very large wave crest, which corresponds to
a wave height of approximately 26 meters.
Acknowledgments
The results show that the vertical motions are weakly nonlinear,
while the vertical bending moment at midship has strong nonlinear This work has been performed in the scope of the research
characteristics since the sagging peaks are much larger than the project Rogue Waves—Forecast and Impact on Marine Structures
hogging peaks, and also larger than the corresponding linear sag- (MAXWAVE), partially funded by the European Commission,
ging peaks. under the program Energy, Environment and Sustainable Devel-
The influence of the wave slope on the vertical responses was opment (contract no. EVK3:2000–00544). Work of the third au-
investigated. It was observed that the wave slope affects the heave thor has been financed by the Portuguese Foundation for Science
and pitch motions, since the ship tends to follow the motion of the and Technology (Fundação para a Ciência e a Tecnologia) under
longer waves. The vertical bending moment depends on the rela- contract SFRH/BD/10527/2002.
tive motions, and it was observed that around the large wave crest
the resulting bending moment for different wave slopes is differ-
ent. References
Since this work was completed and submitted, new results have
ADEGEEST, L., BRAATHEN, A., AND VADA, T. 1998 Evaluation of methods
for estimation of extreme non-linear ship responses based on numerical
simulations and model tests, Proceedings, 22nd Symposium on Naval Hy-
Table 3 Comparison of maximum wave bending moments drodynamics, August, Washington, DC, vol. 1, 70–84.
(Fn = 0.16) BUCHNER, B. 1995 On the impact of green water loading on ship and off-
shore unit design, Proceedings, 6th International Symposium on Practical
Design of Ships and Mobile Units (PRADS ’95), September 17–22, Seoul,
Sagging Hogging Korea, Kim, H., and Lee, J. W., editors, The Society of Naval Architects of
Normalized Normalized Korea, vol. 1, 430–443.
Sagging Hogging by Rule by Rule CLAUSS, G. F., SCHMITTNER, C., JANOU, H., GUEDES SOARES, C., FONSECA, N.,
(MNm) (MNm) Value Value AND PASCOAL, R. 2004 Bending moments on a FPSO in rogue waves,
Proceedings, 23rd International Conference on Offshore Mechanics and
Linear (p ⳱ 10−8, 20 Arctic Engineering (OMAE ’04), ASME, June 20–25, Vancouver, Canada,
years) 1,503.9 1,503.9 1.48 1.91 Paper OMAE2004–51504.
Nonlinear (p ⳱ 10−8, 20 CLAUSS, G. F., SCHMITTNER, C., AND STUTZ, K. 2002 Time-domain inves-
years) 2,730.0 — 2.69 — tigation of a semisubmersible in rogue waves, Proceedings, 21st Interna-
Rules (IACS) 1,016.3 787.9 1.0 1.0 tional Conference on Offshore Mechanics and Arctic Engineering (OMAE
Measured abnormal wave 2,179.6 675.84 2.14 0.86 ’02), ASME, June 23–28, Oslo, Norway, Paper OMAE2002–28450.
FAULKNER, D., AND BUCKLEY, W. H. 1997 Critical survival conditions for

MARCH 2008 JOURNAL OF SHIP RESEARCH 43

 ship design, Proceedings, International Conference on Design and Opera- HAVER, S., AND KARUNAKARAN, D. 1998 Probabilistic description of crest
tion for Abnormal Conditions, RINA, October 21–22, Glasgow, UK, paper heights of ocean waves, Proceedings, 5th International Workshop on Wave
no. 6, 1–25. Hindcasting and Forecasting, January 26–30, Melbourne, FL.
FONSECA, N., AND GUEDES SOARES, C. 1998a Time-domain analysis of large- HENDERSON, K. L., PEREGRINE, D. H., AND DOLD, J. W. 1999 Unsteady wa-
amplitude vertical motions and wave loads, JOURNAL OF SHIP RESEARCH, 42, ter wave modulations: fully nonlinear solutions and comparison with the
2, 100–113. nonlinear Schrodinger equation, Wave Motion, 29, 341–361.
FONSECA, N., AND GUEDES SOARES, C. 1998b Nonlinear wave induced re- JENSEN, J. J., BECK, R. F., DU, S., FALTINSEN, O. M., FONSECA, N., RIZZUTO, E.,
sponses of ships in irregular seas, Proceedings, 12th International Confer- STREDULINSKY, D., AND WATANABE, I. 2000 Extreme hull girder loading,
ence on Offshore Mechanics and Arctic Engineering (OMAE ’98), ASME, Proceedings, 14th International Ship and Offshore Structures Congress
July 5–9, Lisbon, Portugal, paper no. 98 0446. (ISSC 2000), October 2–6, Nagasaki, Japan, Ohtsubo, H., and Sumi, Y.,
FONSECA, N., AND GUEDES SOARES, C. 2004a Experimental investigation of editors, Elsevier Science, vol. 2, 263–320.
the nonlinear effects on the vertical ship motions and loads of a container- KJELDSEN, S. P. 1997 Examples of heavy weather damages caused by giant
ship in regular waves, JOURNAL OF SHIP RESEARCH, 48, 2, 118–147. waves, SNAJ Bulletin of the Society of Naval Architects of Japan, 828,
FONSECA, N., AND GUEDES SOARES, C. 2004b Experimental investigation of 744–748.
the nonlinear effects on the vertical ship motions and loads of a container- LAVRENOV, I. 1998 The wave energy concentration at the Agulhas current
ship in irregular waves, JOURNAL OF SHIP RESEARCH, 48, 2, 147–167. of South Africa, Natural Hazards, 17, 117–127.
FONSECA, N., GUEDES SOARES, C., AND PASCOAL, R. 2001 Prediction of LONGUET-HIGGINS, M. S. 1952 On the statistical distribution of the heights
ship dynamic loads in ship in heavy weather, Proceedings, Conference on of sea waves, Journal of Marine Research, 11, 3, 245–260.
Design and Operation for Abnormal Conditions II, RINA, November 6–7, PASTOOR, W., HELMERS, J. B., AND BITNER-GREGERSEN, E. 2003 Time
London, United Kingdom, 169–182. simulation of ocean going structures in extreme waves, Proceedings, 22nd
FUKUDA, J. 1967 Theoretical determination of design wave bending mo- International Conference on Offshore Mechanics and Arctic Engineering
ments, Japan Shipbuilding and Marine Engineering, 2, 3, 12–22. (OMAE ’03), June 8–13, Cancun, Mexico, ASME, paper no. OMAE03–
GUEDES SOARES, C. 1993 Long-term distribution of non-linear wave in- 37490.
duced vertical bending moments, Marine Structures, 6, 475–483. PELINOVSKY, E., AND KHARIF, C. 2000 Simplified model of the freak wave
GUEDES SOARES, C. 1996 On the definition of rule requirements for wave formation from the random wave field, Proceedings, 15th International
induced vertical bending moments, Marine Structures, 9, 3–4, 409–426. Workshop on Water Waves and Floating Bodies, February 27–March 1,
GUEDES SOARES, C., BITNER-GREGERSEN, E., AND ANTÃO, P. 2001 Analysis Caesarea, Israel, 142–145.
of the frequency of ship accidents under severe North Atlantic weather SAGLI, G., AND MOAN, T. 2001 Application of contour line method to es-
conditions, Proceedings, Conference on Design and Operation for Abnor- timate extreme ship hull loads considering operational restrictions, JOURNAL
mal Conditions II, RINA, November 6–7, London, United Kingdom, 221– OF SHIP RESEARCH, 45, 3, 228–240.
230. ST. DENIS, M., AND PIERSON, W.J. 1953 On the motions of ships in con-
GUEDES SOARES, C., CHERNEVA, Z., AND ANTÃO, E. M. 2003 Char- fused seas. Transactions of the Society of Naval Architects and Marine
acteristics of abnormal waves in North Sea storm sea states, Applied Ocean Engineers, 61, 280–357.
Research, 25, 6, 337–344. TAO, Z., AND INCECIK, A. 1998 Time domain simulation of vertical ship
GUEDES SOARES, C., CHERNEVA, Z., AND ANTÃO, E. M. 2004 Steepness and motions and loads in regular head waves, Proceedings, 12th International
asymmetry of the largest waves in storm sea states, Ocean Engineering, 31, Conference on Offshore Mechanics and Arctic Engineering (OMAE ’98),
1147–1167. July 5–9, Lisbon, Portugal, ASME, paper no. OMAE-98591.
GUEDES SOARES, C., DOGLIANI, M., OSTERGAARD, C., PARMENTIER, G., AND TOMITA, H., AND KAWAMURA, T. 2000 Statistical analysis and inference
PEDERSEN, P. T. 1996 Reliability based ship structural design, Transac- from the in-situ data of the Sea of Japan with reference to abnormal and/or
tions of the Society of Naval Architects and Marine Engineers, 104, 357– freak waves, Proceedings, 10th ISOPE Conference, May 28–June 2, Seattle,
389. WA, 116–122.
GUEDES SOARES, C., FONSECA, N., AND PASCOAL, R. 2006a Characteristics TORHAUG, R., WINTERSTEIN, S. R., AND BRAATHEN, A. 1998 Non-linear
of abnormal waves and their induced global load effects in marine struc- ship loads: stochastic models for extreme response, JOURNAL OF SHIP RE-
tures, Transactions of the Society of Naval Architects and Marine Engineers, SEARCH, 43, 1, 46–55.
114. TRULSEN, K., AND DYSTHE, K. B. 1997 Water wave kinematics computed
GUEDES SOARES, C., FONSECA, N., PASCOAL, R., CLAUSS, G. F., SCHMITTNER, with nonlinear Schrödinger theory, Proceedings, 16th International Offshore
C., AND JANOU, H. 2006b Analysis of the wave induced loads on a FPSO Mechanics and Arctic Engineering Conference (OMAE), ASME, April 13–
due to abnormal waves, Journal of Offshore Mechanics and Arctic Engi- 17, Yokohama, Japan, vol. 1, 7–18.
neering, 28, 241–247. WATANABE, I., AND GUEDES SOARES, C. 1999 Comparative study on the
GUEDES SOARES, C., AND MOAN, T. 1991 Modern uncertainty in the long time-domain analysis of non-linear ship motions and loads, Marine Struc-
term distribution of wave induced bending moments for fatigue design of tures, 12, 3, 153–170.
ship structures, Marine Structures, 4, 294–315. WATANABE, I., AND SAWADA, H. 1986 Effects of elastic responses to the
GUEDES SOARES, C., AND SCHELLIN, T. E. 1996 Long-term distribution of longitudinal bending moment in two-directional irregular waves, Naval Ar-
non-linear wave induced vertical bending moments on a containership, Ma- chitecture and Ocean Engineering, 24, 91–102.
rine Structures, 9, 3–4, 333–352. WHITE, B. S., AND FORNBERG, B. 1998 On the chance of freak waves at sea,
GUEDES SOARES, C., AND SCHELLIN, T. E. 1998 Nonlinear effects on long- Journal of Fluid Mechanics, 355, 113–138.
term distributions of wave-induced loads for tankers, Journal of Offshore XIA, J., WANG, Z., AND JENSEN, J. J. 1998 Non-linear wave loads and ship
Mechanics and Arctic Engineering, 120, 2, 65–70. responses by a time-domain strip theory, Marine Structures, 11, 3, 101–123.

44 MARCH 2008 JOURNAL OF SHIP RESEARCH