Download as pdf or txt
Download as pdf or txt
You are on page 1of 12

Version of Record: https://www.sciencedirect.

com/science/article/pii/S0029801819302999
Manuscript_c4129b9a6d6bfde1da979be11e81ee11

Approximation of capsize probability using a Roll Exceedance


(RE) probability with a threshold chosen in roll phase plane

Romain Macé, DGA Hydrodynamics, romain.mace@intradef.gouv.fr


Jean-Yves Billard, French Naval Academy (École Navale), jean-yves.billard@ecole-navale.fr
Guillaume Lannel, DGA Hydrodynamics, guillaume.lannel@intradef.gouv.fr
Jean-François Leguen, DGA Hydrodynamics, jean-françois.leguen@intradef.gouv.fr

ABSTRACT

The paper investigates the correlation of capsizing and non-capsizing trajectories in an


attempt to identify a relevant parameter for capsize probability assessment. The probability of
capsize of decommissioned French naval ships, a Frigate and helicopter carrier Jeanne D’Arc,
have been determined by a Monte Carlo procedure. Despite the computation cases are limited
to two ships and two sea states, data sets generated tend to show in trajectory phase plane a
convergence of the envelopes of non-capsized trajectories and of capsizing cycles. The
presumed convergence of the envelopes would indicate a correlation between capsized and
non-capsized trajectories. This suggest a faster and more cost effective probability assessment
method is conceivable, using a reduced number of computations and set of data compared to
Monte Carlo Method. Statistical analysis of capsizing cycles shows a normal law can
approximate properly the roll rates distribution before capsize, particularly the roll rate at zero
crossing and the roll rate extrema after zero crossing. However, partial match is obtained
between these criteria associated to the probability obtained from Monte-Carlo method and
the curve of probability of roll rate exceedance obtained with non-capsizing cycles.
Additional computations, that widen the scope of ships and sea states, are in progress.

Keywords: Capsize probability, Monte-Carlo, roll exceedance, attraction basin.

1 Introduction
At first order, capsize probability gives a precise evaluation of the meteorological risk
encompassed by a ship. For given loading conditions, heading, speed and sea state, the
determination of this probability is of major concern and remains an open problem addressed
for a long time in the literature as for example by Rainey and Thompson (1991) or by
Falzarano, Shaw and Troesh (1992). Monte-Carlo method and its derivate give a good
evaluation of this probability, but the associated computation cost are very large and
incompatible with many applications of capsize probability assessment. Beside, probability
of roll exceedance can be evaluated rapidly and used to define a global probability of capsize
for the all life cycle of the ship. Unfortunately, the correlation between those two approaches
is not straightforward. As Er and Iu (1999) or Maki (2017) compare capsize probability with
a Monte-Carlo method, the present paper uses multiple-DOF numerical simulation of a ship in
beam waves instead of analytical results. The paper investigates the correlation of capsizing
and non-capsizing trajectories in the phase plane in an attempt to identify a relevant parameter
for capsize probability assessment using statistical methods as those described by Leadbetter,
Lindgren and Rootzen (2012).

© 2019 published by Elsevier. This manuscript is made available under the Elsevier user license
https://www.elsevier.com/open-access/userlicense/1.0/
2 Ship and numerical tools

The ships.
The study considers two ships: a French frigate designed in the seventies and by now
decommissioned, and the helicopter carrier Jeanne d’Arc (JDA) that was decommissioned in
2010. Data availability and an interest in significantly different stiffness between the ships
considered were crucial in the choice of these vessels as study cases. The natural roll periods
of these ships are 15.8 s and 12.2 s for the frigate and JDA respectively.

Multi-DOF computation.

Ship trajectory is simulated with FREDYN V14, which is a 6-DOF time domain
computational code developed by MARIN in the frame of the CRNAV. The ship evolves in
Beam Sea with frozen yaw and sway. Spectral sea state is simulated using Pierson
Moskowitz spectrum that combines 80 wave components. No wind is taken into account. For
each simulation, the trajectory duration to compute is chosen to be representative of those of a
sea state, and thus set at 20.000 s. Numerical time step is set to 0.5 s for all the simulations.
Initial conditions are zero heel and zero roll rate. The compliance with these initial conditions
while wave excitation starts is achieved using a FREDYN feature that steadily increases wave
height over a user-defined duration set to 20 s in our case.

The Monte Carlo procedure described further in this paper requires the generation of distinct
wave signals for the same sea state. For this purpose, each simulation uses the same wave
component frequencies with specific distribution of phase shifts randomly generated for each
simulation using a seed number that allow the exact replication of each computation.

Beside, statistical analysis of roll trajectories requires avoidance of wave train repetition in a
given simulation. The use of FREDYN repetition time feature prevents such repetition within
a user-defined computed time-period, which is set to 50000 s in this study. The low level of
wave signal autocorrelations, Figure 1, illustrates the absence of obvious correlation within a
given sea signal.

SS1 SS2
Figure 1 : Example of autocorrelation of a wave signal generated

In this study, the contradictory needs for capturing large sets of non-linear behaviours,
capsizing behaviour and non-capsizing behaviour, within the user defined computation time
and with a given sea state, implied to choose specifically for each ship a strong sea state with
a peak period very close to its natural roll period. Sea States SS1 and SS2 were respectively
chosen for the frigate and for JDA. However, both ships are evaluated in both sea states SS1
and SS2 in order to compare ship behaviours and sea state influence. Capsize is arbitrarily
defined as the reaching of 90° roll angle, and simulation is stopped when it occurs. More
details on FREDYN code can be found in Ypma (2012).

Table 1 : Sea state cases for this study

Sea state HS [m] TP [s]

SS1 11 9.5

SS2 15.2 12.5

1-DOF model in still water


In the purpose of comparing the phase trajectories obtained from FREDYN multi-DOF
computation to their respective attraction basins, a fourth order Runge-Kutta algorithm is used
for solving the one-dimension roll equation in still water.
| | 0 (1)

Inertia term , linear damping term and quadratic damping term are
preliminarily determined by the analysis of a FREDYN computation of roll decay in still
water. Righting arm GZ is approximated in a polynomial up to order 5 for JDA, form ∙
∙ ∙ , on the basis of FREDYN hydrostatic computations, figure 2.

1 2

0.8 1.5
0.6 1
GZ [m]
GZ [m]

0.4
0.5
0.2
0
0
0 10 20 30 40 50 60 70 80 90 100 110
0 20 40 60 80 -0.5
-0.2
-0.4 -1
Gîte [°] Gîte [°]
FREDYN Polynomial fitting GZ [m] Polynôme

Frigate JDA
Figure 2 : Righter lever curves

3 Capsize probability

Capsize probability defined by Monte-Carlo computations

Capsize probability assessment of a ship in a given sea state by a Monte Carlo procedure
requires a large set of time domain computations. Therefore, in our case, for a given sea state,
with the same initial conditions excepting a variation of the wave generator phase, a set of
Ntotal = 1.000 different simulations is performed for each ship.
In this study, the probability of survival PSt in an observation duration t < 20.000 s is given
by the ratio of the amount of simulations Nt which have not ended in a capsize before the time
t:
Nt
PCt = 1 − PS t = 1 − (1)
N total

As shown by Belenky and Sevastianov (2007), the time evolution of PC follows exponential
law, which means, basically, that the longer the ship is exposed, the higher the probability of
capsize:
PC = 1 − e −αt (2)

1
αs = 1.18 10-4 / s
0.9

0.8
Probability of capsize [-]

0.7

0.6

0.5 Computational data


Modelisation
0.4
tangent
0.3

0.2

0.1

0
0 5000 10000 15000 20000 25000 30000 35000 40000
total number of comutations = 1000 Time [s]
with 887 commutations with capsize

Figure 3: Evolution of capsize probability of the frigate as function of time

The value of αs is determined by a least square method based on the simulation points. On
Figure 3 the probability of capsize has been plotted as a function of time. Dots represent the
values of the probability PCt and the curve is the exponential fit of those points, PC. The main
interest of this formulation is that the probability of capsize is described by only one
parameter, αs, which is also the tangent at t = 0 s to the curve.

Similarly, it is possible to plot the curve, Figure 4, versus the number of roll cycles
experienced by the ship. This representation allows a direct comparison with the roll
exceedance probability described hereafter.
1

0.9
αc = 1.44 10-3 / cycles
0.8

Probability of capsize [-]


0.7

0.6

0.5 Computational data


Modelisation
0.4
Tangent
0.3

0.2

0.1

0
0 500 1000 1500 2000 2500
total number of comutations = 1000 Number of cycles [-]
with 887 commutations with capsize

Figure 4: Evolution of capsize probability of the frigate as function number of roll cycle.

It is to notice that average cycle duration is close to the ship roll period at small angle in calm
water (13.75 s for the Frigate and 12.2 s for JDA). As Belenky and Sevastianov (2007)
demonstrates, the probability curve as a function of cycle number is more or less a re-scaling
of the one that is a function of observation duration, the scale factor being the mean period.
Hence the tangents of these curves at t=0 s give the probability by cycle or by second, with a
ratio between αc and αs equal to the mean period with an accuracy better than 2%. Table 2
lists the results of Monte-Carlo method for the computation cases performed in this study. For
later development it distinguishes results from the sets of capsizes from up-wave side of the
ship, down-wave side, and both sides. For the frigate, the number of capsizes up-wave side is
too small to consider the case as converged. Its why the values od αs and αc have not been
reported.

Table 2 : Capsize probability per cycle or per second

Frigate JDA
SS1 SS2 SS1 SS2

Down- Down- Down- Down-


wave Up-wave Both Up-wave Both Up-wave
wave wave wave Both sides
side sides side sides side
side side side side

1,44E-03 5,90E-03 7,16E-03 4,87E-05 4,24E-04 5,09E-04 1,56E-04 7,34E-04 1,10E-03

Not 4,47E-
1,18E-04 4,52E-04 5,08E-04 3,73E-05 4,53E-05 1,36E-05 5,93E-05 9,39E-05
converged 06

12,2 13,1 14,1 10,9 11,4 11,2 11,4 12,4 11,7

Nb 887 952 48 1000 75 587 662 189 613 802


Probability of roll exceedance

For given ship and a sea state, the complete set of roll cycles over the 1000 simulations is
analysed to determine the probability of exceedance of a roll rate threshold X, ! "#.
Roll rate extrema of each side are post processed separately and their weight is determined
following Derbanne et al. (2008). Figure 5 gives an example of exceedance probability curve
for the frigate on sea state 1. This curve gives the probability to exceed a given roll angle
during the next wave, hence this probability is determined for a duration similar to the next
oscillation duration.

1.E+00
-40 -30 -20 -10 0 10 20 30 40
Roll rate [°/s]
1.E-01
Down-wave side Up-wave side

1.E-02

1.E-03

1.E-04

Probability
1.E-05 [-]

1.E-06

Figure 5 : Roll rate exceedance probability, Frigate in SS1 (green dots : numerical simulation using
FREDYN / blue line: welbull distribution, best fit)

4 Analysis of phase plane trajectories


A purpose of this study is to identify on the ship trajectory phase plane a roll or roll rate
threshold which exceedance probability accurately approximate exact Monte-Carlo capsize
probability, keeping in mind that the first probability have a much cheaper computational cost
than the second.
Roll cycles are separated in two sets for analysis: capsizing cycles and non-capsizing cycles.
This enables characterization of the outer envelope of non-capsizing cycles (see Frigate in
SS1 for example in Figure 6) and inner envelope of 95% capsizing cycles (see Frigate in SS2
for example in Figure 7).
In figure 6, bleus dotes represent the temporal traces off all the non-capsizing cycles. A part
of these traces are simulations that do not capsize. The other part is composed of simulations
concluded by capsize, excluding their last roll cycles.
Figure 6 : Outer enveloppe (red) of non capsising trajectories (blue), Frigate in SS1

On the contrary, figure 7 plots the temporal traces of the last roll cycles from the simulations
that end in capsize.

Down-wave side

Figure 7 : Set of capsizing cycles and its 95% inner envelope compared to attraction basin in still water,
Frigate in SS2

Figure 8 and figure 9 compare these envelopes and respective attraction basin for both ships
and sea states. Qualitatively, it seems at first sight that there might be a convergence of
capsizing and non-capsizing envelope to a common zero-crossing roll rate $% ° .

Those lines (the two inner and outer envelopes and the attraction basin) are separatrix that
means, boundary that separates initial conditions leading capsizing from initial condition that
do not lead to capsizing in absence of external excitation. Using a separatrix is equivalent to
classic definition of stability, see chapter 5.1 of Belenky and Sevastianov (2007).
Figure 8 : Envelopes of capsizing and non-capsizing cycles compared to attraction basin in still water,
Frigate in SS1 and SS2

Figure 9 : Envelopes of capsizing and non-capsizing cycles compared to attraction basin in still water,
JDA in SS1 and SS2
To confirm and support this observation with quantitative matter, an analysis of the
distribution of $% ° before capsize sets is conducted and extended to the sets of maximum
observed after the last zero crossing and before capsize. The table 3 lists the corresponding
mean values of $% ° and ' ()* ' before capsize and figure 10 presents the repartition of roll
rate and maximum roll rate prior to capsize. The size of capsize set of the frigate on SS2 is not
statistically converged as it has only 49 samples. However, for others, it turns out that normal
laws determined by least square error minimization method could produce an acceptable
approximation of these distributions, as illustrated in Figure 10. Normal law approximate
gives a cumulated density associated to the roll rates of interest (Figure 10 for $% ° and
' ()* '). It must be underline that the mean value of the distributions seems to be few or not
influenced by sea state level.
Due to their different natural roll periods and righting levers, different dynamic responses are
expected between the frigate and JDA. The frigate is much less stiff than JDA, such as 100%
of its trajectories end in capsize in SS1 and SS2, contrary to JDA that ends up in capsize by
58.7% and 80.2% of the trajectories in SS2 and SS2 respectively. For the same reasons the
frigate capsizes only on down waves side in SS1, and she experiences only few capsizes on up
wave side in SS2 (only 5%). It comes to the extent that JDA gets much energy when rolling
down the waves but not enough to capsize. Consequently, as the ship rolls back up the waves,
this energy converts to high roll rate and produces a large proportion of capsizes on up wave
side, by 51.2% and 61.3% in SS1 and SS2 respectively.

FRIGATE
JDA
Figure 10 : Roll rates at zero crossing prior to capsize

Table 3 : Synthesis of mean values encountered for zero-crossing roll rate and maximum roll rate before
capsize

Frigate JDA
SS1 SS2 SS1 SS2
100%
Up wave Down Up wave Down Up wave Down
Down
side wave side side wave side side wave side
wave side
Number of capsize 887 49 951 512 75 613 189
88,7 4,9 95,1 51,2 7,5 61,3 18,9
% to Ntotal
100 100 58,7 80,2
Noncapsizing cycles
enveloppe 30,2 31,9 -27,4 42,4 -41,9 44,8 -39,9
+ extrema
Capsizing cycles
24,2 27,7 25,9 38,4 32,6 41,7 32,2
mean ,,%-°
mean of'+' maxima 27,5 28,1 28,4 39,4 33,9 42,5 33,4

As said previously, ! "# being the probability to exceed X consequently to the next
wave computed with non-capsizing cycles, it is legitimate to attempt locating the extrema
obtained from non-capsizing envelope and the mean of maximum roll rates before capsize
both associated to Monte-Carlo probability to capsize at the next cycle as shown on Figure 11.
The location matches in JDA case and both ship side. In the Frigate case, the location
surprisingly don’t match on down-wave side, that is statistically converged, and the good
match on up-wave side is in all likelihood a coincidence given the poor statistical
convergence of this subcase.

951
887 capsizes
capsizes
49
capsizes

613
512 capsizes
capsizes 189
75 capsizes
capsizes

Figure 11 : Visual comparison of +./0_234563_7/89:;3 ; => # and +./0_3?@3A5883 ; => # to C + ! D

5 Conclusions
The probability of capsize of the frigate and JDA have been determined by a Monte Carlo
procedure. Despite the computation cases are limited to two ships and two sea states, data sets
generated tend to show in trajectory phase plane a convergence of the envelopes of non-
capsized trajectories and of capsizing cycles. The presumed convergence of the envelopes
would indicate a correlation between capsized and non-capsized computations. This suggest a
faster and more cost effective probability assessment method is conceivable, using a reduced
number of computations and set of data compared to Monte Carlo Method. Statistical analysis
of capsizing cycles shows a normal law can approximate properly the roll rates distribution
before capsize, particularly the roll rate at zero crossing and the roll rate extrema after zero
crossing which could be relevant criteria or variables for a probability assessment. However,
partial match is obtained between these criteria associated to the probability obtained from
Monte-Carlo method and the curve of probability of roll rate exceedance obtained with non-
capsizing cycles.
Additional computations, that widen the scope of ships and sea states, are in progress to
confirm the convergence of the capsizing and non-capsizing trajectories envelopes in phase
plane to relevant values such as $% ° and ' ()* ' parameters, and to setup a faster and more
cost effective probability assessment method compared to Monte-Carlo procedure.
Future work is planned also to improve the methodology by trying to include damping in the
attraction basin and to improve the estimation of uncertainty.

6 Acknowledments
The authors want to acknowledge the financial support French navy and DGA and the
valuable discussions during the meetings of CRNAV and ISSW.

7 References
Belenky, V.L., Sevastianov, N.B., 2007. “Stability and Safety of Ships-Risk of Capsizing”,
second edition. The Society of Naval Architects and Marine Engineers (SNAME), Jersey
City, NJ.

Belenky, V. Weems; K. M. & Lin, W. M., 2016. “Split-time method for estimation of
probability of capsizing caused by pure loss of stability”, Ocean Engineering Vol. 122, pp
333-343.

Derbanne, Q., Leguen J.-F., Dupau, T., Hamel, E., 2008, “Long-Term Non-Linear Bending
Moment Prediction”, Proceedings of OMAE 2008.

Falzarano, J. M., Shaw, S. W. and Troesch, A. W., 1992. “Application of global methods for
analyzing dynamical systems to ship rolling motion and capsizing”, International Journal
of Bifurcation and Chaos, Vol. 2, No 1, pp 101-115.

Leabetter, M. R., Lindgren, G., Rootzen, H., 2012, “Extremes and related properties of
random sequences and processes”, Springer series in statistics, springer-Verlag, New York-
Berlin.

Maki, A., 2017. “Estimation method of the capsizing probability in irregular beam seas using
non-Gaussian probability density function”, Journal of Marine Science and Technology
Vol. 22, pp. 351–360.

Rainey, R. C. T., Thompson, J. M. T., 1991. “The transient capsize diagram – a new method
of quantifying stability in waves”, Journal of Ship Research, Vol. 35, No 1, pp. 58-62.

Ypma E., Harmsen E., 2012. “Development of a New Methodology to Predict the Capsize
Risk of Ships”, Proceedings of STAB 2012, pp. 1-10.

Er GK, Iu V.P., 1999, “Probabilistic Solutions to Nonlinear Random Ship Roll Motion”,
Journal of Engineering Mechanics, Vol. 125, n° 5.

You might also like