Available online at [Link].
com
Procedia IUTAM 6 (2013) 132 – 140
IUTAM Symposium on Multiscale Problems in Stochastic Mechanics 2012
Stochastic averaging of roll-pitch and roll-heave motion
in random seas
Leo Dostala,*, Edwin Kreuzera, Navaratnam Sri Namachchivayab
a
Institute of Mechanics and Ocean Engineering, Hamburg University of Technology,
Eissendorfer Strasse 42, 21073 Hamburg, Germany
b
Department of Aerospace Engineering, University of Illinois at Urbana-Champaign, Urbana-Champaign, USA
Abstract
Multi-degree-of-freedom ship motion and ship stability in random seas are of major interest for the development of new
advanced intact stability criteria. The purpose of this research is to improve the safety of new ship designs, but the results are
relevant also for other engineering systems involving multiple scales. We focus on roll-pitch and roll-heave motion in random
seas. The random wave excitation is modeled by a non-white stationary process. This process is derived from a spectral
description of the random seaway using traveling effective wave.
© 2012The
© 2013 TheAuthors.
[Link]
Published
by by Elsevier
Elsevier [Link].
OpenSelection and/or
access under peer-review
CC BY-NC-ND under responsibility of Karlsruhe Institute of
license.
Technology (KIT),
Selection and/or peer Institute of the
review under Engineering
responsibility ofMechanics
Karlsruhe Institute of Technology (KIT) Institute of the Engineering Mechanics.
Keywords: Fluid-structure-interaction; random seas; stochastic averaging
1. Introduction
It is of major interest for the design process of ships to improve the calculation procedures for fluid-ship
interaction in realistic sea states. Modeling of sea states by a random fluid field defined by measured spectral
properties is more realistic then using deterministic fluid fields modeled by Airy or Stokes waves. However,
numerical simulations are time consuming, if a random fluid field model is used, since simulations have to be
repeated many times. In order to make the analysis of ship dynamics in random seas tractable in the early design
process, a reduction of the multi-degree-of freedom state space is needed. This is possible by means of the
stochastic averaging Method. Up to now, asymptotic methods were used for determining the averaged equations
of roll motion subjected to real noise excitation [1,2,3,4]. Exact averaging of strongly nonlinear one-degree-of-
freedom oscillator was done for the case of white noise excitation [5,6,7] and for the white noise excited Duffing
* Corresponding author. Tel.: +49-40-42878-2209; fax: +49-40-42878-2028.
E-mail address: dostal@[Link]
2210-9838 © 2013 The Authors. Published by Elsevier B.V. Open access under CC BY-NC-ND license.
Selection and/or peer review under responsibility of Karlsruhe Institute of Technology (KIT) Institute of the Engineering Mechanics.
doi:10.1016/[Link].2013.01.015
Leo Dostal et al. / Procedia IUTAM 6 (2013) 132 – 140 133
oscillator with a two well potential [8]. Here, we use stochastic averaging as presented in [9,10] and include
additional excitation due to heave or pitch motion of the ship. The purpose of this research is to improve the
safety of new ship designs, but the results are relevant also for other engineering systems involving multiple
scales.
We focus on coupled roll-pitch and roll-heave motion in random seas. The random wave excitation is modeled
by a non-white stationary process. In Section 2 this process is derived from a spectral description of the random
seaway using a traveling effective wave. Section 3 is devoted to roll-pitch and roll-heave equations of motion
which are transformed later into a dynamical system with multiple time scales. The state space dimension of this
system is reduced by means of stochastic averaging in Section 4, followed by possible application of the reduced
equations in Section 5. Before we conclude our results, we calculate probabilistic measures for a real RoRo ship
design.
2. Random Seas Waves
A well-known model for an irregular long crested wave surface is the superposition of infinitely many
harmonic waves with wave numbers k ( ) and frequencies , which correspond to a one-sided sea state
spectral density S ( ) . Such an irregular wave surface is too complex for further analysis, because there are
infinitely many possibilities for the wave pressure field acting on the ship hull. Since high frequent pressure
variations will have only small effect on the total fluid forces acting on the ship hull, we consider only an
averaged incident wave with the same length as the ship length consisting of two harmonic components.
Therefore, we approximate the irregular long crested wave surface by the following effective wave
2 2 2
Zeff ( x, t ) s (t )sin x c (t )cos x (t )cos x (t ) . (1)
L L L
The Gaussian random processes s (t ) and c (t ) are determined by minimizing the error between the irregular
wave surface and the effective wave. Then spectral densities of the above processes are
S s ( ) 2 f s (k ( ))2 S ( ), (2)
S c ( ) 2 fc (k ( ))2 S ( ), (3)
where the transfer functions f s and f c are given by
L L
2 sin k( ) Lk ( )sin k( )
2 2
f s (k ( )) 2
, f c (k ( )) 2
. (4)
2 L 2 L
k( ) k( )
2 2
3. Model of Roll-pitch and Roll-heave coupling
Modeling of coupled 6-degrees of freedom ship motion can be done by many ways, depending on the
necessary level of accuracy. Since our results have to be useable in the early design stage of a ship, we will
include only leading nonlinearities and mode coupling. It follows from linear ship motion analysis, that for
134 Leo Dostal et al. / Procedia IUTAM 6 (2013) 132 – 140
symmetric floating bodies with respect to the x z plane, the surge, heave, and pitch motions are decoupled
from the sway, roll, and yaw motions. On the other hand, we include the coupling resulting from the influence of
ship orientation in waves on the righting lever GZ . This leads to multiplicative coupling in the righting lever
curve approximation. From an order of magnitude analysis of heave-pitch-roll motions, it is common to neglect
the influence of roll on heave and pitch. Then, the heave and pitch motion statistics can be computed beforehand
and the roll motion is then obtained by including these forcing terms in the one-degree of freedom equation of
motion for roll. We use a linear strip method to obtain the response amplitude operator (RAO) for heave and pitch
motion. The spectral densities S z z and S for the heave and pitch motion with corresponding response
amplitude operators RAOz and RAO are given by
2 2
S z z
( ) RAOz S ( ), S ( ) RAO S ( ). (5)
Figure 1. Definition of relevant ship motions
3.1. Roll-pitch and Roll-heave coupling
Heave and pitch motions change the position and orientation of the ship hull in incident waves. Therefore, these
motions will have an influence on the ships restoring capability. The forces and moments on the hull are
determined by integrating the fluid pressure over the wetted surface of the ship hull. This results in a moment
M G with respect to the center of gravity G of the ship, which is determined by M G GZ g . Here, g is
acceleration due to gravity and is the displacement of the ship. For each ship, the righting lever has to be
computed numerically, taking the specific geometry of the ship hull into account. These numerical data are then
fitted by analytical functions. For this, we consider the following approximation GZ app for the righting lever
curve of coupled roll-pitch or roll-heave, where we denote the heave process by 1 : z and the pitch process
by 2 : . Then
3
GZapp ( , i , , ) q1 q2 q3 cos( ) qi i , i 1,2, (6)
where all coefficients are positive. The coefficients q1 , q2 , and q3 are obtained by least squares fitting of the
righting lever curve data GZ , cf. [11]. We have included a cubic nonlinearity in to account for the softening
spring characteristic of roll restoring force. Moreover, the excitation process : c
cos( ) is given by
Leo Dostal et al. / Procedia IUTAM 6 (2013) 132 – 140 135
the amplitude process c defined by the spectral density from equation (3). With the above approximation of
righting lever GZ , the roll dynamics of a ship in head or following long crested waves can be represented by
3
( I xx Axx ) b1 b3 gGZ app ( , , i ) 0, i 1, 2. (7)
Here, I xx is the roll moment of inertia, Axx is the hydrodynamic added mass evaluated at the natural roll
eigenfrequency, b1 , and b3 are linear and cubic damping coefficients.
3.2. Derivation of Multiple Scales Model
Using standard rescalings, equation (7) can be written as the following system of first order differential equations
x y,
2
(8)
y x 3 x3 (2 1 y 2 3 y3 ) ( 4 x k i x i ), i 1, 2.
For the heave-roll and pitch- (8) is much
smaller than one, since hydrodynamic damping and excitation due to sea are small compared to the restoring
forces. Then equations (8) form a weakly perturbed Hamiltonian system, with Hamiltonian
2
y2 x2 3x4
H ( x, y ) . (9)
2 2 4
From the Hamiltonian we obtain
1
Q( x, H ) : y 2 2H 2 2
x 3 x4 . (10)
2
It is clear, that the energy of system (8) changes due to the perturbations and damping. Therefore, the total
derivative of the Hamiltonian (9) does not vanish and we have
dH
(2 1Q( x, H ) 2 3Q( x, H ) 2 ) ( 4 x k i x i ) Q( x, H ). (11)
dt
With this we state a system of differential equations for the energy H and the variable x
x Q( x, H ),
(12)
H (2 1Q( x, H ) 2 3Q( x, H ) 2 ) ( 4x k i x i ) Q( x, H ).
The state variable x and H in equation (12) exhibit different time scales, since the energy H changes slowly
with time compared to the change of the variable x . For fixed H a closed form solution of x can be obtained in
terms of Jacobian elliptic functions by
x(t ) b sn(qt , k ), (13)
136 Leo Dostal et al. / Procedia IUTAM 6 (2013) 132 – 140
where
2 4
4 3H
b , (14)
3
and
4H 3 b
a , q a , k .
b2 3 2 a
With the equality x Q( x, H ) we obtain from equation (13)
Q( x, H ) bq cn(qt , k ) dn qt , k . (15)
For a fixed energy H in the unperturbed system with 0 , the period of one oscillation of the fast variable x
is thus
b
dx
T (H ) 2dx . (16)
b Q ( x, H )
4. Stochastic Averaging
A rigorous proof of a limit theorem concerning stochastic differential equations with multiple time scales was
obtained in 1966 by Khashminskii [12]. This theorem was later extended by Papanicolao and Kohler [13]. We
use the limit theorem from [13] to average the change of energy H in system (12) during one oscillation period
T of the fast variable x . For this purpose, we define the averaging operator
1 T
M f (t ) f (t )dt. (17)
T 0
Application of the limit theorem from [13] yields the following Itô stochastic differential equation
dH m( H )dt ( H )dW , (18)
which determines the energy variable H of system (12). The equations for drift m H and diffusion
( H ) become
Leo Dostal et al. / Procedia IUTAM 6 (2013) 132 – 140 137
m( H ) M Q( x(t ), H )( 2 1 2 3Q( x(t ), H ))
0
2 cnt dnt
4 k ib (R ( ) R ( )) M snt snt d
i i
cnt dnt
0 (19)
2 2 cnt dnt
4 b R ( )M snt snt d
cnt dnt
0
2 2 cnt dnt
k bi R i i ( )M snt snt d ,
cnt dnt
2 2 4
(H ) 4b q2 R ( )M snt snt cnt dnt cnt dnt d
k 2i b 4 q 2 R i i ( )M snt snt cnt dnt cnt dnt d (20)
4 k ib4 q 2 (R i
( ) R i
( ))M snt snt cnt dnt cnt dnt d .
Here R and R i i are the autocorrelation functions of the processes i and c
. The cross correlation
is denoted as R i
. Equations (19) and (20) contain the Jacobian elliptic functions
snt : sn(qt , k ), cnt : cn(qt , k ), dnt : dn(qt , k ),
and Jacobian elliptic functions with time shift
snt : sn(q(t ), k ), cnt : cn(qt , k ), dnt dn(q(t ), k ).
5. Stationary Density and Mean Exit Time
With the results from Section 4 for drift and diffusion of the averaged process for the energy H , we can
calculate further relevant measures like probability density functions (pdf) or mean times until a specific energy
level is reached. For this, we introduce the speed and scale measures, which transform the diffusion process (18)
on its natural scale, where the drift is identically zero. This task is achieved via the transformation S( x) given by
x y m(r )
S( x) s( x)dx, s( y ) exp 2 2
dr , (21)
(r )
where s( y) is called the scale density. Then, the speed density is defined by
1
( y) 2
. (22)
( y ) s( y)
The stationary pdf of energy H can be stated in terms of
138 Leo Dostal et al. / Procedia IUTAM 6 (2013) 132 – 140
pst ( H ) ( H ) c1S( H ) c2 . (23)
The coefficients c1 and c2 are determined by boundary and normality conditions. If we assume a reflecting
4
boundary at H / (4 3 ) (i.e. no capsizing), which is approximately the case up to moderate noise
intensities, then c1 0 . In this case a stationary solution exists and is given by
c2 H m( x)
pst ( H ) 2
exp 2 2
dx . (24)
(H ) ( x)
The mean first passage time Te for a process on x [ xe , xc ] with entrance boundary xe and exit boundary
xc can be calculated by the following formula (cf. [14])
xc xc
Te ( x0 ) 2 s( y)dy ( z )dz, (25)
x0 z
where x0 [ xe , xc ] is the starting point.
2
Pitch RAO [deg/m]
1
Heave RAO [-]
1.5
1
0.5
0.5
0 0
0.2 0.4 0.6 0.8 0.2 0.4 0.6 0.8
Wave frequency [rad/s] Wave frequency [rad/s]
Figure 2. Heave RAO, U=15 kn Figure 3. Pitch RAO, U=15 kn
Leo Dostal et al. / Procedia IUTAM 6 (2013) 132 – 140 139
6. Results for a RoRo ship in random seas
We now consider the case of a RoRo ship travelling in long crested following waves. The data of this ship can
be obtained from [10,11,15]. At ship speed U 15 kn in a sea state S ( ) defined by a Piersson-Moskowitz
spectrum with modal frequency m 0.64 rad / s and significant wave height H s 14 m , the RoRo ship is
in the region of 2:1 parametric resonance, which leads to large amplitude roll motion. The heave and pitch
2 and 3. For the case of roll-pitch coupling we calculate the drift
m( H ) and diffusion ( H ) , which are respectively given in equation (19) and (20), by numerical integration.
The numerical data are interpolated by cubic splines and shown in Figures 4 and 5.
100 1000
(H)
m(H)
-100 500
2
-200
-300 0
0 5 10 15 0 5 10 15
H H
Figure 4. Drift m(H) for roll-pitch motion Figure 5. Diffusion 2
( H ) for roll-pitch motion
We further can calculate an approximate probability density by means of equation (24). After the
transformation from roll energy H to maximal roll angle per roll period b , which is given in equation (14), we
get the corresponding pdf as shown in Figure 6. Further calculations of mean first passage times by means of
formula (25) are shown in Figure 7, where the scaled time for reaching the critical energy level for capsizing is
computed, starting at the initial roll angle 0 . Note, that the scaled time t t is plotted, where we have used
0.1 .
15 150
pst(b)
Stationary density
Mean exit time
10 100
5 50
0 0
0 10 20 30 40
0 0.5 1
b initial angle 0
[deg]
Figure 6. Pdf of maximal angle b Figure 7. Mean time for capsizing
for roll-pitch motion for roll-pitch motion
140 Leo Dostal et al. / Procedia IUTAM 6 (2013) 132 – 140
7. Summary and Conclusions
In this paper we have obtained analytical results for the behavior of the roll-pitch and roll-heave motion of
ships in random seas. It was shown, that the developed theory is applicable for the analysis of large real ships. An
approximation for the probability density in a finite time interval was calculated for a RoRo ship, assuming
reflecting boundary conditions for the corresponding Fokker-Planck equation. The approximate probability
density can be used in the design process of a ship to optimize its stability. Because the roll dynamics are
modeled by a softening spring type Duffing oscillator with additional cubic damping, which is excited by non-
white stochastic processes, the presented results are also applicable to various engineering problems containing
multiple scales.
Acknowledgements
The authors are indebted to the DFG (Deutsche Forschungsgemeinschaft/German Research Foundation)
for funding the project under contract Kr 752/31-1.
References
[1] Moshchuk NK, Ibrahim RA, Khasminskii RZ, Chow PL. Asymptotic expansion of ship capsizing in random sea waves-I. First-order
approximation. Int. J. Non-Linear Mech. 1995;30:727 40.
[2] Moshchuk NK, Ibrahim RA, Khasminskii RZ, Chow PL. Asymptotic expansion of ship capsizing in random sea waves-II. Second-order
approximation. Int. J. Non-Linear Mech.1995;30:741 57.
[3] Roberts JB. A stochastic theory for nonlinear ship rolling in irregular seas. J Ship Res 1982;26:229 45.
[4] Roberts JB, Vasta M. Markov modelling and stochastic identification for nonlinear ship rolling in random waves. Phil Trans R Soc Lond
A 2000;358:1917 41.
[5] Roberts JB. The energy envelope of a randomly excited non-linear oscillator. J of Sound and Vibration 1978;60:177 85.
[6] Sri Namachchivaya N. Co-dimension two bifurcation in the presence of noise. J appl Mech (ASME) 1991;58:259 65.
[7] Proc of the IUTAM symposium on
random vibrations and reliability (ed. K. Hennig). Frankfurt/oder (GDR): Berlin: Akademie-Verlag
[8] Sri Namachchivaya N, Sowers RB, Vedula L. Non-standard reduction of noisy Duffing-Van der Pol equation. Dynamical Systems
2001;16(3):223 45.
[9] Dostal L, Kreuzer E. Sri Namachchivaya N. Ship capsizing analysis in random seas using non-standard stochastic averaging. In: Proc of
7th European Nonlinear Dynamics Conference, Rome, Italy (2011)
[10] Dostal L, Kreuzer E, Sri Namachchivaya N. Non-standard Stochastic Averaging of Large Amplitude Ship Rolling in Random Seas.
Proc. R. Soc. A. 2012;468(2148):4146 4173.
[11] Dostal L, Kreuzer E. Probabilistic approach to large amplitude ship rolling in random seas. Proceedings of the Institution of Mechanical
Engineers, Part C: Journal of Mechanical Engineering Science 2011;225:2464-76
[12] Khasminskii RZ. A limit theorem for the solution of differential equations with random right-hand sides. Theory Probb Appl 1966;11:
390-405
[13] Papanicolaou GC, Kohler W. Asymptotic analysis of deterministic and stochastic equations with rapidly varying components. Commun
Math Phys 1975;45:217 32.
[14] Sri Namachchivaya N. Instability theorem based on the nature of the boundary behaviour for one dimensional diffusion. Solid Mech
Arch 1989;14(3-4):131 41.
[15] Kreuzer E, Sichermann W. Prediction of extreme ship motions in irregular waves. 5th EUROMECH Nonlinear Dynamics Conference.
Eindhoven (2005)