Tank Sloshing

J. Fluid Mech. (2000), vol. 407, pp. 201–234.
Printed in the United Kingdom 201

c 2000 Cambridge University Press
Multidimensional modal analysis of nonlinear

sloshing in a rectangular tank
with finite water depth
By O D D M. F A L T I N S E N1 , O L A V F. R O G N E B A K K E1 ,
I V A N A. L U K O V S K Y2 A N D A L E X A N D E R N. T I M O K H A2
1
Department of Marine Hydrodynamics, Faculty of Marine Technology, NTNU, Trondheim,
N-7491, Norway
2
Institute of Mathematics, National Academy of Sciences of Ukraine, Tereschenkivska, 3 str.,
Kiev, 252601, Ukraine
(Received 29 January 1999 and in revised form 18 October 1999)
The discrete infinite-dimensional modal system describing nonlinear sloshing of an

incompressible fluid with irrotational flow partially occupying a tank performing an
arbitrary three-dimensional motion is derived in general form. The tank has vertical
walls near the free surface and overturning waves are excluded. The derivation is based
on the Bateman–Luke variational principle. The free surface motion and velocity
potential are expanded in generalized Fourier series. The derived infinite-dimensional
modal system couples generalized time-dependent coordinates of free surface elevation
and the velocity potential. The procedure is not restricted by any order of smallness.
The general multidimensional structure of the equations is approximated to analyse
sloshing in a rectangular tank with finite water depth. The amplitude–frequency
response is consistent with the fifth-order steady-state solutions by Waterhouse (1994).
The theory is validated by new experimental results. It is shown that transients and
associated nonlinear beating are important. An initial variation of excitation periods
is more important than initial conditions. The theory is invalid when either the water
depth is small or water impacts heavily on the tank ceiling. Alternative expressions for
hydrodynamic loads are presented. The procedure facilitates simulations of a coupled
vehicle–fluid system.
1. Introduction
A main objective is to describe violent fluid motions (sloshing) in a partly filled
tank forced to oscillate in a frequency domain close to its lowest natural frequency.
The ratio between maximum free surface amplitude and characteristic tank motion
amplitude is then high and significant nonlinearities occur. This has practical interest
for sloshing in ship tanks. By considering sea states that a ship has to operate in,
it is realistic that wave-induced ship motions can cause resonant fluid oscillations.
This can lead to large local structural loads in the tank and has an important
effect on the global ship motions. It is desirable to develop numerical methods
that accurately describe the fluid loading and coupling between ship motions and
sloshing. A necessary requirement is that long time simulations can be performed
and the proper statistical distribution of response variables obtained for various sea
states.
202 O. M. Faltinsen, O. F. Rognebakke, I. A. Lukovsky and A. N. Timokha
Several studies on different numerical approaches to sloshing have been reported
by Su Tsung-Chow (1992), Buechmann (1996), Tanizawa (1996), Chen et al. (1997),
Pawell (1997) and the Loads Committee of the 13th ISSC (Moan & Berge 1997). A
general drawback is the limited ability to perform long time simulations, especially
for coupled ‘liquid–structure’ interactions. It may also be difficult to find water impact
loads and local structural response. One reason is that water impact studies would
often require a very fine discretization in time and space. Hydroelasticity may also
have to be considered. We have instead focused on developing a semi-analytical
method based on modal modelling. The present method assumes a smooth tank.
This implies that potential theory can be used. The method also requires vertical
tank walls near the mean free surface in its equilibrium position. Overturning waves
cannot be described. It will be shown that a high degree of analysis can be performed.
The consequences are both a time efficient and robust method. Water impact is not
studied in detail, but the method can be combined with a local slamming analysis
(see Faltinsen & Rognebakke 1999) and applied to coupled ‘fluid–tank’ simulations.
An example is given illustrating the damping effect of forceful water impact on fluid
motion.
Modal modelling of nonlinear sloshing implies that the equation of the free surface
Σ(t): z = f(x, y, t) is expanded in generalized Fourier series by a set of natural modes.
The free surface elevation and the unknown velocity potential ϕ are expressed as
X 
z = f(x, y, t) = βi (t)(surfacemode)i (x, y), 
X (1.1)
ϕ(x, y, z, t) = Ri (t)(domainmode)i (x, y, z). 
The (x, y, z) coordinate system is fixed relative to the tank; x, y are coordinates in the
plane of the unperturbed water surface and t is the time variable. Generally speaking,
the surface and volume modes are arbitrary known functions. However, they are
typically chosen by the relation
(surfacemode)i (x, y) = (domainmode)i (x, y, z)|Σ0 , (1.2)
where Σ0 is the unperturbed free surface. Since f(x, y, t) is single-valued, (1.1) does not
describe overturning waves. Moreover, (surfacemode)i (x, y) must have a non-varying
domain of definition. This means the tank must have vertical walls near the free
surface in its equilibrium position.
The generalized coordinates βi and Ri are found by a coupled system of nonlinear
ordinary differential equations (modal system). The derivation of the modal system
from the original free boundary problem was first proposed by Narimanov (1957)
based on a perturbation technique. It has been further developed by Dodge, Kana &
Abramson (1965), Narimanov, Dokuchaev & Lukovsky (1977) and Lukovsky (1990).
These and other authors used a perturbation technique combined with variational
(Hamilton–Ostrogradsky) projective method and derived small-dimensional models
(1–3 degrees of freedom in a vertical circular cylinder) in the generalized coordinates
βi (t) or their averaged values (for resonantly excited waves). (See for instance Lukovsky
1976; Miles (1976, 1984a, b).) Using an averaging technique means that βi is written as
X
∞
βi = (hβi i1j (τ) sin (jσt) + hβi i2j (τ) cos (jσt)),
j=0
where σ is the excitation frequency and τ is slowly varying relative to t. The averaged
equations of a hβi ij (τ) have the form of a Duffing equation for a rectangular tank
Multidimensional modal analysis of nonlinear sloshing 203
(see Shemer 1990 and Tsai, Yue & Yip 1990) or a system of four first-order ordinary
differential equations for a vertical circular cylindrical tank (see Miles (1984a, b)).
Funakoshi & Inoue (1991) used Miles’ model in their detailed simulations. The
averaging technique and small-dimensional modal modelling complement each
another in the analysis of the steady-state free surface response due to periodic tank
excitations. But these methods are questionable in modelling coupled fluid–structure
interaction with complicated non-periodic tank motions when transient effects matter.
These complex motions are simulated in engineering applications either numerically
or by phenomenological (usually pendulum) models (see Chapter 5 of Narimanov et
al. 1977 or Pilipchuk & Ibrahim 1997). An alternative is to use Narimanov’s original
technique with the modal representation in the form (1.1) and more general asymp-
totic assumptions of βi and Rn in order to reach reasonable dimensions of the modal
systems. The successful use of this approach is reported by Limarchenko & Yasinsky
(1997) and Lukovsky & Timokha (1995) for simplified models of spacecraft. A similar
method was used by Ikeda & Nakagawa (1997) for analysis of damping of vessel
motions due to sloshing. This suggests that multidimensional modal analysis can
simulate complicated nonlinear wave phenomena coupled with structural vibrations.
The general form of a discrete infinite-dimensional modal system is derived in the
first part of this paper by the Bateman–Luke (pressure-integral Lagrangian) varia-
tional principle. This idea was proposed independently by Miles (1976) and Lukovsky
(1976). They studied forced small-amplitude translatory motions of a vertical circular
cylindrical tank. The surfacemodes and domainmodes were obtained by linear theory
and related by (1.2). Our derivation of a discrete infinite-dimensional modal system
is not restricted to a particular type of body motion. The surface and domain modes
are not associated with natural modes and no asymptotic assumptions are introduced
in the first stage of the derivation. The infinite-dimensional modal system can be
reduced to a finite-dimensional form by assuming small-amplitude forced oscillations
and associate order of magnitudes of the different modes. This is done in the second
part of the paper to analyse nonlinear sloshing in a two-dimensional rectangular
smooth tank with finite water depth. Both forced translatory and rotational body
motions are considered. The lowest natural mode is assumed to dominate and the
three lowest modes interact nonlinearly with each other. Several modes having higher
order are considered by linear theory. The asymptotic theory constructed is a special
multidimensional analogue of the model by Ikeda & Nakagawa (1997) and the direct
generalization of the third-order hydrodynamic theory by Faltinsen (1974).
Experiments on nonlinear sloshing caused by primary mode resonant excitation
have been conducted. The asymptotic modal theory constructed explains the basic
observed phenomena including modulated (‘beating’) waves with a high accuracy
of amplitude and ‘beating period’ characteristics. The beating is a consequence of
transients that do not die out on a very long time scale. The reason is the very
small damping of the fluid motion inside a smooth tank with no internal structures
obstructing the flow and no heavy water impact on the tank ceiling. A consequence is
that steady-state response of the fluid motion can have a limited capability to describe
sloshing quantitatively. However the steady-state response is valuable to understand
important features of the flow like stability and how the response is influenced by
water depth, excitation frequency and amplitude. Since it represents a special case
of our theory, steady-state solutions are used in the verification process. Examples
of steady-state amplitude–frequency response for surge- and pitch-excited nonlinear
waves are presented. The results are consistent with the third- and fifth-order steady-
state solutions by Faltinsen (1974) and Waterhouse (1994) respectively.
The use of a discrete modal system allows us to calculate various kinematic and
dynamic characteristics occurring due to interaction between the fluid and the tank.
We present examples of hydrodynamic force and moment on the tank. The structure
of the equations describing the fluid motion as a function of the rigid body motions
makes it possible to set up an equation system for the coupled tank and fluid motion.
An example could be analysis of a ship tank due to wave-induced ship motions.
Since the wave conditions that cause violent sloshing may not be extreme, we can use
linear time domain theory to describe external hydrodynamic loads acting on the ship.
By setting up the equations of motions for the global rigid ship motions together
with the equations describing sloshing, complex fluid–structure interaction can be
analysed. But the theory does not describe the effect of impact on the tank ceiling.
This can easily occur in practical applications and is an area of further research.
The asymptotic theory is not applicable to shallow water. This is due to secondary
parametric resonance and means that the primary mode is not dominating. The ratio
between water depth and breadth of the tank is 0.173 in the example where the finite
water depth theory does not work. This is not really shallow water in a hydrodynamic
context (see the nonlinear theory by Verhagen & Wijngaarden 1965). What we need
is a theory that can combine the present finite water depth theory with a nonlinear
shallow water theory.
2. Free boundary value problem

A mobile rigid tank partly filled by an inviscid incompressible fluid is considered.
The flow is irrotational. The fluid volume bounded by the free surface Σ(t) and
the wetted tank surface S(t) is denoted Q(t). Let O0 x0 y 0 z 0 be an absolute coordinate
system and Oxyz be a moving coordinate system fixed with respect to the rigid tank.
The origin of Oxyz is in the unperturbed free surface and moves with the velocity
v 0 relative to O0 x0 y 0 z 0 . The tank has an angular velocity ω relative to O0 x0 y 0 z 0 . The
gravity field has the potential
U(x, y, z, t) = −g · r 0 , r 0 = r 00 + r, (2.1)
where r 0 is the radius-vector of a point of the body–fluid system with respect to O0 ,
r 00 is the radius-vector of the point O with respect to O0 , r is the radius-vector with
respect to O and g is the gravity acceleration vector.
Since the flow is irrotational, the fluid velocity can be represented as v a = ∇Φ,
where v a is the fluid velocity vector at the point (x, y, z) in the moving coordinate
system and Φ(x, y, z, t) is the velocity potential. The velocity potential and the free
surface Σ(t) can be found from the following nonlinear free boundary problem:

∂Φ 
∆Φ = 0 in Q(t), = v 0 · ν + ω · [r × ν] on S(t), 

∂ν 



∂Φ ξt
= v 0 · ν + ω · [r × ν] + on Σ(t), (2.2)
∂ν |∇ξ| 

Z 
∂Φ 1 

+ 2 (∇Φ)2 − ∇Φ · (v 0 + ω × r) + U = 0 on Σ(t), dQ = const. 

∂t Q(t)
Here ν is the outer normal to the boundary of Q(t) and ξ(x, y, z, t) = 0 is the equation
of the free surface Σ(t). The last integral condition in (2.2) implies fluid volume
conservation and is also the well-known solvability condition for the Neumann
boundary value problem.
The free boundary problem (2.2) must be completed by initial or periodicity
conditions to get a unique solution. The first type of condition introduces the initial
position of the free surface Σ(t0 ) and the initial distribution of normal derivatives of
Φ, i.e.
∂Φ
ξ(t0 , x, y, z) = ξ0 (x, y, z), = φ(x, y, z). (2.3)
∂ν Σ(t0 )
Here ξ0 (x, y, z) and φ(x, y, z) are given functions. If the flow starts from rest with
sufficiently small tank oscillations, linear theory can be used to formulate the initial
conditions. One way of doing this is in terms of impulse conservation. This means
Φ = 0 on Σ0 and zero free surface elevation for t = t0 . (2.4)
The last free surface boundary condition (dynamic boundary condition) of (2.2) is
obtained by using Lagrange–Cauchy integral for the pressure in the moving coordinate
system. It states that the pressure on the free surface is equal to a constant p0 . The
hydrodynamic pressure p in Q(t) can be obtained by
∂Φ 1 p − p0
+ 2 (∇Φ)2 − ∇Φ · (v 0 + ω × r) + U + =0 in Q(t). (2.5)
∂t ρ
Here ∂Φ/∂t is calculated in the moving coordinate system, i.e. for a point rigidly
connected with the system Oxyz.
There is a set of mechanical characteristics (expressed by integrals of Φ and its
derivatives), which describes the interaction between the vessel and fluid. They are:
(a) the radius-vectors of the mass centre with respect to the points O0 and O (r 01C
and r 1C ) r 01C = r 00 + r 1C , where
Z Z
ρ U dQ = −ρ g · r 0 dQ = −m1 g · r 01C ;
Q(t) Q(t)
(b) the resulting hydrodynamic forces F (t), and moments N (t) on the tank
Z Z
F (t) = (p − p0 )ν dS, N (t) = r × ((p − p0 )ν) dS. (2.6)
S(t) S(t)
3. Derivation of the general modal system by the variational method

Let us consider the boundary value problem (2.2). The unknowns are Φ =
Φ(x, y, z, t) and ξ(x, y, z, t). We will use a Bateman–Luke variational principle and
introduce the pressure in the Lagrangian of the Hamilton principle. The idea of the
pressure integral as the Lagrangian in hydrodynamic problems was first proposed by
Hargneaves (1908). The canonical formulation of this principle is given by Bateman
(1944) and Luke (1967) (for gravity surface waves in infinite basins). We use the
formulation given by Lukovsky (1990).
Pressure-integral Lagrangian variational principle. The boundary value
problem given by (2.2) can be described by examining the necessary conditions for
the extrema of the functional
Z t2
W = L dt, (3.1)
t1
where the Lagrangian L is the pressure integral
Z Z
∂Φ 1 2
L= (p − p0 ) dQ = −ρ + 2 (∇Φ) − ∇Φ · (v 0 + ω × r) + U dQ; (3.2)
Q(t) Q(t) ∂t
and the test functions satisfy
δΦ(x, y, z, t1 ) = 0, δΦ(x, y, z, t2 ) = 0; δξ(x, y, z, t1 ) = 0, δξ(x, y, z, t2 ) = 0. (3.3)
We consider a domain Q having vertical walls in a neighbourhood of the free
surface in the equilibrium position. The normal velocity componentq on the free surface
z = f(x, y, t) is given in the body-fixed system by −ξt /|∇ξ| = ft / 1 + fx2 + fy2 . The
velocity potential is expressed as
Φ(x, y, z, t) = v 0 · r + ω · Ω + ϕ. (3.4)
The vector-function Ω(x, y, z) = (Ω1 , Ω2 , Ω3 ) (Stokes–Zhukovsky potentials) is the
solution of the following Neumann boundary value problem:

∆Ω = 0 in Q(t), 



∂Ω1 ∂Ω2 

= yν3 − zν2 , = zν1 − xν3 ,
∂ν S(t)+Σ(t) ∂ν S(t)+Σ(t) (3.5)



∂Ω3 

= xν2 − yν1 , 


∂ν S(t)+Σ(t)
where ν1 , ν2 , ν3 are the projections of the outer normal ν onto the Oxyz-axes. The
function ϕ is a solution of the Neumann boundary value problem
∂ϕ ∂ϕ ft
∆ϕ = 0 in Q(t), = 0, =q .
∂ν S(t) ∂ν Σ(t) 1 + f2 + f2 x y
The Neumann boundary value problems for Ω and ϕ have unique solutions since
Z Z Z
∂Ωi ∂ϕ ft
dS = 0, dS = p dS = 0
S(t)+Σ(t) ∂ν Σ(t) ∂ν Σ(t) 1 + (∇f)2
are always fulfilled (see Lukovsky & Timokha 1995). These solutions depend para-
metrically on time. By using (3.4) and the boundary value problems for Ω and ϕ it
follows that Φ satisfies the Laplace equation and the Neumann boundary conditions
of (2.2). The dynamic condition (pressure balance) on Σ(t) gives the final equation
connecting f, Ω and ϕ.
Let the function f(x, y, t) be expressed as
X
∞
f(x, y, t) = βi (t)fi (x, y), (3.6)
i=1
where fi (x, y) is a complete (to within a constant)

R orthogonal system of functions
satisfying the condition of volume conservation Σ0 fi (x, y) dx dy = 0. Further,
X
∞
ϕ(x, y, z, t) = Rn (t) ϕn (x, y, z), (3.7)
n=1
where the complete system of functions ϕn (x, y, z) satisfies the Laplace equation in
the whole tank domain Q and zero Neumann boundary condition on the wetted
body surface. Normally, only the wetted body surface below the mean free surface is
considered. Since the system {ϕn (x, y, z)} is complete on any single-connected surface
in the tank domain, it is also a complete system on Σ0 . The Stokes–Zhukovsky
potentials Ωi are assumed to be known functions of βi . Hence, we must only find the
unknown functions βi (t) and Rn (t).
Such a family of harmonic functions ϕn (x, y, z) can be chosen as a set of solutions
of the following boundary spectral problems with spectral parameter λn :
Z
∂ϕn ∂ϕn
∆ϕn = 0 in Q0 , = 0 on S, = λn ϕn on Σ0 , ϕn dS = 0. (3.8)
∂ν ∂ν Σ0
This is the same as the linear eigenvalue problem for sloshing. The solutions can be
found analytically only for a limited class of tank shapes. Examples are a vertical
circular cylinder or a rectangular three-dimensional tank. However, a numerical
method can be used to find ϕn for a general tank shape. This was demonstrated by
Solaas & Faltinsen (1997), where Moiseev’s theory was applied to two-dimensional
sloshing. A different approach is to use a patching procedure and consider for instance
a tank consisting of a cylindrical part near the free surface. Then the solution in the
cylindrical part can be expressed as
X
ϕn (x, y, z) = (bnk exp (−λk z) + ank exp (λk z))φk (x, y) (3.9)
k
with unknown coefficients bnk and ank . Here λk and φk are the solutions of the
following spectral problem:
Z
∂φk
∆2 φk (x, y) + λ2k φk = 0 in Σ0 , = 0 on ∂Σ0 , φk dS = 0, (3.10)
∂n Σ0
where ∂Σ0 is the intersection line between Σ0 and S. The solution in the non-
cylindrical part can be found by a numerical method. When the auxiliary problem
(3.10) is formulated in circular (ring-shaped) or rectangular cross-sections Σ0 , the
solutions φk of (3.10) are expressed by Bessel functions and/or sinusoidal functions.
Otherwise, a numerical procedure for (3.10) is required.
By substituting (3.4) into (3.2) the Lagrangian L takes the following form:
Z
∂
L = −ρ v̇ 0 · r + (ω · Ω) + 12 ∇(ω · Ω) · ∇(ω · Ω) − ω · (r × ∇(ω · Ω))
Q(t) ∂t
i
1 2
− 2 v0 − ω · (r × v 0 ) − ω · (r × ∇ϕ) + ∇(ω · Ω) · ∇ϕ dQ + Lr , (3.11)
where
Z
∂ϕ 1
Lr = −ρ + 2 (∇ϕ)2 + U dQ. (3.12)
Q(t) ∂t
The two last integrand terms in square brackets of (3.11) cancel each other from
Green’s formula, i.e.
Z Z
∂(ω · Ω)
(∇(ω · Ω) · ∇ϕ − (ω × r) · ∇ϕ) dQ = − (ω × r) · ν ϕ dS = 0.
Q(t) S(t)+Σ(t) ∂ν
We also introduce the quadratic symmetric inertia tensor J 1 with components Jij1
defined by the equality
Z
−ρ ( 12 ∇(ω · Ω) · ∇(ω · Ω) − ω · (r × ∇(ω · Ω))) dQ
Q(t)
= − 12 ω12 J11
1
− 12 ω22 J22
1
− 12 ω32 J33
1 1
− ω1 ω2 J12 1
− ω1 ω3 J13 1
− ω2 ω3 J23 .
These components Jij1 can be calculated by Green’s formula, i.e.

Z Z
1 ∂Ω1 ∂Ω1 ∂Ω1
J11 =ρ y −z dQ = ρ Ω1 dS, (3.13a)
Q(t) ∂z ∂y S(t)+Σ(t) ∂ν
Z Z
1 ∂Ω2 ∂Ω2 ∂Ω2
J22 =ρ z −x dQ = ρ Ω2 dS, (3.13b)
Q(t) ∂x ∂z S(t)+Σ(t) ∂ν
Z Z
1 ∂Ω3 ∂Ω3 ∂Ω3
J33 =ρ x −y dQ = ρ Ω3 dS, (3.13c)
Q(t) ∂y ∂x S(t)+Σ(t) ∂ν
Z Z
1 1 ∂Ω1 ∂Ω1 ∂Ω2 ∂Ω2
J12 = J21 =ρ z −x dQ = ρ y −z dQ
Q(t) ∂x ∂z Q(t) ∂z ∂y
Z Z
∂Ω2 ∂Ω1
=ρ Ω1 dS = ρ Ω2 dS, (3.13d)
S(t)+Σ(t) ∂ν S(t)+Σ(t) ∂ν
Z Z
1 1 ∂Ω1 ∂Ω1 ∂Ω3 ∂Ω3
J13 = J31 = ρ x −y dQ = ρ y −z dQ
Q(t) ∂y ∂x Q(t) ∂z ∂y
Z Z
∂Ω3 ∂Ω1
=ρ Ω1 dS = ρ Ω3 dS, (3.13e)
Z Z
1 1 ∂Ω2 ∂Ω2 ∂Ω3 ∂Ω3
J23 = J32 = ρ x −y dQ = ρ z −x dQ
Q(t) ∂y ∂x Q(t) ∂x ∂z
Z Z
∂Ω3 ∂Ω2
=ρ Ω2 dS = ρ Ω3 dS. (3.13f )
The Lagrangian L (3.11) can be rewritten as
L = − [v̇01 l1 + v̇02 l2 + v̇03 l3 + ω̇1 l1ω + ω̇2 l2ω + ω̇3 l3ω + ω1 l1ωt + ω2 l2ωt
+ω3 l3ωt − 12 (ω12 J11
1
+ ω22 J22
1
+ ω32 J33
1 1
) − ω1 ω2 J12 1
− ω1 ω3 J13
1 2 2 2
−ω2 ω3 J23 − 12 m1 (v01 + v02 + v03 ) + (ω2 v03 − ω3 v02 )l1
+ (ω3 v01 − ω1 v03 )l2 + (ω1 v02 − ω2 v01 )l3 ] + Lr , (3.14)
where
Z Z Z 
∂Ωk
m1 = ρ dQ, lkω = ρ Ωk dQ, lkωt = ρ dQ, 


Q(t) Q(t) Q(t) ∂t 
Z Z Z (3.15)


l1 = ρ x dQ, l2 = ρ y dQ, l3 = ρ z dQ. 

Q(t) Q(t) Q(t)
The vectors l = {lk }, l ω = {lkω }, l ωt = {lkωt } depend only on βi (t) and β̇i (t).
It follows from (3.7) that
Z X XX
1
Lr = −ρ Ṙn ϕn + 2 Rn Rk (∇ϕn , ∇ϕk ) + U dQ
Q(t) n=1 n k
X XX
1 0
=− An Ṙn + 2 Ank Rn Rk − g1 l1 − g2 l2 − g3 l3 − m1 g · r 0 , (3.16)
n n k
where
Z Z Z
∂ϕk
An = ρ ϕn dQ, Ank = Akn = ρ (∇ϕn , ∇ϕk ) dQ = ρ ϕn dS (3.17)
Q(t) Q(t) Σ(t)+S(t) ∂ν
are functions of βi (t).
The Lagrangian L is originally a function of two independent variables f(x, y, z, t)
and Φ(x, y, z, t). The independent variables become the time-varying functions βi (t), i >
1 and Rn (t), n > 1 after substituting (3.4), (3.6) and (3.7) into the Lagrangian. The
variations of the functional (3.1) by βi (t) and Rn (t) for given v 0 (t) and ω(t) are
Z t2 X XX X X ∂An
δW = An δ Ṙn + Ank Rk δRn + Ṙn
t1 n n k i n
∂βi
∂l1ωt ∂l2ωt ∂l3ωt XX ∂Ank
+ω1 + ω2 + ω3 + 12 Rn Rk
∂βi ∂βi ∂βi n k
∂βi
∂l1ω ∂l2ω ∂l3ω ∂l1
+ω̇1 + ω̇2 + ω̇3 + (v̇01 − g1 + ω2 v03 − ω3 v02 )
∂βi ∂βi ∂βi ∂βi
∂l2 ∂l3
+(v̇02 − g2 + ω3 v01 − ω1 v03 ) + (v̇03 − g3 + ω1 v02 − ω2 v01 )
∂βi ∂βi
1 1 1 1 1

∂J ∂J ∂J ∂J ∂J ∂J 1
− 12 ω12 11 − 12 ω22 22 − 12 ω32 33 − ω1 ω2 12 − ω1 ω3 13 − ω2 ω3 23 δβi
∂βi ∂βi ∂βi ∂βi ∂βi ∂βi

∂l1ωt ∂l2ωt ∂l3ωt
+ ω1 + ω2 + ω3 δ β̇i dt = 0. (3.18)
∂β̇i ∂β̇i ∂β̇i
The following infinite system of nonlinear differential equations (modal system) cou-
pling modal functions Rn (t) and βi (t) is obtained by integrating by parts in (3.18) and
using (3.3) for test functions:
d X
An − Rk Ank = 0, n = 1, 2, . . . , (3.19)
dt k
X ∂An X X ∂Ank ∂l1ω ∂l2ω ∂l3ω ∂l1ωt ∂l2ωt

Ṙn + 12 Rn Rk + ω̇1 + ω̇2 + ω̇3 + ω1 + ω2
n
∂βi n k
∂β i ∂βi ∂β i ∂β i ∂β i ∂βi

∂l3ωt d ∂l1ωt ∂l2ωt ∂l3ωt ∂l1
+ω3 − ω1 + ω2 + ω3 + (v̇01 − g1 + ω2 v03 − ω3 v02 )
∂βi dt ∂β̇i ∂β̇i ∂β̇i ∂βi
∂l2 ∂l3 ∂J 1
+(v̇02 − g2 + ω3 v01 − ω1 v03 ) + (v̇03 − g3 + ω1 v02 − ω2 v01 ) − 12 ω12 11
∂βi ∂βi ∂βi
1 1 1 1 1
∂J ∂J ∂J ∂J ∂J
− 12 ω22 22 − 12 ω32 33 − ω1 ω2 12 − ω1 ω3 13 − ω2 ω3 23 = 0. (3.20)
∂βi ∂βi ∂βi ∂βi ∂βi
The system of ordinary differential equations (3.19) can be considered as a linear
P
system of algebraic equations Ank (βi )Rk = (d/dt)An (βi ). By using a numerical or
asymptotic technique we can then find a solution of Rn as a function of βi . After
substituting Rn into (3.20) we get a system of second-order nonlinear differential
equations with respect to βi . The values ∂lk /∂βi are given by
Z Z Z
∂l3 2 ∂l2 ∂l1
=ρ fi dS βi = λi3 βi , =ρ yfi dS = λi2 , =ρ xfi dS = λi1 .
∂βi Σ0 ∂βi Σ0 ∂βi Σ0
(3.21)
The constructed infinite-dimensional system of equations (3.19) (3.20) is applicable
to any type of rigid body motion. It is necessary that f(x, y, t) given by (3.6) is
single-valued. This means that plunging breakers cannot be described. There are no
other restrictions on the type of surface wave that can be studied.
4. Modal system for two-dimensional fluid flows

We assume two-dimensional fluid motion in the (x, z)-plane. Then
v 0 = (v0x , 0, v0z ), ω = (0, ω(t), 0), r = (x, 0, z), Ω(x, 0, z) = (0, Ω(x, z, t), 0) (4.1)
and Ω is the solution of the following boundary value problem:

∂Ω
∆Ω = 0 in Q(t), = zν1 − xν3 . (4.2)
∂ν S(t)+Σ(t)
The velocity potential Φ(x, 0, z, t) takes the form

X
∞
Φ(x, 0, z, t) = v0x x + v0z z + ω(t) Ω(x, z, t) + Rn (t) ϕn (x, z), (4.3)
n=1
where ϕn (x, z) is a complete system of harmonic functions satisfying the zero Neumann
condition on the bottom and vertical walls and the Laplace equation in Q.
The general infinite-dimensional modal system of ordinary differential equations
(3.19), (3.20) has in two dimensions the following form:
d X
An − Rk Ank = 0, n = 1, 2, . . . , (4.4)
dt k
X X X ∂Ank
∂An 1 ∂l2ω ∂l2ωt d ∂l2ωt
Ṙn +2 Rn Rk + ω̇ +ω − ω
n
∂βi n k
∂βi ∂βi ∂βi dt ∂β̇i
∂l1 ∂l3 ∂J 1
+(v̇01 − g1 + ωv03 ) + (v̇03 − g3 − ωv01 ) − 12 ω 2 22 = 0. (4.5)
∂βi ∂βi ∂βi
5. Asymptotic modal system for a rectangular tank performing arbitrary

small-amplitude motions
We consider a mobile rectangular rigid tank filled partly by an inviscid incom-
pressible fluid. The mean water depth is h and l is the tank breadth. The flow is
irrotational and two-dimensional (see figure 1). The origin of the coordinate system
is in the mean free surface at the centreplane of the tank. The equation z = f(x, t)
determines the perturbed free surface Σ(t). The fluid domain is
Q(t) = {(x, z) : −h < z < f(x, t); −l/2 < x < l/2}. (5.1)
z
Undisturbed
water plane
z= –h x= l/2
Figure 1. Coordinate system.
Since f(x, t) is expressed by (3.6), the complete (to within a constant) orthogonal
system of functions {fi (x)} should satisfy the volume conservation condition
Z l/2
fi (x) dx = 0. (5.2)
−l/2
The modal system (4.4), (4.5) can be approximated to surface waves with one
primary dominating mode corresponding to the first natural mode. This implies that
the body motions are horizontal and/or rotational and quasi-periodic with average
frequency close to the first resonance frequency. It is also necessary that the water
depth is not shallow and the fluid does not hit the tank ceiling (see, also, physical
arguments presented in Faltinsen 1974 and the book by Mikishev 1978). The rigid
body motions are assumed small relative to the tank breadth and water depth.
The derivation of the finite-dimensional asymptotic analogue of the system (4.4)
and (4.5) requires an asymptotic relation between dominating mode amplitude and
excitation amplitude. It is assumed, as in the theory by Faltinsen (1974), that
O(β13 ) = O(H) = O(ψ0 ) = . (5.3)
Here H is translatory (surge) motion magnitude and ψ0 is angular (pitch) magnitude.
Further β2 = O(2/3 ), β3 = O(). Higher-order terms than will be neglected in the
nonlinear equations. The modes fi (x) in (3.6) as well as ϕi (x, z) in (3.7) can be chosen
as the solutions of the spectral problem

∆ϕi = 0 (−l/2 < x < l/2, −h < z < 0), 


∂ϕi
∂ϕi ∂ϕi (5.4)
= 0, = 0, = λi ϕi (z = 0).  
∂x x=−l/2,x=l/2 ∂z ∂z
z=−h
This means

πi iπ πi 

λi = tanh h , fi (x) = cos (x + l/2) , 
l l l (5.5)
cosh ((πi/l)(z + h)) 

ϕi (x, z) = fi (x) . 
cosh ((πi/l)h)
Equations (3.6) and (3.7) take the following form:
X
∞ X
∞
cosh ((iπ/l)(z + h))
f(x, t) = βi (t)fi (x), ϕ(x, z, t) = Ri (t)fi (x) . (5.6)
i=1 i=1
cosh ((iπ/l)h)
By accounting for the asymptotic relation (5.3) and keeping only terms up to in
the modal system (4.4) and (4.5) we get
d X
An − Rk Ank = 0, n = 1, 2, . . . , (5.7)
dt k
X X X ∂Ank
∂An ∂l2ω ∂l2ωt d ∂l2ωt
Ṙn + 12 Rn Rk + ω̇ +ω − ω
n
∂βi n k
∂βi ∂βi ∂βi dt ∂β̇i
+ (v̇01 − g1 )λi1 + (v̇03 − g3 )βi λi3 = 0. (5.8)
Asymptotic expansions of integrals Ai , Ank , l2 ω, l2ωt have to be used in (5.7) and (5.8).
Here Ai , Ank , l2ω , l2ωt are defined by (3.17) and (3.15) as integrals over the instantaneous
fluid volume position. The integrals are divided into integrals over the mean position
of fluid volume Q0 and over the remaining part Qδ . Qδ is described by βi . Further, the
integrand of the integrals over Qδ can be expanded in Taylor series by βi . Keeping
terms up to gives

ρl
A1 = (β1 + E1 (β1 β2 + β2 β3 ) + E0 (β1 + 2β1 β2 + β1 β3 )), 
3 2 2 


2 


ρl 2 2
A2 = (β2 + E2 (β1 + 2β1 β3 ) + 8E0 β1 β2 ), (5.9)
2 



ρl 

A3 = (β3 + 3E3 β1 β2 + 3E0 β13 ); 
2

A11 = ρl(E1 + 8E1 E0 β12 − (2E0 − E12 )β2 ), A22 = ρl(2E2 ), 




A12 = A21 = ρl((4E0 + 2E1 E2 )β1 + (−4E0 + 2E12 )β3 ),
(5.10)
A33 = ρl(3E3 ), A13 = A31 = 3lρ(2E0 + E1 E3 )(β2 + 2E4 β12 ), 




A23 = A32 = 3lρ(4E0 + 2E2 E3 )β1 ,
where
2
1 π π πi
E0 = , Ei = tanh h , i > 1. (5.11)
8 l 2l l
Further, we express Rn as
X X X
Rn = γi β̇i + γij β̇j βi + γijk β̇i βj βk + · · ·
i ij ijk
and substitute it in (5.7). Explicit values of γi , γij , γijk are found by collecting similar
terms. The result is

β̇1 E0 E0 E0 1 4E0 2 

R1 = + 2 β̇1 β2 − β̇2 β1 + −2 + β1 β̇1 , 

2E1 E1 E1 E2 E1 E1 E2 




1 4E0 β̇3 E0 E0
R2 = β̇2 − β1 β̇1 , R3 = − β̇1 β2 − β̇2 β1 (5.12)
4E2 E1 6E3 E1 E3 E2 E3 





3E2 2E0 E4 4E02 2E0 E2 β̇i 
2
+β̇1 β1 − − E4 + + , Ri = , i > 4; 

2E3 E1 E3 E1 E2 E3 E1 E3 2iEi
and

β̈1 E0 E0 E0 E0 

Ṙ1 = + 2 β̈1 β2 − β̈2 β1 + β̇1 β̇2 − 

2E1 E E1 E2 E12 E1 E2 

1 

E0 1 4E0 E 4E 

+ − + 2
β1 β̈1 + 2
0 1
−2 +
0
β̇12 β1 , 

E1 2 E1 E2 E1 E1 E2 



1 4E0 
2
Ṙ2 = β̈2 − (β1 β̈1 + β̇1 ) ,
4E2 E1 



β̇3 E0 E0 E0 E0 

Ṙ3 = − β̈1 β2 − β̈2 β1 − + β̇1 β̇2 + (β̈1 β12 + 2β̇12 β1 ) 

6E3 E1 E3 E2 E3 E1 E3 E2 E3 





3E2 2E0 E4 4E02 2E0 E2 β̈1 

× − − E4 + + , Ṙi = , i > 4. 
2E3 E1 E3 E1 E2 E3 E1 E3 2iEi
(5.13)
By calculating λij we get

Z l/2 2 
iπ l 

λi1 = ρ x cos (x + l/2) dx = ρ ((−1)i − 1), 

−l/2 l iπ
Z l/2 (5.14)
iπ ρl 

λi3 = ρ cos2 (x + l/2) dx = . 

−l/2 l 2
l2ω and l2ωt (see (3.15)) depend on Ω(x, z, t) which is the solution of the boundary
value problem (4.2). Ω(x, z, t) depends parametrically on βi (t) due to the free surface
Σ(t). Since ∂l2ω /∂βi and ∂l2ωt /∂βi are multiplied by terms of O() in (5.8), it is sufficient
to include only linear terms in βi in the integrals l2ω and l2ωt . The problem (4.2) in a
rectangular tank takes the following form:

∂Ω 
∆Ω = 0 in Q(t), = −x (z = −h), 

∂z

∂Ω l l ∂Ω 1 fx 
= z x = ,− , = −x p − zp (z = f(x, t)).  
∂x 2 2 ∂ν 1 + (fx )2 1 + (fx )2
(5.15)
The solution can be found by a Zhukovsky-type substitution with additional terms
for fluctuations of the free surface. This gives
X
∞
sinh ((π/l)i(z + h/2)) X
∞
cosh ((π/l)i(z + h))
Ω = xz − 2 ai fi + χi (t)fi . (5.16)
i=1
cosh ((π/2l)ih) i=1
cosh ((π/l)ih)
The coefficients ai are found from the condition χi (t) ≡ 0, i > 1, if and only if,
βi ≡ 0, i > 1. Substitution of (5.16) into (5.15) gives
X
N
iπ 2l 2
ai fi =x or ai = [(−1)i − 1]. (5.17)
i=1
l (iπ)3
The functions χi (t) follow from (5.15) after substitution of (5.16) and (5.17) and
performing the Taylor series technique for the free surface Σ(t) (with respect to βi ).
The linear terms of l2ω and l2ωt do not depend on χi (t). To show this we substitute
(5.16) in the corresponding integrals
X∞ Z l/2 X∞ Z l/2
iπ l iπ
l2ω = −2ρ ai tanh h βi fi2 dx + ρ χi (t) tanh h fi dx,
i=1
2l −l/2 i=1
iπ l −l/2
(5.18)
X∞ Z l/2
l iπ
l2ωt = ρ χ̇(t) tanh h fi dx. (5.19)
j=1
iπ l −l/2
It follows from the volume conservation condition (5.2) that

X∞ 3
l i iπ
l2ωt = 0, l2ω = −2ρ βi [(−1) − 1] tanh h . (5.20)
i=1
iπ 2l
The derivatives with respect to βi give

3
∂l2ωt ∂l2ω l i iπ
= 0, = −2ρ [(−1) − 1] tanh h , i > 1. (5.21)
∂βi ∂βi iπ 2l
Finally, by defining the angular position of the mobile coordinate system Oxyz with
respect to O0 x0 y 0 z 0 as ψ(t) we obtain correct to O() that
g3 = −g, g1 = gψ(t). (5.22)
The terms in (5.8) ψ̈∂l2ω /∂βi +(−g3 )βi λ3i +(−g1 )λ1i caused by forced pitch excitation
can be rewritten as
2
l i 2l iπ
−ρ [(−1) − 1] tanh h ψ̈(t) + gψ(t) + gβi . (5.23)
iπ iπ 2l
When substituting above formula in (5.8), we get the following system of ordinary
differential equations describing modal oscillations of a fluid in a rectangular tank
performing arbitrary small-magnitude motions (keeping terms up to third order in
the nonlinear equations):
(β̈1 + σ12 β1 ) + d1 (β̈1 β2 + β̇1 β̇2 ) + d2 (β̈1 β12 + β̇12 β1 ) + d3 β̈2 β1
+P1 (v̇0x − S1 ω̇ − gψ) + Q1 v̇0z β1 = 0, (5.24a)
(β̈2 + σ22 β2 ) + d4 β̈1 β1 + d5 β̇12 + Q2 v̇0z β2 = 0, (5.24b)
(β̈3 + σ32 β3 ) + d6 β̈1 β2 + d7 β̈1 β12 + d8 β̈2 β1 + d9 β̇1 β̇2 + d10 β̇12 β1
+ P3 (v̇0x − S3 ω̇ − gψ) + Q3 v̇0z β3 = 0. (5.24c)
The linear equations describing higher modes are
β̈i + σi2 βi + Pi (v̇0x − Si ω̇ − gψ) + Qi v̇0z βi = 0, i > 4. (5.25)
Here v0x and v0z are projections of the translational velocity onto axes of Oxz, ω(t) is
the value of the angular velocity of coordinate system Oxyz with respect to O0 x0 y 0 z 0 .
The coefficients introduced are calculated by formulas
8E2i−1 l
σi2 = 2giEi , P2i−1 = − 2 , P2i = 0, Qi = 2iEi ,
π (2i − 1)

2l iπ
Si = tanh h , i > 1, (5.26)
πi 2l
where σi is the natural frequency of mode i. Further,

E0 4E0 E0 
d1 = 2 + E1 , d2 = 2E0 −1 + , d3 = −2 + E1 , 

E1 E1 E2 E2 



E0 E0 E2 4E0 6E0 

d4 = −4 + 2E2 , d5 = E2 − 2 2 − , d6 = 3E3 − , 

E1 E1 E1 E1 



E0 E4 E2 E0 E3 
d7 = 9E0 − 12 − 6E3 E4 + 24 0 + 3 , (5.27)
E1 E1 E2 E1 



E0 E0 E0 E0 E3 E3 E1 

d8 = −6 + 3E3 , d9 = −6 − 6 − 6 +3 , 

E2 E1 E2 E1 E2 E2 



12E0 + 6E1 E3 72E02 E3 E1 

d10 = 18E0 − 2E4 + + 12E0 − . 

E1 E1 E2 E1 E2
The first two nonlinear equations of (5.24) couple β1 with β2 and do not depend
on β3 . The third mode component is excited by rigid body motions and the first
and the second modes of sloshing. The second mode response becomes infinite if the
excitation has frequency content at the natural frequency for the second mode; and
similarly for the third and higher modes. The first mode will be finite if it is excited
at the natural frequency of the first mode. This is caused by nonlinear effects and will
become more evident in the next section on steady-state response.
6. Steady-state sloshing in a rectangular tank with a small-amplitude

surge/pitch sinusoidal excitation
The theory of steady-state solutions of the nonlinear sloshing problem in a rect-
angular tank was created by Faltinsen (1974) based on Moiseev’s (1958) method.
The constructed asymptotic discrete theory (5.24) makes it possible to generalize the
main relations of this theory. For surge-excited steady-state waves we express v 0 as
(−Hσ sin (σt), 0, 0), set ω = ψ ≡ 0 and look for periodic solutions
βi (t + 2π/σ) = βi (t), β̇i (t + 2π/σ) = β̇i (t) (6.1)
of the discrete model (5.24).
To construct asymptotically the periodic solutions and to derive analytically the
amplitude–frequency response of nonlinear sloshing in a rectangular tank caused by
forced excitation we express the first approximation of the primary mode in the form
β1 (t) = A cos σt + o(A). (6.2)
The substitution of (6.2) into (5.24b) with periodicity condition (6.1) yields
β1 = A cos σt + o(A), β2 = A2 (l0 + h0 cos (2σt)) + o(A2 ), (6.3)
where
d4 − d5 d5 + d4 σi
l0 = , h0 = , σ̄i = , i = 1, 2. (6.4)
2σ̄22 2(σ̄22 − 4) σ
The amplitude A ∼ 1/3 of the primary mode can be found by substituting (6.3)
into the first equation of (5.24) and collecting Fourier terms of lowest order. The
equation coupling primary mode amplitude, frequency, breadth and depth will be
non-dimensionalized by dividing all length variables by l. This gives
Πh (σ̄1 , σ̄2 , Ā) = (σ̄12 − 1)Ā + m1 (σ̄2 , h̄)Ā3 − P̄1 H̄ = 0, (6.5)
m1 (σ̄2 , h̄) = d̄1 (−l̄0 (σ̄2 ) + 12 h̄0 (σ̄2 )) − 12 d̄2 − 2d̄3 h̄0 (σ̄2 ), (6.6)
where the overbar denotes non-dimensionalized value. The coefficient m1 in equation
(6.5) depends on depth–breadth ratio and frequency of excitation (σ̄i , i = 1, 2). The
last dependence has not been presented earlier for frequency–amplitude response
equations. Usually, the corresponding coefficient depends only on h/l. This means
that our asymptotic technique differs from Faltinsen–Moiseev’s procedure. In order
to compare both techniques we need to give the following remark.
Remark. For any asymptotic theory with one dominating mode the nonlinear equation
describing the dependence of amplitude–breadth ratio Ā and frequency of excitation
σ has the same general form
σ1 σ2
Πh , , Ā = 0,
σ σ
where σ1 is the natural frequency of the primary mode.
The function Π can be expanded in a Taylor series. The approach by Moiseev
(1958) and Faltinsen (1974) gives the expansion near the point (σ̄1 , σ2 /σ1 , 0) (for fixed
h̄). Our approach has no asymptotic restriction on the value of frequency σ and,
therefore, includes only power series in Ā

σ1 σ2 σ1 σ2 ∂Π σ1 σ2
Πh , , Ā = Π , ,0 + , , 0 Ā
σ σ σ σ ∂Ā σ σ

1 ∂2 Π σ1 σ2 2 ∂3 Π σ1 σ2 1 3
+ , , 0 Ā + , , 0 6
Ā + o(Ā3 ).
2 ∂Ā2 σ σ ∂Ā3 σ σ
Since the value σ2 /σ is used to calculate only m1 we in fact make a more precise
calculation of this coefficient.
Equation (6.5) gives infinite response for σ̄1 = 1 and m1 = 0. It implies that the
third-order theory is not valid if
σ2
m1 , h̄ = 0.
σ1
The root of the last equation gives h̄ = h/l = 0.3374 . . . . This is called the critical
depth and coincides with the known value (see Waterhouse 1994). The response
changes from being a ‘hard-spring’ to a ‘soft-spring’ at the critical depth. The detailed
asymptotic analysis of the response near critical depth was done by Waterhouse (1994)
by fifth-order theory based on Faltinsen–Moiseev’s technique. It was shown that the
branches in the amplitude–frequency plane coincide with a third-order theory only
for small amplitude. New turning points on the branches occur at a critical value of
the amplitude/frequency.
In our case m1 = m1 (σ2 /σ, h̄) which means that m1 is a function of σ and h̄. If a
fixed σ is close to the natural frequency σ1 , but σ 6= σ1 the equation

σ2
m1 , h̄ = 0 (6.7)
σ
gives a different value of the critical depth. This means that the critical depth is a
function of σ. If a pair (σ, h̄) satisfies (6.7), then Ā tends to infinity. This effect is
illustrated in figures 2 and 3.
Figure 2 shows the positive and negative solutions (branches P+ , P− ) of the secular
algebraic equation (6.5) for different values of the water depth h and fixed amplitude
of excitation H. The choice of H corresponds to the experimental values reported
later in the paper. Branch O is the set of solutions of (6.5) for H = 0 (no vibration of
1.0 O
(a) P+ (b)
P–
0.8 h/l = 0.173 S h/l = 0.34821
i (2, h) = 0.9 i (2, h) = 0.782
0.6
|A|
l 0.4
0.2
0
0.8 1.0 1.2 1.4 1.6 0.8 1.0 1.2 1.4 1.6
1.0
(c) (d)
0.8 h/l = 0.23121 h/l = 0.40462
i (2, h) = 0.85 i (2, h) = 0.7604
0.6
|A|
l 0.4
0.2
0
0.8 1.0 1.2 1.4 1.6 0.8 1.0 1.2 1.4 1.6
1.0
(e) (f)
0.8 h/l = 0.289 h/l =1.0
i (2, h) = 0.8115 i (2, h) = 0.7085
0.6
|A|
l 0.4
0.2
0
0.8 1.0 1.2 1.4 1.6 0.8 1.0 1.2 1.4 1.6
T1/T T1/T
Figure 2. Amplitude (A)–frequency (σ) response for nonlinear sloshing due to surge excitation
(σ/σ1 = T 1/T ). h is the mean water depth, l is the tank breadth, H is the surge amplitude. i(2, h̄)
is defined by (6.9). H/l = 0.0173.
the tank). This can be interpreted as the amplitude–frequency dependence of free non-
linear (periodic) sloshing. The branches presented differ from diagrams obtained by
Faltinsen’s theory only for large values of |A|/l and far away from the main resonance
σ̄1 = 1. The last difference is due to the change of m1 when varying σ. The results agree
with the fifth-order theory by Waterhouse (1994) for sufficiently small amplitudes.
Similar results are obtained for steady-state sloshing in a rectangular tank excited
by sinusoidal pitch motions. Let us assume the tank is pitching around the point
(0, 0, −z0 ) of the mobile coordinate system. We can correct to O() express that
ψ(t) = ψ0 cos (σt), v̇0x = z0 ψ̈(t), v̇0z = 0.
1.0 O
(a) P+ (b)
P–
0.8 S
ã0 = 0.1 rad ã0 = 0.2 rad
0.6
|A|
l 0.4
0.2
0
0.8 1.0 1.2 1.4 1.6 0.8 1.0 1.2 1.4 1.6
1.0
(c) (d)
0.8
ã0 = 0.1 rad ã0 = 0.2 rad
0.6
|A|
l 0.4
0.2
0
0.8 1.0 1.2 1.4 1.6 0.8 1.0 1.2 1.4 1.6
1.0
(e) (f)
0.8
ã0 = 0.1 rad ã0 = 0.2 rad
0.6
|A|
l 0.4
0.2
0
0.8 1.0 1.2 1.4 1.6 0.8 1.0 1.2 1.4 1.6
T1/T T1/T
Figure 3. Amplitude (A)–frequency (σ) response for nonlinear sloshing due to pitch excitation
(σ/σ1 = T 1/T ). h is the mean water depth, l is the tank breadth, ψ0 is the pitch amplitude, (0, −z0 )
is the position of pitch axis. i(2, h̄) is defined by (6.9). (a, b) z0 /l = 0, h/l = 0.2, i(2, h̄) = 0.874; (c, d)
z0 /l = 0.15, h/l = 0.35, i(2, h̄) = 0.78; (e, f) z0 /l = 0.3, h/l = 0.5, i(2, h̄) = 0.737.
The algebraic governing equation for the frequency–amplitude response takes the
following form:

2 3 z0 S1 g
(σ̄1 − 1)Ā + m1 (σ̄2 , h̄)Ā − P̄1 ψ0 − + 2 = 0. (6.8)
l l lσ
It differs from the equation of forced surge steady-state sloshing (6.5) only by the last
inhomogeneous term and agrees with the corresponding equation of Faltinsen (1974).
All the results are based on the assumption that O(β12 ) = O(β2 ). However, even
for periodic solutions we can find a critical value of σ/σ1 for which the amplitude
of the second mode tends to infinity. It can happen for small h, or for σ̄22 → 4 (see
the asymptotic solution (6.3), (6.4)). In terms of σ the condition of the secondary
resonance takes the form
s
σ tanh (2πh/l)
→ = i(2, h̄). (6.9)
σ1 2 tanh (πh/l)
The value i(2, h̄) characterizes the applicability of the theory constructed (see figures 2
and 3). The ratio T1 /T = σ/σ1 must be close to 1 and not close to i(2, h̄).
Similarly, we can introduce for the third mode
s
tanh (3πh/l)
i(3, h̄) = . (6.10)
3 tanh (πh/l)
However since i(3, h̄) < i(2, h̄), the secondary resonance is the most dangerous.
The trend of the distribution of i(2, h̄) shows for h̄ small enough (but large for
shallow water theory) that i(2, h̄) → 1 as h̄ → 0. This means that the secondary
parametric resonance can occur for small depths and implies that the asymptotic
theory presented is not applicable for shallow water.
The stability analysis for surge/pitch excited waves in a rectangular container was
done by Faltinsen (1974). We can give reliable new treatment of the stability by
introducing branches O and S in figures 2 and 3. The branch O is the relation for the
frequency and amplitude for nonlinear free sloshing, which can be found from the
equation
branch O: (σ̄12 − 1) + m1 (σ̄2 , h̄)Ā2 = 0. (6.11)
The branch O is also the asymptotic curve for P− and P+ as A → ∞.
The branch S is the set of all turning points of the branch P+ (or P− for different
depths) for various amplitudes H̄ (surge excitation) or ψ0 (pitch excitation). The
turning points correspond to when (6.5) has only two solutions. The condition of two
roots of equation (6.5) can be found by differentiating (6.5) with respect to A. This gives
branch S: (σ̄12 − 1) + 3m1 (σ̄2 , h̄)Ā2 = 0. (6.12)
The branch S does not depend on the value of the excitation amplitude and is only
a function of depth–breadth ratio.
Due to the theory of bifurcations the turning point divides the branch P+ or P− into
stable and unstable sub-branches. It was shown by Faltinsen (1974) that the upper
sub-branch of P+ /P− corresponds to unstable solutions and the lower sub-branch to
stable solutions. The branch P− /P+ without a turning point corresponds to stable
solutions. When repeating the averaging asymptotic analysis given by Faltinsen for
our solutions, we arrive at the same result if Ā 1.
When varying the values of the excitation amplitude, the sub-branch situated
between S and O will always correspond to unstable solutions.
7. Comparison between theory and experiments

A series of experiments on nonlinear sloshing in a smooth rectangular tank due
to horizontal (surge) excitation were conducted. Figure 4 shows the tank used in the
experiments. The tank has a front plate made of Plexiglas which is stiffened by two
vertical L-beams. The tank was placed on a wagon that could slide back and forth
Figure 4. Picture of the tank.
0.05 m
FS 1
H = 1.05 m
FS 3 FS 2
l = 1.73 m
Figure 5. Tank dimensions and wave probe positions used in the experiments.
controlled by a hydraulic cylinder. The hydraulic system was strong enough to ensure
that the motion inside the tank had little or no effect on the tank motion.
The tank height, breadth and length were respectively 1.05, 1.73 and 0.2 m. The
observed free surface elevation did not vary in the length direction. The amplitude of
surge excitation was between 0.02 and 0.08 m. The water depth was varied between
0.2 and 0.6 m. The tank was equipped with three wave probes, referred to as FS1,
FS2 and FS3 (see figure 5). Wave probes FS1 and FS2 consist of adhesive copper
tape directly placed on the tank wall. FS3 is made of steel wire and is standing
0.05 m from the left wall. The tank position was measured by a position gauge. The
sampling frequency was 50 Hz and the time series were 50 s long. Video recordings and
visual observation of longer simulations, up to 5 minutes, showed that steady-state
oscillations with the forced oscillation period were not achieved. This implies that
the dissipation in the smooth tank is very small even relative to the small damping
predicted by Keulegan (1959). A reason may be that the boundary layer flow is
laminar in Keulegan’s experiments while it is likely to be turbulent in our case. Since
transients do not die out, a beating effect occurs. The most interesting stage for
analysis is during the first 50 s. During this time the beating parameters are stabilized.
After this time the typical behaviour of the sloshing is repeated. Also, the preliminary
analysis has shown that for beating waves of small amplitude the modulated wave is
stabilized for even shorter time.
The free surface elevation had small amplitudes in the initial period after the tank
was excited. In some of the tests the water was in small-amplitude motion before
starting the excitation. Since the proper initial conditions are unknown two different
sets are used to investigate the influence of initial conditions. One set of initial
conditions is
βi (0) = β̇i (0) = 0, i > 1. (7.1)
The other is based on impulse conservation. If vox = σH cos σt, this gives
βi (0) = 0, β̇i (0) = −σPi H, i > 1. (7.2)
The numerical time integrations were done by a fourth-order Runge–Kutta method

and 11 equations of (5.24) were used. The simulation time on a Pentium II–366
1
computer was 300 of the real time.
The examples of figures 6–9 show the effect of initial conditions on free surface
elevation for different forced excitation periods T , water depth h and excitation
amplitude H. So, for example in figure 6 the effect of initial conditions is not
important. However, for the case of figure 7 the condition of impulse conservation
leads to more reasonable description of free surface elevation. Figures 8 and 9 also
demonstrate good agreement between theory and experiments. The agreement is not
perfect in figure 8, but the difference between experimental and numerical simulation
decreases when initial conditions are based on impulse conservation. Better agreement
between theory and experiment can be achieved by realizing that the forced surge
oscillation is not harmonic and does not have a constant amplitude during the initial
period. This is illustrated in figure 8 where the excitation period T was not a constant
during the first 12 s; it varied from 1.76 s to 1.875 s. This is caused by transient rigid
body motions. We assume that these transient motions decay exponentially. This
effect was simulated by varying the period and the amplitude of forced excitation in
the initial phase. Figure 10 shows the effect of only varying the excitation period.
A better agreement with the experiment is then achieved. Separate numerical results
showed that the amplitude has less effect than variation of the frequency. The effect
of varying the frequency can be found qualitatively by examining the steady-state
response in figure 2.
Our theory assumes that the water does not hit the ceiling. The water touches the
ceiling in the case of figure 8, but this does not have an important effect on the
fluid motion. When comparing theoretical and experimental results for a case when
forceful impact occurs, it is evident that they do not agree. A possible reason is energy
dissipation due to the impact. The impact causes the ceiling to vibrate which represents
energy loss for the fluid motion. Since the tank ceiling is very stiff in the model tests,
this is unimportant in the comparative study with experiments. Furthermore, as the
water hits the ceiling a jet is formed and eventually the free surface overturns and
water hits the free surface. This also causes energy dissipation. An estimate of this
energy loss can be calculated by using a generalization of Wagner’s (1932) theory
(Faltinsen & Rognebakke 1999) and assuming that the kinetic and potential energy in
the jet is dissipated. An equivalent linear damping based on energy conservation can
then be included in the differential equations for the generalized coordinates for the
free surface. The damping coefficients α1 β˙1 , α2 β˙2 and α3 β˙3 are introduced in (5.24a) to
(a)
0.04
(m) 0
– 0.04
0 10 20 30 40 50
0.4
Elevation at FS3 (m)
0.2
– 0.2
0 10 20 30 40 50
t (s)
(b)
0.04
(m)
– 0.04
0 10 20 30 40 50
‘Impulse’
0.4
‘Zero’
0.2
– 0.2
0 10 20 30 40 50
t (s)
Figure 6. (a) Measured and (b) calculated tank position and free surface elevation at wave probe
FS3 (h = 0.6 m, T = 1.5 s). The curve ‘Zero’ corresponds to zero initial conditions, ‘Impulse’ means
initial impulse condition.
(a)
0.04
(m) 0
– 0.04
0 10 20 30 40 50
0.4
0.2
– 0.2
0 10 20 30 40 50
t (s)
(b)
0.04
(m)
– 0.04
0 10 20 30 40 50
0.4 ‘Impulse’
‘Zero’
0.2
– 0.2
0 10 20 30 40 50
t (s)
Figure 7. As figure 6 but at T = 1.3 s.
(a)
0.04
(m) 0
– 0.04
0 10 20 30 40 50
0.4
0.2
– 0.2
0 10 20 30 40 50
t (s)
(b)
0.04
(m)
– 0.04
0 10 20 30 40 50
‘Impulse’
0.4 ‘Zero’
0.2
– 0.2
0 10 20 30 40 50
t (s)
Figure 8. As figure 6 but at h = 0.5 m, T = 1.875 s.
(a)
0.04
(m) 0
– 0.04
0 10 20 30 40 50
0.4
0.2
– 0.2
0 10 20 30 40 50
t (s)
(b)
0.04
(m)
– 0.04
0 10 20 30 40 50
‘Impulse’
0.4 ‘Zero’
0.2
– 0.2
0 10 20 30 40 50
t (s)
Figure 9. As figure 6 but at h = 0.5 m, T = 1.4 s.
0.04
(m)
0
– 0.04
0 10 20 30 40 50
‘Impulse’
‘Zero’
0.4
0.2
– 0.2
0 10 20 30 40 50
1.88
1.84
(m)
1.80
1.76
0 10 20 30 40 50
t (s)
Figure 10. Calculated tank position and free surface elevation at wave probe FS3 for h = 0.5 m.
Effect of varying excitation period exponentially from 1.77 s to 1.875 s.
(5.24c), respectively. Since the average forced excitation is close to the lowest natural
frequency, it is only α1 that matters. Figure 11 shows satisfactory agreement between
theory and experiments by including damping. The damping will vary from cycle
to cycle depending on the severity of the water impact. In the presented case we
calculated approximately 40% loss of energy in the tank for every cycle due to the
two impacts occurring.
The theory will break down for small water depth. Figure 12 presents experimental
data and numerical simulation for h/l = 0.173 and T1 /T = 0.96. Since i2 = 0.9,
the effect of secondary parametric resonance is important. We note that the wave
crest is well predicted, while the theoretical value for the trough is clearly lower
than in the experiments. In order to improve the theoretical predictions we have
to assume that at least the two lowest modes have the same order of magnitude.
This means a complete change of the equation system and higher modes have to
be introduced in the nonlinear equations. The introduction of the fourth mode in
the nonlinear equation system will affect the difference between trough and crest so
that the agreement with experiments may improve. The difference between theory
and experiments is more evident in figure 13 where T1 /T = 1.17 and h/l = 0.173.
The reason is once again that the primary mode is not dominating. This contradicts
our theoretical assumptions. Figure 14 shows that the amplitude of the third mode is
higher than the second mode, which is higher than the first mode.
0.8
Experiment
Theory
0.6

0.4
0.2
– 0.2
– 0.4
0 5 10 15
t (s)
Figure 11. Measured and calculated free surface elevation at wave probe FS3 for T = 1.71 s,
h = 0.5 m and H = 0.05 m. Calculations account for wave impact on tank ceiling.
8. Calculations of hydrodynamic loads on the tank

How to calculate hydrodynamic loads will be illustrated for the surge-excited
rectangular tank. The general expression for the pressure is given by (2.5). By noting
that Φ = v0x x + ϕ and by expressing v0x as −Hσ sin (σt) it follows that

∂ϕ 1 2 2 1 2 2 2
p = p0 − ρ + 2 (∇ϕ) + gz − σ H cos (σt)x − 2 H σ sin (σt) . (8.1)
∂t
Here we use
X N
iπ iπ l cosh ((iπ/l)(z + h))
∇ϕ = Ri (t) − sin x+ , 0,
i=1
l l 2 cosh ((iπ/l)h)

iπ l sinh ((iπ/l)(z + h))
cos x+ , (8.2)
l 2 cosh ((iπ/l)h)

∂ϕ X
N
iπ l cosh ((iπ/l)(z + h))
= Ṙi (t) cos x+ , (8.3)
∂t i=1
l 2 cosh ((iπ/l)h)
where Ri and Ṙi are calculated by (5.12) and (5.13) and N is a number of Fourier
terms (N > 3). When applying these formulas above the mean free surface, a Taylor
expansion about the mean free surface has to be used.
The force F on the tank due to the fluid can be calculated by direct pressure
integration or the compact formula derived by Lukovsky (1990)
F = ml g − ml [v̇ 0 + ω × v 0 + ω × (ω × r 1C ) + ω̇ × r 1C + 2ω × ṙ 1C + r̈ 1C ] (8.4)
where r 1C is radius-vector of the mass centre in mobile coordinate system Oxyz and
ml is fluid mass. We note that F includes the static force component ml g in addition
to hydrodynamic forces.
The formula takes the form
F = ml g − ml (v̇ 0 + r̈ 1C ) (8.5)
in the absence of angular motions (ω = 0).
(a)
0.04
(m) 0
– 0.04
0 10 20 30 40 50
0.6
0.4
0.2
– 0.2
0 10 20 30 40 50
t (s)
(b)
0.04
(m)
0
– 0.04
0 10 20 30 40 50
‘Impulse’
0.6 ‘Zero’
0.4
0.2
– 0.2
0 10 20 30 40 50
t (s)
Figure 12. (a) Measured and (b) calculated tank position and free surface elevation at wave probe
FS3 (h = 0.3 m, T = 2.2 s).
(a)
0.04
(m) 0
– 0.04
0 10 20 30 40 50
0.6
0.4
0.2
– 0.2
0 10 20 30 40 50
t (s)
(b)
0.04
(m)
0
– 0.04
0 10 20 30 40 50
‘Impulse’
0.6 ‘Zero’
0.4
0.2
– 0.2
0 10 20 30 40 50
t (s)
Figure 13. As figure 12 but at T = 1.8 s.
Mode 1
0.6
0.4
b3(t) (m)
0.2
– 0.2
0 5 10 15 20 25 30 35 40 45 50
Mode 2
0.6
0.4
b3(t) (m)
0.2
– 0.2
0 5 10 15 20 25 30 35 40 45 50
Mode 3
0.6
0.4
b3(t) (m)
0.2
– 0.2
0 5 10 15 20 25 30 35 40 45 50
t (s)
Figure 14. Contribution of the three lowest modes to the calculated free surface elevation at wave
probe FS3; h = 0.3 m, T = 1.8 s.
– 0.26
z (m) – 0.27
– 0.28
– 0.29
– 0.30
– 0.2 – 0.1 0 0.1 0.2
x (m)
Figure 15. The position of mass centre for the case in figure 6.
The calculation shows, that if r 1C = (xC (t), 0, zC (t)), then

l X 1 X 2
N n
1 h
xC = − βi (t) 2 (1 + (−1)i+1 ), zC = − + β (t), (8.6)
π h i=1
2 i 2 4h i=1 i
where the point (0, −h/2) corresponds to mass centre of unperturbed fluid.
By introducing the vector F = (Fx , 0, Fz ) we arrive at
! 
l X
N
1 

2
Fx /ml = Hσ cos σt + 2 i+1
β̈i (t) 2 (1 + (−1) ) , 
π h i=1 i 
! (8.7)
1 X
N


Fz /ml = − g + (β̈i βi + β̇i2 ) . 

2h 
i=1
Figure 15 shows the trajectory of the mass centre. Figure 16 presents the trajectory
of the end of the vector F /ml .
The hydrodynamic moment N on the tank can also be calculated by the special
formula derived by Lukovsky (1990) (moment axis coincides with Oy)
N = ml r 1C × (g − ω × v 0 − v̇ 0 ) − J 1 · ω̇ − J̇ 1 · ω − ω × (J 1 · ω)
−l̈ ω + l̇ ωt − ω × (l̇ ω − l ωt ), (8.8)
where the inertia tensor J 1 is defined by (3.13) and l ω , l ωt by (3.15).
For ω = 0
N = ml r 1C × (g − v̇ 0 ) − l̈ ω + l̇ ωt . (8.9)
The time-varying functions l ω , l ωt depend on the solutions of the Neumann boundary
value problem (3.5) and can be expressed mathematically like the Stokes–Zhukovsky
potentials.
By using Green’s formula we get
Z
d
N(t) = ml (xC g − zC v̇0 ) − ρ (zν1 − xν3 )ϕ dS, (8.10)
dt S+Σ
where N = (0, N(t), 0).
–9.792
–9.796
Fz /ml (N kg –1)
–9.800
–9.804
–9.808
–2 –1 0 1 2
Fx /ml (N kg –1)
Figure 16. The trajectory of the vector of the calculated hydrodynamic force (Fx /ml , Fz /ml ) on the
tank for the case in figure 6.
This is not as simple as the formula (8.7) for the force, but is useful in a verification
procedure by comparing with the direct pressure integration of the moment. This
should in both cases be derived correct to O().
9. Conclusions
I. Using the Bateman–Luke variational principle, we generalize the procedure
proposed by Miles (1976) and Lukovsky (1976) to derive a modal system describing
nonlinear sloshing of an incompressible perfect fluid with irrotational flow partly
occupying a tank performing an arbitrary three-dimensional motion. If the tank
has vertical walls near the mean free surface, this procedure leads to an infinite-
dimensional system of nonlinear differential equations coupling the generalized time-
dependent coordinates. No assumptions about the order of smallness are made. It
applies to any type of rigid body motion. The surface and domain modes do not need
to be natural modes. This means that the multidimensional modal discrete system
derived has the most general form of the modal equations and can be used for
modelling different ‘fluid–structure’ problems including the problems associated with
transient sloshing and coupled ‘ship–fluid cargo’ motions.
II. Two-dimensional sloshing in a rectangular smooth tank with finite water depth has
been studied theoretically. The tank is oscillating with arbitrary rigid body motions
of small magnitude with an average frequency close to the lowest natural frequency
of the fluid motion. A finite-dimensional asymptotic model with multiple degrees
of freedom is derived. This is based on the general discrete infinite-dimensional
modal model. The lowest mode is assumed dominant. Each mode has different
order of magnitude. The three lowest modes are interacting nonlinearly with each
other. An important feature relative to other established nonlinear theories is that
transient effects can be described. Since the theory is expressed in terms of a set of
nonlinear ordinary differential equations in time, it is considerably simpler than a
direct numerical solution of the fluid motion.
Periodic solutions are studied analytically. The amplitude–frequency response is
consistent with the fifth-order steady-state solution by Waterhouse (1994).
It is shown that the theory is not valid when the water depth (h) becomes small
relative to the tank breadth (l). This is due to secondary parametric resonance. It
is then necessary to include nonlinearly interacting modes having the same order of
magnitude. This is demonstrated for a tank with h/l = 0.173.
III. We have conducted experimental studies of the free surface elevation for forced
surge oscillations of two-dimensional flow in a rectangular tank. It is demonstrated
experimentally that it takes a very long time for transient fluid motion to die out.
This did not occur during an observation period of 5 minutes, which corresponds to
the order of 150–200 oscillations in terms of the excitation period. The consequence
is that steady-state solutions of nonlinear sloshing in a smooth tank can have limited
applicability. Modulated (‘beating’) waves occurred as a consequence of transient and
forced oscillations. The amplitude/‘beating’ period was stabilized during the first 50 s.
Since we could not exactly state what the initial conditions were in the experiments,
a sensitivity study was performed with different initial conditions in the theoretical
model. The results were not strongly dependent on this, but better agreement between
theory and experiments was in general obtained by using an initial condition based on
impulse conservation. For several experiments we observed fluctuations of the excita-
tion frequency in an initial period up to approximately 10 s. This effect was important
to include in the theoretical model. There is good agreement with experimental free
surface elevation when h/l > 0.28.
IV. The theory was compared with experiments when forceful water impact on the
tank ceiling occurred. The theory assumes no tank ceiling. The experimental free
surface elevations showed a clear influence of the water impact. It was speculated
that this was due to energy dissipation and phenomenological linear damping terms
were introduced in the discrete modal system. Good agreement with the experiments
was demonstrated. This is an area of future research.
V. It is shown how hydrodynamic forces on the tank can be calculated in a simple
way. An alternative formula for the hydrodynamic moment is also presented. The
form of the expressions facilities simulations of a coupled ‘vehicle–fluid’ system.
This work is supported in part by NATO Research Fellowship (Research Council

of Norway) at Norwegian University of Science and Technology, Trondheim (fourth
author), German Research Council (D.F.G.) (third author). The work by the second
author is supported by the Research Council of Norway. The experimental studies
were sponsored by Det Norske Veritas.
REFERENCES
Bateman, H. 1944 Partial Differential Equations of Mathematical Physics. Dover.
Buechmann, B. A. 1996 2D numerical wave based on a third order boundary element model. In 9th
Conf. European Consortium for Mathematics in Industry, Lyngby/Copenhagen, Denmark, June
25–27, 1996, pp. 417–420.
Chen, Sh., Johnson, D. B., Raad, P. E. & Fadda, D. 1997 Surface marker and micro cell method.
Intl J. Numer. Meth. Fluids 25, 749–778.
Dodge, F. T., Kana, D. D. & Abramson, H. N. 1965 Liquid surface oscillations in longitudinally
excited rigid cylindrical containers. AIAA J. 3, 685–695.
Faltinsen, O. M. 1974 A nonlinear theory of sloshing in rectangular tanks. J. Ship. Res. 18,
224–241.
Faltinsen, O. M. & Rognebakke, O. F. 1999 Sloshing and slamming in tanks. In Hydronav’99.–
Manoeuvering’99 Gdansk – Ostroda, 1999, Poland. Technical University of Gdansk.
Funakoshi, M. & Inoue S. 1991 Bifurcations in resonantly forced water waves. Eur. J. Mech.
B/Fluids. 10, 31–36.
Hargneaves, R. 1908 A pressure–integral as kinetic potential. Phil. Mag. 16, 436–444.
Ikeda, T. & Nakagawa, N. 1997 Non-linear vibrations of a structure caused by water sloshing in
a rectangular tank. J. Sound Vib. 201, 23–41.
Keulegan, G. H. 1959 Energy dissipation in standing waves in rectangular basin. J. Fluid Mech. 6,
33–50.
Limarchenko, O. S. & Yasinsky, V. V. 1997 Nonlinear Dynamics of Constructions with a Fluid. Kiev
Polytechnic University (in Russian).
Luke, J. C. 1967 A variational principle for a fluid with a free surface. J. Fluid Mech. 27, 395–397.
Lukovsky, I. A. 1976 Variational method in the nonlinear problems of the dynamics of a limited
liquid volume with free surface. In Oscillations of Elastic Constructions with Liquid, pp. 260–264
Moscow: Volna (in Russian).
Lukovsky, I. A. 1990 Introduction to Nonlinear Dynamics of a Solid Body with a Cavity including a
Liquid. Kiev: Naukova dumka (in Russian).
Lukovsky, I. A. & Timokha, A. N. 1995 Variational Methods in Nonlinear Dynamics of a Limited
Liquid Volume. Kiev: Institute of Mathematics (in Russian).
Mikishev, G. I. 1978 Experimental methods in the dynamics of spacecraft. Moscow: Mashinostroenie
(in Russian).
Miles, J. W. 1976 Nonlinear surface waves in closed basins. J. Fluid Mech. 75, 419–448.
Miles, J. W. 1984a Internally resonant surface waves in a circular cylinder. J. Fluid Mech. 149, 1–14.
Miles, J. W. 1984b Resonantly forced surface waves in a circular cylinder. J. Fluid Mech. 149, 15–31.
Moan, T. & Berge, S. (Eds.) 1997 Report of Committee 1.2 “Loads”. In Proc. 13th Intl Ship and
Offshore Structures Congress, Vol. 1, pp. 59–122. Pergamon.
Moiseev, N. N. 1958 To the theory of nonlinear oscillations of a limited liquid volume of a liquid.
Prikl. Math. Mech. 22, 612–621 (in Russian).
Narimanov, G. S. 1957 Movement of a tank partly filled by a fluid: the taking into account of
non-smallness of amplitude. Prikl. Math. Mech. 21, 513–524 (in Russian).
Narimanov, G. S., Dokuchaev, L. V. & Lukovsky, I. A. 1977 Nonlinear Dynamics of Flying
Apparatus with Liquid. Moscow: Mashinostroenie (in Russian).
Pawell, A. 1997 Free Surface Waves in A Wave Tank. Intl Series Numer. Maths 124, 311–320.
Birkhauser.
Pilipchuk, V. N. & Ibrahim, P. A. 1997 The dynamics of non–linear system simulating liquid
sloshing impact in moving structures. J. Sound Vib. 205, 593–615.
Shemer, L. 1990 On the directly generated resonant standing waves in a rectangular tank. J. Fluid
Mech. 217, 143–165.
Solaas, F. & Faltinsen, O. M. 1997 Combined numerical and analytical solution for sloshing in
two-dimensional tanks of general shape. J. Ship Res. 41, 118–129.
Su Tsung-Chow 1992 Nonlinear sloshing and the coupled dynamics of liquid propellants and
spacecraft. NASA Tech. Rep. AD-A250023.
Tanizawa, K. 1996 A nonlinear simulation method of 3D body motions in waves extended
formulation for multiple fluid domains. In 11th Intl Workshop on Water Waves and Floating
Bodies, March 1996, Hamburg, Germany. Abstracts.
Tsai, W.-T., Yue, D. K.-P. & Yip, K. M. K. 1990 Resonantly excited regular and chaotic motions in
a rectangular wave tank. J. Fluid Mech. 216, 343–380.
Verhagen, J. H. G. & Wijngaarden, L. van 1965 Nonlinear oscillations of fluid in a container. J.
Fluid Mech. 22, 737–751.
Wagner, H. 1932 Uber Stoss- und Gleitvorgange an der Oberflache von Flussigkeiten. Z. Angew.
Math. Mech. 12, 193–235.
Waterhouse, D. D. 1994 Resonant sloshing near a critical depth. J. Fluid Mech. 281, 313–318.

Tank Sloshing

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Tank Sloshing

Uploaded by

Copyright:

Available Formats

J. Fluid Mech. (2000), vol. 407, pp. 201–234.

Printed in the United Kingdom 201

Multidimensional modal analysis of nonlinear

(Received 29 January 1999 and in revised form 18 October 1999)

The discrete infinite-dimensional modal system describing nonlinear sloshing of an

2. Free boundary value problem

3. Derivation of the general modal system by the variational method

where fi (x, y) is a complete (to within a constant)

These components Jij1 can be calculated by Green’s formula, i.e.

The Lagrangian L (3.11) can be rewritten as

X ∂An X X ∂Ank ∂l1ω ∂l2ω ∂l3ω ∂l1ωt ∂l2ωt

4. Modal system for two-dimensional fluid flows

The velocity potential Φ(x, 0, z, t) takes the form

5. Asymptotic modal system for a rectangular tank performing arbitrary

Figure 1. Coordinate system.

By calculating λij we get

It follows from the volume conservation condition (5.2) that

The derivatives with respect to βi give

(β̈2 + σ22 β2 ) + d4 β̈1 β1 + d5 β̇12 + Q2 v̇0z β2 = 0, (5.24b)

6. Steady-state sloshing in a rectangular tank with a small-amplitude

7. Comparison between theory and experiments

Figure 4. Picture of the tank.

The numerical time integrations were done by a fourth-order Runge–Kutta method

Elevation at FS3 (m)

8. Calculations of hydrodynamic loads on the tank

The calculation shows, that if r 1C = (xC (t), 0, zC (t)), then

This work is supported in part by NATO Research Fellowship (Research Council

You might also like