See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/30495699

Dynamics of liquid-filled spacecraft

Article in Journal of Engineering Mathematics · January 2003

DOI: 10.1023/A:1022055916067 · Source: OAI

CITATIONS READS

87 360

2 authors, including:

Arthur E. P. Veldman
University of Groningen
199 PUBLICATIONS 4,405 CITATIONS

SEE PROFILE

All content following this page was uploaded by Arthur E. P. Veldman on 14 August 2014.

The user has requested enhancement of the downloaded file.

 Journal of Engineering Mathematics 45: 21–38, 2003.
© 2003 Kluwer Academic Publishers. Printed in the Netherlands.

Dynamics of liquid-filled spacecraft

J. GERRITS∗ and A.E.P. VELDMAN

Department of Mathematics, University of Groningen, P.O. Box 800, 9700 AV Groningen, The Netherlands
(e-mail: [email protected])

Received 22 May 2002; accepted in revised form 24 September 2002

Abstract. A method is presented for simulating coupled liquid-solid dynamics. An important example of a cou-
pled liquid-solid system is a satellite carrying fuel. The dynamics of the satellite and the onboard fuel influence
each other, which may lead to satellite motion that is uncontrollable. For better understanding of the complex
dynamics of coupled systems, a numerical model is developed. The model consists of two parts. The first part that
solves the liquid motion is only briefly discussed here. The focus in this paper is on the way in which the dynamics
of the liquid and the solid body are coupled. For this, the governing equations are presented in which terms appear
that represent the force and torque on the solid body due to the sloshing liquid. The governing equations are
rewritten such that the discrete approximation of these equations can be integrated in a stable manner for arbitrary
liquid/solid mass ratios. Results are presented demonstrating the stability of the present model. A grid-refinement
study and a time-step analysis are performed. Finally, the flat-spin motion of a satellite, partially filled with liquid,
that flew in 1992 as part of the Wet Satellite Model experiment is studied. Results from the simulation are compared
with the actual flight data.

Key words: coupled dynamics, flat spin, liquid sloshing, microgravity

1. Introduction

Usually, in engineering applications, the motion of a container carrying liquid is not pre-
scribed, but follows from an interaction with the liquid motion: the sloshing liquid induces a
force and torque on the container, which, as a consequence, starts to accelerate. In return, this
container motion influences the liquid motion, and so on. This coupling between the liquid
dynamics and the dynamics of the solid body is called dynamical interaction.
Dynamical interaction is often observed when liquid is transported by vehicles. It is well
known, for example, that the braking distance of a tank lorry is considerably influenced by
the amount of fuel it is carrying [1]. Aboard oil tankers advantage is taken of dynamical
interaction: in so-called anti-roll tanks, water sloshes (due to the roll motion of the vessel)
such that it stabilises the motion of the tanker [2, 3].
Also in space, aboard satellites carrying fuel, dynamical interaction plays an important
role. Regularly, satellites need course corrections to remain in the correct orbit. These are
accomplished by firing small thrusters. This, however, will also induce liquid motion and the
sloshing liquid may affect the satellite motion, such that the desired course correction may not
be accomplished. A recent example of a satellite that got into problems was NASA’s NEAR
(Near Earth Asteroid Rendezvous) satellite, which in 1998 was on its way to the asteroid
433 Eros [4]. A propellant burn that should have put the spacecraft on track was aborted
∗ present affiliation: Institute for Marine and Atmospheric Research Utrecht, Utrecht University, P.O. Box 80005,
3508 TA Utrecht, The Netherlands
22 J. Gerrits and A.E.P. Veldman

after sensors detected accelerations that exceeded safety limits programmed into its onboard
computer. Further investigation revealed that the spacecraft was tumbling due to sloshing of
the onboard liquid. This incident eventually caused a 13 months delay in the mission.
Dynamical interaction in space is a particularly complex problem. Not only because of the
complex dynamics of a coupled system, but also because of the poor understanding of liquid
behaviour under micro-gravity conditions. However, in order to prevent the future loss of
expensive satellites, it is important to improve our knowledge of coupled solid-liquid systems
in an extra-terrestrial environment.
In this paper a method is presented for the numerical simulation of coupled solid-liquid
dynamics. The method consists of both a model for simulating liquid dynamics in a micro-
gravity environment (however, terrestrial liquid dynamics is supported as well) and a model
for the solid-body dynamics in which forces and torques due to the sloshing liquid appear. In
the latter model the governing equations are written in a suitable reference frame such that
the integration is stable for arbitrary solid/liquid mass ratios. The numerical model for the
solid-body dynamics is the focus of this paper; the liquid dynamics is discussed only briefly
(more details can be found in [5, Chapter 2]). The model has been implemented in a computer
program of which results are presented.

2. Liquid dynamics

First the mathematical model for the liquid dynamics is explained. Also, some key aspects of
the numerical model are discussed.

2.1. G OVERNING EQUATIONS

The motion of an incompressible liquid (with density ρ and kinematic viscosity µ) is governed
by conservation of mass

∇ · u = 0, (1)

and conservation of momentum

Du 1
= − (∇p − (∇ · µ∇) u) + F + f . (2)
Dt ρ
In these equations p is the pressure, u = (u, v, w)T is the velocity vector in a moving
reference frame (moving with, for example, the satellite) and F is an acceleration due to
external forces (gravity in this paper). The vector f is the acceleration due to a virtual body
force representing the fact that the equations are written in a moving reference frame. This
acceleration can be written as

f = −q̇ − ω̇ × r − ω × (ω × r) − 2ω × u, (3)

where q̇ = dq/dt + ω × q is the acceleration of the moving origin with respect an inertial
reference frame, ω̇ = dω/dt + ω × ω = dω/dt is the angular acceleration of the moving
reference frame, ω is the angular velocity of the moving reference frame and r is the position
of a liquid particle in the moving reference frame. Note that in the equations above, the vector
u is the velocity in the moving reference frame. If the velocity in the inertial reference frame
is denoted by v, then the following relation holds:
 Dynamics of Liquid-Filled Spacecraft 23
Dv Du
= q̇ + ω̇ × r + ω × (ω × r) + + 2ω × u. (4)
Dt Dt

2.2. F EATURES OF THE NUMERICAL MODEL

Since the coupled dynamics is the focus of this paper, the numerical model for the liquid
dynamics is discussed very briefly.

2.2.1. Discretisation
The continuity equation (1) and Navier-Stokes equations (2) are discretised with a finite-
volume method [6] on a staggered, Cartesian grid [7]. Complex geometries are incorporated
using the cut-cell technique [8, 9, 10]. The discrete operators for convection and diffusion
are constructed such that their symmetry properties correspond to those of the continuous
operators, i.e., convection is approximated by a skew-symmetric operator and diffusion is
approximated by a symmetric, negative definite operator. Further, the pressure gradient is
discretised such that its discrete coefficient matrix is equal to minus the transpose of the
discrete divergence operator. With these properties it can be proven that (in the absence of
external forces) the discrete kinetic energy cannot grow, which makes the model extremely
robust [11].

2.2.2. Free surface

The location of the free surface is captured using a Volume-of-Fluid function F s , which is
defined in every computational cell. It denotes whether a cell contains no fluid (F s = 0), is
completely filled (F s = 1) or contains part of the free surface (0 < F s < 1). The basic
advection algorithm is based on the method of Hirt and Nichols [12]. However, since the
original method of Hirt and Nichols creates a lot of ‘flotsam’ and ‘jetsam’, the method is
adapted with a local height function [13]. In cells that contain part of the free surface (surface
cells) a height function is constructed in a block of 3 × 3 × 3 cells. In these surface cells, the
free surface is advected by changing the value of the local height function instead of using
the method of Hirt and Nichols. By using a local height function most of the ‘flotsam’ and
‘jetsam’ is removed. Moreover, this method conserves mass rigorously.

3. Solid-body dynamics

In this section the solid-body dynamics is discussed. First, the governing equations are stated.
Hereby, the choice of reference frame is extremely crucial for the stability of the numerical
integration as is demonstrated. Thereafter, the solution method is explained.

3.1. G OVERNING EQUATIONS

The model for the solid-body motion consists of an equation for linear momentum

ms q̇ + ω̇ × ms r s + ω × (ω × ms r s ) = F + ms F , (5)

and an equation for angular momentum

ms r s × q̇ + I s ω̇ + ω × I s ω = T + ms r s × F . (6)
24 J. Gerrits and A.E.P. Veldman

In these equations q and ω are the unknown variables, representing the linear and angular
velocity of the solid body, respectively. The mass of the solid body is denoted by ms . Further,
I s is the moment-of-inertia tensor and r s is the centre of mass of the solid body. Note that
the latter two quantities are relative to the moving reference frame. These equations can be
simplified considerably by setting r s = 0. However, the present form is more convenient in
the rest of this section where these equations are rewritten in a different coordinate system.
The last terms in (5) and (6) represent external forces and torques. Finally, F and T are,
respectively, the force and torque that the fluid, via pressure (normal stress) and viscous effects
(mainly tangential stress), exerts on the boundary of the solid body, i.e.,


F = (pE 3 − µ∇u) · n dS,


∂V

T = (r × (pE 3 − µ∇u)) · n dS.

∂V

Here, E 3 is the 3 × 3 identity matrix and n the outward pointing normal on the boundary ∂V
of the total volume V of the solid body. When the divergence theorem is used, these integrals
can be written as integrals over the volume V
   
Dv
F = ∇p − (∇ · µ∇) u dV = − ρ − F dV ,
Dt

V V   
Dv
T = r × (∇p − (∇ · µ∇) u) dV = − ρr × − F dV .
Dt
V V

In the last step the Navier-Stokes equations (2) and Equations (3) and (4) were used. Note that
this integration in the previous equations is not only over the liquid volume, but also over the
void. In the void the density is set to zero, such that the void region does not contribute to the
integral.
Direct time integration of the Equations (5) and (6) would result in a method that is not
stable for arbitrary liquid/solid mass ratios (this is demonstrated in Section
 4). Therefore the
system for the solid-body dynamics is rewritten first. Consider F = − V ρ (Dv/Dt − F ) dV
in the right-hand side of Equation (5). Using relation (4) for the liquid velocity v in an inertial
reference frame and the liquid velocity u in a moving reference frame, the force F is rewritten
as
  
Du
F = − ρ q̇ + ω̇ × r + ω × (ω × r) + + 2ω × u − F dV .
Dt
V

Since the linear velocity q and the angular velocity ω of the solid body are constant over the
volume V , this is equal to
 
     
Du
F = −q̇ ρ dV − ω̇ × ρr dV − ω × ω × ρr dV  − ρ + 2ω × u − F dV .
Dt
V V V V

In short, denoting the liquid mass with ml and the centre of mass of the liquid with r l (with
respect to the moving reference frame), this can be written as
 Dynamics of Liquid-Filled Spacecraft 25
  
Du
F = −ml q̇ − ω̇ × ml r l − ω × (ω × ml r l ) − ρ + 2ω × u − F dV .
Dt
V

Similarly, the torque T on the right-hand side of Equation (6) can be rewritten, which results
in
  
Du
T = −ml r l × q̇ − I l ω̇ − ω × I l ω − ρr × + 2ω × u − F dV ,
Dt
V

where I l is the moment-of-inertia tensor of the liquid (with respect to the moving refer-
ence frame). Substituting these expressions for F and T in the Equations (5) and (6) and
rearranging terms gives an alternative form of the model for the solid-body dynamics, namely
  
Du
mq̇ + ω̇ × mr + ω × (ω × mr) = − ρ + 2ω × u − F dV + ms F , (7)
Dt
V
  
Du
mr × q̇ + I ω̇ + ω × I ω = − ρr × + 2ω × u − F dV + ms r s × F . (8)
Dt
V

In these equations m = ms + ml is the total mass and I = I s + I l the moment-of-inertia

tensor of the coupled system. The centre of mass of the coupled system is denoted with
r = (ms r s + ml r l ) /m. Note that the sloshing liquid makes the latter two quantities time
dependent.
An important difference between Equations (5) and (6) on the one side and Equations (7)
and (8) on the other is the distribution of the solid-body mass and the liquid mass over the
left-hand side and right-hand side of the equations. In the former system (5) and (6) the left-
hand side contains the mass of the solid body, while on the right-hand side the mass of the
liquid, which is moving in the inertial reference frame, appears. Intuitively, one can see that
solving this system iteratively, where the left-hand side and right-hand side are evaluated at
the new and old time level, respectively, can become unstable if the liquid mass is too large
with respect to the mass of the solid body. In the latter system (7) and (8) the total mass of the
coupled system, i.e., the sum of the solid-body mass and the liquid mass, appears on the left-
hand side. Now, the right-hand side contains the liquid mass in the moving reference frame. In
this case, solving this system iteratively can be done in a stable manner, since the liquid mass
is always smaller than the total mass of the coupled system. A more detailed stability analysis
is presented in Section 4.1.

3.2. S OLUTION METHOD

The temporal discretisation of the equations for linear momentum (7) and angular momen-
tum (8) is straightforward. Both the linear acceleration dq/dt and the angular acceleration
dω/dt are discretised at the new time level (indicated by a superscript n + 1). The linear and
angular velocity are discretised at the old time level (indicated by a superscript n). Since the
model for the solid-body dynamics is applied after the model for the liquid dynamics has been
completed (in the same time step), all the quantities from the liquid model carry a superscript
n + 1. This results in
dq n+1 dω n+1
m + × mr n+1 = Ln , (9)
dt dt
26 J. Gerrits and A.E.P. Veldman

dq n+1 dω n+1
mr n+1 × + I n+1 = An , (10)
dt dt
where 
Ln = −mωn × q n − ωn × ωn × mr n+1 − ρa n+1 dV + ms F n+1 , (11)
V

A = −mr
n n+1
× ω ×q
n n
−ω ×I
n n+1
ω −
n
ρr n+1 × a n+1 dV + ms r s × F n+1 . (12)
V

In these equations a n+1 = Dun+1 /Dt + 2ωn × un+1 − F n+1 . Note that the volume V is the
volume of the solid body (recall that the volume integral was deduced, via the divergence
theorem, from a boundary integral over the boundary of the solid body); hence this volume
V does not carry a superscript, since it does not depend on time. However, in that part of the
solid body where liquid is absent, the integrand vanishes, because the density is taken to be
zero in the void region.
The implementation of the non-integral terms in Equations (9)–(12) is trivial. The imple-
mentation of the remaining terms,  although being far from difficult, requires some attention.
All these terms are of the form V ρϕ n dV and are discretised as

.
ρϕ n dV = n
ρϕi,j,k s
Fi,j,k δxi δyj δzk , (13)
V i,j,k

where the summation is over all the computational cells, ϕ is some scalar variable (e.g.,
s
Du/Dt), Fi,j,k is the volume-of-fluid function in cell (i, j, k) and δxi δyj δzk is the volume
of this cell. The variable ϕ is evaluated in cell centres. This means that, sometimes, staggered
velocities have to be averaged before they can be evaluated. Note that also the centre of mass
r l and the moment-of-inertia tensor I l of the liquid are integral terms, since
 
ml r l = ρr dV and I l ψ = ρr × (ψ × r) dV ,
V V
where ψ is some vector (e.g., dω/dt) the moment-of-inertia tensor operates on. Therefore,
these quantities are computed using (13) as well.
In matrix form, the 6 × 6 linear system consisting of Equations (9) and (10) reads
   
m 0 0 0 mr z −mr y Lx
   
 0 m 0 −mr z 0 mr x   Ly 
   
   
 0 0 m mr y −mr x 0  L 
 ξ =  z , (14)
   
 0 −mr z mr y Ixx −Ixy −Ixz   Ax 
   
 mr −mr −I −I  A 
 z 0 x xy I yy yz   y
−mr y mr x 0 −Ixz −Iyz Izz Az
T
where ξ = dqx /dt, dqy /dt, dqz /dt, dωx /dt, dωy /dt, dωz /dt . For presentational reasons
the superscripts have been dropped (it will be clear that the matrix and right-hand side are
evaluated at time level n, while the unknown vector contains quantities at time level n +
1). Further the components of a vector ψ are denoted with subscripts x, y and z, i.e., ψ =
T
ψx , ψy , ψz , and the moment-of-inertia tensor has components
 Dynamics of Liquid-Filled Spacecraft 27
 
Ixx −Ixy −Ixz
 
I = 
 −Ixy Iyy −Iyz  .
−Ixz −Iyz Izz
By Gaussian elimination, the lower-left 3×3 block of the matrix in (14) is eliminated, whence
a decoupled system for dω/dt arises, namely
  2     
r y + r 2z −r x r y −r x r z dωx /dt Rx
      
I − m  −r x r y r 2 + r 2 −r y r z   dωy /dt  =  Ry  , (15)
  x z     
−r x r z −r y r z r 2x + r 2y dωz /dt Rz

where Rx = Ax + r z Ly − r y Lz , Ry = Ay − r z Lx + r x Lz and Rz = Az + r y Lx −
r x Ly . Note that during this process no divisions (in particular through zero) take place. Using
Steiner’s theorem, it follows that the 3 × 3 matrix in (15) is the moment-of-inertia tensor
relative to the center of mass of the coupled system. The 3 × 3 linear system for dω/dt is
solved with Gaussian elimination followed by backward substitution. Once the solution for
dω/dt is known, the solution for dq/dt can be found using backward substitution in (14).
In the model for the solid-body dynamics and also in the Navier-Stokes equations the linear
and angular velocity of the solid body appears. This means that dq n+1 /dt and dωn+1 /dt have
to be integrated in time, which is done using a fourth-order Runge-Kutta method. To this end,
system (14) is rewritten in the form
dy
= g (t, y) , (16)
dt
where y = (q, ω)T and g = M −1 (L, A)T . The matrix M is an abbreviation for the matrix
that appears in (14). Equation (16) is integrated in time from level n to level n + 1 by
1
y n+1 = y n + δt (S 1 + 2S 2 + 2S 3 + S 4 ) ,
6
where the four stages S 1 , . . . , S 4 are defined as
 
1 1
S 1 = g (t , y ) ,
n n
S 2 = g t + δt, y + δtS 1 ,
n n
2 2
 
1 1
S 3 = g t + δt, y + δtS 2 ,
n n
S 4 = g t n + δt, y n + δtS 3 .
2 2
Note that the explicit time dependence in the right-hand side of (16) (i.e., the first argument of
the function g) is entirely due to the liquid model. Since the models for the liquid dynamics
and the solid-body dynamics are treated separately, it is most unconvenient to evaluate the
position and velocity of the liquid at the intermediate time level n+ 12 . Moreover, the quantities
of the liquid model are already known at time level n + 1. Hence, in the Runge-Kutta stages
S 1 , . . . , S 4 , the first argument of g is frozen at time level n + 1.
In the simulations that have been performed with the method presented in this paper, the
solid-body dynamics turns out to be more sensitive to small disturbances than the model for the
liquid dynamics, hence requiring either a smaller time step or a more accurate time-integration
method. Since the latter option has been implemented, in practice, it is possible to set the time
step for the solid-body model equal to the one used in the liquid model. Should this not be
28 J. Gerrits and A.E.P. Veldman

Figure 1. Left: pressure [N/m2 ] at equidistant locations along the z-centerline of the tank as a function of time
[s]. Right: acceleration [m/s2 ] (dashed), velocity [m/s] (dotted) and position [m] (dash-dotted) in z-direction of the
coupled system as a function of time [s]. The solid line represents the same quantities in x- and y-directions.

sufficient, then it is possible to take multiple Runge-Kutta time steps during one single time
step in the liquid model.

4. Results

The method presented in this paper has been implemented in a computer program. Some
results of the present method are discussed in this section. First, the free fall of a container
partially filled with fluid is considered for studying the stability of the method. Then, a grid-
refinement study and a time-step analysis is performed and the effect of the liquid/solid mass
ratio is studied, followed by the simulation of a flat-spin motion of a rectangular container
partially filled with liquid. Finally, the Wet Satellite Model experiment is introduced and
simulated.

4.1. F REE FALL

As a first example, the free fall of a rectangular tank partially filled with liquid is simulated
using the model (7) and (8), with which no stability problems are expected. Consider a hollow,
rectangular tank of dimensions 1 × 2 × 4 m3 with mass ms = 2000 kg. The lower half
(in z-direction) of the tank (i.e., with height h = 2 m) is filled with liquid having a mass
ml = 4000 kg (which corresponds to a density ρ = 1000 kg/m3 ). The acceleration F due to
gravity is set to (0, 0, −g)T with g = 10 m/s2 .
During the first 0·05 seconds of the simulation (on a computational grid of 10 × 20 × 40
cells) the tank is kept at its initial position, whence the pressure settles itself at hydrostatic
values, which serves as a simple test for the liquid model. After 0·05 seconds the tank, together
with the liquid, is released. This, of course, causes the coupled system to fall in the negative z-
direction with an acceleration equal to gravitational acceleration. Also, the pressure becomes
equal to the ambient pressure p0 = 0 N/m2 everywhere in the fluid, since the coupled system
is in a free fall. These results are confirmed in Figure 1. On the left in this figure, the pressure,
at equidistant locations along the centreline in z-direction, is plotted. At the bottom of the
tank, the pressure is equal to p = ρgh = 1000 × 10 × 2 = 20000 N/m2 . Taking into account
that the centre of the first computational cell lies half a mesh size (which is δz/2 = 0·05 m)
above the bottom of the tank, the maximum value of p = 19500 N/m2 in Figure 1 can be
 Dynamics of Liquid-Filled Spacecraft 29

Figure 2. Acceleration [m/s2 ] in z-direction of the coupled system as a function of time [s] as predicted by the
unstable method.

explained. The same reasoning applies to the centre of the last cell below the free surface,
where the pressure is equal to p = 500 N/m2 (at the free surface the pressure has a value of
p = p0 = 0 N/m2 ). Since the tank is kept at its initial position until t = 0·05 s the pressure
in the tank does not change in this interval. Also, one time step after t = 0·05 s (the time step
is set to δt = 0·005 s) the pressure has not changed, since, in the liquid model, the virtual
body force (3) is discretised at the old time level, at which time the tank was not moving. In
the next time step, at t = 0·06 s, the fluid ‘feels’ that the tank is falling with an acceleration
of dq/dt = (0, 0, −10)T m/s2 , whence the pressure in the fluid becomes zero everywhere.
On the right in Figure 1 the motion of the tank is plotted. Most of the quantities, namely
the angular acceleration, velocity and position in all three spatial directions, are equal to zero
during the entire simulation. The linear acceleration (in z-direction) changes from 0 m/s2 to
−10 m/s2 at t = 0·05 s as expected. Also the linear velocity and position, showing linear and
quadratic behaviour, are plotted. Note that, for presentational reasons, the initial position of
the tank is set to zero in this figure.
The previous simulation has been performed with the possibly unstable method (5) and (6)
as well. For this method the acceleration of the coupled system with a liquid/solid mass ratio
of 1·2 is shown in Figure 2. If the liquid/solid mass ratio is further decreased to lower than one,
then the unstable method gives the same results as the stable method. This can be explained
by analysing Equations (5) and (7). In the absence of angular motion these equations reduce
to
  
dq Dv
ms =− ρ − F dV + ms F ,
dt Dt
V

and
  
dq Du
m =− ρ −F dV + ms F ,
dt Dt
V

respectively, (in this case Equation (4) simplifies to Dv/Dt = dq/dt + Du/Dt). These equa-
tions are discretised by evaluating the left-hand side at the new time level n + 1 and the
right-hand side at the old time level n, i.e.,
  n 
dq n+1 Dv
ms =− ρ − F dV + ms F ,
dt Dt
V
30 J. Gerrits and A.E.P. Veldman

and
  
dq n+1 Dun
m =− ρ −F dV + ms F .
dt Dt
V

If both the liquid velocity u in the moving reference frame and the pressure p vanish through-
out the liquid (as is the case in this section), then these equations simplify to
dq n+1 dq n dq n+1
ms = −ml + ml F + ms F and m = ml F + ms F ,
dt dt dt
or
 
dq n+1 ml dq n dq n+1
=− −F +F and = F.
dt ms dt dt
Clearly, the solution for both formulations is dq/dt ≡ F . However, small disturbances in
the solution for dq/dt at time level n are in the first formulation amplified with a factor
ml /ms (this factor is visible in Figure 2, where ml /ms = 1·2), which becomes unstable if
the liquid/solid mass ratio exceeds unity in correspondence with the numerical results. Note
that this instability can not be solved by decreasing the time step in the Runge-Kutta method
for solving system (16). Indeed, the first method is unstable for every time step if ml /ms > 1.

4.2. F ULL TANK

In this section the angular motion of a container with mass ms = 12 kg and moment-of-inertia
tensor I s = diag (1, 2, 4) kg m2 is studied. The container is completely filled with liquid. A
simple calculationshows that the moment-of-inertia tensor of the liquid in this case is equal
√ 3 7 √ 
to I l = 12 ml diag 2 − 2 2, 2 , 2 − 2 kg m . The liquid mass is chosen as ml = 120 kg
1 7 2

(i.e., a liquid/solid mass ratio of 10) and the initial angular velocity of the system is set to
ω (0) = (3, 2, 1)T s−1 . The coupled system settles itself in a state of minimum kinetic energy,
which corresponds to rotation around the axis with maximum moment of inertia (the fluid
is responsible for damping the rotation around the x- and y-axes). Thus, in the steady-state
solution, the system rotates around the z-axis only. The results for this particular simulation
are shown in Figure 3. On the left, the angular velocities around the x-, y- and z-axis are
plotted versus time as well as the theoretically predicted final angular velocity around the z-
axis. On the right, the projection of the angular velocity on the unit sphere is plotted. From
these figures the transition to rotation around a single axis becomes very clear. The numer-
ical simulation predicts a final angular velocity around the stable moment-of-inertia axis in
close correspondence with the theoretical value (which follows from conservation of angular
momentum); the difference is less than half a percent.

4.2.1. Grid-refinement and time-step analysis

The above simulation is repeated on grids consisting of 20×20×20, 40×40×40, 60×60×60
and 80 × 80 × 80 computational cells. The results are shown in Figure 4 (for presentational
reasons only the angular velocity around the x-axis on the four different grids is considered).
On the right in this figure the solutions on subsequent grids are subtracted from each other,
giving an indication of the error in the numerical solution. Clearly, the error decreases when
the grid is refined. Note that the large size of the error can be explained by the phase difference
between the solutions on the various grids.
 Dynamics of Liquid-Filled Spacecraft 31

Figure 3. Left: evolution of the angular velocity [s−1 ] (solid lines) around the three Cartesian axes for a com-
pletely filled tank. The dashed line represents the theoretically predicted final angular velocity around the z-axis.
Right: projection of the angular velocity on the unit sphere.

Figure 4. Left: angular velocity [s−1 ] around x-axis of a full tank for different grids. Right: indication of the error
in the angular velocity given by the difference in the solutions on subsequent grids.

All the simulations in this section have been performed with a time step of δt = 10−3 s. In
Figure 5, the previous simulation is repeated on a grid of 40 × 40 × 40 cells but with different
time steps, namely δt = 5 × 10−3 s, δt = 1 × 10−3 s and δt = 2 × 10−4 s. The right-hand
figure shows that the error in the angular velocity decreases when the time step is reduced.

4.2.2. Liquid/solid mass ratio

Since the grid-refinement study and the time-step analysis revealed that the method is consis-
tent, it is possible to examine how the coupled system that has been examined in this section
reacts to a change in physical parameters, such as the liquid/solid mass ratio. In the previous
simulations the liquid/solid mass ratio was equal to 10; the liquid mass and the solid-body
mass were set to ml = 120 kg and ms = 12 kg, respectively. If the liquid/solid mass ratio is in-
creased (by increasing the density of the fluid) the fluid has more influence on the motion of the
coupled system, whence the system settles itself faster into a steady state. This phenomenon
is illustrated on the left in Figure 6, where the angular velocity of the coupled system is shown
for liquid/solid mass ratios of 0·1, 1, 10 and 100. Note that the angular velocity converges to
different values for different mass ratios, which is caused by a change in the liquid mass and
thus the moment-of-inertia tensor. For presentational reasons only a detail of the time history
32 J. Gerrits and A.E.P. Veldman

Figure 5. Same as Figure 4 but now for different time steps.

Figure 6. Left: evolution of the angular velocity [s−1 ] around z-axis of a full tank for different liquid/solid mass
ratios. Right: comparison of the theoretical and numerical final angular velocity [s−1 ] around z-axis.

of the angular velocity is shown in this figure. In the actual simulation the computation was
continued until the amplitude of the oscillations in the angular velocity around the z-axis was
less than 10−2 s (for the simulation with a mass ratio of 0·1 this happened after approximately
400 seconds simulation time). On the right in Figure 6 the computed angular velocity in the
steady-state solution is compared to theory. Only the absolute value of the angular velocity is
considered, since the coupled system may be rotating in either positive or negative direction in
the body-fixed coordinate system. The difference between simulation and theory is less than
half a percent for all four mass ratios.

4.3. PARTIALLY FILLED TANK

After having studied a completely filled tank, a partially filled tank is discussed in this section.
As an example a flat-spin motion is simulated. The same rectangular container that has been
studied before is rotating around the x-axis with angular velocity ωx = 5 s−1 (initially, both
ωy and ωz are equal to zero). The container is filled with fluid for |y| ≥ c, where c [m]
determines the filling ratio. Since Ixx < Iyy < Izz , rotation around the x-axis and z-axis
is stable. However, because of the viscous liquid, kinetic energy is dissipated. Hence, in
the steady state, the system will be rotating around the z-axis only, which corresponds to
a state of minimum kinetic energy. This phenomenon – transition from rotation around the
axis with minimum moment of inertia to rotation around the axis with maximum moment of
inertia – is known as a flat spin. In the computation the filling ratio was set to 65% and the
 Dynamics of Liquid-Filled Spacecraft 33

Figure 7. Evolution of the angular velocity [s−1 ] around the three Cartesian axes for a partially filled tank.

density of the fluid was chosen such that the liquid/solid mass ratio equals ml /ms = 10. In
Figure 7 the angular velocities around the Cartesian axes are shown as a function of time. This
figure clearly shows that some time (more than ten seconds) is needed before the transition is
triggered. Once rotation around the y- and z-axes is excited, the change of rotation axis takes
place in about ten seconds, although quite some time passes before the angular velocities
around the x- and y-axes decrease to zero.
In Figure 8 a more detailed study of the steady rotation around the x-axis at the beginning
and around the z-axis at the end of the simulation is shown. In this figure snapshots of the free
surface (solid lines) are drawn; on the left the results from the simulation and on the right the
corresponding plots based on theory are shown. The results in two different cross sections,
namely in the xy- and the yz-plane, and at two different time levels, namely at the beginning
and at the end of the simulation, are presented. This figure clearly shows that initially the
system is rotating around the x-axis and finally around the z-axis. Indeed, in the xy-plane the
shape of the free surface changes from a straight line into a circle, while in the yz-plane the
opposite behaviour takes place.
When the system is rotating around one axis only, i.e., at the beginning and at the end of the
simulation, it is easy to compare the pressure contours to theory, since in this case the pressure
contours are concentric circles around the centre of rotation. If the pressure at the free surface
is equal to zero, then the pressure p at a distance R from the centre of rotation is equal to
1
p = ρ ω2 R 2 − r 2 ,
2
where ρ is the density of the fluid, ω is the angular velocity and r is the distance between
the free surface and the centre of rotation. This theory is compared to the results from the
numerical simulation in Figure 8 as well, where the pressure contours are plotted as dashed
lines. The correspondence between simulation and theory is very good.
In Figure 9 three-dimensional snapshots of a flat-spin simulation are shown for which the
filling ratio was decreased to 40%. In the upper-left snapshot the container is rotating steadily
around the axis of minimum moment of inertia. In the second snapshot the beginning of the
flat spin can be seen; liquid starts to flow due to rotation around the y- and z-axes. Some
intermediate snapshots show the violent behaviour of the transition. Finally, in the bottom-
right snapshot, the angular velocities around the x- and y-axes have vanished and the container
is rotating steadily around the axis of maximum moment of inertia.
34 J. Gerrits and A.E.P. Veldman

Figure 8. Snapshots of the free surface (solid lines) and pressure contours [N/m2 ] (dashed lines) for the flat-spin
motion in the xy- and yz-planes at two different time levels, namely at the beginning (top four) and at the end
(bottom four) of the flat spin. On the left, results from the simulation; on the right, results based on theory.

4.4. T HE W ET S ATELLITE M ODEL EXPERIMENT

For studying the dynamics of a liquid-filled spacecraft, the Wet Satellite Model (WSM) ex-
periment was performed in 1992 [14]. Following the launch of the MASER 5 rocket on April
9, 1992, a small spacecraft, the Ejectable Ballistometer (EB), was separated from the MASER
payload. After separation the EB continued on its own trajectory until impact with earth after
370 seconds. Two photographs of the launch of the EB are shown in Figure 10. During its
lifetime the EB performed a flat-spin manoeuvre.
The EB consists of an experiment tank that contains approximately 2·7 kilograms of water,
which is about 50% of the total volume of the tank. The experiment tank is an annular cylinder
 Dynamics of Liquid-Filled Spacecraft 35

Figure 9. Three-dimensional snapshots (at unequal time intervals) of the flat-spin simulation with a filling ratio
of 40%.

Figure 10. Photographs of the separation from MASER of the Ejectable Ballistometer taken during
the Wet Satellite Model experiment (these photographs were taken from http://www.nlr.nl/public/
hosted-sites/mgcd/EXPMNT/A-015.html).

of which the inner and outer radius are equal to 0·125 m and 0·145 m, respectively, and the
height is equal to 0·323 m. The dry mass of the EB is 18·8 kg and its moment-of-inertia tensor
equals diag (0·25, 0·38, 0·37) kg m2 . Since the annular width of the tank is much smaller
than the axial and azimuthal dimensions of the tank, the radial velocity of the liquid can be
neglected with respect to the axial and azimuthal velocities. Also, the radius of curvature in
the radial direction is much smaller than the radius of curvature in the axial and azimuthal
directions. Hence, the free surface will deform in the axial and/or azimuthal directions mainly
and the flow can be assumed to be two-dimensional. This simplification to a two-dimensional
36 J. Gerrits and A.E.P. Veldman

Figure 11. Angular velocities [s−1 ] around x- (left), y- (middle) and z-axes (right) in the Wet Satellite Model
experiment. Comparison of simulation (top) to experiment (bottom).

liquid dynamics problem was an important feature of the WSM experiment, since in 1992
no numerical models were available that could simulate a three-dimensional analogon of the
WSM experiment.
When the EB was released from the MASER payload, its axis of symmetry was aligned
with the direction of the separation push. Hence, all the liquid inside the experiment tank was
initially located in the bottom side of the tank. The initial motion of the EB was induced by
transferring angular momentum from a carried flywheel. The flywheel axis was aligned with
the axis of symmetry, i.e., the axis of minimum moment of inertia. Thus, the initial rotation
of the EB was around this axis, whence a flat-spin transition could be expected. The amount
of angular momentum transferred from the flywheel had been calibrated on the ground, such
that the initial angular velocity around the axis of minimum moment of inertia was known
accurately.
A simulation of the WSM experiment has been performed (actually the very first simula-
tion of the WSM experiment). As in the experiment, the lower half of the annulus was filled
with liquid and an initial angular velocity of 1·2 radians per second around the cylindrical
axis was prescribed. For this simulation a grid consisting of 60 × 60 × 60 cells was used
and the time step, which was automatically adjusted based on the CFL number, varied from
δt = 2·5×10−4 s to δt = 4×10−3 s. The total simulation time was 370 seconds corresponding
to the experiment time. Results from the simulation and the actual experiment are shown in
Figure 11.
Although the geometry of the EB is not very suitable for a method based on Cartesian grids
– a cylindrical grid would be more appropriate – the qualitative agreement between experiment
and simulation is satisfactory. Indeed, the interval in which the angular velocity around the x-
axis decreases from its initial value to zero is of the same order (approximately 200 seconds)
in both the simulation and the experiment. Further, the frequencies of the oscillations in the
angular velocities around the y- and z-axes in the simulations are comparable with those in
the experiment. In particular the higher frequency oscillations in the first half and the lower
frequency oscillations during the second half of the flat spin appear both in the experiment
and the simulation. Also, the amplitude of the oscillations in the simulation is of the same
magnitude as in the experiment. In the experiment, after about 200 seconds, the angular
velocity around the y-axis is approximately constant for almost 40 seconds. This phenomenon
 Dynamics of Liquid-Filled Spacecraft 37

is also present in the simulation, however, not in the angular velocity around the y-axis but
around the z-axis and somewhat earlier than in the experiment.
In the simulation, the angular velocity around the z-axis vanishes in the final stage of the
flat spin. This behaviour is not visible in the experiment, where the EB seems to settle itself
into a steady state where it is rotating both around the y- and z-axes. Since the rotation axis
in the steady state depends on the moment-of-inertia tensor of the coupled system, and thus
on the liquid distribution, apparently, the free-surface configuration that is predicted by the
simulation does not agree with the experiment. Unfortunately, the free-surface configuration
was not monitored during the WSM experiment. Hence, no further conclusions about this
disagreement between simulation and experiment can be drawn.

5. Discussion

In this paper a method has been presented for the numerical simulation of coupled solid-liquid
dynamics. The model consists of two parts. One part solves the liquid motion and another
part takes care of the solid-body motion. Details of the numerical model for the solid-body
dynamics have been presented in this paper.
The governing equations for the solid-body dynamics are naturally written in an inertial
reference frame. Hence, the force and torque on the solid body due to the sloshing liquid
also is written in this reference frame. However, discretisation of these equations results in a
numerical method that is unstable if the liquid mass becomes too large with respect to the mass
of the solid body. This has been demonstrated in a simulation of the free fall of a liquid-filled
container. This stability problem was solved by decomposing the sloshing force and torque in
two contributions.
The first contribution is the force and torque resulting from linear and angular accelerations
of the fluid in the inertial reference frame, hereby neglecting the accelerations in the moving
reference frame. The second contribution is the force and torque due to accelerations of the
fluid in the moving reference frame only. The former contribution is then treated simultane-
ously with the motion of the solid body. Discretisation of this alternative mathematical model
for the solid-body dynamics results in a numerical model that is stable for arbitrary liquid/solid
mass ratios.
An important example of coupled solid-liquid dynamics is the flat-spin motion of a (par-
tially) liquid-filled spacecraft. A flat-spin motion of a three-dimensional rectangular container
partially filled with liquid has been simulated. This simulation showed that the numerical
method can predict a flat spin correctly. A quantitative comparison of a flat spin was made
by simulating the Wet Satellite Model (WSM) experiment. An adequate agreement between
simulation and experiment was found.

Acknowledgements

J.G. wishes to acknowledge financial support from the Netherlands Organisation for Scientific
Research (NWO-SRON). Also, the author wants to thank dr. Jan Vreeburg of the Nationals
Aerospace Laboratory (NLR) in Amsterdam for providing the data of the WSM experiment.
 38 J. Gerrits and A.E.P. Veldman

References

1. W. Rumold, Modeling and simulation of vehicles carrying liquid cargo. Multibody System Dynamics 5 (2001)
351–374.
2. E.F.G. van Daalen, J. Gerrits, G.E. Loots, and A.E.P. Veldman, Free surface anti-roll tank simulations
with a volume of fluid based Navier-Stokes solver. In: T. Miloh and G. Zilman (eds.), Proceedings 15th
International Workshop on Water Waves and Floating Bodies, Caesarea, Israel, February 27 – March 1
(2000).
3. E.F.G. van Daalen, K.M.T. Kleefsman, J. Gerrits, H.R. Luth, and A.E.P. Veldman, Anti-roll tank simu-
lations with a volume of fluid (VOF) based Navier-Stokes solver. In: Twenty-Third Symposium on Naval
Hydrodynamics. Washington D.C.: National Academy Press (2001) pp. 457–473.
4. B. Iannotta, Slosh, rattle and roll. New Scientist, May 27 (2000) 32–35.
5. J. Gerrits, Dynamics of Liquid-Filled Spacecraft. Ph.D. thesis, University of Groningen, The Netherlands
(2001) 112pp.
6. F.J. Kelecy and R.H. Pletcher, The development of a free surface capturing approach for multidimensional
free surface flows in closed containers. J. Comp. Phys. 138 (1997) 939–980.
7. F.H. Harlow and J.E. Welch, Numerical calculation of time-dependent viscous incompressible flow of fluid
with free surface. Phys. Fluids 8 (1965) 2182–2189.
8. D.K. Clarke, M.D. Salas, and H.A. Hassan, Euler calculations for multielement airfoils using Cartesian grids.
AIAA J. 24 (1986) 353–358.
9. K. Morinishi, A finite difference solution of the Euler equations on non-body fitted Cartesian grids. Comp.
Fluids 21 (1992) 331–344.
10. H.S. Udaykumar, H.-C. Kan, W. Shyy, and R. Tran-Son-Tay, Multiphase dynamics in arbitrary geometries
on fixed Cartesian grids. J. Comp. Phys. 137 (1997) 366–405.
11. R.W.C.P. Verstappen and A.E.P. Veldman, Numerical computation of a viscous flow around a circular
cylinder on a Cartesian grid. In: E. Oñate, G. Bugeda, and B. Suárez (eds.), Proceedings ECCOMAS 2000,
Barcelona, Spain, September 11 – 14 (2000) paper 423.
12. C.R. Hirt and B.D. Nichols, Volume of fluid (VOF) method for the dynamics of free boundaries. J. Comp.
Phys. 39 (1981) 201–225.
13. Z.A. Sabeur, J.E. Cohen, J.R. Stephens, and A.E.P. Veldman, Investigation on free-surface flow oscillatory
impact pressures with the Volume-of-Fluid method. In: M.J. Baines (ed.), Numerical Methods for Fluid
Dynamics VI. Oxford: Will Print (1998) pp.493–498.
14. J.P.B. Vreeburg, The Wet Satellite Model Experiment. Technical Report TP 94196 L, NLR, Amsterdam, The
Netherlands (1994) 15pp.

View publication stats