Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
Paul THIRIAT
Introduction
ABSTRACT
This report presents the work done within the framework of my master thesis in the program Infrastructure
Engineering at KTH Royal Institute of Technology, Stockholm. This project has been proposed and sponsored by the
French company Setec TPI, part of the Setec group, located in Paris.
The overall goal of this study is to investigate fluid-structure interaction1 and particularly sloshing in liquidcontaining structures subjected to seismic or other dynamic action. After a brief introduction, the report is
composed of three main chapters. The first one presents and explains fluid-structure interaction equations. Fluidstructure interaction problems obey a general flow equation and several boundary conditions, given some basic
assumptions. The purpose of the two following chapters is to solve the corresponding system of equations. The first
approach proposes an analytical solution: the problem is solved for 2D rectangular tanks. Different models are
considered and compared in order to analyze and describe sloshing phenomenon. Liquid can be decomposed in
two parts: the lower part that moves in unison with the structure is modeled as an impulsive added mass; the
upper part that sloshes is modeled as a convective added mass. Each of these two added mass creates
hydrodynamic pressures and simple formulas are given in order to compute them. The second approach proposes a
numerical solution: the goal is to be able to solve the problem for any kind of geometry. The differential problem is
resolved using a singularity method and Gauss functions. It is stated as a boundary integral equation and solved by
means of the Boundary Element Method. The linear system obtained is then implemented on Matlab. Scripts and
results are presented. Matlab programs are run to solve fluid-structure interaction problems in the case of
rectangular tanks: the results concur with the analytical solution which justifies the numerical solution.
This report gives a good introduction to sloshing phenomenon and gathers several analytical solutions found in the
literature. Besides, it provides a Matlab program able to model effects of sloshing in any liquid-containing
structures.
Keywords: fluid-structure interaction, sloshing, tank, seismic action, hydrodynamic pressure, velocity potential,
boundary element method, Housner, resonant sloshing
Authors note: This version of the thesis, edited to be published on KTH Database, does not include the appendices
and Matlab scripts.
Fluid-structure interaction
TABLE OF CONTENT
1
Introduction ........................................................................................................................................................... 7
Introduction ................................................................................................................................................... 8
2.2
2.2.1
2.2.2
2.3
2.3.1
2.3.2
2.3.3
2.3.4
2.4
3
Conclusion .................................................................................................................................................... 13
Introduction ................................................................................................................................................. 14
3.2
3.2.1
3.2.2
3.3
3.3.1
Free oscillations...................................................................................................................................... 17
3.3.2
3.4
3.4.1
3.4.1.1
3.4.1.2
Hydrodynamic forces..................................................................................................................... 22
3.4.2
Transient and steady state response: Hunt & Priestleys method ......................................................... 24
Introduction
3.4.2.1
3.4.2.1.1
3.4.2.1.2
3.4.2.1.3
Interpretation............................................................................................................................ 31
3.4.2.2
3.4.2.2.2
3.4.3
Conclusion ..................................................................................................................................... 34
3.4.3.1
3.4.3.2
Convective part.............................................................................................................................. 37
3.4.3.3
Model ............................................................................................................................................ 38
3.4.4
Conclusion .............................................................................................................................................. 38
3.5.1
3.5.2
3.5.3
3.5.4
Conclusion .............................................................................................................................................. 48
3.6
4
3.4.2.2.1
3.4.2.3
3.5
Conclusion .................................................................................................................................................... 50
Introduction ................................................................................................................................................. 51
4.2
4.3
4.3.1
Formulation ............................................................................................................................................ 54
4.3.1.1
4.3.1.2
Green function............................................................................................................................... 54
4.3.1.3
Fluid-structure interaction
4.3.2
4.4
4.3.2.1
4.3.2.2
4.4.1
2D tanks.................................................................................................................................................. 57
4.4.1.1
4.4.1.2
4.4.2
4.5
3D tanks.................................................................................................................................................. 62
4.4.2.1
Internal problem............................................................................................................................ 62
4.4.2.2
Conclusion .................................................................................................................................................... 67
Introduction
TABLE OF FIGURES
Figure 1: typical fluid-structure interaction problem................................................................................................... 10
Figure 2: rectangular tank fixed to the ground ............................................................................................................ 15
Figure 3: rectangular tank fixed to the ground ............................................................................................................ 20
Figure 4: Graham & Rodriguez model impulsive mass.............................................................................................. 23
Figure 5: Graham & Rodriguez's method - Impulsive mass ratio ................................................................................. 23
Figure 6: Hunt & Priestley's method - impulsive pressure at the bottom of the tank ................................................. 26
Figure 7: EN 1998-4 Figure A.6 - impulsive pressure at the bottom of the tank ......................................................... 27
Figure 8: Hunt & Priestley's method - Impulsive pressure ........................................................................................... 27
Figure 9: EN 1998-4 Figure A.7 - impulsive pressure ................................................................................................... 28
Figure 10: Hunt & Priestley's method - Convective pressure, sloshing mode 1 .......................................................... 29
Figure 11: Hunt & Priestley's method - Convective pressure, sloshing mode 2 .......................................................... 29
Figure 12: E EN 1998-4 Figure A.7 - convective pressure, sloshing mode 1................................................................. 30
Figure 13: EN 1998-4 Figure A.7 - convective pressure, sloshing mode 2 ................................................................... 30
Figure 14: Hunt & Priestley's method - Impulsive mass ratio for 1, 2, 3 and 10 sloshing modes ................................ 32
Figure 15: Hunt & Priestley's method Convective mass ratio for the sloshing mode 1, 2 and 3 .............................. 33
Figure 16: Hunt & Priestley model impulsive mass and convective mass (2 sloshing modes) ................................. 34
Figure 17: fundamental natural sloshing period for rectangular tanks ....................................................................... 35
Figure 18: Housner's method - lamina fluid theory ..................................................................................................... 36
Figure 19: Housner's method - Impulsive mass ratio, comparison with Graham & Rodriguezs method ................... 36
Figure 20: Housner's method - Convective mass ratio, comparison with Hunt & Priestley's method ........................ 37
Figure 21: Housner's model - Impulsive and convective mass..................................................................................... 38
Figure 22: resonant sloshing ........................................................................................................................................ 41
Figure 23: impulsive added mass for high frequencies as a function of L/h ................................................................ 42
Figure 24: resonant amplification coefficient as a function of L/h .............................................................................. 43
Figure 25: frequency window coefficient of the resonant amplification ..................................................................... 44
Fluid-structure interaction
Introduction
INTRODUCTION
When subjected to external excitation like earthquake, liquid-containing structures are challenging to design due to
sloshing effects. Indeed, fluid-structure interaction is the source of free surface fluctuation and hydrodynamic
pressure loads that can cause unexpected instability or even failure of these structures. Thus, it is paramount to
carry out a thorough investigation about sloshing phenomenon. The main purpose of this report is to characterize
the dynamic response of liquid storage tanks.
In the first chapter of this study, equations that rule fluid-structure interaction are derived. The system to solve can
be summarized to a differential problem with boundary conditions, often called Dirichlet problem. Different
assumptions are discussed in order to obtain a simple model of the fluid behavior. Both analytical and numerical
methods are then proposed in order to solve this problem.
The first approach proposes an analytical solution: the problem is solved for 2D rectangular tanks. Different models
are considered and compared in order to analyze and describe sloshing phenomenon. This study is based on the
results found in the literature dealing with this subject. It is interesting to focus on rectangular tanks because they
represent the major part of practical cases. Besides, an analytical solution often provides a good understanding of
the general aspects ruling the phenomenon. A detailed investigation of resonant sloshing is also conducted in this
part.
The second approach proposes a numerical solution: the goal is to be able to solve the problem for any kind of
geometry. The differential problem is resolved using a singularity method and Gauss functions. It is stated as a
boundary integral equation and solved by means of the Boundary Element Method. Analytical and numerical
solutions are then compared and performance is discussed.
The whole study assumes that the mass of liquid is lumped on the wall, based on rigid wall boundary condition in
the calculation of hydrodynamic pressures. This assumption is widely used in practice, particularly for concrete
structures which have relatively high stiffness.
As a conclusion, several concepts that are not studied in detail in this paper are put into perspective. We can state
for example the development of Tuned Sloshing Dampers or the influence of flexible wall on the dynamic response
of liquid storage tanks.
Fluid-structure interaction
FLUID-STRUCTURE INTERACTION
2.1
INTRODUCTION
The goal of this chapter is to present and explain fluid-structure interaction equations. Fluid-structure interaction is
the interaction of a movable structure with an internal or external fluid.
In the first part, we will focus on the fluid. Different assumptions will be discussed in order to obtain a simple model
of the fluid behavior. We will demonstrate that, under certain conditions, the fluid obeys Laplaces equation. It will
be done by considering the velocity potential function associated to the velocity field.
Then, in a second part, we will analyze several boundary conditions which are likely to occur in fluid-structure
interaction problems. In particular, we will present free surface and fluid-structure interface boundary conditions.
Finally, we will expose the system of equations to solve in order to model fluid-structure interaction.
2.2
FLUID EQUATIONS
In this part, we will derive the fundamental equations of the fluid. It will be done by writing local fluid equations.
Then these equations will be simplified thanks to basic assumptions.
( )
( )
represents body forces acting on the fluid, such as the gravity force for example.
is the
stress tensor.
It is possible to assume that the flow is incompressible. Indeed, we study liquid with low flow velocity. This
property implies that the density is constant and homogeneous. From the continuity equation, we obtain the
well-known criterion of incompressibility:
( )
Fluid-structure interaction
With
and
The frequency range for seismic action is between 0 and 50 rad.s-1 so the incompressible flow assumption is valid.
The stress tensor
as follow [4]:
(
( )
( ))
Eulers equation can be simplified thanks to the perfect fluid model, which implies the following assumptions:
))
( )
that satisfies:
( )
If we now use the property of incompressibility, expressed in the previous part by:
( )
It yields that:
In the case of an irrotational and incompressible flow, the fluid can be represented by its velocity potential
function, which satisfies Laplaces equation.
If we now consider a compressible flow, we can show that the velocity potential function verifies Helmholtz
equation [6]:
10
Fluid-structure interaction
This is the wave equation that expresses the pressure propagation in the fluid (with c the velocity of the sound in
the fluid). When we use Laplaces equation, we consider that this propagation is instantaneous. This assumption is
correct when the velocity c is high before the dimension of the problem. It is the case for the study of water
sloshing in tanks for example.
(
Body forces derive from a potential. For example, gravity field force gives
). Thus, Navier-
After integration of this equation in the fluid domain, we get Bernoullis equation:
( )
Fluid-structure interaction problems always consider small displacement. Thus, it is possible to neglect . It is
( )because
possible to include the time-dependent function k in the velocity potential (we still have
( )
) and we finally obtain linearized Lagranges equation where p is the total pressure (sum of the
This formula is paramount because we want to represent the fluid-structure interaction by the velocity potential
function. So it will allow us to compute the value of the pressure with the knowledge of the velocity potential.
2.3
BOUNDARY CONDITIONS
Now that we have derived the equations that describe the fluid behavior (in particular Laplaces equation), we
need to express boundary conditions in order to solve fluid-structure interaction problems. A typical and general
fluid-structure interaction problem is depicted below (Figure 1).
Structure
fs
s
Fluid
x
Ground
Fluid-structure interaction
Fluid density:
Sound velocity in the fluid: c
Ground density:
This figure can describe, for example, a dam. In this case, it is of great important to be able to model fluid-structure
interaction when a seismic action occurs in order to assess the response of the structure. The velocity of the ground
due to the seismic action is .
( )
, it holds that:
is the velocity of the structures particles at the fluid-structure interface. In the case of a rigid structure, this
velocity is equal to the ground velocity which is known. However, if we have a deformable structure, we have to
add the velocity due to structure deformation and we get:
. In this case, is unknown and we have
a coupled system.
In this formula, we can replace the pressure by its expression as a function of the velocity potential found in the
previous part.
Besides, by definition of the velocity potential function, we have:
11
12
Fluid-structure interaction
It is now possible to express the free surface condition with the velocity potential.
This condition allows us to study waves and surface fluctuation. In particular, we need to know the waves height if
we want to avoid overflow. However, it is sometimes possible to take a simpler condition which does not involve
time. In fact, we consider only high frequency response (for seismic action for example), so it holds that [6]:
and the condition is the same than the one for the
One has to be careful with the distance between the structure and the imaginary boundary. It is recommended to
choose a distance of at least 5h, h being the height of the wet structure.
Fluid-structure interaction
2.4
CONCLUSION
Basically, we can summarize fluid-structure interaction problems to the mathematical resolution of Laplaces
equation with Neumann boundary conditons.
For a perfect fluid, it is convenient to model fluid-structure interaction with its velocity potential. Thus, we have the
following system of equations:
This kind of system of differential equation can be really challenging to solve. That is what is done in detail in the
following chapters of this paper.
The first approached presented is an analytical solution. It is doable for problems with simple geometries only. The
second one is a numerical solution that uses Boundary Elements Method and Gauss functions.
13
Fluid-structure interaction
14
3
3.1
In this chapter, we solve the fluid-structure interaction problem in the simple case of a 2D rectangular tank. The
starting point is the Neumann problem derived in the previous part. Thanks to the simple geometry, it is possible to
solve analytically the sloshing problem. Actually, an analytical solution is possible for rectangular and circular tanks
only. These kinds of tanks represent most practical cases, that is why it is useful to focus on them. Moreover, an
analytical solution will give us a good understanding of the solution we get at the end.
We are going to study fluid-structure interaction when some water, contained in a rectangular tank, is subjected to
an external action (a seismic action for example). During an earthquake, the mass of water contained in the
structure will move because of the displacement of the solid: that is what we call sloshing.
Water sloshing induces hydrodynamic (or fluctuating) pressures on the vertical tank walls because of the horizontal
acceleration of the structure. The goal of this study is to assess the value of these hydrodynamic loads.
The simplest way to compute the horizontal pressure forces when the tank is subjected to a seismic action,
represented by horizontal ground acceleration, is to accelerate the whole mass of water as if it were acting like a
rigid body. Therefore, the hydrodynamic forces will be proportional to the total mass of water. However, the water
does not behave like a rigid body: it is a conservative assumption that is way too detrimental to structure design.
That is why we have to model the fluid flow and the interaction between the fluid and the solid that are both
moving.
The literature presents several methods in order to express the value of the fluctuating pressure [1]. This part
explains and details these different approaches and compares the assumptions and the results obtained. The whole
study assumes that the mass of liquid is lumped on the wall, based on rigid wall boundary condition in the
calculation of hydrodynamic pressures. This assumption is widely used in practice, particularly for concrete
structures which have relatively high stiffness.
In the first part, we will present the problem and give the system of equation to solve. We will explain the different
assumptions made. In the second part, we will study the sloshing phenomenon and the time-history response with
free and forced oscillations. Then, in the third part, we will solve the sloshing problem using several approaches.
The results will be compared to the one we can find in the literature (Eurocode 8.4 for example).
3.2
PROBLEM PRESENTATION
Xs
(0; 0)
ex
h
L/2
With:
, in
is the horizontal displacement of the structure, in meter. The acceleration of gravity is noted g.
We assume that the fluid has the following properties:
Incompressible fluid;
Inviscid flow;
Small displacement;
Non-turbulent regime;
Irrotational flow.
These are the usual assumptions for a perfect fluid. These assumptions are realistic in the case of water contained
in a tank and they will allow us to solve analytically the problem.
We define everywhere in the fluid:
(
(
)
(
)
(
): pressure in the fluid.
15
16
Fluid-structure interaction
In this case, we have two types of boundary conditions: free surface and fluid-structure interface.
With the gravity wave theory, the free surface condition is:
(
( )
( )
)
(
The first two conditions describe the interface with the edges of the tank (x = -L/2 and x = L/2). The last one is the
condition at the bottom (z = h).
The problem to solve is described by the following equations which are respectively Laplaces equation, the free
surface condition and the boundary fluid-solid interface conditions:
(
(
(
{
( )
( )
)
(
3.3
This system is going to be solved with a modal decomposition of the velocity potential. So we are looking for
solution of this type:
(
( ) (
( ) ( )
verifies:
(
So it yields:
(
)( )
)( )
The functions cos, sin, sh and ch are the solutions of this equation. a and b have the same frequency.
The interface conditions gives:
( )
( )
The 2 first conditions give the expression of a:
(
))
17
18
Fluid-structure interaction
))
))
))
( )
)
)
( )
This is the equation of a harmonic oscillator without damping. It implies that is a sinusoidal function whose
frequency
satisfies:
( )
( )
And we find:
( )
)
(
n is sloshing mode of the fluid in the tank. The solution of the free oscillation problem is the sum of the
contribution of every sloshing mode:
(
( )
The shallower the tank is, the smaller the natural sloshing frequency is.
]. This is an orthogonal family for the usual inner product. We are looking for a velocity field of this type:
(
( )
( )
With:
(
So (
))
( )
)
(
)
( )
So
))
The projection of the free surface condition on the n sloshing modes (which are orthogonal) gives n equations:
] ( )
] ( )
]( )
( )
( )
( )
With:
(
)
(
19
20
Fluid-structure interaction
3.4
We still consider the 2D rectangular tank fixed to the ground (Figure 3):
Xs
(0; 0)
ex
h
L/2
Figure 3: rectangular tank fixed to the ground
This part proposes frequency-domain based solutions. Thus, the imposed displacement is a harmonic oscillation
along ex:
( )
The frequency-domain based resolution will not give us access to the time response of the system but we will get
the maximum deformation (and loads) induced by a given elastic ground acceleration response spectrum.
The problem is going to be solved using different method. In the first one, we neglect the transient state. We find
the results proposed by Graham & Rodriguez [1]: the fluid is assimilated to an equivalent impulsive added mass
that moves in unison with the tank. In a second part, when we take into account the transient state, an oscillating
mass is added to the model: this is equivalent to the solution proposed by Hunt & Priestley [1]. The results are then
compared with the one given in the Eurocode 8.4 [2]. And finally, we take a look at Housners method [1], which is
an approached method that gives simple expressions for the added impulsive and oscillating masses. In the
conclusion of this part, the results are compared and the differences are explained.
(
(
And we have:
(
( )
So when we consider only the steady state of the response of the tank due to harmonic oscillations, ( ) is given
by:
( )
( )
(
))
] ( )
))
(
With:
(
(
)
(
And we find:
(
))
] ( )
))
(
( ) ( )
So we it yields:
( )
[ (
))
]
(
The seismic action has a huge importance for high frequencies. Besides, we will prove later that we only have to
take into account the first sloshing modes (so low frequency modes) to have a good estimate of sloshing effects.
Thus, we can consider that:
(
This assumption will be discussed afterward in the part Study of resonant sloshing.
21
22
Fluid-structure interaction
) )
((
)
(
))
We have to multiply by 2 because the pressure is acting on the 2 vertical tank walls.
And we find the hydrodynamic pressure force on the vertical tank wall:
( )
((
] ( )
(
With:
(
((
The value of the horizontal pressure force applied on the vertical tank wall can be written:
( ) ( )
( )
With:
( )
((
]
(
This expression represents the virtual mass of water which exerts a force on the structure when subjected to a
harmonic solicitation of frequency . The force caused by this mass
and proportional to the acceleration ( )
is called the impulsive force.
It proves that it is not necessary to take the whole mass of water contained in the tank (
), but we just
need to consider a certain ratio of this mass. This is due to the interaction between the fluid and the structure. So
finally, it is possible to model the phenomenon of sloshing in the tank by replacing the water by the added
impulsive mass mi.
Therefore, it is possible to see the system as depicted in the Figure 4:
Xs
mi
ex
hi
ez
L
Figure 4: Graham & Rodriguez model impulsive mass
):
The abacus that gives the percentage of impulsive mass to take into account for the calculation is presented below
(Figure 5):
mi/mw
0,7
0,6
0,5
0,4
0,3
0,2
0,1
0
0
= L/h
Figure 5: Graham & Rodriguez's method - Impulsive mass ratio
10
23
24
Fluid-structure interaction
3.4.2 TRANSIENT AND STEADY STATE RESPONSE: HUNT & PRIESTLEYS METHOD
In this part, we do not neglect the transient part of the solution. So we want the exact solutions of the differential
equations ( ) which are given by Duhamels equation:
( )
( )
))
This equation is valid because we can assume that the system is initially at rest. That gives initials conditions:
( )
( )
{
We are still studying the case of harmonic oscillations:
( )
The solution is the sum of a particular and a general solution of the differential equation, which leads to:
( )
(
(
))
))
] ( )
))
(
))
))
(
(
With:
(
)
(
))
] ( )
))
(
))
))
(
(
The hydrodynamic pressure consists of two terms; one proportional to the acceleration of the structure ( ), the
other is proportional to
(
) for each element of the sum (i.e. for each sloshing mode).
Actually, if we compare this result to the one of Graham & Rodriguezs method, we note that the term that is
proportional to ( ) is the pressure factor ( ) on the vertical tank wall (for x= L/2) found previously. This is the
impulsive pressure.
(
( ) ( )
[ (
))
]
(
The only difference with Graham & Rodriguezs method is the presence of an oscillating term. For each mode, we
have an oscillating hydrodynamic pressure with a frequency equal to the natural sloshing frequency of this mode.
This is the convective pressure defined for each sloshing mode:
(
( )
( )
With the convective pressure factor on the vertical tank wall (for x= L/2):
( )
))
(
( )
It is possible to compare this expression with the abacus that we find in the Eurocode. It is however necessary to be
careful with the notations. Indeed, the Eurocode considers a tank with a width of 2L and z = 0 corresponds to the
bottom of the tank (while z = h corresponds to the free surface).
Lets use these notations, that we write L and z (for the half width and the new vertical coordinate).
In the Eurocode, the dimensionless parameter used is:
25
Fluid-structure interaction
Besides, the results presented in the Eurocode consider only high frequencies, so:
(
( )
( )
( )
((
) )
( )as a function of
as a function of
)
(
and :
0,8
0,6
q0(0)
26
0,4
0,2
0
0
= h/L'
Figure 6: Hunt & Priestley's method - impulsive pressure at the bottom of the tank
The abacus given by the Eurocode [2] is presented below (Figure 7):
Figure 7: EN 1998-4 Figure A.6 - impulsive pressure at the bottom of the tank
( )
graph (Figure 8) represents the shape of the impulsive hydrodynamic pressure along the vertical wall of the tank.
0,8
0,6
z'/h
h/L' = 0.1
h/L' = 1
0,4
h/L' = 3
h/L' = 5
0,2
0
0
0,2
0,4
0,6
q0(z'/h)/q0(0)
0,8
27
28
Fluid-structure interaction
The abacus given by the Eurocode [2] is presented below (Figure 9):
( )
( )
( ) is the acceleration response function of a simple oscillator having frequency of the n mode.
( )
((
) )
)
(
as a function of
and :
)
)
We can now draw the hydrodynamic convective pressure for the first and second sloshing mode and for different
value of (Figure 10 and Figure 11). These graphs represent the shape of the convective hydrodynamic pressure
along the vertical wall of the tank.
0,8
h/L' = 0.3
z'/h
0,6
h/L' = 0.5
h/L' = 1
0,4
h/L' = 2
h/L' = 5
0,2
0
0
0,2
0,4
qc1(z'/h)
0,6
0,8
Figure 10: Hunt & Priestley's method - Convective pressure, sloshing mode 1
0,8
0,6
z'/h
h/L' = 0.3
h/L' = 0.5
h/L' = 1
0,4
h/L' = 2
h/L' = 5
0,2
0
0
0,02
0,04
0,06
0,08
qc2(z'/h)
Figure 11: Hunt & Priestley's method - Convective pressure, sloshing mode 2
29
30
Fluid-structure interaction
The abacuses given by the Eurocode [2] are presented below (Figure 12 and Figure 13):
3.4.2.1.3 INTERPRETATION
The different graphs depicted in the previous part give results in accordance with the Eurocode. The work done in
this study gives analytical expressions for the impulsive and convective pressures. These formulas are much more
accurate and easier to use than the abacus proposed by the Eurocode.
We have to multiply by 2 because the pressure is acting on the 2 vertical canal-bridge walls.
And we find:
( )
((
)
( )
(
((
)
(
(
With:
(
((
The transverse force consists of two terms; one proportional to the acceleration of the structure ( ), the other is
proportional to
(
) for each element of the sum (i.e. for each sloshing mode).
( )
With:
( )
((
]
(
31
Fluid-structure interaction
32
):
The abacus that gives the percentage of impulsive mass to take into account for the calculation is presented below
(Figure 14):
1 sloshing
mode
0,7
2 sloshing
modes
mi/mw
0,6
0,5
3 sloshing
modes
0,4
0,3
10 sloshing
modes
0,2
0,1
0
0
10
= L/h
Figure 14: Hunt & Priestley's method - Impulsive mass ratio for 1, 2, 3 and 10 sloshing modes
The value of the added impulsive mass has been plotted for 1, 2, 3 and 10 sloshing modes in order to focus on the
influence of the modes.
The first thing to note is that the more sloshing modes we take into account, the smaller the impulsive mass is. That
is logical because if we do not consider any sloshing at all, the impulsive mass would be equal to the total mass of
water contained in the tank.
The second observation is the rapid convergence of the value of the impulsive mass. The value obtained when we
take into account 3 sloshing modes is almost the same than the one with 10 sloshing modes. Therefore, a good
assessment of the added impulsive mass can be done by considering only a few sloshing modes, 3 for example.
With
( )
)) given
by:
((
)
(
((
( )
for high frequencies (i.e.
):
(
Depicted below is the convective mass ratio for the first 3 sloshing modes (Figure 15).
mc,n/mw
0,7
Sloshing
mode 1
Sloshing
mode 2
Sloshing
mode 3
0,6
0,5
0,4
0,3
0,2
0,1
0
0
5
6
= L/h
10
Figure 15: Hunt & Priestley's method Convective mass ratio for the sloshing mode 1, 2 and 3
33
34
Fluid-structure interaction
Once again, what stands out of this graph is the small contribution of the modes other than the fundamental mode.
A good assessment of the convective force can be obtained by taking into account a few sloshing modes, like the
first and second one for example.
3.4.2.3 CONCLUSION
For high frequencies, Hunt & Priestleys method decomposes the horizontal pressure force into two distinct parts:
An impulsive force, the same than the one proposed by Graham & Rodriguez, which represents the ratio
of water that moves in unison with the tank:
( )
An convective force, which is the sum of the oscillation of every sloshing modes:
( )
The only difference with the previous approach is that the solution of the differential equation is composed of the
general solution and a particular solution. Thats why we have the convective forces, which stand for the transient
state of the response. And the hydrodynamic pressure force can be written as the contribution of these two
forces:
Therefore, it is possible to model the system as depicted in the Figure 16. It is important to remember that this
mechanical analogy is only possible when we consider that the added masses are independent of the oscillation
frequency (i.e. for high frequencies compared to natural sloshing frequencies).
ko,2/2
Xo,2
ko,2/2
mc,2
ex
ez
Xo,1
mc,1
ko,1/2
Xs
ko,1/2
mi
Figure 16: Hunt & Priestley model impulsive mass and convective mass (2 sloshing modes)
The graphs that give the values of the impulsive and convective hydrodynamic pressures prove that the impulsive
mass is mainly acting on the lower part of the tank while the added convective masses are mainly acting on the
upper part of the tank. Indeed, the waves that cause the convective forces are developed on the surface of the
fluid.
Besides, the graphs that give the values of the added impulsive and convective mass show that the shallower the
tank is, the more important the convective forces are (it is the opposite for the impulsive mass). Once again, that
can be explained by the fact that the waves that cause the convective forces are developed on the surface of the
35
fluid: their influence will be even more important than these waves represent a large part of the fluid. This is the
case for shallow tanks.
In a practical case, it will be interesting to compare the impulsive and convective forces. In particular, is it possible
to neglect one of them? When we look at the fundamental natural sloshing frequency of water for different tank
geometry, we get the following graph (Figure 17):
14-16
12-14
21
25
29
10-12
8-10
17
13
1
5
h (m)
6-8
L (m)
4-6
5
91
9
h (m)
5
11
13
15
17
19
21
23
25
27
1
29 L (m)
18-20
16-18
14-16
12-14
10-12
8-10
6-8
4-6
2-4
0-2
The fundamental period of water sloshing is, for most of the cases, several seconds long. At such long periods, the
generated hydrodynamic convective pressures are much smaller than the hydrodynamic impulsive pressures. The
convective forces are generally negligible and can be ignored [9]. It is however necessary to be careful when L is
small (less than 5m) even if the failure risks when an earthquake occurs for such a small tank are not very high.
Besides, we know that the added convective mass ratio is important for shallow tanks, so when L is large in
comparison with h.
Fluid-structure interaction
movement is proportional to the acceleration of the tank. Besides, the upper part of the fluid starts to slosh and
causes vibration forces on the tank walls.
Actually, this description of the fluid movement corresponds to the expression of the pressure forces found in the
previous part with Hunt & Priestleys method. Housner used the lamina fluid theory to calculate the impulsive and
convective hydrodynamic pressure [1]. The liquid is assumed to be incompressible and undergo small
displacements.
L
dx
h
u(x,t)
Housners method is an approximate method that gives simple expressions for the impulsive and convective
masses mi and mc.
percentage of added impulsive mass to take into account for the calculation is presented below (Figure 19). The
results of Housners method are compared with the results of Graham & Rodriguezs method.
mi/mw
36
1
0,9
0,8
0,7
0,6
0,5
0,4
0,3
0,2
0,1
0
Housner
5
= L/h
10
Figure 19: Housner's method - Impulsive mass ratio, comparison with Graham & Rodriguezs method
The result of the impulsive mass given by Housner is very close to the one obtained previously, in particular for
shallow tanks. For < 2, it might be better to use Graham & Rodriguezs method.
Housner method provides also a simple formula that gives the point of application of the resulting pressure force
[1]:
)
. The abacus that gives the
percentage of added convective mass to take into account for the calculation is presented below (Figure 20). The
results of Housners method are compared with the results of Hunt & Priestleys method (1st sloshing mode).
0,7
0,6
Housner
0,5
Hunt &
Priestley
0,4
0,3
0,2
0,1
0
0
5
= L/h
10
Figure 20: Housner's method - Convective mass ratio, comparison with Hunt & Priestley's method
Once again, the approximate approach of Housner gives really good results in comparison with Hunt & Priestleys
method.
The point of application of the resulting pressure force is given by the following formula [1]:
37
38
Fluid-structure interaction
Housner proposes an expression for the value of the natural frequency of the first sloshing mode [1]:
(
We can compare it to what we had with the analytical resolution of the sloshing problem:
(
3.4.3.3 MODEL
Housners method is a simple approach based on lamina fluid theory. It gives results in accordance with the ones
found with the other methods. It takes into account only one sloshing mode for the added convective mass. Thus,
the system can be modeled as follow (Figure 21):
Xo
ex
mc
k0/2
k0/2
ez
hi
Xs
mi
hc
It will be interesting to specify in which case this approximate and simple method can be used.
3.4.4 CONCLUSION
In this part, we have presented 3 different methods that describe the sloshing phenomenon when a tank, fixed to
the ground, is subjected to harmonic oscillations. Nevertheless, every signal can be decomposed into harmonic
signals with the Fourier transform so it can be seen as a general solution.
Graham & Rodriguez and Hunt & Priestleys methods are similar in the approach of the problem: assumptions and
equations are the same. The difference lies in the resolution of the equation of sloshing.
Graham & Rodriguez neglect the transient state of the response of the fluid. Thus, with this model, it was like no
waves were created due to the movement of the tank. The consequence is that the pressure force is proportional
to the ground acceleration with a coefficient called the added impulsive mass. The impulsive mass represents the
quantity of water that is moving in unison with the structure.
Hunt & Priestley take into account the transient state. The waves formed are oscillating in the tank without
damping (so the transient state is not really transient), which leads to a convective force acting on the tank walls.
This convective force is the sum of n added convective mass (n being the number of sloshing modes) that are
oscillating like a simple oscillator having the frequency of the sloshing mode considered. The steady state gives the
same impulsive force than in Graham & Rodriguezs method.
Housner proposes an approximate method that gives simples formulas to assess both of impulsive and convective
forces. Actually, Housner takes into account only one sloshing mode for the computation of the added convective
mass. The results are very close to the ones found with Hunt & Priestleys method.
We observe in the abacus that gives the added impulsive and convective mass ratio that the shallower the tank is,
the more important the oscillating part is. Indeed, the waves are formed on the surface and the oscillating force
has an influence on the upper part of the fluid only. It will be interesting to compare the contribution of the
impulsive and the oscillating force in order to conclude about the validity of Graham & Rodriguezs method.
The table below presents the 3 different approaches with their hypothesis, results, limitations and model.
Method
Hypothesis
Results
Limits
Housner
Perfect fluid described by
the lamina fluid theory
Approximate method: fluid
movement decomposed in
two part
Impulsive and convective
part of the solution, 1st
sloshing mode only for the
convective part
Solution for high
frequencies only
Results independent of
frequency: no analysis for low
frequencies possible
Results less accurate for
L/h < 2
Diagram
In most of the case, it is possible to neglect the convective forces. Indeed, for shallow tanks (with L large), the
fundamental natural period of water sloshing is several seconds long. At such long periods, the generated
hydrodynamic convective pressures are much smaller than the hydrodynamic impulsive pressures. The convective
forces are generally negligible and can be ignored.
The key assumptions of this part are that there is no fluid damping and that external oscillations are harmonic.
It is also important to note that the walls are supposed to be rigid in this study. Their deformation might induce
higher impulsive force.
39
40
Fluid-structure interaction
3.5
The results presented in the previous parts are only valid for high frequency. Indeed, formulas given for the added
mass ratio consider that (
In this part, we analyze the response of a 2D rectangular tank subjected to a given seismic action. We compare the
results of a frequency domain analysis with the high frequency assumption.
( )
( )
is the liquid damping factor [5]. For water, we can consider that, for the first sloshing mode:
When there is a damping factor, even very low, it is possible to neglect the transient state. Thus, the solution of this
equation can be found easily (with a complex resolution for example) for a harmonic oscillation. We find:
) )
Therefore, the ratio impulsive mass over total mass of water becomes:
( )
( )
) )
((
)
(
) )
With the natural sloshing frequency for the mode n given by:
((
10
-200
L=1
-400
L=5
L = 20
-600
-800
-1000
Oscillating frequency (rad/s)
Figure 22: resonant sloshing
The table below gives the first and second natural sloshing frequencies:
L (m)
1
5
20
0 (rad.s-1)
5.6
2.5
1.0
1 (rad.s-1)
9.7
4.3
2.2
: oscillating frequency is very small and the water does not slosh. Thus, the whole water contained
in the tank is moving in unison with the structure and the added mass ratio is equal to 100%;
: oscillating frequency is equal to the fundamental sloshing frequency of the tank. Resonance
induces a huge fall of the added mass: this is the resonant amplification;
: oscillating frequency is equal to the second natural sloshing frequency. Resonance effect is much
smaller than for the first sloshing mode;
: oscillating frequency becomes high and the added mass ratio reaches quickly an asymptotic
value. Its the impulsive added mass found in the previous part with the assumption of high frequencies.
Two main conclusions can be extracted from these graphs for the different geometries. Firstly, we note that the
shallower the tank is, the higher the fundamental sloshing frequency is. In the same time, amplification decreases
when natural frequency increase.
The different methods presented in this document assume that it is possible to take the impulsive added mass for
high oscillating frequencies. It is of great importance to justify this assumption. It can be done by studying the
resonant amplification of the added impulsive mass. Indeed, this graph shows that resonant water in the tank can
induce a huge impulsive mass for certain geometry. The force associated with this impulsive added mass would be
huge as well. What are the resonance effects when our structure is subjected to a seismic action?
41
Fluid-structure interaction
The fundamental sloshing mode is predominant. The other ones are cushioned and do not have any significant
effect on the impulsive mass value. Thus, it is possible to consider the first sloshing mode only for the calculation.
The added mass ratio becomes:
( )
( )
( )
(
( ) for
) )
( )
This is the value of the asymptotic added mass by taking into account only one sloshing mode (Figure 23). The
resonant amplification has to be compared with this expression.
42
1
0,9
0,8
0,7
0,6
0,5
0,4
0,3
0,2
0,1
0
0
10
= L/h
Figure 23: impulsive added mass for high frequencies as a function of L/h
( ) for
( )
This is the resonant amplification value. It is necessary to be very careful with the magnitude found. Indeed, the
model can be put in into question. In particular, sloshing equations established concerned a perfect inviscid fluid. In
the resonant sloshing study, we have added an artificial damping coefficient in the final equation with a certain
value that is purely theoretical. Somehow, using this formula, we get the following graph (Figure 24):
amplification
Resonant amplification
90
80
70
60
50
40
30
20
10
0
0
10
= L/h
It is of course very unlikely to have a resonant amplification up to 80. It would mean that resonant sloshing could
induce hydrodynamic forces 80 times larger than if we considered water as a rigid body. The useful information
that can be extracted from this graph is that resonant amplification increase with .
In order to now if it is reasonable to take the asymptotic value of the added mass ratio, it is important to have a
frequency domain based description of seismic action. If the typical frequency window of earthquake corresponds
to sloshing resonance, we will have to study this phenomenon very carefully.
So the first question to ask is: when do we have amplification due to resonance? What is the frequency window of
resonant sloshing? In order to answer this question, we have to find the frequency range for which:
( )
( )
With:
( )
)
(
( )
)
]
This coefficient k is the resonant amplification peak width (Figure 25). It characterizes the frequency window
where there is resonant amplification.
43
Fluid-structure interaction
1,5
1
0,5
0
0
10
= L/h
Figure 25: frequency window coefficient of the resonant amplification
We note that ( ) is close to 1: it means that the amplification is significant in a thin frequency window.
Resonance will be significant when the structure is subjected to a seismic action if oscillating frequencies of the
earthquake are close to the tank fundamental sloshing frequency.
This fundamental sloshing frequency is given on the graph below for different geometries (Figure 26).
6
5
4
w0 (rad.s-1)
44
L=1
L=2
L=5
L = 10
L = 20
3
2
1
0
0
5
= L/h
10
It means that for geometry with high water loads (i.e. and L large), even if we have a big amplification, the natural
sloshing frequency is very low. Thus, in order to know if earthquake can cause resonant sloshing, we have to
analyze power spectral densities of earthquake and compare them to tanks natural sloshing frequencies. If the
frequency window of the earthquake has a higher range than natural sloshing frequencies, it is possible to use the
asymptotic value of the added mass.
The graph depicted below (Figure 27) sums up important features of resonant amplification.
Added mass
Resonant amplification
45
46
Fluid-structure interaction
Lets now analyze the response of a tank filled with water subjected to this earthquake (Figure 30).
Xs
(0; 0)
ex
h
L/2
Figure 30: rectangular tank fixed to the ground
In the case of a tank fixed to the ground, it is very simple to get the harmonic response of the structure. The inertia
force of water has the following expression:
( )
With
( )
Figure 31: amplitude spectrum of the earthquake (blue) and the transfer function (green)
47
48
Fluid-structure interaction
Knowing the DFT, it is now possible to compute this inertia force [5]:
( )
)
]
Depicted in the next page are the graphs that represent the inertia force for 2 cases:
( ).
The 3 graphs represent respectively the case for L = 20, 5 and 1 m (Figure 32).
As expected, the 2 responses are almost identical. The maximum forces are equal which justifies the fact that it is
possible to take the asymptotic value of ( ) in the calculations. This is true for seismic action, but for other
external loads such as wind for example, it is not always possible to neglect resonant sloshing. In some other cases,
it is even possible to use the resonant sloshing with Tuned Sloshing Dampers.
3.5.4 CONCLUSION
It is not relevant to take into account resonant sloshing when we study dynamic response of structure subjected to
seismic actions. It is consequently possible to use the formula presented in this chapter.
However, resonant sloshing can have positive effects in other case, as it is the case for wind-induced vibration that
can be reduced thanks to Tuned Sloshing Dampers.
49
50
Fluid-structure interaction
3.6
CONCLUSION
In this part, we have thoroughly studied water sloshing in 2D rectangular tanks. We have used the equations of
fluid-structure interaction derived in the first part. In this particular case, it is possible to solve the corresponding
Dirichlet problem analytically. Actually, it is only possible to do so for rectangular and circular tanks.
Different models are studied and compared. In the end, we can propose a quite simple model that describes water
sloshing in 2D rectangular tank. When the tank is subjected to a horizontal acceleration, the lower part of the water
moves together with the tank walls: the pressure force induced by this movement is proportional to the
acceleration of the tank. Besides, the upper part of the fluid starts to slosh and causes vibration forces on the tank
walls. That is why we can decompose hydrodynamic pressure forces in two parts: an impulsive and a
convective force that corresponds respectively to the water that moves in unison with the tank and the water
that is sloshing. These two forces can be computed with the knowledge of their associated added mass. The models
assume that the mass of liquid is lumped on the wall based on rigid wall boundary condition in the calculation of
hydrodynamic pressures.
We have also demonstrated that effects of resonant amplification of sloshing water are not relevant in the case of
seismic action.
4.1
INTRODUCTION
Analytical solutions presented earlier are really useful in order to understand phenomenon that occur in sloshing
problems. They provide good results and do not require any complicated theory: most of the time, the simple
Housner formula is sufficient to estimate the impulsive added mass. However, it only allows solving problems in a
very limited number of cases. Indeed, analytical solutions are only possible for simple geometry, namely
rectangular and circular tanks. Even if these 2 cases encompass a great majority of practical cases, it is possible to
encounter more specific geometry.
The purpose of a numerical method is to solve sloshing problems for every kind of structure.
Lets take the example of a tank as depicted below (Figure 33):
y
x
Figure 33: example of non-rectangular 2D tank
It can move vertically and horizontally along z and y and it can rotate around x. The goal of numerical methods is to
obtain the added mass matrix. In other words, for each degree of freedom, we want to know the effective mass of
water that we have to take into account to assess the loads.
Then, the hydrodynamic force torsor will be expressed as follow:
[ ] ( )
51
52
Fluid-structure interaction
4.2
GENERAL METHOD
We assume that the fluid is incompressible and we consider only the case of high frequency oscillations. In this
case, the flow is characterized by its velocity potential which obeys the following equation:
( )
{
The notations are explained in the graph below (Figure 34):
fs
n
f
n
We define
The system being linear, it is possible to solve elementary problems corresponding to each degree of freedom. For
a 2D problem as presented in the introduction, we will resolve the 3 following problems, for i = 1, 2 and 3.
(
{
{
The structure rotates around point O.
Once we have solved these elementary problems, we have access to the total velocity potential. In this case:
And then, using the relation between the hydrodynamic pressure and the velocity potential, we can express the
value of this pressure acting on the solid:
(
And finally, we get the forces and moment caused by these hydrodynamic pressures:
{
That leads to the expected formulation:
[ ] ( )
In order to solve the given system of equations, we write Laplace equation with an integral form:
( )
Where
so that
( )
(which are
on the fluid-solid
The numerical method proposes a frequency domain based solution. Consequently, we will only get the impulsive
added mass in our results. If we wanted to compute convective added masses, it would be necessary to put an
alternative input signal, such as a harmonic signal equal to 0 before t = 0 (so that there is a transient state). Then,
we should compute the Fourier transform of this signal to analyze the effects of the transient state on the dynamic
response of our system.
53
Fluid-structure interaction
54
4.3
SINGULARITY METHOD
The singularity method is a way to solve differential problems. The trick is to state the problem as a boundary
integral equation, whose support lies on the bounded boundary. The use of Greens integral theorem allows us to
solve the radiation problem by only calculating the boundary values. Thus, we just need to discretize the contour of
the fluid domain. This method, used to solve a boundary equation by means of finite element method is known as
the Boundary Element Method (BEM).
4.3.1 FORMULATION
4.3.1.1 GREEN INDENTITIES
Lets apply the first Green identity to W; we get:
If we permute
and
And the third Green identity gives, by subtracting the 2 equations established above [6]:
( )}
( )
( )
( )}
with
and
( )
( )
with .
. Using
( )
( ( )
( )
)
( )
( )
( )
The potential flow in P can be seen as a mixed distribution of sources and doublets on the surface S. So the
potential flow in P is the superposition of potentials created by sources and doublets with the following surface
density:
( )
( )
Once its done, we can find the potential flow on the surface S with the following equation:
( )
We know
on
and
on
( ( )
( )
)
( )
( )
This equation is to be solved by discretization of the surface S. We assume that on each element
[
] of S,
( ) and ( ) are constant and equal to their mean value and . Integrals become sums of integrals on each
element. The equation to solve is now, for each element i:
( )
( )
( )
( )
The next step is to find the Greens function associated to our problem, as described in the previous part. Greens
functions for 2D and 3D problems are computed afterward. Once we have the desired function, we have to solve
the linear equation with a computer. The tricky part is to compute the matrices coefficients.
55
56
Fluid-structure interaction
Once we have found the value of the potential flow on the surface S of our domain, it is possible to compute it
everywhere on
with the following formula:
( ( )
( )
( )
( )
( )
( )
, we show that:
( ( ))
, we show that:
4.4
The numerical problem is solved using Matlab. Scripts and user guides are given in Appendix 1 and 2. In this part,
we present results for several examples, including 2D and 3D problems. We choose rectangular geometries in our
example in order to compare numerical results with analytical results found in the previous chapter.
4.4.1 2D TANKS
Matlab function slosh2D.m analyzes sloshing effects for 2D tanks. For any 2D tank, defined by its walls and its
free surface, slosh2D returns the value of the hydrodynamic pressure acting on the walls and the total added mass
for horizontal and vertical translations and rotation.
We have respectively the added mass for a vertical and horizontal translation and for a rotation.
When we apply the analytical formula for the horizontal added mass (see chapter 2), we find m = 0.73 t, which is
the same value than the one given by the program.
We can compare the added mass ratio for rectangular tanks with other geometries. The following graph shows that
there is a perfect match between the numerical and analytical solution (Figure 35).
1
0,9
0,8
0,7
0,6
0,5
0,4
0,3
0,2
0,1
0
Numerical
solution
Analytical
solution
L/h
Figure 35: numerical and analytical added mass for 2D rectangular tanks
57
58
Fluid-structure interaction
Depicted below are the pressure distributions acting on the vertical wall for horizontal and vertical actions for the
rectangular 2D tank with L = h =1 m (Figure 36 and Figure 37).
Lets now compare the numerical horizontal pressure distribution given by slosh2D with the analytical pressure
factor given in chapter 2 (Figure 38).
0,5
0,4
Numerical
solution
0,3
Analytical
solution
0,2
0,1
0
0
0,2
0,4
Depth (m)
0,6
0,8
The 2 curves are really close: the numerical model gives excellent results in this case. The only significant difference
is found close to the toe of the wall where there seems to be a numerical singularity. It can be explained by the
numerical approximation of Gauss integral for each finite element. This numerical singularity disappears when the
integral is computed more precisely, but it is not worth it because it is not a time-efficient computation. If needed,
it possible to adapt the last value.
The graph below is the added mass for a horizontal oscillation as a function of the frequency given by the program
(Figure 39). These results are really useful if we want to study resonant sloshing.
59
Fluid-structure interaction
10
8
6
4
Added mass (t)
Numerical
solution
Analytical
solution
0
-2
0,1
0,6
1,1
1,6
2,1
2,6
3,1
3,6
4,1
4,6
-4
-6
-8
-10
Frequency (Hz)
Figure 40: resonant sloshing for numerical and analytical solutions
Once again, the two curves are really close, especially for the first sloshing modes which are the most important.
For the 3 first sloshing modes, the natural sloshing frequencies are really close. After that, there is a discrepancy
between analytical and numerical values.
As a conclusion of this part, lets say a few words about program performance. On the graphs depicted below, the
blue curve shows the ratio between the added mass provided by the program and the expected analytical
asymptotic added mass (Figure 41). The red curve represents the time needed by the program to compute the
value of the added mass.
1
0,9
0,8
0,7
0,6
0,5
0,4
0,3
0,2
0,1
0
0,1
0,09
0,08
0,07
0,06
0,05
0,04
0,03
0,02
0,01
0
0
40
80
120
160
200
240
280
Number of elements
320
360
400
Convergence test
Convergence
60
At least 100 elements are necessary in order to obtain satisfying results. After that, it is not worth it to increase this
number if we compare the gain of accuracy with the computation time. However, in any case, the program needs
less than 0.1 s to run, so time-efficiency is not a big issue here.
-2
meters
-4
-6
-8
-10
-12
-15
-10
-5
10
meters
The numerical solution is much more powerful because it allows us to compute the impulsive pressure in a
rectangular tank regardless of the degree of filling. Besides, we do not need to apply complicated formulas such as
the ones given in the Eurocode.
61
Fluid-structure interaction
62
3
Numerical
solutions
2,5
2
Eurocode
1,5
1
0,5
0
0
0,2
0,4
0,6
0,8
1,2
1,4
1,6
angle (rad)
Figure 43: pressure factor for 2D rectangular tank - Eurocode and numerical solution
4.4.2 3D TANKS
4.4.2.1 INTERNAL PROBLEM
Matlab function slosh3D.m analyzes sloshing effects for 3D tanks. For any 3D tank, defined by its walls and its
free surface, slosh3D returns the value of the hydrodynamic pressure acting on the walls and the total added mass
for translations.
It is interesting to use the program for a 3D rectangular tank. Indeed, we know the expected results since we have
an analytical solution of the sloshing problem in this case. This is an effective way to test the program.
The added mass matrix for the 3 translations (respectively x, y and z see figure below) is given by the program (in
t):
[ ]
These results are exactly the same than the ones given by the analytical formula for a rectangular tank.
The program gives also access to the hydrodynamic pressures acting on the tank walls. The following figures display
these pressure distributions (Figure 44, Figure 45 and Figure 46).
Figure 44: pressure acting on the walls for horizontal oscillation along x
Figure 45: pressure acting on the walls for horizontal oscillation along y
63
64
Fluid-structure interaction
And we can study resonant sloshing with the graph of the added mass as a function of frequency (Figure 47 and
Figure 48).
As a conclusion of this part, lets say a few words about program performance. On the graphs depicted below, the
blue curve shows the ratio between the added mass provided by the program and the expected analytical
asymptotic added mass (Figure 49). The red curve represents the time needed by the program to compute the
value of the added mass.
Convergence test
10
0,9
0,8
0,7
0,6
0,5
0,4
0,3
0,2
0,1
0
0
96
192
288
384
480
576
Number of elements
672
768
864
Convergence
65
66
Fluid-structure interaction
In this case, even a very low number of elements are sufficient to get decent results. It is particularly important to
adapt the meshing since the computation time increases quickly to several seconds long.
The difference of convergence between the 2D and 3D programs can be explained by the numerical integration of
Gausss functions, required to compute the influence coefficients. Indeed, in the program slosh2D, integral values
are estimated roughly by taking the mean value of the 2 nodes of the corresponding element. In the program
slosh3D, integral values are computed exactly thanks to the formula derived in Appendix 3.
L
Figure 50: external problem geometry
One has to be careful with the distance between the structure and the imaginary boundary. It is recommended to
choose a distance of at least L = 5h, h being the height of the wet structure.
The program gives the pressure factor exerted on the dam. In the literature [15], hydrodynamics pressure factor on
a dam subjected to seismic action is given by Westergaards formula:
( )
On the graph depicted below, numerical results are compared with Westergaards formula that is widely used
(Figure 51). The height of the dam is taken equal to 1 m.
Values are almost identical in the upper part. However, a discrepancy appears when we get closer to the ground.
The shape of the numerical solution seems to be more accurate because we recognize the usual curve of pressure
distribution.
z (m)
0,6
Westergaard's
formula
0,5
0,4
Numerical
solution
0,3
0,2
0,1
0
0
0,2
0,4
0,6
0,8
4.5
CONCLUSION
In this chapter, we have presented a way to solve numerically fluid-structure interaction. The resolution of Laplace
equation is done with an integral method. The Boundary Element Method is also used to obtain a linear system
which gives the velocity potential. Gauss functions are proposed to solve 2D and 3D problems.
The method is implemented in Matlab. The efficiency of the program has been studied by comparing the results
with the one found analytically in the previous chapter. Both programs are very effective. The only difficulty can be
the meshing of a 3D tank. It can require the use of other software.
67
Fluid-structure interaction
68
GENERAL CONCLUSION
In this study, different approaches are presented in order to analyze fluid-structure interaction which obeys a
Neumann problem. The equations have been derived in the first part, based on the assumption of a perfect fluid.
Two methods are presented: an analytical and a numerical solution.
We have presented 3 different ways to solve analytically the sloshing problem when a 2D rectangular tank, fixed to
the ground, is subjected to harmonic oscillations. Graham & Rodriguez and Hunt & Priestleys methods are similar
in the approach of the problem: assumptions and equations are the same. The difference lies in the resolution of
the equation of sloshing. Graham & Rodriguez neglect the transient state of the response of the fluid. Thus, with
this model, it was like no waves were created due to the movement of the tank. The consequence is that the
pressure force is proportional to the ground acceleration with a coefficient called the added impulsive mass. The
impulsive mass represents the quantity of water that is moving in unison with the structure. Hunt & Priestley take
into account the transient state. The waves formed are oscillating in the tank without damping (so the transient
state is not really transient), which leads to a convective force acting on the tank walls. This convective force is
the sum of n added convective mass (n being the number of sloshing modes) that are oscillating like a simple
oscillator having the frequency of the sloshing mode considered. The steady state gives the same impulsive force
than in Graham & Rodriguezs method. Housner proposes an approximate method that gives simples formulas to
assess both of impulsive and convective forces. Actually, Housner takes into account only one sloshing mode for
the computation of the added convective mass. The results are very close to the ones found with Hunt & Priestleys
method. We have also demonstrated that effects of resonant amplification of sloshing water are not relevant in the
case of seismic action.
Analytical solutions are really useful in order to understand phenomenon occurring in sloshing problems. They
provide good results and do not require any complicated theory: most of the time, the simple Housner formula is
sufficient to estimate the impulsive added mass. However, it only allows solving problems in a very limited number
of cases. Indeed, analytical solutions are only possible for simple geometry, namely rectangular and circular tanks.
Even if these 2 cases encompass a great majority of practical cases, it is possible to encounter more specific
geometry. That is why we propose a numerical solution that allows us to solve the problem for any kind of
geometry. The resolution of Laplace equation is done with an integral method. The Boundary Element Method is
also used to obtain a linear system which gives the velocity potential. Gauss functions are proposed to solve 2D and
3D problems. The program is implemented on MATLAB. The BEM is very efficient and provide accurate results.
The main point that still needs thorough investigation is fluid-structure coupling. All the work done in this study is
based on the assumption that structures have rigid walls. This is not always the case (in particular for steel tanks).
Structure deformations can influence on hydrodynamic pressure, and hydrodynamic pressure can then affect
structure deformations It is thus necessary to couple the program proposed in this paper with structure software.
This coupling would require FEM-BEM coupling. In this case, resonant sloshing can become important.
General conclusion
REFERENCES
[1] DAVIDOVICI, V. and HADDADI, A. Seismic calculation of tanks (in French), Institut technique du btiment
et des travaux publics, 1982.
This document gives the results of the methods developed by Graham & Rodriguez, Hunt & Priestley and Housner.
It also proposes practical calculations. Despite several mistakes, it is a useful document that summarizes the
different known method at the time.
[2] EUROCODE 8 Design of structures for earthquake resistance part 4: silos, tanks and pipelines, European
Committee for Standardization, 1988.
The Eurocode 8.4 proposes an approach to compute the seismic response of tanks with graphs and abacus.
[3] DE LANGRE, E. Fluids and Solids (in French), Les ditions de lEcole Polytechnique, 2006.
This book deals with the fluid-solid interactions and explains the phenomenon with equations. It also treats the
case of an oscillating tank filled with water.
[4] HUERRE, P. Fluid mechanics (in French), Les ditions de lEcole Polytechnique, 1998.
This book presents assumptions and equations dealing with fluid mechanics problems.
[5] PECKER, A Structural dynamic of infrastructures (in French), ENPC, 2007.
This document deals with the effects of seismic actions and gives tools to study the dynamic response of structures
(modal decomposition in particular).
[6] PESEUX, B. Introduction to fluid-structure coupling (in French), Ecole Centrale Nantes, 2010.
This handout presents Greens equations and the singularity method in order to solve fluid-structure problems.
[7] HEIN HOERNIG, R.O. Greens functions and integral equations for the Laplace and Helmholtz operators in
impedance half-spaces, Ecole Polytechnique, 2010.
This thesis presents the mathematical tools to compute Greens function for Laplace equation. This is useful to
solve the problem with the singularity method.
[8] SEGHIR, A. Numerical modeling of seismic response of structures with ground-structure and fluid-structure
interaction (in French), Universit Paris-Est Marne-la-Vall, 2011.
This thesis presents equations and numerical solutions for fluid-structure interaction problems. In particular, the
author gives boundary conditions in order to solve the external problem of radiation.
[9] US Army Corps of Engineers Time-History Dynamic Analysis of Concrete Hydraulic Structures, Department
of the Army, 2003.
This manual describes procedures for the linear-elastic time-history dynamic analysis for seismic design and
evaluation of concrete hydraulic structures.
69
70
Fluid-structure interaction
[10] SCORLETTI, G. Signal processing (in French), Ecole Centrale Lyon, 2011.
This handout deals with signal processing, in particular presents how to program the Fast Fourier Transform on
Matlab.
[11] Pacific Earthquake Engineering Research Center PEER Ground Motion Database:
<http://peer.berkeley.edu/peer_ground_motion_database> (valid link February 20, 2013).
This website provides numerous data about earthquake, such as time-history recording.
[12] LEBON, J-D. Housners method for conical tanks in seismic area (in French), Annales de linstitut
technique du btiment et des travaux publics N491, February 1991.
This article shows how to use Hounsers method for certain kind of tanks.
[13] GEORGAKIS, C.T., KOSS H.H. and DE TOFFOL W. Tuned Liquid Dampers for the New European Court of
Justice, Luxembourg, Structural Engineering International, 4/2005.
Presentation and experimental verification of the performance of TLD put in place in the New European Court of
Justice.
[14] FUJII, K., TAMURA, Y., SATO, T. and WAKAHARA T. Wind-Induced vibration of tower and practical
applications of Tuned Sloshing Dampers, Journal of Wind Engineering and Industrial Aerodynamics, 33 (1990)
263-272.
Presentation and experimental verification of the performance of TLD put in place in the Nagasaki Airport Tower
and Yokohama Marine Tower.
[15] http://www.physicsforums.com/showthread.php?t=129837 (valid link February 20, 2013).
Assumptions for the validity of the incompressible flow.
[16] Geodynamics and structures Dynamic risks for river marine structures. Fascicle n4: Seismic action (in
French), Centre dtudes techniques maritimes et fluviales ER QG 94.05, August 1995.