���������������������������������������������

�����������������������������������������������

3D dynamic soil–fluid–structure interaction analysis in the time domain

O. von Estorff & C. Hagen
Modelling and Computation, Hamburg University of Technology, Hamburg, Germany

ABSTRACT: A procedure for the coupling of an arbitrary number of three-dimensional subregions into one
single computational model is presented. Each subregion may be modelled using the FEM or the BEM in the
time domain, and the coupling can be formulated in the conventional direct way or by means of iterations.
The approach allows to take into account nonlinearities, unbounded regions, and incident earthquake waves in
soil–fluid–structure interaction analyses. Numerical results are presented and compared to analytical solutions,
where available.

1 INTRODUCTION and subjected to some typical transient excitations, is

investigated.
For the investigation of complex dynamic problems,
such as offshore structures or dam–reservoir–soil sys- 2 FEM AND BEM IN THE TIME DOMAIN
tems under earthquake excitation, often various sub-
systems with very different characteristics have to The formulation of the time domain FEM for a solid
be taken into account within a single computational medium (Bathe 1996) yields for each time step m an
model. This can include subsystems, where nonlinear- equation of the form
ities may occur, soil or water regions, or unbounded m
domains. In such cases, it usually is advantageous to K mu = mf + mr , (1)
split the domain under consideration into several sub- where m K = m K(mu ) is the (effective tangent) stiff-
domains. Afterwards, each subdomain can be treated ness matrix, mu is the vector of nodal displacements,
with best suited methodologies, such as the Finite m
f is the vector of applied nodal forces, and mr is a
Element Method (FEM) for regions, where inhomo- vector that comprises influences from inertial forces,
geneities exist or nonlinear behaviour is expected to from nonlinearities and from constrained displace-
appear, and the Boundary Element Method (BEM) for ments.
unbounded water or soil subregions. The time domain BEM for an elastic solid and a
In the present contribution, a time domain proce- compressible fluid domain (Domı́nguez 1993, Bonnet
dure for the coupling of an arbitrary number of three- 1995) yields algebraic equations of the form
dimensional subdomains within a single global model
is proposed. Each subdomain may be treated either by
1
T mu = 1 U mt + mh + mu inc (2)
the FEM or by the BEM. The BEM formulation al- and
lows the inclusion of incident wave fields in a rather
accurate manner, making the approach especially in-
1
Q mp = 1 P mq + mh + mpinc , (3)
teresting for earthquake engineering problems. respectively, where 1 T, 1 U and 1 Q, 1 P are the so-
A direct coupling, as well as an iterative coupling called influence matrices for the first time step, mu
formulation are discussed and applied to some repre- and mt are the vectors of nodal displacements and
sentative examples. The validity of the formulation is tractions, mp and mq are the vectors of nodal pres-
shown by means of two simple problems, for which sures and pressure derivatives, while mh contains the
analytical solutions exist. As a more realistic exam- (known) influence of previous time steps, and mu inc
ple, the influence of soil–fluid–structure interaction and mp inc represent incident displacement and pres-
on a water retaining structure, resting on a soft ground sure wave fields, respectively.

1261
3 COUPLING OF THE SUBDOMAINS Jr., von Estorff, and Mansur 2004), where the sys-
tems of equations of the subdomains are solved in-
Each subdomain in the global model is treated using dependently of each other. The interaction effects are
either the FEM or the BEM, ie., the respective equa- taken into account as boundary conditions which are
tion (1), (2) or (3) is used. imposed on the coupling interfaces. Within each time
The coupling is performed by imposing the appro- step, an iteration is performed in order to enforce sat-
priate compatibility and equilibrium conditions at the isfaction of the coupling conditions. The convergence
interface boundaries. In the proposed procedure, these characteristics can be improved by applying a relax-
conditions are formulated as compatibility of the dis- ation operator to the interface boundary conditions
placements and equilibrium of the nodal forces at the (Elleithy and Tanaka 2003).
interface nodes (von Estorff 1992). The application of an iterative coupling algo-
The use of nodal displacements and forces requires rithm to a transient, hybrid FEM–BEM formulation
some modifications in equations (2) and (3), namely was recently proposed and implemented for two-
the transformation of surface tractions and pressures, dimensional problems by Soares Jr., von Estorff, and
respectively, to equivalent nodal forces, and, for fluid Mansur (2004). The algorithm allows to use different
domains, the transformation of pressure derivatives to time step durations in the subdomains and, moreover,
displacements. The latter is accomplished by taking performs the coupling iterations very effectively to-
into account that the pressure derivatives are linearly gether with the iterations which are needed to solve
related to the normal accelerations (Batchelor 1967), the nonlinear FEM system.
For the analyses presented in this contribution, the
un
q = −ρ ü , (4) algorithm is further modified in order to allow for
an arbitrary number of subdomains. Thus, a parallel
and by using an appropriate time integration scheme, Dirichlet–Neumann algorithm (Elleithy and Tanaka
such as Newmark’s method (Bathe 1996), to trans- 2003) is obtained. Moreover, the focus is placed on
form the accelerations to displacements. three-dimensional systems.
Re-arranging the matrices and vectors, and some
additional algebra, yields from (2) and (3) the expres- 4 NUMERICAL EXAMPLES
sion
The outlined coupling procedures have been applied
AB mx B = my B + mr B , (5) to several three-dimensional wave propagation prob-
lems. These include simple models, in order to prove
where the system matrix AB is constant for all time the accuracy of the new methodology, as well as more
steps. The vector mx B contains the unknown nodal realistic systems like a typical water retaining struc-
forces and displacement or pressure derivative values, ture.
respectively, while my B contains the according known In all cases, elements with quadratic shape func-
nodal values, and mr B contains the influences from tions are used (FEM: 20-node solid elements; BEM:
previous time steps, from incident waves and from the 8-node surface elements). For the fluid regions mod-
time integration of the nodal accelerations. elled with the BEM, multiple nodes are employed at
corners and edges in order to account for the non-
3.1 Direct Coupling uniqueness of the normal flux in these points. Addi-
tional equations at those nodes are obtained, where
Taking into account the abovementioned coupling necessary, based on the assumption of a unique pres-
conditions, equations (1) and (5) can be assembled sure gradient (Sládek and Sládek 1992).
into a single, coupled system of equations. The BEM
subdomains are treated as linear finite “super ele-
ments” (Czygan and von Estorff 2000, Firuziaan and 4.1 Elastic rod
von Estorff 2002). By solving the coupled system, An elastic rod (cross section 4 m × 4 m, length 12 m),
all subdomain systems are solved simultaneously. consisting of a homogeneous, linearly elastic mate-
This way, many of the sophisticated solution schemes rial with Young’s modulus E = 1.042 × 105 kN/m2 ,
which exist for the FEM can also be used for the hy- Poisson’s ratio ν = 0.25 , and mass density ρ =
brid formulation. 2.0 t/m3 (P-wave velocity cP = 250.0 m/s ), is ana-
lyzed.
3.2 Iterative Coupling The boundary conditions are chosen such that a
one-dimensional problem, for which an analytical
Alternatively, a partitioned solution scheme can be solution exists (Schanz 2001), is defined. The rod
used (Park and Felippa 1983, Le Tallec 1994, Soares is subjected to a Heaviside forcing function p(t) =

1262 �����������������������������������������������
1.0 kN/m2 H(t) in vertical direction at the bottom 1
top (BEM)
surface, while at all other surfaces, the normal dis- top (analytical)
interface (BEM)
placements are restrained to zero. 0.8 interface (analytical)
The model is decomposed into three subdomains.
The lower half is modelled using the FEM, while the

Nodal forces [kN]

0.6
upper half consists of two parts, each having the di-
mensions 4 m × 2 m × 6 m and being modelled 0.4
by means of the BEM (Figure 1). All element edge
lengths are 1.0 m. The time step duration for the cal- 0.2
culations is chosen as ∆t = 0.0019 s.
0

0 0.1 0.2 0.3 0.4 0.5

� 2 (BEM)
Time [s]
Figure 2. Time histories of the nodal forces at the top of the rod
� 3 (BEM) and at the central interface, respectively.

interface (BEM)
0.00025 interface (analytical)
� 1 (FEM)
bottom (BEM)
bottom (analytical)
0.0002

Vertical displacements [m]

Z

X Y 0.00015

Figure 1. Coupled FEM–BEM model of an elastic rod consisting

0.0001
of three subdomains.

5e-05
The response of the rod is investigated at three
points, namely at the centers of the (restrained) top 0
surface and of the horizontal interface, respectively, of
subdomain Ω3 , and at the center of the (loaded) bot- 0 0.1 0.2 0.3 0.4 0.5
tom surface. The nodal forces and the displacements Time [s]

at these points are shown in Figures 2 and 3. It was Figure 3. Time histories of the vertical displacements at the cen-
found that the results obtained from the direct cou- tral interface and at the bottom of the rod, respectively.
pling approach are identical to those obtained from
the iterative one. All results are reasonably close to normal displacements are again restrained to zero.
the analytical solutions. The smoothing of the sharp The time step duration for the calculations is chosen
corners which occur in the solutions is caused by the as ∆t = 0.00091 s.
coarse discretization rather than by the coupling pro-
cedure. The time histories of the hydrodynamic pressures at
the fluid–structure interface and at the bottom of the
4.2 Fluid–structure interaction fluidic region are shown in Figure 5. Figure 6 depicts
the vertical displacements of the loaded top surface
The next example deals with dynamic fluid–structure and of the fluid–structure interface. As in the previous
interaction. Again, one-dimensional wave propaga- example, the results are in good agreement with the
tion is studied. The employed model consists of an analytical solutions. Wave reflections at the interface
upper, linearly elastic region, which is modelled us- are not observed.
ing the FEM, and a lower, fluidic region, for which
the BEM is employed (Figure 4). Both regions have
the dimensions 6 m × 6 m × 6 m. The material pa-
rameters of the elastic region are: Young’s modulus elastic�region
E = 1.0 × 106 kN/m2 , Poisson’s ratio ν = 0.25 , and (FEM)

mass density ρ = 1.0 t/m3 . The fluid region has a

compression modulus K = 1.2 × 106 kN/m2 and a
fluidic�region
mass density ρ = 1.0 t/m3 . Thus, the P-wave velocity (BEM)
in both regions is cP = 1095.45 m/s . z
The top surface of the elastic region is subjected y
to a Heaviside force function p(t) = 1.0 kN/m2 H(t), x

acting in vertical direction. At all other surfaces, the Figure 4. Coupled FEM–BEM fluid–structure interaction model.

���������������������������������������������� 1263
 3
interface (BEM)
early elastic material ( Er = 1.0 × 107 kN/m2 , νr =
interface (analytical) 0.25 , ρr = 1.8 t/m3 ), and the reservoir water is given
2.5 bottom (BEM)
bottom (analytical) by Kw = 2.06 × 106 kN/m2 and ρw = 1.0 t/m3 .
2 The material parameters of the far field of the soil
are Es = 5.0 × 105 kN/m2 , νs = 0.40 , and ρs =
Pressure [kN/m^2]

1.5 1.8 t/m3 . In the near field of the soil, a Drucker–

1
Prager material law with kinematic hardening is em-
ployed (Belytschko, Liu, and Moran 2000). The elas-
0.5 tic material parameters are the same as in the far field,
while the friction angle is ϕs = 0.436 rad and the
0 hardening parameter is Hs = 1.0 × 103 kN/m2 . The
-0.5
cohesion is taken to be either cs = 150 kN/m2 or
0 0.02 0.04 0.06 0.08 0.1 0.12 cs = 200 kN/m2 . Alternatively, a linear material be-
Time [s] haviour in this region is also taken into account.
Figure 5. Time histories of the pressures at the fluid–structure
interface and at the bottom of the fluidic region, respectively.
The time step duration for the calculations is cho-
sen as ∆t = 0.0015 s. At those sides of the reservoir
which are not adjacent to the wall, zero flux boundary
3e-05
top (BEM) conditions (rigid boundaries) are applied. At the free
top (analytical)
2.5e-05 interface (BEM) surface, the pressure is assumed to vanish.
interface (analytical)
Two load cases are considered. First, the response
Vertical displacements [m]

2e-05 of the system due to an impact load at the top of the

wall is observed. The impact load acts during a time
1.5e-05
period of 0.0075 s in horizontal (x) direction along
1e-05
the top surface of the wall. Its variation in time is of
triangular shape, and the peak load is 30 000 kN.
5e-06 The response of the wall due to this load case is
shown in Figures 8 to 10. First, a linear material be-
0 haviour is assumed in the near field soil, and the re-
0 0.02 0.04 0.06 0.08 0.1 0.12 sponses considering the full reservoir are compared
Time [s] to those for the empty one. It can be seen that the am-
Figure 6. Time histories of the displacements at the top of the plitudes at the top of the reservoir are slightly smaller
elastic region and at the fluid–structure interface, respectively. when the reservoir is full (Figure 8). Whereas, the am-
plitudes at the bottom are smaller for the empty reser-
4.3 Soil–fluid–structure interaction voir (Figure 9). Clearly, the reservoir has a consider-
able influence on the vibration period of the system.
As a more realistic earthquake engineering example, a The effect of plastic deformations of the soil can be
dynamic system consisting of a slender retaining wall, seen from Figure 10. They cause displacements of the
a reservoir, and a soft subsoil is investigated. All in- wall which are considerably larger than in the linear
teraction effects between these subdomains are taken case.
into account in the model. This example demonstrates
the applicability of the proposed computational model retainment�wall�(FEM) reservoir
to sophisticated earthquake engineering problems. inhomogeneous
(BEM)

The considered system is given in Figure 7. The soil�region�(FEM)

wall is discretized with 32 finite elements, while the

reservoir, which is considered to be infinite in the up- surface�of�the
halfspace�(BEM)
stream direction, is represented by 160 boundary ele-
ments. The subsoil is partitioned into a near field and
a far field region. In the near field region, close to the
wall, large strains are expected to appear. This part is
discretized with 224 finite elements, allowing to con-
sider a nonlinear material law. A reasonable part of X

the far field of the soil (i.e. the coupling interface be- Y
tween near field and far field and a part of the half- Z

space surface) is discretized with 302 boundary ele-

ments. Figure 7. Coupled FEM–BEM model of a retaining wall–
The retaining wall consists of a homogeneous, lin- reservoir–soil system.

1264 �����������������������������������������������
 0.1 cillations of the system, where the observed vibra-
Displacements in x direction [m]

tion periods are considerably different from the dom-

0.05 inant period of the incident wave. This also indicates
that the resonance period of the wall–reservoir–soil
0
system is different from the resonance period of the
rigidly supported wall. It can also be seen that the
-0.05
full reservoir
reservoir water, due to the added mass, causes a re-
empty reservoir duction of the vibration frequency of the system.
-0.1
0 0.1 0.2 0.3 0.4 0.5 The ratio of the damping, which is caused by radi-
Time [s] ation of energy to infinity, seems to be only slightly
Figure 8. Time histories of the x displacements at the top of the influenced by the reservoir. Since the reservoir itself
retaining wall due to impact load. is infinite, reservoir resonances are not present.
The hydrodynamic pressure which acts on the wall
Displacements in x direction [m]

0.003
0.002
0.020
0.001

Amplitude�[m]
0
0.010
-0.001
-0.002
0.000
-0.003
-0.004 full reservoir
empty reservoir -0.010
-0.005 -0.025 0.000 0.025 0.050 0.075 0.100 0.125 0.150
0 0.1 0.2 0.3 0.4 0.5
Time�[s]
Time [s]
Figure 9. Time histories of the x displacements at the foot of the Figure 11. Transient behaviour of the incident wave (Ricker
retaining wall due to impact load. wavelet).
0.004 full reservoir (direct)
Displacements in x direction [m]

0.04 full reservoir (iterative)

0.002 empty reservoir (direct)
empty reservoir (iterative)
Displacements in x direction [m]

0 0.02
-0.002

-0.004 0
linear
-0.006 c = 200
c = 150 -0.02
-0.008
0 0.05 0.1 0.15 0.2 0.25
Time [s] -0.04
Figure 10. Time histories of the x displacements at the foot of the
retaining wall due to impact load, considering the empty reser- -0.06
voir: Influence of cohesion c. 0 0.1 0.2 0.3 0.4 0.5 0.6
Time [s]
The second load case which shall be considered Figure 12. Time histories of the x displacements at the top of the
here is the case of an incident plane SH wave. The retaining wall due to incident wave. Results obtained by means
propagation direction of the wave is given by p = of the direct and the iterative coupling approach, respectively.
(0, 1, −1)T , and the direction of particle displace- 0.03 full reservoir (direct)
ments (polarization) is d = (1, 0, 0)T , such that the full reservoir (iterative)
empty reservoir (direct)
wall is excited in its weak direction (x). The time vari- empty reservoir (iterative)
Displacements in x direction [m]

0.02
ation of the input motion is given by a Ricker wavelet
(Figure 11) with an amplitude of 0.02 m and a domi-
0.01
nant period of 0.0536 s, where the latter corresponds
to the second mode of vibration of the wall if consid-
ered as a rigidly supported cantilever beam. 0

The displacement responses of the system due to

the incident wave are surveyed at the top as well as -0.01

at the foot of the wall, considering the full and the

empty reservoir, respectively. The time histories of the -0.02
0 0.1 0.2 0.3 0.4 0.5 0.6
displacements at these points in x-direction are given Time [s]
in Figures 12 and 13. Purely linear material behaviour Figure 13. Time histories of the x displacements at the foot of the
is assumed. retaining wall due to incident wave. Results obtained by means
The single incident wavelet causes resonance os- of the direct and the iterative coupling approach, respectively.

���������������������������������������������� 1265
due to the full reservoir is shown in Figure 14. These ACKNOWLEDGEMENTS
pressures represent the interaction forces between the
wall and the reservoir. The largest amplitudes occur The authors would like to thank the Deutsche
only during the first 0.2 s. Later, when the system Forschungsgemeinschaft (DFG) for the financial sup-
is vibrating freely, the interaction forces are much port of their studies through a scholarship, granted to
smaller. the second author, within the graduate research pro-
gram (Graduiertenkolleg) ”Ocean Engineering Struc-
tures”.
Hydrodynamic pressure [kN/m^2]

1000
REFERENCES
500

Batchelor, G. K. (1967). An Introduction to Fluid Dynamics.

0
Cambridge: Cambridge University Press.
-500 Bathe, K.-J. (1996). Finite Element Procedures. Englewood
Cliffs: Prentice-Hall.
-1000 (direct) Belytschko, T., W. K. Liu, and B. Moran (2000). Nonlinear
(iterative)
Finite Elements for Continua and Structures. Chichester:
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 John Wiley & Sons.
Figure 14. Time histories of the hydrodynamic pressures on the Bonnet, M. (1995). Boundary Integral Equation Methods for
retaining wall at 2m above ground level due to incident wave. Solids and Fluids. Chichester: John Wiley & Sons.
Results obtained by means of the direct and the iterative coupling Czygan, O. and O. von Estorff (2000). FEM/BEM coupling
approach, respectively. for fluid–structure interaction including nonlinear ef-
fects. In C. A. Brebbia and H. Power (Eds.), Boundary
Elements XXII. Southampton: Computational Mechanics
5 CONCLUSION Publications.
Domı́nguez, J. (1993). Boundary Elements in Dynamics.
Southampton: Computational Mechanics Publications.
A modelling approach for three-dimensional dynamic
Elleithy, W. M. and M. Tanaka (2003). Interface relaxation al-
analyses of soil–fluid–structure interaction problems gorithms for BEM–BEM coupling and FEM–BEM cou-
in the time domain has been presented. The approach pling. Computer Methods in Applied Mechanics and En-
combines the respective merits of the finite element gineering 192, 2977–2992.
and the boundary element methods, and thus is well von Estorff, O. (1992). Coupling of BEM and FEM in the
suited to handle inhomogeneities, nonlinearities as time domain: Some remarks on its applicability and effi-
well as unbounded subregions. Moreover, the occur- ciency. Computers & Structures 44(1/2), 325–337.
rence of incident wave fields can be accounted for in Firuziaan, M. and O. von Estorff (2002). Transient 3D
soil/structure interaction analyses including nonlinear ef-
a highly accurate manner. fects. In H. Grundmann and G. I. Schuëller (Eds.), Struc-
The presented approach has been applied to some tural Dynamics – EURODYN2002, Volume 2, pp. 1291–
representative, three-dimensional wave propagation 1296. Lisse: Swets & Zeitlinger.
problems. These include some simple examples in or- Le Tallec, P. (1994). Domain decomposition methods in
der to show the validity of the proposed method. The computational mechanics. Computational mechanics ad-
vances 1(2), 121–220.
results were compared to to the analytical solutions,
and good agreement was found. Park, K. C. and C. A. Felippa (1983). Partitioned analysis
of coupled systems. In T. Belytschko and T. J. Hughes
The applicability of the method to sophisticated (Eds.), Computational Methods for Transient Analysis,
soil–fluid–structure interaction problems, involving Chapter 3, pp. 157–219. Amsterdam: North-Holland.
nonlinear material behaviour, was demonstrated by Schanz, M. (2001). Wave Propagation in Viscoelastic and
the investigation of a retaining wall–reservoir–soil Poroelastic Continua – A Boundary Element Approach.
system, subjected to transient loadings. The responses Berlin: Springer.
of the system to an impact load and to an incident Sládek, V. and J. Sládek (1992). Why use double nodes
in BEM? Engineering Analysis with Boundary Ele-
wave field was analysed, illustrating the employment ments 8(2), 109–112. Technical Note.
of the approach for earthquake engineering problems.
Soares Jr., D., O. von Estorff, and W. J. Mansur (2004). Iter-
From the numerical examples it became obvious ative coupling of BEM and FEM for nonlinear dynamic
that the presented approach works very well. The analyses. Computational Mechanics 34, 67–73.
three-dimensional formulation allows the investiga-
tion of rather realistic problems, eg. the dynamic be-
haviour of water retaining structures or offshore plat-
forms.

1266 �����������������������������������������������