Engineering Structures 85 (2015) 206–218

Contents lists available at ScienceDirect

Engineering Structures
journal homepage: www.elsevier.com/locate/engstruct

An efficient computational procedure for the dynamic analysis of liquid

storage tanks
R.O. Ruiz a,b, D. Lopez-Garcia a,⇑, A.A. Taflanidis b
a
Department of Structural & Geotechnical Engineering, Pontificia Universidad Catolica de Chile, and National Research Center for Integrated Natural Disaster Management
CONICYT/FONDAP/15110017, Av. Vicuna Mackenna 4860 Macul, Santiago, RM 782-0436, Chile
b
Department of Civil & Environmental Engineering & Earth Sciences, University of Notre Dame, 156 Fitzpatrick Hall, Notre Dame, IN 46556, USA

a r t i c l e i n f o a b s t r a c t

Article history: A computationally efficient numerical model is developed in this study for evaluating the dynamic
Received 4 March 2014 behavior of liquid storage tanks. This model has higher complexity than the Housner model (which cor-
Revised 31 August 2014 responds to the simplest and most popular approach for approximating the behavior of rectangular and
Accepted 8 December 2014
circular tanks) but still enjoys high computational simplicity to facilitate implementation in practice,
while it is applicable to virtually any kind of tank geometry, providing at the same time a high degree
of accuracy. In the proposed model, the liquid is assumed to be inviscid, incompressible and irrotational,
Keywords:
and its motion is completely characterized by a velocity potential function. Thus, the Continuity and
Liquid storage tanks
Seismic analysis
Equilibrium equations characterizing this motion take the form of Laplace and Bernoulli equations,
Dynamic analysis of liquids respectively. The Laplace equation is solved through a 2D finite element scheme, and is then combined
Tuned liquid dampers with the Bernoulli equation through the velocity potential function condensed at the free surface of
the liquid. Numerical details for the practical implementation of the proposed scheme are discussed,
whereas the approximation is shown to provide results with high accuracy for the dynamic behavior
of different type of tanks when compared to the Housner model and a full finite element implementation.
As shown in the examples considered the computational efficiency of the proposed model is such that
extensive parametric studies can be performed with small numerical effort, which in turn makes the
proposed model very attractive not only for analysis purposes but also for the design of liquid storage
tanks and other related devices such as tuned liquid dampers.
Ó 2014 Elsevier Ltd. All rights reserved.

1. Introduction storage tanks, based on a simple mechanical model (combination

of mass-spring systems with different characteristics) that repre-
The study of the dynamic behavior on liquid storage tanks has sents the fluid. The computation of the physical constants in this
gained significant attention in the last years as the seismic vulner- procedure is based on the separation of the hydrodynamic
ability of these tanks represents a potential source of significant behavior into two components: (1) the impulsive component that
economic loss due to structural failures, leakages or environmental is related to the mass that moves together with the structure; and
accidents (caused by the liquid spilled out) [1–3], whereas such (2) the convective component that takes into account the free
tanks have been proposed to be used as mass dampers to mitigate surface oscillations [6,7]. This is a broadly adopted model in civil
the vibration induced by wind or seismic excitations [4]. Several engineering since it provides closed form solutions for the
approaches have been proposed in this setting to model the transmitted force due to the liquid sloshing, and represents the
dynamical behavior of such liquid storage tanks. In earlier studies, basis of many design codes, i.e. API 650 [8], AWWA D100 [9] and
the fluid was taken into account by adding a mass to the structure, the New Zealand recommendation guidelines NZSEE [10], that
with characteristics computed by an analytical solution based on establish procedures for the seismic response analysis of liquid
simplified geometries [5]. Later, Housner developed an analysis tanks based on this linear model proposed by Housner. At the same
and design procedure, primarily for cylindrical and rectangular time, it is an approximation that is based on the assumption that
simplified flows can represent the actual fluid movement, restrict-
⇑ Corresponding author. Tel.: +56 2 2354 7684. ing its use to tanks with simple geometries (such as rectangular or
E-mail addresses: [email protected] (R.O. Ruiz), [email protected] (D. Lopez-Garcia), circular tanks).
a.tafl[email protected] (A.A. Taflanidis).

http://dx.doi.org/10.1016/j.engstruct.2014.12.011
0141-0296/Ó 2014 Elsevier Ltd. All rights reserved.
 R.O. Ruiz et al. / Engineering Structures 85 (2015) 206–218 207

To obtain higher accuracy solutions, various high-fidelity proce- as a second order linear system of equations, where the indepen-
dures [11–13] have been also established, applicable to arbitrary dent variables are the vertical elevation of the free surface and
tank geometries, using the Finite Element Method (FEM) to model the excitation is directly related to the ground acceleration. The
the fluid utilizing (depending on the numerical scheme) displace- fluid is assumed ideal while the tank walls and bottom are
ment, pressure or potential variables to characterize the fluid assumed rigid. As the fluid is considered ideal, it is possible to
motion. For generalized fluid–structure interaction applications, adopt a FEM formulation based on potential variables, reducing
as the equations of the structure are expressed in terms of dis- the number of unknowns and avoiding problems with spurious fre-
placements it is convenient to also express the fluid equations quencies. The rigid tank assumption simplifies the fluid–structure
through displacement variables. With respect to the fluid motion coupling since it is not necessary to generate a mesh for the tank
modeling, such a FEM approach, based on a displacement formula- walls and bottom and to match it with the fluid mesh. In this sense,
tion, leads to a symmetric eigenvalue problem but it produces non- the procedure is easier to implement than the ones cited previ-
zero spurious frequencies that are difficult to identify [12,14] and ously [12–20]. Furthermore, the proposed procedure allows study-
suppress [13,15]. Additionally, this approach requires discretizing ing the sloshing effect over the tank support rather than the
a vector field (displacement) instead of a scalar field (pressure or sloshing effect over the tank itself [12–20]. Although the method-
potential variables), increasing the number of degrees of freedom. ology is standard, the numerical procedure offers significant
In contrast, a FEM approach based on pressure or potential vari- advances and physical insight as: (1) the equations are expressed
ables involves fewer unknowns, increasing the computational effi- in terms of physical variables (free surface elevation and ground
ciency and avoiding physical inconsistencies [16]. In this case, the acceleration), (2) the system of equations is similar to that of
fluid–structure system leads to a non-symmetric eigenvalue prob- mass-spring systems, (3) the approach is valid for any tank geom-
lem, though it is possible to keep the symmetry if the fluid is etry, (4) the formulation is suitable for both time-history and fre-
described in a redundant way using both pressure and potential quency analysis, (5) it allows for a straightforward coupling
variables [17–19]. In particular, Olson and Bathe presented such between rigid tanks and elastic structures, (6) the procedure is rel-
a linear formulation based on velocity potentials and pressures ative easy to implement or understand, and (7) it offers significant
[17], expanded later to take also into account gravity loads [20]. advances over the models suggested by the design codes. Ulti-
An important aspect of this formulation is its suitability for both mately, the proposed numerical procedure, from now on denoted
time-history and frequency analysis of fluids with free surface. as Simplified Sloshing Model (SSM), enjoys such computational
More recently, an increased number of even more complex proce- simplicity and efficiency that it can be used for various tasks such
dures have been proposed, for example taking into account non- as parametric studies, preliminary dimensioning of tanks, seismic
linear sloshing due to large free surface motions [21–24] and performance identification, or even design and dimensioning of
including identification of damping effects introduced at the tank tuned liquid dampers.
walls due to viscosity effects in the thin interface layer [25]. How-
ever, implementation of these procedures is almost exclusively rel-
2. Description of the Simplified Sloshing Model (SSM)
egated to scientific and research professional environments due to
the complexity of the formulations and the high level of expertise
A 2D schematic diagram of a liquid storage tank is presented in
required for their implementation.
Fig. 1. It is important to mention that, although the formulation of
Despite such high-fidelity modeling developments and
the proposed Simplified Sloshing Model could be either 2D or 3D,
advances in computer and computational science, the philosophy
this paper presents in detail only a 2D formulation because it is
of the analysis methods of design codes is still based on analytical
the case most extensively studied in the literature, and the valida-
expressions and equivalent mechanical models. Though undoubt-
tion will be made considering several 2D existing examples ana-
edly some practitioners are utilizing commercial software to solve
lyzed independently by other authors. An inertial system of
multi-physic problems under seismic loads, avoiding the use of
reference x–z is located at the middle of the non-perturbed free
simplified models proposed by the design codes or even the com-
surface and an auxiliary coordinate g is defined to measure the rel-
plex implementation of the procedures described above, such
ative displacement between the free surface and the coordinate
approaches are still not widely used (presumably because a signif-
system. Let X represent the volume of liquid, Co the non-perturbed
icant background is required not only in the software know-how
free surface (at z = 0), Cs the free surface at any time t, and Cp the
but also in the theoretical knowledge about the involved physics).
walls and bottom surfaces (all these variables are also shown in
Furthermore, many traditional software packages used for seismic
Fig. 1). The liquid motion is modeled using principles of Mass
and structural analysis lack fluid–structure interaction modules,
and Momentum Conservation, while the tank walls and bottom
enforcing engineers to work with alternative packages that were
are considered to be rigid. The liquid is assumed to be inviscid,
not designed to perform seismic analysis. There is a gap for a meth-
incompressible, and irrotational, allowing its motion to be com-
odology that is more simple and attractive than the commercial
pletely defined by a velocity potential function u. Additionally,
packages but still maintains the accuracy of the advanced methods
body forces are assumed conservative and nonlinear terms are
presented in the literature.
Motivated by this realization, the main novel contribution of
this work is to develop a simplified, computationally efficient Γs
framework, utilizing a FEM modeling based on potential variables Γo
and a static condensation approach while assuming the tank walls
x
as rigid, for describing the dynamic behavior of arbitrary geometry
η
liquid storage tanks under seismic excitation. This approach can
H
facilitate a computationally efficient description of the dynamic Γp
behavior of tanks (supporting frequency and time domain analysis Ω

as well as eigenvalue analysis) including its interaction with a sup-

porting structure (as needed for TLD design applications), though it L

cannot provide detailed predictions for localized failure phenom-

ena related to the tank walls (which are considered rigid). The pro- Fig. 1. Scheme of the liquid storage tank: (left) arbitrary geometry; (right)
posed numerical procedure expresses the linear sloshing problem rectangular.
208 R.O. Ruiz et al. / Engineering Structures 85 (2015) 206–218

neglected. Thus, Mass and Momentum Conservation equations A typical FEM discretization (more details on this later) is then
take the form of Laplace and Bernoulli equations, which are given, applied to the velocity potential u, the auxiliary coordinate g and
respectively, by [26]: the weighting function w, such that:
Du ¼ 0 ð1Þ uðx; z; tÞ ¼ Nðx; zÞuðtÞ ð8Þ

@u p gðx; tÞ ¼ Ng ðxÞgs ðtÞ ð9Þ

€g x ¼ 0
þ þPþu ð2Þ
@t q
where p is pressure, q is the fluid density, G is the potential of the con- wðx; z; tÞ ¼ Nðx; zÞwðtÞ ð10Þ
servative forces and üg is the horizontal acceleration of the tank. The It is important to mention that u is a vector built with the values of
term üg x must be introduced because the motion of the liquid is
the nodal velocity potential in the volume X, and that gs is a vector
expressed with respect to a coordinate system (x–z) that moves along
that contains the values of the nodal displacements of the free sur-
with the tank, whereas the horizontal acceleration üg is expressed
face (here, the bar under the Greek letters is used to distinguish vec-
with respect to a fixed coordinate system (full details can be found
tors and functions, indicating vector variables when the bar is used
in [4]). It is important to notice that Du defines the velocity of any
and scalar functions when it is not used). Also, N is the vector of
particle of the liquid. Therefore, ru  n corresponds to the velocity
interpolation functions for the velocity potential u, while Ng is
of the liquid projected over the normal vector n. Two normal vectors
the vector of interpolation functions for the auxiliary coordinate g
are defined: no, perpendicular to the free surface Co; and np, perpen-
within the FEM scheme. In this study linear functions are adopted
dicular to the tank walls and bottom. Small displacements at the free
for both N and Ng. By partitioning u into components associated
surface are assumed, in order to simplify the problem. Thus, the lin-
and non-associated with the free surface (us and ur , respectively),
earized boundary condition at the free surface is given by [27]:
(7) is transformed into a linear system of ordinary differential
@u equations given by:
ru  n o ¼ ¼ g_ at Co ð3Þ
@z
 " g_ #  " #
In turn, the normal velocity of the liquid at Cp is zero due to the G 0 s Dss Dsr us
 ¼0 ð11Þ
rigid condition of the tank walls and bottom, and the boundary con- 0 0 g_ r Drs Drr ur
dition at Cp is given by [27]:
with:
ru  np ¼ 0 at Cp ð4Þ Z
G¼ NTg Ng dCs ð12Þ
A specific manipulation of the Laplace and Bernoulli equations is Cs
required to combine both equations such that the sloshing dynamics
  Z
could be expressed only by the elevation of the free surface. Section Dss Dsr
2.1 shows the procedure based on the FEM to handle the Laplace D¼ ¼ rNT rNdX ð13Þ
Drs Drr X
equation. It is important to note that Laplace equation defines only
the spatial solution, in other words, the solution for the velocity of The vector gr is included only to complete the second block of equa-
any particle of the fluid (as a function of the walls and free surface tions, allowing to express (11) in a matrix form. A static condensa-
velocities) but does not give information about how the velocities tion is finally performed in order to express (11) only in terms of the
change over time. Section 2.2 shows how to manipulate the Ber- surface variables, such that the condensation is defined by:
noulli equation to make it compatible with the Laplace equation.
J ¼ Dss  Dsr D1
rr Drs ð14Þ
The combination of these equations addresses the temporal varia-
tion, allowing for the modeling of the time-history evolution of the and then becomes:
liquid motion. Section 2.3 then discusses how the shear forces trans-
Gg_ s  Jus ¼ 0 ð15Þ
mitted by the tank to the supporting ground can be estimated. Such
a connection, ultimately between the free surface elevation and the which is the final expression of the Laplace Equation, where the free
transmitted forces to the ground, allows the analysis of liquid stor- elevation velocity and velocity potential of the free surface are
age tanks supported by elastic structures (which for example facili- related such that the coupling with the Bernoulli Equation is possi-
tate the analysis of sloshing dampers). The numerical details of all ble. It is important to notice that the static condensation introduced
these components are summarized in Section 2.4. here [shown in (14)] is key in this procedure because: (a) it will
make possible in the next section to express the equations of
2.1. Laplace equation modification motion in terms of variables whose physical meaning can be readily
appreciated (vertical displacements of the liquid surface); and (b) it
Galerkin’s Weighted Residual Method is applied to solve the makes the method computationally efficient as it considerably
Laplace equation. In particular, (1) is expressed in a weak formula- reduces the number of variables involved.
tion using w as a weighting function, leading to [26]:
Z 2.2. Bernoulli equation modification
wDudX ¼ 0 ð5Þ
X
In order to solve the Bernoulli equation, it is necessary to define
Then, Green’s First Identity is applied to include boundary condi- the potential G. Under this setup, the potential of volume forces is
tions in an explicit way: defined as P = gz, where g is the gravity acceleration. The Bernoulli
Z Z Z equation is evaluated at the free surface, imposing z = g, pjz¼g ¼ 0
 
wðru  no ÞdCo þ w ru  np dCp  ðrw  ruÞdX ¼ 0 ð6Þ and u_ jz¼g  u
_ jz¼0 . Then (2) is simplified as:
Co Cp X

and, by including boundary conditions, we have: u_ þ g g þ u€ g x ¼ 0 at Cs ð16Þ

Z Z
wg_ dCo  rwT rudX ¼ 0 ð7Þ Furthermore, (16) is particularized for each node located at the free
Co X surface using the previous FEM discretization, leading to:
 R.O. Ruiz et al. / Engineering Structures 85 (2015) 206–218 209

u_ s þ ðgIIÞgs þ ðIIXs Þu€ g ¼ 0 ð17Þ the total horizontal force transmitted to the ground is by multiply-
ing Fp by a ‘‘1  m’’ vector of ones, denoted S herein:
where II is the identity matrix and Xs is a vector containing the x
coordinate of each node located at the free surface. Then (15) is F ¼ ½ 1 1    Fp ¼ SFp ð24Þ
introduced in (17), leading to:
Finally, the total force transmitted to the ground is expressed as:
Ms g
€ þ Ks gs ¼ Rs u
€g ð18Þ  
s  
F ¼  qSRp Q J1 G g€  qSRp Xp u
s
€ g ¼ Ag
€  Bu
s
€g ð25Þ
with:

Ms ¼ J1 G This expression ultimately addresses the shear force between the
ð19Þ tank and the ground or, more generally, to the surface on which
Ks ¼ g II
the tank rests. Hence, this force can be used to connect the tank
Rs ¼ II Xs with a supporting elastic structure.
This numerical procedure allows expressing the dynamic behavior
of liquid storage tanks as a second order lineal system of equations, 2.4. Summary of the numerical approach
such that the sloshing problem could be interpreted as a mass-
spring system as the resultant equation of motion involves equiva- This Section summarizes the numerical approach for analyzing
lent mass and stiffness characteristics. Moreover, (18) is expressed the sloshing of a liquid storage tank either on its own or when it is
in physical variables since it uses the ground acceleration as input supported by an elastic structure (leading to a fluid structure inter-
and the free surface elevation as the dependent variable. This ulti- action problem). An overview is presented in Fig. 2 where the
mately allows the solution of the sloshing problem to be computed liquid storage tank is analyzed by the SSM and the support struc-
with respect to the same physical variables (displacement and ture by a traditional FEM approach.
acceleration) as typically used in structural modeling, circumvent- Sloshing Model (SSM): First, is necessary to define the geometry
ing the velocity potential variables. Note that the obtained solution of the tank and the physical characteristics of the fluid. Then, the
completely characterizes the temporal variation, facilitating ulti- fluid is discretized (meshed) with respect to the Equilibrium posi-
mately both time-history or (if preferred) frequency analysis. tion, remarking that is not necessary to re-mesh at any step of the
time integration (size of mesh is discussed later). Though any
2.3. Pressure on the tank walls approach can be implemented for this discretization, in the exam-
ples presented later the Delaunay Triangulation Algorithm [28] is
The methodology described previously only estimates the free used. Second, the FEM is applied to the fluid by computing (12)
surface elevation when the tank is subjected to a specific base and (13) to find G and D. Next, the matrix J in (14) is computed
motion, but does not give information about the force transmitted to condensate the equations at the free surface. Third, the equiva-
to the ground. The coupling between the fluid storage tank and the lent model is generated by the identification of Ms, Ks and Rs in
support structure may be then described by the reaction force (18) and the eigenvalue problem (knowing Ms and Ks) is solved
between them. This requires a proper estimation of the dynamical to identify the natural sloshing frequencies xi. The accuracy of
pressure on the walls due to the sloshing effect. For this purpose these frequencies may be used as a criterion to select the mesh
(2) is evaluated at the tank wall and bottom in order to obtain size. The original mesh should be iteratively refined and the eigen-
the dynamic pressure over surface Cp: value problem revisited until small variations of the sloshing fre-
quencies of interest (typically the first few frequencies) is
Pp ¼ qu
_  qXp u
p
€g ð20Þ obtained within subsequent iterations. In general, the relative dif-
ference between the sloshing frequencies decreases exponentially
where Pp, up and Xp are vectors with the pressures, velocity poten-
when the number of elements on the mesh is increased and con-
tials and x coordinates of the nodes located at the surface Cp,
vergence is quickly established. After the final mesh of the fluid
respectively. Then, Q is defined as a transformation matrix, in order
has been determined, damping effects can be incorporated in the
to express up as function of us , such that up = Q us . The pressure at
formulation by selecting matrix Cs through any standard approach,
Cp is then expressed as a function of u € g and g
€ by the combination
s for example by establishing a pre-specified modal damping [29].
of (20) and (15):
Finally, the resultant equation of motion (18) for the liquid slosh-
    ing can be integrated in time by any conventional technique, for
Pp ¼  qQJ1 G g€  qXp u
s
€g ð21Þ
example through Runge–Kutta family solvers like the Dormand–
Nodal forces at the nodes located at Cp are calculated by integrating Prince method [30]. The parameters A and B to estimate the trans-
the pressure along the edge of the corresponding element (only the mitted force on the supporting structure can be also calculated
edge that coincides with Cp). The vector of the horizontal compo- using (25). These parameters will be needed if the ultimate goal
nents of these nodal forces at the nodes of the i-th element is given is to address a fluid structure interaction problem.
by: Fluid–Structure model: In this case a model for the structure that
Z supports the storage tank needs to be first adopted (for example
through a FEM approach). This ultimately leads to evaluation of
FpðiÞ ¼ sinðaðiÞ Þ NðiÞT ðiÞ ðiÞ ðiÞ
g Ng dCp Pp ð22Þ
CðiÞ
p
the mass M, damping C and stiffness K matrices and of the influ-
ence vector Do (Section 3.2 later offers more details). The SSM
where a(i) is the angle between the surface CðiÞ
p and the x–y plane. It and the structural model are then coupled by the dynamic pressure
is recalled that (22) is computed considering only those elements on the tank walls. The resultant coupling equation is a second
where one of the edges coincides with Cp. After computing all nodal order system of equations, which can be similarly integrated
forces, the system is expressed by: through any standard approach (as for the equation of sloshing
Fp ¼ Rp Pp ð23Þ motion). As the input of the SSM corresponds to its base accelera-
tion üg, for a problem with a multi-story structure a new vector is
If ‘‘m’’ is the number of nodes on Cp the dimension of Rp is ‘‘m  m’’ required (Ls) defining the location (floor) of the tank within the
and the dimension of Pp is ‘‘m  1’’. Note that an easy way to obtain structure as the tank base acceleration ultimately is defined by
210 R.O. Ruiz et al. / Engineering Structures 85 (2015) 206–218

Define Sloshing Model (SSM)

Geometry

No Yes Damping Model

Meshing Converged?
Cs

FEM Compare ωi
with previous
Yes No
G Tank rest on the
J iteration
D floor?
condensation

Transmitted
Solve Eigenvalue Problem ready Force
Equivalent Model for dynamic
Problem A, B
Ms, Ks, Rs Find ωi analysis

Fluid-Structure Ms, Cs, Ks, Rs

Define the location of Define the
Model A, B
the tank Ls and the support structure
Problem
ready for M, C, K, Do, Ls excitation üg
dynamic üg
analysis FEM
M, C, K, Do

Fig. 2. Procedure to use SSM in fluid–structure interaction models.

the acceleration of that floor (more details are offered again in that the springs transmit to the wall due to the movement of the
Section 3.2 through an illustrative example). convective masses. The most common implementation (and the
one that is used in the present work) of this procedure only takes
3. Validation/implementation of the model into account one impulsive and one convective mass, such that the
effect of the higher modes are neglected and the equivalent system
The validation of the SSM is considered here in a variety of two is defined only with a single degree of freedom. The linear spring
dimensional applications. The fluid inside the tank is meshed by a stiffness, the impulsive and convective mass values are then given
Delaunay Triangulation Algorithm [28] with the size of the trian- in a closed form as described in [7], which is the reason why this
gles controlled by imposing a maximum admissible value for its model is primarily used to approximate the fundamental sloshing
side length. The equivalent model generated by the SSM (Ms, Ks, mode of simple tank geometries.
Rs, A and B) is evidently affected by the quality of the mesh, and ANSYS implementation: the element FLUID79 is used in the anal-
the refinement procedure described in Section 2.4 is implemented ysis, which consists of a modification of a plane element to model
to adaptively choose the mesh size. Convergence is defined when solid structures. It uses 4 nodes with 2 degree of freedom on each
the variation in the first sloshing frequency is smaller than 1% (this node, where the degrees of freedom are the horizontal and vertical
applies to all cases studied). displacements of the node [31]. The tank walls and bottom are not
discretized since they are considered non-deformable. The mesh is
3.1. Tanks resting on the ground refined in the same way as for the SSM to obtain a variation in the
fundamental sloshing frequencies smaller than 1%. To facilitate the
In this section, harmonic and earthquake excitations are transient analysis, a reduced model approximation is adopted
imposed to different tank geometries. The fluid response obtained defining the nodes on the walls, bottom and free surface as master
by the SSM, the Housner model and a full FEM implementation nodes. In contrast to the SSM (in which the equations are reduced
through the commercial software ANSYS [31] are compared in to use only the free surface nodes), the reduced ANSYS model
terms of accuracy and computational efficiency. The tank walls requires explicitly the use of all border nodes (free surface, walls
and bottom are considered rigid and the liquid used is water. and bottom nodes) to impose the boundary and initial conditions.
Before moving forward, a brief review of the Housner model and This ultimately increases the computational time since ANSYS is
ANSYS implementation is given. enforced to use more degrees of freedom than the SSM. The fluid
Housner model: as described earlier this approximate model is displacements at the tank walls and bottom are constrained such
typically used for describing the sloshing in tanks with simple that the fluid movement is parallel to the surface.
geometries (primarily cylindrical and rectangular tanks that lead
to closed form solutions for all quantities of interest). Sloshing is 3.1.1. Rectangular tanks
modeled by a linear system formed by a group of masses that A rectangular tank (Fig. 1 right) of length L and water depth H
are attached to the walls. The first mass (named impulsive mass) (parametric investigations will be performed) fixed to the ground
is attached rigidly to the tank walls and moves along with the base. is considered first. R is defined as the aspect ratio R = H/L facilitat-
The additional masses (named convective masses) are attached to ing the intended parametric investigation. Table 1 presents the
the walls by linear springs. When only one such mass is added, it fundamental sloshing period of a specific geometry (L = 9.144 m
approximates explicitly the effect of the fundamental sloshing and R = 0.5) calculated via three different methods. The first
mode. The transmitted force to the ground is calculated as the method is the analytical solution presented in [32], which is valid
sum of the force generated by the impulsive mass and the force only for rectangular tanks. The second method is the Housner
 R.O. Ruiz et al. / Engineering Structures 85 (2015) 206–218 211

Table 1 M. If Fn is higher than 1 then the sloshing increase the total force
Sloshing periods for a rectangular tank with L = 9.1440 m and R = 0.5. transmitted to the ground, and if Fn is lower than 1 then the oppo-
Analytic Housner SSM site behavior occurs.
solution [32] model The numerical analysis is conducted with a tank of length
Fundamental period (s) 3.5755 3.5834 3.5734 L = 9.144 m and results are presented for three different aspect
Second sloshing period (s) 1.9771 – 1.9698 ratios, R = 0.3, R = 0.5 and R = 1. The liquid is assumed to have
Third sloshing period (s) 1.5314 – 1.5158 0.5% modal damping while the excitation period varies between
0.1 and 2.0 times the fundamental sloshing period. It is observed
model (used only for approximating the fundamental mode) and that for r  1, the normalized transmitted force is close to 1 so
the third method corresponds to the SSM. The three methods pre- the convective component is not important. The maximum value
dict essentially the same fundamental sloshing period. Table 1 also of Fn is reached when the excitation and the fundamental sloshing
includes higher sloshing periods computed by the analytical solu- periods are close (r  1) which is expected since this condition cor-
tion and by the SSM, which are also very similar to each other (dif- responds to resonance. On other hand, the minimum value of Fn is
ferences are in all cases less than 1%). The ANSYS estimation is not reached when r < 1. In this case, the impulsive mass moves at the
included here since the availability of an analytical solution is con- same phase as the excitation but the convective mass moves with
sidered more appropriate to compare the sloshing frequencies a 180 degrees phase difference with respect to the excitation. Then,
obtained by the SSM (the ANSYS model will be used later only in the forces related to the impulsive and convective masses counter-
the time-history analysis comparison). act each other leading to a reduced force transmitted to the
In order to investigate the behavior of the sloshing periods for ground, Fig. 6.
different rectangular tank dimensions, iso-period curves are plot- The SSM and the Housner model predict similar values of Fn
ted in Figs. 3–5. The value of the tank length L varies between when the period ratio is close to 1. The SSM additionally provides
1 m and 6 m while the aspect ratio R varies between 0.2 and 2.0. local maxima of Fn located at the second and third sloshing periods.
For the tank dimensions studied, the fundamental sloshing period The Housner model is of course not capable of estimating these
depends only on L if the aspect ratio is greater than 0.8 (Fig. 3). The local maxima since it only takes into account one degree of free-
second and third sloshing periods are, though, practically indepen- dom. It is also evident that the differences between the models
dent of the aspect ratio as shown in Figs. 4 and 5, respectively. In depend not only on the period ratio r but also on the aspect ratio
other words, the sloshing periods of a rectangular tank is indepen- R. The biggest differences occur when the aspect ratio R is small
dent of the water depth unless R < 0.8. For all cases presented, the and the excitation period is close to the second and third sloshing
maximum error found is less than 1.1% when the sloshing periods periods, that is, in Fig. 6, the difference between the model
of the analytical and the SSM solutions are compared. responses is much more dramatic when R = 0.3 and r < 1 (Fig. 6a),
The harmonic response is then studied by imposing harmonic while the differences decrease when the aspect ratio R increases
accelerations at the base of the tank. Here, the normalized trans- (Fig. 6b and c).
mitted force and the period ratio are utilized to examine the The seismic response of liquid storage tanks is finally studied by
results, defined, respectively, as: imposing two different ground motions: the 1940 El Centro earth-
quake (Component S90W) and the 1985 Mexico City SCT1 record
jFðtÞjmax
Fn ¼   ð26Þ (Component N90W). These seismic excitations are chosen to
M u€ g ðtÞmax
directly compare the SSM with the model implemented by Chen
et al. [32], in which a finite difference scheme is applied to resolve
T exc the non-linear and linear sloshing problem, where the non-linear-
r¼ ð27Þ
T slo ities are primarily related to the second order differential terms
where M is the total mass of liquid, Texc is the excitation period and that appear in the Bernoulli equation and the border conditions
Tslo is the fundamental sloshing period. The normalized transmitted at the free surface due to large sloshing amplitudes. Table 2 shows
force is an indicator of the force amplification due to the sloshing the maximum transmitted force in a rectangular tank
effects. In a sloshing tank, part of the liquid mass moves together (L = 9.1440 m and R = 0.5) subjected to the aforementioned seismic
with the walls (impulsive component) and part moves together excitations. In addition to the Housner model and the SSM, the
with the free surface (convective component). Notice that Fn = 1 maximum transmitted force is also calculated by the numerical
corresponds to a system for which the sloshing effect is non-impor- procedures described in [32] and by an ANSYS implementation.
tant since all the liquid mass moves together with the walls. In Chen et al. [32] assumed that the modal damping of the liquid is
these systems, the liquid could be modeled by a rigid solid of mass 0.5%, and the same value is adopted here for the Housner and

Fundamental Sloshing Period [s]

6.0 6.0 6.0
5.5 Analitycal Model 5.5 Housner Model 5.5 SSM
2.6
2.6 2.6
2.6 2.6
2.6 2.6
2.6 2.6 2.6 2.6
5.0 22.6
.6
5.0 22.6
.6
5.0
4.5 4.5 4.5
Length L [m]

Length L [m]
Length L [m]

4.0 4.0 4.0

2.6
22..66

22..66

2.2 2.2 2.2 2.2 2.2 2.2 2.2 2.2 2.2

3.5 3.5 3.5
3.0 3.0 3.0
2.2
2.2

2.2

2.5 1.8 1.8 1.8 2.5 1.8 1.8 1.8 2.5 1.8 1.8 1.8

2.0 2.0 2.0

1.8
1.8

1.8

1.5 1.4 1.4 1.4 1.5 1.4 1.4 1.4 1.5 1.4 1.4 1.4

1.0 1.0 1.0

0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
Aspect Ratio R Aspect Ratio R Aspect Ratio R

Fig. 3. Fundamental sloshing period of rectangular tanks.

212 R.O. Ruiz et al. / Engineering Structures 85 (2015) 206–218

Second Sloshing Period [s]

6.0 6.0
Analitycal Model SSM
5.5 5.5
5.0 5.0
4.5 1.4 1.4 1.4 1.4 1.4 1.4
4.5

Length L [m]
Length L [m] 4.0 4.0
3.5 1.2 1.2 1.2 3.5 1.2 1.2 1.2
3.0 3.0
2.5 2.5
1.0 1.0 1.0 1.0 1.0 1.0
2.0 2.0
1.5 0.8 0.8 0.8 1.5 0.8 0.8 0.8
1.0 1.0

0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
Aspect Ratio R Aspect Ratio R

Fig. 4. Second sloshing period of rectangular tanks.

Third Sloshing Period [s]

6.0 6.0
1.2 Analitycal Model 1.2 1.2 SSM
5.5 5.5
5.0 5.0
4.5 4.5
Length L [m]
Length L [m]

4.0 1.0 1.0 1.0 4.0 1.0 1.0 1.0

3.5 3.5
3.0 3.0
2.5 0.8 0.8 0.8 2.5 0.8 0.8 0.8

2.0 2.0
1.5 0.6 0.6 0.6 1.5 0.6 0.6 0.6
1.0 1.0

0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
Aspect Ratio R Aspect Ratio R

Fig. 5. Third sloshing period of rectangular tanks.

SSM models and for the ANSYS implementation. An interesting Five different tank configurations were used in the analysis,
result mentioned by Chen et al. [32] is that the linear theory pre- with each tank identified by a letter and geometry specification
dicts the transmitted force relative well compared to the non-lin- as presented in Table 3. The geometry of tanks A and B were
ear model. The SSM provides good agreements with the Linear obtained from Idir et al. [33] while the geometry of tanks C, D
model and the FEM through ANSYS while the accuracy of the Hous- and E were obtained from Gardarsson [34]. Table 4 presents the
ner model shows a significant excitation dependency. The average comparison between the fundamental sloshing period obtained
computational time required to conduct the time-history analysis by the SSM, ANSYS and the experiments conducted in the afore-
on a PC with a 3.3 GHz Xeon quad processor and 8 GB RAM is mentioned two studies. The period obtained by the SSM and the
0.5 s, 6.8 s and 738 s for the Housner model, the SSM and ANSYS experimental tests are in good agreement for all geometries with
simulation, respectively. This shows that the proposed SSM offers the maximum difference being close to 4%, except for tank A for
high accuracy results with a remarkable computational efficiency. which the difference is higher. Idir et al. [33] also reported signif-
icant differences between the experimental and numerical value
3.1.2. Non-rectangular tanks of the fundamental sloshing period for this tank. A reason of this
Extending now the comparison to non-rectangular tanks, three difference is a potential inaccuracy in the experimental details pro-
different types of tanks (also shown in Fig. 7) are chosen to com- vided. With respect to the computational time required to calcu-
pare the fundamental sloshing period obtained by the SSM with late the sloshing periods presented in Table 4, the slowest
experimental results found on previously published studies calculation time (this corresponds to tank A) is 5.2 s using the
[33,34]. The first one is a W-type tank corresponding to a rectangu- SSM and 78 s using ANSYS. Note that these calculation times do
lar tank with a triangular obstacle located at the bottom, where its not take into account the time used to remesh the domain in the
location and size are defined by a and h. The other two correspond convergence process. This further validates the previous claims
to the U-type and V-type tanks, both corresponding to a tank with for the computational efficiency of the SSM.
sloped walls. The only difference is that the U-type tank satisfies As discussed in the Introduction section this computational effi-
the condition of h < H while the V-type tank satisfies the condition ciency/simplicity of the SSM can be exploited to further perform
of h > H. For all cases, H is the non-perturbed water depth and L is sensitivity studies of such non-rectangular tanks, which is one of
the length of the tank. the motivations for establishing this new approach. An interesting
 R.O. Ruiz et al. / Engineering Structures 85 (2015) 206–218 213

2
10
Housner Model

Normalized Force Fn
SSM
1
10

0
10

−1
10
R = 0.3
−2
10
0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
Period Ratio r
2
10
Normalized Force Fn

1
10

0
10

−1
10
R = 0.5
−2
10
0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
Period Ratio r
2
10
Normalized Force Fn

1
10

0
10

−1
10
R = 1.0
−2
10
0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
Period Ratio r

Fig. 6. Harmonic linear response of the transmitted force to the walls (rectangular tank, L = 9.144 m).

Table 2
Comparison of transmitted force to the ground obtained by different methods.

Earthquake record Maximum transmitted force to the ground (kN)

Non-linear model [32] Linear model [32] Housner model SSM ANSYS
El Centro 52.32 51.40 53.92 50.21 51.72
Mexico City 48.70 47.52 36.20 47.19 47.93

W-Type U-Type V-Type

H
H h H
h h

a a a
L L L

Fig. 7. Description of three different non-rectangular tanks.

Table 3
Details of the tank geometries selected from [33,34].

Tank ID Tank type L (mm) H (mm) a (mm) h (mm) Total water mass (kg)
A W-type 380 76 52.62 50 0.594
B U-type 380 76 137.38 50 0.594
C V-type 590 40 250 144.34 2.134
D V-type 590 70 250 144.34 4.954
E V-type 590 100 250 144.34 8.817
214 R.O. Ruiz et al. / Engineering Structures 85 (2015) 206–218

Table 4 reducing vibration of civil structures [35]. This implementation,

Fundamental sloshing period in non-rectangular tanks (s). corresponding ultimately to a liquid storage tank located on an
Tank ID SSM ANSYS Experimental elastic structure, is examined here. In this case the numerical
A 1.226 1.183 1.024 [33] model needs to capture the fluid–structure interaction with a high
B 0.953 0.984 0.970 [33] level of accuracy. An elastic structure with a TLD on top is thus
C 0.779 0.832 0.7805 [34] studied to determine the effectiveness of the SSM to simulate the
D 0.904 0.976 0.9412 [34] response of this type of systems. The numerical scheme for this
E 1.022 1.115 1.0667 [34]
evaluation was presented earlier in Fig. 2.
The chosen structure corresponds to the linear 9-story
benchmark building presented by Ohtori et al. [36]. This building
related question is to identify how variations of critical character- was first modeled in a commercial software and then the mass
istics affect the fundamental sloshing period. In particular, the and stiffness matrix were condensed at the lateral floor displace-
change of the fundamental sloshing period due to variations in a ments producing a planar model with 9 translational degrees of
and h for a W- and U-type tank (both with L = 380 mm and freedom. The first four natural periods were calculated to
H = 76 mm) is presented in Fig. 8. Results show that the sloshing correspond to 2.10 s, 0.80 s, 0.47 s and 0.32 s. Also a 2% of modal
period of the W-type tank exhibits greater variations in compari- damping is assumed for the structure for all modes of vibration.
son to the U-type tank. Also, a variation of h has greater effect on The equation of motion for the coupled TLD-structure system is
the sloshing period than a variation of a. This discussion shows obtained following the procedure described in Section 2.4, result-
the utility of the established methodology in examining the behav- ing to:
ior of tanks with different geometric characteristics. Once the capa-
" #" #  " _ #  " #
bilities of the SSM to predict the sloshing periods and perform M þ LTs BLs LTs A €
u C 0 u K 0 u
sensibility analysis on non-rectangular tanks are identified, it is þ þ
Rs Ls Ms g
€
s
0 Cs g_
s
0 Ks g s
necessary to study its performance on time-history analysis. "  # ð28Þ
T
The seismic response of tanks A and B is finally examined for M þ Ls BLs Do
¼ €g
u
the same ground motions considered in Section 3.1.1. Again, it is
Rs
assumed that the modal damping of the liquid is 0.5%. In particular,
the time-history of the transmitted force obtained by the SSM and
where M, C and K are mass, damping and stiffness matrix of the
ANSYS are compared in Fig. 9. The maximum transmitted force
structure, respectively, u is the vector with the displacement of
predicted by both methods is practically the same for the cases
each story relative to the ground, Do is the influence coefficient vec-
studied (but with a computational time required by ANSYS greater
tor, in this case corresponding to a nine-dimensional vector (num-
by a factor of 300 compared to the SSM), further validating the
ber of degrees of freedom of the planar structural system) of ones
accuracy of the new model. It is interesting to note that even
and Ls being a vector used to define the location of the TLD within
though both tanks (A and B) have the same volume of water and
the structure (degree of freedom it impacts). The TLD is assumed to
the same external dimensions (H and L), meaning same space
be at the top of the building leading to Ls as a vector of zeros with
requirements, tank A exhibits lower sloshing amplitudes that lead
the 9th element being equal to 1 so that Ls u = u9, where u9 repre-
ultimately to smaller forces acting on the walls. The difference in
sents the lateral displacement of the top floor (which is the floor
the transmitted force is related to the amount of water that moves
that contains the TLD in this case).
(convective mass) and does not move (impulsive mass) inside the
A rectangular tank and a U-type tank are used as TLDs. The
tank, features that ultimately depend on the tank geometry. For
fundamental sloshing period of both tanks is tuned to the same
maximizing (or minimizing, depending on the application of inter-
fundamental period of the structure, while the total mass of water
est) this force an optimization problem can be further formulated
is fixed as 2% of the total mass of the structure. Additionally, the
by changing the tank geometry, something that can be easily facil-
damping matrix Cs is defined so that a 0.5% of modal damping is
itated through the computational efficiency of the SSM.
obtained for the fluid.
The SSM is utilized here to predict the response of the structure
3.2. Tank supported by a linear structure and is compared against the predictions provided by ANSYS or by
the Housner model. Before describing the comparisons the
The SSM can be readily implemented to evaluate the behavior coupling of the latter two models with the structural model is
of Tuned Liquid Dampers (TLD), which are popular devices for briefly discussed. The coupling of the Housner model with the

Sloshing Period [s] Sloshing Period [s]

W Type Tank U Type Tank
0.50 0.50
1.12
1.08 4 0
4

0.45 1.1 1.0 1.0 0.45

0.9

98
1.08 1.06 2
0.
1.0
96

0.40 0.40
0.

1.06 4
1.0 8
0 0.9
0.35 1.0 0.35
1.04 1.02
h/H
h/H

4
0.9

0.30 0.30
1.02 1.00 8 96
0.9 0.
0.25 0.96 0.25
1.00
0.20 0.98 0.20
4

0.98
0.96
9
0.

0.15 0.15
0.96
0.10 0.10
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
a / (L/2) a / (L/2)

Fig. 8. Sensitivity of the sloshing period to a and h in W- and U-type tanks, L = 380 mm and H = 76 mm.
 R.O. Ruiz et al. / Engineering Structures 85 (2015) 206–218 215

Tank A Response
El Centro Earthquake Record
150

Transmitted Force [N/m]

SSM
100 ANSYS

50

−50

−100

−150
0 5 10 15 20 25 30 35 40
Time [s]

Tank B Response
El Centro Earthquake Record
150
Transmitted Force [N/m]

100

50

−50

−100

−150
0 5 10 15 20 25 30 35 40
Time [s]

Fig. 9. Seismic response histories of tanks A and B.

0.4
3rd Floor SSM
0.2 ANSYS

0.0

−0.2

−0.4
0 5 10 15 20 25 30 35 40
Time [s]
0.4
6th Floor
Displacement [m]

0.2

0.0

−0.2

−0.4
0 5 10 15 20 25 30 35 40
Time [s]
0.4
9th Floor
0.2

0.0

−0.2

−0.4
0 5 10 15 20 25 30 35 40
Time [s]

Fig. 10. Displacements response of the 9-floor building with a rectangular TLD.

structure is realized by (28) but taking into account the following Section 3.1, the only difference being that the TLD is now sup-
modifications: (i) the impulsive mass is added to the mass matrix ported by a frame structure which is modeled using the elements
element M9,9 due to the 9th floor which contains the TLD; (ii) B, A BEAM3 and PLANE42. Spurious frequencies are expected in this
and Rs are now scalars and all equal to the convective mass, (iii) gs, case since the formulation is based on displacement variables
Ks and Cs are also scalars, gs represents only the movement of the [12,14] as indicated in [31]. Many spurious frequencies were
convective mass and not the elevation of the free surface, Ks is indeed observed between the first three sloshing frequencies, with
defined by the spring proposed by Housner whereas Cs is similarly the problem ultimately resolved by adopting a modal reduction
defined to correspond to 0.5% of modal damping. For ANSYS, the [29] to reduce the computational burden (i.e., the time needed
characteristics of the analysis are similar to those described in by ANSYS to integrate the equations of motion).
216 R.O. Ruiz et al. / Engineering Structures 85 (2015) 206–218

0.6
3rd Floor SSM
0.4
ANSYS
0.2
0.0
−0.2
−0.4
−0.6
0 5 10 15 20 25 30 35 40
Time [s]
0.6
6th Floor
Acceleration [g]

0.4
0.2
0.0
−0.2
−0.4
−0.6
0 5 10 15 20 25 30 35 40
Time [s]
0.6
9th Floor
0.4
0.2
0.0
−0.2
−0.4
−0.6
0 5 10 15 20 25 30 35 40
Time [s]

Fig. 11. Acceleration response of the 9-floor building with a rectangular TLD.

Table 5
Peak response values of the structure with a rectangular TLD.

Peak displacement (m) Peak absolute acceleration (g)

Without TLD Housner SSM ANSYS Without TLD Housner SSM ANSYS
3rd floor 0.1664 0.1470 0.1452 0.1475 0.4087 0.4351 0.4399 0.4316
6th floor 0.2745 0.2401 0.2394 0.2412 0.5400 0.5198 0.5277 0.5108
9th floor 0.3723 0.3350 0.3321 0.3381 0.7041 0.6510 0.6481 0.6550

0.4
3rd Floor SSM
0.2 ANSYS

0.0

−0.2

−0.4
0 5 10 15 20 25 30 35 40
Time [s]
0.4
Displacement [m]

6th Floor
0.2

0.0

−0.2

−0.4
0 5 10 15 20 25 30 35 40
Time [s]
0.4
9th Floor
0.2

0.0

−0.2

−0.4
0 5 10 15 20 25 30 35 40
Time [s]

Fig. 12. Displacements response of the 9-floor building with a U-type TLD.
 R.O. Ruiz et al. / Engineering Structures 85 (2015) 206–218 217

0.6
3rd Floor SSM
0.4
ANSYS
0.2
0.0
−0.2
−0.4
−0.6
0 5 10 15 20 25 30 35 40
Time [s]
0.6
6th Floor
Acceleration [g]

0.4
0.2
0.0
−0.2
−0.4
−0.6
0 5 10 15 20 25 30 35 40
Time [s]
0.6
9th Floor
0.4
0.2
0.0
−0.2
−0.4
−0.6
0 5 10 15 20 25 30 35 40
Time [s]

Fig. 13. Acceleration response of the 9-floor building with a U-type TLD.

3.2.1. Rectangular tuned liquid damper

Table 6
As illustrated in Fig. 3 earlier it is possible to find many different Peak response values of the structure with a U-type TLD.
combinations of R and L that result in the same fundamental slosh-
Peak displacement (m) Peak acceleration (g)
ing period. For this illustrative application the dimensions of the
tank are chosen as R = 0.6 and L = 3.29 m, satisfying the tuning Without SSM ANSYS Without SSM ANSYS
TLD TLD
requirements to the fundamental structural period. The building
is subjected to the El Centro ground motion and results are pre- 3rd floor 0.1664 0.1458 0.1495 0.4087 0.4395 0.4292
sented for the response of floors 3, 6 and 9 of the structure. The 6th floor 0.2745 0.2408 0.2447 0.5400 0.5254 0.5108
9th floor 0.3723 0.3327 0.3483 0.7041 0.6503 0.6571
floor displacements and absolute accelerations are presented in
Figs. 10 and 11, respectively, where the green line is obtained by
solving (28) and the red line is obtained by using ANSYS. The
results show that the TLD modeled by the SSM or ANSYS generate problem (TLD and building together), the fluid–structure analysis
similar responses. Table 5 then presents the floor peak response using the SSM requires to have the structure already modeled
values (accelerations and displacements) obtained by the different (on the implementation, the structural model of the building is
methods examined. All approaches yield similar results, verifying taken as an input and not as part of the calculation).
their accuracy, whereas in general good efficiency is obtained by Overall the discussion shows the computational efficiency of
incorporation of the TLD in reducing the structural response (com- the proposed method for evaluating the behavior of structures
pare the results with and without the TLD). It should be stressed equipped with TLDs.
that the differences observed between the ANSYS and the SSM
may be attributed to the contribution of higher sloshing modes 4. Conclusions
that cannot be taken into account in the ANSYS model due to the
modal reduction implemented. In this study, a numerical procedure (SSM) to represent the lin-
ear sloshing of liquid storage tanks was described and validated.
3.2.2. Non-rectangular tuned liquid damper The proposed model is based on the potential flow theory, where
The same analysis is implemented for the structure equipped the Continuity equation is resolved by a FEM scheme and then
with a U-type TLD with dimensions of R = 0.625, L = 3.29, a = 0.5/ combined with the Equilibrium equation generating a second order
(L/2), h = 0.3H. The time history is only compared with the results linear system of equations. The implementation of the new
obtained by ANSYS, Figs. 12 and 13 (the Housner model is not approach requires simply the computation of the equivalent fluid
applicable for such tank geometries). Results are reported in Table 6 mass, damping and stiffness matrices additionally to the computa-
which shows good agreement between the SSM and ANSYS. More- tion of the terms related to the transmitted force that finally allows
over, the behavior estimation of the building with a rectangular the numerical coupling with the support structure.
TLD and non-rectangular TLD are similar since the geometries Contrary to the Housner model, that represents the sloshing
are not significantly different. For both TLD geometries, the SSM only through a single degree of freedom while, the SSM approxi-
estimates 10% on peak displacement reduction and 8.0% on peak mates it as a full system of n degrees of freedom facilitating a more
acceleration reduction when the time-histories of the building accurate description (higher modes of vibration can be accounted
with and without TLD are compared. For this case, comparative for). For a rectangular tank sloshing periods obtained by the SSM,
analysis in terms of computational time are not presented since by the Housner model and by analytical approximations have a
the fluid-interaction using the SSM and ANSYS operate under maximum difference of 1.0% (for different tank geometries). When
different conditions: while ANSYS applied the FEM to the whole looking at the response under harmonic excitations the Housner
218 R.O. Ruiz et al. / Engineering Structures 85 (2015) 206–218

and SSM models were similar only if the excitation period was [8] American Petroleum Institute (API). Welded storage tanks for oil storage, API
650. Washington (DC): American Petroleum Institute Standard; 2003.
higher than 0.8 times the fundamental sloshing period. This should
[9] American Water Works Association (AWWA). Welded steel tanks for water
be attributed to the inability of the Housner model to represent storage. AWWA D-100. Denver (CO): American Water Works Association;
higher modes of vibration. Under seismic excitations the examined 2005.
methods (including a full FEM implementation through ANSYS) [10] Priestley MJN, Davison BJ, Honey GD, Hopkins DC, Martin RJ, Ramsey G, et al.
Seismic design of storage tanks, recommendations from the New Zealand
were found to yield comparable results, whereas similar trends National Society for Earthquake Engineering. Wellington, New Zealand: New
were also found for non-rectangular tanks. Zealand National Society for Earthquake Engineering; 1986.
In addition, the utility of the proposed model to evaluate the [11] Love JS, Tait MJ. Equivalent linearized mechanical model for tuned liquid
dampers of arbitrary tank shape. J Fluid Eng 2011;133(6). 061105 (9pp).
performance of rectangular and non-rectangular Tuned Liquid [12] Bermudez A, Duran R, Rodriguez R. Finite element solution of incompressible
Dampers (TLDs) was examined. For this purpose the seismic fluid–structure vibration problems. Int J Numer Meth Eng
response of a 9-story building equipped with a TLD was simulated 1997;40(8):1435–48.
[13] Chen HC, Taylor RL. Vibration analysis of fluid–solid systems using a finite
through ANSYS and SSM, with results showing good agreements element displacement formulation. Int J Numer Meth Eng 1990;29(4):683–98.
since the maximum differences were lower than 3%. [14] Kiefling L, Feng GC. Fluid–structure finite element vibrational analysis. AIAA J
In general, the proposed model was validated in terms of slosh- 1976;14(2):199–203.
[15] Bermudez A, Rodriguez R. Finite element computation of the vibration modes
ing period estimation, harmonic and seismic response of rectangu- of a fluid–solid system. Comput Methods Appl Mech Eng 1994;119(3–
lar and non-rectangular tanks. The SSM offers a more accurate 4):355–70.
representation of the sloshing dynamic than the Housner model [16] Mellado M, Rodriguez R. Efficient solution of fluid–structure vibration
problems. Appl Num Math 2001;36(4):389–400.
but also a faster model and free of inconsistences (that is, presence
[17] Olson LG, Bathe K-J. Analysis of fluid–structure interactions. A direct
of spurious frequencies) than the ANSYS simulation. Ultimately, symmetric coupled formulation based on the fluid velocity potential.
the SSM establishes a simple, (second order lineal system of equa- Comput Struct 1985;21(1–2):21–32.
tions that can be interpreted as a mass-spring system), generalized [18] Morand H, Ohayon R. Substructure variational analysis of the vibration of
coupled fluid–structure systems. Finite element results. Int J Numer Meth Eng
(any tank geometry) mathematical modeling of the liquid sloshing 1979;14(5):741–55.
based on variables that have physical meaning (free surface eleva- [19] Bermudez A, Duran R, Rodriguez R. Finite element analysis of compressible
tion and ground acceleration), and facilitates an easy coupling and incompressible fluid–solid systems. Math Comput 1998;67(221). 111-
111.
between rigid tanks and elastic structures that support them. Fur- [20] Kock E, Olson L. Fluid-structure interaction analysis by the finite element
thermore, its high computational efficiency and accuracy make it a method. A variational approach. Int J Numer Meth Eng 1991;31(3):463–91.
valuable tool that could be used by designers to conduct paramet- [21] Hernandez-Barrios H, Heredia-Zavoni E, Aldama-Rodriguez AA. Nonlinear
sloshing response of cylindrical tanks subjected to earthquake ground motion.
ric studies, preliminary dimensioning of storage tanks, seismic per- Eng Struct 2007;29(12):3364–76.
formance identification, or even preliminary design and [22] Frandsen JB, Borthwick AGL. Simulation of sloshing motions in fixed and
dimensioning of tuned liquid dampers. vertically excited containers using a 2-D inviscid -transformed finite difference
solver. J Fluids Struct 2003;18(2):197–214.
[23] Ali Goudarzi M, Reza Sabbagh-Yazdi S. Investigation of nonlinear sloshing
Acknowledgements effects in seismically excited tanks. Soil Dynam Earthquake Eng
2012;43:355–65.
[24] Ikeda T, Ibrahim RA, Harata Y, Kuriyama T. Nonlinear liquid sloshing in a
Financial support was provided by the Pontificia Universidad square tank subjected to obliquely horizontal excitation. J Fluid Mech
Catolica de Chile, by the Department of Civil & Environmental Engi- 2012;700:304–28.
[25] Wang W, Li J, Wang T. Damping computation of liquid sloshing with small
neering & Earth Sciences, University of Notre Dame, and by the amplitude in rigid container using FEM. Acta Mechanica Sinica/Lixue Xuebao
National Research Center for Integrated Natural Disaster Manage- 2006;22(1):93–8.
ment CONICYT/FONDAP/15110017 (Chile). This financial support [26] Reddy JN. An introduction to the finite element method. 3rd ed. New
York: McGraw Hill; 2005.
is gratefully acknowledged. [27] Komatsu K. Non-linear sloshing analysis of liquid in tanks with arbitrary
geometries. Int J Non-Linear Mech 1987;22(3):193–207.
[28] Lee DT, Schachter BJ. Two algorithms for constructing a Delaunay
References triangulation. Int J Comput Inform Sci 1980;9(3):219–42.
[29] Clough Ray, Penzien J. Dynamics of structures. 2nd ed. New York: McGraw
[1] Zareian F, Sampere C, Sandoval V, McCormick DL, Moehle J, Leon R. Hill; 1993.
Reconnaissance of the Chilean wine industry affected by the 2010 Chile [30] Dormand JR, Prince PJ. Family of embedded Runge–Kutta formulae. J Comput
offshore Maule earthquake. Earthquake Spectra Jun. 2012;28:503–12. Appl Math 1980;6(1):19–26.
[2] Tang AK, Eng P, Eng C, Asce F. Lifelines performance of the Mw 8.8 off shore [31] R. ANSYS. Release 11.0 Documentation for ANSYS; 2007.
Biobio, Chile earthquake. In: 12th East Asia-Pacific conference on structural [32] Chen W, Haroun MA, Liu F. Large amplitude liquid sloshing in seismically
engineering and construction, EASEC12, January 26, 2011–January 28, 2011, excited tanks. Earthquake Eng Struct Dynam Jul. 1996;25(7):653–69.
2011, vol. 14. p. 922–30. [33] Idir M, Ding X, Lou M, Chen G. Fundamental frequency of water sloshing waves
[3] Buratti N, Tavano M. Dynamic buckling and seismic fragility of anchored steel in a sloped-bottom tank as tuned liquid damper. In: Proceeding of the 2009
tanks by the added mass method. Earthquake Eng Struct Dynam structures congress; 2009. p. 831–40.
2014;43(1):1–21. [34] Gardarsson SM. Shallow-water sloshing. PhD thesis. University of
[4] Ibrahim RA. Liquid sloshing dynamics: theory and applications. 1st Washington; 1997.
ed. Cambridge University Press; 2005. [35] Maravani M, Hamed MS. Numerical modeling of sloshing motion in a tuned
[5] Hoskins LM, Jacobsen LS. Water pressure in a tank caused by a simulated liquid damper outfitted with a submerged slat screen. Int J Numer Meth Fluids
earthquake. Bull Seismol Soc Am 1934;24. 2011;65(7):834–55.
[6] Housner G. Dynamic pressure on accelerated fluid containers. Bull Seismol Soc [36] Ohtori Y, Christenson RE, Spencer BF, Dyke SJ. Benchmark control problems for
Am 1957;47:15–35. seismically excited nonlinear buildings. J Eng Mech – ASCE
[7] Housner G. The dynamic behavior of water tanks. Bull Seismol Soc Am 2004;130(4):366–85.
1963;53(2):381–7.