EARTHQUAKE ENGINEERING AND STRUCTURAL DYNAMICS

Earthquake Engng Struct. Dyn. (2009)

Published online in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/eqe.945

Experimental investigation on the seismic response of a steel liquid

storage tank equipped with floating roof by shaking table tests

M. De Angelis1 , R. Giannini2 and F. Paolacci2, ∗, †

1 Department of Structural and Geotechnical Engineering, University of Rome ‘La Sapienza’, Italy
2 Department of Structures, University of Roma Tre, Italy

SUMMARY
In this paper, the effectiveness of the base isolation on steel storage tanks has been investigated through
numerical models and then checked by shaking table tests on a reduced scale (1:14) model of a real
steel tank, typically used in petrochemical plants. In the experimental campaign the floating roof has also
been taken into account. The tests have been performed on the physical model both in fixed and isolated
base configurations; in particular two alternative base isolation systems have been used: high-damping
rubber bearings devices and sliding isolators with elasto-plastic dampers. Finally, a comparison between
experimental and numerical results has also been performed. Copyright q 2009 John Wiley & Sons, Ltd.

Received 7 November 2008; Revised 16 June 2009; Accepted 20 June 2009

KEY WORDS: steel storage tanks; floating roof; shaking table test; base isolation

1. INTRODUCTION

One of the causes of the most serious accidents occurring in the past to major-hazard industrial
plants is represented by seismic action. In fact, unlike accidents caused by internal events, which
generally happen in specific points of the plant and, at the limit, during an earthquake might be
extended to a limited surrounding region, the input could cause the simultaneous damaging of
different apparatus of the installation, whereas the effects can be amplified by the failure of the
safety systems or by the simultaneous occurrence of numerous accidental chain.
Among the different types of industrial installations, except for the nuclear plants, the process
industry is one of the most vulnerable groups of apparatus, which has to be treated with particular
attention for several reasons: the large number of installations in the national territories, the
important allied industries (petrochemical poles), the frequent location in high seismic hazard

∗ Correspondence to: F. Paolacci, Department of Structures, University of Roma Tre, Via C. Segre 6, 00146 Rome, Italy.
†
E-mail: paolacci@uniroma3.it

Contract/grant sponsor: ISPESL-D.P.I.A.; contract/grant number: B106/DIPIA/03

Copyright q 2009 John Wiley & Sons, Ltd.

 M. DE ANGELIS, R. GIANNINI AND F. PAOLACCI

Figure 1. Tupras refinery: (a) Tanks farm and (b) fire during Koaceli Earthquake ’99 [7].

areas, the seismic vulnerability demonstrated in several occasions [1–4], (Figure 1(a)) and finally
the consequences, often serious, of losses of dangerous substances (inflammable, explosive, toxic).
On the basis of the above-mentioned observations, the importance of using techniques for
reducing the seismic risk of new or existing plants appears clearly.
Among the techniques employed for reducing the seismic action, the base isolation systems
with HDRB or SIEPD devices are those most widely used; the first is usually adopted for the base
isolation of buildings, while the second one is frequently employed for the seismic isolation of
bridges. On the contrary, similar methods have been used, until now, for a very limited number of
industrial applications; for example, in Europe the isolation technique has been adopted only in a
few cases: the seismic protection of Petrochemical LNG terminal of Revythousa, Greece and of
ammoniac tanks, at Visp, in Switzerland [5, 6].
The Department of Structures of the University of Roma Tre and the Department of Structural
and Geotechnical Engineering of the University of Rome ‘La Sapienza’ are involved in a research
project, funded by the Italian National Institute for Occupational Safety and Prevention (ISPESL),
whose aim is study of the applicability of the base isolation technique in petrochemical plants
exposed to seismic risk. Among industrial components, the big steel tanks (Figure 1(b)) for
the storage of petroleum and its products have been taken into account for their high seismic
vulnerability and suitability for the application of base isolation technique. In fact, in the recent
past (e.g. Izmit earthquake in 1999, in Turkey) most of this kind of tanks have been destroyed,
especially as consequence of fires caused by earthquakes [7].
In this paper, the effectiveness of the base isolation on steel storage tanks has been investigated
through numerical models and then checked by shaking table tests on a reduced scale (1:14)
physical model of a real steel tank (diameter 55 m, height 15.6 m), typically used in petrochemical
plants. In the experimental campaign the floating roof has also been taken into account. In practice,
in order to evaluate the effect of the sloshing, many seismic design codes adopt the formula relative
to free surface conditions [8]. This is justified by theoretical studies [9], where it is shown that
the effect of the floating roof has a negligible influence on the frequency and the amplitude of the
first vibration mode and it is confirmed by the experimental study presented in the following.
The tests have been carried out using the six d.o.f. 4×4 m shaking table installed in the laboratory
of ENEA (Italian National Agency for New Technologies, Energy and the Environment) Research
Centre ‘La Casaccia’ at Rome, Italy. The tests have been performed on the physical model both
in fixed and isolated base configurations; in particular two alternative base isolation systems have

Copyright q 2009 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2009)
DOI: 10.1002/eqe
 THE SEISMIC RESPONSE OF A STEEL LIQUID STORAGE TANK

been used: high-damping rubber bearings devices (HDRB) and PTFE-steel sliding isolation devices
with c-shaped elasto-plastic dampers (SIEPD).
In the following, after a brief presentation of the dynamic behavior of tanks, with and without
base isolation systems, the main results of the experimental tests are shown and discussed. Finally,
a comparison between experimental and numerical results has been illustrated.

2. DYNAMICS OF CYLINDRICAL LIQUID STORAGE TANKS

2.1. Fixed base tanks

The dynamics of cylindrical tanks subjected to a base motion has been extensively studied by
several authors. Starting from the earliest work of Housner [10], the hydrodynamic pressure,
induced by the liquid to the tank wall due to the base motion, has been determined taking into
account the deformability of the tank wall; see for example [11–13].
Under the hypotheses of inviscid, incompressible and irrotational liquid, the problem is governed
by the continuity equation:
 = 0 (1)
in which  is the velocity potential, such that
grad  = v (2)
with v the velocity of the fluid. Two types of boundary conditions have to be added to
Equation (1): (i) at the contact between fluid and structure, the velocity of the liquid along the
normal of the wall surface has to be equal to the velocity of the wall along the same direction
(Figure 2)
*
= vn (3)
*n
and (ii) on the free liquid surface, where the pressure of the fluid has to be equal to the atmospheric
pressure (that can be taken as zero for simplicity). It can be shown that ( is the fluid density)
*
p = − (4)
*t
Therefore if  is the height of the wave on the free surface of the liquid, the condition (ii), for
waves of a height lower than the liquid height H , can be expressed equating the pressure at z = H

Figure 2. Geometric and kinematic parameters of the tank model.

Copyright q 2009 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2009)
DOI: 10.1002/eqe
 M. DE ANGELIS, R. GIANNINI AND F. PAOLACCI

Figure 3. (a) Impulsive pressure; (b) convective pressure; and (c) flexible (continuous lines) versus
impulsive (dotted lines) pressure.

with the weight of a liquid column of height . Taking into account Equation (4) and deriving
both the sides, it follows that:

* *  ∼ * 
2 2
* 
g + 2 =g  + 2 =0 (5)
*t *t *z z=H *t
The solution of Equations (1), (3) and (5) can be conveniently divided into three components,
each one solution of Equation (1):
• fluid pressure due to the ground acceleration (considering the tank wall as being rigid), named
‘impulsive pressure’; in this case vn is the velocity impressed to a rigid body by the ground
motion, whereas Equation (5) is substituted by the simpler equation *(H )/*t = 0;
• fluid pressure, named ‘flexible pressure’, caused by the deformability of the tank wall; it is
obtained assuming for vn the velocity of the wall excited by the impulsive pressure component,
the pressure generated by the wall deformations and the inertia forces of the wall mass;
• fluid pressure, named ‘convective’ or ‘sloshing pressure’ obtained by imposing zero velocity
on the wall (vn = 0), and the satisfaction of the (5) on the free liquid surface, where  is the
sum of the impulsive and convective components of the velocity potential.
In the above-mentioned concerns, the interaction between the motion due to the deformability of
the wall and the convective motion has been neglected, since their frequencies are very different
and thus their mutual influence is limited [14].
For cylindrical tanks subjected to a base motion, the impulsive and convective solutions of
motion may be expressed in closed form (see for example [13]). In Figures 3(a) and (b) the
pressures, generated along the height by these two components are shown for different values of
height–radius ratio (H/R). The pressure is normalized with respect to fluid density , radius R
and ground acceleration ag .
In order to determine the pressure due to the deformability of the wall (second component), a
fluid–structure interaction problem has to be solved. The solution depends on the geometrical and
mechanical characteristics of the tank: dimensions, thickness, fluid density and elastic modulus.
The problem is uncoupled into infinite vibration modes, a few of which have a significant mass.

Copyright q 2009 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2009)
DOI: 10.1002/eqe
 THE SEISMIC RESPONSE OF A STEEL LIQUID STORAGE TANK

Figure 4. (a) Percentage of mass relative to the impulsive (continuous line) and convective motion (dashed
line). The bottom line (Total) is the sum of the four terms and (b) comparison between impulsive
(continuous line) and flexible (first and second mode) masses.

In Figure 3(c) a comparison between the first and second component of the pressure is shown.
It is evident that in slender tanks the distributions along the height are very different, whereas for
broad tanks are practically coincident.
In Figure 4(a) the normalized masses of the impulsive and the first three vibration modes of
the convective pressure are represented versus H/R. The percentage of mass associated with the
higher modes of the convective pressure is very small (<3%); in addition, while in the slender
tank a large amount of liquid mass moves rigidly with the tank, in the broad tanks a large part of
the mass oscillates in the convective motion.
The flexible mass is also distributed among the different vibration modes of the tank wall. In
Figure 4(b) the normalized values of the mass of the first two vibration modes is shown versus
the impulsive mass (continuous line).
In the broad tanks (H/R1) the entire mass is practically coincident with the mass of the first
vibration mode; moreover, as already shown, the pressure distribution is the same as for the rigid
case. In fact, in the slender thanks, the mass associated with the higher mode is significant and the
pressure distribution of the first vibration mode is significantly different from that of the impulsive
pressure.
The above results show that the study of the dynamic pressure in a tank subjected to a seismic
motion may be reduced to the study of the simpler model with lumped masses and springs,
represented in Figure 5(a). m i is the impulsive mass and m ik is the mass of the kth vibration
mode of the tank wall; whereas m ck is the mass of the kth convective mode. The total pressure is
obtained by summing the effects of the mass m i subjected to ground acceleration, the masses m ik
subjected to the relative acceleration of the oscillators with stiffness kih = m ik 2ik and masses m ck
subjected to the absolute acceleration of oscillators with stiffness kck = m ck 2ck .
For broad tanks the model of Figure 5(a) can be simplified, as illustrated in Figure 5(b). In fact,
the contribution of the higher vibration modes of the wall may be neglected and the entire mass is
practically coincident with the mass of the first mode; moreover, the distributions of the impulsive

Copyright q 2009 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2009)
DOI: 10.1002/eqe
 M. DE ANGELIS, R. GIANNINI AND F. PAOLACCI

Figure 5. Lumped mass models: (a) fixed base tank (general); (b) fixed base broad
tank; and (c) isolated base tank.

and flexible pressure are almost coincident. Therefore, the effects of the impulsive and flexible
component can be globally evaluated in terms of absolute acceleration of a simple oscillator with
mass m i and stiffness ki1 . If the higher modes of the convective motion are also neglected, the
dynamics of a broad tank could be studied by mean of two simple oscillators of frequency c and
i , where:
  
g H
c = 1.841 tanh 1.841 (6a)
R R

C Es
i ≈ (6b)
H R

While the expression of c derives exactly from the previous equations, the expression of i
is approximated [8];  is the fluid density, s is the thickness of the tank wall and E is the young
modulus of the material of which the tank wall is made; the coefficient C depends on the ratio
H/R and, for usual values, varies between 0.17 and 0.13.
The difference between the low frequencies of the convective motion and the high frequencies
of the wall vibrations (the ratio is between 20 and 30) justifies the usual hypothesis of neglecting
the interaction between these two phenomena.
The resultant forces on the tank wall can be easily calculated integrating the pressure on the
contact surface between tank and liquid. These may be expressed as follows:
Q j k (t) = Mm j k A j k (t) (7)
in which M is total mass of the liquid, j = 0, 1, 2 indicates the impulsive, flexible and convective
pressure components, k1 indicates the vibration mode, whereas m j k is the percentage of the
mass of vibration mode k. For the impulsive component A j k (t) is the acceleration of the tank base,
whereas for the flexible and convective components it is, respectively, the relative and absolute

Copyright q 2009 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2009)
DOI: 10.1002/eqe
 THE SEISMIC RESPONSE OF A STEEL LIQUID STORAGE TANK

accelerations of the oscillators of Figure 5(a). For simplicity, hereafter the sum of the impulsive
and flexible contribution will be termed as ‘impulsive’.

2.2. Base-isolated tanks

The idea of seismic protection of tanks and similar industrial structures through the base isolation
technique is not new. Starting from 1990 many works on this subject have been presented [15–17].
Unfortunately, few practical applications have been realized [5] and a limited number of experi-
mental activities have been carried out [18, 19].
On the basis of the above-mentioned observations, a dynamic model of a base-isolated tank
can be easily built. A simple model of broad tanks with base isolation is shown in Figure 5(c).
It has been derived from the model of Figure 5(b) adding a Kelvin unit with stiffness kiso and
damping coefficient ciso representative of the base isolation system. Moreover, two dashpots with
constants cc and ci have been added to account for the energy dissipation of impulsive and
convective motions. The equivalent Kelvin model of the base isolation devices has been here used
only for design purposes, whereas the real nonlinear behavior has been taken into account in the
numerical–experimental comparison discussed in Section 5.
The vibration period of the impulsive component of pressure generally falls in the maximum
amplification field of the response spectra. This implies a high effectiveness of the base isolation
system, which can be profitably used to reduce the base shear due to the impulsive pressure
component.
On the contrary, the periods of the convective motion fall in a very different range. Apart from
the case of small
√ tanks, the oscillation period is much greater than the isolation period (3–12 s). In
fact, Tc ≈ 1.48 R (R in meter); thus for R>7 m, the period is greater than 3.9 s. Therefore, from
the base isolation point of view, the tanks are in an atypical situation with respect to the usual
structures: the isolation period falls between the period of the two main vibration periods of the
structure (flexible and convective motion). Under these conditions, the natural vibration period of
an isolated broad tank is approximatively given by

Mi + Mw + Mb
Tiso ≈ 2 (8)
kiso

in which Mi is the impulsive part of the liquid mass, Mw and Mb are, respectively, the wall and
base tank masses. Mi is a relatively small part of the total mass and this allows a significant
reduction of the devices’ dimensions. Moreover, in the case of big tanks, for which Tc  Tiso , the
first period of the convective motion is practically unmodified. Consequently, the base isolation
system does not show any important mitigation effects on the sloshing pressures. This is not
relevant, since the convective pressure is very small because of the long period of the fluid
oscillations.
The negative effect of the sloshing is related only to the free surface motion, because either the
height of the wave may exceed the upper limit, causing overtopping phenomenon, or the floating
roof motion may cause a breaking of the gaskets and the leakage of dangerous vapors of inflammable
substances. Unfortunately, the base isolation technique does not modify this phenomenon.
Moreover, the base isolation can cause high displacements between tank and ground, which
may induce dangerous damage to the pipes–tank connections.

Copyright q 2009 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2009)
DOI: 10.1002/eqe
 M. DE ANGELIS, R. GIANNINI AND F. PAOLACCI

3. SHAKING TABLE TESTS

3.1. Description of the physical model

The full-scale structure considered here is a big steel liquid storage tank equipped with metallic
floating roof typically installed in petrochemical plants. A sketch and the main dimensions of the
tank are shown in Figure 6.
The real dimensions of industrial tanks do not allow the dynamic tests of natural-scale models
using shaking table. For this reason, it has been necessary to realize a reduced-scale model capable
of simulating in a satisfactory way the behavior of the real structure.
The dependency of the natural frequencies of convective and flexible motions on the characteristic
of the tank can be easily deduced from Equations (6a) (6b). Equation (6b) shows that using the
same material, the natural period of the flexible motion is proportional to the geometrical-scale
factor, whereas from Equation (6a) the dependency of the natural period of the convective motion
on the square root of the geometrical scale factor is clearly evident. This second observation is
justified by the gravitational nature of the convective motion [14].
Thus, a correct dynamic equivalence between reduced- and full-scale tanks cannot be obtained
using geometrical scaling only. A correct scaling could be achieved, in fact, by varying the elastic
modulus of the tank wall or varying the liquid density. In any case some physical limits exist. For
example the wall thickness cannot be reduced beyond a low limit (lateral instability). Moreover,
the use of liquid other than water could be too expensive because of the high volume of liquid to
be used.
On the basis of the above considerations only the scaling with respect to the sloshing motion
has been adopted here, which corresponds to a scaling factor (in terms of times) equal to
T = 0.27. The geometrical scaling factor ( L = 13.7) indicated in Table I has only been used
for the diameter and the height of the tank, which are in the right geometrical proportion with
the real tank, whereas the thickness of the wall does not respect the scale ratio. Consequently,
the dimensions of the mock-up are: diameter D = 4 m, height Hw = 1.45 m and wall thickness
s = 1 mm (Figure 7(a)). The bottom is realized with a steel plate having a thickness of 25 mm,
connected to the wall by bolts and stiffened using a grid of four steel beams welded to the bottom
plate. These beams are supported by four short steel columns 25 cm high. The liquid used in the
experimental campaign is water with a fill level of H = 1 m; the total mass of the liquid is then
about 13 t.
During the experimental campaign several configurations have been analyzed, some of which
were equipped with floating roof (Figure 7(b)). The latter has been realized using a wood grid

Figure 6. Sketch of full-scale tank and main dimensions.

Copyright q 2009 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2009)
DOI: 10.1002/eqe
 THE SEISMIC RESPONSE OF A STEEL LIQUID STORAGE TANK

Table I. Geometrical characteristics of the mock-up.

Geometrical-scale factor L 13.70
Time-scale factor T 0.27
Diameter (m) D 4.00
Wall thickness (mm) s 1
Steel plate thickness (mm) b 25
Height (m) Hw 1.45
Maximum filling level (m) H 1.00
Liquid density (kg/m3 )  1000
Total mass of liquid (kg) 13 000
Elastic modulus of the tank (MPa) E 200 000
Floating roof+bottom plate mass (kg) mt 120
Weight of the bottom beams (kg) 180

Figure 7. Images of the mock-up: (a) without floating roof and (b) with floating roof.

Figure 8. Floating roof: (a) wall connection in a real tank and (b) mock-up roof.

structure with plywood layer, whose mass is about 127 kg. In order to simulate the mechanical
behavior of the roof–wall connection (see Figure 8(a)), a rubber tube was placed along the perimeter
of the roof, as shown in Figure 8(b).

Copyright q 2009 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2009)
DOI: 10.1002/eqe
 M. DE ANGELIS, R. GIANNINI AND F. PAOLACCI

Table II. Design parameters of the isolation devices.

Isolation mass M (kg) 7509
Design soil type C
Design acceleration (g) 0.35
Importance factor 1.4
Local amplification factor 1.25
Isolation period for HDRB (s) 0.756
Equivalent damping ratio for HDRB (%) 10
Friction coefficient for SMD (%) 2

Table III. Characteristics of HDRB isolators.

Diameter  Rubber thickness Steel plate thickness Total height H Rubber modulus G
(mm) (mm) (mm) (mm) (MPa)
90 23.8 6.2 30 0.4

Table IV. Parameters of SIEPD isolators.

R bmax S L E 0 y
(mm) (mm) (mm) (mm) (MPa) (degree) (MPa)
50.66 17 8 90 2 06 000 −27◦ 355

3.2. Design of base isolation systems

For the seismic protection of the mock-up two isolation systems have been used: HDRB and PTFE-
steel isolation devices with metallic c-shaped dampers (SIEPD). The global design parameters
are presented in Table II. Both the base isolation systems have been designed by choosing an
appropriate isolation period (Tiso ) and then evaluating the stiffness kiso using Equation (8). For
HDRB the damping ratio
iso was assumed to be equal to 10%, whereas the yielding force of the
metallic dampers has been designed using the method proposed in [20]. According to Equation (8)
the isolation mass M is equal to 7509 kg, the main characteristics of the devices are summarized
in Tables III and IV and shown in Figures 9(a), (b).
The prototypes of the isolator devices, properly realized for the experimental activity by
the Company Alga Spa (Milan, Italy), are shown in Figures 9(c) and (d). For each isolator
typology four devices have been used, which are placed at the cross of the bottom metallic beams
(Figures 9(e) and (f)).
The mechanical behavior of the isolators has been characterized by cyclic-imposed displacement
tests carried out in the experimental laboratory of the University of Roma Tre using a 500 kN
oleodynamic MTS universal machine. Moreover, in order to reproduce the gravitational load
conditions of the isolator during the shaking table tests, a transversal prestressing force of 40 kN
has been applied.
In Figures 9(g), (h) the experimental cycles obtained, for the HDRB and SIEPD systems,
respectively, are shown. The high dissipation capability of the sliding isolators is clearly apparent.

Copyright q 2009 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2009)
DOI: 10.1002/eqe
 THE SEISMIC RESPONSE OF A STEEL LIQUID STORAGE TANK

Figure 9. Sketch (a–b), in situ placement (c–d), in-plane arrangement (e–f), experimental cyclic response
(g–h) of HDRB and SIEPD isolators, respectively.

Copyright q 2009 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2009)
DOI: 10.1002/eqe
 M. DE ANGELIS, R. GIANNINI AND F. PAOLACCI

Table V. Characteristics of the natural records used during the experimental tests.
Accelerogram Magnitude Epicenter distance (km) PGA (g) Duration (s)
Irpinia, Italy, 11/23/80,Sturno, 270 6.5 32 0.313 40.00
Duzce, Turkey,11/12/99, Duzce, 180 7.2 8.2 0.482 25.88
Kocaeli, Turkey, 08/17/99, Arcelik, 090 7.4 17.0 0.149 28.00
Chi-Chi, Taiwan, 09/20/99, TCU120, W 7.3 8.1 0.268 90

The strong control on the maximum force of this device is also noticed. On the contrary, the rubber
isolator shows a quasi-elastic behavior, together with a reduced energy dissipation capability.

3.3. Test set-up

A series of dynamic tests have been carried out on the tank using the shaking table installed at the
Research Center of ENEA ‘La Casaccia’ (Rome). The mock-up has been tested in four different
configurations: fixed base tank without floating roof (case A), fixed base tank with floating roof
(case B), isolated tank with HDRB and floating roof (case C), isolated base tank with SIEPD
devices and floating roof (case D).
Six different base motion histories have been used in each configuration (four natural and
two synthetic accelerograms, generated by Simqke [21], according to the European code spectra
(soil C), scaled to different intensity levels). For all accelerograms the time scale has been changed
according to the indications of Section 3.1. The natural accelerograms have been selected from the
Pacific Earthquake Engineering Research center database, between time histories recorded for soil
C and generated by seismic events with magnitude lower than 8 and epicenter distance lower than
50 km. Another more selective criterion, adopted here, consists of a selection of signals filtered
using a high-pass filter with cut-off frequency greater than 0.1 Hz. As already seen, this is almost
the frequency of the sloshing motion of the full-scale tank. The records used during the tests are
reported in Table V. The tank has been also tested using white-noise and harmonic signals with
variable frequency (sine-sweep) for the identification of its dynamic characteristics.
The response signals were measured using numerous sensors: pressure transducers, strain-gauges
and laser transducers placed on the tank wall. Laser transducers have also been used for the
sloshing motion of the liquid or floating roof, whereas the motion of the table has been monitored
by several accelerometers. In the isolated base configurations, the motion of the structure with
respect to the base has been measured by wire LVDT sensors. The arrangement of the sensors
has changed between the several series of tests. In Figure 10 the arrangement of the main sensors
used in the isolated base configuration is shown.

4. ANALYSIS OF RESULTS

4.1. Identification of dynamic characteristics

Before evaluating the seismic effects on the tanks, the main dynamic characteristics (frequencies
and damping of the main vibration modes) have been identified. The frequencies of the systems
have been determined by means of transfer functions between input (base acceleration) and output
signals (liquid pressure, wall deformations and displacements, vertical displacements of the free

Copyright q 2009 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2009)
DOI: 10.1002/eqe
 THE SEISMIC RESPONSE OF A STEEL LIQUID STORAGE TANK

Figure 10. Arrangement of the sensors.

Figure 11. Transfer functions: (a) pressure transducer, case A, White Noise; (b) pressure transducer,
case B, White Noise; and (c) LVDT, case C, ARC090.

liquid surface or floating roof). The damping ratio has been evaluated using the logarithmic
decrement method applied to the time histories of free vibrations (after the motion stopped).
In Figures 11 the transfer functions between the table acceleration and one of the pressure
sensors (near the bottom of the tank) are shown for the fixed base configurations (A and B),
without (Figure 11(a)) and with floating roof (Figure 11(b)). For both cases, the frequency of
the sloshing motion does not change and is almost coincident with the theoretical value (0.4 Hz),
whereas a resonant frequency of around 18 Hz is also shown, which corresponds to the main
natural frequency of the motion due to the deformability of the wall.
In Figure 11(c) the transfer function of the base displacement, for case C and the Koaceli
accelerogram (PGA = 0.35g), is represented. The graph shows a sloshing frequency placed at

Copyright q 2009 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2009)
DOI: 10.1002/eqe
 M. DE ANGELIS, R. GIANNINI AND F. PAOLACCI

Figure 12. Free vibrations time history of: pressure (a) case A and B; roof
displacements (b) case C; and (c) case D.

0.40 Hz and, more clearly, the isolation frequency equal to 1.22 Hz, in good agreement with the
design indications.
The damping of the sloshing motion has been measured on the basis of the amplitude decrement
of the free vibrations of fluid. Figure 12(a) shows the time-history of the wall pressure relative
to the two fixed base cases; it has been measured after the table was stopped, and filtered with a
low-pass filter using a cut-off frequency of 0.4 Hz. The damping is significantly increased by the
floating roof, even if the first oscillation frequency does not change. In fact, in case of free liquid
surface, the damping is very low and is not in practice measurable, whereas using the floating
roof a damping of 1% has been obtained. This effect is probably due to the interaction between
the liquid surface and the floating roof together with the friction developed by the rubber gasket
during contact with the tank wall.
The free vibrations of the floating roof for the isolated tank are shown in Figures 12(b) and (c).
The signals have been filtered in order to remove the frequencies greater than 1 Hz. Comparison
with the Figure 12 clearly shows a considerable increasing of the sloshing damping, variable in
the range 2.5–3.0%, for both isolated configurations. This increment was expected by numerical
models, carried out with non-classical damping theory, but with lower values than the experimental
ones.

4.2. Response of the base-isolated tank

The evaluation of seismic effects on the tank wall has been made in terms of both maximum base
shear induced by the dynamic pressure and relative impulsive and convective components.
These resultants have been evaluated interpolating the experimental values of the pressure with
the relative theoretical functions and then integrating along the height. The separation of the
sloshing and impulsive plus flexible components was obtained by filtering the signals around the
corresponding frequencies.
The responses of the table, measured in the different experimental tests, because of some
problems with the control system, were different, even if the same accelerograms were used.
Because of these differences, comparison was difficult. To overcome this problem the comparison
was done in terms of mean spectra accelerations Sa , (5% of damping), calculated in the period
range 0.05–0.1 s (10÷20 Hz) for the impulsive component and 2.2–2.8 s (∼ 0.35÷0.45 Hz) for
the sloshing component. Impulsive and sloshing forces have been represented as functions of the
relative spectra acceleration (Figures 13(a) and (b)).

Copyright q 2009 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2009)
DOI: 10.1002/eqe
 THE SEISMIC RESPONSE OF A STEEL LIQUID STORAGE TANK

Figure 13. (a) Impulsive base shear versus spectral acceleration of the impulsive motion; (b) convective
base shear versus spectral acceleration of the convective motion; (c) total base shear versus spectral
acceleration of the impulsive motion; and (d) maximum displacements of the floating roof versus spectral
acceleration of the convective motion.

In the same figures the interpolation curves of the experimental data are shown for the fixed
base (case B) and isolated base cases (case C, case D). For the fixed base case the trend is linear,
whereas the isolated base cases clearly show a nonlinear behavior of the phenomenon, especially
for high values of spectral acceleration.
The total shear has been plotted as a function of the same spectra ordinate as that used for
the impulsive component, the latter being a significant part of the total one (Figure 13(c)). The
figure clearly shows the high effectiveness of both the isolation systems, which induce considerable
reductions of the total shear affecting the tank wall. However, whereas the effectiveness of HDRB
isolators does not practically vary with Sa , the effectiveness of SIEPD devices is reduced for low
values of spectral acceleration, and it is practically negligible for Sa <0.3g.
This is due to the friction effects, often neglected for this kind of device, but particu-
larly significant for the tanks, in which the gravity mass, represented by the entire liquid
volume, is greater than the dynamic mass, which is represented by the impulsive mass
only (in the present case only 30% of the total mass). Therefore, to overcome the friction
appreciable levels of acceleration are needed. This problem is not present for elastomeric
devices.
This behavior, if appropriately controlled, could represent an interesting advantage recognizable
to SIEPD devices compared with the HDRB devices. In fact, for low levels of seismic intensity,

Copyright q 2009 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2009)
DOI: 10.1002/eqe
 M. DE ANGELIS, R. GIANNINI AND F. PAOLACCI

Figure 14. Liquid pressure on the wall along the height: (a) total; (b) impulsive+flexible; and (c) convective.

it could be convenient to use an isolation device with a reduced amount of slip, which instead is
required for more intense events in order to achieve a full employment of the dissipation capability
of the devices.
For the impulsive component (Figure 13(a)), the reduction is quite important, as already observed
for the total base shear, especially for high values of Sa ; on the contrary, the sloshing component
(Figure 13(b)) remains practically unchanged for HDRB, whereas a slight increase has been
observed using SIEPD. This increase has no influence on the stresses of the tank wall, at least for
large broad tanks. The negative effects are related to the floating roof oscillations, which could
generate, in some cases, the breaking of the gaskets and the leakage of inflammable materials.
Unfortunately, the base isolation does not reduce the amplitude of the superficial waves, although
the sensible increasing of damping reduces its duration and then probably the risk of failure of
the gaskets. However, for case D only an increase of the convective pressure has been measured,
which has induced a rather moderate increase of the amplitude displacement of the floating roof
that does not seem to be not particularly worrying (Figure 13(d)).
Further interesting information is provided by the pressure along the height of the tank wall,
measured in the four points indicated in Figure 10, and placed, respectively, at 0.15, 0.35, 0.45
and 0.75 m (Figure 14). The values have been interpolated using a polynomial function, imposing
the appropriate boundary conditions.
For example, Figure 14(a), relative to Arcelik Earthquake (PGA= 0.5g), demonstrates the high
effectiveness of the HDRB isolation system, which greatly reduces the total pressure on the tank
wall. For the single components of the fluid motion, Figures 14(b) and (c) show that the impulsive
pressure increases with depth while the convective pressure decreases, according to the theory
presented in Section 2. Moreover, the HDRB devices reduce both these quantities; in particular,
the impulsive one for which it is more effective. The almost zero value of the total pressure at the
liquid surface, case (B), is due to the predominance of the impulsive component that, in practice,
hides the convective pressure.
Finally, Figure 15 shows the maximum values of displacements of the tank base, recorded during
the tests. It may be noticed that the displacements of case D are lower than the displacements
of case C. This is another advantage for the SIEPD isolation system, since large displacements,
which have to be absorbed by the pipes–tank connections, represent a problem, the solution of
which is not always easy.

Copyright q 2009 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2009)
DOI: 10.1002/eqe
 THE SEISMIC RESPONSE OF A STEEL LIQUID STORAGE TANK

Figure 15. Tank base displacement, comparison between HDRB and SIEPD isolation systems.

5. COMPARISON BETWEEN NUMERICAL AND EXPERIMENTAL RESULTS

The numerical model defined in Section 2 has also been used to carry out step-by-step nonlinear
analyses using as base motion the accelerograms recorder on the shaking table during the tests.
The isolator devices have been modeled by the classical Bouc–Wen model [22]. The aim is to
confirm the reliability of the analytical and numerical formulation of the problem, also when the
floating roof is present, generally missing in the evaluation of the dynamic response of tanks.
As a matter of fact, hereafter a numerical–experimental comparison of the results of cases B, C
and D is presented and discussed. For brevity only the results of the Arcelik accelerogram, scaled
to the nominal acceleration nearest to the design value (0.5g), are shown.
Figure 16 shows the comparison, in terms of base shear for case B. First, a good agreement
between the experimental and numerical responses may be noted. A good predictive capacity of
the model emerges when the floating roof is present as well.
The numerical–experimental agreement can be better highlighted by analyzing the single compo-
nent of motion. Figure 16(a) shows, for case B, the base shear due to impulsive pressure. The
agreement is quite good and sufficient to correctly describe the phenomenon.
For the convective component of motion, although the value of the fundamental frequency was
very close to the theoretical value (0.4 Hz), the comparison between numerical and experimental
results proved unsatisfactory, because of non-zero initial conditions due to a low damping of the
liquid motion. The numerical–experimental agreement is improved in the cases C and D. In fact,
in these cases the fluid–structure interactions become less important and the high damping of the
liquid motion arrests oscillations more rapidly, allowing zero initial conditions.

Copyright q 2009 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2009)
DOI: 10.1002/eqe
 M. DE ANGELIS, R. GIANNINI AND F. PAOLACCI

Figure 16. Total base shear. Numerical–experimental comparison: (a) case B;

(b) case C; and (c) case D, Arcelik 0.5g.

Figure 17. Displacements of the isolator devices. Numerical–experimental comparison:

(a) HDRB and (b) SIEPD, Arcelik 0.5g.

For the isolated tank, Figures 16(b) and (c) show the comparison between theoretical and
experimental results. The good agreement proves the reliability of the model.
For completeness, Figures 17(a) and (b) show the time-history of displacements of the tank
base for case C and D. It may be noticed that both the maximum value of displacements and their
residual values are different in the two cases. This is due to the accumulation of displacements
during the test with SIEPD devices and to a rotational motion of the tank due to a non-perfect
dynamic control of the shaking table. Nevertheless, the numerical model is sufficiently accurate
and may be considered as a good tool for evaluating the response quantities.
To summarize, the comparison between the shaking table test results and the response of simple
numerical models described in Section 2 shows the suitability of the latter to simulate the behavior
of the fluid–structure system, both for fixed base and isolated base tank, considering the floating
roof as well. Actually, the slight discrepancies between numerical and physical models are probably
due to some drawbacks in the experimental activity, caused by the high mass of the filled tank and
the frequency range investigated.

Copyright q 2009 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2009)
DOI: 10.1002/eqe
 THE SEISMIC RESPONSE OF A STEEL LIQUID STORAGE TANK

6. CONCLUSIONS

This paper deals with the main results of an experimental campaign carried out on a mock-up of
a steel liquid storage tank, equipped with floating roof, excited at base by a shaking table. The
mock-up is a reduced-scale model (1:14) of a big steel tank used in industrial plants for the storage
of petroleum products. The tests have been performed using different experimental configurations:
fixed base with and without floating roof, isolated base with HDRB or SIEPD. During the tests,
the tank has been subjected both to accelerogram histories, which reproduce records of natural and
artificial seismic events, and to appropriate signals for dynamic identification of the system. The
results show the effectiveness of both isolator typologies in reducing the total pressure generated by
the earthquake on the tank wall. On the contrary, a low increase of the oscillation amplitude of the
liquid surface, consequently of the floating roof, has been observed. This is partially compensated
by an increase of damping, which drastically reduces the number of the high amplitude oscillations
of the floating roof, after the seismic event is over. Finally, the comparison between the shaking
table test results and the response of simple numerical models shows the suitability of the latter
to simulate the behavior of the fluid–structure system, both for fixed base and isolated base tank
with and without floating roof.

ACKNOWLEDGEMENTS
The authors are grateful for the financial support provided by ISPESL-D.P.I.A. under project
B106/DIPIA/03. The Italian Company A.L.G.A. s.p.a. (Milan) and in particular Agostino Marioni are
gratefully acknowledged for the realization of the isolation systems. Finally, the ENEA (Italian National
Agency for New Technologies, Energy and the Environment) Research Centre ‘La Casaccia’ (Rome) is
gratefully acknowledged for the experimental apparatus and in particular Gerardo de Canio and Nicola
Ranieri for their participation to all test phases.

REFERENCES
1. Applied Technology Council. ATC-25: seismic vulnerability and impact of disruption of lifelines in the
conterminous United States. Redwood City, CA, 1991.
2. Beavers JE, Hall WJ, Nyman DJ. Assessment of earthquake vulnerability of critical industrial facilities in
the central and eastern United States. Proceedings of the National Earthquake Conference, vol. II, Memphis,
Tennessee, 1993; 81–90.
3. Nielsen R, Kiremidjian AS. Damage to oil refineries from major earthquakes. Journal of Structural Engineering
1986; 112(6):1481–1491.
4. Sezen H, Whittaker AS. Seismic performance of industrial facilities affected by the 1999 Turkey
earthquake. Journal of Performance of Constructed Facilities 2006; 20(1):28–36. DOI: 10.1061/(ASCE) 0887-
3828(2006)20:1(28).
5. Tajirian FF. Base isolation design for civil components and civil structures. Proceedings of Structural Engineers
World Congress, San Francisco, CA, July 1998.
6. Marioni A. The use of high damping rubber bearings for the protection of the structures from the seismic risk.
Jornadas Portuguesas de Engenharia de Estruturas, LNEC, Lisboa, 25–28 November 1998.
7. Sezen H, Elwood KJ, Whittaker AS, Mosalam KM, Wallace JW, Stanton JF. Structural engineering reconnaissance
of the 17 August 1999 Kocaeli (Izmit), Turkey earthquake. Technical Report PEER 2000/09, Pacific Earthquake
Engineering Research Center, University of California, Berkeley, CA, December 2000.
8. European Committee for Standardization (CEN). Eurocode 8: design of structures for earthquake resistance—part
4: silos, tanks and pipelines. UNI EN 1998-4:2006.
9. Sakai F, Nishimura M, Ogawa H. Sloshing behaviour of floating-roof oil storage tanks. Computer and Structures
1984; 19(1–2):183–192.

Copyright q 2009 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2009)
DOI: 10.1002/eqe
 M. DE ANGELIS, R. GIANNINI AND F. PAOLACCI

10. Housner GW. The dynamic behavior of water tanks. Bulletin of the Seismological Society of America 1963;
53:381–387.
11. Fischer D. Dynamic fluid effects in liquid-filled flexible cylindrical tanks. Earthquake Engineering and Structural
Dynamics 1979; 7:587–601.
12. Haroun MA, Hausner GW. Earthquake response of deformable liquid storage tanks. Journal of Applied Mechanics
1981; 48:411–418.
13. Veletsos AS, Tang Y. Rocking response of liquid storage tanks. Journal of Engineering Mechanics 1987;
113:1774–1792.
14. Fischer FD, Rammerstorfer FG. A refined analysis of sloshing effects in seismically excited tanks. International
Journal of Pressure Vessels and Piping 1999; 76:693–709.
15. Malhotra PK. New method for seismic isolation of liquid-storage tanks. Earthquake Engineering and Structural
Dynamics 1997; 26:839–847.
16. Shrimali MK, Jangid RS. Non-linear seismic response of base-isolated liquid storage tanks to bi-directional
excitation. Nuclear Engineering and Design 2002; 217:1–20.
17. Wang Y, Teng M, Chung K. Seismic isolation of rigid cylindrical tanks using friction pendulum bearings.
Earthquake Engineering and Structural Dynamics 2001; 30:1083–1099.
18. Summers P, Jacob P, Marti J, Bergamo G, Dorfmann L, Castellano G, Poggianti A, Karabalis D, Silbe H,
Triantafillou S. Development of new base isolation devices for application at refineries and petrochemical facilities.
Thirteenth World Conference on Earthquake Engineering, Vancouver, B.C., Canada, 1–6 August 2004; Paper
No. 1036.
19. Bergamo G, Castellano MG, Gatti F, Marti J, Poggianti A, Summers P. Seismic protection of petrochemical
facilities: main results from Indepth project. Tenth World Conference on Seismic Isolation, Energy dissipation
and Active vibrations Control of Structures, Istanbul, Turkey, 28–31 May 2007.
20. Ciampi V, De Angelis M, Valente C. Una metodologia di progetto di dispositivi per la protezione sismica di
ponti. 7◦ Convegno Nazionale l’Ingegneria Sismica in Italia, Siena, Italy, 25–28 Settembre 1995.
21. Vanmarcke EH, Gasparini DA. Simulated earthquake motions compatible with pre-scribed response spectra.
Technical Report R76-4, Department of Civil Engineering, Massachusetts Institute of Technology, Cambridge,
1976-01; 99 (420/G32/1976).
22. Wen Y. Method for random vibration of hysteretic systems. Journal of the Engineering Mechanical Division
1976; 102:150–154.

Copyright q 2009 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2009)
DOI: 10.1002/eqe