1983-Vibration Studies and Tests of Liquid Storage Tanks
1983-Vibration Studies and Tests of Liquid Storage Tanks
1983-Vibration Studies and Tests of Liquid Storage Tanks
SUMMARY
Theoretical and experimental investigations of the dynamic behaviour of ground-supported, deformable, cylindrical
liquid storage tanks were conducted. The study was carried out in three phases: (I) a detailed theoretical treatment of the
coupled liquid-shell system for tanks rigidly anchored to their foundations; (11) an experimental investigation of the
dynamic characteristics of full-scale tanks; and (111) a development of an improved seismic design procedure.
INTRODUCTION
The dynamic behaviour of ground-based liquid storage tanks has been the subject of an extensive study at
Caltech during the past 5 years. Theoretical and experimental investigations have been conducted to seek
possible improvements in the design of such tanks to resist earthquakes. This paper records the principal
results obtained during this research programme and presents a comprehensive review of the subject which
makes it more understandable and useful to both researchers and practising engineers.
Historical background
Early developments of seismic response theories of liquid storage tanks considered the container to be
rigid and focused attention on the dynamic response of the contained liquid.I3. l 4 HousnerZ3formulated an
idealization for estimating liquid response in seismically excited rigid, rectangular and cylindrical tanks. The
study presented values for equivalent masses and their locations that would duplicate the forces and
moments exerted by the liquid on the tank. In 1964, the Alaska earthquake caused large scale damage to
tanks of modern design, l 9 and initiated many investigations into the dynamic characteristics of flexible
containers. In addition, the evolution of both the digital computer and various associated numerical
techniques has significantly enhanced solution capability. Several studies4*6 9 2 1 were carried out to
investigate the dynamic interaction between the deformable wall of the tank and the liquid by the finite
element method. The tank was regarded as anchored to its foundation and restrained against cross-section
The tank was
distortions. A different approach to the solution of the problem was developed by Velet~os.~
assumed to behave as a single degree of freedom system and to vibrate in a prescribed mode. Later, Veletsos
and Yang obtained the natural frequencies of the liquid-filled shell by the Rayleigh-Ritz method. In the
past, experimental data were obtained by testing reduced scale models; however, most of these studies were
concerned with dynamic problems associated with aerospace applications. It was not until recently that
experimental investigations of the seismic response of aluminium tank models were carried out at the
University of California, Berkeley., l 6
Outline of the present study (see Figure I )
A necessary first step in analysing seismic response of structures is to compute the natural frequencies of
vibration and the associated mode shapes. These are determined by means of a discretization scheme in
* Assistant Professor; formerly Graduate Student and Research Fellow at California Institute of Technology
0098-8847/83/020 179-28$02.80
@ 1983 by John Wiley & Sons, Ltd.
180
M. A. HAROUN
VIBRATION STUDIES AND TESTS OF LIQUID STORAGE TANKS
y R I N G SHELL ELEMENT
I
2
MECHANICAL
3
LIQUID
REGION
I=
1
Theoretical
II.
Study
Vibration Tests
of Ful I - Scale
Liquid Storage Tanks
m. Seismic
PARAMETERS
OF MECH.
MODELS
Design
Simplified Analysis
and Design Curves
which the elastic shell is modelled by finite elements and the fluid region is treated as a continuum by the
boundary solution technique. In this approach, the number of unknowns is substantially less than in those
analyses where both tank wall and fluid are subdivided into finite elements.
Having established the basic approach to be used, the analysis is applied to investigate the influence of the
static hoop stress on wall vibrations, the effect of the coupling between liquid sloshing modes and shell
vibration modes, the effect of the flexibility of the foundation soil, and the influence of the rigidity of the roof.
The remainder of the theoretical phase of the study was devoted to analysing tank response to earthquake
excitation. Special attention was first given to the cos &type modes for which there is a single cosine wave of
deflection in the circumferential direction. The significance of the cosne-type modes on the response of
irregular tanks was then investigated.
The second phase of research involved vibration tests of full-scale tanks. The vibrations of three water
storage tanks with different types of foundations were measured. Ambient as well as forced vibration
measurements were made of the natural frequencies and mode shapes. Measurements were made at selected
points along the shell height, at the roof circumference and around the tank bottom. Comparison with
computed mode shapes and frequencies showed good agreement with the experimental results, thus
confirming the reliability of the theoretical analysis.
The main objective of the final phase of research was to close the gap between academic studies and
practical design considerations, and to provide practising engineers with a simple and sufficiently accurate
tool for estimating seismic response of tanks. A mechanical model, which takes into account the
deformability of the tank wall, was developed. The parameters of such a model can be found from charts and
the maximum seismic loading can be predicted by means of a response spectrum characterizing the design
earthquake. The outline of the overall study is presented diagramatically in Figure 1.
181
The free lateral vibration modes of an anchored circular cylindrical tank can be classified as the cost?type modes for which there is a single cosine wave of deflection in the circumferential direction, and as the
cos n8-type modes for which the deflection of the shell involves circumferential waves having n higher than 1.
Figure 2 illustrates the circumferential and the vertical nodal patterns of these modes. It should be noted that
earthquake motions tend to strongly excite the cos @type modes. For a tall tank, the cos &type modes can be
denoted beam-type modes because the tank behaves like a vertical cantilever beam. This is not true for a
broad tank because both the amplitude and the axial distribution of the radial displacement at 8 = 0 are
different from those of the circumferential displacement at 8 = n/2.
In addition to the shell vibrational modes, there are the low-frequency sloshing modes of the contained
liquid shown also in Figure 2.
I
I
I
I
I
m=l
m.2
m=3
I e r t i c a l nodal p a t t e r n
r
I
I
I
I
I
I
I
I
Circumferential
nodal p a t t e r n
( a ) Shell Modes
'
I
I
I
F i x e d base
ANCHORED
TANK
UNANCHORED TANK
182
M. A. HAROUN
CYLINDRICAL
SHELL
W E T T E D SURFACE
OF SHELL ( S e )
1/
W E T T E D SURFACE
OF BOTTOM
PLATE (S31
PHASE (1)-THEORETICAL
ANALYSIS
(1)
in the region occupied by the liquid (0< r < R, 0 < 8 < 2n, 0 <z < H). Since the velocity vector of the liquid is
the gradient of the velocity potential, the liquid-container boundary conditions can be expressed as follows:
1. At the rigid tank bottom, z = 0, the liquid velocity in the vertical direction is zero
84
-(r,
az
8,0, t ) = 0
2. The liquid adjacent to the wall of the elastic shell, r = R, must move radially with the same velocity as
the shell
183
34
PIdt(r, e, H , t )+ PIg a r , 0, t) = 0
(5)
5) = I:'UAt ) d t
(6)
in which V indicates the volume occupied by the liquid and ( ' ) means differentiation with respect to the time,
t . By requiring that the first variation of J c be identically zero, the Laplace equation and the linear boundary
conditions [equations (2), (3), (4) and (5)] can be obtained.
U ( t )=
{crJT{E}Rd8dz
2 0
and
{E}
fl,M,, Mo, A1
(9)
CE,,
Eel
EZe,
(10)
In equation (9), N , and No are the membrane force resultants; and M , and M , are the bending moment
resultants.The quantities fl and A? are referred to as the effective membrane shear force resultant and
the effective twisting moment resultant, respectively.
The shell material is assumed to be homogeneous, isotropic and linearly elastic. Hence, the force and
moment resultants can be expressed in terms of the normal and shear strains in the middle surface E,, E~ and
E , ~ ; in terms of the midsurface changes in curvature K , and KO; and in terms of the midsurface twist K , , as
follows:
(01 = CDl { E l
(1 1)
{d}T = [u, v, wl
(13)
and [PI is a differential operator matrix relating the generalized strain vector to the shell displacement
vector.
184
M. A . HAROUN
Lastly, the kinetic energy of the shell, neglecting rotary inertia, can be expressed as
j1
L
T(t)=
2
257
By selecting suitable trial functions which satisfy the Laplace equation identically, the volume integral in
equation ( I 5) can be replaced by a surface integral using Green's theorem
+ +
where V is the volume occupied by the liquid and bounded by the surface S = S , S, S,; and v is the
outward normal vector.
In the basic analysis, only the impulsive pressure of the liquid is considered; and, therefore, the shell
vibrational motion becomes independent of the free surface motion. Given this new situation, the functional
J takes the form
Expansion ofthe velocity potential function. Once a set of trial functions, fii(r, 8, z), which are solutions of
the Laplace equation, are identified, then one can assume that
I
4(r, Q, z , t ) =
c ~ ~ (fii(r,
t ) .8, z )
i= 1
(18)
where I is the number of trial functions. These functions can be obtained by the method of separation of
variables. Those which have vanishing derivative with respect to z at z = 0 and satisfy that the time derivative
of the potential function at z = H be zero for all time, are given by
185
where I , are the modified Bessel functions of the first kind of order n;
ff.
(2i- 1 ) n
; and
2H
i = 1,2, ..., I
Idealization of the shell. The shell is divided into an appropriate number of finite ring-shaped elements as
shown in Figure 4. The displacement components u, u and w are first expanded in Fourier series as follows:
c
00
u(O,z, t ) =
n= I
00
o(0, z, t ) =
~ ( eZ,, t ) = C
n= 1
This displacement functions u,(z,t), u,(z,t) and w,(z,t) are then expressed in terms of the nodal
displacements of the finite elements by means of an appropriate set of interpolation functions. The shape
functions associated with the axial and tangential displacements are taken to be linear between the nodal
points. However, those associated with the radial displacement are cubic Hermitian polynomials to assure
slope continuity at the nodes.
RING
ELEMENTS
Figure 4. Finite element definition diagram: (a) finite element idealization of the shell; (b) shell element
Matrix equation of motion. Employing the variational functional J , one can derive the matrix equation of
motion of the liquid-shell system. With the aid of the shell displacement model, the scalar energy quantities
can be expressed in terms of the assemblage nodal displacement vector of the shell, ( q } , as follows:
where [ K , ] and [ M , ] are the stiffness and mass matrices, respectively, of the shell.
The velocity function expression [equation (18)] can be rewritten in a matrix form as
186
M. A. HAROUN
Inserting this expression into the third term of the functional J , and noting that the trial functions satisfy the
conditions that 4 = 0 along S , and t?$/az = 0 along S,, one can write
I,
w4 ds = {4ITCc^l{A)
(25)
The calculation of the matrix [el is straightforward' and will not be presented herein.
Inserting equations (21), (23) and (25) into the variational functional, and applying the variational operator
yields
CMSI i41 +[&I
{d+c c^l{4= ( 0 )
[ C l W - IIQTM= ( 0 )
Since the matrix [C] is non-singular, one can write
( A ) = cc1- 'CC^IT{(i)
(28)
The governing matrix equation of the free lateral vibration of the liquid-filled shell is therefore given by
(CMSl+ CDW) { 4 }+ C K J ( 4 ) = ( 0 )
(29)
where [DM] is an added mass matrix due to the effect of the liquid. The matrix [ D M ] is symmetric and
partially complete (i.e. not banded); the elements are well distributed over the matrix.
Numerical examples. A digital computer program was written to compute the natural frequencies and
mode shapes of vibration of the coupled liquid-shell system. The shell nodal displacements (eigenvectors) are
a direct result of the solution, and these are then used to solve for the shell force and moment resultants, and
for the hydrodynamic pressure acting on the wall of the tank. Because of the efficiency of the method of
analysis, the program consists only of about 500 cards and the execution time on the Caltech digital
computer (IBM 370/3032 system) is less than 10sec.
Throughout this investigation, attention is given to two particular tanks: a broad tank (B) and a tall tank
(T). Tank (B) has the following dimensions: R = 18.3m (60 ft), L = 12.2m (40 ft) and h = 254 cm (1 in) while
the dimensions of tank (T) are: R = 7.32m (24ft), L = 21.96m (72ft) and h = 2.54cm (1 in). Both tanks are
Ib . sec2/in4). The tank wall is
assumed to be completely full of water of density p1 = lo3 kg/m3 (0.94 x
made of steel whose properties are: E = 20.67 x lo7kPa (30 x lo6Ib/in2), v = 0.3 and ps = 7,84 x lo3kg/m3
(0.733 x 1 O - j Ib. sec2/in4). The analysis was applied to compute the natural frequencies and modes of
vibrations of tank (B). The number of shell elements was taken to be 12; and, therefore, the number of
expected modes is 4 x 12 = 48. The fundamental mode shape of the cos &type modes is shown in Figure 5(a).
To investigate the influence of the aspect ratio (length to radius ratio), the dynamic characteristics of tank (T)
were computed and the fundamental mode shape was displayed in Figure 5(b). Inspection of Figure 5 shows
that the mode shapes of broad and tall tanks are indeed quite different. The hydrodynamic pressure
187
distributions for these two tanks and for similar rigid tanksz4 are shown in Figure 6 for comparison. It is
important to note that, in the numerical examples under consideration, attention is given to the cos 8-type
modes only; these modes are unaffected by the hydrostatic pressure of the liquid.
(a)
(b)
Figure 5. Fundamental vertical modes of full tanks: (a) broad tank (B) ( L / R = 0.67); (b) tall tank (T) (L/R = 3.00)
TRNK
L / R = 0.67
L/R = 3.00
HYORODYNRMIC
(
PRESSURE
DISTRIBUTION
FUNDRMENTRL MODE
188
M. A. HAROUN
To incorporate such an effect, it was necessary to modify the strain energy expression of the shell, and to
generalize accordingly the equations of motion. The strain energy can be written conveniently as
U t ) = U,(t)+
(30)
U2(d
where U,(t) is the strain energy employed in deriving the stiffness matrix of equation (21) and U2(t)is given by
where NL is the initial membrane force resultant in the circumferential direction, and cO is the midsurface
strain which can be expressed as
&
-R
ae +
.) +;{$(;$
I).
+[(; $+
[ -g)T}
+ (;
The non-linear terms in equation (32) are given by Washizu.26 However, it should be mentioned that the
linear terms of the strain-displacement relationships developed by Washizu are identical to those of
Novozhilov theory which has been used in U,(t).
With the aid of the finite element displacement model of the shell, the strain energy expression U2(t)can be
rewritten as
where [KB] is an added stiffness matrix due to the presence of the initial hoop stress. The matrix equation of
motion of the tank wall now takes the familiar form
(34)
where [ K ] = [ K , ] [KL].
Numerical examples. The analysis was applied to compute the natural frequencies and mode shapes of tank
(B) and tank (T). As expected, the influence of the hoop stress on the coso-type modes of vibration is
insignificant. The stiffening effect due to the hydrostatic pressure has a considerable influence upon the
frequencies of the cosn0-type modes of tall tanks. Such an effect becomes more significant as the
circumferential wave number n increases. On the other hand, the stiffening effect on the cos no-type modes of
broad tanks is, for practical purposes, negligible.
To illustrate the effectiveness of the analysis under consideration, a comparison between the computed
dynamic characteristics and those found experimentallyz2 is made. The physical model employed in the
test was partly filled with water, and had the following dimensions and properties: R = 10.2cm (4in),
L= 31.8cm (12.5in), H = 27.9cm (llin), h = 0.013cm (0.005in), E = 5 0 6 x lOkPa ( 0 . 7 3 5 ~1061b/in2),
p s = 1.42 x lo3 kg/m3 (0.133 x
Ib. sec2/in4) and v = 0.3. As is seen from Figure 7, the computed
characteristics are in good agreement with the experimental results. This confirms the accuracy of the
analysis, and the significant role played by the initial hoop tension during the cos no-type vibrations of tall
tanks.
Interaction between liquid sloshing modes and shell vibration modes
The dynamic interaction between liquid sloshing waves and shell vibrations has not yet been investigated.
The coupling is usually neglected on the ground that the significant sloshing modes are of much lower
natural frequencies than those of the vibrating shell. In the following analysis, emphasis is placed on the
question of when the coupling effectcan be significant; in other words, is it necessary to consider the liquidshell-surface wave system, or only the two uncoupled cases: (i) the liquid-shell system plus (ii) the free surface
gravity waves in a similar rigid tank?,,
I89
-&--A-INITIAL
i
I
.-.
SHAKING
TESTS
n.6
I
6
.n
.J
9
(a)
(b)
Figure 7. Comparison between calculated and measured characteristics of the cos no-type modes: (a) mode shape; (b) natural frequencies
As was mentioned earlier, a finite element discretization of the liquid region itself is not necessary; only its
boundaries need to be discretized. The shell is modelled by a series of ring-shaped elements as before, and the
quiescent liquid free surface is represented by concentric annular rings. The trial functions are chosen from
the solutions of the Laplace equation which are non-singular at r = 0 and have a vanishing derivative with
respect to z at z = 0. These can be written as
J,(kr) cosh ( k z )
(ne)
( n > 1)
rn
I&)
(35)
cos (kz)
(0)
(36)
where the overall mass and stiffness matrices are written in the following partitioned forms:
and {x} is the nodal displacement vector for the entire assemblage which can be written as
where (4) is the assemblage nodal displacement vector of the free surface.
Numerical examples. The analysis was applied to the tall tank (T). The quiescent liquid free surface is
divided into 12 elements, and the elastic shell is modelled by 12 elements; therefore, the number of expected
modes is 60. The lowest natural frequencies of the coupled system are in good agreement with the sloshing
frequencies in a similar rigid tank. Furthermore, the 13th, 14th, ..., etc. ascending frequencies are, for
practical purposes, the same as those computed for the liquid-filled shell. Therefore, it may be concluded that
the coupling effect is negligible. This is further substantiated by the mode shapes. Figure 8 displays the
190
M. A. HAROUN
fundamental mode shape of the coupled system; it is clear that this mode has predominantly free surface
motion.
.I
(39)
and
84
-(r, 8,0, t ) =
8Z
- rdr(t)cos (8)
(40)
where x(t) is the horizontal translation of the tank base; u(t) is the angle of rotation about a transverse axis
through the base; and w(z, t ) is the deflection of the beam axis. The finite element analysis of the liquid-shell
interaction problem revealed that the fundamental mode shape of completely full, moderately tall tanks can
191
(E)
w(z, t ) = y ( t ) .sin -
By neglecting liquid sloshing modes, one can express the velocity potential function in terms of the time
derivatives of x(t), ~ ( tand
) y(t). The hydrodynamic pressure is related to the velocity potential function by
Pd(r,
84
e, z, t , = -pl-(r,
e, z, t,
at
(42)
and the lateral force exerted on the tank at any elevation z above the base can be computed from
(43)
The variation of the hydrodynamic pressure acting on the base produces an overturning moment M , which
can be calculated from
M, = JRJ2'..(..8,O,
0
t)r2cos(0)dedr
(44)
Having obtained the hydrodynamic forces, the base shear force, Q(t),and the overturning moment, M(t),
can be computed. These are related to the base motion by"
192
M. A. HAROUN
where K , and K , are the real parts of the foundation impedance functions, and C, and C , are the imaginary
parts. Taking y(t), x ( t ) and $(t) = Ru(t) as the generalized degrees of freedom, one can write the matrix
equation which governs the free vibration of the soil-flexible tank system as
Numerical examples. A water storage tank is anchored to a 61 cm (2ft) thick R.C. slab on deep alluvium.
The tank, which was tested experimentally, is 7.32 m (24ft) in radius, 21.64m (71 ft) in height, and consists of
a thin steel shell of varying thickness; the maximum thickness at the bottom is 1.75 cm (0.69 in) and the
minimum thickness at the top is 0.64cm (0.25 in). The fundamental natural frequency of the soil-tank system
is computed for different values of shear wave velocity of the soil, and the results are shown in Table I.
Inspection of this table indicates that the deformability of the foundation soil reduces the natural frequency
of the tank; this was observed experimentally in phase (11). As the shear wave velocity increases, the
fundamental natural frequency approaches the value computed assuming a rigid foundation.
304.8
(1000)
2.79
365.8
(1200)
3.03
426.7
(1400)
3.21
487.7
(1600)
3.34
609.6
(2000)
3.49
731.5
(2400)
3.59
1097.3
(3600)
3.12
(;
-, L, t
(n
1)
(47)
In addition, it restrains the cos no-type modes against cross-sectional deformations at the tank top; i.e.
w(0, L, t ) = u(o, L, t ) = 0 (n 3 2)
(48)
Furthermore, by virtue of its thinness, the plate has very little stiffness in the z-direction transverse to its
plane; consequently, it will generate negligible moment M , and membrane force N , at the shell top as the
tank vibrates. Although the foregoing boundary conditions are highly simplified, the computed frequencies
and mode shapes of real full-scale tanks are in good agreement with those measured by vibration tests.
Numerical exumples. The effect of the roof in-plane rigidity on the cos &type modes is generally negligible.
A slight reduction in the values of the natural frequencies is observed due to the additional mass of the roof.
On the other hand, the roof has a significant effect on the cosno-type modes of the shell as illustrated in
Figure 10.
193
+K-+K-
ROOF EFFECT
INCLUDED
I
1
rl
I
5
I
7
(49)
where [MI and [ K ] are the mass and stiffness matrices defined in equation (29); [ C ] is the damping matrix;
and (Perf}is the effective earthquake load vector resulting from a given ground motion G(t). For a perfect
circular cylindrical shell, the effective earthquake load vector takes the form
and consequently, the earthquake response can be obtained by superposition of the vertical modes
corresponding to n = 1 only.
Efjiective force vector
The total displacement vector of the shell can be considered as the sum of two components: the relative
displacement vector { d } defined by equation (13), and the displacement vector { d g } , associated with the
ground displacement G(t), which can be expressed as (Figure 11)
M. A. HAROUN
194
The external forces acting on the shell due to ground motion G(t) include: (i) the distributed inertia force of
the shell which is given by
-Psh{J,j
(52)
and (ii) the hydrodynamic pressure on the wall of a similar rigid tank. This pressure can be expressed as
V g }
The work done by these external loads during arbitrary virtual displacements
dU
,cOS (e)
(54)
can be expressed as
6W =
lLj
2z({Fg)T(6d})
R d8 dz + ]oH]02z(pg(R,8, Z,t ) 6w,cos (8))R dB dz
(55)
With the aid of the finite element model of the shell, this expression becomes
6W = - & ) { 6 q } T { F }
and, therefore, the effective earthquake load vector is given by
{Peff}= - if-)
(56)
(57)
Numerical examples
The analysis was applied to estimate the earthquake response of the tall and broad tanks (T and B) presented
in the preceding sections. Their free lateral vibrational modes were superimposed to evaluate the response to
the N-S component of the 1940 El Centro earthquake. Although a modal damping ratio of about 2 per cent
seems appropriate for the vibration of the tank wall, the maximum values of the seismic response were
computed for different values of damping ratio. Table I1 presents the impulsive response of the tall tank and
displays that of a similarly excited rigid tank for comparison. It is found that the flexibility of the tank wall
has a significant effect on the response of both broad and tall tanks. This is due to the fact that the impulsive
loads arise through acceleration of the shell. If the shell is flexible, two acceleration components must be
considered: (i) the acceleration of the undeformed shell, i.e. the ground acceleration, and (ii) the relative
acceleration due to shell deformations. In a rigid tank, only the acceleration of the undeformed shell is
considered which introduces the noticeable difference in the magnitude of shell stresses. These results were
further substantiated by comparing the computed responses of aluminium tank models with those
measured. 5 * This comparison also indicated that the computed fundamental frequency is higher than the
measured frequency.
195
Damping
Mechanical
modell
2%*
5%t
lO%t
Rigid tank
1.13
(0.445)
0.87
(0.344)
0.75
(0296)
1.17
(0.462)
1,467
(8,375)
1,134
(6,473)
975
(5,5641
603
(3,444)
1,655
(9,454)
379
(2,166)
1.75 x lo7
(3.95 x 106)
1.51 x lo7
(3.39 x 106)
1.21 107
(2.72 x lo6)
2.30 107
(5.17 x lo6)
2.27 x 107
(5.1 1 x lo6)
EXPERIMENT. FREE
1.2
a
LK
W
00
0
I-
TO!
H/L
R/h
833
1250
0.93
0.89
1.c
_I
093
0.E
zA
Y
ox
m
n
0.4
o\
H / L = 0.92
zP
0.2
0.2
0.4
0.6
0.8
I .O
NORMALIZED FREQUENCY, w / w ,
1.2
196
M. A. HAROUN
fixing the frequency of excitation and increasing the amplitude of the shaking table motion until buckling
occurred. Theoretically, the buckling was assumed to occur when the axial membrane stress at the base,
computed from the present analysis, reached the classical value.
PHASE (11)-FULL-SCALE
VIBRATION TESTS
Adequate understanding of the dynamic behaviour of complex structures is dependent upon the use of both
theoretical and experimental techniques to support each other. Therefore, a series of ambient and forced
vibration tests of three, full-scale, water storage tanks was conducted to determine the natural frequencies
and, if possible, the mode shapes of vibrations, and to select two tanks on which permanent instruments
would be installed to record future earthquakes. Figure 13 shows schematic sections of these tanks and their
foundations. Tank no. (1) is anchored to a 61 cm (2 ft) thick R.C. slab on deep alluvium, and tank no. (2) rests
on a 3.66 m (12 ft) deep concrete ring wall without anchor bolts. The third tank is anchored to a 153 m (5 ft)
thick R.C. foundation slab supported by 100 R.C. caissons.
NO. ( 2 )
NO. ( 1 )
NO. ( 3 )
I
-I
I
I
1
/R.C.
SLAB
/RING
WALL
onchor ' b o l l s
/COMPACTED
GRAVEL
L!
R. C. CAISSONS
197
vibrations of the tank. Figure 14 is a schematical diagram showing the experimental set-up and the
instrumentation used in testing tank no. (1). Slightly different arrangements were made for the other two
tanks. During the tests, the tanks were maintained full whenever possible.
oscillograph
recorder
detail ( a I
198
M. A. HAROUN
8
2
FREWENCY
- nz.
(a)
1.m-
0.75
2!
- HE.
(b)
Figure 15. Ambient and sinusoidal tests of tank no. (1): (a) Fourier amplitude spectrum of the radial component of shell velocity at the
tank top; (b) Fourier amplitude spectrum of the vertical velocity at station (8), and sample traces made simultaneously during forced
vibration test
FREMNCY
I99
One phenomenon, clearly observed in the motion of the shell away from the roof, was that cos no-type
vibrations of the tank wall were developed. These modes were anticipated only in the ambient tests because
of the nature of the excitation which tends to excite many modes. However, in the forced vibration test of
tank no. (l), cosno-type modes were also excited by rigid base motion, presumably because of the initial
irregularity of the shell. Figure 16 shows the axial pattern of the cos %-mode based on ambient and forced
vibration measurements. It is clear that the roof does restrain the tank top against radial deformations. The
computed natural frequency is 2.46cps, which is in close agreement with the measured one of 2-42cps. The
computed mode shape is also presented in the same figure for comparison.
Tank no. (2), which is not anchored to the foundation, exhibited behaviour slightly different from tank no.
(1). However, it is believed that it would behave much differently with a high level of excitation. This is a nonlinear problem for which the relationship between small amplitude vibration characteristics and strong
motion seismic response is not well defined.
F U L L - S C R L E V IBRRT ION
TEST
FORCED VIBRATION
0
C I R C U M F E R E N T I A L WAVE NUMBER
AMBIENT VIBRATION
TRNK NUMBER !
Figure 17 displays the Fourier amplitude spectrum of the radial component of shell velocity recorded at
station no. (4) on tank no. (3). The circumferential modes with n up to 5 were identified from the ambient
measurements. The availability of the computed frequencies and the good correlation between the measured
and the computed frequencies helped in identifying the mode number with n>6. It should be mentioned that
the low-pass filter of the signal conditioner was set, by mistake, to 4 cps in testing the third tank and therefore
the peaks in the range 4 to 5 cps do not appear in their respective magnitude. Also, the high peak at 3.45 cps is
attributed to environmental noise which was also observed in the calibration test. Figure 18 shows a
comparison between the computed and measured frequencies. It clearly emphasizes the significant role
played by the roof and the initial hoop stress in estimating the natural frequencies of the cos &type modes. It
is also evident that the roof effect is more pronounced for small n, while the initial stress influence is more
significant for large n.
PHASE (111)-SEISMIC
DESIGN ANALYSIS
The two most common standards and codes currently used for the design of tanks are the API 6502 and the
AWWA
Seismic loads in these standards are based on the mechanical model derived by HousnerZ3for
rigid tanks. The design procedure considers two response modes of the contained liquid: (1) the impulsive
200
M. A. HAROUN
=
-
II
d
II
c
t 4
Not
0.03
"
1.m
o m
-.--A
2.m
to scale
u.m
3.m
5 .m
FAEQUENCY - H2.
NO
ROOF
--
itSt ROOF
-&--A-
FULL
NU
INITIHL
IhlTIRL
INITIQL
STRESS
STRESS
STRESS
SCALE V I B R A T I O N TEST
/A
- i - - -i _ ~ - J
-L
-_
I-_--_
_1
4
Figure 18 Comparison between calculated dnd measured natural frequencles of the cos no-type modes
response caused by the portion of the liquid accelerating with the tank, and ( 2 ) the 'convective' response
caused by the portion of the liquid sloshing in the tank. In a recent investigation of the effect of wall flexibility
on the seismic response of anchored tanks, V e l e t ~ o proposed
s~~
an approximate method for the computation
of the hydrodynamic forces associated with tank deformations. However, the method did not provide explicit
values for the fundamental natural frequency of the system which is needed for the determination of the
20 1
spectral acceleration from a response spectrum. Later, Veletsos and Yang25provided a diagram between the
fundamental natural frequency of the liquid-shell system and the height/radius ratio; however, such a relation
is applicable only to a shell thickness/radius ratio of 0.001. Recognizing the importance of wall flexibility,
recent codes have adopted an increase in the acceleration coefficient to an ad hoe value representing the
short period amplified acceleration due to shell deformation. It should be noted that such acceleration is
specified independent of the tank dimensions and of the support condition.
The principal aim of the final phase of research was to devise a practical approach which would allow,
from the engineering point of view, a simple, fast and sufficiently accurate estimate of the seismic response of
storage tanks. A mechanical model, shown in Figure 19, was developed. The equivalent masses m,,rnf and m,
correspond to the forces associated with ground motion, wall deformation relative to the ground, and liquid
sloshing, respectively.
Convective component
I t has been shown that the coupling between liquid sloshing modes and shell vibrational modes is weak;
and, consequently, the convective dynamic pressure can be evaluated with reasonable accuracy by
considering the tank wall to be rigid. The maximum convective pressure due to the fundamental sloshing
mode only is given by
where S,, is the spectral value of the pseudo-acceleration corresponding to the fundamental sloshing
frequency which is given by
w, = 1 . 8 4 tanh
~
(y)
The equivalent mass m, can be evaluated from the hydrodynamic pressure [equation
(I*:H)
= 0.455npIR 3 tanh -
(59)
(%)Iby
202
M . A. HAROUN
and its centre of gravity is at a distance H , from the base which is given by
Figure 20 displays the ratios (m,/m) and ( H J H ) for different values of (HIR),where m is the total mass of the
liquid.
I
I
I
I
I '
(62)
Xf
(63)
Since the first term in equation (62) is proportional to the relative acceleration of the shell, one must
rearrange equation (62) before estimating the maximum seismic loads by means of a response spectrum; i.e.
rewrite equation (62) as
Q(t) = rnf(Xf(t)
(64)
One recognizes that the maximum value of I X,(t)+ G(t)I is the spectral acceleration Safcorresponding to the
natural frequency 0,.
The overturning moment, due to seismic forces, applied to the bottom of the shell can
203
be expressed as
M ( t ) = mf Hf x,(t)
+ rn, H , %(t)
It is of interest to note that the fundamental mode shape is normalized in such a way that the maximum
amplitude of the radial component of shell displacement is 1.0 and, therefore, one can also estimate the
maximum radial component of shell displacement by
I wmax I = P I Sdf
(66)
where PI is the modal participation factor of the fundamental mode of vibration and S d f is the spectral
displacement corresponding to the fundamental natural frequency of.Figure 2 l(a) displays the nondimensional parameter (ofH J ( p , / E ) for different values of ( H / R ) and (h/R), where p , and E are the mass
density and Youngs modulus, respectively, of the shell material. These frequencies are for tanks completely
filled with water. The remaining parameters PI, (m,/m), (H,/H), (m,/rn) and ( H J H ) are displayed in Figures
21(b)-(f), respectively. It should be noted that the effect of shell mass on the fundamental frequency of full
tanks is negligible; consequently, one can estimate the natural frequency c5, of a tank filled with liquid of
density p , by
f;
(67)
= o f J(Pw/PJ
where ofis the natural frequency of the same tank when filled with water and pw is the mass density of water.
Numerical examples
Consider the tall tank (T) discussed in the preceding sections. The analysis of the tank involves the
following steps:
(a) Parameters ofthe mechanical model. The fundamental natural frequency of sloshing, o,= 1.57 rad/sec,
is obtained from equation (59); and consequently, T, = 2n/o, = 4sec. The convective mass m, and its
elevation H , are given by (Figure 20) rn, = 5,585N. sec2/cm (3,1881b. sec2/in) and H , = 0.82H = 18m
(59.04ft). The fundamental natural frequency of vibration of the liquid-filled shell can be determined from
Figure 21(a) for a value of (H/R)= 3.0 and a value of (h/R)= 0900347; hence, of= 33.25 rad/sec.
The remaining parameters can be found from Figure (21): m, = 0.735 m = 27,251 N . sec2/cm
(15,5551b.sec2/in); H , = 0.5548 = 12.17m (39.88ft); m, = 0.892rn = 33,073 N.sec*/cm (18,8771b.sec2/in);
and H , = 0.442H = 9.71 m (31.82ft).
(b) Spectral values. The maximum ground acceleration of the N-S component of the 1940 El Centro
earthquake is 0.3489. The spectral acceleration S,, corresponding to the sloshing frequency o,and for a
damping ratio [, of 0.5 per cent can be found from the response spectrum; it is given by S,, = 0.063g. Lastly,
the spectral displacement and acceleration corresponding to the frequency ofand for a damping ratio of 2
per cent are Sdf = 0.757 cm (0.298in) and S,, = 0.8569.
(c) Seismic response. The maximum free surface wave height is estimated from
&,,ax
Sas
= 0837R-
38.6cm (15.19in)
In addition, the maximum radial component of shell displacement is given by [equation (66)]
IwrnaxI = PISdf= (155)(0.757) = 1.17cm(0.462in)
Because the maximum values of the convective, impulsive and short-period components of the base shear do
not occur, in general, at the same time, one can estimate the maximum value of base shear from
I QT
Imax
= J((m, S a J 2
+ (mf
Saf12
+ C(m,-mr)
%max12)
204
M. A. HAROUN
I
3
ZLLi1
0 . U 0.60 0.80
I00
1.60
!.UO
I20,
I
1.60
2.00
2.20
2.UO
2.60
2 60
HEILHT-RRDIUS m r i o
L-..1
3.0C 3 . 2 0 3 . W
(HIR)
I
o.uoo.so
0.80
2.00
2.m
2 . ~ 0 2.60
2.80
3.00
1
320
1
3 . ~ 0
0.vOO.m
0.m
1.03
1.20
I.VO
2.00
1.80
1 . 1
2.20
2.W
HEIMT-RADIUS RRTIO ( H / R )
H E I M T - R A D I U S RATIO ( H / R )
(4
(4
2.60
2.BO
--011
3.20
3.W
-I
i
-1
?j
o.uOO.60
0.80
I.00
1.20
l
LVO
160
1.80
2.00
2.20
2.C
2.60
l
2.80
l
3.00
3.20
,
3.K
?I
, I0.80/ >.DO
, ,
O.VOO.60
1.20
HFIGHI-RADIUS R R T l O ( H I R )
(0
(e)
, , , ,
2.60
2.80
'
3.00
3.20
3.W
205
Similarly, the maximum overturning moment applied to the bottom of the shell is given by
IM,
lmax
J((ms
S a J 2 +(mf Hf Sa,12
~s
+ ~ ( m Hr-mf
r
H,) Grnax12)
and, consequently, the maximum axial force resultant at the bottom of the shell is given by
Nz
lmax
=
=
Table I1 shows a comparison between the maximum response values obtained by the simplified mechanical
model and those found by time integration of the modal equations of motion.
Design charts were also used to estimate the response of the broad tank (B) to the N-S component of the
1940 El Centro earthquake; the results are shown in Table 111.
Table 111. Seismic response of tank (B)
r,
-4
(sec)
(sec)
_-_
~~~~
Parameters
0.162
6.89
73,980
(42,227)
52,140
(29,761)
53,428
(30,496)
..
S,,
Spectral values
Seismic
response
t,,,
w,,,
Q,,
N,I,,,
= 0.0289
S,, = 0.8289
198N/mm (1,1281b/in)
6.77
(22.2)
4.89
(16.1)
4.94
____
( 16.2)
S,, = 0544cm
(0.214 in)
Exact solution
0.620 cm (0.244 in)
39.7 x lo6 N
(8.92 x lo6lb)
190 N/mm
(1,085 lb/in)
CONCLUSION
A simple and computationally effective method for computing the dynamic characteristics of anchored,
ground-supported, cylindrical liquid storage tanks is presented. The elastic shell was modelled by finite
elements and the liquid region was treated by the boundary solution technique.
It was concluded that the initial hoop stress due to the hydrostatic pressure and the in-plane rigidity of the
roof system may affect considerably the cos &type modes of the tank wall. In addition, the flexibility of the
foundation soil can reduce measurably the fundamental natural frequency of the cos 8-type modes due to the
rocking motion. It was confirmed that the coupling between liquid sloshing modes and shell vibrational
modes is weak; and, consequently, the convective pressure can be evaluated with reasonable accuracy by
considering the tank wall to be rigid, and the impulsive pressure can be determined by analysing the
liquid-shell system and neglecting the sloshing motion. The validity of the method of analysis was
confirmed by both scale model testing and field measurements of the vibrational characteristics of full-scale
tanks, Earthquake response of deformable storage tanks was obtained by superposing the free lateral
vibrational modes; the dynamic stresses were much greater than those computed assuming rigid walls.
Finally, a mechanical model which takes into account the deformability of the tank wall was developed and
its parameters were displayed in charts. Comparison with the exact solution of the problem confirmed the
validity of the model.
206
M. A. HAROUN
ACKNOWLEDGEMENTS
The author wishes to acknowledge the encouragement and advice of Professor G. W. Housner of the
California Institute of Technology. The co-operation of the Metropolitan Water District of Southern
California in making available its facilities for conducting tests is gratefully acknowledged. This research was
supported by grants from the National Science Foundation and the Earthquake Research Affiliates program
at the California Institute of Technology.
REFERENCES
1. H. N. Abramson (Ed.), The dynamic behaviour of liquids in moving containers, N A S A SP-106, National Aeronautics and Space
Administration, Washington, D.C., 1966.
2. API Standard 650, Welded steel tanks for oil storage, 6th ed., American Petroleum Institute, Washington, D.C.
3. AWWA D100, AWWA Standard for welded steel tanks for water storage, American Water Works Association, Denver, Colorado.
4. T. Balendra, and W. A. Nash, Earthquake analysis of a cylindrical liquid storage tank with a dome by finite element method,
University of Massachusetts, Amherst, Mass., 1978.
5. D. P. Clough, Experimental evaluation of seismic design methods for broad cylindrical tanks, Report No. EERC 77-I0,
Earthquake Engineering Research Center, University of California, Berkeley, CA, 1977.
6. N. W. Edwards, A procedure for dynamic analysis of thin walled cylindrical liquid storage tanks subjected to lateral ground
motions, PhD Thesis, University of Michigan, Ann Arbor, Michigan, 1969.
7. R. D. Hanson, Behaviour of liquid storage tanks, The Great Alaska Earthquake of 1964, Engineering, National Academy of
Sciences, Washington, D.C., 1973, pp. 331-339.
8. M. A. Haroun, Dynamic analyses of liquid storage tanks, Report N o . EERL 80-4, Earthquake Engineering Research Laboratory,
California Institute of Technology, Pasadena, CA, 1980.
9. M. A. Haroun and G. W. Housner, A procedure for seismic design of liquid storage tanks, Report N o . EERL 80-5, Earthquake
Engineering Research Laboratory, California Institute of Technology, Pasadena, CA, 1982.
10. M. A. Haroun and G. W. Housner, Seismic design of liquid storage tanks, J . tech. councils ASCE, 107, 191-207 (1981).
1 I. M. A. Haroun and G. W. Housner, Earthquake response of deformable liquid storage tanks, J . appl. mech. A S M E , 48, 411418
(1981).
12. M. A. Haroun and G. W. Housner, Dynamic interaction of liquid storage tanks and foundation soil, Proc. 2nd A S C E / E M D
specialty con$ dynamic response struct., Atlanta, Georgia (1981).
13. L. S. Jacobsen, Impulsivehydrodynamics of fluid inside a cylindrical tank and of a fluid surrounding a cylindrical pier, Bull. seism.
SOC. Am. 39, 189-204 (1949).
14. L. S . Jacobsen and R. S. Ayre, Hydrodynamic experiments with rigid cylindrical tanks subjected to transient motions, Bull. seism.
soc. Am. 41, 313-346 (1951).
15. J. C. Luke, A variational principle for a liquid with free surface, J . Juid mech. 27, 395-397 (1967).
16. A. Niwa, Seismic behaviour of tall liquid storage tanks, Report N o . EERC 7 8 4 4 , Earthquake Engineering Research Center,
University of California, Berkeley, CA, 1978.
17. V. V. Novozhilov, Thin Shell Theory, P. Noordhoff, Groningen, The Netherlands, 1964.
18. F. E. Richart Jr., J. R. Hall and R. D. Woods, Vibrations ofsoils and Foundations, Prentice-Hall, Englewood Cliffs, N.J., 1970.
19. J. E. Rinne, Oil storage tanks, The Prince William Sound, Alaska, Earthquake of 1964, and Aftershocks, Vol. 11, Part A, 1J.S. Coast
and Geodetic Survey, Washington, D.C., 1967, pp. 245-252.
20. F. Sakai, M. Nishimura and H. Sakoda, Studies on earthquake-resistance of liquified natural gas storage tanks, 6th int. conJ
exhibit. LNG, Kyoto, Japan (1980).
21. S. H. Shaaban and W. A. Nash, Finite element analysis of a seismically excited cylindrical storage tank, ground supported and
partially filled with liquid, University of Massachusetts, Amherst, Mass., 1975.
22. C. Shih and C. D. Babcock, Scale model buckling tests of a fluid filled tank under harmonic excitation, 1980 press. vessels piping
techn. conj, ASME, San Francisco, California (1980) (Preprint 8GC2/PVP-66).
23. U.S. Atomic Energy Commission, Nuclear reactors and earthquakes, TID-7024, Washington, D.C., 1963, pp. 367-390.
24. A. S . Veletsos, Seismic effects in flexible liquid storage tanks, Proc. 5th world con$ earthquake eng., Rome, Italy, 1, 63G639 (1974).
25. A. S. Veletsos and J. Y. Yang, Earthquake response of liquid storage tanks, Advances in Civil Engineering through Engineering
Mechanics, Proc. annual EMD specialty conj, A X E , Raleigh, N.C., 1-24 (1977).
26. K. Washizu, Variational Methods in Elasticity and Plasticity, Pergamon Press, New York, 1975.
27. R. S. Wozniak and W. W. Mitchell, Basis of seismic design provisions for welded steel oil storage tanks, presented at the 43rd
Midyear Meeting, API, Toronto, Ontario, Canada (1978).