Академический Документы
Профессиональный Документы
Культура Документы
Abstract: In this paper, a procedure for computing hydrodynamic pressures in rectangular tanks is proposed. The pro-
cedure, which is referred to as the sequential method, considers the effect of the flexibility of the tank wall in deter-
mining the hydrodynamic pressures. In this study, only the impulsive response of the tank is considered. Based on a
two-dimensional model of the tank wall, dynamic time-history analysis is carried out to study the effect of wall flexi-
bility on the response. In the analysis, both a tall tank and a shallow tank are considered. The results of analysis are
compared with those obtained based on current design practice codes and standards. The well-known Housner’s model,
which assumes that the mass of liquid is lumped on the wall based on rigid wall boundary condition in the calculation
of hydrodynamic pressure, is widely used in practice. A comparison shows that in most cases, the lumped mass ap-
proach overestimates the base shear. The effect of wall flexibility on wall displacements, base shears, and moments are
also discussed.
Key words: reinforced concrete, liquid containing, rectangular tank, seismic, dynamic analysis, tank flexibility.
Résumé : Le présent article propose une procédure pour calculer les pressions hydrodynamiques dans les réservoirs
rectangulaires. Cette procédure, connue sous le nom de méthode séquentielle, tient compte de l’effet de la flexibilité de
la paroi du réservoir lors de la détermination des pressions hydrodynamiques. La présente étude ne considère que la ré-
ponse impulsive du réservoir. L’analyse dynamique variant en fonction du temps basée sur un modèle bidimensionnel
de la paroi du réservoir est effectuée afin d’étudier l’effet de la flexibilité de la paroi sur la réponse. Dans l’analyse, un
réservoir haut et un réservoir peu profond sont examinés. Les résultats de l’analyse sont comparés à ceux obtenus selon
les codes et les normes de pratique de conception actuels. Le modèle bien connu de Housner, qui présume que la
masse de liquide est amassée contre la paroi en fonction d’une condition limite à la paroi rigide pour calculer la pres-
sion hydrodynamique, est largement utilisé en pratique. Une comparaison montre que, dans la plupart des cas, l’ap-
proche de la masse amassée surestime le cisaillement à la base. L’effet de la flexibilité de la paroi sur les mouvements
de la paroi ainsi que les efforts et les moments de cisaillement à la base sont également abordés.
Mots clés : béton armé, stockage de liquide, réservoir rectangulaire, sismique, analyse dynamique, flexibilité des réservoirs.
[Traduit par la Rédaction] Chen and Kianoush 752
Fig. 1. Schematic of rectangular tank (a) three-dimensional (3-D) model of rectangular tank (b) two-dimensional (2-D) model of rect-
angular tank.
Fig. 2. Finite element model of rectangular tank (a) 2-D tank configuration (b) finite element model.
Housner’s model has been adopted in most of the current obtain the dynamic response of fluid storage tanks subjected
codes and standards for calculating the hydrodynamic pres- to earthquakes. Both impulsive and convective effects were
sures in concrete tanks. It is expected that the effect of the considered. Later they presented an analytical method with
flexibility of the tank wall is more significant in steel tanks formulas for the three-dimensional (3-D) hydrodynamic
than that for concrete tanks. For this reason, most of the re- pressure calculation and applied Rayleigh–Ritz method us-
search investigations have been carried out for steel tanks. In ing assumed vibration modes of rectangular plate with
this regard, little attention has been paid to concrete tanks. boundary conditions as admissible functions (Kim et al.
Yang and Veletsos (1976) used Flügge’s shell theory to ana- 1996).
lyze circular tanks. It was found that for tanks with realistic In this paper, a procedure, referred to as the sequential
flexibility, the impulsive forces are considerably higher than method, is used to determine hydrodynamic pressures for
those in a rigid wall. Veletsos (1984) considered the effect rectangular tanks. The method considers the effect of wall
of the wall flexibility on the magnitude and distribution of flexibility on impulsive pressures.
the hydrodynamic pressures and associated tank forces.
They assumed that the tank-fluid system behaved like a sin- 2. Hydrodynamic pressure
gle degree of freedom system and the base shear and mo-
ment were evaluated for several prescribed modes of Figure 1a shows a 3-D rectangular tank. It is assumed that
vibration. Most of the research conducted on liquid storage the liquid storage tank is fixed to the rigid foundation and a
tanks, as mentioned above have been of circular configura- Cartesian coordinate system (x, y, z) is used with the origin
tions. For rectangular tanks, Haroun (1984) gave a very de- located at the center of the tank base. Furthermore, it is as-
tailed analysis method in the typical system of loadings. The sumed that the width to the length ratio of the tank is so
hydrodynamic pressures were calculated using the classical large that the unit width of tank can represent the tank and
potential flow approach. The formula for hydrodynamic the corresponding two-dimensional (2-D) model, as shown
pressures only considered the rigid wall condition. Specific in Fig. 1b. It is noted that Lx and Lz are assumed to be paral-
to concrete tanks, Park et al. (1992) studied the dynamic be- lel and perpendicular to the direction of the earthquake, re-
haviour of rectangular tanks using boundary element model- spectively.
ing for the fluid motion and finite element modeling for the The fluid filled in the rectangular tank is of height Hl
solid walls. The time-history analysis method was used to above its base. The fluid, which is incompressible, inviscid,
Fig. 3. Comparison of impulsive pressure distribution in rigid Fig. 4. Procedure of sequential analysis.
wall models.
[3]
∂ φ
2
∂φ
(x, Hl , t) + g (x, Hl , t) = 0
[7] p= ∑ Ri u&& (t) dy
i =1
∂t 2 ∂y
where λ i = (2i–1)π/2Hl, u && (t) is the acceleration along the
(2) The boundary condition at the flat bed at y = 0, the ver- height of the wall, and Ri is a time dependent factor given by
tical velocity becomes zero in the horizontal ground mo-
∞
tion condition, that is 2ρl tanh( λ i Lx ) H
∂φ
[8] Ri = ∑ λ i Hl
cos (λ i y)∫ cos (λ l y)
0
[4] (x, 0, t) = 0 i =1
∂y
&& (t) dy 1
× u
(3) At the surface of fluid in contact with the tank wall x = && (t)
u
±Lx: The detailed derivation of the above equation is described
∂φ by Chen (2003). As the series in eq. [6] converges very fast,
[5] (± Lx , y, t) = u& (t) = u& g (t) + u& r (t) = u& (t)
∂x only the first term of the series may be used for practical ap-
plications.
where u& g (t) is the velocity of horizontal ground motion, and && (t) = u
For the rigid tank, u &&g (t). It means that the accelera-
u& r (t) is the transverse velocity relative to the ground motion tion along the height of the wall is same as the ground accel-
due to the flexibility of tank wall. eration, then eq. [6] becomes
Fig. 6. North–south component of El Centro accelerogram: 1940 Imperial valley earthquake record.
∞
2(−1) iρl In this study, two types of time-history analysis are used
[9] p= ∑ λ2i Hl
&&g (t)
tanh( λ i Lx ) cos (λ i y) u to investigate the dynamic response of liquid storage tanks,
i =1 the mode superposition method with the added mass ap-
proach and direct step-by-step integration method, which in-
This equation is the same as the hydrodynamic pressure
cludes the sequential method. The sequential method is
equation for the rigid wall derived by Haroun (1984).
discussed subsequently in this paper. The unconditional sta-
ble Wilson θ method is used for direct integration.
3. Dynamic analysis In design application, usually the impulsive hydrodynamic
pressure is applied on the wall using the added mass ap-
3.1. Finite element model proach. This can be calculated using the Housner method.
In this investigation, the dynamic response of the tank is Using this approach, the added mass can be either lumped at
obtained by using both empty and full conditions. The tank the one node or distributed over the height of the wall. In
wall is considered as 2-D plane elements fixed at the base terms of distributed mass this is expressed as
and free at the top, as shown in Fig. 2. The walls parallel to
the direction of horizontal ground motion are assumed to be 3ρl Hl ⎡ y2 ⎤ ⎛ 3Lx ⎞
rigid and have no significant effect on the flexibility of two [11a] m i (y) = − ⎢1 − 2 ⎥ tanh ⎜ ⎟
2 ⎣ Hl ⎦ ⎝ Hl ⎠
long sidewalls on which hydrodynamic pressures are ap-
plied.
or using the lumped mass
The hydrodynamic pressures are considered as external
forces in the proposed model, while they are treated as tanh [0.866(Lx /HL )]
[11b] M i ,total = ML
added masses in the Housner model. The advantages of us- 0.866(Lx /HL )
ing the proposed model will be explained in the following
section. Although only the impulsive pressure is considered The mass distribution can also be derived from eq. [9] using
in this paper, basically both impulsive and convective pres- the proposed model assuming that the walls are rigid. In this
sures can be calculated, and the dynamic response of a liq- case, the distributed mass can be expressed as:
uid storage tank can be analyzed using the proposed model.
∞
2(−1) iρl
3.2. Time-history analysis
[12a] m i (y) = ∑ λ2i Hl
tanh( λ i Lx ) cos(λ i y)
i =1
The governing equation of motion using the finite element
method can be expressed by and the lumped mass is expressed as
[10] [Mw]{ür} + [C]{ u& r } + [K]{ur} ∞
2
[12b] M i ,total = ∑ λ3 H 2 L tanh( λ i Lx ) M L
= –[Mw]{üg} + {P} i =1 i l x
where {ur}, {u& r }, {ür} are displacement, velocity, and accel- Figure 3 compares the hydrodynamic pressure distribution
eration of rectangular wall relative to the ground motion; for the Housner model and the proposed model for the rigid
{üg}is the horizontal ground acceleration in x direction; {P} wall condition (same as the model by Haroun (1984)).
is the hydrodynamic pressures on the wall surface; [K] is the Although there are some differences between the two formu-
stiffness matrix of rectangular tank wall; [Mw] is the mass lations, the difference in pressure distribution is not signifi-
matrix of rectangular tank wall; and [C] is the damping ma- cant. The difference in the total pressure, which represents
trix of rectangular tank wall. the area under the pressure curve, does not exceed 10%. The
Fig. 7. Displacement time-history response for the empty tank: (a) model 1, (b) model 2, and (c) model 3.
input parameters for these results are provided in the exam- Therefore the equation of motion with respect to the added
ple for tall tank analysis later in this paper. mass method can be expressed by
For the general condition, the hydrodynamic pressure in
terms of added mass method can be expressed by [14] [Mw + Mi]{ür} + [C]{u& r } + [K]{ur}
Schematic of model
As the added mass matrix representing the impulsive pres- hydrodynamic pressure P is determined, which also includes
sure is derived from the rigid wall boundary condition, this the effect of flexibility of the tank wall. Finally, the hydro-
approach cannot consider the change in hydrodynamic pres- dynamic pressure is applied on the tank wall at the next time
sure distribution because of the effect of wall flexibility. step t + ∆t. The procedure is then repeated at each time step
until the analysis is complete. Figure 4 shows in a flowchart
format the procedure for analysis, and Fig. 5 shows how the
4. Sequential method data is transferred between rectangular tank wall and fluid.
The equation of motion in terms of sequential method at
To consider the effect of flexibility of the tank wall on hy- time t can be expressed as
drodynamic pressures in dynamic analysis, the sequential
method is used in this investigation. The sequential method [15] &&r , t} + [C]{u& r , t} + [K]{ur , t} = −[M w ]{u
[M w ]{u &&g , t}
is a technique in which the two fields of fluid and structure
&&g , t − ∆t + u
− {Ri(t − ∆t)}{u &&r , t − ∆t}
are coupled by applying results from the first analysis as
loads or boundary conditions for the second analysis. Ba- where Ri(t-∆t) is time dependent function for impulsive hy-
sically the dynamic response of a liquid storage tank must drodynamic pressure.
be solved by strong coupled method, which is that the data If Ri(t-∆t) is treated as Mi(t-∆t), the time dependent func-
must be transferred or shared at each step of the solution to tion of added mass matrix related to hydrodynamic pressure
maintain accuracy of dynamic response analysis. From at time t-∆t, then eq. [15] becomes
eq. [10], the hydrodynamic pressure can be treated as the ex-
ternal forces applied on the rectangular tank wall, and the [16] &&r , t} + [C]{u& r , t} + [K]{ur , t} = − [M w ]{u
[M w ]{u &&g , t}
boundary conditions of a rectangular tank wall determines &&g , t − ∆t + u
− {M i(t − ∆t)}{u &&r , t − ∆t}
the hydrodynamic pressure in eq. [6]. Actually, these two
equations must be solved simultaneously because the inter- It can be seen that if the impulsive hydrodynamic pressure
action between the rectangular tank wall and the hydrody- due to the acceleration of the flexible wall is moved to right
namic pressure occurs at the same time. Since it is difficult hand side, eq. [14] is similar to eq. [15] except that the effect
to solve the dynamic response of wall and hydrodynamic of wall flexibility in calculating the hydrodynamic pressure
pressure directly from eq. [10], we can apply the sequential is ignored. After the hydrodynamic pressure at time t – ∆t is
method to approximate it. established, it can be easily applied as external force on the
The sequential method is carried out by the following pro- tank wall to obtain the dynamic response of tank wall at
cedure. First the dynamic response of the flexible tank wall time t. Furthermore, if the time interval ∆t is decreased in
subjected to an earthquake is analyzed at time t. Then the the calculations, the accuracy of final results may be im-
T1 = 0.3413 – –
T2 = 0.06365 – –
5% α = 1.552 α =0
β = 0.0027 β =0
2.69 2.65 2.16
4.76 4.56 5.030
314.8 319.0 284.0
290.0 338.1 202.9
32.68 23.04 0
30.11 26.90 (t=4.57 s) 0
Fig. 8. Displacement time-history response for the full tank: (a) model 1, (b) model 2, (c) model 3, (d) model 4, and (e) model 5.
Fig. 9. Dynamic response of wall and impulsive pressure distribution at the time of maximum response (model 5) (a) acceleration of
wall and (b) hydrodynamic pressure.
5.1.1. Response of empty tank This represents a typical model for tank wall used in most of
For the empty tank, three models are considered: the current codes and standards for design of concrete liquid
(i) equivalent lumped mass model, where the total mass of containing structures.
rectangular wall is lumped at the centroid of the wall; (ii), In model 2, the mass of wall is distributed over the height
and (iii) distributed mass model in which the mass of the of the wall whereas the impulsive mass of the liquid is
wall is distributed at nodal points. Both models 1 and 2 are lumped at height hi, as given by Housner (1963). Model 3 is
analyzed using the mode superposition method, while the the same as model 2, but the impulsive mass is determined
third model is analyzed using the direct step-by-step integra- using the proposed method assuming rigid wall condition. In
tion method. In the direct step-by-step integration method, a this case, the impulsive mass is lumped at height hi. Models
time step ∆t = 0.01 s is used for analysis. The results of 4–6 are also based on the proposed method. Model 4 is the
analysis are summarized in Table 1. The time-history re- same as model 3 except that the impulsive mass is distrib-
sponse in terms of displacement dA at the top of tank wall is uted over the height of the wall. In model 5, the hydrody-
shown in Fig. 7. namic impulsive pressure is determined considering the wall
Both Table 1 and Fig. 7 show that the results are very sim- flexibility. This model is expected to provide the most accu-
ilar with respect to the mode superposition method and the rate results among the six different models. Model 6 is the
direct step-by-step integration method for the mass distribu- same as model 5 except that the wall is assumed to be infi-
tion models (models 2 and 3). However, in the equivalent nitely rigid. In this case, the hydrodynamic pressure is deter-
lumped mass model, the maximum base shear FB is larger mined assuming a rigid wall condition. In models 1 to 4, the
than the mass distribution model. The mass distribution mode superposition method is used for dynamic analysis
model is expected to yield more accurate results than the whereas the last two models are analyzed using the direct
lumped mass model. The displacement dA at the top of step-by-step integration method including the sequential pro-
equivalent beam in model 1 is less than that of mass distri- cedure.
bution model. A summary of the analytical results for all models is
listed in Table 2. This is in terms of the maximum base shear
5.1.2. Response of full tank and top displacement during the time-history analysis. Peak
For the full tank, six conditions referred to as models are values corresponding to two different times at which maxi-
considered. In both models 1 and 2, the impulsive mass of mum values occur are shown. Fig. 8 also shows the displace-
liquid is determined using the procedure described by ment time-history response for all models. These results
Housner (1963). In model 1, both the impulsive mass and indicate that the response is considerably less for model 5 as
the inertial mass of wall is lumped at an equivalent height h, compared with the other models. The results also show that
which is determined by values obtained from the proposed model assuming rigid
M i hi + M whw boundary conditions (model 3) are similar to that of Housner
[19] h = model (model 2).
Mi + Mw
From calculation results, it is observed that for model 1,
where hi is the height from the base of the wall to the centre the results for base shear in particular are much higher than
of gravity of the impulsive lateral force, hw is the height those observed in other models. This gives an indication that
from the base of the wall to the centre of gravity of the tank the procedure adopted in the codes and standards are too
wall, Mi is the equivalent mass of impulsive component of conservative.
stored liquid, and Mw is the equivalent mass of wall. It is observed that the results of model 2 and 3 are very
In model 1, the period of vibration of the tank wall is de- similar. The difference in results between these two models
termined using the classical approach for a cantilever wall. does not exceed 10%.
Schematic of model
Fig. 10. Time-history response of shallow tank — proposed method (model 5) (a) top displacement and (b) base shear.
T1 = 0.1476 – –
T2 = 0.02798 – –
5% α = 3.579 α =0
β = 0.0012 β =0
2.69 2.28 2.16
3.59 3.57 3.40
4.95 4.56 4.87
73.2 78.7 74.5
78.9 54.6 49.2
74.3 64.6 59.4
278.2 241.8 179.5
300.0 167.8 119.2
282.4 179.2 143.0
5.48 4.54 0
5.91 3.15 0
5.56 3.25 0
For model 4, although the period of first two modes are The direct step-by-step integration method together with
not much different from models 2 and 3, there is a major dif- the proposed sequential analysis is applied to model 5. In
ference in the values of the base shear that is less for this case, the effect of flexibility of wall on hydrodynamic
model 4. This effect is due to the mass distribution. In these pressure and tank wall is considered. The base shear is in-
three models (2–4), the rigid wall boundary condition is creased because of the flexibility of tank wall as compared
used to calculate the impulsive pressure, which is approxi- with model 4 but the maximum displacement is somewhat
mated by added masses placed on the wall. smaller. It is observed from the results of calculations that
Fig. 11. Effect of flexibility of tank wall on response: (a) nor- Fig. 12. Effect of tank flexibility on wall acceleration and impul-
malized displacement spectrum, (b) normalized base shear spec- sive pressure distribution: (a) acceleration of wall and (b) hydro-
trum, and (c) normalized base moment spectrum. dynamic ressure.
cantly higher for model 1. In this case, the difference is that the dynamic response of liquid storage tanks calculated
195% as compared with 230% observed in the tall tank. using the current design codes or standards in terms of base
shear is too conservative. It is concluded that the effect of
the flexibility of the tank wall should be considered in the
5.3. Effect of wall flexibility
calculation of hydrodynamic pressures in concrete rectangu-
The shallow tank is used to study the effect of flexibility lar tanks.
of rectangular tank wall. In this case, the stiffness of rectan-
gular tank wall is varied using the modulus of elasticity of 7. Acknowledgements
concrete material, Ec, used in the tank wall.
The tank is analyzed using the direct step-by-step integra- The authors would like to express their appreciation to the
tion method and the sequential method (model 5). The Natural Sciences and Engineering Research Council of Can-
damping ratio, which is assumed to be 5% of critical, is ada (NSERC) for research grant to the second author that
based on the first two modes of vibration. enabled this work to be done. The authors would also like to
Under horizontal ground motion, the tank wall may de- thank Ryerson University for financial support provided to
velop cracking of concrete and yielding of reinforcement. As the first author during graduate studies that enabled him to
a result, the stiffness of the wall could be reduced consider- complete the M.A.Sc. degree, which is the basis for the re-
ably. For this reason, the tank wall is analyzed under differ- sults in this paper.
ent Ec values ranging from Ec = E0 for the fully uncracked
section up to Ec = 0.4E0 for the fully cracked section. For References
the latter case, the concrete is expected to be totally dam-
aged. It may be unrealistic to use such a small value of Ec Bathe, K.J., Wilson, E.L., and Peterson, F.E. 1974. SAP IV A
for liquid containing structures. However, the results provide structural analysis program for static and dynamic response of
linear system. College of Engineering, University of California,
some useful information on how the fully flexible tank wall
Berkeley, EERC 73-11.
could behave as compared with the more rigid walls.
Chen, J.Z. 2003. Dynamic analysis of concrete rectangular liquid
The results of analysis are presented in Table 4. Also, storage tanks. M.A.Sc. thesis, Ryerson University, Toronto,
Fig. 11 shows plots of the normalized response with respect Ont., Canada.
to E0. To reflect the dynamic response of the full tank sys- Currie, I.G. 1973. Fundamental mechanics of fluid. McGraw-Hill,
tem more precisely, the normalized response of these tanks Inc., New York, N.Y.
can also be presented in the response spectrum form. The EERI. 1976. Earthquake Engineering Research Institute, California
normalized natural frequencies of tanks with respect to f0 for Institute of Technologies, Report No. EERI 76-02.
the variable flexibility are used in the x coordinate as well. Haroun, M.A. 1984. Stress analysis of rectangular walls under
Both the first and the second natural frequencies have the seismically induced hydrodynamic loads. Bulletin of the Seis-
same proportion to f0 of which the tank has the modulus of mological Society of America, 74(3): 1031–1041.
elasticity equal to E0. Similar graphs are shown for base Housner, G.W. 1963. The dynamic behavior of water tanks. Bulle-
shear and moment. tin of the Seismological Society of American, 53(2): 381–387.
Figure 11 shows that the natural frequencies of vibration Kim, J.K., Koh, H.M., and Kwahk, I.J. 1996. Dynamic response of
decrease when the tank wall becomes more flexible. The rectangular flexible fluid containers. ASCE Journal of Engi-
base shears and top displacement increase with the increase neering Mechanics, 122(9): 807–817.
of flexibility of tank wall. Park, J.H., Koh, H.M., and Kim, J. 1992. Liquid-structure interac-
Figure 12 shows the acceleration and impulsive pressure tion analysis by coupled boundary element-finite element
distribution along the tank wall when the maximum base method in time domain. Proceedings of the 7th International
shear is reached. Figure 12a indicates that as the flexibility Conference on Boundary Element Technology, BE-TECH/92,
Computational Mechanics Publication, Southampton, England.
of tank wall increases, the acceleration along the height of
89–92.
tank wall increases. Figure 12b shows the increase in impul-
Veletsos, A.S. 1984. Seismic response and design of liquid storage
sive pressure distribution with the increase of the flexibility tanks. In Guidelines for the seismic design of oil and gas pipe-
of the tank wall. line systems, 1992, Albuquerque, N. Mex. Tech, Council on
Lifeline Earthquake Engineering, ASCE, New York, N.Y.
6. Conclusions Yang, J.Y., and Veletsos, A.S. 1976. Dynamic behavior of fluid-
tank system. Ph.D. thesis, Civil Engineering, Rice University,
The dynamic response of liquid storage tanks is studied in Houston, Tex.
this investigation. Using the proposed model, the hydrody-
namic pressures are no longer required to be approximated
List of symbols
by added mass, but they can be treated as external forces.
The advantage of the proposed model is that it can consider [C] matrix of damping coefficient
the effect of the flexibility of the wall in the calculation of Ec modulus of elasticity of concrete
hydrodynamic pressures during the time-history analysis. To FB base shear of tank wall
analyze the dynamic response of liquid storage tanks, the se- g acceleration due to gravity
quential method is applied. hi height above the base of the wall to the center of the
To demonstrate the efficiency of the proposed model us- gravity of the impulsive lateral force
ing the sequential method, the response of a rectangular tank hw height from the base of the wall to the center of gravity
is studied based on a time-history analysis. This study shows of tank wall