You are on page 1of 8

ANALYSIS OF SHELL WALLS OF CIRCULAR TANKS

UNDER AXISYMMETRICAL PRESSURE

FREDW. BEAUFAlTt
Department of Civil Engineering, Vanderbilt University, Nashville, TN 37235,U.S.A.

and

SOCRATES A. IOANNIDES and MAURKIA A. GERLEINS


Stanley D. Lindsey & Associates, Ltd., Nashville, TN 37215,U.S.A.

(Received 30 January 1978: received for publication 13 June 1978)

Abstract-The midpoint difference method is developed for solving the basic differential equation for the elastic
deformation of a thin shell wall of a circular tank subjected to axisymmetrical pressures. The method of analysis
permits the consideration of varying wall thickness, changes in strength of wall material (modulus of elasticity), the
secondary bending effect due to the presence of axial forces (PA effect), and any variation of lateral pressure with
respect to depth. The method of analysis is organized for a computer solution. Several examples are presented to verify
the method and to illustrate its versatility.

INTRODUCTION point difference method [3] for solving the basic


The elastic deformation of a thin shell wall of a circular differential equation of the structure. The method has the
cylinder subjected to an axisymmetrical pressure can be advantage that the influence of the axial load in the wall,
described by the differential equation i.e. the PA effect, can be taken into consideration. The
analysis is developed in terms that are familiar to most
$ (D$ )+$!Y=P. structural engineers.

SIGNCONvENTION
where D = Et3/12(1 - CL*).Defining the parameters: E is The method of analysis leads to the evaluation of the
the modulus of elasticity of the wall material, CLis the bending moments (M,, MB), the shear force (S,) and the
Poisson’s ratio of the material, pX is the lateral pressure circumferential axial force (NB) as defined in their
at any depth x, r is the radius of the cylinder, t is the
assumed pbsitive senses in Fig. 1. It is assumed that the
wall thickness and y is the lateral displacement of the
wall at any depth x, where x is measured from the top of axial force N, is known. In addition to determining the
the wall or cylinder [ 11.The solution of this basic differen- internal actions, the lateral deflection and the slope of the
tial equation for specific boundary conditions and specified wall at specified depths x are evaluated. The unit strip of
loading can be rather difficult, especially if the thickness wall to be used in the analysis is shown in Fig. 2 with
of the wall or the modulus of elasticity vary over the respect to the assumed x-y axis system. An outward
height of the structure or should the variation of the pressure is assumed positive; also, the assumed positive
pressure be complex. sense of any applied boundary moment, shear or axial
Billington[l] has suggested a method for analyzing’ force is as indicated in Fig. 2.
shell walls of constant thickness and molulus of elasticity
utilizing the flexibility approach (force method). A table
of functions required for computing the flexibility
coefficients of the wall at the boundaries is made avail-
able. However, the basic displacements of the wall, with
no constraints at the top or bottom, must be obtained for
the particular loading, this may limit the usefulness of
the method. The shell wall can be analyzed by the finite
element method [2] using an axisymmetric element, but
this approach will be of limited value to the engineer who
is not very familiar with the finite element method and
does not have access to a working program.
It is the purpose of this paper to present a method for
analyzing the shell wall utilizing a numerical solution of
eqn (1). The method is rather general in nature in that it
presents no special problems to the user when analyzing
shell walls with varying thickness and modulus of elasti-
city, with a complex variation of axisymmetric loading
and with elastic boundary constraints. The proposed Fig. 1. Internal actions shell wall element.
approach to the analysis of shell walls utilizes the mid-
FINMULATION OF METHODOF AML.YSIS

tprofessor of Civil Engineering, Vanderbilt University, Nash- Equation (1) is seen to be of the same form as the
ville, TN 37235,U.S.A. basic uerential equation’ for thi elastic curve of a beam
Wructural Engineer. on an elastic foundation. With this in mind, eqn (1) can
531
532 F. W. BEAUFAIT et al.

Pressure
-44

Segnnmtnumbcr--I ~Statimk2bel
x
Fig. 3. Partitioning and stationing of wall strip.

By partitioning the wall strip into R segments, n + 1


stations will be established (Fig. 3). It is not necessary that
the segments be of equal length; the strip should be
lx
partitioned so that stations will be located at important
Fig. ?. Unit wall strip. points of change in either the loading, the geometry, or the
material properties of the wall. Equation (4) can now be
be rewritten as [41 written for the midpoint of each segment. For any segment
i each of the basic variables, defined at the midpoint of the
segment, in eqn (4) can be approximated as an average of
the values of the variable evaluated at the stations defming
the boundaries of the segment, i.e.
where Z=t’/12(1-~z) and K=Et/?. The term
N, d2y/d.r2which has been inserted takes into account the Y!+v2 = 1/2(y{+,t yi’, (51
influence of the presence of the axial load in the wall. Both
the axial force N, and the lateral pressure pXmay vary with where f is 1,2,3 or 4. The derivative of each of the basic
the depth x. Equation (2) takes into consideration only the variables, evaluated at the midpoint of the segment i can be
flexural deformation of the shell wall.
Since the structure is assumed to be axisymmetrically
loaded, a unit vertical strip can be used in analyzing the
wall (Fig. 21.
At any depth x four variables can be defined:

Y”Y (deflection) (3a)

dy (slope)
Y2'&

d2y
y'=E+=M, (bending moment) (3c)
(a) Top of wall

and

(shear). (Ml

DitIerentiating the relationships of eqns (3) with respect to


x and recalling eqn (2), results in

Y’ 0 1 0 0 y’ lo
Y2
y3
Y4
=
0

-K
0
0
-N,
0
l/EI
0
0
0
1
0
y2
y3 ’
y4
1
k
0
0
1 px
(4)

L J L Jl- J L _I
where the prime denotes differentiation with respect to x.
Now, the basic differential equation for the shell wall Cb) Bottom of wail
reduces to the lirst order vector, differential equation
defined in eqn (4). Fig. 4. Elastic boundary constraints and fixed end actions.
Analysisof shetl wallsof circulat tanks under a$symmetricJ pressure s33

approximateas ry’l

-
dyf+,,,, 1
= 3;: (YL - yff (6)
dr I

whereht is the lengthof segmenti, The errorassociated


andfor a free end where moment and shear equal zero
withboth of these a~ro~ons is of the orderof ck*,
where c is a constant.
Usingthe approximations givenby eqns(5)and(6),eqn
(4) can be written for the burnt of a segment, e.g.
segment i, in termsof the basic variablesdefinedat the
bokiary stationsi and it 1: L

Y’ Y'

or
; -1
-&0K21 -_
!+.
i

-2f);i
2EJt

----xi
0O*1 ;I
---f0;,
Y2

YJ
Y”
Ji I_
Y2

Y’
Y’

If the top of the shell wall were subjected ta elastic


constraints and applied actions, as would be the case if
the top of the tank were built rnonol~~~iywith the wag
Lsegmentiabei Lstationlabel (Fig. 4af, the two ~undary equations would take the
form I- -I
whereI$, A+1%; Nsi and pXiare evaluatedat the midpoint
of the segment, i.e. segment i, or at station i + X/2.The
parameter pXhas not changed its originaldefinition;it is
simply tbe m~itude of the lateralpressureat the mid-
point of the segment. lotion (7) is referred to as the L J
rni~poiff# ~~~erenc~ equation; it can be writtenfor each and if the base of the wall were subjected to elastic
and every segment specified when the wali, strip is constraints (Fig. 4b), the two boundary equations would
p~tion~. This results in 4n ~qu~ns in termsof 4n + 4 be
unknowns(the basic variables at each station). The four
additional equations are obtained from the boundary r,;l -mix
y3 = c 1
1Y4y2
conditions. 0 k,,, 1 0
(13)
Two equationscan be written for each ~un~y of the k, 0 01 F&l‘
wall strip, i.e. at station 1 and station n + 1. If a boundary n+i
is hinged, the two constraint equations (deflection and L J

moment equal zero) can be expressed as In eqns (12)and (13), k,,, is the ro~tion~ stiffnessof the
boundary constraint, k, is the translational stiffness of
the boundary constraint, Fhf is the applied moment and

][I
Y’ EP is the applied shear at the boundary. Any of the
10 0 0 y* = 0 various boundary constraints can be expressed as
(9)
ootoy3 II0’
[ YJ USI{Y’jj = (Plj (141
wherei=lorn+l.
For a fixed boundary, the two constraint equations Writingthe midpoint difference equation, eqn (7Xfor
(deflectionand slope equal zero) would be each segment and the appropriate boundary equations
for either end of the wall strip,yieldsa set ~f*equations

(1s)

t_

or (161
534 F. W. BEAIJFAITet al.

The coefficient matrix [/3] is nonsymmetric; however it analysis, S2, is carried out using segment lengths of h/2,
has a narrow band width of 11 which can be taken an improved and essentially exact solution, S,. can be
advantage of with the equation solver. Because of the found for any function (or value) by substituting into the
weak nature of the system of equations, any equation expression
solver must be capable of interchanging columns, as well
as rows, when searching for the best pivotal element[3].
Otherwise, a zero or almost zero term might be encoun-
tered in the diagonal position.
The midpoint difference method for solving eqn (2) is EXAMPLES
ideally suited for programming on the computer. The To illustrate the use of the midpoint difference method
reader is directed to Ref. [3] for a discussion on how a in analyzing shell walls, three examples are presented.
general computer program might be organized.
Solving eqns (16) yields the lateral deflection y, the Example 1
slope dy/dx, the bending moment M, and the shear force In order to show a comparison of the proposed mid-
SYat each station specified along the height of the wall. point difference method with a more classical solution,
In addition, the circumferential force Ne can be deter- the liquid storage tank analyzed by Billington[l] was
mined from the relationship [ 1] selected for analysis. The tank is described in Fig. 5. The
modulus of elasticity was taken as 3000 ksi
(20685000kN/m2) and Poisson’s ratio was assumed to be
l/6. The structure was analyzed for the condition where
the tank is filled with water. The base of the wall was
and the circumferential bending moment MB can be assumed to be fixed; the top of the wall is free.
evaluated from the relationship [ 1] The shell wall was partitioned into 60 segments for the
analysis. Twenty segments of 0.25 ft (76.2 mm) length
MB= I.LM~ (18) were used at the top and bottom of the tank and 10
segments of 0.5 ft (152.4 mm) were used in the middle
at any station. As mentioned earlier, the axial load NX portion of the wall. The smaller segments were used at
must be known before solving eqn (2). the top and bottom in order to get a better evaluation of
As stated earlier, the midpoint method of analysis has the four variables within the end regions of the tank. The
an error associated with it of the order of ch2. A relative influence of boundary conditions is greatest near the
high degree of accuracy can be achieved by using a large boundaries; it dampens out in the wall rather quickly.
number of segments. It is also possible to improve on the Also, the gradient of any of the variables might be high
accuracy of the solution by employing an extrapolation in the vicinity of the boundary.
technique, such as Richardson’s method: If one analysis, The results of the analysis are given in Table 1; plots
S,, is made using segments of length h and a second of the variation of lateral pressure, circumferential force
N,,, and bending moment M, are shown in Fii. 6. (These
graphs were drawn using a Calcomp Plotter.) The
maximum value of the circumferential force NB(Fig. 6b),
is 17,430Ibs (77,529N) at a depth of 11.5ft (3.51 m). This
t-1.25’ r=27’ agrees with a value of 17,260lbs (76,772N) at ap-
proximately the same location found by Billington; these
values are within 1%. The maximum negative bending
moment (Fig. 6c), is - 2601 ft-lbs (- 3527 Nm) at a depth
of 13.5ft (4.11 m); Bill&ton gives a value of
- 2590 ft-lbs (3512 Nm). The maximum positive bending
Fig. 5. Liquid storage tank. moment, Fig. 6(c), occurs at the base of the tank. A value

n
m.00 2o.w

l6.W 16.00 1600


II

pressnxld N-F- x lti Mommtr102


p,, tldft2 N,. Ibs y, ft-lbs
(a) (b) w
Fig. 6. Analysis of water tank, top free and based fixed.
Analysis of shell walls of circular tanks under axisymmetrical pressure 535

Table I. Analysis of liquid storage tank-example I.

of 9591 ft-lbs (13,005 Nm) was found by the midpoint The translational stiffness of the roof slab is[l]
diierence method; the value given by Billington is
9650 ft-lbs (13,085Nm) which agrees within 1%.
k, = F (1 - cc)= 19,200,000~ (280,180,980N/m).
Example 2
BiBington[ll also analyzed the concrete storage tank The fixed end moment acting at the top of the wall due to
shown in Fig. 7 for a roof loading. Again the modulus of <he roof loading is [ 11
elasticity, E, is taken as 3000 ksi (20,685,OOO
kN/m’) and
Poisson’s ratio, p, is l/6. A uniform roof load of 19Opsf FMT _ 4? _ (190~27)~
(9097 N/m? was assumed; this includes 40 psf (1915 N/m’) 8 8
snow load. The top of the wall is monolithic with the roof
= - 17,313ft-lbs (- 23,477Nm).
slab: the base of the wall is assumed, in accordance with
Billington, to be hinged.
Results of an analysis of the tank by the midpoint
difference method are given in Fig. 8; the loading,
deflected shape, and variation of I$, and M, ace shown.
The maximum bending moment (Fig. &I), was found to
be - 16,447ft-lbs (- 22,302 Nm) at the top ‘of the wall
where it joins the roof slab. This compares within 1%
with the value of - 16,340ft/lbs (- 22,157Nm) obtained
by Billingtontll. The maximum value of cricumferential
force Nh Fig. 8(c), was found to be 13,426lbs (59,719N)
at a depth of 3.75 ft (1.143m); this compares within 2%
with the value of - 16,340ft-lbs (- 22,157Nm) obtained
Fig. 7. Water storage tank. Billington[l].
The value of the circumferential force, Nb, at the top
of the wall was computed to be - 3549lbs (- 15,785N);
For the analysis, the tank wall was partitioned in the Billington computed a value of - 4256 lbs (- 18,930N) at
same manneT as was done in Example 1, using 60 seg- this boundary. There is a 16.6% difference between these
ments. The rotational stiffness of the circular roof slab is two values. In order to check tbe midpoint difference
found to bet11 analysis, the autbors analyzed the tank using 120
segments (halving the segment length used in the tirst
analysis). From this second analysis a value of - 3549Ibs
(- 15,785N) was found, again, which indicates that the
536 F. W. BEAUFAIT et al

!zlefkxthl x lo-2 N-Force x IO* Moment x IO3


P, Y. ft N,, ES Mz, ft-lbs
(a) (b) Cc) (d)

Fig. 8. Analysis of water tank for roof load

Rarsvc XIO’ N-Fwce x IO3 Momentx103


&.s bs/fta A$, Ibs M, ft-lbs
(a) (b) (cl

Fig. 9. Analysis of full water tank with roof load.

midpoint difference method is giving a nearly exact Some results of the analysis of the tank for this
answer. Upon examining Billington’s solution, it was loading condition are given in Fig. 9. The deflected shape
seen that he had used only 3 significant figures in his of the wail follows the same curve as that of the circum-
computations. The authors found that a 1.7% change in ferential force (refer to eqn 17).
the value of only one of his redundants, the shear at the
top of the wail, results in a 16.2% change in his value for Example 3
Ne at the top of the wall. Thus, the difference between The reinforced concrete silo described in Fig. 9 was
the two methods of analysis is actually not as bad as it designed for the storage of run-of-the-mine coal having a
first appears. Studying Fig. 8(c), it is seen that the rate of unit weight of 50pcf (801kg/m’). The silo wall is 180ft
change in NB in the region of the top of the wall is very (54.86m) high and has a diameter of 60 ft (18.29m); the
high; hence, NB would be sensitive to round-off. wall is 0.75 ft (228.6mm) thick. For the first 28 ft (8.53 m)
The storage tank of Fig. 7 was also analyzed for the of height of wall, the concrae has a strength of 4500 psi
case where it was full of water along with the roof load (31.027.5kN/m*); for the next 24ft (7.32m) it has a
of 190psf (9097 N/m?. strength of 4000psi (27580kN/m’); and for the remainder
This analysis was carried out taking into consideration of the wall, the concrete has a strength of 3OOOpsi
the presence of axial load acting on the wall; the axial (20,685kN/m’). The roof of the silo is not connected to
load is produced by the roof load and the weight of the the wall; it simply rests on the top of the wait. Hence, the
concrete wall. The magnitude of the axial load must be top of the tank is assumed to be free. At the bottom, the
computed at the midpoint of each segment. In this silo wall is keyed into the foundation but not monolithic
analysis, the base was assumed to be fixed. with the foundation; thus, the base of the wall is
Analysis of shell walls of circular tanks under axisymmetrical pressure 531

assumed to be hinged. Poisson’s ratio is taken as l/6 for V = 750X,- 58,870(1- e-0~0’274x0).
the concrete.
The internal pressure and vertical wall force produced The vertical force is given in units of lbs/ft of width of
by the stored coat were computed using the equations wall. In evahratjng the lateral pressure and vertical force,
proposed by Safarian[5]. The .lateral internal pressure the following values were used for the parameters: the
was determined from the exnression (from ean 13. Ref. hydraulic radius of the silo is 15ft (4.57 m), the
coefficient of friction between the concrete wail and the
L = G{1071.45(1 -e-“.o’274x0)I. stored coal equals 0.7, the angle of repose of the stored
coal equals 35”, and the ratio of lateral to vertical pres-
Where X0 is measured down the wall starting at sure for the stored coal was taken as 0.273.
a point 12ft (3.66 m) from the top. (When the top The silo wall was analyzed taking into account the
at a point 12ft (3.66 m) from the top. (When the top presence of an axial loading resulting from a roof load of
surface of the stored material is sloping toward the walls 22OOIbs/ft of wall (32,105N/m), the weight of the
of the silo, the depth X0 is measured from the centroid of concrete wall, itself, and the vertical force caused by the
that material above the level where the material first stored material. The wall was partitioned into 160 seg-
strikes the wall.) The coefficient Cd is taken as 1.35 for ments; segment lengths of 1 ft (0.305 m) were used for
0 5 X0 5 42, 1.75 for 56 s X,, 5; 168, and varies linearly the upper 12ft (3.66 m) and lower 20 ft (6.10 m) of wall
between 1.35 and 1.75 for 425 X0= 56. The lateral pres- and 2ft (0.61 m) segment lengths were used over the
sure, L, is given in units of psf. The vertical wall force interior portions of the span. Some of the results of the
produced by the stored coal was determined from the analysis are given in Fig. 11; these curves were
expression (eqn 18, Ref. [S]) developed using a Calcomp Plotter. Of particular interest
is the influence the change in concrete strength has on
the behavior of the silo, i.e. the development of internal
forces and moments.

7 B To illustrate more clearly the influence of the variation


in concrete strength, i.e. the modulus of elasticity, the
silo of Fig. 9 was modified by assuming a constant
strength of 3000psi (20,685kN/m*) for the concrete over
the height of the structure. This was the only change
p-o.7
k-0273 made. The important results of this analysis are shown in
w-50 pcf
Fig. 12. Notice the less erratic variation of the circum
ferential force, Nh (Fig. 12b) and the bending moment,
M, (Fig. 12~).

CONCLUSION
w The midpoint difference method offers a numerical
method for solving the basic differential equation of a
shell wall which is well suited for programming on the
computer. It has the advantage over other difference
methods in that segments of varying length can be used
without any difficulty. This makes it easier to use finer
meshes in regions where one or more of the four basic
variables might have a high gradient, e.g. at boundaries,
in order to better account for localized effects. The
method also requires very few special sets of equations
Fig. IO. Reinforced concrete silo. to formulate an appropriate system of simultaneous al-

I cew
.tz

P 72.W

.2.37

Mocl*lr x IO7 Rapnrs XIO’ N-Force xd l8hWftX102


E. wft2 p,. bdft2 4. h Me ft-its
(a) lb) (cl (d)
Fig. 11.Analysis of silo, varying modulus of elasticity.
538 F. W. BEAUFAIT et al.

3&00- 1.67

/
,
ODO SW0 lexu .m
Raauc x IO’ N-Fwcex103 rvlcinmx102
p,. lbd N, ~JS M, ft-lbs
(a) bl (cl

Fig. 12. Analysis of silo, constant modulus of elasticity.

gebraic equations required for the analysis of a shell RDEREhTEs


wall. The obvious disadvantage of the method is the 1. D. P. Billington,Thin Shell Concrete Structures. McGraw-Hill,
number of equations that must be solved in analyzing a New York (1965).
shell wall. This disadvantage is counterbalanced by the 2. K. C. Rockey. H. R. Evans, D. W. Griffiths and D. A.
narrow band width of the resulting equations. Another Nethercot, The Finite Element Method, A Basic Introduction.
Crosby Lockwood Staples, London (1975).
disadvantage is that the equation solver must be able to
3. F. W. Beaufait and G. W. Reddien, Midpoint difference
search both rows and columns to find the appropriate method for analyzing beam structures. Comput. Structures
pivot element. The development of an equation solver to 8(6). 745-751 (1978).
take into account the constant narrow band width and to 4. W. G. Godden, Numerical Analysis of Beam and Column
use full pivotal condensation is not that difficult a task. Structures. Prentice-Hall, Engiewood Cliffs, New Jersey
The authors have developed an efficient equation solver (1%5).
as part of a general computer program for analyzing shell 5. S. S. Safarian. Design pressure of granular meterials in silos.
walls by the midpoint difference method. AC1 J. Proc. 66(8) (Aug. 1%9).

You might also like