Vous êtes sur la page 1sur 30

First of two Papers Paper No.

6827
for joint discussionat an Ordinary Meeting
on Tuesday, 23 February, I965 at 5.30 p m . ,
, and for subsequent written
discussion*

DESIGN OFTHE DOME SHELL ROOF FOR


SMITHFIELD POULTRY MARKET
by
Povl Ahm, M.Sc., M.Ing.F., M.I.C.E.
Associate Partner, Messrs Ove Arup & Partners
and
Edwin John Perry, M.E., A.M.I.C.E.
Senior Engineer, M e w s Ove Arup & Partners

SYNOPSIS
The shell roof described in this Paper covers the hall of the new Poultry
Market at Smithfield. In shapeit is part of an elliptical paraboloid.The
horizontal projection is a rectangle 225 ft 0 in. X 127 ft 9 in. and the rise is
30 ft from corners to centre. It is supported on columns along all four sides.
The shell is of reinforced concrete with prestressed edges. The thickness is
generally 3 in., increased to 63 in. at the edges.The Paper deals with the
conditions leading to the concept of the shell and describes the preliminary
design, based generally on the membrane theory, which was carried out to get
an approximate idea of the stresses involved.This led to the decision to
carry out a model test with the principal purpose of determining the load at
which the shell would fail by buckling. A description is given of the more
rigorous analysis based on the bending theory and the results are compared
with the results of the preliminary analysis and in particular with the results
obtained from the model test. The Paper can be read as a separate entity,
but a more complete picture is presented if it is read in conjunction with the
associated Paper No. 6716 by L. L. Jones and G . D. Base, which describes
the model test.

* Written discussion on this and the following Paper should reach the Institution
before 15 March, 1965, and will be published after July 1965. Contributions should
not exceed 1200 words.
79
80 AHM AND PERRY ON DESIGN OF DOME SHELL

NOTATION*
U,, u2, u3, u4 arbitrary constants
coefficient -
1
Q:
l+@
as suffix indicating edge beam
constant in equations for buckling load
modulus of elasticity
rises of shell
coefficients, see equation (15)
stiffness of shell = Et3/12
bending moments in shell
twisting moment in shell
root of characteristic equation
the coefficient Q, d ( r x / r y )
order of harmonic in Fourier series
direct forces in shell
tangential shear force in shell
as suffix indicating particular integral value
maximum prestress force in edge beam
coefficient, see equation (15)
principal radii of curvature for shell
transverse shear forces in shell
as suffix indicating shell
thickness of shell
displacements of shell
displacement of shell edge or edge beam
force at shell edge or in edge beam
co-ordinates of shell corners
vertical load
buckling load
coefficient nn/2xo in Fourier series
coefficient 3/t2r;
coefficient ry/rx
stress function
slope of shell surface at edge
angle of rotation of shell or edge beam
Poisson’s ratio
subtended angle at buckling
Laplace operator
the (--+-7)
operator
r1y ddx22 1 drax d y

* The same notation is used in the associated Paper No. 6716.


ROOF FOR SMITHFIELD POULTRY MARKET 81
INTRODUCTION
SMITHFIELD has been used as a market probably since 1150 A.D. It served as
a live marketfor 700 years until 1855 when thisfunction wasmoved to
Islington. It was then decided to build a market for meat and poultry on the
site of Smithfieldand thebuildings were designedand built in the period 1866-
1875 by Sir Horace Jones, who was Architect and Surveyor to the City of
London.
2. Interrupted only by World War 11, the market functioned right up to
1958 when the Poultry Market was destroyed by fire. In 1959 it was decided
to rebuild the Poultry Market on the same site with the possibility later of
extending the new building gradually to replace the existing ones. Sir Thomas
Bennett was commissioned to carry out the design.

THENEW MARKET
3. Meatis transported to and from the market by road. Theproblems
governing the design of the building are therefore the movements of vehicles
in and aroundthe market, loading and unloading, and transport andstorage
of the meat within the market.
4. The Architect first proposed a two-storey building in which the road
traffic andthemarketproper wereseparated.However, it proved too
difficult to solve the transport of meat within the market, and it was decided
to revert to the traditional type of market with the traffic and loading bays on
the periphery and the market hall in the centre.
5. Offices are located above the loading bays, with direct access to the stalls
in the market hall. The basement is used for cold storage, smaller rooms on
the outside are occupied by the market tenants, and a large storage space in
the middle is let to a cold-storage company.
6. Because of the natureof the subsoil, which is London clay, a heated mat
was introduced between the basement and the surrounding soil to prevent
frost-heave of the foundations.
7. TheMarket Committeewanted astructuremore fire-proof than the
previous one, and reinforced concrete was therefore chosen as the natural
material for the structural frame. The new building covers a total area of
263 ft X 269 ft (Figs 1 and 2).

ROOFOVER THE MARKET HALL


8. The Market Hallis 225 ftlongand 127 ft 9 in. wide (Fig. 3). For
functional and aesthetical reasons it was decided not to have any supports for
the roof inside the Hall. The possible solutions for the structurewere limited
to those which could span the whole of the Hall, supported only along the
periphery.
9. Varioustypes of frameswithsecondarybeamswere investigated but
generally work was concentrated on shell structures, which are able to cover
large areas in a light and elegant way. Cylindrical shells as well as doubly-
curved shells spanning between frames were considered, but the frames or tie
beams cutting into the space of the Hall proved unsatisfactory.
10. The dome covering the wholearea of theMarket Hall was finally
selected as the optimum solution from functional, aesthetical and cost con-
siderations without necessarily being the cheapest solution.
82 AHM AND PERRY ON DESIQN OF DOME SIIBU.
84 OF DOME
AHM A N D PERRY O N DESIGN SHELL

1 1 . A shell of these dimensionswasdifficult to balance with the rest of


the building and the rise had to be kept to an absolute minimum.
12. Working to conventionalspan-to-height ratios the rise should have
been 45-50 ft which was considerably in excess of what could be accepted.
After some approximate calculations it wasdecided to use a rise of 30 ft
(Fig. 4) which was less than any shell built so far, although it seemed possible
from consideration of the maximum stresses and a preliminary assessment of
the buckling load. The shell is shaped as an elliptical paraboloid expressed
analytically by the formula
2 = ( x / r 2 + (Y/rY)z

to 6:' at
edge beam

FIG.3: PLANOF ROOF

FIG.4: SECTION AT CENTRE-LINEOF ROOF


FOR ROOF SMITHFIELD POULTRY MARKET 85
13. The rise in the long direction of the shell (the Y-axis) is 20 ft while the
rise in the short direction (the X-axis) is 10 ft, giving a total rise of 30 ft from
the corners to the top. The difference in rise in the two directions is made
in order to even out the maximum stresses occurring in thesurface of the shell.
14. The shell is supported on columns along the periphery of the Market
Hall and an attempt was made to make all columns carry approximately the
same load. The edges of the shell are prestressed to counteract the tension in
the edge beams and to reduce the incompatibility between compression in the
edge zones ofthe shell and tension inthe edge beams.
15. The edge problem might have been solved by using a bow-string girder
with horizontal ties, but then all the.weight of the shell would have beencarried
by the four corner-columns, which would have been disproportionately large.
It would also have increased the cost of the shell plus the edge beams.
16. The roof is provided with a small overhang on all\four sides, which
was added after the shell was constructed and thus has little influence on the
stress distribution in the shell. If it had been constructed with the shell much
of the effect of the prestressing would have been lost.
17. The shell is generally 3 in. thick but the thickness is increased in the
edge zones to a maximum of 63 in. because of the bending moments. The
edge beams are 2 ft 11 in. deep X 1 ft 6 in. wide and contain the l + in. strand
cables which are used to prestress the edges.
18. The roof is covered with copper, laid on an insulation of lightweight
screed on expanded PVC bonded to the concrete shell by asphalt. The insula-
tion serves the dual purpose of reducing expansion and contraction of the
large shell surface with varying outside temperatures and preventing condensa-
tion on the soffit of the concrete shell.
19. A number of 5-ft dia. holes is provided in the shell surface for lights.
The concrete round the edges is thickened to make up for the loss in strength
caused by the holes.

PRELIMINARY
DESIGN
General
20. The shell consists of 3-in. reinforced concrete, covered with a layer of
expanded PVC and copper sheeting on a 3-in. screed. This gives a weight of:
3-in. reinforced concrete 38 lb/sq. ft
Expanded PVC lb/sq.
ft 1
Screed 10 Ib/sq. ft
Copper lb/sq. 3 ft

52 lb/sq. ft
21. At the time the designwas carried out the insulation consisted of a
bituminized cork compound which was considerably heavier than the PVC.
The total dead load used for the design was 61 lb/sq. ft.
22. Thelive load consists of wind forces and snow, with the snow load
of 15 lb/sq. ft governing the design.The fact that the snow load can act
asymmetrically is taken into account.
23. Theshortcomings of the membranetheory for the analysis of shell
structures are wellknown.However, the theoryis useful to obtain an
approximate idea of the maximum direct stresses occurring in the shell.
86 AHM A N D PERRY ON DESlGN OF DOME SHELL

I t I I

FIG. 5: PLANSHOWING GAUGE AND COLUMN POSITIONS AND AXIS

(a) DIRECT FORCES

(b) BENDING FORCES AND DISPLACEMENTS


FIG.6: SIGNCONVENTION FOR FORCES AND DISPLACEMENTS
ROOF FOR SMITHFIELD POULTRY MARKET 87
24. It is emphasized that although the bending moments in the shell surface
are important, shells carry the major partof their load through direct forces.
This is particularly true of doubly curved shells of positive gaussian curvature
where the curvatures in the two principal directions have the same sign. The
importance of the bending moments, in particular in relation to the ultimate
failure of the shell, will not be sufficiently known until much more research
has been carried out in this field.

Membrane theory (Figs 5 and 6)


25. The assumption having been made that no bending moments occur in
the shell, the direct or membrane forces are governedby the well-known
differential equation

26. This equation is derived from the equations of equilibrium only and
thus expresses a statically determinate solution to the distribution of forces,
internal as well as external.
27. Airy’s stress function (4) is defined as follows

and Z is the vertical load.


28. If 2 is developed into Fourier series as follows
4 1 nn
Z = 2 Z, = .
n n
2 0 2.
nn
sin - cos a,x (n = 1,3,5,
2 .. .) . (3)
nn
a, = -
2x0
a particular solution to the differential equation can be found:
. . . . .(4)
29. The general solution to the homogeneous equation is given by
3
4 = (AnlemnY+A,2e-mnY)cos a,x . . . . (5)

m, = d G ya,
30. The arbitrary constants A,, and A,, are determined by the boundary
conditions
Ny = 0 for y = fyo.
31. The complete solution to the differential equation is then1.

from which the forces can be found.


32. The solution expresses the stress function 4 in terms of theratio
(xo/yo)(ry/rx)f.For shallow shells the stress function and the forces can be

l The references are listed on p. 107.


88 AHM A N D PERRY ON DESIGN OF DOME SHELL

LB/FT
28000

\ 24000

20000

I6000

I2000

8000

Edge
ROOF FOR SMITHFIELD POULTRY MARKET 89
expressed approximatelyby the ratiory/rxonly and may thus be plotted against
the ratio of the two rises h,/h, to enable a wide range of shells of varying rises
to be dealt with. A. L. Parmel has done this and has also given correction
factors for non-shallow shells.
33. The values of the forces in this shell were found from Parme’s tables
by graphical interpolation and are shown in Fig. 7.
34. The ratio of the rises of 2:1, which corresponds roughly to the plan
dimensions, was selected to give the most satisfactory distribution of stresses
in the membrane and also to give the shell a pleasant appearance.

Bending moments in the edge zone


35. An approximate analysis of the bending moments in the edge zone of
the shell perpendicular to the edge was required to determine the extent to
which it was necessary to thicken the edge. The membrane theory naturally
yields no information about the bending moments which werefound by using
the results from theanalysis of a cylindrical shell supported along its two ends.
In other words, curvature of the shell perpendicular to the edge was neglected.
36. The basic differential eauation is3.

The thickness of the shell ( t ) is assumed to be constant.


37. For no moment restraint at the edge (W= 0,M = 0 at edge) the solution
is

38. For full fixity at the edge ( W= 0, dw/dy = 0 at edge) the solution is
Z
M = - 2p- - fl(yo-y) {cos p(y0 -Y) - sinp(yo-v)} . . (8b)
39. Fig. 8 shows the bending moments for the two extremes. The actual
bending moment will lie between the two curves as far as the approximation
used is valid.
40. The approximate results are compared later in this Paper with more
exact calculations and with the experimental results.

Buckling
41. The preliminary design of a structure must contain an assessment of
the ultimate strength as well as an analysis of the elastic condition. For a
shell of the given dimensions it was reasonable to expect that the ultimate
failure would take place by buckling of the shell surface.
42. When an attempt was made to predict the failure load it was realized
that theanalysis of buckling was still in avery preliminarystage and the failure
load could not be reliably calculated. If the shell were to be built, a model
test would have to be carried out. It would have to be tested to failure and
at the same time the elastic condition should be observed and compared with
the theoretical analysis.
90 AHM AND PERRY O N DESIGN OF DOME SHELL

-t
m
_1
H -
t 8
0
2 0
m
t 7
I

I
!
i

I
-1-

0
I
0
N
-
E
m
0
8
I I !

0
N

l
D FOR ROOF POULTRY MARKET 91
43. Bucklingof tubes under axial loadhas been investigated by many
authors, includingTimoshenko and Gere5 and Fliigge.s The general
expression for the critical stress is
t 1
ucr = E -
.? 1 / [ 3 ( 1 - P ) ]
which for p = 0 gives

44. The result isbased on smalldeflexions.Experimental results have


given much smaller values and Lundgren7 has suggested using a factor of 0.2
for concrete. Thisagreeswith a theoretical result by Karmknand Tsien8
based on large deflexions.They found
t
u0, = 0.195E;.
The value to assume for E has naturally been the subject of considerable
discussion.
45. For spherical shells similar results havebeen found. Love’stheory
assuming small deflexions gives

Experimental values are again lower and one might use a value of
t
U,, = 0.2E;.

46. Based on these considerations a maximum stress in the shell of the


order of 700 Ib/sq. in. was thought to be reasonable. This corresponds to the
maximum membrane stress under the given load occurring in a 3-in. shell with
a total rise of 30 ft, rises ratio of 2:l and radii of curvature of 316.4 ft and
204.8 ft.
47. From the formula above is found
z,, = U,,- 2t = 1.156E(;)2.
r
Assuming large deflexions, Karman and Tsien8 have found

with half angle subtended by buckled segment


I-

Tsieng found later

with
W = 8O.3
92 AHM A N D PERRY ON DESIGN OF DOME SHELL

Edge beams
48. The edge beams carry the load of the shell which in the preliminary
analysis is transferred only through membrane forces. The beams follow the
line of the edge of the shell and are supported on columns alongthe whole of
their length, the tension increasing from zeroat theends to amaximum in the
middle. Prestressing was used to counteract the tension and the prestressing
force varied accordingly.
49. The edge beams were analysed as normal continuous beams and the
analysisshowed that the load wouldbe transferred almost evenly to the
columns. In the preliminary analysis the prestressing force was designed also
to counteract the bending moments in the beams for full load on theshell and
to produce a residual compression in the edge beam compatible with that in
the edge of the shell for dead load plus half live load acting on the structure.
This would reduce the bending moments in the edge of the shell to aminimum.
50. A description is given later of how the model test showed the influence
of the prestressing on the ultimate load for the shell. Thus the prestressing
force was finally determined by the ultimate load instead of as originally
intended by minimum incompatibility at working load.
SpeciJication f o r model test
51. The principal purpose of the modeltest was to show the mode of failure
of the shell and the load at which failure would take place. It was expected
that the failure would be by buckling of the surface. To assess the reliability
of the model test as a tool of design and to compare it with the theoretical
analysis it was decided to test the shell also through the elastic phase.
52. The following requirements were therefore decided upon.
(a) the model should be a scale model of the final structure;
(b) the model should be in reinforced mortar with strength and elastic
properties similar to those of the concrete to be used for the shell;
( c ) the edgebeams of the model should represent the actual support
conditions as closely as possible;
( d ) the load should be scaled so that stresses in the model would be the
same as those in the final structure;
(e) live load on the model should beplacedsymmetrically as well as
asymmetrically about both centre lines ;
(f)strain and deflexion measurementsshould be carried out on the model
for comparison with calculated values ;
( g ) the model should be tested for the following loading:
(i) dead load,
(ii) dead load +live load,
(iii) dead load + 2.0 X live load,
(iv) 1.5 dead load + 2.0 live load,
(v) load to failure,
(vi) being the ultimateload which the shellgenerallywas
required to carry; and
(h) column loads should be measured.
53. Since the material for the model had to be reinforced mortar and $-in.
is the minimum thicknessfor that material, the scale was fixed as 1 :12.*
* A full description of the model test is presented in the associated PaperNo. 6716.
which could be read with benefit at this stage.
ROOF FOR SMITHFIELD POULTRY MARKET 93

RIGOROUS ANALYSIS
General
54. To enable one to assess the results of the model test and to draw the
correct conclusions for thefinal design, a more rigorousanalysis than the one
based on the membrane theory was carried out simultaneously with the test.
5 5 . The analysis is based on the bending theory using boundary conditions
which were as close as possible to those of the actual structure. It was based
on work by WlassowlO in which he has neglected all second order terms and
made geometrical simplifications as follows :
(a) the coefficients in the first quadratic form for thesurface of the shell
+
are equal to one, ds2 =dxa dya, and
(b) the principal curvatures are assumed constant over the whole of the
surface.
56. Wlassow has shown thatthese assumptions arevalid for surfaces where
the rise is small compared with the length (1 :5). The two 4' order differential
equations can then be written
+
V4$ EtV,w = 0
K V 4 w - V# = Z . . . . .(9)
where

by eliminating $ :
KVew+EtViw = V4Z . . . . . . (10)
57. A solution to this differential equation can be found by developing Z
into single Fourier series
.
z xz' -. n
4 1 na
- Z o 1 - s i n -cosa,x
,,n 2
(n .
1, 3, 5 , . . .), a .
nv
-
- 2x0
58. A particular integral is then

W, = E BnZnCOS QI,X . . . . (11)

where

and
59. To find a solution for the homogeneous equation,
W = aelny cos ax . . . . . . (12)
yields the characteristic equation
(m2 - 4 4 + 4p4(yma - a2)2 = 0 . . . . (13)
94 AHM AND PERRY ON DESIGN OF DOME SHELL

U,
2

jt"
II
4
h

5
ROOFFOR SMITHFIELD POULTRY MARKET 95

4
*
I
U"
3
96 OF DOME
AHM AND PERRY O N DESIGN SHELL

which has the following roots


m - 4 = k(jl+ikl)
nz5-8 = +(j2+ik2) . . . . . . (14)
where

.
j2 =
J 2/[(a2-p)z+(q+py)2~+(a2-P)
2
2jlkl = 4-82>.
- p+u2
j f - k 21 -
(15)

k2 = Jl/[(n2-p)2+(q+~2v)z~-(n2-p) = q+Py
2 jz"- k - u 2 - p
60. The cylindrical shell is a special case of this general equation and by
introducing y = 0 and p = q = ap the familiar roots of the Jenkins equation are
obtained.
61. The expressions above may be developed into a series form for manual
calculation but this is unnecessary when the analysis is programmed for com-
puter. The first term in the expansion gives the approximate roots derived
by Bouma."
62. The homogeneous solution can now be written
+ + +
W = [e-jlY(blcos k,y b2 sin kly) ejly(b3cos k,y bq sin kl y )
+
e-j2Y(bs cos k 2 y + b6 sin k 2 y )+ e'2Y(b7 cos k z y + bs sin kzy)] cos ax
. , . . . (16)
63. For a symmetrical shell this can be reduced to
W = [a, cos hj, y cos kl y + a2 sin hjl y sin kl y
+a3 cos hj2y cos k2y +Q sin hjzy sin k 2 y ] cos ax . . . . (17)
where U,, u2, u3, and u4 are arbitrary constants.

(a) FORCES

(b) DISPLACEMENTS
FIG.9 : FORCES
AND DISPLACEMENTS AT EDGES
ROOF FOR SMITHFIELD POULTRY MARKET 97
64. Now that the complete solutionforthe has beendeveloped, the
W
displacements and forces can be expressed in terms of W or 4 by linking 4
with W with:
V:$ = KV'w
65. The results are given in Tables 1 and 2, the contents of which can be
written in matrix form
{ W = HNKCI{d; { G P }
{X.} = {k'}{MKCHa); {xm} . . . . . (18)
where {k}and {k'} contain themultipliers, { N } and { M }the rootsof the charac-
teristic equation, { C } the trigonometrical terms and{a}the arbitrary constants.
The displacements and forces considered are
{v.} = {aulax, 0, e}

as shown in Fig. 9.
66. {a} is determined from the boundary conditions at edges y = +yo.
The edge beams are cut free from shellthe and the continuity
expressed through
the displacements.
67. The stiffness relationship for the edge beam is, in matrix form,
{ub} E {K){xb+xbp} S . * (19)
where {Xbp} denotes effects from external forces acting on theedge beams and
{K}is the stiffness matrix.
68. The edge beam is prestressed to counteract the loadbending moments,
thus w b =0. For horizontal effect the columns were replaced by a continuous
wall of uniform height having the equivalent stiffness.
69. Thehorizontal bending andtorsional stiffnessis assumed to be
provided solely by the columns. Thus the shear forcebetween shell and edge
beams may be considered to act in the centroid of the beam. Thus {K}=

where A = area of edge beam,


I = equivalent height of wall,
I, = moment of inertia for columns,
n = number of columns, and
d = total distance alongside.
70. These assumptions arevalid for theshell analysis only and do not apply
for the analysis of the edge beams.
71. { x b p } contains effects fromthe prestressing force only. This force,
which increasestowards themiddle, is expressedas P cos ( m / 2 ) ( x / x 0 )and thus
has no lower Fourier terms.
4
98 AHM AND PERRY ON DESIGN OF DOME SHELL

72. Equations of equilibrium and continuity give


{Xd = - WHXB + X*,}
{U,) = - {L){U. + USPI . . . . . (20)
where {Xsp}and { Clsp} are particular integrals and

. cos#, sin+, .
L =
. -sin #, cos +,.
. 1 . J
4, is slope of surface at edge,
LB/FT
.- 36000 L ~ ~ F T
/-m0

- 32000 l !

- 28000

- 24000

\
- 20000
\
. ____ LBI FT
---_- -----_____
.c____-

- l2000
- 80M)
h - ZBWO
-IQ30

Edge . Nr for .v= 0


-Full derlgn load
-- -- Own weight on!y,
(a'

FIG. 10: FORCES


FOUND BY
BENDINO THEORY
ROOF FOR SMITHFIELD
MARKETPOULTRY 99
73. It should be noted that the use of single harmonic series governs the
boundary conditions at edges X = & X,,. The support conditions here will be
similar to those for a shell hinged to inextensible diaphragms alongthe edges
(M, = 0, N, = 0, U =0, W = 0). A complete analysis of the shell will therefore
involve two calculations with the axis changed.
74. Equations (18), (19), and (20) give

for the determinationof {a}.


75. The solving of this equation requires the manipulation of ten 4 X 4
matrices which can be set up on a computer programme.
76. When {a} has been determined, forces and displacements can be found
from the expressions in Tables 1 and 2. For forces and displacements which
are not given in the tables, similar expressions can easily be found from the
basic equations. This calculation can also be expressed in matrix form and
incorporated in the computer programme.
77. A Ferranti ‘Sirius’ computer was used and the time for calculating five
loading terms was 6 min.
78. N,, Ny and M, for y = 0 and N,, along both edges are shown on
Fig. 10. They have been calculated for the boundary conditions applied to
edges X = + x o and the load developed in series in the y-direction, i.e. by
using the described theory with X and y interchanged in all formulae.

COMPARISON OF ANALYSIS WITH TEST RESULTS


General
79. A comparison between experimental and calculated results can be
madeonly when one knows theaccuracy withwhich the test results are
100 AA NH DM PERRY ON DBSION OF DOME SHELL

obtained. The accuracy is mainly governed by the accuracy with which the
model correspondsto the actual structure and the accuracywith which strains
are measured. Construction tolerances onthe modelwere kept within
, .
& 1/16-in. which corresponds to ++in. on the actual shell.
80. The scatter of readings was not fully assessed but it was shown that
it was independent of the strain value. When stabilized voltage supply was
used it was found that 90% of the readings had a variation of less than.
+ 15 lb/sq. in. about the mean stress. This represented about k 3% of the
stress values near the edges or -t 6% of the average smaller stresses in the
middle of the shell. The final scatter whichincludes scatter onreadings
obtained from non-stabilized voltage supply is probably larger.
81. The effect of the roof-light openings was not taken into account in
the calculation. The tests showed that their effect is only local.

Elastic stage
82. Columnloads. Table 3 gives the load in each column-expressed as
a percentage of the total load-obtained from the twotests on themodel and
from analysis. The measured values agreed well with those calculated except
at the corners where the total load on the three corner columns agreed well
but was distributed in a different way to that calculated. This may be due to
the. special corner conditions of the model.

TABLI!
3 : COLUMN
LOADS

Column load a % of total load


on roof
-
column Tests
Calculated
Loading I Loading II
-_
A 5.0 1.9 3.5
1 1.1 2.4 1*6
2 3.3 3.1 3-6
3 4.1 3.9 4.1
4 4.2 3.8 4.5
5 4.6 4.1 4:5
6 3.8 3.7 4.1
7 3*9 3.6 3.6
8 0.9 2.7 1*6
B 4.9 2.6 3.4
9 1.o 2-0 1-4
10 3 *O 3.1 2.4
11 3.6 3.5 3.9
26 4.1 4.3 4.2
27 2.5 4.0 3-2
28 0.1 1-5 0.2

Loading I-prestress and dead load applied.


Loading 11-load increment of 60 Ib/sq. ft
ROOF FOR SMITHmBLD POULTRY MARKET 101
83. When the three corner columns are considered together the measured
values of load in all columns lie within +28% of those calculated and all
except two within -I 18%. The variation between measured loads on sym-
metrical columns indicates the accuracy of the test and is 9% about themean. +
84. Thus reasonable agreement is found between measured and calculated
values, which implies that the load is being carried by the structure in a way
close to that predicted by the analysis.
85. Forces. Curves forinternal forces found by analysis and tests are
shown in Fig. 11. The test curves are foundby combining curves for prestress
and loading whichwere found separately. The laboratory readings of the
strains were converted into forces by assuming a value of the elastic modulus.
The forces shown in Fig. 11 are based on a value of E = 4.5 X 106lb/sq. in.

-j8000
-.-.-
- - - - - - - - Calcu;aed bending theory
membrane theory
Testa j-
Curver’a’reprerent forces for lnltial prestress L
on model Edge A‘, along row 4
Curves% correspond to final prestress
LB/FT

4000
1
l
:I
I
Edge Nxralong row 4

Fro. 1l : COMPAJUSON
OF TEST R E ~ U L ~
WITH RESULTS OBTAINED FROM
“BRANB AND BENDINO THEORY
102 AHM AND PERRY ON DESION OF W M E SHBU
which was found by considerations of equilibrium of one quarter of the shell.
This shows reasonable agreement with the value of E obtained in the laboratory
which was E = 4 . 0 x 10elb/sq. in.
86. The curves for N , and Ny show good agreement between tests and
analysis except for a local reduction in the test results on the centre lines.
This can possibly be explained by the use of air bags for loading the shell.
The air bags did not meet along the centrelines and calculations have shown
that a reduction in load along the centre stripwould be sufficient to produce
effects of this magnitude.
87. The N, and Ny test curves also have a sharp depression up to the
boundary, which is not reflected i n the calculated curves. Also although it
appears consistently in the combined curves it does not appear in the experi-
mental curves for load and prestress taken separately. Several explanations
have been attempted, such as inaccuracy in combining theresults and also the
effect of edge thickening of the shell, but no satisfactory solution has been
found.
88. Perhaps it is not necessary to look fsr a solution, but onewould expect
the values of internal forces created from different loading conditions but in
the same structure to be in phase. If they were in phase the combined curve
would have the same characteristic shape as the individual curves from which
it was formed.
89. The curves for Nxy again show reasonable agreement, except for some
distortion adjacent to the cable anchorages. The high calculated value at the
corner is not confirmed by the test, aswas expected, since corners tend to reveal
the shortcomings of simplified assumptions.
90. To compare the bending moments in the shell, the curves for M, along
gauge row 4 are shown. The bending moments for load and prestress have
opposite signs, and as the curves slope steeply near the edge, it is difficult to
assess accurately the maximum values. Nevertheless the combined curve
agrees well with the analysis.
91. Deflexions. The upwards deflexion at the centre of the shell for pre-
stress and dead load of concrete was found to be 0.7-in. in calculations aswell
as tests. The deflexion on the shell proper was 0.5-in. Since the E-value for
the shell was larger than that for the model, the agreement is good.

Ultimate stage
92. The shell failed in buckling at a load of190 lb/sq. ft and the angle
subtended by the buckled segment was 9.5". The load-&flexion curve was a
straight line right up to failure, which indicates that the structure was still in
the elastic condition when it failed.
93. The effect of long-term loading of the structure could not be assessed
from the model but the total dead-load had been applied continually for a
period of four months when the load was increased to failure.
94. It is thought, however, that the shell was mainly carrying the load
throughmembrane forces and creep would therefore have no significant
effect on the ultimate load. The suddenness of the failure without any prior
btnding distortion of the surfaceseems to be evidenceof this. Further evidence
is required to draw any firm conclusions on this point.
95. Based on the theoretical work by Karman and Tsiene for buckling
ROOF FOR BMTrHFIELD POULTRY MARKRT 103
of spherical surfaces Schmidtla has suggested an empirical formula for the
buckling of translation surfaces

where RIand Ra arethe principal radii of curvature andC is a constant, which


he attempts to determine experimentally.
96. Z,,= 190 lb/sq. ft and E = 4 5 X 106lb/sq. ft give a value of C in the
above formula of 0.308 which compares with Tsien’slO constant for spheres
of 0.312 derived theoretically. An ultimate load of 190 lb/sq. ft gives a factor
+
of safety of 1.5 X dead load 6.5 X live load, but since the factor is an expression
of uncertainty of design method rather thanuncertainty of strength of
materials and loading it would be correct to express it as a factor of safety
of 2.5 on the total load.
97. During the model test the edge beams showed signs of cracking under
dead load+ 2 X live load. It was decided to increase the prestressing by 40%.
+
98. Cracks appeared again underthe load of 1 X dead load 2 X live load
which was the factor of safety aimed at for the design generally,not including
buckling.
99. The corners of the shell were then fixed to a stiff frame as described in
the associated Paper No. 6716 and this frame stayed in position until failure
occurred by buckling.
100. Failure would no doubt have taken place earlier if the prestressing
had not been increased and the shell subsequently restrained.

CQNCLUSIONS AND FINAL DESIGN


Conclusions
101. The column loads measured in the model test agree reasonably well
with those calculated. The distribution of forces found from the test was
similar to that obtained by analysisusing the bending theory. There was
also good agreement between direct forces found by the bending theory and
those found by the membrane theory. In other words, the membrane theory
gave areasonable answer tothe problem.This is due to the boundary
conditions beingclose to those required by the membrane theory, mainly
because of the prestressed edges.
102. It is therefore considered that the model gave a reasonably accurate
representation of the behaviour of the final structure and that the method of
calculation was satisfactory.
103. As the tests showed that thelive load on the entire shell created larger
stresses than the load on one half of shell, it was decided not to analyse the
asymmetrical loading case.
104. The initial prestressing forces in the edge beams of the model were
probably lower than estimated becauseof friction losses.Buteven if this
had not been the case the prestressing force would still have been inadequate
and the most important single result of the model test was to show that it was
not correct to determine the amount of prestress by considerations of minimum
incompatibility at working load. The prestress should be increased beyond
that to improve the ultimate strength of the shell and to obtain as near as
possible a straight load-deflexion curve right up to failure.
104 AHM AND PERRY ON DESfGN OF DOME SHEU
Final design (Figs 12 and 13)
105. The final design of the shell was based on the bending theory. The
boundary conditions were adjusted for the actual stiffness of the columns.
The prestressing force was increased to 1.5 times the initial value after losses
to prevent tension in the edge beam and to produce a straight load-deflexion
curve near to the failure point. The bending moments in the edge zone were
adjusted accordingly.
106. The prestressing forceproduced an average compression stress of
1700 lb/sq. in. at transfer in the edge beams. Theelasticshortening was
therefore considerable and displacements of column-tops of up to +-in. were
expected.

- Due
---_ Due
to prerrress\
to total load
Due t o own weight of concrete

FIG. 12: BENDINGMOMENTS IN EDGE BEAMS


ROOF FOR SMITHmELD POULTRY m 105
107. Thecolumns weregenerallyconsideredfixed at the levelof the
surrounding flat roof as this would provide an additional tie force and give
extra safety. The stiffness of the columns did not, however, permit the neces-
sary movement at top.
108. A joint, which would act as a hinge during prestressing and provide
fixity afterwards, was therefore designed (Fig. 14). Freyssinet bearing pads
carry the vertical load and allow horizontal movement and rotation. Column
starter-bars pass through slots in the pads into
steel tubes cast into thecolumns.
After prestressing, fixity was obtained by grouting the ducts and concreting
the pockets around the rubberpads.

Starter bars

FIG.14: DETAILS
AT CORNERCOLUMNSSHOWINGBEARINGSDESIGNEDTO
ACCOMMODATE
MOVEMENTS DUE TO
PRESTRESSING '

109. It was estimatedthat creep of theconcrete wouldincrease the


curvature of the shell as tests and calculations showed that the shell would lift
from the shuttering when the prestressing force was applied. A calculation
allowing for the effect of the additional dead load applied subsequently and
the reduction of prestressing force confirmed this.

CONSTRUCTTON
110. Construction work began on site in January 1960 and the scaffolding
for the dome began to rise in April 1961.
111. The formwork supported on the scaffolding was designed to fit the
8 ft X 4 ft standard sheet of plywood. Since plywood can curve in one direc-
tion only, the shape in the other direction had to be formed from 4 ft long
straight lines. To follow the theoretical surface as close as possible the ply-
wood was curved in the short direction, held by formers which were cut to
the smaller radius. The plywood sheets werecut to shape as they were placed.
112. Care was taken to ensure the accuracy of the formed surface.
Tolerances in level of ?&-in. were worked to generally, but in addition, a
maximum rate of change in tolerance of +-in. in10 ft wasspecified. Spot
levels were taken on a grid of 12 ft 6 in. X approximately 11 ft over the whole
surface.
113. The reinforcementin the membraneconsists of two layers of &in. dia.
mild steel bars, parallel to the edges, in each face, with +-in. dia. bars placed
106 AHM AND PERRY ON DESlGN OF DOME SHELL

centrally to space the layers accurately. When the +-in. spacerbars were
placed to coincide with the crossing pointsof the +in. bars, the reinforcement
mat was extremely rigid, and thus reduced the likelihood of displacement of
thebarsduringconstructionandensuredthatthe coverwas maintained
accurately.
114. Within the edge zone, the spacing of the +in. bars perpendicular to
the edge was reduced to 6-in. Across the corners the size and spacing of the
diagonal bars was adjusted to carry the heavy tensile stresses.
115. The edgebeamswerestressed by I t i n . strand cables,using the
Gifford-Burrow anchoragesystem. The longeredge beams contain 22 cables,
of which four extend the full length of the beam; the short beams have 16
cables, withtwo extendingthe full length. The cables werelaid in metalsheath-
ing and held in position on bars welded to the links.
116. The casting sequence adopted for concreting the membrane was to
progress from the four corners towardsthe crown, so that fresh concrete was
always cast against theupper face of previously-placed concrete. This
prevented the concrete from slumping away from the old face, and gave the
most satisfactory construction joint. The roof area wasdivided into three
. longitudinal strips, each approximately 40 ft wide, and each strip divided into
11 panels, 20 ft wide. The casting of adjacentpanels in the longitudinal
direction of the shell was delayed by a minimum of one day, and in the short
direction by two days, to reduce the effect of shrinkage. The concrete was
made with rapid-hardeningcement.
117. The scaffolding was not perfectly rigid and membrane action could
develop before the shell was complete. In order to prevent cracking of the
edge beams the prestressing cables extending thefull length of the beams were
lightly stressed prior to the casting of the shell.
118. To find the elastic modulus, special test specimens were made from
the concrete mixea in addition tothe normal cubesused for testing the
strength. The average value obtained at 28 days was 5.3 X 106lb/sq. in., and
the lowest value 4.6 X 106Ib/sq. in. The average compression strength of the
concrete was 7100 lb/sq. in.
119. The final prestressing of the edgebeamsbeganwhen the concrete
had reached a strengthof 5000 lb/sq. in. Working from thecorners, sufficient
cables to carry the dead weight of the structure were stressed in a strictly
symmetrical sequence. The intermediatecables were then stressed with much
greater latitude in the symmetry of the operation.
120. During the stressing operation, it was possible to observe the lifting
of the shell surface from the shutter at the roof-light openings. It was found
that lifting became noticeable when the prestress calculated to balance the
dead weight had been applied and increased steadily as the prestresswas
increased. Over some areas of the roof, the shutter adhered to the under
surface, and was lifted away from the scaffold. The weight of these sections
added to'the dead weight of the structure, and an increase, in the upward
deflexion of the roof was observed as the shutter was removed.
121. The prestressing of theroof was completedtowardstheend of
September 1962 and the Market was officially opened in May 1963.
ROOF FOR SMITHFIEID POULTRY M A ” 107

ACKNOWLEDGEMENTS
122. The new Poultry Market was designed by SirThomas Bennett & Son.
The Consulting Engineers wereOve Arup & Partners and the main contractors
were Sir Robert McAlpine & Sons Ltd. The model was built and tested by
at Wexham Springs,
the Cement and Concrete Association in their laboratory
as described in the following Paper.

REFERENCES
1. A. L. PARME. Hyperbolicparaboloids and other shells of doublecurvature.
, Proc. Amer. Soc. civ. Engrs, Paper No. 1057, 1956.

2. H. TOTTENHAM.Approximatesolutions to shell problems. Proc. 2nd Symp.


Concr. Shell Roof Constr. Norwegian Engineering Society, Oslo, 1958.
3. K. GIRKMANN.‘Flachentragwerke’. 5th ed., Springer Verlag, Vienna, 1959.
4. R. S. JENKINS.‘Theory and design of cylindrical shell structures’. Ove Arup
& Partners, London, 1947.
5. S. P. TIMOSHENKO and J. M. GERE. ‘Theory of elastic stability’. 2nd ed.,
McGraw-Hill, New York, 1961.
6. W. FLUGGE.‘Statik und Dynamikder Schalen’. 2nd ed., Springer Verlag,
Berlin, 1957.
7. H.
LUNDGREN. ‘Cylindrical shells’. Danish Technical Press, Copenhagen,
1949.
8. TH. VON K~RMAN and HSUE-SHENTSIEN. Buckling of spherical shells by
external pressure. J. Aeronaut. Sci., vol. 7, December 1939, pp. 43-50.
9. HSUE-SHEN TSIEN. Atheory of the buckling of thin shells. J. Aeronaut. Sci.,
vol. 9, August 1942, pp. 373-384.
10. W. S. WLASSOW.‘Allgemeine Schalentheorie und ihre Anwendunginder
Technik’. Akademie Verlag, Berlin, 1959.
11. A. L. BOUMA. Someapplications of the bendingtheory regarding doubly
curved shells. Proc. Symp. Theory of Thin Elastic Shells, Deljt. North-Holland
Publishing Co., Amsterdam, 1960.
12. H. SCHMIDT.Ergebnisse von Beuhersuchen mit doppelt gekriimmten Schalen-
modellen aus Aluminium. Proc.Symp. Shell Res,Deljt, North-Holland
Publishing Co., Amsterdam, 1961.
108 AHM A N D PERRY ON DBSIGN OF DOMB SHELL
ROOF FOR SMITHFIELD POULTRY MARKET

The Paper, which was reccivcd on 3 March, 1964, is accompanied by two photo-
graphs and twenty-six sheets of drawings, from which the Figures in the text have been
prepared.

Vous aimerez peut-être aussi