ARTICLE IN PRESS

International Journal of Pressure Vessels and Piping 85 (2008) 728 738

Contents lists available at ScienceDirect

International Journal of Pressure Vessels and Piping

journal homepage: www.elsevier.com/locate/ijpvp

Stress and deection analyses of oating roofs based

on a load-modifying method
Xiushan Sun, Yinghua Liu , Jianbin Wang, Zhangzhi Cen
Department of Engineering Mechanics, Tsinghua University, Beijing 100084, China

a r t i c l e in f o

a b s t r a c t

Article history:
Received 22 June 2007
Received in revised form
24 March 2008
Accepted 27 March 2008

This paper proposes a load-modifying method for the stress and deection analyses of oating roofs
used in cylindrical oil storage tanks. The formulations of loads and deformations are derived according
to the equilibrium analysis of oating roofs. Based on these formulations, the load-modifying method is
developed to conduct a geometrically nonlinear analysis of oating roofs with the nite element (FE)
simulation. In the procedure with the load-modifying method, the analysis is carried out through a
series of iterative computations until a convergence is achieved within the error tolerance. Numerical
examples are given to demonstrate the validity and reliability of the proposed method, which provides
an effective and practical numerical solution to the design and analysis of oating roofs.
& 2008 Elsevier Ltd. All rights reserved.

Keywords:
Floating roof
Rainwater load
Load-modifying method
Deection
Nonlinear analysis

1. Introduction
Floating roofs are widely used in the middle- and large-scale
cylindrical tanks for crude oil and other liquid hydrocarbon
storages around the world because of their advantages such as
reducing product evaporation, improving safety, overall operating
economy, etc. After a history of over 80 years with continual
development and improvement, modern oating roofs with larger
diameters for open-top tanks can be classied usually into two
common types: single-deck type and double-deck type [15]. The
single-deck oating roof consists of characteristically a circular
deck plate and a pontoon (i.e. a compartmented buoyant ring)
which are both constructed with thin plates and jointed together
by a connection component, e.g. an angle-iron ring. To meet the
increasing capacity of oil storage tanks and to improve the
performance of the traditional single-type and double-type
oating roofs, a new-style oating roof with continuous
beams was also developed [6]. This oating roof has more
complex components, which increases somewhat the difculty
of structural analysis.
In the practical operation, the oating roof is usually subjected
to rainwater loading resulting from the accumulated rainfall.
The rainwater loading will result in a much larger deformation
(or deection) in the deck compared with the plate thickness. In
many codes for the design of oating roofs, the whole structure is

 Corresponding author. Tel.: +86 10 62773751; fax: +86 10 62781824.

E-mail address: yhliu@tsinghua.edu.cn (Y. Liu).

0308-0161/$ - see front matter & 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ijpvp.2008.03.003

required to possess good performances such as strength and

stability under a standard rainfall of 250 mm over the tank [7,8],
i.e. no failure modes such as fracture, buckling or sinking should
occur in this rainwater loading. Accordingly, stress and deformation analyses of oating roofs under rainwater loading are
practical problems to be solved.
However, the oating roof is actually subjected to complex
loads and deformations during the operation. The loads and
deformations of oating roofs are nonlinearly coupled with each
other, which results in the difculty of analysis. Mitchell [9]
investigated the problem of oating roofs with pontoon, in which
the deck plate was treated as membrane and the membrane large
deection equations were solved numerically by assuming a range
of starting values. But the proper selection of these values was
usually difcult and, of course, important to the solution. A similar
method was also used by Epstein et al. [3,10,11] to analyze
deformations and stresses for different types of oating roofs,
including pan oating roofs, pontoon oating roofs with accumulated rainwater loading or with punctures in the deck, in which
the effects of various parameters such as tank diameter and
pontoon geometry were also examined. Umeki and Ishiwata [12]
improved Epsteins solution and better computational efciency
was achieved, and they replaced the original RungeKutta
numerical method by the Milne method. Another analytical
method, i.e. the ODE-solver (ordinary differential equation solver)
method, was proposed by Yuan et al. [13]. This method was used
to solve the large deection equation of oating roofs based on the
bending theory rather than the membrane theory. To simplify the
problem, some authors [4,14] also presented calculating formulas

ARTICLE IN PRESS
X. Sun et al. / International Journal of Pressure Vessels and Piping 85 (2008) 728738

Nomenclature

R1,R 1

a
c1,c2

R2,R 2

C
E
f(r)
fmax
F
g
Gs
h0
hc
hs
hw
ha
H0
H1, H2
Hg
i
KL, KNL
M
Mc
Nr
Na
Nv
pb(r)
q(r)
qc
qs
qw
Q
r
R0

displacement vector in the FE equation

deformation coefcient of the outer rim and inner rim
of the pontoon, respectively
ratio of increments of water and liquid heads
Youngs modulus of the oating roof
deection of the deck plate
maximum deection of the deck plate
restoring load vector in the FE equation
9.8 N/kg, gravitational acceleration
weight of the oating roof excluding the deck plate
typical rainfall
equivalent deection of the deck plate
liquid head in the tank
water head on the deck plate
sinking depth of the oating roof due to slope of the
pontoons bottom plate
installing height of the deck plate
heights of the outer rim and inner rim of the pontoon,
respectively
sinking depth of the oating roof due to its weight
number of iteration in the load modication
linear and nonlinear stiffness matrices in the FE
equation, respectively
total mass of the oating roof
mass of the deck plate
number of radial continuous beams
number of annular continuous beams
number of vertical ribs
net pressure on the bottom plate of the pontoon
net pressure on the deck plate
weight of deck plate per unit area
liquid pressure applied on the deck plate in the tank
rainwater loads on the deck plate
applied load vector in the FE equation
radial coordinates of the oating roof
radius of the tank

for the large deection of the deck in oating roofs. These

formulas, however, were based on a water test condition in which
the loads on the deck plate distribute uniformly. In addition, with
the development of computer modeling and corresponding
numerical methods in modern engineering and sciences, the
nite element method (FEM) was also employed in the structural
analysis of oating roofs. Uchiyama et al. [15] and Yoshida [16]
analyzed oating roofs under rainwater load by a nonlinear
axisymmetric FEM, and special program codes for analysis of
oating roofs, THANKS V-III and KOSTRAN, were, respectively,
used in these two studies to compute the deformation and stress.
The above methods for analysis of oating roofs were usually
based on the axial symmetry theory, and the oating roof is
simplied to a plane structure with this theory and the
components such as bulkheads (necessary to divide the pontoon
into several compartments) in the pontoon were neglected. These
methods would be no more applicable when oating roofs with
nonaxial symmetry or with 3-D complex structures such as the
newly developed oating roof with continuous beams mentioned
above, are used. Moreover, the rainwater was usually assumed to
ll the whole deck plate in these methods. The rainwater,
however, would ll only part of the deck plate if the oating roof
has a large enough diameter. On the other hand, although some
FEM solutions were used to conduct the analysis of oating roofs,

Rm
Rw
t
t1, t2
t3, t4
V1, V2
Ve
wA, wB
z
a
d1, d2
Dh0
Dhw
Dhs
DH
e
l0
lw
n
y
r0
r1
t; t
tw
f

729

radius of the outer rim of the pontoon before and after

deformation, respectively
radius of the deck plate or the inner rim of the
pontoon before and after deformation, respectively
mean radius of the pontoon
radius of the area of rainwater lling the deck plate
thickness of the deck plate
thicknesses of the outer rim and inner rim of the
pontoon, respectively
thicknesses of the top and bottom plates of the
pontoon, respectively
two parts of water volume on the deck plate due to
redistribution
water volume on the deck plate
vertical displacements of the bottom of the outer rim
and inner rim, respectively
vertical coordinates of the oating roof
tilt angle of the pontoons bottom plate
radial displacements of the outer rim and inner rim of
the pontoon, respectively
rainfall increment
water head increment
liquid head increment
difference between installing height of the deck plate
and sinking depth of the oating roof
error tolerance
coefcient of determining the water distribution
status on the deck plate
ratio of equivalent water volumes on the deck plate
Poissons ratio of the oating roof
time of deformation progression
water density, 1.0  106 kg/mm3
liquid (oil) density in the tank
ratio of the inner rims and outer rims radii of the
pontoon before and after deformation, respectively
ratio of water distributions and deck plates radii
rotation angle of the pontoon

these solutions were based on a simple axisymmetric method and

only simple plane problem was dealt with. Accordingly, it is
necessary to develop a general numerical method for practical
analysis of oating roofs with 3-D structures in order to ease and
aid implementations of structure design, analysis and optimization of oating roofs.
This paper proposes a general and practical nite element (FE)based numerical method, i.e. the load-modifying method (LMM),
for the 3-D structural analysis of oating roofs under rainwater
load. A relationship between loads and deformations is developed
rstly according to the equilibrium of the oating roof, in which
two cases of rainwater distribution on the deck plate are
considered, one case in which the rainwater lls only part of the
deck plate and the other case in which the rainwater lls the
whole deck plate. Then the FE analysis of the oating roof with
this relationship is conducted based on the LMM. In the analysis
procedure with the LMM, an initial condition (e.g. the condition
with no deformation) is assumed to begin the nonlinear FE
analysis with iterative computations, and then the load magnitudes in the current iteration are modied with computational
results in the previous iteration and are ready for a new iterative
analysis if necessary. Before each iterative analysis, the case of
rainwater distribution on the deck plate is determined by results
of the previous iteration. This analysis process is carried out

ARTICLE IN PRESS
730

X. Sun et al. / International Journal of Pressure Vessels and Piping 85 (2008) 728738

continuously until the computational results converge to the real

solutions. In numerical examples for applications of the proposed
method, the present computational results are compared with
those from other numerical or experimental methods. The validity
and reliability of the proposed method are demonstrated.

2. Equilibrium analysis of the oating roof

A typical single-deck oating roof consists of a circular plate,
i.e. deck, and a compartmented buoyant ring, i.e. pontoon, and its
geometry and characteristic dimensions are sketched in Fig. 1.

Fig. 2. Simplied loads on the oating roof.

2.1. Loads on the oating roof

In the practical operation, the loads on the oating roof
includes: weight (including that of appurtenances), rainwater
loading on the deck, buoyancy and linearly distributed side
pressures along the inner rim and outer rim. In general, the side
pressures along the inner rim and outer rim are slight and
contribute little to the deformation and stress of the oating roof,
and these pressures can be neglected in order to simplify the
analysis. In other words, only three typical loads, i.e. weight,
rainwater loading and buoyancy, are considered in the following
analysis. Additionally, the weight of the deck, buoyancy on the
deck and the rainwater loading can be further simplied with a
net pressure q(r) on the deck, as shown in Fig. 2.
The weight of the oating roof and rainwater loading on
the deck are balanced with the buoyancy produced by liquid in
the tank, and the buoyancy is relevant to the liquid head (i.e. the
liquid surface height, hs, see Fig. 3 or Fig. 4). As it has been pointed
out, the deck plate is usually subjected to a large deection under
rainwater loading, and this deection also has an inuence on the
value of buoyancy. Accordingly, the loads on the oating roof are
coupled with its deformations. The proper relationship between
the loads and the deformations is necessary to carry out the
analysis of the oating roof.
2.2. Relationship between deformations and loads
When there is a rainfall h0 on the top of the tank, the
rainwater will accumulate into the deck plate and redistribute its
volume. If the oating roof is too large, the rainwater will ll
only part of the deck plate since the deck plate is subjected to a
larger deection (as shown in Fig. 3); but if the oating roof is not
very large, the rainwater will ll the whole deck plate (as shown
in Fig. 4).

Fig. 3. Deformation of the oating roof in case the rainwater lls only part of the
deck plate.

Fig. 4. Deformation of the oating roof in case the rainwater lls the whole deck
plate.

In these two cases, the pressures on the bottom plate of the

oating roof are similar, but the load distributions on the deck
plate are a little different.
For the case that the rainwater lls only part of the deck plate
(Fig. 3), the net pressure q(r) on the deck plate can be written with
two different load distributions as
(
qc qw qs ; 0prpRw
qr
(1a)
qc qs ;
Rw orpR2
where Rw is the radius of the area of rainwater lling the deck
plate (Fig. 3); qc, qw and qs are loads resulted from the deck
weight, rainwater on the deck plate and liquid pressure in the
tank, respectively, and
qc 

Fig. 1. Geometry of the oating roof.

Mc g
2

pR 2

(2)

qw r0 gf r  t=2 hw 

(3)

qs r1 gf r t=2 hs 

(4)

where the negative sign  denotes the loading direction

opposite to that of z-coordinate; Mc is the mass of the deck plate
and g 9.8 N/kg is the gravitational acceleration; R 2 is the radius

ARTICLE IN PRESS
X. Sun et al. / International Journal of Pressure Vessels and Piping 85 (2008) 728738

of the deck plate after deformation; r0 1.0  106 kg/mm3 and r1

are the rainwater density and liquid density, respectively; hw and
hs are the water head and liquid head (Fig. 3), respectively;
f(r) is the deection of the deck plate in the middle plane and
always takes a positive value; and t is the thickness of the deck
plate. The water head hw takes a negative value in Eq. (1a). Both
the water head hw and the liquid head hs are relevant to the
deection f(r).
For the case that the rainwater lls the whole deck plate
(Fig. 4), the net pressure q(r) on the deck plate can be written as
qr qc qw qs

2p

Rw

qrr dr 2p
0

2p

R 2

qrr dr  M  Mc g
Rw

R 1
R 2

pb rr dr 0

pb r r1 ghs H0  R 1  r tana f

(6)

where a and f are the tilt angle of the pontoons bottom plate and
rotation angle of the pontoon, respectively.
The liquid head hs can be derived from Eq. (5) as





r
r
t r0 2
hs 0 t2w hw hc 0 lw  1 
tw 1 t 2
2 r1
r1
r1
2

 DH  ha 1  t

(7a)

and

R R 2
0

Hg

2
2
pr1 R 1  R 2

rf rdr

(9)

2
R 2

1 2t
R 1  R 2 tana f
3 3t

c2
R 2
t
t ;
c1
R 1

R2
;
R1

 
dk 
ck 1 
; R k ck Rk
Rk
R Rw

lw R0
R2
0

rf r dr

(8)

tw

(10)
Rw
R 2

k 1; 2

(11)

(12)

(14)

R20 h0

If Rw R 2 , which results in tw 1 and lw 1, Eq. (7a) can be

simplied to the case in which the rainwater lls the whole deck
plate, which can be rewritten as





r
r
t r0
hs 0 hw hc 0  1 
1 t 2
2 r1
r1
r1
 DH  ha 1  t 2

(15a)

hw f Rw

where Rw can be computed with the following formula of

equivalent volume:
Z

Rw

rf r  f Rw  dr pR20 h0

(16)

This integral equation can be solved with the numerically

iterative method.
For the case in which the rainwater lls the whole deck
plate (l0p1), the volume Ve can be divided into two parts,
one for capacity related to the water head hw and the other for
capacity related to the deection f(r), i.e. Ve V1+V2, as shown
in Fig. 5.
According to the equivalent volume in Fig. 5, the following
equation can be obtained:
hw

R20
2
R 2

h0  hc

t
2

(15b)

where t/2 is the additional term resulting from the thickness of

the deck plate and usually can be neglected since it is a very small
value.

(13a)

rf r dr

where H0 is the installing height of the deck plate (vertical

distance between the deck plate and the outer rims bottom
of the pontoon); R 1 is the outer rims radius of the pontoon

(7b)

Eqs. (7a) and (7b) provide the relationships between the water
head hw and the liquid head hs.
As it is pointed out, the redistributions of the rainwater due to
deection of the deck plate are different for cases that rainwater
lls whole or part of the deck plate. For the case that the rainwater
lls only part of the deck plate (l041), the water head hw can be
obtained as (Fig. 3)

2
R 2 hc

where R0 is the radius of the tank. If l0p1, the rainwater lls the
whole deck plate; otherwise, the rainwater lls only part of the
deck plate.
Additionally, the ratio of equivalent volumes, lw, can also be
computed with the following formula:
8
1;
l0 p1
>
>
< 2
2
(13b)
lw R0 h0 Rw f Rw ; l 41
0
>
2
>
:
R 2 hc

2p

DH H0  Hg ;

ha

l0

(5)

where M is the total mass of the oating roof; pb(r) is the pressure
applied on the bottom plate of the oating roof (Fig. 2) and

hc

after deformation; d1 and d2 denote respectively the horizontal

displacement or radial displacement, of the outer rim
and inner rim (Figs. 3 and 4). The equivalent deection hc in
Eq. (9) can be computed with the regular numerical integration
method.
In Eqs. (8) and (10), Hg is the sinking depth due to weight of the
oating roof and ha is the sinking depth of the oating roof due to
slope of the bottom plate of the pontoon. The case that the
rainwater lls whole or part of the deck plate can be determined
with the following ratio:

(1b)

where qc, qw and qs have the same meanings as in Eq. (1a), but it
should be noted that the water head hw takes a positive value in
Eq. (1b).
In Eq. (3), the water head hw is dened as positive value when
the rainwater surface is above the installing position of the deck
plate and as negative value when the rainwater surface is below
this installing position (Figs. 3 and 4). It can be found that Eq. (1b)
can be obtained with Eq. (1a) when Rw R 2 . Accordingly, without
loss of generality, Eq. (1a) is used to carry out the following
derivations unless specied otherwise.
The deection of the deck plate can be treated as same along
the circumference at a given position, r, and the equilibrium
equation of the oating roof in the vertical direction can be
written as

731

Fig. 5. Redistribution of rainwater due to deection.

ARTICLE IN PRESS
732

X. Sun et al. / International Journal of Pressure Vessels and Piping 85 (2008) 728738

"



R20 r0
t
DH   hc t 2 Hg
2r
2
R 1 1

2

ha 1  t  R1  r tana f ; R 2 prpR 1

pb r r1 g h0

hs h0

(19b)



R20 r0
t 2
t  DH  ha 1  t 2
 hc
2r
2
R 1

(20b)

Fig. 6. Rotation of the pontoon.

The rotation angle of the pontoon f in Eq. (10) can be

computed numerically with the following equation (Fig. 6):
f

180 wB  wA
p R1  R2

(17)

where wA and wB are the vertical displacements of point A in the

bottom of the outer rim and point B in the bottom of the inner
rim, respectively, and they are computed numerically. It should be
noted that wA and wB are positive values in the direction of
z-coordinate (Fig. 6). The rotation angle f is selected as a positive
value in the counterclockwise direction and it usually has a
negative value in Eq. (17). The value of f in Eq. (17) includes two
parts that result from the deformation and rigid rotation of the
pontoon, respectively. Actually, Eq. (17) is an approximation on
the rotation of the pontoon since the bottom plate of the pontoon
would curve inward under the liquid pressure. This approximation, however, is reasonable because the angle f usually is a tiny
value [10,11].
By substituting Eqs. (7a) and (15a) into Eqs. (1a) and (6), or by
substituting Eqs. (7b) and (15b) into Eqs. (1b) and (6), the net
pressure q(r) on the deck plate, the pressure pb(r) on the bottom of
pontoon and the liquid head hs for two cases of load distribution
(l041 or l0p1) can be obtained:
(1) For the case that the rainwater lls only part of the deck
plate (l041):
h 
8



>
r0 g rr1 DH  2t  ha 1  t 2  f Rw 2t 1  t2w t 2
>
0
>
>
>


>
>
>
r1
r1
Mc
2
>
>
< hc t lw  r0 pR 22 r0 f r 1  r0 ; 0prpRw
h 
qr



>
>
r0 g rr1 DH  2t  ha 1  t 2 f Rw 2t t2w t 2
>
0
>
>

>

>
>
>
>
hc t 2 lw  rr1 M2 c  f r rr1 ; Rw orpR 2
:
0
0
pR r
2 0

Eqs. (18a)(20b) show that the variables q(r), pb(r), hs and also
hw in Eqs. (15a) and (15b) are all relevant to the deection f(r).
Usually, the net pressure q(r) distributes nonuniformly along the
radial direction on the deck plate and is a function of the
deection f(r) when the liquid density is not equal to the water
density, i.e. r16r0.
Generally speaking, the radial deformations of the pontoon,
i.e. d1 and d2, are much smaller compared with the radii R1
and R2, and the coefcients c1 and c2 can be taken as c1 c2 1 in
order to simplify the computation. That is to say, the variable
values in all above equations can be computed with the initial
dimensions (no deformation) of the oating roof, i.e. R 1 R1 , R 2
R2 and as a result that t t. In addition, the rotation angle of
pontoon, f, is also a tiny value. If f is small enough and
contributes little to the pressure pb(r) in Eqs. (19a) and (19b) or
to parameter ha in Eq. (10), it can also be neglected in order to
simplify the computation.
2.3. Water test condition
Now consider a special but practical condition, the so-called
water test condition, when liquid in the tank is water and r1 r0.
For example, for the case that the rainwater lls the whole deck
plate (l0p1), substituting r1 r0 into Eq. (18b), we see the net
pressure
"
!
#
R2
t
Mc
qr r0 g h0 02 DH   hc  ha 1  t 2 2
(21)
2
pR r
R
2

(18a)



 
r
r
t r0 2
pb r r1 g hc 0 lw  1 DH  0 t2w f Rw 
tw 1 t 2
2 r1
r1
r1

Hg ha 1  t 2  R 1  r tana f ; R 2 prpR 1 (19a)

(20a)

(2) For the case that the rainwater lls the whole deck
plate (l0p1):
("


#
r1
t

h

DH

1  t 2
a
2
2
R 2 r0



)
r
Mc
r
hc 1  1 t 2 2 f r 1  1
r0
r0
pR 2 r0

qr  r0 g

h0

hs h0



t 2
t  DH  ha 1  t 2

h

c
2
2
R

R20

(23)

The water head hw has the same formulation as that in Eq. (15b).
3. Load-modifying method

 



r
r
t r0 2
tw 1 t 2
hs hc 0 lw  1  0 t2w f Rw 
2 r1
r1
r1
 DH  ha 1  t 2

2 0

Eq. (21) demonstrates a uniform load applied on the deck

plate in the water test condition when r1 r0. However, the net
pressure q(r) is still relevant to the deection f(r) since hc in
Eq. (21) must be computed with f(r) and Eq. (9). The pressure pb(r)
on the bottom plate of the pontoon in Eq. (19b) and the liquid
head hs in Eq. (20b) can also be simplied respectively to
"


R2
t
pb r r0 g h0 02  hc  DH t 2 Hg
2
R 1

2
ha 1  t  R 1  r tana f ; R 2 prpR 1
(22)

R20

Eqs. (18a)(20a) or Eqs. (18b)(20b) provide the basic

formulations for analysis of the oating roof with FE simulation.
This analysis is actually a geometrically nonlinear problem
resulting from the deection of the deck plate. Followed with a
standard modeling and discretization in the FE analysis [17,18],
the typical equation of solution can be established with the
principle of virtual displacement in the total Lagrange formulation
y K L y K NL a yDy Q  y F

(18b)

(24)

where KL and KNL denote the linear and nonlinear stiffness

matrices, respectively; y+DyQ and yF are the applied and restoring

ARTICLE IN PRESS
X. Sun et al. / International Journal of Pressure Vessels and Piping 85 (2008) 728738

load vectors, respectively; the superscript y denotes the time of

deformation progression.
Eq. (24) can be solved with an equilibrium iteration method,
such as NewtonRaphson method [18], in which step-by-step
loading (or loading with sub-steps) usually is necessary to obtain
a satisfactorily converged solution.
However, the loads of q(r) and pb(r) are coupled nonlinearly
with the deformations of f(r) and f, and the above regular FE
analysis faces difculty. To solve this problem, an LMM is
presented here to carry out the FE analysis. The concept of this
method is based on an idea of iterative computation with the
loaddeformation relationship and the analysis consists of three
basic sequential parts. Firstly, assume an initial condition, e.g. the
undeformed condition when f0(r) 0 and f0 0, and then the
loads of q(r) and pb(r) are computed with these initial values
according to Eqs. (18a) and (19a), or Eqs. (18b) and (19b),
respectively. Secondly, a nonlinear FE analysis on the oating roof
is carried out using the loads and conditions given above, and then
solutions including f(r) and f are obtained. Lastly, the loads of q(r)
and pb(r) are modied with new values of f(r) and f, and a new
analysis is restarted with the modied q(r) and pb(r) until the
difference between the current modied and previous values is
within the error tolerance specied by the user.
In the above analysis procedure, after the deformations, i.e. f(r)
and f, are determined (obtained initially by assumption and then
by computation), the case of rainwater distribution on the deck
plate must be determined successively with Eq. (14), which is
necessary for computation of loads q(r) and pb(r).
This analysis can be summarized as follows:
(1) construct the FE model according to the geometry of the
oating roof;
(2) compute Hg, DH and t with Eqs. (8) and (11), respectively;
(3) assume an initial conditions, e.g. f0(r) 0 and f0 0;
(4) determine if parameter l041 or l0p1 (the case of rainwater
distribution on the deck plate) with Eq. (14);
(5) compute hc and ha with Eqs. (9) and (10), respectively, then
determine the values of R 1 , R 2 and t , and usually the
simplications that R 1 R1 , R 2 R2 and t t are used;
(6) if l041, i.e. the rainwater lls only part of the deck plate,
compute Rw with Eq. (16), then compute tw with Eq. (11) and
lw with Eq. (13a) or (13b);
(7) compute net pressure on the deck plate q(r) with Eq. (18a) or
(18b), and compute pressure on bottom plate of the pontoon
pb(r) with Eq. (19a) or (19b);
(8) load on the FE model and start a geometrically nonlinear
solution;
(9) modify the loads of q(r) and pb(r) with the solutions of f(r)
and f according to the steps above (4)(7); and
(10) check whether the inequality |qi(r)/qi1(r)1|oe holds
true, where qi(r) and qi1(r) respectively denote load in the
current and previous iterative computation (iteration or
modication i and i1) and e is the user-specied error
tolerance; if it does, qi(r) (and also pib(r)) is the nal solution;
otherwise, return to step (8) and restart a new computation
with the current modied loads until the inequality holds
true.
After q(r) and f(r) are obtained with this method, the water
head hw and the liquid head hs also can be computed with Eq.
(15a) or (15b) and Eq. (20a) or (20b), respectively.
In each modication and iteration, the computation is based on
the undeformation condition with the modied loads. Usually, the
solution converges to the real value with a process of uctuation
between the maximum and minimum. For example, in the water
test condition, the rst modied load q(r) according to the initial

733

values that f0(r) 0 and f0 0 has a maximum value qmax(r)

which also results in a maximum deection fmax(r), and the
next modied load according to this maximum deection will
have a minimum value qmin(r). The convergence process is shown
in Fig. 7.

4. Numerical examples
Two typical oating roofs are presented in this section to
demonstrate the applications of the proposed method. The rst
example is about a traditional single-deck oating roof with larger
diameter [12,13], which is used to demonstrate the reliability of
the present method for the analysis of general single-deck oating
roofs. The other example is an experimental model of oating roof
with continuous beams in both the top and the bottom plates of
the pontoon [6]. This example is used to demonstrate the validity
of the present method for analysis of oating roofs with 3-D
complex structures or components. The operation condition and
water test condition are considered in these two examples,
respectively. Additionally, the case of rainwater lling only part
of the deck plate is demonstrated in the rst example and the case
of rainwater lling the whole deck plate is demonstrated in the
other example.
4.1. A traditional single-deck oating roof
A traditional oating roof with a larger diameter in operation
condition under a total 250 mm rainfall is examined in this
example to demonstrate the application of the present method.

Fig. 7. Process of the solution convergence in water test condition.

Table 1
Parameters of the single-deck oating roof
Parameter

Value

R0
R1
R2
H0
H1
H2
t
t1
t2
t3, t4
a
r1
Mc
M
E
n

40,000 mm
39,600 mm
34,600 mm
300 mm
800 mm
450 mm
4.5 mm
8.0 mm
12.0 mm
4.5 mm
01
0.7  106 kg/mm3
152,000 kg
307,000 kg
210,000 MPa
0.3

ARTICLE IN PRESS
734

X. Sun et al. / International Journal of Pressure Vessels and Piping 85 (2008) 728738

The design parameters of this oating roof are listed in Table 1

[12,13], where ti (i 1,2,3,4) denotes the thickness of outer rim,
inner rim, top and bottom plates of the pontoon, respectively.
Since the oating roof is a rotationally periodic structure which
usually is simplied to axisymmetic problem, one thirty-sixth of
the whole structure is adopted as a compartment in the 3-D FE
analysis, and the 3-D shell elements are adopted to simulate all
the plate components in the FE model. The displacements along
z-direction in the bottom of the outer rim are constrained in order
to avoid the rigid body displacement in this direction, and the
symmetric boundary conditions are applied to the circumferential
sections of the concerned one thirty-sixth (1/36) part structure.
The initial values for computation are f0(r) 0, f0 0 and the

error tolerance for iterative load-modifying computation is 1%.

The change of net pressure on the deck plate, q(r), with iteration i
is plotted in Fig. 8. It can be found that the net pressure q(r)
changes little after 16 iterations, and therefore the equilibrium
solution of the oating roof is obtained. Fig. 9 plots the loads of
the oating roof, q(r) and pb(r), in the equilibrium state. The radius
of the area of rainwater lling the deck plate, Rw, is about
31,000 mm (oR2 34,600 mm), which means that the rainwater
lls only part of the deck plate.
The numerical results of deection and radial stress of the deck
plate by the present method are compared with the results from
other methods in Figs. 10 and 11, respectively. The deections and
radial stresses near and away from the bulkheads are both given

net pressure on deck plate, q(r) (x10-3MPa)

1.5
i=0
i=1
i=4
i=7
i=8
i=10
i=13
i=15
i=16

1.0
0.5
0.0
-0.5
-1.0
-1.5
-2.0
-2.5
0

5000

10000

15000
20000
radius r (mm)

25000

30000

Fig. 8. Procedures of the modied computation for net pressures on the deck plate.

Fig. 9. Loads of the oating roof.

35000

ARTICLE IN PRESS
X. Sun et al. / International Journal of Pressure Vessels and Piping 85 (2008) 728738

735

1600
present method (near bulkhead)
present method (away from bulkhead)

deflection of the deck plate, f (mm)

1400

Milne method (Umeki and lshiwata, 1985)

KOSTRAN (Umeki and lshiwata, 1985)

1200
1000
800
600
400
200
0
0

5000

10000

15000
20000
radiums, r (mm)

25000

30000

35000

Fig. 10. Deection of the deck plate under rainfall h0 250 mm.

pontoon results in an additional rotation of the deck plate, the

deections near the center area of the deck plate from the present
numerical method are a little larger than those from the methods
given by Umeki and Ishiwata [12] in which the pontoon is
assumed as a rigidity.

radial stress of the deck plate, r (MPa)

180
160

present method (near bulkhead)

present method (away from bulkhead)
Milne method (Umeki and lshiwata, 1985)
KOSTRAN (Umeki and lshiwata, 1985)

140
120

4.2. A new type of oating roof with continuous beams

100
80
60
40
20
0
0

5000

10000

15000

20000

25000

30000

35000

radius, r (mm)
Fig. 11. Radial stress of the deck plate under rainfall h0 250 mm.

to demonstrate the present numerical results. The deection of

the deck plate near the bulkhead is nearly the same as that away
from the bulkhead. However, there are different stress distributions near and away from the bulkheads, and a signicant radial
stress appears near the bulkhead. The results by the present
method agree well with the results given by Umeki and Ishiwata
[12], except for the stress in the edge area of the deck plate where
the deck plate is jointed with the pontoon. It should be noted that
the bending effect is neglected in the results by Umeki and
Ishiwata [12] since only the membrane stress is concerned and
the bulkhead is also not considered. It is known that the bending
effect appears locally near the edge area of the deck plate, and
there is usually a signicant bending stress near the connections
of deck plate and pontoon, especially the area near the bulkhead
with considerable bending tendency. The effect of the bulkhead
on the stress in the deck plate will be further demonstrated in the
next example. Moreover, because the elastic deformation of the

In this example, an experimental model of oating roof with

continuous beams is numerically simulated under rainwater
loading in the water test condition, and the numerical results
are compared with the experimental ones. This model is one fth
scale of the designed oating roof with continuous beams [6]. This
oating roof has a pontoon with radial and annular continuous
beams in both of its top and bottom plates, and the intersections
of these beams are jointed by vertical ribs between the top and
bottom plates, which results in a more complex structure (Fig. 12).
The characteristic geometric properties and material properties of
the oating roof model are listed in Table 2, where Nr, Na
and Nv denote the number of radial, annular continuous beams
(in both top and bottom plates of the pontoon) and vertical
ribs (beam structure), respectively, all beams with L-section
10 mm  10 mm  1 mm.
This model is also tested with the experimental method. Water
is lled into a cylindrical container to lift up the model roof, which
is used to simulate the oating roof in operation (Fig. 12a).
A controlled-volume water pump is used to ll water onto the
deck plate to simulate rainfall. The deections and strains are
measured by the staff gauges and the electrical resistance stain
gauges, respectively, and then the stresses are computed with
the strains obtained from the experiment. A load of rainfall
h0 50 mm is applied to the oating roof (i.e. R0 R1) to simulate
a typical rainfall required in the design codes. Through the above
process, we have obtained the experimental results in the water
test condition [6].
In the numerical analysis, one-eighteenth of the oating roof,
i.e. one compartment, is used for the FE model since the oating
roof is a rotationally periodic structure consisting of 18 identical
compartments, and the rotation angle of pontoon, f, is neglected

ARTICLE IN PRESS
736

X. Sun et al. / International Journal of Pressure Vessels and Piping 85 (2008) 728738

Fig. 13. FE mesh of the oating roof with continuous beams: (a) one-eighteenth
structure (one compartment); and (b) pontoon with continuous beams (view
without the top plate).
Fig. 12. Model of the oating roof with continuous beams: (a) experimental
model; and (b) pontoon with continuous beams (topoblique view without the
top plate).

Table 2
Parameters of the model of oating roof with continuous beams
Parameter

Value

R1
R2
H0
H1
H2
t
t1, t2, t3, t4
a
Nr
Na
Nv
r1
Mc
M
E
n

4010 mm
3414 mm
57.5 mm
170 mm
140 mm
1.0 mm
1.0 mm
01
4  2  18
22
8  18
1.0  106 kg/mm3
286 kg
734 kg
200,000 MPa
0.334

because it is a very tiny value. In the FE model, the continuous

beams, vertical ribs and all the plates are simulated with the 3-D
beam elements and 3-D shell elements, respectively, and the
meshes of the one-eighteenth structure and corresponding
pontoon are shown in Fig. 13(a and b). It should be noted that
the mesh of the pontoons top plate is not illustrated in Fig. 13b in
order to demonstrate the inside structures of the pontoon. The
displacement boundary conditions are similar to those in the rst
example.
Table 3 shows the maximum deection fmax, the equivalent
deection hc and the net pressure q(r) (constant in water test

Table 3
Procedures of the modied computation under rainfall h0 50 mm
Modication i

fmax (mm)

hc (mm)

q(r) (  103 MPa)

0
1
2
3
4
5
6

0
55.032
48.881
49.659
49.561
49.574
49.572

0
30.210
26.770
27.206
27.151
27.158
27.157

0.27646
0.19499
0.20427
0.20309
0.20324
0.20322
0.20323

condition) in each modied procedure under rainfall h0 50 mm.

It can be found that the values of q(r) or hc in the current and
previous iterations are nearly unchanged after about ve iterative
modications, and these values can be treated as the nal results.
Table 4 lists the water head increment Dhw, the liquid head
increment Dhs and the ratio C Dhw/Dhs (R1/R2)2 in different
rainfall increments. The results from the present method, the
approximate method [4] and the experimental method are
compared, and good agreements are observed. The numerical
results are more consistent with the theoretical values since the
numerical model is more similar to the theoretical model.
Fig. 14 plots the deections of the deck plate under different
rainfalls, i.e. h0 50, 65 and 80 mm, and the present numerical
results are compared with the experimental results. It can be
found that the results from the present numerical method
agree well with those from experiment. Fig. 15 shows comparison of the deections of the deck plate for the roofs with
and without bulkheads under h0 50 mm. The deck plate is
usually subjected to larger deection for the roof without
bulkheads since the bulkheads somewhat strengthen the rigidity
of the pontoon.

ARTICLE IN PRESS
X. Sun et al. / International Journal of Pressure Vessels and Piping 85 (2008) 728738

737

Table 4
Comparison of coefcient C with different rainfall increments
Parameter

Methods

Rainfall increments Dh0

50 mm-65 mm

65 mm-80 mm

50 mm-80 mm

Dhw (mm)

Present method
Approximate method
Experimental method

18.8
18.53
18.35

18.6
18.79
17.65

37.4
37.32
36.00

Dhs (mm)

Present method
Approximate method
Experimental method

13.7
13.43
11.50

13.4
13.63
13.75

27.1
27.06
25.25

Present method
Approximate method
Experimental method
Theoretical method

1.372
1.380
1.596
1.38

1.388
1.378
1.284
1.38

100
radial stress of the deck plate, r (MPa)

deflection of the decak plate, f (mm)

60

1.380
1.379
1.426
1.38

50

40

30

20
numerical results

10

experimental results

h0 = 50 mm
h0 = 65 mm
h0 = 80 mm

h0 = 50 mm
h0 = 65 mm
h0 = 80 mm

numerical results (near bulkhead)

numerical results (away from bulkhead)
numerical results (without bulkhead)
experimental results (near bulkhead)

80

60

40

20

-20
0

500

1000

1500

2000

2500

3000

3500

500

1000

50

50

40

stress of the radial beams, (MPa)

deflection of the decak plate, f (mm)

60

40

30

20

the floating roof with bulkhead

the floating roof without bulkhead

0
0

500

1000

1500

2000

2000

2500

3000

3500

Fig. 16. Radial stress of the deck plate under rainfall h0 50 mm.

Fig. 14. Deections of the deck plate under different rainfalls.

10

1500

radius, r (mm)

radius, r (mm)

2500

3000

3500

radius, r (mm)

numerical results (top beam)

numerical results (bottom beam)
experimental results (top beam)
experimental results (bottom beam)

30

20

10

-10
3400

3500

3600

3700

3800

3900

4000

4100

radius, r (mm)

Fig. 15. Deections of the deck plate for the roofs with and without bulkheads
under rainfall h0 50 mm.

Fig. 17. Stress distributions of the radial beams under rainfall h0 50 mm.

The radial stress in the deck plate is shown in Fig. 15 with the
present numerical results and experimental results. Both the
numerical and experimental results indicate that there are
signicant stresses near the bulkhead at the edge of the deck
plate, as shown in Fig. 16. These signicant stresses, however, are

eliminated near the edge which is away from the bulkhead. The
comparison of the radial stresses of the deck plate for the roof
with and without bulkheads under h0 50 mm is also shown in
Fig. 16. These results demonstrate that the bulkhead has
signicant effects on the radial stress near the edge of the deck

ARTICLE IN PRESS
738

X. Sun et al. / International Journal of Pressure Vessels and Piping 85 (2008) 728738

stress near the edge of the deck plate. Bending stress usually
appears near the connections of deck plate and pontoon, whereas
the membrane stress dominates the stress states at locations
away from the edge of the deck plate.
The proposed method provides an effective and practical 3-D
numerical solution to the design and analysis of oating roofs. The
present solution can be further applied to structural analysis of
oating roofs with more complicated components and load
conditions, and a detailed analysis in a local structure or
component of oating roofs can even be potentially conducted
with the present method.

80
numerical results
vertical stress
annular stress

stress of the inner-rim, (MPa)

60

experimental results
vertical stress
annular stress

40

20

-20

Acknowledgments

-40
0

20

40

60

80

100

120

140

height, H (mm)
Fig. 18. Stress of the inner rim under rainfall h0 50 mm.

plate, which was not considered in the other methods forementioned since, wherein the bulkhead was usually neglected.
Furthermore, at locations away from the edge of the deck plate,
the membrane stress dominates the stress states, and the stress
distributes relatively uniformly at these locations.
Figs. 17 and 18 demonstrate the stress distributions in typical
components of the pontoon under rainfall h0 50 mm. The stress
of the radial beams in the top and bottom plates of the pontoon is
illustrated in Fig. 17. The results indicate that the radial beams in
the pontoon are tensioned and there are signicant stresses near
the connections with the inner rim. Fig. 18 gives the vertical and
annular stress distributions of the inner rim of the pontoon. The
present numerical results are compared with the experimental
results, and good agreements are observed, which demonstrates
the present numerical method can simulate the structural
problems of the oating roof with reasonable and reliable results.

5. Conclusions
The LMM is proposed in this paper for analysis of oating roofs
under rainwater loading. According to the equilibrium analysis of
oating roofs in practical operation, the relationship between
loads and deections is derived, and two cases of the rainwater
distribution on the deck plate are considered. The analyses of
stress and deformation of oating roofs are developed with a
geometrically nonlinear FE simulation based on the LMM. The
loading magnitudes are modied with a series of iterative
computations until the computational results meet the given
accuracy requirements within the user-specied error tolerance in
the analysis process. Numerical examples demonstrate that the
proposed method is valid and reliable for analysis of oating roofs
based on 3-D model. The numerical results indicate that the deck
plate is usually subjected to larger deection under rainwater
loading and the bulkhead has signicant effects on the radial

The research work in this paper is funded by the National

Foundation for Excellent Doctoral Thesis of China (No. 200025)
and the National Tenth Five-Year Key Technology Research
Special Funds of China (No. 2001BA803B). The authors are grateful
to Mr. Wang Fuan in Design Institute of China Petroleum Pipeline
Engineering Co. for his generous help with the oating roof model
data referred in the research.

References
[1] de Wit J. Floating roof tanks. Engineering 1970;210:558.
[2] Young WB. Floating roofstheir design and application. In: Proceedings of
the ASME petroleum mechanical engineering conference, Los Angeles, CA,
73-Pet-44, 1973.
[3] Epstein HI. Stresses and displacements for oating pan roofs. Comput Struct
1982;15(4):4338.
[4] Pan JH. Design of cylindrical metal tanks. Beijing: Petroleum Industry Press;
1984 (in Chinese).
[5] Gallagher TA. Floating-roof technology advances with lessons learned from an
80-year history. Hydrocarbon Process 2003;82(9):637.
[6] Wang FA, Sun XS. Research on the technology of the oating roof with
continuous beams and dome frames. Technical Report, Design Institute of
China Petroleum Pipeline Engineering Co., Langfang, Hebei Provice, China,
2004 (in Chinese).
[7] ANSI/API STD 650-1988. Welded steel tanks for oil storage. Washington, DC,
USA: American Petroleum Institute; 1988.
[8] SH3046-92. Chinese standard specication for design of vertical steel welded
storage tanks for the petroleum industry. Beijing, China: China Petroleum and
Chemistry Co.; 1992 (in Chinese).
[9] Mitchell GC. Analysis and stability of oating roofs. J Eng Mech Div 1973;
99(EM5):103752.
[10] Epstein HI, Buzek JR. Stresses in oating roofs. J Struct Div 1978;104(ST5):
73548.
[11] Epstein HI, Buzek JR. Stresses in ruptured oating roofs. J Pressure Vessel
Technol 1978;100(2):2917.
[12] Umeki T, Ishiwata M. Deection and stress analyses of oating roofs under
rainwater loading. Piping Eng 1985;27(5):649 (in Japanese).
[13] Yuan S, Wang JL, Zhong HZ. Analysis of oating roofs by ODE-solver method. J
Eng Mech Div 1998;124(10):112934.
[14] Institute of Mechanics, CAS. Calculation formulas of strength and stability of
oil tanks with oating roofs. Mech Pract 1982;4(2):3640 (in Chinese).
[15] Uchiyama S, Oka T, Oikawa T. The stress analysis of oating roof in the
petroleum storage tank under rain-water load by FEM. J High Pressure Inst
Japan 1981;19(2):816 (in Japanese).
[16] Yoshida S. Geometrically nonlinear stress analysis of oating roofs by nite
element method. Piping Eng 1983;25(8):517 (in Japanese).
[17] Bathe KJ. Finite element procedures. Englewood Cliffs, NJ: Prentice-Hall;
1996.
[18] Zienkiewicz OC, Taylor RL. The nite element method. 5th ed. Oxford:
Butterworth Heinemann; 2000.