Você está na página 1de 21

Copyright by SIAM. Unauthorized reproduction of this article is prohibited.

SIAM J. APPL. MATH. c 2011 Society for Industrial and Applied Mathematics
Vol. 71, No. 1, pp. 2040
MODELING THE EQUILIBRIUM CONFIGURATION OF A
PIECEWISE-ORTHOTROPIC PNEUMATIC ENVELOPE WITH
APPLICATIONS TO PUMPKIN-SHAPED BALLOONS

MICHAEL C. BARG

, JIEUN LEE

, AND FRANK BAGINSKI

Abstract. Large superlight structural systems that, for functional reasons, require large surfaces
are composed at least in part of structural membranes. For eciency of design, components that
experience low stress can be made of lighter material, while those expected to experience high stress
can be reinforced with tendons or made from a stronger, albeit heavier, material. The design engineer
seeks an ecient design without compromising structural performance and safety. The undercon-
strained nature of such structural membranes poses analytical diculties and leads to challenging
mathematical problems in modeling, analysis, and numerical simulation. Motivated by the problem
of modeling the shape of a high-altitude large scientic balloon, we present a model for a tendon-
reinforced piecewise-orthotropic thin pressurized membrane. Using direct methods in the calculus
of variations, a variational principle for a quasi-convex Caratheodory Lagrangian is developed, and
rigorous existence theorems are established. Our model is implemented into a numerical code which
we use to explore equilibrium congurations of a strained pumpkin-shaped balloon at low pressure
where the symmetric shape becomes unstable.
Key words. membranes, calculus of variations
AMS subject classications. 74K15, 58E30
DOI. 10.1137/100795875
1. Introduction. Due to their size, it may be dicult to build appropriately
scaled physical models of large light-weight structures that will respond in a way
that is representative of the actual structure. For example, stratospheric large sci-
entic balloons can be over 130 meters in diameter and are constructed from thin
polyethylene (PE) lm 2040 microns thick. It would be impractical to build a scale
model 2 meters in diameter, since a suciently thin lm does not exist. Similarly,
it would be impossible to simulate a zero-gravity environment in a ground test of a
large-aperture antenna (say, 10 meters in diameter) that is to operate in outer space.
In situations such as these, where physical tests cannot provide a complete picture of
the full-scale structure, an accurate mathematical model is critical.
Motivated by the problem of modeling the shape of a high-altitude large sci-
entic balloon, we present a mathematical model for a tendon-reinforced piecewise-
orthotropic pressurized membrane. This model is readily adaptable to other struc-
tures. For example, this approach was adapted to carry out a stress analysis of a thin
nylon spherical pressure vessel that contained the scintillator for the solar neutrino ex-
periment Borexino (see [8]). Although the scintillator was shielded by a thick barrier
of buer uid of similar density and the temperature was regulated, experimenters
were concerned that, in the event of a power failure, a substantial temperature dif-

Received by the editors May 19, 2010; accepted for publication (in revised form) October 13,
2010; published electronically January 4, 2011. The research presented here was supported in part
by NASA awards NAG5-5353, NNX07AQ49G, and NNX07AR67G.
http://www.siam.org/journals/siap/71-1/79587.html

Department of Mathematics, Niagara University, Niagara, NY 14109 (mbarg@niagara.edu). This


authors work was supported in part by a grant from the Niagara University Research Council in
Summer 2009.

Department of Mathematics, The George Washington University, Washington, DC 20052 (zieun@


gwmail.gwu.edu, baginski@gwu.edu).
20





































































Copyright by SIAM. Unauthorized reproduction of this article is prohibited.
ORTHOTROPIC PNEUMATIC ENVELOPES 21
ference between the inside and outside uids would lead to a buoyant scintillator,
possibly damaging the nylon shell or the supporting structure. Another application
arises in spacecraft missions that require a large radio frequency (RF) antenna and
high-power transmitters to accommodate elevated data rates. An inatable antenna
system has unique features that make it a viable means to support such requirements
while meeting the mass and packing constraints of a deep space mission (see [18]).
Surface features directly aect RF performance, and to thoroughly probe the design
space, it is important to have an accurate mathematical model (see [13]).
To assess design validity and performance of inatable membrane structures, a
mathematical model that captures the essential features of the structure is needed.
Moreover, the model should allow for rigorous analysis with existence results that
ensure the validity of numerical solutions. In [4], we established existence results
for pressurized isotropic membranes inated with a lifting gas and reinforced with
inextensible tendons. The results in [4] apply to closed and open balloon systems. In
the present paper, we extend the results in [4, 7] to piecewise-orthotropic membranes
with elastic tendons. The tendons can be modeled as narrow ribbons located along
the seam between adjacent gores. We derive an expression for the strain energy in the
ribbon that is equivalent to the strain energy in a linearly elastic tendon, enabling us
to derive estimates that are needed to establish our existence results.
In section 2, we provide background for the balloon problem. In section 3, we
formulate the problem of determining the shape of a strained elastic balloon with re-
inforcing tendons. In section 4, we establish theoretical existence results. In section 5,
we present numerical solutions for pumpkin-shaped balloons. For a stable design and
a suciently high pressure, the cyclically symmetric fully developed pumpkin shape is
achieved. When the pressure is reduced and positive lift is maintained, the symmetric
shape becomes unstable, and the balloon seeks a nonsymmetric equilibrium congu-
ration. We simulate these scenarios with a computer model and nd good qualitative
agreement with test observations.
2. Background. The balloon is modeled as a nonlinearly elastic membrane
shell (see [12, section 9.4]). Our strain energy density, W, is equivalent to the two-
dimensional strain energy density of Koiters nonlinearly elastic membrane shell (see
[12, p. 548]). In our formulation, we can ignore the contribution of the exural
energy since the balloon has negligible bending stiness. In [4], the balloon was
assumed to be constructed of a single layer of an isotropic material. In the model
presented here, the membrane is constructed of piecewise-orthotropic materials. The
total energy of the balloon system is
(2.1) I(x) =
_

(f
P
(x, x) +W

(u, x) +f
w
(u, x) +) dA,
where x is an admissible deformation and u = (u, v) , a bounded, open, connected
set in R
2
. In (2.1), dA denotes area-measure on , f
P
is the hydrostatic pressure
potential density, f
w
is the lm and tendon gravitational potential energy density,
W

is the relaxed lm strain energy density, and is a constant to be specied later.


is referred to as the at reference conguration. A tendon is treated as a thin ribbon
with one of its edges attached to the right edge of a gore (see Figure 2.1(a)(b)). In
reality, a tendon is contained in a attened sleeve of PE. The right edge of the ith
gore, the right edge of the attened sleeve, and the left edge of the (i +1)th gore are
sealed together. We assume that the tendon does not slide within the sleeve. We can
model a tendon as a narrow ribbon, and it provides a realistic representation of how a





































































Copyright by SIAM. Unauthorized reproduction of this article is prohibited.
22 MICHAEL C. BARG, JIEUN LEE, AND FRANK BAGINSKI
(a) (b) (c)
Fig. 2.1. (a) G
i
is the reference conguration of the ith gore; subregions G
i

l
(lm) and
G
i

l

(tendon ribbon). (b) Film facet T G


i

l
and tendon facet T G
i

l

. In a real
balloon, T G
i

is essentially a right triangle with base L and height h. (c) Material directions
{ex, ey} and principal directions {n
1
, n
2
}.
real tendon behaves. When is triangulated, each facet T corresponds to either
a membrane component or a tendon ribbon component (see Figure 2.1(b)).
Mechanical properties of importance are lm thickness, Youngs modulus, and
Poissons ratio. In [4], the balloon lm was assumed to have a constant thickness and
constant mechanical properties so that W

(u, x, x) W

(x). It follows that the


integrand in (2.1) is continuous, enabling us to apply [15, Theorem 2.9, p. 180] to
establish existence results. However, this theorem does not apply in the present case,
where the mechanical properties of the lm may vary from region to region and the
integrand in (2.1) (denoted f

Tot
) is no longer continuous. In addition, we assume that
the lm is orthotropic and not isotropic. We will show that f

Tot
is a Caratheodory
function and obtain weak lower semicontinuity as a consequence of [1, Theorem II.4,
p. 137]. In the piecewise-orthotropic case, W

depends explicitly on u . We will


use a modied version of [15, Theorem 2.9, p. 180], where weak lower semicontinuity
is guaranteed by [1, Theorem II.4, p. 137]. The next step is to show that f

Tot
satises
[A[
p
f

Tot
(u, x, A) q(u) + C([x[
p
+ [A[
p
), where p > 1, > 0, and C 0
are constants and q is a nonnegative locally summable function. In our analysis, we
will show that W

has upper and lower bounds proportional to [x[


4
, leading to a
problem formulation in W
1,4
(, R
3
). Next, we record a few basic denitions.
Definition 2.1. A continuous function f : R
23
R is quasi-convex if for every
A R
23
, for every open subset R
2
, and every function W
1,
0
(, R
3
),
(2.2) f(A)meas()
_

f(A+ (u)) dA.


Definition 2.2. If f : R
3
R
23
R is continuous in the third argument for
almost every u and every x R
3
, then f is quasi-convex in the third argument
if (2.2) holds for almost every u and every x R
3
.
For our model, we consider a complete shape. Constraints are built directly
into our solution space of admissible deformations. Here, there is only one boundary





































































Copyright by SIAM. Unauthorized reproduction of this article is prohibited.
ORTHOTROPIC PNEUMATIC ENVELOPES 23
condition: the bottom of the balloon and tendons are attached to a rigid nadir end-
tting whose center is located at the origin of a Cartesian coordinate system, i.e.,
(2.3) x(u, v) = (x
0
(u, v), y
0
(u, v), 0) for (u, v) ,
where parametrizes the boundary of the end-tting. Equation (2.3) can be written
in the form g = 0 R
3
with g(u) = x(u) (x
0
(u), y
0
(u), 0); it is an example of a
local constraint. A volume constraint is an example of a global constraint. In [4], we
derived estimates to show that a volume constraint can be included in the solution
space. We set g
4
(x) = V (x)
0
, where V (x) is the volume of the lifting gas and

0
is the target volume. In general, a global constraint has the form g
i
(x) = 0, where
g
i
C(X, R) for some i and X is the solution space (see section 3). X is a closed
subset of the Sobolev space W
1,4
(, R
3
) and incorporates the constraints directly. We
use the convention g(x) = 0 to indicate all local and global constraints.
3. Mathematical formulation. The balloon design problem is to determine
the at cutting pattern for a gore lobe so that when the complete balloon is assembled
and pressurized, it will attain the desired shape (see Figure 5.1(a)). For our purposes,
the cutting pattern is known. See [3, 17] for more on the design process. In this
section, we present results that are needed to establish Theorem 4.3.
3.1. Preliminaries. Normally, a large scientic balloon is made by sealing a
number of long tapered panels of PE edge-to-edge to form a complete shape. Such
panels are called gores and are constructed as follows. Long circular cylinders of PE
lm are formed using a slot-die extrusion process (imagine blowing a long Delaunay
surface of PE that is attened, then rolled up onto a spool). The axis of the cylinder
is called the machine direction. The cylinder is cut along its axis, then laid out into a
plane. The nal gore pattern G
F
is then cut from these panels. See Figure 2.1. The
number of gores is n
g
. Tendons are located along the seams between adjacent gores.
Suppose =

ng
i=1
G
i
, where G
i
is the at reference conguration of the ith gore.
The reference conguration of a gore is positioned so that e
y
and e
x
are parallel to
the machine and transverse directions, respectively.
In the present work, we will allowthe thickness of the lm, as well as its mechanical
properties, to vary over subregions of . Let u = (u, v) . The properties of
interest are lm thickness t(u), lm weight density w
film
(u), Youngs moduli E
0
i
(u),
and Poissons ratios
0
i
(u) for i = 1, 2, which are assumed to be piecewise-constant
functions. The superscript 0 indicates that the material constants are specied with
respect to the unstressed principal material frame (see section 3.4). We assume that
there is a nite collection of distinct values t
l
, E
0,l
1
, E
0,l
2
,
0,l
1
,
0,l
2
, w
l
for l = 1, . . . , m
such that
(3.1)
l
= u [ t(u) = t
l
, w
film
(u) = w
l
, E
k
(u) = E
0,l
k
,
k
(u) =
0,l
k
, k = 1, 2
forms a partition of , i.e., =

m
l=1

l
. It is convenient to set
l
1
=
l
21
and
l
2
=
l
12
.
Using
l
from (3.1), a partition of the ith gore consists of the sets
l
i
= G
i

l
for
l = 1, . . . , m, i = 1, . . . , n
g
. Each
l
i
is triangulated. T denotes a triangle in
l
i
.
Strictly speaking, along the boundary where two
l
i
s are joined together, there
is some overlap. However, for most interfaces this overlap is insignicant, as it does
not inuence the overall response of the structure. The exception to this occurs along
a seam where a load tendon is attached. The contribution of the load tendon will be
handled separately. Without loss of generality, we can assume that each
l
i
is open





































































Copyright by SIAM. Unauthorized reproduction of this article is prohibited.
24 MICHAEL C. BARG, JIEUN LEE, AND FRANK BAGINSKI
and
l

k
= when l ,= k. By construction, the material properties of the lm are
constant in
l
i
.
In our present model, the edges of the gores are identied as they were in [4]. A
narrow ribbon of constant width h is attached along the seam between every pair of
adjacent gores. The complete membrane is =

ng
i=1
G
i
, where each G
i
is a copy
of the known cutting pattern: G
F
= (u, v) [ H(v) < u < H(v), 0 < v < L
c
.
See Table 5.1(b). We assume that H C
0,1
((0, L
c
), R) so that G
F
is a C
0,1
domain.
Since a balloon normally has a bottom end-tting, we assume that G
F
has an edge
at the bottom. Let
F
= (u, v) [ H(0) < u < H(0), v = 0 be the bottom edge of
G
F
. For this exposition, we assume that a typical gore comes to a point at the apex,
and so u = H(L
c
) = 0. In Figure 2.1(a), G
i
is displayed. Suppose G
1
= G
F
, > 0,
and G
i
= (u + (i 1), v) [ (u, v) G
1
, i = 2, . . . , n
g
. The balloon is modeled by
considering the union of the G
i
s with the following identications. The right seam of
G
i
is identied with the left seam of G
i+1
for i = 1, 2, . . . , n
g
1, and the right seam
of G
ng
is identied with the left seam of G
1
. Thus, =

ng
i=1

m
l=1

l
i
.
Due to the identications of the G
i
s along their respective seams, we can treat
as a two-dimensional manifold. However, to avoid the complicated notation of the
manifold setting, we will treat the at reference conguration as an open subset in R
2
.
In an eort to clarify our presentation, we also use the following conventions. We write
x C
1
(

, R
3
) to mean that x is dierentiable in the interior of the open sets G
i
for
1 i n
g
. With an appropriate metric dened on , we can consider the closure
of , denoted by

. The boundary of is denoted by and represents the edge of
the balloon that is attached to the bottom end-tting. With these conventions, we
consider mappings x C
1
(

, R
3
) C(

, R
3
).
A typical mapping is x : o R
3
, where x() = o and x is an admissible
deformation (see Denition 3.1). Let o
0
= x
0
() be an initial, possibly strained, con-
guration that is not necessarily in equilibrium. x
0
(u, v) = (x
0
(u, v), y
0
(u, v), z
0
(u, v))
is a deformation that satises (2.3). As in [4] and [7], W
1,4
(, R
3
) = x [ |x|
1,4
<
is the proper Banach space setting for the balloon problem. The norm | |
1,4
is equiv-
alent to the standard Sobolev norm (see, e.g., [4]). In the following, we will study
the balloon problem in a subset of W
1,4
(, R
3
). We can now dene an admissible
deformation mapping.
Definition 3.1. Let x
0
(u, v) = (x
0
(u, v), y
0
(u, v), z
0
(u, v)) W
1,4
(, R
3
). An
admissible deformation mapping x(u, v) = (x(u, v), y(u, v), z(u, v)) is such that
(a) x C
1
(

, R
3
) C(

, R
3
),
(b) x(u, v) = x
0
(u, v) for (u, v) , and
(c) x(0, L
c
) = y(0, L
c
) = 0.
The set of admissible deformations is denoted by T. Implicit in Denition 3.1(c)
is the fact that the top of the balloon (z
top
= z(0, L
c
)) is free to slide up and down
the z-axis. We dene our solution space,
(3.2) X =
_
x = x
0
+ x [ x W
1,4
0
(, R
3
)
_
,
to be the completion of T with respect to the norm | |
1,4
. This is consistent with
the denition of Sobolev spaces in the manifold setting [2, p. 32]. For the balloon
problem, a solution is in the form x = x
0
+ x X, where x W
1,4
0
(, R
3
). The space
W
1,4
0
(, R
3
) is the completion of C
1
(

, R
3
) C
0
(

, R
3
) with respect to | |
1,4
.





































































Copyright by SIAM. Unauthorized reproduction of this article is prohibited.
ORTHOTROPIC PNEUMATIC ENVELOPES 25
3.2. Hydrostatic pressure potential. We consider a position dependent hy-
drostatic pressure in the form P(z) = bz + p
0
. P(z) > 0 means that the inside
of the gas bubble is pushing outward at a height z units above the xed base of the
balloon, z = 0. At its base, an open balloon system has pressure p
0
. The volume will
adjust to meet this condition. In a closed balloon system, p
0
is unknown a priori, and
it must be determined from the volume constraint. The zero-pressure natural-shaped
(ZPNS) balloon system is open, and the pumpkin-shaped balloon system is closed. If
the density of the atmosphere is denoted by
a
, the density of the lifting gas is
g
,
and the gravitational acceleration is g, then the specic buoyancy of the lifting gas,
b = g(
a

g
), is positive. From [4], with P(x) = (bx k+p
0
), the potential for the
hydrostatic pressure is
E
P
(x, x) =
_
D
P(x) dV =
_

f
P
(u, x, x) dA, where
f
P
(u, x, x) =
_
1
2
b(x k)
2
+p
0
(x k)
_
k adj
2
x, (3.3)
D R
3
is the gas bubble, and dV is volume-measure in R
3
. The adjugate adj
2
is
dened in [15, section 4.1]. Suppose [x[ R. Since [k adj
2
x[ [x[
2
, we have
(3.4)
_
1
2
bR
2
+[p
0
[R
_
[x[
2
f
P
(u, x, x)
_
1
2
bR
2
+[p
0
[R
_
[x[
2
.
f
P
is shown to be quasi-convex in [4]. The condition [x[ < R is not a serious restriction.
R is chosen suciently large so that the constitutive relations that we employ remain
valid for a solution x.
3.3. Gravitational potential energy due to weight. The gravitational po-
tential energy due to the lm weight is
E
film
(x, x) =
_

f
w
(u, x, x) dA,
where f
w
(u, x, x) = w
film
(u)(x k). Note that tendon ribbons also contribute to
E
film
. Assuming [x[ R and setting w = max
l
w
l
, we see
(3.5) wR f
w
(u, x, x) wR.
For almost every u and for all x R
3
, the mapping A f
w
(u, x, A) is constant
and hence convex. Thus, f
w
is quasi-convex in the third argument.
3.4. Constitutive relations for orthotropic materials. We present a brief
survey of key denitions related to orthotropic materials here. We will follow the
conventions of [16, Chapter 1]. The reference state of an elastic material is a natural
state without stress or strain; i.e., the components of the stress tensor are
ij
= 0,
and the components of the strain tensor are
ij
= 0. In three-dimensional linear
elasticity, is a linear function of ; i.e., = C : , and
(3.6)
ij
= C
ijkl

kl
for i, j, k, l = 1, 2, 3.
In a homogeneous medium, the C
ijkl
elements are the 81 components of the elastic
moduli tensor or stiness tensor C. Inverting (3.6), we nd = S : , where
(3.7)
ij
= S
ijkl

kl
for i, j, k, l = 1, 2, 3.





































































Copyright by SIAM. Unauthorized reproduction of this article is prohibited.
26 MICHAEL C. BARG, JIEUN LEE, AND FRANK BAGINSKI
The S
ijkl
elements are the 81 components of the elastic compliance in the compliance
tensor S. Using the symmetries of the tensors as previously dened, and following
[16, Chapter 1], we have the contracted notation I = i for i = j, I = 9 (i + j) for
i ,= j, and J = k for k = l, J = 9 (k + l) for k ,= l. Thus, (3.6) can be rewritten

I
= C
IJ

J
for I, J = 1, . . . , 6, where

I
=
ij
with I = i for i = j,

I
=
ij
with I = 9 (i +j) for i ,= j,

J
=
kl
with J = k for k = l,

J
= 2
kl
with J = 9 (k +l) for k ,= l.
With these conventions, (3.7) can be expressed in the form
(3.8)

S
11
S
12
S
13
0 0 0
S
12
S
22
S
23
0 0 0
S
13
S
23
S
33
0 0 0
0 0 0 S
44
0 0
0 0 0 0 S
55
0
0 0 0 0 0 S
66

(see [21, (2.20)]). The 66 matrix in (3.8), denoted by [S


ij
], involves nine independent
constants and is called the compliance matrix. We nd
(3.9) [S
ij
] =

1
E1

21
E2

31
E3
0 0 0

12
E1
1
E2

32
E3
0 0 0

13
E1

23
E2
1
E3
0 0 0
0 0 0
1
G23
0 0
0 0 0 0
1
G31
0
0 0 0 0 0
1
G12

,
where E
1
, E
2
, E
3
are Youngs moduli in the 1, 2, 3 directions, respectively;
ij
are
Poissons ratios for transverse strain in the j direction when stressed in the i direction;
and G
23
, G
31
, G
12
are shear moduli in the 2-3, 3-1, and 1-2 planes, respectively. Since
the compliance matrix is symmetric,
(3.10)
ij
/E
i
=
ji
/E
j
.
Using the notation in [16], where
4
=
23
,
5
=
31
,
6
=
12
,
4
=
23
,
5
=
31
, and

6
=
12
, we see that for a lamina in the 1-2 plane, a state of plane stress is dened
by setting
3
= 0,
23
= 0, and
31
= 0 in (3.8). This implies that auxiliary equations

3
= S
13

1
+ S
23

2
,
23
= 0, and
31
= 0 can be used to reduce (3.8) to
(3.11)

12

1
E1

21
E2
0

12
E1
1
E2
0
0 0
1
G12

12

.
If the matrix in (3.11) is nonsingular, we have
(3.12)

12

E1
11221
12E2
11221
0
21E1
11221
E2
11221
0
0 0 G
12

12

.





































































Copyright by SIAM. Unauthorized reproduction of this article is prohibited.
ORTHOTROPIC PNEUMATIC ENVELOPES 27
(a) (b)
Fig. 3.1. (a) E
1
/Ex, E
2
/Ey, G
12
/Gxy, and xy as functions of , where E
1
= 137 MPa,
E
2
= 164 MPa,
12
= 0.58,
21
= 0.69, and G
12
= 30 MPa; (b) (
1
,
2
)-domain for W
l
and (W
l
)

.
If a membrane is taut, then the 3 3 matrices in (3.11) and (3.12) are invertible.
However, in the presence of compressive stresses, a thin membrane will wrinkle and
the matrices in (3.11) and (3.12) are singular. To model wrinkling, we will follow the
energy relaxation work of Pipkin [23] as described in section 3.5. It is important to
note that applying standard membrane theory and allowing small stresses may not
accurately describe the response of a thin membrane under all relevant loading condi-
tions. The beauty of Pipkins approach is that it avoids the complications of dealing
with a singular stiness matrix. For other applications using Pipkins approach, see
[20, 24].
Properties such as Youngs modulus and Poissons ratio for a thin membrane
must be determined via mechanical testing on the lm of interest. Otherwise, the
constitutive relation will not describe the appropriate membrane response to a given
strain. For example, invoking the assumption of isotropy and applying standard three-
dimensional elasticity theory, one is led to the conditions 0 < <
1
2
(see [11, p. 129]).
However, a Poissons ratio greater than
1
2
is to be expected for thin PE lms (see, e.g.,
[9, 19]). Two-dimensional membrane and shell theories can be developed by asymp-
totic methods [12, Chapter 1] by letting the thickness go to zero. However, meaningful
results can be obtained using appropriate mechanical properties and assuming that
the thickness is piecewise-constant.
If the natural coordinates are not aligned with the principal material coordinates,
then it is possible to relate quantities in the dierent frames. In the case of plane
stress, where the 1-2 principal frame is rotated an amount from the xy-frame, we
have
(3.13)

xy

cos
2
sin
2
2 sin cos
sin
2
cos
2
2 sin cos
sin cos sin cos cos
2
sin
2

12

(see [16, section 1.5]). In the case of plane stress, the authors in [21, (2.97)] derived
relations for an orthotropic membrane that is stressed in nonprincipal xy-coordinates,
e.g., 1/E
x
= cos
4
/E
1
+(1/G
12
2
12
/E
1
) sin
2
cos
2
+sin
4
/E
2
. Using [21, (2.97)],
we present graphs of E
1
/E
x
, E
2
/E
y
, G
12
/G
xy
, and
xy
with 0
1
2
for a typical
PE lm at room temperature in Figure 3.1(a). In principle, one should be able to
estimate the mechanical properties of the membrane for the principal material direc-
tions and then use the matrix in (3.13) to transform the stresses of (3.12) when the





































































Copyright by SIAM. Unauthorized reproduction of this article is prohibited.
28 MICHAEL C. BARG, JIEUN LEE, AND FRANK BAGINSKI
membrane is stressed in nonprincipal xy-coordinates. However, the mechanical prop-
erties are only estimates, and they may lead to nonrealistic values in the transformed
frame. Moreover, G
12
is dicult to measure in the lab for thin PE, and the esti-
mates for G
12
are arguably unreliable. In our analysis, we will assume that E
1
(u, ),
E
2
(u, ), and
12
(u, ) for u
l
are known functions of determined by mechanical
testing on
l
-specimens (in practice, they will be determined for several directions:

i
=
1
2
(i 1)/k for i = 1, . . . , k +1). For the situation that is of primary interest to
us, thin membranes with no resistance to bending, there are four independent material
properties E
1
, E
2
,
12
, and G
12
when we account for the relation
12
/E
1
=
21
/E
2
.
In a state of plane stress, normal stresses dominate, and there is negligible resistance
to shearing. For these reasons, we do not include the shear stiness in our model. We
are led to
(3.14)
_

1

2
_
=
1
1
12

21
_
E
1

12
E
2

21
E
1
E
2
_ _

1

2
_
.
Assuming that the membrane thickness t(u) is piecewise-constant, it will be conve-
nient to relate the strains to stress resultants (see (3.15)). The principal directions
are based on the geometry as dened by (x, x). The response of the membrane to
strain will depend on (u, x). Strictly speaking, we should write E
1
(u, G), E
2
(u, G),
and
12
(u, G), where G is the Green strain. However, to simplify notation, we will
sometimes suppress G (or ) and write E
1
(u), E
2
(u), and
12
(u). If the dierence
between the E
i
s and the dierence between the
i
s are small, it may be more com-
putationally ecient to average the Youngs moduli and Poissons ratios and treat
the lm as if it were piecewise-isotropic. We follow this approach when we present
numerical results in section 5.
3.5. Film strain energy. The Green strain is G =
1
2
(C I), where C = F
T
F
is the right CauchyGreen strain and F = x. The eigenvalues of C are called the
Cauchy strains, denoted by
2
i
for i = 1, 2. The corresponding unit eigenvectors are
denoted by n
i
and are referred to as the principal directions. The eigenvalues of G
are called the principal strains and are denoted by
1
and
2
, where
i
=
1
2
(
2
i
1)
for i = 1, 2. Multiplying (3.14) by t(u) and using the Green strain tensor, we nd
that the tensor of the second PiolaKircho stress resultants (denoted by S) can be
written in the form
(3.15)
_

1
(u, G)

2
(u, G)
_
= (u, G)
_
E
1
(u, G)
12
(u)E
2
(u, G)

21
(u, G)E
1
(u, G) E
2
(u, G)
_ _

1
(G)

2
(G)
_
for u and (u, G) = t(u)/(1
12
(u, G)
21
(u, G)). It will be convenient to have
S in matrix form:
(3.16) S(u, G) = a(u, G)
_
A(u, G)G+b(u, G)Cof(G)
T
_
,
where
A(u, G) =
_
1
E2(u,G)
0
0
1
E1(u,G)
_
, a(u, G) =
E
1
(u, G)E
2
(u, G)
1
12
(u, G)
21
(u, G)
t(u),
and
(3.17) b(u, G) =
12
(u, G)/E
1
(u, G) =
21
(u, G)/E
2
(u, G).





































































Copyright by SIAM. Unauthorized reproduction of this article is prohibited.
ORTHOTROPIC PNEUMATIC ENVELOPES 29
The eigenvalues of S(u, G) are the principal stress resultants:

1
(u, G) = a(u, G) (
1
/E
2
(u, G) +b(u, G)
2
),

2
(u, G) = a(u, G) (
2
/E
1
(u, G) +b(u, G)
1
).
Since G and S are symmetric, they have the spectral representations
G =
1
n
1
n
1
+
2
n
2
n
2
, (3.18)
S(u, G) =
1
(u, G)n
1
n
1
+
2
(u, G)n
2
n
2
, (3.19)
respectively, where a b = (a
i
b

)
1im,1n
for a R
m
and b R
n
.
The balloon lm strain energy density can be written W(u, G) =
1
2
G: S(u, G),
which, in terms of the principal strains, gives
(3.20) W(u,
1
,
2
) =
1
2
a(u, G)
_
1
E
2
(u, G)

2
1
+
1
E
1
(u, G)

2
2
+ 2b(u, G)
1

2
_
.
Our aim is to obtain estimates on W in terms of [F[
2
=
2
1
+
2
2
. To accomplish this,
we rst estimate W on some
l
. Let u
l
. Let a
l
=
E
l
1
E
l
2
1
l
12

l
21
t
l
, A
l
=
_
1/E
l
2
0
0 1/E
l
1

,
and b
l
=

l
12
E
l
1
=

l
21
E
l
2
. Dene W
l
:= W(u,
1
,
2
) for u
l
, i.e.,
(3.21) W
l
=
1
2
a
l
_
1
E
l
2

2
1
+
1
E
l
1

2
2
+ 2b
l

2
_
for u
l
.
In terms of the Cauchy strains, W
l
is
W
l
(u,
1
,
2
) =
a
l
8
__
1
E
l
2

4
1
+
1
E
l
1

4
2
+ 2b
l
_

2
1

2
2
(
2
1
+
2
2
)
_
_
(3.22)
2
_
1
E
l
2

2
1
+
1
E
l
1

2
2
_
+
1
E
l
1
+
1
E
l
2
+ 2b
l
_
.
Suppose E
l
1
< E
l
2
. It is straightforward to show
(3.23)

l
21
[F[
4
/E
l
2

4
1
/E
l
2
+
4
2
/E
l
1
+ 2b
l

2
1

2
2
[F[
4
/E
l
1
,
(1 +
l
21
)[F[
2
/E
l
2
b
l
(
2
1
+
2
2
) +
2
1
/E
l
1
+
2
2
/E
l
2
(1 +
l
12
)[F[
2
/E
l
1
.
Combining (3.23) and (3.22), we obtain
a
l
8
_

l
21
E
l
2
[F[
4
2
(1 +
l
12
)
E
l
1
[F[
2
+
_
1
E
l
1
+
1
E
l
2
+ 2b
l
__
W(u,
1
,
2
)

a
l
8
_
1
E
l
1
[F[
4
2
(1 +
l
21
)
E
l
2
[F[
2
+
1
E
l
1
+
1
E
l
2
+ 2b
l
_
.
(3.24)
It is helpful to observe
l
12
/E
l
1
=
l
21
/E
l
2
< 1/E
l
2
and
l
21
/E
l
2
=
l
12
/E
l
1
< 1/E
l
1
. If
E
l
2
< E
l
1
, we can interchange the 1 and 2 indices in (3.24) to obtain similar inequalities.
Upon dening E
l
= min
i
E
l
i
and E
l
= max
i
E
l
i
, we nd
a
l
b
l
8
[F[
4

a
l
4
_
1
E
l
+ b
l
_
[F[
2
+
a
l
8
_
1
E
l
2
+
1
E
l
1
+ 2b
l
_
W
l

a
l
8
_
1
E
l
[F[
4
2
_
1
E
l
+b
l
_
[F[
2
+
_
1
E
l
1
+
1
E
l
2
+ 2b
l
__
.
(3.25)





































































Copyright by SIAM. Unauthorized reproduction of this article is prohibited.
30 MICHAEL C. BARG, JIEUN LEE, AND FRANK BAGINSKI
In an eort to further clarify our exposition, we set
(3.26)
E = min
l
E
l
,
E = max
l
E
l
,
a = min
l
a
l
,
a = max
l
a
l
,
b = min
l
b
l
,
b = max
l
b
l
.
Since a thin membrane is unable to support compressive stresses, we follow Pip-
kins approach [22, 23] and replace (3.20) with its quasi-convexication. The balloon
shape is approximated by a faceted surface based on a triangulation of . T ;
we consider T x(T) and calculate the corresponding principal stresses and strains.
The membrane can be decomposed into three disjoint sets:
Slack: S = u [
1
,
2
< 0,
Taut: T = u [
1
(u) 0,
2
(u) 0,
Wrinkled: U = u [ u / T and u / S.
U can be decomposed further as U = U
1
U
2
, where U
1
= u [
1
0,
2
(u) 0 and
U
2
= u [
2
0,
1
(u) 0.
The relaxation of W is the largest convex function not exceeding W. In [22],
Pipkin shows that the relaxation of W is the quasi-convexication of W if W is a
convex function of G. The convexity of (
1
,
2
) W(u,
1
,
2
) is equivalent to the
convexity of the function (u, x, y) = x
2
/E
2
(u)+y
2
/E
1
(u)+2b(u)xy. In this context,
we are speaking of convexity for xed u and xed . Let , 0, + = 1,
u
l
, and 0 <
l
12
,
l
21
< 1. Since 0 <
l
12

l
21
< 1 or, equivalently,
(3.27) 0 < E
l
1
E
l
2
(b
l
)
2
< 1,
we have det(D
2
) = 4(1
l
12

l
21
)/(E
l
1
E
l
2
) > 0. We nd that is convex; i.e.,
(u, x
1
+ x
2
, y
1
+ y
2
) (u, x
1
, y
1
) + (u, x
2
, y
2
). We conclude that W is
a convex function of G.
Let W

be the the relaxed strain energy density. On T, W

= W; on S, W

0;
on U, Pipkin shows that there exists a matrix G

such that W

(u, G) = W(u, G

),
(3.28) S

(u, G) = S(u, G

).
We can write the matrix G

as
(3.29) G

= G+
2
n n, where R, n R
2
, [n[ = 1.

2
n n is the wrinkling strain, and G

is the elastic strain (see [23]). This leads


to a uniaxial tension on U in the form
(3.30) S

= t t,
where > 0 and t R
2
is a unit vector orthogonal to n. One can demonstrate (3.30)
by determining and n. To this end, suppose
n S

n = 0, (3.31)
n S

t = 0. (3.32)
Here (3.31) ensures that the elastic strain produces a uniaxial tension on U; i.e., the
eigenvalues of S

(u, G) are nonnegative principal stress resultants. Equation (3.32)







































































Copyright by SIAM. Unauthorized reproduction of this article is prohibited.
ORTHOTROPIC PNEUMATIC ENVELOPES 31
is true since t and n are orthogonal directions. Adapting the analysis in [14] to an
orthotropic membrane, we let K(u) = a(u)
_
A(u)n n + b(u)Cof(n n)
T
_
. Using
(3.16), (3.28), and (3.29), we nd
(3.33) S

(u) = S(u) +
2
(u)K(u).
From the orthogonality of n and t, one has the identity Cof(n n)
T
= t t. It
follows that K(u) = a(u)A(u)n n +b(u)a(u)t t, where
K(u)n = a(u)A(u)n, (3.34)
K(u)t = b(u)a(u)t. (3.35)
From (3.31), (3.33), and (3.34), we have
(3.36)
2
(u) =
n S(u)n
a(u)n A(u)n
.
Now, if u U
2
=
2
0,
1
(u) 0, then we should take t = n
2
and n = n
1
.
In this case,
2
(u) =
1
E
2
(u)b(u)
2
=
1

21
(u)
2
. Similarly, if u U
1
=

1
0,
2
(u) 0, then take t = n
1
and n = n
2
so that
2
(u) =
2

12
(u)
1
.
Since G

(u) = G+
2
(u)n n, we have
(3.37) G

(u) =

0 if u S,

2
(n
2
n
2

21
(u)n
1
n
1
) if u U
2
,

1
(n
1
n
1

12
(u)n
2
n
2
) if u U
1
,

1
n
1
n
1
+
2
n
2
n
2
if u T.
On U, the principal strains of G

are in the form


2
,
21
(u)
2
or
1
,
12
(u)
1
.
Following Pipkins approach, we can model wrinkling by replacing W with W

, where
(3.38) W

(u,
1
,
2
) =

0,
1
,
2
< 0,
1
2
t(u)E
2
(u)
2
2
,
1
(u) 0,
2
0,
1
2
t(u)E
1
(u)
2
1
,
2
(u) 0,
1
0,
a(u)
2
_

2
1
E2(u)
+

2
2
E1(u)
+ 2b(u)
1

2
_
,
1
(u),
2
(u) 0.
System (3.38) is obtained by substituting the principal strains of G

into (3.20). We
can calculate S

=

G

or use (3.16) and (3.28) with (3.29), obtaining


(3.39) S

(u) =

0,
1
,
2
< 0,
t(u)E
2
(u)
2
n
2
n
2
,
1
(u) 0,
2
0,
t(u)E
1
(u)
1
n
1
n
1
,
2
(u) 0,
1
0,

1
(u)n
1
n
1
+
2
(u)n
2
n
2
,
1
(u),
2
(u) 0.
Next, we establish upper and lower inequalities for W

. Since W

is the relaxation
of W, we have the pointwise estimate W

W and (see [23, section 5])


(3.40) (W
l
)

W
l
.
To obtain a lower bound on W

, we will argue as in [4]. Figure 3.1(b) shows admissible


subsets of the principal strain space (
1
,
2
). For xed l, consider the following subsets:

l
= (
1
,
2
) [
1

1
2
,
2

1
2
,

l
1
= (
1
,
2
)
l
[
2
>
1
/
l
21
,
2
>
l
12

1
,

l
2
=
l

l
1
.





































































Copyright by SIAM. Unauthorized reproduction of this article is prohibited.
32 MICHAEL C. BARG, JIEUN LEE, AND FRANK BAGINSKI
For each l,
l
1
is unbounded, but (W
l
)

= W
l
for (
1
,
2
)
l
1
, and we have
already established a lower bound on W
l
. Furthermore, the compactness of
l
2
implies
that there exists a real number d
l
such that
d
l
= max
(1,2)
l
2
W
l
(
1
,
2
) (W
l
)

(
1
,
2
).
W
l
(
1
,
2
) has a critical point when
1
+
l
21

2
= 0 and
2
+
l
12

1
= 0, which im-
plies that (
1
,
2
) = (0, 0) is the only such critical point. Thus, d
l
is achieved on
the boundary of
l
2
at a corner point. By inspection, one nds that d
l
is achieved
at (0.5, 0.5), (W
l
)

(0.5, 0.5) = 0, and d


l
=
1
8
a
l
_
1
E
l
1
+
1
E
l
2
+ 2b
l
_
. Therefore,
W
l
(W
l
)

+
1
8
a
l
_
1
E
l
1
+
1
E
l
2
+ 2b
l
_
for (
1
,
2
)
l
. From (3.40), we have
(3.41) W
l

1
8
a
l
_
1
E
l
1
+
1
E
l
2
+ 2b
l
_
(W
l
)

W
l
.
Applying (3.25)(3.26) yields
(3.42)
1
8
a b[F[
4

a
4
_
b +
1
E
_
[F[
2
W


a
8E
[F[
4
+
1
4
a
_
1
E
+b
_
, u .
3.6. Tendons. In a real balloon, the load tendon is xed within a PE sleeve.
Only one edge of the attened sleeve is attached to the edge of the gore. For this
reason, we will model the tendon as a narrow ribbon of lm attached along a seam.
The strain energy in a ribbon segment (see Figure 2.1(b)) consisting of two identical
right triangles T G
i

l

with Poissons ratio


l

and Youngs modulus E


l

can be
approximated by
(3.43) S
film
= 2
_
T
W

dA 2
_
1
2
t
l

E
l

2
1
_
_
1
2
hL
_
=
1
2
t
l

E
l

2
1
hL.
The strain energy in a tendon segment with stiness K
t
and length L is
(3.44) S
t
=
_
L
1
2
K
t

2
(S) dS
1
2
K
t

2
1
L,
where (S) =
1
2
([

(S)[
2
1), and (S) is a parameterization of the deformed tendon.
The dierential of arc length along (S) is dS. Expression (3.44) follows from the
approximations (ds+dS)/(2dS) 1,
1
, and = (dsdS)/dS (ds+dS)/(2dS)
(dsdS)/dS. Choosing E
l

= K
t
/(t
l

h), we see that (3.43) and (3.44) are equivalent.


Remark 3.2. The approximation in (3.43) is valid because the lm, sleeve, and
tendon are very compliant. For stier membrane elements, these simplications may
not be realistic.
4. Existence results. In this section, we will establish rigorous existence re-
sults. The total energy of the balloon system is
(4.1) I(x) =
_

(W

(u, x) +f
P
(x, x) +f
w
(u, x)) dA.
Combining (3.4), (3.5), and (3.42) and simplifying, we obtain
1
8
a b[F[
4

_
1
4
a
_
b +
1
E
_
+
_
1
2
bR
2
+[p
0
[R
__
[F[
2
wR
W

+f
P
+f
w

1
8
a
1
E
[F[
4
+
_
1
2
bR
2
+[p
0
[R
_
[F[
2
+
1
4
a
_
1
E
+b
_
+wR.
(4.2)





































































Copyright by SIAM. Unauthorized reproduction of this article is prohibited.
ORTHOTROPIC PNEUMATIC ENVELOPES 33
Next, we restate a technical result that is proved in [4, Lemma 2].
Lemma 4.1. Let
i
> 0 for i = 1, . . . , 4. (i) There exist
1
,
1
such that
(4.3)
1
[u[
4
+
2
[u[
2

1
[u[
4
+
1
for 0 <
1
<
1
and
1
=
2
2
/4(
1

1
).
(ii) There exist constants ,
2
,
2
such that
(4.4) [u[
4

2

3
[u[
4

4
[u[
2

2
[u[
4
for 0 < <
3
<
2
and
2
=
2
4
/4(
3
).
Let
3
=
1
8
a b,
4
=
1
4
a
_
1
E
+b
_
+
1
2
bR
2
+[p
0
[R, and apply Lemma 4.1(ii) to the
rst inequality in (4.2). Choose such that 0 < <
1
8
a b and dene b

by =
1
8
ab

.
Then 0 < b

< b, and by Lemma 4.1(ii)


1
8
ab

[F[
4

2

3
[F[
4

4
[F[
2
,
where
2
=
_
a(b +
1
E
) +2bR
2
+4[p
0
[R
_
2
/ (8a(b b

)). Consider the second inequality


in (4.2). Choose

> 1 and apply Lemma 4.1(i) to obtain


1
8
a
1
E
[F[
4
+
_
1
2
bR
2
+[p
0
[R
_
[F[
2

1
8
a
1
E

[F[
4
+
1
,
where
1
=
_
bR
2
+ 2[p
0
[R
_
2
/
_
2a
1
E
(

1)
_
. Using the estimates in (4.2), we have
(4.5)
ab

8
[F[
4

2
wR W

+f
P
+f
w

a

8E
[F[
4
+
1
+
a
4
_
1
E
+b
_
+wR.
To obtain the proper form, we rewrite (4.5) as
1
8
ab

[F[
4
W

+f
P
+f
w
+
2
+wR

8E
[F[
4
+
2
+
a
4
_
1
E
+b
_
+
1
+ 2wR
(4.6)
and, with =
2
+wR, dene
(4.7) f

Tot
(u, x, x) = W

(u, x) +f
P
(x, x) +f
w
(u, x) +.
We can now state and prove a lemma needed for our existence results.
Lemma 4.2. If f

Tot
is as dened in (4.7) and [x[ R, then
(i) [A[
4
f

Tot
(u, x, A) C (1 + [A[
4
), where > 0 and C 0 are constants;
(ii) f

Tot
is a Caratheodory function;
(iii) A f

Tot
(u, x, A) is quasi-convex for every x and almost every u.
Proof. With C = max
_
a

/(8E),
2
+ a
_
1/E + b
_
/4 +
1
+ 2wR
_
, part (i)
follows from (4.6) with =
1
8
ab

. To show part (ii), we begin as follows. For a xed


(x, A) R
3
R
23
, we claim that u f

Tot
is measurable. Since any nite sum
of measurable functions is measurable, it suces to show that f
P
, , f
w
, and W

are measurable. The mappings u f


P
and u are continuous and, therefore,
measurable. u f
w
is measurable since f
w
(u, x, A) = w
film
(u)(x k) is piecewise-
constant. To show that u W

is measurable, we need an expression for the principal







































































Copyright by SIAM. Unauthorized reproduction of this article is prohibited.
34 MICHAEL C. BARG, JIEUN LEE, AND FRANK BAGINSKI
strains
i
(A) for i = 1, 2. We then obtain an expression for W

(u, A) from (3.38).


With
A = (A
1
, A
2
) =

A
1
1
A
1
2
A
2
1
A
2
2
A
3
1
A
3
2

R
23
,
let A
1
A
2
=

3
j=1
A
j
1
A
j
2
. It follows that

1
(A) =
1
4
([A[
2
2) +
1
4
_
([A
1
[
2
[A
2
[
2
)
2
+ 4(A
1
A
2
)
2
,

2
(A) =
1
4
([A[
2
2)
1
4
_
([A
1
[
2
[A
2
[
2
)
2
+ 4(A
1
A
2
)
2
.
To determine W

(u, A), we substitute


1
(A),
2
(A) into (3.38) and nd
(4.8) W

(u, A) = W

(u,
1
(A),
2
(A)).
If
1
< 0 and
2
< 0, then u W

0 is measurable. For all other cases in (3.38),


u W

is measurable since it is a linear combination of nite products of measurable


functions t(u), a(u), b(u), and E
i
(u). Thus, u W

(u, A) is measurable, and we


conclude that for xed A R
23
, u f

Tot
is measurable.
We now verify the second condition that a Caratheodory function must satisfy.
Let =

l,i

l
i
, where
l
i
is a partition of as dened in section 3.1. Note that
meas() = 0. For a xed u , we claim that (x, A) f

Tot
is continuous. To
show this, we will observe that each of f
P
, , f
w
, and W

is continuous. (x, A)
is continuous since it is a constant mapping. Since the
l
i
s are disjoint, u
implies that u
l
i
for some i and l. For u
l
i
, (x, A) f
w
(u, x, A) = w
l
(x k) is
continuous since it is linear. (x, A) f
P
(u, x, A) =
_
1
2
b(x k)
2
+p
0
(x k)
_
k adj
2
A
is continuous since A adj
2
A is continuous (see [4, Lemma 3]). Finally, one can
verify that W

is a continuous function of (
1
,
2
). Since A
i
(A) for i = 1, 2 are
continuous, (x, A) W

is continuous. Thus, for almost every u , (x, A)


f

Tot
(u, x, A) is continuous, and f

Tot
is a Caratheodory function.
For the proof of part (iii), x u and x R
3
. As above, u
l
i
for some i
and l. We will show that f

Tot
is quasi-convex in the third argument by showing that
each of f
P
, , f
w
, and W

is quasi-convex in the third argument. From Denitions


2.12.2, one veries that A and A f
w
(u, x, A) = w
film
(u)(x k) are quasi-
convex since both are constant in A. By construction, A W

is quasi-convex. Since
A f
P
is quasi-convex and the nite sum of quasi-convex functions is quasi-convex,
(iii) is satised, and the proof of the lemma is complete.
Since all of the forces are conservative, equilibrium is achieved at a minimum
of the energy functional over X. As the term only increases W

+ f
P
+ f
w
by a
constant, the minimizer of I in (4.1) is the same as the minimizer of I with integrand
W

+f
P
+f
w
+. With f

Tot
given by (4.7) and I(x) redened, we state an existence
result for a closed balloon system. The arguments are easily adjusted to accommodate
an open balloon system and other boundary conditions.
Theorem 4.3. If [x[ R, then
(B) inf
_
I(x) =
_

Tot
(u, x, x) dA

x X
_
admits at least one solution.





































































Copyright by SIAM. Unauthorized reproduction of this article is prohibited.
ORTHOTROPIC PNEUMATIC ENVELOPES 35
Proof. T is nonempty since we can always nd an x x
0
+W
1,4
0
(, R
3
) param-
eterizing a spherical cap with volume
0
. To complete the proof, we need to show
that f

Tot
is a quasi-convex Caratheodory function and verify Theorem A.2(i). This is
accomplished in Lemma 4.2. Since X without the volume constraint is weakly closed,
there exists x X without the volume constraint such that I( x) = infI(x) [ x X.
We must show that x satises the volume constraint. As in the proof of Theorem A.2
(see [7, p. 23]), there is a minimizing sequence x
m
X satisfying x
m
x weakly
in W
1,4
(, R
3
). By Lemma A.1, V ( x) =
0
. Thus, x X when g = 0 includes the
volume constraint. We conclude that (B) has at least one solution.
Remark 4.4. The boundary condition from (2.3) is a convex constraint. It
follows that x X when g includes the boundary condition dened in (2.3).
5. Numerical solutions. To demonstrate the robustness of our model, we sim-
ulate balloon ascent shapes and investigate low pressure regimes where the symmetric
conguration becomes unstable and undergoes a localized collapse. We used the de-
sign code Planetary Balloon [17] developed by Farley/NASA Goddard Space Flight
Center to determine the cutting pattern G
F
(see Table 5.1). We used Brakkes Sur-
face Evolver [10] to determine stability of a symmetric state. At the design pressure
P
0,d
, the pumpkin shape is fully developed and the symmetric shape is stable. To
mimic ascent shapes, we decrease the nadir pressure P
0
. We nd that the symmetric
shapes are stable for P
0
> 1 Pa and unstable for P
0
< 1 Pa.
Table 5.1
Design parameters for a pumpkin-shaped balloon.
(a) Key properties. (b) Cutting pattern for G
F
. Units are in meters.
Variable Value
ng 60
P
0,d
155 Pa
h 38 m
0.636

E 150.5 MPa
Et 693 kN
v 2H(v)
0.0000 0.0346
1.6401 0.1217
4.9295 0.2931
6.5743 0.3782
8.2192 0.4632
9.8641 0.5481
11.5134 0.6332
14.8033 0.8014
16.4483 0.8839
18.0933 0.9647
19.7383 1.0435
21.3878 1.12
23.0329 1.1932
24.6781 1.2629
v 2H(v)
27.9687 1.3891
29.6141 1.4422
32.9095 1.528
34.5552 1.5716
36.2007 1.6189
37.846 1.6658
39.4912 1.7072
42.7835 1.7414
44.4271 1.7278
46.0706 1.7042
47.714 1.6712
49.3574 1.6293
51.0006 1.5792
52.6481 1.5216
v 2H(v)
55.9339 1.3884
59.2192 1.2364
60.8617 1.1555
62.5086 1.0721
64.1509 0.9871
65.7931 0.901
69.0774 0.727
70.7196 0.6398
72.3661 0.553
74.0082 0.4669
75.6502 0.3811
77.2921 0.2953
78.9341 0.2094
82.2221 0.0351
We assume that the balloon is constructed from PE lm with
1
= 0.58,
2
=
0.692, E
1
= 137 MPa, and E
2
= 164 MPa. These values correspond to a lm
temperature of 253 K and are based on empirical tests carried out at the Balloon Lab
at NASAs Wallops Flight Facility. Note that (3.17) is only approximately satised.
Since the values of
i
and E
i
dier by no more than 10%, we will assume that the
balloon lm is isotropic and use a Youngs modulus of

E = (E
1
+E
2
)/2 and a Poissons
ratio of = (
1
+
2
)/2. See Table 5.1(a) for other parameter values. The mass
density of PE is 920 kg/m
3
and the cap length is 30 m. The equatorial bulge radius
of a fully inated strained gore lobe is estimated to be 2.87 m, and the corresponding
equatorial bulge angle is 70 degrees. The tendon has a mass density of 0.0127 kg/m.
A tendon is assumed to be 0.3% slack, so it must strain this amount before coming
under tension. For our calculations, we use a standard linearly elastic string model





































































Copyright by SIAM. Unauthorized reproduction of this article is prohibited.
36 MICHAEL C. BARG, JIEUN LEE, AND FRANK BAGINSKI
Table 5.2
Summary of strained balloon shape calculations.
Description/Pressure P
0
(Pa) 155 50 10 2.5 0.5 -2.00
Max. height (m) 39.4333 39.5344 40.6711 43.0638 46.4921 52.0328
Apex height (m) 39.3381 39.4350 40.5465 42.89486 45.7803 51.0624
Diameter (m) 63.4805 62.8063 62.1420 60.9320 59.5393 56.0676
Max. principal stress 1 (MPa) 8.2900 3.6901 1.9318 0.2416 1.0558 0.4579
Max. principal stress 2 (MPa) 9.1129 4.1770 2.2101 1.4754 1.3963 1.1406
Max. principal strain 1 (m/m) 0.02716 0.00827 -0.00058 -0.00197 0.00039 0.00135
Max. principal strain 2 (m/m) 0.03675 0.02177 0.01293 0.00317 0.00596 0.00307
Max. tendon tension (N) 7539 2474 609 277 229 135
Max. tendon strain (m/m) 0.0109 0.0036 0.0009 0.0004 0.003 0.0002
Volume ratio (/
0
) 1.000 0.977 0.964 0.944 0.904 0.7724
for the tendons. The design shape has a height of 40.8 m and a diameter of 63 m. We
assume b
d
= 0.1034 N/m
3
. The apex tting has a mass density of 42 kg. The nadir
end-tting has a diameter of 1.32 m.
The balloon is approximated by a faceted surface consisting of triangles whose
(x, y, z) coordinates are free to move. Let q = (q
1
, q
2
, . . . , q
N
) be the vector consisting
of the (x, y, z) coordinates of the facet nodes, and let I(q) represent the resulting
energy of the balloon system based on the discretization of I(x). With 320 triangles
per gore, there are 28,981 degrees of freedom. The quantities I(q), DI = [I/q
i
],
D
2
I =
_

2
I/q
j
q
i

are calculated analytically and are used by Matlabs optimization


toolbox function fminunc to solve problem (B) when there are only linear constraints.
Our model is also incorporated into Surface Evolver [10]. See [5] for other Evolver
balloon applications. We used Evolver to determine that the cyclically symmetric
shape is stable for the design conditions at oat, i.e., P
0,d
= 155 Pa. A solution q

is
stable if all the eigenvalues of D
2
I(q

) are positive. At P
0
1 Pa, D
2
I(q

) has one
negative eigenvalue.
We rst calculated the strained shape for P
0
= 155 Pa. The numerical val-
ues of quantities such as the height, diameter, maximum principal stresses, etc., are
presented in Table 5.2. The corresponding equilibrium shape is presented in Figure
5.1(a). We varied P
0
and determined the corresponding strained shapes. Data is
presented in Table 5.2 for P
0
= 155, 50, 10, 2.50, 0.50, 2.00 Pa. These shapes are
representative of ascent congurations. In each case, the buoyancy b was adjusted
so that b1 b
d
1
d
. To the eye, there is very little dierence between the shapes for
10 P
0
155. However, quantities do change signicantly (see Table 5.2). The
shape at P
0
= 2.5 Pa is cyclically symmetric (see Figure 5.1(b)), but it is not at at
the nadir. At about P
0
= 1 Pa, all 60 gores are not able to fully deploy, and in the case
of our simulation, two gores collapse and are swallowed into the interior of the bal-
loon. This is a typical feature in partially inated balloons. See Figure 5.1(c) where
an equilibrium shape with P
0
= 0.50 Pa is presented. Figure 5.2 contains images of
a 27 m diameter pumpkin balloon with n
g
= 200 that were taken during ination
tests carried out by NASAs Balloon Program Oce at the TCOM Manufacturing
and Flight Test Facility in Elizabeth City, NC in 2007. Figure 5.2(a) is an image
of a pumpkin balloon inated to less than 5% of full capacity. It is characteristic of
the shape of an ascending balloon shortly after launch. In Figure 5.2(b), the same
balloon at about 66% of full capacity is displayed. There are a number of locations
about the circumference of the balloon where contiguous gores are swallowed inward.
In general, these locations are not distributed symmetrically. In Figure 5.2(c), the
nadir pressure has been increased, and all the gores are deployed, even though the





































































Copyright by SIAM. Unauthorized reproduction of this article is prohibited.
ORTHOTROPIC PNEUMATIC ENVELOPES 37
(a) (b)
(c) (d)
Fig. 5.1. Family of ascent shapes. (a) P
0
= 155 Pa; (b) P
0
= 2.5 Pa; (c) P
0
= 0.5 Pa, two
gores collapse; (d) P
0
= 2.0 Pa, taut regions are shaded dark; wrinkled regions are bright.
pumpkin shape is not fully developed. In Figure 5.2(d), the nadir pressure has been
increased to P
0,d
, and the pumpkin shape is fully developed. If we attempt to calculate
equilibrium congurations for 1/1
d
< 0.66, the self-contact regions are more compli-
cated, requiring more sophisticated contact models. In the present work, we handle
the contact problem by manually gathering excess material along certain curves and
evolving equilibrium shapes with local constraints as described in [6]. Gathering does
not impede the solution process and provides a realistic way to store excess material
in a uniaxial state.
When 50 P
0
155, most triangles are taut. However, when P
0
is near 10 Pa,
one notices an increase in wrinkling near the tendons in the southern hemisphere of the
shape. As P
0
decreases, the amount of wrinkling becomes more pronounced. When
P
0
= 2.0 Pa, we color the triangles according to their state (see Figure 5.1(d)). A
taut triangle is colored dark blue. The edge of the cap is indicated by the red curve.
Wrinkled triangles are colored according to the wrinkle strain
2
from (3.36). If
2
is small, the triangle is colored light blue. Triangles with increasing values of
2
are
colored cyan, green, yellow, and red, respectively. A few triangles are slack, and these
are colored white. The wrinkle pattern in our numerical solution is very consistent
with what is observed in a real balloon (see Figure 5.2). It would be natural to assume
symmetry when modeling shapes such as those in Figure 5.2(c)(d). However, such
an assumption leads to the exclusion of nonsymmetric shapes like those in Figures





































































Copyright by SIAM. Unauthorized reproduction of this article is prohibited.
38 MICHAEL C. BARG, JIEUN LEE, AND FRANK BAGINSKI
(a) V/V
d
< 0.05. (b) V/V
d
0.66.
(c) V/V
d
0.95. (d) V/V
d
1.00.
Fig. 5.2. Ascent shapes: (a) at launch, (b) some gores are internally ingested, (c) not fully
developed, (d) pumpkin shape is fully developed. Images courtesy of NASAs Balloon Program Oce.
5.2(a)(b). As Figure 5.1 shows, we are able to capture a variety of features by
modeling a complete shape. Since there may be many wrinkle/fold patterns that give
rise to the same stress distribution in the lm, solutions like those of Figure 5.1(c)(d)
need not be unique. However, they provide the balloon designer with estimates of
lm stresses in these o-design states.
6. Conclusions. In this paper, we present a mathematical model for a tendon-
reinforced piecewise-orthotropic pressurized membrane. We apply our methods to
the problem of determining the equilibrium shape of a pumpkin-shaped balloon with
an external cap, and calculate numerical solutions for a complete balloon without
symmetry assumptions. We establish estimates on the relaxed lm strain energy
density and show that it is quasi-convex. We apply direct methods in the calculus of
variations to rigorously show that solutions exist to this class of problems. Our results





































































Copyright by SIAM. Unauthorized reproduction of this article is prohibited.
ORTHOTROPIC PNEUMATIC ENVELOPES 39
hold when the constant dierential pressure is unknown and a volume constraint is
applied. Linear constraints, like those used to handle contact problems, can also be
included. We nd that our numerical model is able to capture many nonstandard
features, such as wrinkling and self-contact, that are observed in real balloons.
Appendix A. Direct methods in the calculus of variations. For the con-
venience of the reader, we record here results that appear with proof in [7].
Lemma A.1. If x
k
x weakly in W
1,4
(, R
3
), then V (x
k
) V ( x), where
V (x) =
1
3
_

x x
u
x
v
dA is the volume functional.
To prove Lemma A.1, one can show for a constant K > 0
3[V (x
k
) V ( x)[ K|x
k
x|

x (adj
2
x
k
adj
2
x) dA

.
Letting k , the proof follows as a consequence of the compact imbedding
C
0,
(, R
3
) W
1,4
(, R
3
) with 0 <
1
2
, the uniform boundedness of [x
k
[ in
L
4
() and L
2
(), and the weak convergence adj
2
x
k
adj
2
x in L
2
(, R
3
).
Theorem A.2. Let R
n
be a bounded open set. Let f : R
m
R
nm
R
be a quasi-convex Caratheodory function satisfying (i) [A[
p
f(u, x, A) q(u) +
C([x[
p
+[A[
p
), where p > 1, > 0 are constant, q is a nonnegative locally summable
function, and C is a nonnegative constant. If
(P) inf
_
I(x) =
_

f(u, x(u), x(u)) dA

x x
0
+W
1,p
0
(, R
m
)
_
,
then (P) admits at least one solution.
Theorem A.2 is a modication of [15, Theorem 2.9, p. 180]. To establish the
result, one must demonstrate that I(x) is coercive and sequentially weakly lower
semicontinuous. The coercivity follows from the rst estimate in (i). Sequential weak
lower semicontinuity of I(x) is guaranteed by [1, Theorem II.4, p. 137]. The proof of
Theorem A.2 is standard and can be found in [7, p. 23] and [15, p. 181].
REFERENCES
[1] E. Acerbi and N. Fusco, Semicontinuity problems in the calculus of variations, Arch. Ration.
Mech. Anal., 86 (1984), pp. 125145.
[2] T. Aubin, Nonlinear Analysis on Manifolds. Monge-Amp`ere Equations, Springer-Verlag, New
York, 1982.
[3] F. Baginski, On the design and analysis of inated membranes: Natural and pumpkin shaped
balloons, SIAM J. Appl. Math., 65 (2005), pp. 838857.
[4] F. Baginski, M. Barg, and W. Collier, Existence theorems for thin inated wrinkled mem-
branes subjected to a hydrostatic pressure load, Math. Mech. Solids, 13 (2008), pp. 532570.
[5] F. Baginski and K. Brakke, Exploring the stability landscape of constant stress pumpkin
balloon designs, AIAA J. Aircraft, 47 (2010), pp. 849857.
[6] F. Baginski and K. Brakke, Simulating clefts in pumpkin balloons, Adv. Space Res., 45
(2010), pp. 473481.
[7] M. Barg, Direct Methods in the Calculus of Variations with Applications to Tendon-Reinforced
Piecewise-Isotropic Membranes, Ph.D. thesis, Department of Mathematics, The George
Washington University, Washington, DC, 2007.
[8] J. Bensinger, L. Cadonati, F. Calaprice, E. de Haas, R. Fernholz, R. Ford, C. Galbiati,
A. Goretti, E. Harding, A. Ianni, S. Kidner, M. Leung, F. Loeser, K. McCarty, A.
Nelson, R. Parsells, A. Pocar, T. Shutt, A. Sonnenschein, and R. B. Vogelaar,
The nylon scintillator containment vessels for the Borexino solar neutrino experiment,
Nucl. Instr. Meth. Phys. Res. Sect. A, 582 (2007), pp. 509534.





































































Copyright by SIAM. Unauthorized reproduction of this article is prohibited.
40 MICHAEL C. BARG, JIEUN LEE, AND FRANK BAGINSKI
[9] J. Blandino, J. Sterling, F. Baginski, E. Steadman, J. T. Black, and R. Pappa, Opti-
cal strain measurement of an inated cylinder using photogrammetry with application to
scientic balloons, in Proceedings of the 45th AIAA/ASME/ASCE/AHS/ASC Structures,
Structural Dynamics, and Materials Conference, Palm Springs, CA, 2004, AIAA, Reston,
VA, paper AIAA-2004-1500.
[10] K. Brakke, The Surface Evolver, Experiment. Math., 1 (1992), pp. 141165; see also
http://www. susqu.edu/brakke/evolver.
[11] P. G. Ciarlet, Mathematical Elasticity Volume I: Three Dimensional Elasticity, North
Holland, New York, 1993.
[12] P. G. Ciarlet, Mathematical Elasticity Volume III: Theory of Shells, NorthHolland, New
York, 2000.
[13] M. Coleman and F. Baginski, A shape deformation study of large aperture inatable elastic
parabolic antenna reectors, in Proceedings of the 51st AIAA/ASME/AHS/ASC Struc-
tures, Structural Dynamics, and Materials Conference, Orlando, FL, 2010, AIAA, Reston,
VA, paper AIAA-2010-2501.
[14] W. Collier, Estimating stresses in a partially inated high altitude balloon using a relaxed
energy, Quart. Appl. Math., 61 (2003), pp. 1740.
[15] B. Dacorogna, Direct Methods in the Calculus of Variations, Springer-Verlag, New York,
1989.
[16] C. Decolon, Analysis of Composite Structures, Hermes Science Publication, Paris, 2000.
[17] R. Farley, Planetary Balloon: Balloon Design Software Manual, NASA Goddard Space Flight
Center, Greenbelt, MD, 2007.
[18] E. M. Flint, J. E. Lindler, D. L. Thomas, and R. Romanofsky, RF performance of mem-
brane aperture shells, in Proceedings of the 48th AIAA/ASME/ASCE/AHS/ASC Struc-
tures, Structural Dynamics, and Materials Conference, Honolulu, HI, 2007, AIAA, Reston,
VA, paper AIAA 2007-1831.
[19] C. Guo and L. Wheeler, Extreme Poissons ratios and related elastic crystal properties, J.
Mech. Phys. Solids, 54 (2006), pp. 690707.
[20] E. Haseganu and D. J. Steigmann, Analysis of partly wrinkled membranes by the method of
dynamic relaxation, Comput. Mech., 14 (1994), pp. 596614.
[21] R. M. Jones, Mechanics of Composite Materials, 2nd ed., McGrawHill Book Company, New
York, 1975.
[22] A. C. Pipkin, Relaxed energy densities for small deformations of membranes, IMA J. Appl.
Math., 50 (1993), pp. 225237.
[23] A. C. Pipkin, Relaxed energy densities for large deformations of membranes, IMA J. Appl.
Math., 52 (1994), pp. 297308.
[24] D. J. Steigmann and A. C. Pipkin, Finite deformation of wrinkled membranes, Quart. J.
Mech. Appl. Math., 42 (1989), pp. 427440.

Você também pode gostar