Você está na página 1de 10

A new approach to geometric nonlinearity of cable

structures
A.S.K. Kwan
School of Engineering, University of Wales Cardi, PO Box 917, Cardi CF2 1XH, UK
Received 8 January 1997; received in revised form 1 January 1998
Abstract
The basic structural principles surrounding nonlinear behaviour of cable networks are explained through the
example of a two-link structure. The nonlinear static response to load for this structure is then derived explicitly
using the proposed simple approach, and results are compared with those obtained from a general two-dimensional
non-linear bar element (derivation given), and to results quoted in the literature. The proposed approach to
geometric nonlinearity is then tested on three three-dimensional cable networks and the results compared with those
obtained by three other techniques, namely geometric stiness matrix, dynamic relaxation and general minimum
energy. The proposed technique has been found to be comparable to established techniques in accuracy, stability
and speed of solution while at the same time exhibiting the key features of separation of the numerical computation
from the underlying structural mechanics, and the requirement of understanding only the most elementary of
structural mechanics. The proposed technique is thus also most suitable for introducing cable structures to
undergraduate courses. # 1998 Elsevier Science Ltd. All rights reserved.
Keywords: Cable structures; Geometric nonlinearity; Innitesimal mechanisms; Large displacement analysis; Nonlinear analysis
1. Introduction and purpose of paper
It is clear from the emergence of small cable nets and
fabric structures over doorways to shopping malls,
supermarkets and petrol stations, that such structures
are considered to be elegant and aesthetically pleasing.
There are, however, only a few large-scale examples of
such structures adorning prestigious structures. From a
structural engineering point of view, there can be two
reasons for this. Firstly, cable networks are seen as
highly exible structures. The problem however, is not
in the exibility per se, but in the nature of the exi-
bility, which is not widely expounded in teaching insti-
tutions. The second reason why cable network
examples are few is because they have an entirely non-
linear response to load. Where nonlinear compu-
tational mechanics is taught in undergraduate courses,
the emphasis is on the algorithms, aspects of the com-
putation and computer programming. Such an
approach does not encourage the development of an
intuitive feel for nonlinear behaviour. For these
reasons, it is possible to believe cable networks rather
mystical structures that can only be analysed by some
complex computer programs restricted to a few highly
specialized engineering rms. Such a misconception is
usually a sucient reason to place more conventional
structural forms over cable networks. One of the key
emphases of this paper therefore is the clarity of the
fundamental characteristics of cable networks.
The aim of this paper is to both oer a simple expla-
nation of the nonlinear exibility inherent in cable net-
works through a straightforward example, and to
propose an analytical technique for the nonlinear static
behavior of cable networks which separates the com-
plexity of a nonlinear numerical algorithm from the
underlying structural principles. However, far from
being a purely demonstrative teaching technique, the
proposed technique has been found to be comparable
to established techniques in speed and accuracy of sol-
utions. A brief review of the main existing techniques
is given in Section 2. Section 3 provides a discussion
on the nature of the exibility in cablenet structures;
Computers and Structures 67 (1998) 243252
0045-7949/98/$19.00 # 1998 Elsevier Science Ltd. All rights reserved.
PII: S0045- 7949( 98) 00052- 2
PERGAMON
this forms the basis of the proposed technique
described in Section 4. Some computational aspects are
examined in Section 5 and the technique is illustrated
with comparative examples in Section 6. Concluding
remarks are made in Section 7.
2. Review of existing techniques
Several techniques exist for the static (and dynamic)
analysis of cable structures and a brief outline is pro-
vided in this section. The purpose of this short review
is not to present a detailed comparison but to illustrate
the relative complexity previous researchers have
applied to this problem.
2.1. Dynamic relaxation
Dynamic Relaxation presents itself as an attractive
approach to the design of prestressed cable networks
because both form-nding and analysis can be carried
out within a single analytical framework. The tech-
nique traces the motion of the structure from the point
of loading to the nal steady equilibrium state (mini-
mum energy) through use of the D'Alembert principle:
pt M

d C
_
d Kd 1
where p(t) is the time-dependent vector of externally
applied load, M is a ctitious mass matrix, C the
matrix of ctitious damping coecients, K is the sti-
ness matrix and d is the vector of displacement. The
rst two terms on the right-hand side of Eq. (1) rep-
resent the residual, or out of balance force, which is
iteratively minimized to an acceptably low value, at
which point the structure reaches steady equilibrium.
Iteration is carried out in small time steps using a nite
dierence approach to nd values for changes in nodal
displacements. The speed and stability of the whole
process is critically governed by the ctitious mass
assigned to the nodes, the choice of damping coe-
cients, as well as the size of the time step. Lewis et
al. [9] and Barnes [1] give details on how values can be
assigned to these parameters.
Although dynamic relaxation is generally seen as a
specialist technique, it is a popular choice in the design
of cable networks and membrane structures. Lewis [10]
reported it to be more stable and ecient for struc-
tures with large degrees of freedom than the stiness
matrix approach outlined below.
2.2. Stiness matrix method
The stiness matrix approach, solved iteratively by
the NewtonRaphson method, is fully described by
Krishna [6] who also supplied a computer program
listing. For a prestressed cablenet in equilibrium,
instead of the usual p = Kd stiness equation for a
geometrically linear structure (where symbols have
meanings dened above), the equivalent relationship is
p = Kd R(d) where R(d), the vector of residual
forces, is a nonlinear function of d. In addition to the
NewtonRaphson iteration, Krishna also advocated
incremental loading for situations where error accumu-
lation is too large.
2.3. Minimizing total potential energy
The approach of minimizing total potential energy
has been described by Coyette and Guisset [3], Suan
and Templeman [15], and Stefanou et al. [14]. The
theoretical approach is essentially the same in each
case, but the choice of the minimizing algorithm dif-
fers. Coyette and Guisset favoured a reduced-gradient
algorithm in the MINOS library; Suan and
Templeman chose a quasi-Newton algorithm from the
NAG library; and Stefanou et al. used the conjugate
gradient method.
2.4. Approximate linear methods
Pellegrino [12] proposed an approximate linear
approach to geometric nonlinearity by observing that a
general load on a cablenet can be decomposed into
two parts, i.e. one part which causes extensional dis-
placements, and the second which excites inextensional
displacements (innitesimal mechanisms). Geometric
nonlinearity arises from the latter which Pellegrino ver-
ied, by experiments, to be small for prestressed
cablenets and hence a linear approximation he pro-
posed to determine the magnitude of the second type
of displacements was suciently good for practical
purposes. Vilnay and Rogers [16] also approached the
problem in the same way.
2.5. Finite element method
While the geometric stiness matrix method outlined
above is strictly speaking a form of nite element
method (FEM) in which each cable segment is rep-
resented by a single element, some researchers have
applied a more general FEM approach. Gambhir and
de Batchelor [5], for example, developed a curved
member for shallow cablenets. Their technique, which
ensured continuity of slope across nodes, was devel-
oped primarily for dynamic analysis but also produced
static response to load. Mitsugi [11] formulated a sti-
ness matrix for the ``hypercable'' (a cable connected to
intermediate pulleys along its length, also known as
``active cable'' after Kwan et al. [7]) element to be
incorporated into an FEM solver.
A. Kwan / Computers and Structures 67 (1998) 243252 244
3. Inherent characteristics of prestressed mechanisms
While we can instinctively say that cable networks
are more exible than other structural forms, the
nature of that exibility and how it can be controlled
are often less intuitive. We shall examine this exible
behaviour through the simple two-link planar structure
shown in Fig. 1. Links AB and BC shall be treated as
capable of sustaining both tensile and compressive
axial forces, even though cable elements are unable to
carry compressive forces, because the structure shall be
prestressed to a suciently high level such that the lar-
gest compressive axial bar force due to load can be
overcome by the prestress; a compressive bar force is
thus sustained through a reduction in prestress tensile
force.
The well-known but unsophisticated Maxwell's rule
for testing statical/kinematic determinacy (3j-b-k,
where j = number of joints, b = number of bars, and
k = number of degrees of freedom removed by sup-
ports) classies this two-link structure as a statically
determinate structure with no mobility, which it clearly
is not. There can be no vertical component of bar ten-
sion in the two bars to equilibrate a vertical load P
applied at B. In this conguration, the structure is
unable to resist the load; it is a mechanism. The struc-
ture however, undergoes ``large deformation'', and in a
displaced conguration (dotted line in Fig. 1), the
inclined bars AB and BC are capable of equilibrating
the load P. Since the mechanism is stiened by the dis-
placement, it is an ``innitesimal mechanism'', as
opposed to the more common large deformation,
``nite mechanism''.
Such structures with simultaneous statical and kin-
ematic indeterminacy prompted Calladine [2] to modify
Maxwell's rule to 3j-b-k = s-m (s = number of states
of self-stress, and m = total number of nite and in-
nitesimal mechanisms). For the structure in Fig. 1,
there is a single innitesimal mechanism (as shown by
the dotted line) and a single state of self-stress which
involves the two collinear bars having equal bar ten-
sion. Note that this state of selfstress requires no exter-
nal load for equilibrium and hence the term self-stress.
We should therefore clearly attribute the ``exibility''
of cable networks not to the low axial stiness of the
constituent cables (which is often not true), but to the
geometry of the structure. If point B had been located
some distance below the level of A and C, the structure
would indeed be statically determinate with no mech-
anism. Since the two-link structure in its original con-
guration is unable to carry the load P, but the
mechanism is stiened by a vertical movement of B,
then the PD relationship is clearly nonlinear. The
stiness of the structure to an applied load is depen-
dent on its displaced geometry. The nonlinearity how-
ever, is not to be attributed to a nonlinear stressstrain
relationship of the cable material (which is often not
true), but is due to the geometry of the two-link struc-
ture. The nonlinear behavior arises not so much out of
the structure undergoing large deformation beyond the
limit of small deection theory, but precisely because
small deection theory is not applicable to cable net-
works.
Having established the fact that cable structures
have a nonlinear response to load because they are in-
nitesimal mechanisms in their undeformed congur-
ation, and that the nonlinearity is directly dependent
on the geometry of the structure, we shall now devise
the nonlinear ``large deection'' PD relationship for
the two-link structure in Fig. 1, which is preloaded
with a state of self-stress of t
o
in the two bars. It will
be seen that the treatment outlined below, following
the aim of this technique, requires only understanding
of elementary structural mechanics. We shall examine
the compatibility (elongationdisplacement), equili-
brium (loadtension), and constitutive (elongationten-
sion) relationships in turn, and then combine these
relationships to form the PD equation.
3.1. Compatibility
Consider that point B has lowered by D under the
load P, so that bar AB now has length
L' = L
o
+e = L
o
secb. The expression for secb can be
substituted by its Taylor's series so that
L
o
secb L
o
1
b
2
2!

5b
4
4!
. . .
_ _
; and that
e L
o

L
o
b
2
2!

5L
o
b
4
4!
. . .
_ _
L
o
and when we ignore the fourth order term,
Fig. 1. Planar two-link structure composing of bars AB and
BC, both initially pretensioned by t
o
. Joint B is displaced by
D under the vertical load P, and the bar tensions increase by
t.
A. Kwan / Computers and Structures 67 (1998) 243252 245
e
L
o
b
2
2!
; but b1tan b
D
L
o

L
o
D
2
L
2
o
2!

D
2
2L
o
: 2
In this large deection consideration then, the bar
elongation e is a second order function of the nodal
displacement.
3.2. Equilibrium
Unlike small deection theory, the equilibrium of
the structure is considered at the deformed congur-
ation. Vertical equilibrium at B gives:
2t t
o
sinb P; but sin b1b1tanb
D
L
o
and hence
t t
o

PL
o
2D
: 3
3.3. Constitutive relationship
The relationship between the elongation e (due to
the load alone) and the bar tension is:
e
tL
o
AE

PL
o
2D
t
o
_ _
L
o
AE
: 4
3.4. Governing PD equation
Eqs. (2)(4) can be combined together to form a
cubic relationship between the load and the displace-
ment D.
EA
L
o
D
3
2t
o
L
o
D PL
2
o
0: 5
The PD relationship in Eq. (5) provides even more in-
formation on the behavior of the cablenets than noted
above. If t
o
=0 then the PD curve has zero slope at
the origin, and the two-link structure thus has no re-
sistance to load; it is a mechanism. Away from the ori-
gin, PD does have a positive slope, and hence the
mechanism is stiened by displacement; it is thus an
innitesimal mechanism. When a state of self-stress is
applied, t
o
$0, the PD curve has slope at the origin
proportional to t
o
and hence the structure does indeed
have initial resistance against load; the innitesimal
mechanism is stiened by self-stress. Note though that
this initial stiness is dependent on the prestress, but
not the axial stiness AE of the bars.
4. Present formulation
We shall now derive the equivalent equations for a
general bar in two-dimensional space and then apply
these equations as an example to the two-link structure
in Fig. 1. Consider a bar 12, with end coordinates
(x
1
, y
1
) and (x
2
, y
2
), and initial length L undergoing
deformation so that it ends up in new position 1'2'
with a length of L', see Fig. 2a.
4.1. Compatibility
Adopting the short hand notation ( )
21
=( )
2
( )
1
and taking into account joint displacement, the new
bar length
L
0
x
21
dx
21

2
y
21
dy
21

2
_ _
1=2
L
2
1
A
L
2
_ _ _ _
1=2
where
A dx
2
21
2x
21
dx
21
dy
2
21
2y
21
dy
21

1L 1
1
2
A
L
2
_ _

1
8
A
L
2
_ _
2
. . .
_ _
L
dx
2
21
2L

dy
2
21
2L

x
21
dx
21
L

y
21
dy
21
L

x
2
21
dx
2
21
2L
3

y
2
21
dy
2
21
2L
3

x
21
y
21
dx
21
dy
21
L
3
:
Hence
e
dx
2
21
2L

dy
2
21
2L

x
21
dx
21
L

y
21
dy
21
L

x
2
21
dx
2
21
2L
3

y
2
21
dy
2
21
2L
3

x
21
y
21
dx
21
dy
21
L
3
: 6
We thus have an expression for the bar elongation as a
Fig. 2. (a) Bar 12 of length L in two-dimensional space is
displaced to new position 1'2'; the displaced bar has length
L'. (b) Equilibrium of the bar 1'2' in displaced position.
A. Kwan / Computers and Structures 67 (1998) 243252 246
second order function of the nodal displacements, hav-
ing ignored all higher order terms.
4.2. Equilibrium
We consider equilibrium in the displaced congur-
ation, taking into account also the change in the bar
length up to rst order displacements. Firstly, ex-
pressions for sin y and cos y are:
siny
x
21
dx
21
L
0

x
21
dx
21
L
x21dx21y21dy21
L

x
21
dx
21
L 1
x21dx21y21dy21
L
2
_ _
1
x
21
dx
21
L
_ _
1
x
21
dx
21
y
21
dy
21
L
2
. . .
_ _
1
y
21
L

y
21
x
21
dx
21
L
3

y
2
21
dy
21
L
3

dy
21
L
and similarly,
cos y1
x
21
L

x
2
21
dx
21
L
3

x
21
y
21
dy
21
L
3

dx
21
L
:
The equilibrium equations at the two nodes are as
follows.
P
x1
t t
o
cos y
P
y1
t t
o
siny
P
x2
t t
o
cos y
P
y2
t t
o
sin y: 7
4.3. Constitutive relationship
The relationship between the bar elongation and the
bar tension, being entirely independent of assumptions
made in equilibrium and compatibility equations, is
still e = tL/AE.
4.4. Two-link structure
Having derived the governing equations for a gen-
eral two-dimensional bar, we can now apply the
equations to the two-link structure in Fig. 1, and
expect the same numerical results as those computed
by Eq. (5). We shall have P = 311.38 N, AE = 564.92
kN, L
o
=5080 mm and t
o
=4448.2 N because these
values represent metric equivalents to imperial values
used by Levy and Spillers [8] analysing the same pro-
blem. In the two-link structure, the two equilibrium
equations at joint B are:
t
AB
t
BC
0; and t
AB
4448:22
dy
b
5080
_ _
t
BC
4448:22
dy
B
5080
_ _
311:38: 8
The two compatibility equations can be combined with
the constitutive relationships, to eliminate bar
elongations as variables, and the resulting equations
are:
dy
2
B
10160
dx
B
t
AB
5080
564:92 10
3
; and
dy
2
B
10160
dx
B
t
BC
5080
564:92 10
3
: 9
Eqs. (8) and (9), representing the PD relationships
for the two-link structure, are four equations (two
of which are nonlinear) in four unknowns X = (dx
B
,
dy
B
, t
AB
, t
BC
). The nonlinear equations necessitate
some form of iterative solution and a standard Gauss
Newton algorithm (see e.g. Dennis [4]) has proved to
work very well. The algorithm requires the equations
(linear and nonlinear) to be written as a function F(X)
that returns the residuals for a trial vector X. The
Jacobian of F can also be explicitly provided, and in
our case this is not dicult, but the algorithm will
work without one by estimating the Jacobian using
nite dierence. As it turned out, the nonlinearity we
have in our formulation is very well behaved, and the
GaussNewton algorithm converges without diculty,
even though the initial starting point for X is always
given as the null-vector.
The GaussNewton algorithm gives the solution to
the current problem as X = (0, 166.449, 303.246,
303.246), which is identical to the solution
(D = 166.45 mm) obtained from Eq. (5). Since both
answers are derived from the same theoretical
approach, this nding is rstly not very surprising, and
secondly provides comforting conrmation that the
GaussNewton algorithm is well-behaved. These two
answers also compare very well with that provided by
Levy [8], dy
B
=166.54 mm.
The formulation for a general three-dimensional
bar, being a straightforward extension of the two-
dimensional bar, is not shown but has been used in the
illustrative examples in Section 6.
5. Solution techniques
Clearly dierent techniques can be employed to
solve the family of non-linear equations typied by
Eqs. (8) and (9). Since the form of the nonlinear
equations is known, and the highest order of non-
linearity is two, it is possible to develop an algorithm
that specically exploits these features, but such a
A. Kwan / Computers and Structures 67 (1998) 243252 247
development is not treated in this paper. The present
author has used GaussNewton in its most elementary
form because it has been found to be both ecient
and well-behaved, even with complicated examples.
Three simple enhancements though could be supplied
to the GaussNewton approach if they were deemed
necessary by the analyst.
Firstly, the initial vector for trial solution is given as
the null-vector, whereas a more reasonable initial esti-
mate for the displacements would be a multiple of the
mechanism vector. A basis for the mechanism vector
can be found from the nullspace of the small-
deformation compatibility matrix (see Pellegrino [13]
for more details). Secondly, an explicit Jacobian to the
governing equations could be supplied with little extra
dierential algebraic eort. This would eliminate the
need to estimate the Jacobian by repeated calculations
using nite dierence approximations at every iter-
ation. Thirdly, if convergence proved to be elusive
because of a large amount of nonlinearity for a par-
ticular problem, incremental loading could be used
where the displaced parameters after each load incre-
ment would be used as initial values for the next
Fig. 3. 3 3 at net in three-dimensional space is loaded by
three equal vertical loads of 150 N.
Fig. 4. Results from the present technique compared to results
given by Lewis [10] (in italics, where available). (a) Bar ten-
sion in N. (b) Nodal displacements in the x-, y- and z-direc-
tions.
Fig. 5. Saddle net with underlying grid showing joint numbers; z-coordinates are given in Table 1.
A. Kwan / Computers and Structures 67 (1998) 243252 248
increment. None of these enhancements, however, have
been found to be necessary in the examples in the
following section, although they might be useful for
more complicated problems.
6. Examples
We shall now examine three three-dimensional
examples using the governing equations for the three-
dimensional bar, whose derivations follow exactly the
pattern described above for the two-dimensional bar.
The examples have been chosen to compare results
from the present technique with those quoted in the
literature.
6.1. Flat net
The at net example consists of a cablenet lying on
a 3 3 square grid with cell side lengths of 400 mm,
see Fig. 3. The cables have AE = 97.97 kN, and pre-
tension of t
o
=200 N in all cables. The cables are
rmly anchored at the perimeter leaving four free
joints. There are thus 4 3 = 12 equilibrium
equations and 12 compatibility-constitutive equations,
in 24 unknowns (12 joint displacements and 12 bar
tensions). Starting with a null vector, the Gauss
Newton algorithm returns nodal displacements and
bar tensions shown in Fig. 4, which are exactly the
results obtained by Lewis [10] using both the tech-
niques of dynamic relaxation and stiness matrix.
Table 1
Nodal z-coordinates and nodal displacements for selected joints in the saddle net
z Present results Results from Lewis (1987)
Coor dx dy dz dx dy dz
Node (mm) (mm) (mm) (mm) (mm) (mm) (mm)
1 3632 0 0 0
2 2568 0 0 0
3 1808 0 0 0
4 1352 0 0 0
5 1200 0 0 0
10 5000 0 0 0
11 3968 15.55 4.46 81.70 15.5 4.5 82.0
12 3165 11.50 5.55 61.22
13 2592 7.38 4.20 33.31
14 2248 5.34 3.11 17.88
15 2133 4.11 2.80 11.16 4.1 2.9 11.2
21 5000 +0 0 0
22 4208 14.43 3.53 97.14
23 3592 11.27 4.47 72.90
24 3152 7.25 2.97 31.98
25 2882 5.67 2.12 10.54
26 2800 4.77 0.60 11.34
32 5000 0 0 0
33 4352 11.71 1.71 92.44
34 3848 9.55 2.11 66.94
35 3488 6.30 1.15 20.21 6.2 1.2 19.8
36 3272 4.92 0.23 14.05
37 3200 4.65 0.52 35.79
43 5000 0 0 0
44 4400 10.63 0 88.73 10.6 0 88.7
45 3933 8.80 0 62.83
46 3600 5.83 0 13.99
47 3400 4.64 0 22.52
48 3333 4.55 0 45.89 4.5 0 46.7
52 4400 0.92 0 5.86 0.9 0 6.0
72 3152 3.85 0.78 30.12 3.9 0.7 30.2
81 2133 4.11 2.80 11.16 4.0 2.9 11.0
85 3968 5.40 1.87 32.17 5.5 1.9 32.6
A. Kwan / Computers and Structures 67 (1998) 243252 249
6.2. Saddle net
Lewis [10] compared the stiness matrix and
dynamic relaxation techniques on ve examples of
cable nets of increasing complexity. The most complex
example quoted was the saddle net shown now in
Fig. 5. Although both techniques converged on similar
solutions (Lewis reported a 0.5% discrepancy) for the
rst four examples, Lewis found the stiness matrix
approach unable to converge on a solution for the sad-
dle net; ill-conditioning of the p = Kd system (where
K = K
E
+K
G
, K
E
is the elastic stiness matrix, and K
G
is the geometric stiness matrix) was highlighted as the
problem for such a ``large system''. The dynamic relax-
ation however, converged onto a solution and the pre-
sent formulation is thus compared to that technique.
The saddle net example is shown in Fig. 5 with an
underlying square grid with superimposed joint num-
bers where convenient. The individual square cells
have side lengths of 5.0 m. The net has mirror sym-
metry about both centerlines and the z-coordinates for
a quarter of the net is given in Table 1. All cable seg-
ments have AE = 44.982 MN, and pretension of
t
o
=60 kN. Some joints, forming a quarter of the net
in total, are loaded by 1.0 kN in the x- and z-direc-
tions. They are joints 11 415, 22 426, 33 437,
44 448, 55 459, 66 470 and 77 481. All perimeter
joints are anchored, giving thus a total of 63 free
joints. These, together with 142 cable segments, pro-
vide 331 nonlinear equations in 331 unknowns. The
GaussNewton algorithm, using the 331-dimensional
null vector as a starting point converges on a sol-
ution very similar to that produced by Lewis. Nodal
displacements are compared in Table 1 and cable
tensions in Table 2, for specic values given by
Lewis.
6.3. Hyperbolic paraboloid net
A hyperbolic paraboloid net, shown in Fig. 6, has
been numerically analysed by several researchers as
well as experimentally tested by Lewis [9]. The struc-
Table 2
Cable tension after load on selected cables in the saddle net
Present results Lewis' results
Cable Tension (kN) Tension (kN)
1112 53.70 54.14
2324 57.80 58.12
4748 62.51 62.78
6061 59.90 60.18
7273 54.83 55.16
8586 50.23 50.60
111 75.43 75.52
2435 69.30 69.56
2839 48.00 48.28
6273 57.37 57.61
6778 78.69 78.89
8595 63.55 63.66
Fig. 6. Hyperbolic paraboloid net with dimensions in mm. All edge nodes are supported by pin foundations and 10 internal nodes
are applied equal vertical loads.
A. Kwan / Computers and Structures 67 (1998) 243252 250
ture consists of 31 cable segments (AE = 100.72 kN)
loaded at some of the joints by a load of 157 N. The
cables were pretensioned to carry 200 N prior to the
application of load. A table of vertical deections
obtained by dierent researchers is shown in Table 3.
The second column consists of experimentally
measured deections. Lewis [10] and Suan et al. [15]
analysed this problem by the stiness matrix method,
and a minimum energy method, respectively.
Lewis [9, 10] performed a dynamic relaxation calcu-
lation in Fortran on a Prime 750 mainframe computer
and her results are shown in the fth column. The pre-
sent author has also computed the problem with
dynamic relaxation on a desktop microcomputer. The
results of this computation, shown in the sixth column
of Table 3, are very similar to those provided by
Lewis. The nal column of Table 3 shows results from
the proposed technique. It can be seen from Table 3
Fig. 7. Displacement of joint 22 of the hyperbolic paraboloid shown in Fig. 6 during three dynamic relaxation computations. As
the ctitious damping value is increased, the nal value of displacement is obtained with fewer oscillations.
Table 3
Comparison of displacements (in mm) and computation time (in CPU seconds)
Stiness Minimum Dynamic Dynamic Proposed
Experiment matrix energy relaxation relaxation technique
Joint (Lewis) (Krishna) (Suan) (Lewis) (present) (present)
5 19.5 19.6 19.3 19.3 19.38 19.52
6 25.3 25.9 25.5 25.3 25.62 25.35
7 22.8 23.7 23.1 23.0 22.95 23.31
10 25.4 25.3 25.8 25.9 25.57 25.86
11 33.6 33.0 34.0 33.8 33.79 34.05
12 28.8 28.2 29.4 29.4 29.32 29.49
15 25.2 25.8 25.7 26.4 25.43 25.79
16 30.6 31.3 31.2 31.7 31.11 31.31
17 21.0 21.4 21.1 21.9 21.28 21.42
20 21.0 22.0 21.1 21.9 21.16 21.48
21 19.8 21.1 19.9 20.5 19.79 20.00
22 14.2 15.7 14.3 14.8 14.29 14.40
CPU 150 120
A. Kwan / Computers and Structures 67 (1998) 243252 251
that there is close agreement in joint displacements
between all the dierent methods, and the presently
proposed technique thus compares well with all the
established techniques.
The purpose of reproducing the dynamic relaxation
calculation is to have a comparison of the computation
time required by the dynamic relaxation technique,
and that by the present technique. Computation time
in dynamic relaxation is highly dependent on three
parameters: the time increment, the ctitious nodal
masses and the damping coecients at the nodes.
When Lewis' [9] original parameters were used, the
computation time on the present author's desktop
computer was exceedingly large (>70,000 cpu s).
Alternative parameters were thus investigated and
Fig. 7 shows that a stable solution could be attained
after about 150 cpu s. This value, rather surprisingly,
is slightly larger than that obtained for the presently
proposed technique, computing the same problem
without any of the enhancements discussed in Section
5, on the same desktop computer. It would seem there-
fore, that the present technique, although developed in-
itially for its simplicity in the treatment of geometric
nonlinearity, is also comparable in speed to the more
established technique of dynamic relaxation.
7. Conclusions
A new approach to the statical behavior of geome-
trically nonlinear cable structures has been presented.
The key features of this approach are the separation of
the numerical computation from the underlying mech-
anics of the problem and understanding of only the
most elementary of structural mechanics is required.
This technique is thus eminently suitable for presen-
tation at undergraduate courses on cable structures.
On the other hand, in spite of its lack of sophisti-
cation, the proposed technique has been proved to be
comparable to more established techniques in its accu-
racy of solution, in its speed of calculation and in its
stability of solution, even when optional enhancements
are not used.
References
[1] Barnes MR. Form-nding and analysis of pre-
stressed nets and membranes. Computers & Structures
1988;30(3):68595.
[2] Calladine CR. Buckminster Fuller's `Tensegrity' struc-
tures and Clerk Maxwell's rules for the construction of
sti frames. International Journal of Solids and
Structures 1978;14:16172.
[3] Coyette JP, Guisset P. Cable network analysis by a non-
linear programming technique. Engineering Structures
1988;10:416.
[4] Dennis JE Jr. Nonlinear least squares. In: Jacobs D,
editor. State of the Art in Numerical Analysis. New
York: Academic Press, 1977;269312.
[5] Gambhir ML, de Batchelor BV. A nite element for
3-D prestressed cablenets. International Journal of
Numerical Methods in Engineering 1977;11:1699718.
[6] Krishna P. Cable-suspended roofs. New York: McGraw-
Hill, 1978.
[7] Kwan ASK, You Z, Pellegrino S. Active and passive
cable elements in deployable/retractable masts.
International Journal of Space Structures 1993;8(1/2):29
40.
[8] Levy R, Spillers WR. Analysis of geometrically nonlinear
structures. London: Chapman & Hall, 1995.
[9] Lewis WJ, Jones MS, Rushton KR. Dynamic relaxation
analysis of the non-linear static response of pretensioned
cable roof. Computers & Structures 1984;18(6):98997.
[10] Lewis WJ, A comparative study of numerical methods
for the solution of pretensioned cable networks. In:
Topping BHV, editor. Proceedings of the International
Conference on Design and Construction of Non-
Conventional Structures. Edinburgh: Civil-Comp Press,
vol. 2, 1987:2733.
[11] Mitsugi J. Static analysis of cable networks and their
supporting structures. Computers & Structures
1994;51(1):4756.
[12] Pellegrino S. A class of tensegrity domes. International
Journal of Space Structures 1992;7(2):12742.
[13] Pellegrino S. Structural computations with the singular
value decomposition of the equilibrium matrix.
International Journal of Solids and Structures
1993;30(21):302535.
[14] Stefanou GD, Moossavi E, Bishop S, Koliopoulos P.
Conjugate gradients method for calculating the response
of large cable nets to static loads. Computers &
Structures 1993;49(5):8438.
[15] Suan FMA, Templeman AB. On the non-linear analysis
of pretensioned cable net structures. Structural
Engineering 1992;4(2):14758.
[16] Vilnay O, Rogers P. Statical and dynamical response of
cable nets. International Journal of Solids and Structures
1990;26(3):299312.
A. Kwan / Computers and Structures 67 (1998) 243252 252

Você também pode gostar