Structural Eng./Earthquake Eng., JSCE, Vol.21, No.

1, 27s-41s, 2004 January

（J. Struct. Mech. Earthquake Eng., JSCE, No.752/I-66)

A SHAPE-FINDING ANALYSIS OF
SUSPENDED STRUCTURES
ON THE DISPLACEMENT-METHOD EQUILIBRIUM

Masahiro AI 1 and Hironori IMAI 2

1 Mem. of JSCE, Dr.Eng., Professor, Dept. of Civil Eng., Hosei Univ.
(Koganei Tokyo 184-8584, Japan)
2 M.Eng., Teacher, Dept. of Civil Eng., Iidaosahime High School
(Kanaenagokuma Iida Nagano 395-0804, Japan)

By its geometrical nonlinearity, the realization of a desired suspension structure depends on the
shape-finding procedure. In an actual suspension structure, cable members are usually combined
with bending members. For a beam-cable structure, in this study, the displacement method is
employed to deal with the structural equilibrium, but in which each cable member is treated as
the elastic catenary through a force-method computation. Thereafter an iterative scheme of shape
finding is developed for such a mixture of cable and beam members, which is based on the tangent
coefficients to change of the cable natural lengths on each updated equilibrium configuration.

Key Words : shape finding, displacement method, elastic catenary, iteration

1. INTRODUCTION in the shape finding. In Ref. 6), after the same

The suspension bridge has a long history since
before the computational age. In the cable-stayed
bridge, from its earlier stage, the statically inde-
terminate forces have been estimated with the aid
of computers. In the recent discrete methods, the
static behaviors of such a beam-cable structure
can be known with a sufficient accuracy. On the
other hand, beyond those established ones, there
could be other suspension forms remaining to be
developed according to circumstances.
In principle, the configuration of a cable struc-
ture is known as a result of equilibrium analysis.
So a general suspension structure necessitates the
“shape-finding” procedure in its erection: to ful-
fill the design requirements, the equilibrium state
needs to be settled in variation of structural pa-
rameters such as the cable lengths. There have
been various approaches to that problem. 1) — 9)
The force density method in Ref. 4) is based on
one notable feature of a general network: once
the force densities (tension/length ratios) are as-
sumed in members, the discrete equilibrium equa-
tions become linear for the joint coordinates.
Through the linear solution, those force densi-
ties have a direct correspondence to the equilib-
rium shape, and so could be favorable parameters
parameters are employed, an arbitrary number
of requirements such as more than the variable
parameters are dealt with upon a suggested ob-
jective function through the optimization proce-
dure. Those existing methods are developed for
an assembly of straight cable and axial members.
But, in a usual suspended structure, the cables
are combined with column and beam members.
Practically, a pretensioned cable is represented
by the simple tension member. But, a long self-
weighted cable such as in a large-scale struc-
ture needs to be divided into several segments.
For the accuracy, the sag-embodied elements are
available: a shallow cable is described in the
parabolic profile e.g. 10) ; and the catenary theory
can deal with a deeply hanging cable in one
segment. e.g. 11), 12) In either of them, the expres-
sion for tension components is implicit in terms
of its span coordinates. This compatibility con-
dition in each element has to be treated numeri-
cally, prior to the structural computations.
Because of its capability to various discrete
models, the displacement method is employed for
the present equilibrium computation of a beam-
cable structure, in which both the simple tension
element and the elastic catenary element are pre-
pared for each cable member. This study is to de-

27s
 Fig. 1 Catenary cable
 q 
velop a shape finding scheme for such a mixture  2 +T2 +T
T0x 
T0x 0y 0y
of cable and beam members, based on the tan- + log q
w  2 + (T − ws)2 + (T − ws) 
gent coefficients of equilibrium quantities to the T0x 0y 0y
variable natural lengths of cable and axial mem- (2a)
bers. The shape settlement is executed in com- w 2
T0y s − s
bination with the equilibrium computation: the y(T 0 ; s) = 2
responsive equilibrium state is determined after EA
½ q ¾
each correction of member lengths; and the next 1 q 2 2 2 2
+ T0x + T0y − T0x + (T0y − ws)
correction is estimated tangentially upon that up- w
dated configuration. The two structural itera- (2b)
tions are layered. In each of them, the fractional
correction technique 7), 8) is adopted so that in- in which { T0x , T0y } act as parameters in the
the-midst states are not thrown far away from an spatial curve. This expression is the so-called
expected correction path. “elastic catenary,” which has been presented by
Rough. 13)
Through relation { Tl x , Tl y } = { T0x , T0y −
2. ELASTIC CATENARY ELEMENT wl } , either T 0 or T l can be taken as the inde-
pendent parameters. By differentiating the above
In the 2-D space, consider a uniform flexible (2) with respect to T l (δT l = δT 0 = δT (s)) , we
cable of natural length l , extension rigidity EA have the tangent flexibility matrix between xl =
and weight w per unit length. Firstly, as shown x( T l ; l ) and T l :
in Fig. 1, let this cable be anchored at one end
to the origin of rectangular coordinates { x , y } . ½ ¾ ½ ¾
δxl δTl x
After material coordinate s is taken along its nat- = [ Q( T l ) ] (3)
δy l δTl y
ural length, the configuration is here described in
the form { x(s) , y(s) } . l
Qxx =
When tensioned by { T0x , T0y } at s = 0, the EA
( Ã ! )
internal components under w into y-direction are 1 T0 + T0y Tl y T0y
{ Tx (s),qTy (s) } = { T0x , T0y − ws } . Submitted to + log + − (4a)
w Tl + Tl y Tl T0
T (s) = Tx (s)2 + Ty (s)2 , differential element ds µ ¶
T0x 1 1
is elongated to ds̄ = ( 1 + T (s)/(EA) ) ds ; and, Qxy = Qyx = − (4b)
w T0 Tl
by the flexibility, this ds̄ is laid in the same direc- µ ¶
tion to { Tx (s), Ty (s) } . By those two, the spatial l 1 T0y Tl y
Qyy = + − (4c)
components of ds is eventually written in terms EA w T0 Tl
of { Tx (s), Ty (s) } : q q
where T0 = T0x 2 + T 2 and T = Tl 2x + Tl 2y .
0y l
½ ¾ µ ¶½ ¾
dx 1 1 Tx (s) So far as { Tx (s), Ty (s) } 6= { 0, 0 } in 0 ≤ s ≤ l, ma-
= + ds (1)
dy T (s) EA Ty (s) trix [ Q(T l ) ] is kept positive definite. 7) If Tl x = 0
and −wl ≤ Tl y ≤ 0 (a line segment in { Tl x , Tl y }-
By the actual integration, we have space), the cable has a point of { Tx (a), Ty (a) } =
{ 0, 0 } , and is bent back in a vertical hanging
T0x s (see Fig. 1). In this configuration, derivative ma-
x(T 0 ; s) = trix [ Q(T l ) ] becomes singular into the direction
EA

■
2
28s
normal to the line segment, ∂xl / ∂Tl x → ∞ . But,
function xl (T l ) itself is continuous over the en-
tire domain. By the symmetry of [ Q(T l ) ] , dif-
ferential δF = xl (T l )δTl x + yl (T l )δTl y is path-
independently integrable, and, by its positive def-
initeness, the resulting potential F (T l ) is properly
convex :
½ µ ¶
1 2 T0 + T0y
F (T l ) = T0x log + T0y T0
2w Tl + Tl y Fig. 2 Isolated cable for increment δl
¾ ½ ¾
l 2 1 V(c) ( T l , yi ) . When joint i is fixed in the
−Tl y Tl + T0x + ( T 3 − Tl 3y ) (5)
2EA 3wl 0y space (see Fig. 1), by the positive definiteness
Consider joint i and j to which cable element of [ k̄( T l ) ](c) (kept from [ Q( T l ) ] ), the sum
(c) is attached, with their spatial positions be- U(c) ( T l ) +V(c) ( T l ) is convex with respect to T l .
ing denoted by xi and xj . In the displacement- In Sec. 5, the equilibrium response to varia-
method computation, those joint positions are tion of the cable lengths will be argued to de-
iteratively approximated toward the structural velop a shape-finding procedure. With regard to
equilibrium. On such momentary joint positions, this structural response, an essential behavior is
all the catenary elements have to be computed for seen in an isolated cable. As shown in Fig. 2, let
{ Tl x , Tl y }(c) corresponding to their span coordi- an infinitesimal increment δl of natural length be
nates, x̄ = xj −xi . In this compatibility problem assumed in the catenary equilibrium, with span
of each element (c), a total complementary poten- coordinates { x̄, ȳ } being fixed. If the tension
tial is defined by W ∗ (T l ) = F (T l ) − x̄T T l , and components are unchanged, the cable is simply
the unique solution can be found numerically as lengthened at its fore end: according to Eq.(1)
½ ¾ µ ¶½ ¾
a point to minimize W ∗ (T l ) . 12) δxl 1 1 Tl x
After the above force-method computation, the = + δl (8)
δy l |Tl | EA Tl y
tension components to span x̄ are rewritten as a
displacement-method element into In the cable spanned to { x̄, ȳ } , those {δxl , δy l }
are offset by a simultaneous change of T (s) . In
{ T }(c) (= { T i , T j } ) addition to { δTl x , δTl y } , the self weight is in-
= {(−T0x , −T0y ), ( Tl x , Tl y )} (6) creased by wδl :
½ ¾ ½ ¾ ½ ¾
Matrix [ Q( T l ) ] is inverted into tangent stiffness δTx (s) δTl x 0
= + δl : const.
[ k̄( T l ) ](c) , and which is rewritten into the full δTy (s) δTl y w
4 × 4 matrix [ k( T l ) ](c) between { δT i , δT j } and (9)
{ δxi , δxj } . According to Ref. 14), the strain en- Applying this { δTx (s), δTy (s) } to the flexibility
ergy U(c) and the self-weight potential V(c) are matrix, we have the compatibility equations
µ½ ¾ ½ ¾ ¶
obtained as follows : δTl x 0
à Z ! [ Q( T l ) ] + δl
T2 l δTl y w
U(c) ( T l ) = ds µ ¶½ ¾ ½ ¾
0 2EA
1 1 Tl x 0
½
+ + δl = (10)
1 1 ³ 3 ´¾ |Tl | EA Tl y 0
2 3
= T0x l + T − Tl y (7a)
2EA 3w 0y The responsive { δT }(c) is eventually written as
à Z ! follows :
l
½ ¾
V(c) ( T l , yi ) = − wy(s) ds ψ0
−ψ
0 δ{ T }(c) = { ψ }(c) δl(c) = δl(c)
µ ¶ ψl
1 (11)
= −wl yi − T0 l − Tl Tl y − T0 T0y µ ¶
2w 1 1
à ! ψ0 = − + [ Q( T l ) ]−1
T2 T0 + T0y |Tl | EA
+ 0x log ½ ¾
2w Tl + Tl y Tl x
· (12a)
à ! Tl y
w T0y l 2 wl 3 µ ¶
1 1
−
EA 2
−
6
(7b) ψl = − + [ Q( T l ) ]−1
|Tl | EA
½ ¾ ½ ¾
The work done by { T i , T j } during change Tl x 0
· − (12b)
of { xi , xj } is stored into those U(c) ( T l ) and Tl y w

■
3
29s

Fig. 3 Beam-cable structure
3. CONTINUITY RELATIONS
IN STRUCTURAL STIFFNESS
Consider a two-dimensional assembly of cable
and beam members, as shown in Fig. 3. The free-
dom of a beam joint is represented by its transla-
tional and angular coordinates in the space, X n =
{ x, y, θ }n , with the force components being de-
noted by F n = { Fx , Fy , M }n . At an intersection
of two or more cables, a joint of another class is
placed, which has joint position xi = { x, y }i and
joint force F i = { Fx , Fy }i . Those joint positions
are collected into one column vector, { X J } =
{ X 1 , · · · , X N , x1 , · · · , xI } ( J = N + I ). Ex-
cepting the constrained ones at the supports, we
have the variable components of { X J } listed into
another { dΓ } = { d1 , d2 , · · · , dΓ } . The corre-
sponding joint forces are denoted by { F J } and
{ FΓ } , respectively.
The spatial configuration of each element is
defined by its two joint positions: {X}[ e ] =
{ X m , X n } for a beam element, and {x}(c) =
{ xi , xj } for a cable element. The force com-
ponents into {X}[ e ] and {x}(c) are denoted by
{F }[ e ] and {T }(c) , respectively. Those element
positions are collected into their respective sets:
{X [ E ] } = {{X}[ 1 ] , · · · , {X}[ E ] } and {x(C) } =
{{x}(1) , · · · , {x}(C) } . Similarly, {F [ E ] } and
{T (C) } stand for the element forces. Column vec-
tor {X [ E ] } and {x(C) } are related to {X J } in the
form†
{X [ E ] } = [ SX ]{ X J } , 0
{x(C) } = [ SX ]{ X J }
(13)
0 ] are the continuity matrices,
where [ SX ] and [ SX
consisting of 0 and 1.
As for the 2-D beam element, there have
been presented certain nonlinear discretiza-
tions. The stiffness relations developed in the
method of separation into rigid displacement
and deformation 15) are employed in our elastic
† £
A beam element is attached to its two joints with cer-
tain fixed angles. In case the tangent angles at the both
ends of its axial line are employed in {X}[ e ] rather than
the joints’ angular coordinates, no changes in other rela-
tions but {X [ E ] } = [ SX ]{ X J } + { const. } .
■
4
30s
finite-displacement analysis: the element force
{F }[ e ] (= { F m , F n }) , the tangent stiffness ma-
trix [ K ][ e ] and the strain energy U[ e ] are eval-
uated for an assumed {X}[ e ] . In each cable
element, positioned at {x}(c) , the cable tension
{T }(c) (= { T i , T j }), the tangent stiffness [ k ](c) ,
the strain energy U(c) and the self-weight poten-
tial V(c) are obtained through the force-method
computation, Sec. 2.
In the structure, those element forces are accu-
mulated into the joint forces, namely
{F J } = [ SF ]{F [ E ] } + [ SF0 ]{ T (C) } (14)
Those [ SF ] and [ SF0 ] are determined by the
element-to-joint continuity, and so are not inde-
pendent of the former [ SX ] and [ SX0 ] : by the
contragredience
[ SF ] = [ SX ] T and [ SF0 ] = [ SX
0 T
] (15)
Let the external forces into the { dΓ }-directions
be denoted by { P̄Γ } . By extracting the corre-
sponding components out of { F J } , we have the
equilibrium equations for { dΓ } :
( { FΓ } = ) [ C F ] { F J } = { P̄Γ } (16)
where [ C F ] is a matrix of Γ × (3N + 2I), con-
sisting of 0 and 1.
We next consider the tangent stiffness for dif-
ferential δ{ dΓ } . The variation of {X J } is
written as δ{X J } = [ C F ] T δ{ dΓ } . The dif-
ferential relations of the former (13), (14) and
(16) are apparent. Let the element tangent
stiffness matrices be collected into δ{F [ E ] } =
[ K[ E ] ]δ{X [ E ] } and δ{F (C) } = [ k(C) ]δ{x(C) } ,
where
 
h i
[ K ][ 1 ]
 .. 
K[ E ] =  .  (17a)
[ K ][ E ]
 
h i
[ k ](1)
 .. 
k(C) =  .  (17b)
[ k ](C)
Then, by the chain rule, the tangent stiffness ma-
trix on freedom { dΓ } is written as follows :
δ { FΓ } = [ KΓ ({dΓ }) ] δ { dΓ } (18a)
[ KΓ ] = [ CF ] [ SF ] [ K[ E ] ] [ SX ] [ CF ] T
¤ £ ¤
+ [ CF ] SF0 [ k(C) ] SX
0
[ CF ] T (18b)
The total potential energy is given by the sum
of the elements’ strain energies, the self-weight
potentials in cables, and the potentials of external
joint forces :
X X³
W ({dΓ }) = U[ e ] + U(c) + V(c)
[e] (c)
− { P̄Γ } T { dΓ }
4. ITERATION FOR EQUILIBRIUM
The characteristics of cable members are, of
course, reflected onto the equilibrium behavior of
their suspended structure. First, let a mere as-
sembly of the simple tension members be summa-
rized. The stiffness relations as a discrete element
are given in Appendix—A, together with the
compressive axial member. Since each tension
member has a clear domain of singularity, ¯l (=
p
x̄ + ȳ 2 ) ≤ l , the assembly also has certain
2 {dΓ }0 has an actual distance to {dΓ }E , the ap-
singular domains of {dΓ } , in general. But, in
the displacement method, such singular elements
do not necessarily mean the structural singular-
ity: so far as every free joint is chained through
tensioned members to one or more supporting
joints (anchored), irrespective of other slackened
members, the tangent stiffness matrix between
{FΓ } and {dΓ } is kept positive definite. 8) Only if
some free joints are left not anchored, the struc-
tural tangent stiffness becomes singular. It is un-
der those circumstances that the assembly has a
semi-convex total potential with respect to {dΓ } .
The above sayings are true of the catenary ca-
bles also. However, as stated on its flexibility
in Sec. 2, the singularity exhibited by a cate-
nary element is much weak: only into the normal
direction on a line segment of { x̄ , ȳ } , namely
∂Tl x / ∂ x̄ → 0 for x̄ = 0 and −l (1 + wl/2EA) ≤
ȳ ≤ l (1 + wl/2EA) . In such a case, its potential
U(c) + V(c) still remains properly convex with re-
spect to { x̄ , ȳ } , and so is the total potential of
their assembly, with respect to {dΓ } .
In our structure consisting of beam and cable
members, on the other hand, the total potential
energy is not convex at an arbitrary {dΓ } : the
structure could become unstable beyond its buck-
lings. The stability analysis on certain loading
paths is essential to estimate the load-carrying ca-
pacity. But, in a shape-finding analysis, the prin-
cipal loads are kept unchanged except for the ca-
bles’ self-weights. The equilibrium configuration
is settled through successive changes of the cable
lengths. After each change of those lengths, the
displacement-method computation is executed to
determine the corresponding equilibrium state.
Usually, desired in the shape finding is a sta-
ble equilibrium configuration. If equilibria in the
midst of the settlement are allowed to go across
the stability boundaries, the handling must be-
come much complicated. The execution of our
´ shape finding is confined to the extent where the
tangent stiffness [ KΓ ({dΓ }) ] is kept positive def-
inite.
(19) Suppose the cable lengths are changed upon
a known equilibrium {dΓ }0 . The equilibrium is
shifted to another position, say {dΓ }E . From
the unbalanced joint forces, {FΓ }0 (6= {P̄Γ }), es-
timated at {dΓ }0 , the equilibrium computation is
started to find {dΓ }E . With those renewed cable
lengths, it has been postulated that our struc-
ture has {dΓ }0 and {dΓ }E in one stable subdo-
main. The Newton-Raphson method seems suit-
able to such a local one-to-one between {FΓ } and
{dΓ }. Indeed, provided they are located close to-
gether, {dΓ }E is attained without fail. But, when
proximated {dΓ }’s in that method can be thrown
away beyond the one-to-one boundary.
For a consistent convergence to {dΓ }E , we here
employ the fractional correction method. 7), 12)
Consider the h i i-th cycle where error vector
{∆FΓ }h i i (= { FΓ }h i i − { P̄Γ }) , tangent stiffness
[ KΓ ]h i i and total potential Wh i i are obtained for
{dΓ }h i i . By the Taylor’s expansion, the total po-
tential is written for increment ∆{dΓ } as follows :
W ({dΓ }h i i +∆{dΓ }) = Wh i i + {∆FΓ }Th i i ∆{dΓ }
1
+ ∆{dΓ }T [ KΓ ]h i i ∆{dΓ } + 0(∆{dΓ } 3 ) (20)
2
If the whole {∆FΓ }h i i is put on the tangential
correction, namely ∆{dΓ } = −[ KΓ ]−1 h i i {∆FΓ }h i i ,
we have
³ ´
1
∆W = − Rh2i i + 0 {∆FΓ }h3i i (21a)
2
r
Rh i i = {∆FΓ }Th i i [ KΓ ]−1
h i i {∆FΓ }h i i (21b)
This 1/2 · Rh2i i is representing the first approxi-
mation of potential rise at {dΓ }h i i from the (rel-
ative) minimum at {dΓ }E . But, unless {∆FΓ }h i i
is small enough, that decrease can be overcome
by the remaining terms.
Then, by the use of factor θh i i ( 0 < θh i i ≤ 1 ),
let a fraction of error {∆FΓ }h i i be submitted to
the tangent correction :
{dΓ }h i+1 i = {dΓ }h i i − θh i i [ KΓ ]−1
h i i {∆FΓ }h i i
(22)
Trivially, the resulting { FΓ }h i+1 i is related to
{ FΓ }h i i and { P̄Γ } in the form
{ FΓ }h i+1 i = ( 1 − θh i i ){ FΓ }h i i + θh i i { P̄Γ }
+ { 0 ( θh2i i ) } (23)
■
5
31s
And, by the substitution into expression (20), we
have the total potential at { dΓ }h i+1 i written as
³ θ2 ´
Wh i+1 i = Wh i i − θ− R 2 + 0 ( θh3i i )
2 hii hii
(24)
For a small θh i i , as is seen in (23), the shift from
{ FΓ }h i i is directed the more straightly to { P̄Γ } :
that is, sequence { { FΓ }h 0 i , { FΓ }h 1 i , · · · } can
be kept close, to any extent, to the line segment
from { FΓ }0 to { P̄Γ } . At the same time, since
θ − θ2 /2 > 0 for 0 < θ ≤ 1 in (24), inequality
Wh i i > Wh i+1 i is kept for such a small θh i i .
It is not until factor θh i i ’s are actually specified
in a sequence that formula (22) is accomplished.
The fractional corrections are steady for smaller
θh i i , but the rapidity of convergence comes from
larger ones. If θh i i are fixed to one fraction
such as 1/2 , theoretically, the whole correction
is not attained in a finite iteration. Since prod-
uct θh i i {∆FΓ }h i i is submitted to the tangent cor-
rection, it is rational for factor θh i i to be taken
larger with decrease of error {∆FΓ }h i i . We have
mentioned quantity Rh i i in relation to the total
potential. That Rh i i defined by (21b) is a norm
of vector {∆FΓ }h i i weighted by [ KΓ ]−1 h i i , and so
scalar θh i i Rh i i stands for vector θh i i {∆FΓ }h i i .
We now generate factor θh i i such that θh i i Rh i i is
kept constant, but to the extent θh i i ≤ 1 :
à !
θ0 Rh 0 i
θh i i = min ,1 (25)
Rh i i
in which θ0 is a quantity called “basic correction
factor,” initially assumed within 0 < θ0 ≤ 1 .
A descent in the total potential means that
{dΓ }h i+1 i is relatively improved from {dΓ }h i i .
But, those differences might become numerically
invisible on the final cycles in approaching to
{dΓ }E ; for, as is seen in expression (21), Wh i+1 i −
Wh i i is consisting of the quadratic and higher
terms of {∆FΓ }h i i . On the contrary, the scalar
Rh i i has an increasing accuracy near the solution.
As well as the descent in potential, the shifted
{dΓ }h i+1 i is required to stay in the same one-
to-one domain with {dΓ }h i i : for an insufficient
θ0 , {dΓ }h i+1 i can be thrown beyond the stabil-
ity boundary; or can be dropped into the singular
domain, in case of the simple tension members.
Thus, after each correction (22), {dΓ }h i+1 i is ex-
amined for the following two :
A) det [ KΓ ] h i+1 i > 0 (26a)
B) Wh i+1 i < Wh i i for θh i i < 1
Rh i+1 i < Rh i i for θh i i = 1 (26b)
■
6
32s
If A) or B) is not true, the last {dΓ }h i+1 i is aban-
doned. With a renewed correction factor, the iter-
ation is resumed from the previous h i i-th: for our
θh i i generated by Eq.(25), the basic θ0 is changed
into a smaller one such that θh i i is halved, namely
θh i i false Rh i i
θ0 new = · (27)
2 Rh 0 i
Requirement A) is for {dΓ }h i i ’s to be con-
fined to a stable subdomain, in which the
convex W ({dΓ }) and the norm R({dΓ }) are
bounded from below. By the other B), se-
quence { Wh 0 i , Wh 1 i , · · · , Wh L i } and { Rh L+1 i ,
Rh L+2 i , · · · } are kept monotone-decreasing.
Hence the sequence is convergent.
The convergence comes mostly from Rh i i → 0 .
But, if the sequence {dΓ }h i i is kept by the tan-
gent coefficients to cut right across the stability
boundary, factor θ0 is halved repeatedly. This
θ0 → 0 leads to a trivial convergence, {dΓ }h i+1 i −
{dΓ }h i i → { 0 } , but the limit of {dΓ }h i i is to a
boundary point of the stable subdomain, not to
{dΓ }E .
Strictly speaking, the above A) is not sufficient
to keep {dΓ }h i i ’s within one stable subdomain. If
two eigenvalues are turned negative in [ KΓ ]h i i →
[ KΓ ]h i+1 i , the sign of det [ KΓ ] does not change.
When the iteration is started with a relatively
large θ0 , it can happen that another equilibrium
than {dΓ }E is attained as the result of Rh i i → 0 .
(This convergence is excluded if the eigenvalues
are observed instead.)
Let the line segment from {FΓ }0 to {P̄Γ } be
expressed by {FΓ ( ρ )} = ( 1 − ρ ){FΓ }0 + ρ {P̄Γ }
( 0 ≤ ρ ≤ 1 ). To this {FΓ ( ρ )} , there exists a cor-
responding equilibrium curve {dΓ ( ρ )} continued
from {dΓ }0 . Provided the entire {dΓ ( ρ )} is lying
in a stable subdomain ( det [ KΓ ({dΓ ( ρ )}) ] > 0
for 0 ≤ ρ ≤ 1 ), the line {FΓ ( ρ )} is closely followed
by { { FΓ }h 0 i , { FΓ }h 1 i , · · · } with a smaller θ0 ,
and so the solution {dΓ }E is approached by our
{ { dΓ }h 0 i , { dΓ }h 1 i , · · · } .
5. SETTLEMENT OF EQUILIBRIUM
CONFIGURATION
The following shape finding is preceded by Ref.
8), in which the essential scheme is developed for
an assembly of the simple tension members.
( 1 ) Tangent coefficients
Suppose the cable natural lengths are infinites-
imally changed on an equilibrium of our sus-
pended structure. In order to derive the struc-
tural response, firstly, we assume that the free
joints are fictitiously fixed at their preceding
equilibrium positions. As for an isolated cable
spanned between two fixed points, we have ob-
tained the response of tension components to
a differential increment of its natural length,
Eqs.(11) and (12) of Sec. 2. Those expressions
for our cable elements are collected into a matrix
form : Fig. 4 Chord force

δ{ T (C) }Fixed = [ ψ (C) ] δ{ l(C) } (28a) The substitution leads to the variation of tension
components written in terms of the independent
 
ψ }(1)
{ψ δ{ l(C) } :
 .. 
[ ψ (C) ] =  .  (28b)
δT l (c) = ψ l (c) δl(c)
ψ }(C)
{ψ
+ [ Q ]−1
(c) ( [ ΦX ] j − [ ΦX ] i ) δ{ l(C) } (32)
δ{ l(C) } = { δl(1) , δl(2) , · · · , δl(C) } (28c)
For the later expansion, let this result be given a
in which subscript “Fixed” means a quantity on short notation :
the temporary joint positions.
By continuity relation (14) and (16), the above δT l (c) = [ Φ T ](c) δ{ l(C) }
δ{T }(c) ’s are superposed on the joints :
or ∙ ¸
h ΦTx i
δ{FΓ }Fixed = [ C F
] [ SF0 ] δ{T (C) }Fixed (29) = δ{ l(C) } (32)’
h ΦTy i (c)

Those δFγ ’s are reactive forces to the fictitious In a sagged cable, the chord force given by
constraints. The free joints are now released: p
by the use of the structural tangent stiffness ∗ x̄2 + ȳ 2
[ KΓ ({dΓ }) ], the incremental equilibrium equa- T = Tl x · (33)
x̄
tions for the joint displacements are written as
can be employ as a tensile member force (see
[ KΓ ] δ{dΓ } + [ C F ] [ SF0 ] [ ψ (C) ] δ{ l(C) } = { 0 } Fig. 4). Differential δT ∗ is written as
(30) ¯l Tl x ȳ ³ ´
Thus we have the structure’s own response to δT ∗ = δTl x + ¯ 2 −ȳ δx̄ + x̄ δ ȳ
x̄ l x̄
δ{ l(C) } :
By substituting the preceding results, we have
δ{X J } = [ ΦXJ ({ l(C) }) ] δ{ l(C) } (31a) δT ∗ for our element (c) :
[ ΦXJ ] = µ ¯
∗ l Tl x ȳ
−[ C F ]T [ KΓ ]−1 [ C F ] [ SF0 ] [ ψ (C) ] (31b) δT(c) = h Φ T x i(c) + ¯ 2
x̄ l x̄
³ ´¶
where δ{dΓ } in Eq.(30) has been rewritten into · h −ȳ, x̄ i [ ΦX ] j − [ ΦX ] i δ{ l(C) } (34)
the freedom of {X J } .
In addition to δ{T }(c) Fixed , the tension com- When the magnitude of tension at s = l (or =
ponents of each element (c) have been changed by 0) is chosen instead ofq the chord force, by the
the joint displacements: at its fore end (s = l ) differentiation of Tl = Tl 2x + Tl 2y , we have
½ ¾ µ³
δTl x 1 1 ´
= − + [ Q( T l ) ]−1 T lT
δTl y (c) |Tl | EA δTl (c) = [ Φ T ](c) δ{ l(C) } (35)
½ ¾ ½ ¾¶ Tl
Tl x 0
· + δl(c)
Tl y w ( 2 ) Shape-finding scheme
ý ¾ ½ ¾ ! In our shape finding, the objective values are
δx δx
+ [ Q( T l ) ]−1 − prescribed to certain equilibrium quantities se-
δy j
δy i lected from among
For {δx, δy}i and {δx, δy}j , their 2×C submatri- (a) spatial coordinates of joints,
ces are extracted from the coefficient matrix (31 (b) spatial components of cable tensions, and
b) with notation [ ΦX ] i and [ ΦX ] j , respectively. (c) magnitudes of cable tensions.

■
7
33s
So the preceding {X J }, { T l (C) } and { T(C) ∗ } (or equilibrium {dΓ }E . With no regard to the equi-
{ Tl (C) }) are appointed to the above (a), (b) librium conditions, first, the free joints listed in Y
and (c), where we have formulated their tan- are thrust to their objective coordinates, and the
gent coefficients to change of the cable lengths. member forces are shifted accordingly. Next, in
The individual subjects taken from those vec- those cable elements which are prescribed in Y ∗ ,
tors are listed into a set Y = { Y1 , Y2 , · · · , YS } , the relevant tension components or magnitudes
with their required values being denoted by Y ∗ . are replaced by the required values. Those dis-
To realize Y ∗ in the structural equilibrium, to turbed member forces are accumulated into the
the same number S , we specify the active ca- unbalanced joint forces, say {∆FΓ }S (= {FΓ }S −
ble members which have variable natural lengths. {P̄Γ }). Finally, in the same way to Eq.(21b), the
Their lengths are collected into another z = magnitude of vector {∆FΓ }S is estimated by
{ z1 , z2 , · · · , zS } . Thus our problem is reduced q
to solve Y (z) = Y ∗ for z . In Appendix—B, it RS = {∆FΓ }S T [ KΓ ]E −1 {∆FΓ }S (36)
is indicated that our shape-finding becomes more
in which [ KΓ ]E is the tangent stiffness matrix
practicable if the objective values are set on linear
at the preceding equilibrium {dΓ }E . Immedi-
combinations of the above equilirium quantities.
ately after this estimation of R S , the structure
Through the equilibrium computation, Sec. 2 is brought back to {dΓ }E .
to 4, the functional relation Y (z) is known nu- The fractional correction method is now exe-
merically. By collecting the relevant coefficients cuted from an initial z 0 ( = z h 0 i ), with a basic
from Eqs.(31), (32), (34) and (35), we can have factor α 0 assumed within 0 < α 0 ≤ 1 . Suppose
the derivative matrix, [ ∂Y (z)/∂z ] . The equa- the h j i-th cycle where ∆Y h j i ( = Y h j i − Y ∗ ),
tion Y (z) = Y ∗ is now prepared to be iterated [ ∂Y / ∂z ]h j i and R hSj i are evaluated on the equi-
upon the tangent coefficients. However, as is well
known, the shape-finding problem has its own dif- librium {dΓ }E h j i for z h j i . The formula of frac-
ficulties: an arbitrary Y ∗ is not necessarily real- tional correction is written as
izable; and, even for a possible Y ∗ , we are faced ∙ ¸−1
∂Y
with which members to be taken active. This z h j+1 i = z h j i − αh j i ∆Y h j i (37)
∂z hji
feasibility problem for a general cable structure
seems quite difficult to be clarified. But, once a where the current αh j i is generated by
specific structure is presented to us, we can de- Ã !
duce its possible equilibrium shapes to a certain α 0 R hS0 i
αh j i = min ,1 (38)
extent. From this view point, it is here assumed R Sh j i
that Y ∗ is lying in a realizable range for our z .
After the change of cable lengths to z h j+1 i , the
The stiffness relations of our cable elements are
responsive equilibrium {dΓ }E h j+1 i is computed in
quite nonlinear: for instance, the longitudinal co-
efficient in stiffness [ k̄( T l ) ](c) ( = [ Q( T l ) ]−1 the displacement method, Sec. 4.
(c) ) is
To the line segment Y ( ξ ) = ( 1 − ξ )Y 0 + ξ Y ∗
comparable to EA/l for ¯l ≥ l , but becomes much ( 0 ≤ ξ ≤ 1 ), there is considered a correspond-
smaller for ¯l < l . That matrix can even be singu- ing z( ξ ) continued from z 0 , where Y 0 = Y (z 0 ) .
lar as stated in Sec. 2. Thus, similarly to the For the convergence to Y ∗ , function Y (z) is re-
equilibrium computation, the full correction of quired to be a one-to-one on the entire path, i.e.
error vector Y − Y ∗ on the tangent coefficients det [ ∂Y / ∂z ] 6= 0 for 0 ≤ ξ ≤ 1 . In the present
can yield an excessive change of z . The method non-potential problem, the magnitude of vector
of fractional correction is conceptually valid to difference Y h j i − Y ∗ is estimated by the scalar
the present problem also, but, in practice, it is
R hSj i . Thus, the resulting equilibrium after each
necessary for vector Y − Y ∗ to be transformed
into a suitable scalar. The function Y (z) does correction (37) is observed for the following two :
not have its own potential. The absolute value ³ ∙ ¸ ´
∂Y
| Y − Y ∗ | might be considered as an alternative. A) sgn. det
∂z h j+1 i
But, this numeric magnitude is inadequate as a ³ ∙ ¸ ´
physical quantity; for, first of all, it is a mixture ∂Y
= sgn. det (39a)
of dimensionally different components, i.e. trans- ∂z hj i
lational and rotational coordinates of joints, and B) R hSj+1 i < R hSj i (39b)
cable tensions.
We here consider an energy-based conversion If A) or B) is not true, the factor α 0 is regarded
of Y − Y ∗ . The differences are existing in an still too large. To resume the iteration from the

■
8
34s
previous z h j i , similarly to Eq.(27), the basic cor-
rection factor is changed into
αh j i false R Sh j i
α 0 new = · S (40)
2 Rh 0 i
Differently from the equilibrium problem, it
is not always the case that a solution z ∗ ex-
ists for a required Y ∗ . When iterated for an
impossible Y ∗ , the derivative matrix [ ∂Y / ∂z ]
comes to be singular with a repeated reduction
of α 0 . On such a nearly singular [ ∂Y / ∂z ] ,
the correction (37) yields a large change of z . Fig. 5 Initial equilibrium shape of a guyed cantilever
Even if the equilibrium is obtained for those ca-
ble lengths, factor α 0 will be halved further for
violating the requirement A) or B). After such
extreme changes of z , it can usually happen that
the equilibrium is driven out of the stable do-
main. Our displacement-method computation is
unable to pursue an equilibrium beyond the sta-
bility boundary.
Once a cable structure is presented to our
shape finding, an unconsciously imposed Y ∗ is
not likely to be attainable. But, a preliminary
inspection makes it possible to array admissi-
ble objectives and active members. So far as
det [ ∂Y / ∂z ] 6= 0 on the aforesaid path z( ξ ) ,
our sequence { Y h 0 i , Y h 1 i , · · · } computed with
a small enough α 0 is kept close to Y ( ξ ) =
( 1 −ξ )Y 0 + ξ Y ∗ . The difference of z h j i from so-
lution z ∗ is represented by the scalar R hSj i . So, in
addition to the above one-to-one between Y and
z , function R S (z) is required to be monotone-
Fig. 6 Convergence to the initial equilibrium
decreasing on the entire 0 ≤ ξ ≤ 1 . In its defini-
tion, R S (z) seems to embody this monotonicity
considerably, unless tangent stiffness [ KΓ (z) ]E in which “normalized” means that the outermost
has enormous changes on z( ξ ) . value in each line is scaled to ±1 . Started with
θ 0 = 0.25 , the basic factor is changed twice to
0.0625 for a long-distant equilibrium state with
6. NUMERICAL EXAMPLES the deeply hanging cables.
(1) A cantilever supported by deeply From the above equilibrium state, the cable
hanging cables lengths are computed so that the lumped beam
The cantilever shown in Fig. 5 has elastic mod- weights are precisely upheld by the cable tensions,
ulus 20600 kN/cm2 , area 63.53 cm2 and moment i.e. Tl y (1) ∼ (4) → 5 and Tl y (5) → 2.5 kN . The
of inertia 1600 cm4 of cross-section, and weight convergence process from basic factor α 0 = 0.065
per unit length 0.5 kN/m . After segmented into is indicated in Fig. 8, and the equilibrium shape
five elements of length 10 m, the joints are guyed is settled as shown in Fig. 7: the achieved ca-
to another joint fixed at { x, y } = { 30, −15 } m ble lengths are l(1) = 25.2294 , l(2) = 18.1368 ,
with catenary cables of E = 15700 kN/cm2 , A = l(3) = 14.9998 , l(4) = 18.1366 and l(5) = 25.4390 m .
45.24 cm2 and w = 0.364 kN/m . Initially, an
equal length l = 30 m is assumed to all the ca- ( 2 ) A vertically suspended girder
ble members. Shown in the figure is the equilib- The girder shown in Fig. 9 is divided into
rium configuration for those cable lengths with eight elements of length 10 m , and which are
the beam weights being lumped at their joints, assumed to have elastic modulus 20600 kN/cm2 ,
which is computed from the joint positions in- area 1200 cm2 and moment of inertia 2667000 cm4
dicated with broken lines. The process of our of cross-section, and weight per unit length
fractional correction method is shown in Fig. 6, 9.24 kN/m. E = 19620 kN/cm2 , A = 22.04 cm2

■
9
35s
 Fig. 7 After settlement of cable lengths
and w = 0.1697 kN/m are given to the main-cable
elements; and E = 15695 kN/cm2 , A = 3.50 cm2
and w = 0.02698 kN/m , to the hanger cables.
Cable joint 10 and 18 are fixed at { x, y } =
{ 0, −20 } and { 80, −20 } m , and beam joint 1
and 9 are supported by horizontal rollers at height
y = 0 . At the beginning of computation, the
joints are located at x = 0, 10, · · · , 80 m on level
y = 0, −20 m , and where the initial cable lengths
are assumed by their joint-to-joint distances.
The objective conditions for all the 15 cables:
the beam joints are required to be on height
y = 0 m , with the cable joints staying at x =
10, 20, · · · , 70 m ; and, in the chained (1) to (8),
the degree of sag is specified by y = −4 m at the
center joint 14.
When computed as the catenary elements, the
shape finding was not successful. This result
comes from the following: in the initial equilib-
rium, all the hanger cables are (nearly) in the sin-
gular state, that is, are almost in vertical hanging
with the joint-to-joint distances less than their
member lengths (see Fig. 1). Our displacement
method can deal with such a singular state in
equilibrium computation. 12) But, in the shape
finding, the tangent coefficients to variation of
their natural lengths become quite different in
quality (far small) from the ordinary configura-
tions. The tangent correction by (37) with a usual
αh j i yields an extreme change of cable lengths
such as to negative values.
Next, the same shape finding is executed for the
simple tension members. Since the hanger cables
come to be slackened toward the initial equilib-
rium, a vertical roller support is added to joint 5
to stabilize the girder. This equilibrium state is
shown by the broken lines in Fig. 9. As stated in
Appendix—A, the slackened cable lengths found
in a structural equilibrium are immediately re-
placed by the spatial joint-to-joint lengths, and
the tangent approximation (37) is continued for
the objective values. In this shape finding for
α 0 = 0.2 , the convergence of R Sh j i is obtained as
■
10
36s
Fig. 8 Convergence of the shape finding
shown in Fig. 10. Shown in Fig. 9 is the result-
ing equilibrium configuration. The achieved cable
lengths and their tensions are listed in Table 1.
This result is followed by the shape computation
as the elastic catenary elements: after only one
full correction (α 0 = 1), the structure is shifted
within the same allowable error 10−6 (kN·m)1/2
for R S . Those quite small changes are invisible
in the decimals of Table 1 (chord force T ∗ of a
catenary cable corresponds to axial tension T of
a straight cable).
( 3 ) A cable-trussed bridge
The cable-trussed bridge shown in Fig. 11 is
made with the use of compressive axial members,
namely (5), (7), (12) and (14). The girder has
area 0.2784 m2 and moment of inertia 0.1618 m4
of cross-section, elastic modulus 20600 kN/cm2
and weight per unit length 32.15 kN/m , which is
divided into l[ 1 ] = l[ 8 ] = 16 , l[ 2 ] = l[ 7 ] = 20 , l[ 3 ] =
l[ 6 ] = 23 and l[4] = l[5] = 15 m . Tower member [ 9 ]
has length 20 m , elastic modulus 20600 kN/cm2 ,
area 0.0672 m2 , moment of inertia 0.006656 m4
and weight per unit length 5.17 kN/m , which is
rigidly connected to the girder at joint 5 instead of
to the ground. E = 15700 kN/cm2 , A = 57.9 cm2
and w = 0.466 kN/m are assumed to cable mem-
ber (1), (4), (8) and (11) ; and E = 15700 ,
A = 32.57 and w = 0.262 , to the remaining ca-
ble members. Those cable members are treated
as the elastic catenary. Ē = 20600 , Ā = 288 and
w̄ = 2.218 are for axial member (5) and (12); and
Ē = 20600 , Ā = 160 and w̄ = 1.232 for (7) and
(14). Joint 5 is hinged to the ground at the ori-
gin of { x, y }-coordinates, and joint 1 and 9 are
 Fig. 9 Initial and settled equilirium shapes of a suspended girder

Table 1 Result of shape finding

cable l (m) T (kN)

(1) (8) 12.1940 576.07
(2) (7) 11.1661 527.45
(3) (6) 10.4275 492.52
(4) (5) 10.0387 474.14
(9) (15) 12.9717 92.58
(10) (14) 7.9816 92.51
(11) (13) 4.9900 92.47
(12) 3.9933 92.45

Fig. 12. In the settled equilibrium shape (see

Fig. 10 Iteration process for the simple tension members
Fig. 11), the disagreements of the respective Ys
from Ys∗ are less than 10−14 m , and the beam
bendings are correspondingly negligible for their
self-weights lumped at the joints. The attained
member lengths, and the chord and axial forces
are given in Table 3.

kept on y = 0 by horizontal rollers. 7. CONCLUDING REMARKS

Initially, joint 2 to 8 are placed on y = 0 with
no elongations in the beam elements; and joint
12, 11, 10, 13 and 14 are just above joint 3, 4,
5, 6 and 7 on height y = −20 , respectively. The
spatial distances between those joint positions are
assumed to the cable and axial members as their
initial lengths. Released from those joint posi-
tions, the equilibrium configuration is obtained
as indicated by broken lines in Fig. 11, where the
deflections at the beam joints are y 2 = y 8 = 7.64 ,
y 3 = y 7 = 9.75 and y 4 = y 6 = 3.18 cm .
The followings are required for the 14 variable
member lengths : the beam joints are kept ex-
actly on y = 0 ; the cable joints are positioned at
height y 11 = y 13 = −16 and y 12 = y 14 = −10 m ;
and the compressive axial members are inclined
with diagonal proportion 1 : ± 2 . Those require-
ments are listed specifically in Table 2. The
convergence process for α 0 = 0.1 is shown in
The “elastic catenary” was formulated more
than a hundred years ago, but which had been
held from practical utilization until the computa-
tional age. The force-method treatment for the
compatibility can be extended directly to an as-
sembly of those catenary members. 7) But, the
displacement method is employed for our struc-
ture compounded of cable and beam members,
in which the fractional correction method is ap-
plied to the layered numerical problems, namely,
for the compatible tension components in each
catenary cable and for the structural equilibrium.
Those equilibrium computations are based on an-
alytical treatment of the nonlinearity exhibited
by the consisting members: the convergence to
solution is secured for a convex potential prob-
lem, so except for the structural instability.
The present shape finding for a beam-cable

■
11
37s
 Fig. 11 Initial and settled equilirium shapes of a cable-trussed bridge

Table 2 Requirements in the shape finding

Ys Ys∗ (m)
Y1 = y 2 0.
Y2 = y 3 0.
Y3 = y 4 0.
Y4 = y 6 0.
Y5 = y 7 0.
Y6 = y 8 0.
Y7 = y 11 −16.
Y8 = y 12 −10.
Y9 = y 13 −16.
Y10 = y 14 −10.
Y11 = x 11 − x 4 − 0.5y 11 + 0.5y 4 0.
Y12 = x 12 − x 3 − 0.5y 12 + 0.5y 3 0.
Y13 = x 13 − x 6 + 0.5y 13 − 0.5y 6 0.
Y14 = x 14 − x 7 + 0.5y 14 − 0.5y 7 0.
Fig. 12 Iteration process of the shape finding
structure is developed from the preceding one
for a mere assembly of straight cables. 8) In Table 3 Settled member lengths and chord/axial forces
that extension, no fundamental difficulties are
encountered beyond the structural complexity. Cable and Compressive Members ( in kN, m )
This is indicating a potential capability of the
(c) length chord/axial force
displacement-method system. The objective
state is approached also by the fractional tech- (1) (8) 23.2810 2433.3
nique. Even in a non-potential problem, if an ad- (2) (9) 20.8367 1060.4
equate scalar is available for the deviation, that (3) (10) 17.9907 1047.5
iteration can be applied for the solution exist- (4) (11) 24.9418 1931.4
ing in a one-to-one correspondence. The present (5) (12) 17.8916 —1015.8
shape finding is a problem of simultaneous alge- (6) (13) 21.8732 1356.0
braic equations with quite nonlinear behaviors of (7) (14) 11.1814 —322.3
each catenary element. In that iteration scheme,
there might be considered other additional tactics Beam Members ( in kN )
for a more persistent achievement over their tight- [e] axial force [e] axial force
ened and slackened states. On the other hand,
the shape determination for the straight tension [1] [8] 0.0 [2] [7] —871.6
members yields a steady convergence owing to the [3] [6] —1943.1 [4] [5] —3556.4
simple treatment of their relaxation (Appendix— [9] —3996.9
A), and from which the computation could be
continued as the catenary cables, if necessary.

■
12
38s
APPENDIX A.
SIMPLE TENSION MEMBER AND
COMPRESSIVE AXIAL MEMBER
In cable element (c), let the distributed self-
weight be lumped into halves wl/2 at its both
ends. When spanned to { x̄, ȳ } , this straight
cable is lengthened by
Fig. 13 Straight cable
³ ´ q
∆¯l = ¯l − l = x̄2 + ȳ 2 − l (41) With substitution of (42b), the quadratic form
of matrix [ k̄ ](c) can be developed into
With its no resistance to compression, the tension ³ ´
£ ¤
vector T̄ is written as follows : δ 2 I(c) = δ x̄T k̄ (c) δ x̄
½ ¾ ½ ¾ 
T̄x T x̄  EA
= ¯ (42a) 

 ( x̄ δ x̄ + ȳ δ ȳ ) 2
T̄y l ȳ  ¯
l3
 = T
+ ¯ ( δ x̄2 + δ ȳ 2 ) for ∆¯l ≥ 0 (46)
 EA ∆¯

l for ∆¯l ≥ 0


 l


T = l (42b) 0 for ∆¯l < 0

 ¯ 0 for ∆l < 0
So far as tensioned ( ∆¯l > 0 ), matrix [ k̄ ](c) is
Consider joint i and j to which element (c) is positive definite; and becomes singular
p into the
attached: x̄ = xj − xi and {x}(c) = { xi , xj } . circumferential direction at limit x̄2 + ȳ 2 → l .
As shown in Fig. 13, the joint forces, {T }(c) = In the circular domain of ∆¯l < 0 , this tension
{ T i , T j } , are here defined after inclusion of the element is singular into any directions.
lumped self-weights : Similarly to Eqs.(11) and (12) for a catenary
element, we now consider the tangent coefficients
½ ¾ ½ ¾ ½ ¾
Tx T x̄ 0 to variation δl of natural length with x̄ being
=−¯ − (43a) fixed. From Eq.(42b), δT = −EA ¯l/ l2 · δl for
Ty i l ȳ wl/2
½ ¾ ½ ¾ ½ ¾ ∆¯l ≥ 0 . We have the differentiation of Eqs.(43)
Tx T x̄ 0 written as
= ¯ − (43b)
Ty j l ȳ wl/2 ½ ¾
ψi
δ{ T }(c) = { ψ }(c) δl = δl(c) (47a)
Accordingly, the element potential is given by the ψj
sum of the strain energy and the self-weight po- ½ ¾ ½ ¾
EA x̄ 0
tential : ψi = − (47b)
l2 ȳ w/2
½ ¾ ½ ¾
T 2l wl EA x̄ 0
U(c) + V(c) = − (yj + y i ) (44) ψj = − − (47c)
2EA 2 l2 ȳ w/2

By differentiating the above relations with re- When the simple tension elements are active in
spect to x̄, we have the tangent stiffness matrix : the shape finding, it can happen that some of
them are slackened ( ∆¯l < 0 ) in the passing struc-
δ T̄ = [ k̄(x̄) ](c) δ x̄ (45a) tural equilibria. Their tangent coefficients have
£ ¤ £ ¤ £ ¤ no validity any more. As a counter treatment, we
k̄ (c) = k̄ M (x̄) (c) + k̄ G (x̄) (c) (45b) have their natural lengths replaced by the spatial
£ ¤
∙ ¸ joint-to-joint lengths ( ∆¯l = 0 ), prior to the si-
EA x̄x̄ x̄ȳ
k̄ M = (45c) multaneous approximation of all the cable lengths
(c)
l ¯l2 x̄ȳ ȳȳ
upon [ ∂Y (z)/∂z ] , Eq.(37).
∙ ¸
£ ¤ T ȳ ȳ −x̄ȳ When the incompressibility for ∆¯l < 0 is elim-
k̄ G (c) = ¯3 (45d)
l −x̄ȳ x̄x̄ inated, we have the correspondent relations for
the usual axial-force element. For simplicity, let
If ∆¯l < 0 , zero matrix stands for both [ k̄ M ] and the same notations such as {T }(c) be employed
[ k̄ G ] . Through δ x̄ = δxj − δxi and δ T̄ = δT j = even for those compressible members. In the do-
−δT i , those [ k̄ ], [ k̄ M ] and [ k̄ G ] are rewrit- main of ∆¯l < 0 , this element is not singular, but
ten into the 4 × 4 matrices, [ k ](c) , [ k M ](c) and unstable : in Eq.(46), δ 2 I(c) < 0 for δ x̄ such that
[ k G ](c) , respectively. x̄T δ x̄ = 0 .

■
■
13
39s
APPENDIX B.
FOR PRACTICAL SHAPE FINDING
In Sec. 5, the components selected among vec-
∗ } and { T
tor {X J }, { T l (C) }, { T(C)
l (C) } are indi-
vidually subjected to their objective values. But,
those conditions are not sufficient for an actual
structure: for instance, some cable tensions con-
nected to a joint might be required to offset their
horizontal or vertical components to each other;
and some free joints are, to be located on the same
height. With no difficulty, the capability of our
shape finding is much widened by imposing the
objective values on linear combinations of those
equilibrium quantities :
Y1 = C 1 1 Y10 1 + C 1 2 Y10 2 + · · · + C 1 K1 Y10 K1
..
. R hSj i =
Ys = C s 1 Ys0 1 + · · · + C s Ks Ys0 Ks (48)
..
.
YS = C S 1 YS0 1 + · · · · · + C S KS YS0 KS
where Ys0 k ’s are the direct components taken from
the above {X J } to { Tl (C) } ; and C s k ’s are rel-
ativity constants. The tangent coefficients be-
tween Y and z are obtained in the same way:
[ ∂Y / ∂z ] = [C ][ ∂Y 0 / ∂z ] . In the iterative com-
putation, the first Ys0 1 is treated a little differ-
ently from others in each Ys : when error vector
∆Y h j i = Y h j i − Y ∗ is found on the h j i-th cycle,
scalar R hSj i by Eq.(36) is calculated for those un-
balanced joint forces which result from the shift
of only Ys0 1 h j i ’s :
∆Ys h j i
Ys0 1 h j i → Ys0 1 h j i − , s = 1, · · · , S (49)
Cs1
In our fractional iteration upon the tangent
coefficients, the member lengths in z are im-
proved along by the linear path Y ( ξ ) = ( 1 −
ξ )Y 0 + ξ Y ∗ . Even if the initial Y 0 has a long
distance to Y ∗ , we expect the objective state
to be attained in so far as det [ KΓ ( ξ ) ]E > 0
and det [ ∂Y / ∂z ] 6= 0 on the exact path z( ξ )
( 0 ≤ ξ ≤ 1 ). However, in principle, our struc-
tural analysis has a difficulty in handling finite
strain states: when subjected to finite deforma-
tions, the beam elements can easily become un-
stable. The discretized beam elements themselves
are prepared not so much for finite strain states
as for finite displacements: the stiffness relations
by the separation method 15) can deal with any
large displacements (as a rigid body), but are
still a disputable approximation for finite defor-
mations. So the equilibrium states corresponding
to the exact z( ξ ) are required to stay in a range
of relatively small strains. In our actual compu-
tation, those small-strain states are kept by the
■
14
40s
fractional corrections with a sufficiently small α 0 .
But, to convert error vector ∆Y h j i into the scalar
R hSj i , once in each cycle h j i, the structure can
be deformed extremely by the aforesaid shift of
Ys0 1 h j i . By the use of a factor such as the previ-
ous αh j−1 i taken in z h j−1 i → z h j i , we can have
R hSj i also for a small disturbance: Ys0 1 h j i’s are
shifted for a fractional αh j−1 i ∆Y h j i instead of
by (49), that is
∆Ys h j i
Ys0 1 h j i → Ys0 1 h j i − αh j−1 i (50)
Cs1
and, from the resulting {∆FΓ }Sh j i , error R Sh j i for
the total ∆Y h j i is estimated by
q
1
{∆FΓ }S T [ KΓ ]E −1 {∆FΓ }S h j i
αh j−1 i
(51)
REFERENCES
1) Avent, R.R.: Nonlinear field analysis of structural
nets, J. Struct. Div., ASCE, Vol.95, No.ST5, pp.889—
907, May, 1969.
2) Namita, Y. and Nakanishi, H.: A method of compu-
tation for determining the shape of cable structures,
Trans. Japan Architectural Inst., No.212, pp.37—45,
Oct., 1973 (in Japanese).
3) Ohyama, H. and Kawamata, S.: A problem of surface
design for prestressed cable nets, Proc. IASS Pacific
Symposium—Part II, Tokyo and Kyoto, pp.103—115,
1972.
4) Schek, H.J.: The force density method for form
finding and computation of general networks, Com-
puter Methods in Applied Mechanics and Engineering,
Vol.3, pp.115—134, 1974.
5) Argyris, J.H., Angelopoulos, T. and Bichat, B.: A
general method for the shape finding of lightweight
tension structures, Computer Methods in Applied Me-
chanics and Engineering, Vol.3, pp.135—149, 1974.
6) Nakanishi, H. and Namita, Y.: Shape determination
analysis of cable structures by means of the methods
for optimization problem, Trans. Japan Architectural
Inst., No.238, pp.13—23, Dec., 1975 (in Japanese).
7) Ai, M., Nishioka, T. and Okumura, T.: A theoreti-
cal analysis of cable assemblies, Proc. JSCE, No.260,
pp.17—32, Apr., 1977 (in Japanese).
8) Ai, M. and Nishino, F.: An equilibrium and a shape-
determination analysis of cable assemblies upon the
displacement method, Proc. of Symposium on Com-
putational Methods in Structural Engineering, Vol.12,
JSSC, pp.109—114, Jul., 1988 (in Japanese).
9) Nishino, F., Duggal, R. and Loganathan, S.: Design
analysis of cable networks, J. Struct. Engng., ASCE,
Vol.115, No.12, pp.3123—3141, Dec., 1989.
10) Maeda, Y., Hayashi, M. and Maeda, K.: Analyti-
cal calculation of sagging cable, Proc. JSCE, No.257,
Jan., 1977 (in Japanese).
11) Goto, S.: Tangent stiffness equation of flexible cable
and some considerations, Proc. JSCE, No.270, pp.41—
49, Feb., 1978 (in Japanese).
12) Ai, M. and Ohsumi, Y.: The elastic catenary as
a displacement-method element, Structural Eng./
Earthquake Eng., JSCE, Vol.17, No.2, pp.241s—246s,
Oct., 2000.
13) Rough, E.J.: A Treatise on Analytical Statics, Cam-
bridge Univ. Press, Cambridge, England, Vol.1, 2nd
Edition, 1896.
14) Ai, M., Nakano, T. and Masuda, S.: A considera-
tion on the slackened and tightened cables, Structural
Eng./Earthquake Eng., JSCE, Vol.11, No.3, pp.141s—
144s, Oct., 1994.
15) Ai, M. and Nishino, F.: Mechanics in geometrically
nonlinear problem of discrete system and application
to plane frame-works, Proceedings of JSCE, No.304,
pp.17—32, Dec., 1980 (in Japanese).

(Received March 19, 2003)

■
15
41s