Sunteți pe pagina 1din 10

Shooting--optimization

technique for large


deflection analysis of
structural members
C. M. Wang
Department of Civil Engineering, National University of Singapore, Kent Ridge, 0511,
Singapore

S. K i t i p o r n c h a i
Department of Civil Engineering, Universityof Queensland, St. Lucia, Queensland, 4072,
Australia
(Received April 1990)
Large deflection and postbuckling responses of structural members
may be obtained either by using the finite element method or by
solving the exact nonlinear governing differential equation. This paper
deals with the latter approach. An easy-to-use technique for solving
the nonlinear two-point boundary value problem is proposed. The
technique combines the idea of the shooting method and optimization
algorithms. Using the shooting method, the two-point boundary value
problem is converted into a set of initial value problems which are
then integrated forward using the Runge-Kutta algorithm. For a
solution, the terminal boundary conditions are satisfied by adjusting
the unknown initial values and parameters of the differential equations. The adjustments are carried out through the minimization of the
errors in the terminal boundary conditions using an optimization
aigor;~;~m. The technique is demonstrated on a large class of
problems involving large deflection and postbuckling of beams,
columns, rings, arches and frames.

;~,~ywords: beam bending theory, large deflection, nonlinearity,


optimization, shooting method, structural member analysis
Most analyses of structural members are based on the
conventional 'Engineers' theory of bending'. The theory
should not be used to determine the large deflection
response of structural members, however, because of
the underlying approximations. When it is applied to
determine the large deflection behaviour of members,
the theory yields results which are inaccurate and, in
some cases, grossly in error. The theory is also limited
to giving deflections normal to the original member
position.
Essentially, there are two approaches for solving large
deflection problems. The common approach uses the
finite element method 1.2 Many elements are necessary
for an accurate solution, however, because of the
iinearization of the energy functional. Further, the solution must be obtained iteratively because the tangent
stiffness matrix has to be repeatedly updated to account
for changes in the shape of the member.
The second approach is to formulate the exact governing differential equation from static and kinematic considerations, but the resulting nonlinear two-point
boundary value problem is difficult to solve. Common
techniques for solving linear differential equations such
as the finite difference method 3-s, the finite integral
0141-0296/92/040231

method 6, the least mean squares error method 7 and


Galerkin's method s cannot easily be applied to nonlinear two-point boundary problems because it is
necessary to solve a large system of nonlinear algebraic
equations. Semi-analytical techniques such as the elliptic
integrals method 9 and the power series method 9 may
be used, but the procedure involved with these methods
can be rather tedious.
A more convenient technique for solving highly
nonlinear ordinary differential equations is the shooting
method t-t~. In this technique, the two-point boundary
value problem is converted into a set of initial value problems. The unknown initial boundary values are then
determined iteratively until the terminal boundary conditions are satisfied. Essentially, it involves solving a set
of implicit nonlinear algebraic equations which express
the boundary conditions. For the analysis of structural
members, this method is computationally far more efficient than the finite element approach since it does not
require matrix manipulations or inversions, and the
number of equations involved is significantly smaller.
Surprisingly, in general, this technique has not been
commonly applied to study the nonlinear large
deflection behaviour of structural members.

- 10

1992Butterworth-Heinemann Lid

Eng. Struct. 1992, Vol. 14, No 4

231

Large deflection analysis of structural members: C. M. Wang and S. Kitipornchai


This paper proposes a technique which combines the
idea of the shooting method and optimization
algorithms. Instead of solving the algebraic equations of
the boundary conditions, the sum of the L~ norm errors
of the terminal boundary conditions is minimized with
respect to the unknown initial values and any unknown
parameters in the differential equation. The minimization process can be carried out using a standard
optimization algorithm. For convenience, the direct
search optimization methods are used here because they
do not require any gradient computations. The proposed
shooting-optimization technique will be illustrated by
various examples of large deflection problems involving
beams, columns, rings, arches and frames.

equation (4) may be solved (or integrated forward) using


the fourth-order Runge-Kutta algorithm t2 given below
(1)

Specify the step size of integration, h, the initial


values of x(0) and ~,, i = 1, 2 . . . . .
N.
(2) For n = 0, 1, 2 . . . . .
(l/h - 1) do:

x(n + 1) = x(0) + (n + 1)h


ci, (n) = h f [x(n), zl (n), z2(n) . . . . .

ZN- | (n)]

h
x(n) + ~, zi(n) + 2 2 , z2(n)

c,2(n) = h f
+

c21

, ZN(n)+C~ -]

Shooting-optimization technique
Consider a general Nth order ordinary differential equation of the form

dNY
~ - f

(x

dY d2Y
,Y,--,-....
dx dx 2'

+ ~ .....

dN-'Y )
dx N-~' ~

ci4(n ) ----hf

k = 1, 2 . . . . .

zN(n) +

(1)

where X = {hi, X2. . . . .


kMIr is the unknown
parameter vector which may represent unknown stress
resultant parameters and/or eigenvalues (as in stability
and vibration problems). The function f ( . ) may be
linear or nonlinear, but it must be continuously differentiable with respect to each argument. The initial and
terminal boundary conditions, after appropriate normalization, are

N-~
dJy
~a atJdxJ],=
-/=0
-- ako=0,

z,n,+7

c~3(n ) = hfi

-t- h, z|(n)

Ix(n)

+ c23. . . . .

+ cl 3,

zz(n)

z~(n) + cs3 ]

zi(n + 1) = zi(n) + 1/6[cil + 2ci2 + 2ci3+ci4]


Using equation (4), the right-hand sides of equations
(2) and (3) become, respectively

i=|

(2)
N-'

dJy

i=1

k=r+l
j=O

k = r + l

.....

N+M

(3)

To solve the two-point boundary value problem given by


equations ( 1 ) - (3), the Nth order differential equation
is first reduced to a set of more readily solved first-order
differential equations using the auxiliary function
z; +~ = the ith derivative of y with respect to x where
i = 0, 1,2 . . . . N - 1. Thus equation (1) may be rewritten as
dz~
-dx
dz2
dx

= z2

z~ ( 0 ) = 1-~

z2(0)

23

1"2

(4)

d z ~ = f ( x , zl, z2 . . . . .
dx

ZN, X)

zN(0) =1-N

where, in general, some of the initial values of ~" are


unknowns. Let this set of Unknowns be represented by
the vector ~ . By assuming some arbitrary values of ~ ,
the set of first-order differential equations given by

232

.....

N+M

(6)

Eng. Struct. 1992, Vol. 14, No 4

Note that z;(l) is given by zi(llh) in the Runge-Kutta


algorithm. For a solution, the value of ek(k = 1, 2 . . . .
N + M) in equations (5) and (6) must be zero. Thus
equation (5) furnishes r simple linear algebraic equations in ~ while equation (6) gives (N + M - r) nonexplicit functions of 2 and ~.
The natural course for solving the nonlinear differential equation is to solve the (N + M) algebraic (equations
(5) and (6)) iterafively. For efficient computation, it is
necessary to use a systematic procedure for changing
successive values of X and ~ so that equations (5) and
(6) are satisfied. A l~opula~"method is that of Newton
which, however, requires evaluation of the gradients of
the functions ek with respect to ~ andif.*~These gradients
can be derived on the basis of variational equations ~0
or based on optimal control theory |3.
An alternative approach to solving the nonlinear
algebraic equations (equation (6)) is as follows. Let the
objective function, formed by taking the sum of the
absolute values of the functions ek, k = r + 1. . . .
,
N + M, be minimized with respect to X and ~'*, i.e.

N+M
M
in~=
~,.~
-

k=r+ I

lek(~, N1-)1

(7)

Large deflection analysis of structural members: C. M. Wang and S. iKitipornchai


Note that some of the functions ~k may be described by
interior point constraints or geometrical constraints (see
example 4). It is clear that the desired value of must
be zero for solution.
The optimization problem defined by equation (7) is a
simple unconstrained one and may be solved by direct
search techniques such as the Nelder and Mead simplex
method, the Hooke-Jeeves pattern search method or
the Powell conjugate direction method 14. If the problem
involves only one unknown, a simple golden section
method t4 may be used for rapid solution. For those
familiar with optimization techniques, the proposed
combined shooting-optimization method is relatively
easy to use for solving any form of two-point boundary
value problem. The primary reason for using direct
search techniques is to avoid the need for gradient computations and thus reduce the preparation of the computer code; thereby making the technique more
attractive.
In the following sections, various nonlinear large
deflection problems are solved to demonstrate the proposed technique. For all the problems considered, the
following assumptions are made: (i) the material is
linearly elastic; (ii) the axial and shear deformations are
negligible, (iii) the members are perfect and prismatic
and (iv) the structure and loads are coplanar.

Numerical examples
Example 1: large deflection of cantilever beam subject
to an inclined end point load
Figure I shows a cantilever beam of flexural rigidity E1
and length L. The beams carries an inclined end point
load P whose inclination angle o~ is measured with
respect to the horizontal. The proposed technique will
now be used to determine large deflections of the cantilever for any given values of P and or.
Let the distance measured along the beam from the
fixed end be s and the local slope be 0 (see Figure 1).
Adopting the following nondimensional quantities
s
g=L '

PL 2
~ = E1

(8)

the exact equilibrium equation for shear forces is given


by
d20(j)
- + B(cos o~ sin O(g) + sin o~ cos O(g)) = 0 (9)
dg 2

Mo [

Rj~_. ~jO

yR2 ~

L
X1

J~

x~01

Figure 1 Cantilever beam subject to an inclined end point load

The boundary conditions are

0(0) = 0
d0(0)
- ~dg
d0(l)
- 0
dg

(10a)
(10b)

(10c)

where the unknown initial value ~"= MoL/(EI). Mo is


the moment at the fixed end. The nonlinear differential
equation (9)) and its boundary conditions (equation (10))
can readily be solved using the proposed shootingoptimization method. The second-order differential
equation is first converted into two first-order differential equations using equation (4). The satisfaction of the
terminal boundary condition is then ensured by minimizing the following objective function
Min = --d~21)

(11)

Using a step size of h = 0.005 for the Runge-Kutta


algorithm (this step size will be used for solving all
subsequent problems), the deflection curves can be
obtained in terms of the rectangular coordinates (x, y)
using the following relationships

df
Y ~Iosin O(/j)d~
d-~ = s i n O - - Y = ~-

(12a)

- - = cos 0 -- ~ = - =
dg
L

(12b)

cos 0(~)d~

Table 1 presents numerical results for the fixed end


moment, ~', the deflected position of the free end (y~/L,
x~/L) and the slope 0~ for given values of applied load
magnitude, /3, and the load inclination, ct. Figure 2
shows the deflection curves of the cantilever beam for
= ~'/4 and various values of/3. The solutions derived
have been found to be in very close agreement with
those obtained by Mattiasson z5 who used the elliptic
integrals method.
It should be noted that if the Engineers' theory of bending had been used, it would have furnished an end
deflection of
Yt _ B sin a
L
3

and

xt =0
L

(13)

Figure 3 compares the end vertical deflections of the


cantilever beam (or = lr/2) obtained from solving the
exact nonlinear differential equation using the proposed
method with those obtained using the Engineers' theory
of bending (equation (13)). It is seen that for small
deflections (i.e. yl/L. less than 0.3) the Engineers'
theory of bending is adequate. However, the two curves
then diverge as the deflection increases (see Figure 3).
For example when /~ = 6, the Engineers' theory of
bending predicts an end vertical deflection equal to twice
the length of the cantilever beam which is totally

Eng.Struct.1992,Vol. 14, No4 233

Large deflection analysis o f structural members: C. M. Wang and S. Kitipornchai


Table I

Large deflection results for a cantilever beam subject to an inclined end point load
pL 2

~ -

El

MoL
El
Yl IL
~" -

7r

x1/L

01
T

Yl/L

xl/L
01
3~"

~"

-4

Yl/L
x1/L
01

1.0

2.0

3.0

4.0

6.0

8.0

O.8776

1.8987

2.7369

3.3892

4.3786

5.1479

0.2956

0.5831

0.7330

0.7976

0.8358

0.8386

0.9455
0.4575

0.7595
0.9808

0.5572
1.3574

0.4007
1.6019

0.1962
1.8839

0.0714
-270-361-

0.9436
0.3017
0.9436
0.4614

1.6787
0.4935
0.8394
0.7818

2.2367
0.6032
0.7456
0.9860

2.6842
0.6700
0.6711
1.1212

3.3925
0.7446
0.5654
1.2837

3.9614
0.7850
0.4952
1.3744

O. 5675
O. 1783
0.9808
0.2660

0.9505
0.2801
0.9522
0.4172

1.2379
0.3439
0.9274
0.5107

1.4708
0.3873
0.9074
0.5730

1.8447
0.4431
0.8779
0.6487

2.1485
0.4779
0.8577
0.6915

unrealistic, while the actual vertical deflection is only


about 0.74L.
Example 2: post-buckling of cantilever column subject
to an end concentrated load and uniformly distributed
axial load

Let the local slope of the buckled column be 0 and the


coordinate axes be taken as shown in Figure 4 with the
distance s measured along the axis of the column from
the origin at the clamped end. Adopting the following
nondimensional quantities
S

Consider a cantilever column of length L, subject to a


combined end concentrated load P and uniformly
distributed axial load q as shown in Figure 4. The
column is prismatic and has a flexural rigidity of El. The
proposed technique will be used to study the postbuckling behaviour of such a column.

qL 3

PL 2
(14)

8,

/
I

Accurate large /
~efle

0.2
5
--J

~=1

c" 0 . 4
.9

II

// Engineers' theory
of beam bending

,n

g3

no
-a

0.6

~=2

0
J

~=8

~=6

1.0 _

pL2

1.2 0

I
0.2

I
0.4

I
0.6

I
0.8

x/L

Figure 2 Large deflections of a cantilever beam (c~ = x/4)

Eng. S t r u c t .

1 9 9 2 , V o l . 14, N o 4

x-

P=-~"

234

1P

2
0.8

1.0

0.5

1.0

1.5

2.0

2.5

Yl/L

Figure 3 Comparison of end vertical deflections of cantilever


beam calculated using Engineers' theory of bending and large
deflection theory

Large deflection analysis of structural members: C. M. Wang and S. Kitipornchai


X

and (15)) may be solved using the proposed


shooting-optimization method. The objective function
for the optimization routine may be written as

d0(1)I

Min ~I, = 10(1)- c~l + - ~ -

(17)

Upon optimization of equation (17), the buckling load k


may be obtained for given values of ~ and B. Numerical
results are presented in Table 2. The solutions are found
to be in very close agreement with those published for
the case of/3 = 0 in Timoshenko and Gere's book ~6
(Table 2 - 4 , p 79) and in Reference 17. The negative
values of X imply a tensile end concentrated load which is
necessary for equilibrium if the column is subject to a
large uniform compressive distributed load/3. Two sets
of column deflectionl6 corresponding to/~ = 0 and 10
are shown in Figure 5 for various specified end slopes,
c~. The variations of the end point load, k, with the ratio
of the end horizontal deflection, yf/L are plotted in
Figure 6. It is seen that the column exhibits considerable
post-buckling strength for all,combi~aations of/3 and
values.

L
X

by

Example 3: large deflection of a circular ring subject


to two diametrically opposite forces
Figure 4 Cantilever column under end concentrated and
uniformly distributed axial loads

the exact governing differential equation for the


equilibrium of shear forces is given by
d20
cL5 + (h +/3(1 - g))sin 0 = 0

(15)

The boundary conditions are given by equation (10). To


obtain the post-buckling characteristics, it is necessary
to prescribe the slope at the free end
0(1) = o~ = specified value

(16)

Consider a perfect circular ring of radius R and uniform


flexural rigidity El. The ring is subject to two
diametrically opposite forces, 2P, at points B and D as
shown in Figure 7a. Because of symmetry, only a
quadrant of the ring needs to be analysed. The supporting points (B, C) of the quarter ring can be treated
as fixed supports with an allowance for vertical and
horizontal translations, respectively, (see Figure 7b).
The large deflection response of the ring for a given
force, 2P, is to be determined.
Using the notations 0 and s to represent the local slope
and the coordinate along the deformed ring axis, the
exact governing differential equation for the quarter ring
is given by
d20

The nonlinear differential equation (equation (15))


together with its boundary conditions (equations (10)
Table 2

d~ 2

+ / 3 sin 0 = 0

(18)

Large deflection results for cantilever columns under end concentrated and uniformly distributed loads

qL3

X -

pL 2
El

MoL
= El

x214

2~

4~

2x

8x

2.5054

2.6245

3.1925

4.6506

9.9438

~0

0.5497

1.1082

2.2970

3.7352

6.2110

yl/L
xl/L

=0
~- 1.0

0.2194
0.9697

0.4222
0.8812

0.7195
O. 5 5 9 4

0.8032
O. 1232

0.6246
- 0.3403

5.0

k
~"
F~IL
xl IL

0.9204
-~ 0
sO
,= 1.0

0.9620
0.6752
0.2361
0.9660

1.0745
1.3523
0.4516
0.8674

1.6110
2.7335
0.7531
0.5190

2.9953
4.2746
0.8166
0.0714

8.1121
6.7231
0.6172
- 0.3730

10.0

k
~"
y~ IL
Xl IL

-0.7152
=0
~0
,= 1.0

-0.6944
0.8325
0.2557
0.9612

-0.5760
1.6530
0.4850
0.8499

-0.0301
3.2420
0.7869
0.4720

1.3264
4.8623
0.8264
0.0169

6.3012
7.2484
0.6082
- 0.4048

Eng. Struct. 1992, Vol. 14, No 4

235

Large deflection analysis of structural members: C. M. Wang and S. Kitipornchai

1.0

l
,
) ~/ / = = I t / 9

////2n/9
o.s

..I
x

qLa

P=

0.6

----- ~=10
T[ 19

//

0.4

~ ..--""
- "" "" "" "-

0.2

-'"--~

I
0.2

n
Ii

._ 0

0
O.
10
c
UJ

\\

,--~,, ~

LU

1012m~[~
-0.6

0
0
.J

xl

-0.4

p=o

'///

//
~

'l

qL3
~= El

\~8~/9

I
0.4

I
0.6

-4

I
0.8

I
1.0

_61

I
0.2

Figure 6

Large deflections of a cantilever column

in which g = 2s/(xR) and i~ = (Pr2R2)/(4EI). Note that


positive values of/3 correspond to compressive forces
and negative values of B tensile forces. The boundary
conditions are
0(0)=0
dO(O)
--=
d;

(19a)

I
0.6

(19b)

0(1) = w
2

(19c)

Note that the fixed end moment at C is given by


= Mc~cR/(2EI) while the moment at the point of load
application (point B) is furnished by tiw value dO(l)/
d~ =7 = MaxR/(2EI). The moments at B and C are
necessary to maintain a horizontal and a vertical end
tangent, respectively.

IB

0l
D

ylMl~'p

I-

2P

(a)
Figure 7 Circular ring subject to diametrically opposite forces

236

Eng. Struct. 1992, Vol. 14, No 4

1.0

Load-lateral deflection curves for a cantilever column

xT

2P

I
0.8

End Horizontal Deflection, y l / L

Deflected Position, y l L
Figure 5

I
0.4

(b)

Large deflection analysis of structural members: C. M. Wang and S. Kitipornchai


1.8

To determine the deformation of the ring for a given


/3 value, the shooting-optimization method requires

1.2

Min b = I0(1) - ~a'[

(20)

Upon minimization, the unknown moment values at both


ends of the quarter ring are given directly by the values
of ~"and ~( corresponding to the location at g = 0 and 1.
The numerical results obtained are presented in Table 3
while the deformed shapes of the ring in rectangular
coordinates are shown in Figure 8 for various values of

Table 3 Large deflection results for a circular ring subject to


diametrically opposite forces

Px 2R 2
{3 -

- 5.0
-2.5
0
2.5
5.0
7.5

4El

Mc ~rR
~=

3' -

2El

0.7928
1.1006
x12
2.2830
3.2599
4.3643

2El

3.2601
2.4922
~12
0.4604
-0.7918
- 2.0117

Yl IR

x~ IR

0.7751
0.8744
1.0000
1.1451
1.2729
1.3354

1.1976
1.1205
1.0000
0.8102
0.5430
0.2488

t
0.8

I
1.2

p-2

-0.4

~= 4El

-1.8
I
J
-1.8-1.2-0.8-0.4

t
0

t
0.4

1.6

y/R
Figure 8

Large deflections of a circular ring

where 0 is the local slope, g = 2sl(xR) = the coordinate


along
the
arch
centreline,
h = pR2xZI(4EI),
{3 = Hx2R2/(4EI) and H is the horizontal reaction component of the arch. Note that negative values of h and
/3 indicate force directions opposite to those shown in
Figure 9. The boundary conditions are
0(0) = 0
dO(O)
- J"
dg"

(22a)
(22b)

0(1)=-2

Ms lrR

-1.2

(21)

~-0

13=-5.0,

(X:

Example 4: post-buckling and snap-through of semicircular arch subject to a midspan point load

- + X sin O(g) -/3 cos O(g) = 0


dg-'

0.4

-0.8

d20(g)

0.8

/3.

Consider a semi-circular arch of radius R with fixed supports. The arch shown in Figure 9a is subjected to a
midspan point load, 2P, at point B. The symmetrical
large deflection response of the arch under the applied
point load will be investigated using the proposed
shooting-optimization technique.
Because of symmetry, only half of the arch (see
Figure 9b) needs to be analysed. The support condition
at the top end may be treated as fixed but with an
allowance for vertical movement. The exact governing
differential equation for equilibrium of shear forces of
the half arch is

(22c)

Note that ~"= MorRI(2EI) where Mc is the fixed support moment and d0(1)/dg=,y = MnxRI(2EI) is the
unknown moment at the arch apex. In view of the constant half arch span of R, the following geometrical constraint must be imposed, i.e.

t l sin 0 dg = - 2
0

(23)

7r

q.
xt

2P
B

_.

I_

-I H

L
i

(a)
Figure 9

-q

'

(b)

Semi-circular arch subject to a mid-span point load

Eng. Struct. 1992, Vol. 14, No 4 2 3 7


t

Large deflection analysis of structural members: C. M. Wang and S. Kitipornchai


Table 4 Large deflection results for a semicircular arch subject to
a midspan point load

HTr2R 2
-

4El

-12
-6
0

plr2R
X-

4El

-13.7284
-6.7431
0
6.1468
19.6537
10.8731
12.7414
12.8959
7.3537
10.0299
6.1067

6
12
18
24

M c TrR
~ -

1,6

2El

3.1091
2.4084
r/2
0.5622
8.4707
-0.5579
6.4133
-1.4672
3.9602
-1.2935
0.5352

x~ IR

2El

3.6226
2.6994
~/2
0.1494
-6.6668
-1.6016
-6.2966
-3.5158
-6.0808
-5.4418
-6.0061

1.0768
1.0477
1.0000
0.9164
-0.6872
0.7695
-0.6019
0.5377
-0.4677
0.1464
-0.1737

0.8
0.4

-!

~<

0
-0.4
-0.8

(c),(d),lf),lg)

The objective function then becomes


Min ~ =

0(1)-~

sinOdg-

LX

(24)

The unknown moments at the fixed supports and at the


apex of the arch are given by the values of ~" and 7,
respectively. These values and the vertical load parameter are presented in Table 4. The variations of the
horizontal support reaction/3, with respect to the vertical
load, X, are shown in Figure 10. The curve shows a
reversal of the horizontal reaction force during
unloading. The reason for this becomes apparent when
examining the load-deflection curve in Figure 10, where
it is seen that a snap-through phenomenon occurs under
large deformations. The deformed shapes of the arch
under different load levels are shown in Figure 11. The
deformed shapes are plotted for varying horizontal support reaction/3 values. The corresponding points in the
load-deflection curve for deformed shapes (c), (d), (f)
and (g) (see Figure 11) are marked in Figure 10.

Horizontal Reaction, 13= Hn 2 R2/(4EI)


4
8
12
16 20 24 28
i ",\ l
i
i
I
i
i/

2()
18 -

\~

16-

--~12-tU

~.~ 8 -

13curve

\\

(d)
'
\

~/

/-

/-

(g)~-/(g)-

/ / ~ / S c u r v e ~

n." 4 (//"
2P --40 / )(c
.~)-19
-8
H_..t~-~--~--~~
--0.2

Load-deflection and load-horizontal support reaction


curves for a semicircular arch

238

Eng. Struct. 1992, Vol. 14, No,4

-1.2
-1.2
Figure 1 1

I
-0.8

I
-0.4

0
y/R

I
0.4

I
0.8

1.2

Large deflections of a semicircular arch

Example 5: large deflection of diamond-shaped


frame subject to two diagonally opposite
forces
Figure 12a shows a diamond-shaped frame of side
length L with a flexural rigidity E1 for each of its four
members. The joints at the top and bottom corners of the
frame are pinned while those at the two sides are rigid.
The angles between members at the pinned joints are 2or.
The frame is subject to two diagonally opposite forces,
2P, as shown in Figure 12a. The proposed shootingoptimization technique will be used to study the large
deflection response of the frame for a given load value.
This problem has been investigated by Mattiasson ~5,
using the elliptic integrals method.
Because of symmetry, only a quarter of the frame
needs to be analysed. The supporting points of the
quarter frame (Member BC in Figure 12a) can be treated
as fixed at end C with a specified slope of (1r/2) - or, and
as pinned at end B. Ends B and C are free to move vertically and horizontally, respectively (see Figure 12b).
The exact governing differential equation for member
BC is similar to the ring problem (Example 3) and is
given by equation (18). The/3 parameter in equation (18)
becomes/3 = pL2/(EI). Negative values of/3 imply tensile applied forces. The boundary conditions are

71"

0.4
0.8
1.2
1.6 1.8
Vertical Deflection, h= (R-x 1)/R

Figure 10

1.2

Me r R
-y -

0(0) = - - o~
2

(25a)

dP(O)
- ~"
d;

(25b)

dO(l)
- 0
d~

(25c)

Large deflection analysis of structural members: C. M. Wang and S. Kitipornchai

2P

B
P
B
C

Xl

t2P
(a)
F i g u r e 12

(b)

Pinned-fixed diamond-shaped frame subject to diagonally opposite forces

The objective function for the shooting-optimization


method is
M i n = d0(l)
~-

(26)

Table 5 gives numerical values of the moment at joint C,


= McL/(EI), the rectangular coordinates of the ends

the set of first-order differential equations and the errors


at the terminal boundary conditions are minimized by a
direct search optimization technique which does not
require the evaluation of gradients, The proposed technique is powerful and easy to use. It has been applied to
a large class of problems involving large deflections of
a beam, and post-buckling of a column, a ring, an arch
and a frame. The numerical results obtained are

B and C and the slope 01 at corner C for ot = ~r/4 and


various values of/3. The deformed shapes of the frame
(ct = x/4) are shown in Figure 13 for various values of
/3. The solutions obtained are in close agreement with
those obtained by Mattiasson 15.

1.2

~2p I

0.8

Conclusions
A numerical technique combining the shooting method
and optimization algorithms for solving any Nth-order
linear or nonlinear ordinary differential equation has
been proposed. The method uses the accurate fourthorder Runge-Kutta algorithm for integrating forward

0.4

x
Table 5 Large deflection results for a pinned-fixed square
diamond-shaped frame subject to diagonally opposite forces

PL 2
/3 -

-6
-4
-2
0
2
4
6

El

Mc L
~ -

El

-1.8447
-1.4708
-0.9505
0
1.8987
3.3892
4.3786

y l /L

X l IL

-0.4

01

-0.8
0.3074
0.3677
0.4753
x/2/2
0.9493
0.8473
0.7298

0.9341
0.9155
0.8714
~2/2
0.1247
-0.2806
-0.4523

0.1367
0.2124
0.3682
x/4
1.7662
2.3873
2.6693

~= pL2

El
-1.2
I
-1.0-0.8

I
-0.4

I
0.4

I
0.81.0

y/L
F i g u r e 13

Large deflections of diamond-shaped frame

Eng. Struct. 1992, Vol. 14, No 4

239

Large deflection analysis of structural members: C. M. Wang and S. Kitipornchai


presented in a tabulated form which should provide
useful information to researchers who wish to check
finite element software that incorporates a large deflection analysis.
The proposed technique has the following advantages
over the finite element method: (i) it does not require
matrix inversion or manipulation and hence the computational effort and memory space is reduced, allowing
the problem to be solved using a small personal computer; (ii) when solving stability or vibration problems,
it generates the eigenfunctions (buckling and vibration
modes) simultaneously with the eigenvalues (buckling
loads and frequencies) through the solving of the differential equation whereas the eigenvalues and eigenfunctions are obtained in a separate process in the finite
element method; and (iii) it does not require any
discretization and thus can provide a more accurate solution, especially for nonprismatic structural members.
However, the finite element method is more versatile in
that it can be used to analyse large complex structures
consisting of many structural members, whereas the
proposed technique can only be used where there are
ordinary differential equations available to describe the
problem.

Acknowledgments
This work was conducted in the Department of Civil
Engineering at The University of Queensland, Australia
while the first author was on sabbatical leave from the
National University of Singapore. It forms part of the
project 'Stability of beams and beam-columns' supported by the Australian Research Council (ARC) under
Project Grant no 834. The authors wish to thank Mr
Warren H. Traves of Gutteridge Haskins and Davey Pry
Ltd for proof-reading the manuscript.

240

Eng. Struct. 1992, Vol. 14, No 4

References
1 Kitipornchai, S. and Chan, S.L. "Stability and nonlinear finite clement analysis of thin-walled structures'. Ch 4 Finite element applica
tions to thin-walled structures (Ed J. W. Bulll Elsevier Applied
Science, London 1990
2 AI-Bermani, F. G. A. and Kitipornchai, S. Nonlinear analysis of thinwalled structures using least element/member', J. Struct. Eng., ASCE
1990, 116(1), 215-234
3 Richardson, L. F. 'The approximate arithmetical solution by finite
differences of physical problems involving differential equations with
an application to the stresses in a masonry dam'. Phil. Trans. Roy.
Soc. London, Ser. A, 191 I, 210, 307-357
4 Salvadori, M. G. 'Numerical computation of buckling loads by finite
differences', Trans. ASCE, 1951, 116, 590
5 Salvadori, M.G. "Numerical methods in engineering" (2nd edn)
Prentice-Hall, Englewood Cliffs, NJ, 1961
6 Brown, P. T. and Trahair, N. S. 'Finite integral solution of differential equations', Cir. Eng. Trans., Inst. Eng. Australia, 1968,
CEI0, (2), 193-196
7 Fraser, R. A., Jones, W.P. and Skan, S. W. "Approximations to
functions and to the solutions of differential equation', Tech. Rep.
Aero. Res. Com., Rep. and Mem. No. 1799, 1937, 517-549
8 Duncan, W. J. "Galerkin's method in mechanics and differential
equations', Tech. Rep. Aero. Res. Com. Rep. and Mem. No. 1798,
1937, 484-515
9 Frisch-Fay, R. 'Flexible bars" Butterworths, London, 1962
10 Keller, H. B. "Numerical methods for two-point boundary-value
problems', Blaisdell Publishing, USA, 1968
11 Na, T. Y. "Computational method~ in engineering boundary value
problems" Academic Press, New York, 1979
12 Kreyszig, E. 'Advanced engineering mathematics' f5th edn) John
Wiley New York, 1983
13 Goh, C. J. and Wang, C. M. 'Generalised shooting method for elastic
stability analysis and optimization of structural members', Comp.
Struct., 1991, 38(1), 73-81
14 Reklaitis, G. V., Ravindran, A. and Ragsdell, K. M. "Engineering
optimization: methods and applications' John Wiley New York, 1983
15 Mattiasson, K. 'Numerical results from large deflection beam and
frame problems analysed by means of elliptic integral', Int. J. Num.
Meth. Eng., 1981, 17(I), 145-153
16 Timoshenko, S. P. and Gere, J. M. "Theory of elastic" stability'"
McGraw-Hill, New York, 1961
17 Wang, C. Y. 'Buckling and postbuckling of heavy columns', J. Eng.
Mech., 1987, 113(8), 1229-1233

S-ar putea să vă placă și