1 s2.0 S0045794996004002 Main

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 4

Computers & Srrucrures Vol. 63. No. 5, pp.

1033-1036, 1997
0 1997 Elsevier Sciena Ltd. All rights reserved
Printed in Great Britain
PII: SOO45-7949(96)00400-Z OLW-7949197 $17.00 + 0.00

TECHNICAL NOTE

A PROPOSAL FOR OPTIMUM STRUCTURAL DESIGN WITH


THE LARGEST BUCKLING LOAD

R. Takagi,t M. Maeda,$ S. J. Duan$tt and K. Nakagawag


tThe Takigami Steel Construction Co., Ltd, 2-1 Kiyokawacho, Nakagawa-ku, Nagaya 454, Japan
$Central Japan Railway Co., Ltd, l-6-5 Marunouchi, Chiyoda-ku, Tokyo, 100 Japan
§Department of Civil Engineering, Gifu University, l-1 Yanagido, Gifu, 501-l 1 Japan

(Received 3 1 March 1995)

Abstract-A design technique for the frame structure with the largest buckling load will be proposed in
this paper. The reciprocal sum of the buckling loads of the structure is minimized with a constraint that
the volume of the structural materials should be kept constant. Such an optimal method also means that
the structure will approximately take the largest buckling load. The structural problem is formulated by
using the m,atrix displacement method and the solution is given by an iterative scheme, while it is not
necessary to compute the buckling eigenvalues. Some numerical examples are included. 0 1997 Elsevier
Science Ltd.

danger factors may be designed by the proposed method,


1. INTBODUCI’ION while only rl (or 51) is considered in the present design
Until now, many studies have been done on the standard.
optimization problem of finding the minimum volume (or Some plane structures are taken as numerical examples.
weight) design of structure satisfying stability requirements, The cross-sectional area of each member is chosen as the
variable and all loads are considered acting at the nodes.
such as in Ref. [I]. For linear stability, the buckling
The optimization problem is solved by the iterative method,
eigenvalues must be calculated and often only the
fundamental buckling eigenvalues are considered on the step by step.
structural optimum designs.
Differently to the above concepts, we have proposed
a method to minimize the sum of the squares of the 2. BASIC FORMULATION
natural periods of vibration in dynamics [2-4] or to
The plane framed structures are considered in this
minimize the sum of the squares of the combined stresses of
paper. The relative tensional rigidity EA,/L,, and flexural
the applied loads in statics [5] with one constriant of
rigidity EZo/Lu of the members are chosen as design
constant value of the weight or volume of the structural
variables, where EA,, EZq are the tensional and the flexural
materials.
rigidity of the ijth frame member, and L, is the 0th member
This paper is concerned with the structural optimization
length. As it is well known, the equilibrium equation of
of the buckling load problem. Following the authors’
static forces is given by the matrix displacement method, as
previous studies [2-Q we will find the minimum reciprocal
follows:
sum of the buckling loads with constant values of the weight
of the structural materials. In other words, an optimum
structure will be designed which can take the largest su = P, (1)
buckling loads. As the linear stability problem, the
governing equation is established by the matrix displace- where, II is the unknown nodal displacement vector, P
ment method directly based on the small rotations the nodal load vector, and S the general elastic stiffness
assumption. The structural stiffness matrix S is expressed in matrix.
the form of a triple square matrix product, even if the Following the authors’ previous paper [4,5], S can be
structure is either statically determinate or indetenninate. expressed in the form of the triple square matrix product
The buckling eigenvalues need not be computed, because the for both statically determinate and indeterminate structures
reciprocal sum of their values can easily be obtained for as
the present objective. Therefore, we can say that not only
the fundamental, but also all of the buckling modes are S = C?DC, (2)
considered in our study.
If the applied load at the structure is P, and the in which, D is a diagonal matrix that represents the stiffness
fundamental buckling load is P,, = &P, then 5, is called properties of the members, and C the geometrical
the safety factor of the sturcture and then rl = l/C is connections of the members. It should be noticed that C
equivalent to the danger factor of the fundamental buckling must be an extended square matrix if the considered
mode. So that a structure with the minimum sum of the structure is indeterminate [4, 51. Therefore, the flexibility
matrix S-’ can be given by the product of the inverse
tt Visiting Researcln Fellow, from North China Institute element matrices as follows:
of Water Conservancy and Hydroelectric Power, Handan,
Hebei, 056021 China. S-1 = (CQC)-1 = C-‘D-C-1. (3)

1033
1034 Technical Note

I I I I I Table 1. Example l-P,& = rP


AP
I I I I I
A Initial (l&J Optimized (l/t,) Ratio (&&,p)
-
L
t i
Z115k
l/t;1
4.019
2.956
x
x
10-S
lO-5
3.641
2.480
x
x
10-S
lo-’
0.899
0.838
Fig. 1. The analysed flexible column. l/C2 8.716 x lO-6 2.841 x 1O-6 0.423
lit 2.596 x lO-6 2.839 x lO-6 1.093
Iterative cycle-6
[5 x 10-51

r-------7 3. OPTIMIZATION PROCEDURE


By Euler (N = 00) In this study, the cross-sectional area Ak of each member
is defined as the independent design variable, while b is
approximately expressed by

Ix = BA:, (‘3)
By Authors
in which /l is a parameter which depends on the
cross-sectional shape.
The optimum values are obtained by the Lagrange
multiplier method. The present objective is to minimize
equation (7) under the constraint of constant value of the
structural materials.

ZAkLk = V = Const. (9)


01’ 1
10
I I
20
I I
30
I I
40 Then the Lagrangian
N
F = Z + c(ZAkLk - V), (10)
Fig. 2. Comparison of El/Z% by the Euler theory and the
present method. presents a calculus variation in mathematics, in which L is
the Lagrange multiplier. So that, the optimality criterion is
obtained by
Consider the system is linear, the buckling eigenvalues
Pc,k = &P can be expressed in equation (1): g=o,$0.
Su - &Hu = 0, (4)
Then the largest buckling load is approximately obtained
in which, H is the geometric matrix, and & and buckling and a stability structure should be designed.
mode u = [&, &. $3, . ] are the kth eigenvalue and For the member with both rigid ends, the optimum
eigenvector, respectively. That is, cross-sectional area Ah can be approached by the following
iterative formula

S[$V, +2,+3,. . .I = MI,42.&. .I . (5)


(Ac),+ I = J& k,k + (a, $,)“};, (12)

Hence, in which, the coefficients ark, a% and aa can be determined


from C and H, and assumed to non-variables in the different
procedure; v is the iteration number.
If the member with one rigid end and one hinged end, the
(6) iterative formula is

Therefore, the reciprocal sum of the eigenvalues can be (A*),+1 =&FM + ‘(‘$;)L:J:“, (13)
obtained as
While for the member hinged at both ends,
Z = X(1/&) = Trace(S-‘H) = Trace(C-ID-V-‘H). (7)

This is equivalent to the reciprocal sum of the elastic (14)


buckling loads of the structure.

--* : Initial cross-section


- : Optimizod crosr-section
Fig. 3. Initial and optimized cross-sectional areas for example 1.
Technical Note 1035

P=lt P=lt Table 3. Example 2.2-P,* = (P


-4m Initial (l/&J Optimized (l/r,) Ratio (&/&,,,)

1%
2.663 x lo-’ 1.482 x lO-4 0.556
9.625 x IO-’ 5.950 x 10-S 0.618
-!-I- 6.587 x 1O-5 3.594 x 10-S 0.545
)8m 5.834 x 10-S 2.484 x 1O-5 0.425
Iterative cycle-3.

I so that,
’ T (a)

Initial
+--&XT& (16)
Optimized
The computed results by equation (16) or (7) respectively,
are shown in Fig. 2. Where, L = 100 m, E = 2.1 x 10’ t m-2,
A = 0.9 m* and I = 1.578 m4 are assumed.
From Fig. 2, we can see that the results by the present
method approaches that by Euler formula, as the number
of the elements increases. So equation (7) can be used to
compute the reciprocal sum of the buckling loads
approximately.

5. STRUCTURAL OPTIMUM EXAMPLES

In the following numerical examples, all of the


cross-sectional areas are assumed, or chosen from the
established structures.
Initial Optimized
5.1. Example 1
A simply supported column is divided into five elements
and then with six joints (reference to Fig. 1). In Fig. 3, the
initial relative cross-sectional areas are shown by dashed
lines, while the optimized relative cross-sectional areas by
solid lines. The reciprocal sum of the buckling loads and the
reciprocal fundamental buckling loads for the initial and
the optimized designs and their comparisons are presented
in Table 1, while the results for high order buckling modes
are omitted.

5.2. Example 2
Fig. 4. Initial and ‘optimized cross-sectional areas for
example 2: (a) model; (b) whole buckling; (c) whole and The analysed plane truss is shown in Fig. 4(a). Only
localized buckling. vertical loads are acted at the hinged joints in this example.
Fig. 4(b) and Table 2 are for the whole buckling analysis,
where only the hinged joint displacements are considered.
In this study, the whole and the localized buckling
4. COMPARISON WITH EULER LOAD problem can be taken into account simultaneously, if the
nodes are also set up in the middle of each member (see the
Based on the above algorithm, a computing program is previous section and Example I). For the present example,
written. In this section, we will compare the reciprocal sum the numerical results by such method are given in Fig. 4(c)
of the buckling loads by the Euler theory, with those in the and Table 3, where every member is divided into three
present study; while the optimal numerical examples will be elements.
shown in the next section.
Consider a flexible column hinged at both ends (see 5.3. Example 3
Fig. l), the Euler load is Fig. 5(a) shows a designed framed structure and Fig. 5(b)
its relative cross-sections. The optimized relative cross-sec-
tions are shown in Fig. 5(c), in which both the whole and
the localized buckling are considered. Their comparisons are
demonstrated in Table 4. It should be noticed that the

Table 2. Example 2.1-Pc,k = <P Table 4. Example 3-P,* = 0’


Initial (l/&i) Optimized (I/&) Ratio (I;i&,,) Initial (l/&J Optimized (l/L) Ratio (ti&m)
1.689 x lo-” 9.452 x 1O-5 0.559 zl/tk 1.128 x 10-Z 9.333 x IO-’ 0.721
9.651 x lo-’ 5.991 x 10-S 0.620 l/r1 5.400 x 10-1 4.319 x 10-l 0.799
, _- 6.603 x lo-: 2.401 x 10-5 0.363 1152 3.455 x 10-s 3.020 x lo-’ 0.874
l/t;3 6.365 x lo-” 1.051 x 10-s 1.651 114 1.223 x lo-’ 1.072 x 10-l 0.932
Iterative cycle-+. Iterative cycle-7.
1036 Technical Note

+ 18.0 m --+ 6. CONCLUSIONS


582 t 214 t 690 t In the present paper, a new optimal design method for
frame structures is proposed, based on the structural
stability. The objective is to minimize the reciprocal sum of
the buckling loads, subjected to the constraint of constant
volume of the structural materials. So, this study may be
said to be a link in the chain of the authors’ past studies,
Refs [4, 51, respectively, for dynamics or statics, while the
present study is for the stability problem. All of the
structural analyses are based on homogeneous, linearly
elastic ones, so the dynamic [4] or the stability problem is
equivalent to solve the eigenvalue problem, but the
eigenvalues need not be computed in the proposed otpimum
methods.
In the stability problem, lateral load subjection was
generally assumed, and the whole buckling was analysed [6].
There is no difference between this technique and other
methods in the case of the whole buckling problem analysis
without lateral loads. In this study, both the whole and the
localized buckling problems can be, approximately, taken
into account simultaneously. The reason is that we assumed
that the cross-sectional rotations by bending moments are
very small and so are omitted, while only the nodal
displacements are considered for the critical equilibrium
position under buckling loads. Therefore, as a numerical
method, enough nodes must be set up for the member if its
localized buckling is to be considered. In other words, if the

El El
initial deflection or bending moment by the applied loads is
comparatively large and must be considered, then this is not
an eigenvalue problem and the present method is not
0’) suitable.
It may be concluded that the present method is more
practical in engineering structural designs, because all of the
buckling modes can be taken into account, especially for the
case <I and 52, &, . . which have very close values (see
Example 3). It should also be noticed that the iterative cycle
must increase in computing for convergence to the solution
in such cases.
Future studies will extend the present method to the
non-elastic buckling problem and systematically estimate
the optimized structure by statics [S], dynamics [4] and
stability (present).

REFERENCES

1. G. M. Barsan, Optimal design of planar frames based


on structural performance criteria. Comput. Strucr.
53(6), 1395-1400 (1994).
2. K. Nakagawa, Studies on eigenvalues in beam
vibration. Proc. JSCE 150, l-8 (1968) (in Japanese).
3. K. Nakagawa, Design of a cantilever beam section
Fig. 5. Initial and optimized cross-sectional areas for under minimizing the square sum of natural periods.
example 3: (a) model; (b) initial designed cross-section; (c) Proc. JSCE 169, 15-20 (1969) (in Japanese).
optimized cross-section. 4. K. Nakagawa, K. Andoh and S. J. Duan, A design
method of structures to minimize the square sum of
natural periods of vibration. Comput. Struct. 45(3),
587-591 (1992).
magnitudes of the first terms (l/t,, l/&, l/t, . . .) are very 5. R. Takaki, M. Maeda, K. Nakagawa and S. J. Duan,
close individually. This means that the second, the third and A proposal of design technique for optimum section
other high buckling modes have a great effect in this with average stresses throughout frame structures.
structure on the buckling. This shows it may not be enough Comput. Struct. 53(2), 233-240 (1994).
to design a structure if only the first buckling mode is 6. C. H. Norris and J. B. Wilbur, Elementary Structural
considered. Analysis. McGraw-Hill, New York (1960).

You might also like