Topology optimization of beam cross sections

Yoon Young Kim*, Tae Soo Kim

School of Mechanical and Aerospace Engineering, Seoul National University, San 56-1, Kwanak-Gu, Shinlim-Dong, Seoul 151-742,
South Korea
Received 6 August 1998; in revised form 14 December 1998
Abstract
Perhaps, this paper reports the rst successful applications of the topology optimization in the design of (thin-
walled) beam sections. In particular, topologically dierent thin-walled beam cross sections can be obtained by the
present approach, which is very useful in identifying the direction and location of stieners. In formulating the
topology optimization problems, a simple power law is used for the relation between the density of an element with
a hole and the mechanical properties of the element. The sensitivity of the torsional rigidity is obtained by
developing a nite element model of a St. Venant torsion problem, and the Euler beam theory is used for the
sensitivity analysis of the bending rigidities. # 1999 Elsevier Science Ltd. All rights reserved.
Keywords: Topology optimization; Cross-section; Torsion; Bending
1. Introduction
The beam cross section optimization problems have been very important as beams are widely used as
ecient load-carrying structural members. Earlier investigations were based on analytic approaches
(Banichuk, 1976; Banichuk and Karihaloo, 1976; Parbery and Karihaloo, 1977), but the optimization
problems in more complicated geometries need to be solved numerically.
The section shape optimization based on the nite element formulation was carried out by Dems
(1980) and Na et al. (1983). As a more ecient alternative method for section shape optimization, Mota
Soares et al. (1984) and Gracia and Doblare (1988) used the boundary element method. Schramm and
Pilkey (1993) developed a shape optimization technique using direct integration and B-splines. More
literature on section optimization problems can be found in the references cited above. However, the
aforementioned techniques are not applicable in obtaining (thin-walled) beam cross sections that are
topologically dierent from original cross sections.
International Journal of Solids and Structures 37 (2000) 477493
0020-7683/00/$ - see front matter # 1999 Elsevier Science Ltd. All rights reserved.
PII: S0020- 7683( 99) 00015- 3
www.elsevier.com/locate/ijsolstr
* Corresponding author. Fax: +82-2-883-1513.
E-mail address: yykim@plaza.snu.ac.kr (Y.Y. Kim)
One important problem in beam section design is to nd the location and direction of stieners,
particularly in thin-walled beams. As the introduction of a stiener in a closed beam may result in a
topologically dierent cross section from the original cross section, the existing section shape
optimization cannot be used. The purpose of this paper is to formulate a section topology optimization
technique and apply it to various practical problems. In this work, the objective function is taken as a
weighted sum of bending and torsional rigidities and a topology optimization technique is proposed to
nd the optimal cross section conguration.
The material homogenization concept (Bendse and Kikuchi (1988)) has been introduced to overcome
the ill-posedness of optimal material layout problems. With this approach which converts the
optimization problems into sizing optimization problems, the optimal material layout problems can be
fairly relaxed. The present topology optimization technique is based on the technique and formulation
developed for elastic bodies, not directly for beam structures (Bendse and Kikuchi, 1988; Guedes and
Kikuchi, 1990; Suzuki and Kikuchi, 1991; Olho et al., 1991; Jog et al., 1994). A complete treatment of
this subject and a number of related references can be found in Bendse (1995). In this optimization
problem setting, the density of each element may be used as the design variable. A brief summary of the
topology optimization that is necessary for the present analysis will be given in the next section.
The beam bending rigidity appearing in the objective function is based on the classical beam theory.
The torsional rigidity is calculated by the nite element analysis of the St. Venant torsion problem for
which the Prandtl stress function formulation is employed. To simplify the form of the objective
function, an inequality relation stating that the torsional rigidity is always smaller than the mean
bending rigidity is derived. The expression for the sensitivity of the torsional rigidity with respect to the
design variable is explicitly obtained. Several cross sections including some practical ones are
investigated as numerical examples. In particular, the direction and the location of a stiener in a thin-
walled closed beam section can be determined by applying the present topology optimization technique.
The successful applications in this class of problems may be a major contribution of the present work.
2. Review of topology optimization using material density
In this section, a brief review of the topology optimization method using the material density as the
design variable will be given. Fig. 1 illustrates an overview for a simple version of the topology
optimization procedure. The design domain is uniformly discretized by nite elements having internal
holes. The size of the hole governs the element or cell density and thus the homogenized mechanical
Fig. 1. Overview of the topology optimization procedure.
Y.Y. Kim, T.S. Kim/ International Journal of Solids and Structures 37 (2000) 477493 478
properties of the element. The density r appearing in Fig. 1 is dened as
r 1
_
b
a
_
2
, 1
where a and b are the side lengths of the square element and hole, respectively.
In the homogenization-based topology optimization approach, the optimal element density
distribution is obtained rst. Then the obtained result is postprocessed to determine the optimal
conguration for the given problem. Depending on the value of the element density, the element is
regarded either full or void.
The rigorous homogenization technique (see e.g., Guedes and Kikuchi, 1990 and Bendse, 1995) can
be used to nd the relation between the density r and the material properties. However, a simple
articial relation between the density and the material properties can be also used in the topology
optimization problems. Furthermore, it is assumed in this work that only Young's modulus E varies as
the function of the density as
E Er E
0
r
n
n 2, 0<rR1, 2
where E
0
is the Young's modulus of the material of the original element without any hole. Poisson's
ratio n and the area A are assumed to be independent of the density r. Therefore only the element
stiness changes as the hole size varies. For more rigorous treatments of this subject, see Bendse
(1995). Fig. 2 illustrates graphically the present assumptions used in relating the density and the
mechanical and geometrical properties.
The material and geometrical properties expressed as the functions of the element density will be used
in determining the bending and torsional stiness of a beam. The following sections derive the
expressions for the bending and torsional rigidities that are needed for the topology optimization of
beam cross sections.
3. Bending and torsional rigidities
Fig. 3 shows the general cross section of a beam along with a discretized model of it. The beam is
subject to bending and twisting moments. The positive moment directions and the coordinate systems
are also shown.
The bending rigidities, D
x
, D
y
and D
xy
, may be found from the EulerBernouilli beam theory:
D
x

_
A
Ey
2
dA, D
y

_
A
Ex
2
dA and D
xy

_
A
Exy dA: 3
In Eq. (3), the Cartesian coordinate system (x, y) that has its origin at a centroid is dened as
x X X
c
and y Y Y
c
, 4
where (X, Y) is an arbitrarily located Cartesian coordinate system and X
c
and Y
c
are dened as
X
c

_
A
EX dA
_
A
E dA
and Y
c

_
A
EY dA
_
A
E dA
: 5
Y.Y. Kim, T.S. Kim/ International Journal of Solids and Structures 37 (2000) 477493 479
The bending moments (M
X
, M
Y
) and the curvatures (k
x
, k
y
) are related as
_
M
x
M
y
_

_
D
x
D
xy
D
xy
D
y
__
k
x
k
y
_
: 6
To carry out the topology optimization, we consider the beam cross section discretized with two-
dimensional nite elements. Consequently, Eq. (3) must be rewritten as
D
x

N
e
e1
E
_
A
e
y
2
dA, D
y

N
e
e1
E
_
A
e
x
2
dA and D
xy

N
e
e1
E
_
A
e
xy dA, 7
where Young's modulus E is now regarded as the function of the element density r
e
as in Eq. (2). The
location of the centroid is rewritten as
Fig. 2. The variation of the mechanical and geometrical properties as the function of the element density r: (a) r: Density of el-
ement; (b) E: Young's modulus; (c) n: Poisson's ratio; (d) A(V): Geometric area (volume).
Y.Y. Kim, T.S. Kim/ International Journal of Solids and Structures 37 (2000) 477493 480
X
c

N
e
e1
E
_
A
e
X dA

N
e
e1
E
_
A
e
dA
and Y
c

N
e
e1
E
_
A
e
Y dA

N
e
e1
E
_
A
e
dA
, 8
where N
e
is the total number of nite elements used to discretize the design domain.
Once the bending rigidities D
x
, D
y
and D
xy
are determined, it is straightforward to nd the maximum
and minimum bending rigidities along the principal axes (Crandall et al., 1978):
D
max
D
mean
R
M
and D
min
D
mean
R
M
, 9
where D
mean
is the mean bending rigidity dened as
D
mean

D
x
D
y
2
10
and R
M
is given by
R
M

D
x
D
y
2
D
2
xy
_
: 11
Unlike the bending rigidity, the torsional rigidity for general cross sections must be obtained by
solving a two-dimensional boundary value problem (see Timoshenko and Goodier, 1970; Sokolniko,
1955). The torsion problem may be formulated in terms of either the warping function or the Prandtl
stress function, but the nite element analysis based on the Prandtl stress function formulation will be
employed here as it gives a simpler expression for the torsional rigidity. Recent references on the
solution of torsion problems may be found in Kim and Yoon (1997).
Fig. 3. (a) A beam under bending and torsional moment; (b) A discretized model of the cross section.
Y.Y. Kim, T.S. Kim/ International Journal of Solids and Structures 37 (2000) 477493 481
In the Prandtl stress formulation, the shear stress components are expressed as the derivatives of the
stress function f(x,y):
s
zx

@fx,y
@y
and s
zy

@fx,y
@x
, 12
where the stress function f(x,y ) must satisfy the following dierential equation
@
@x
_
1
G
@f
@x
_

@
@y
_
1
G
@f
@y
_
2y: 13
In Eq. (13), y is the twist rate and G(x, y) is the shear modulus. The traction-free beam wall
condition along the boundary curve C may be given in terms of f:
df
ds
0 or f constant 0 for simply connected regions: 14
Once the solution to the stress function is found, the torsional rigidity is determined as (for simply
connected regions)
D
z
M
z
2
_
A
f dA with y 1: 15
The weak form of the Prandtl stress function formulation may be written as
_
A
1
G
ru rf dA 2y
_
A
u dA, 16
where u is an arbitrary function which lies in the admissible space (u=0 on C). For the nite element
analysis, the stress function f is discretized, which may be written as
f NF: 17
In Eq. (17), N is the displacement interpolation matrix and F F represents the column vector consisting
of the nodal values of the stress function.
If a symbol B is introduced to designate the corresponding strain interpolation matrix, one can use
Eq. (16) to obtain the following system of equations in the element level:
K
e
FFF
e
f
e
, 18
where the element stiness matrix K
e
and the load vector f
e
are
K
e

_
A
e
B
T
e
D
e
B
e
dA 19
and
f
e
2y
_
A
e
N
T
e
dA: 20
In Eq. (19), D
e
is the matrix dening the constitutive relation
Y.Y. Kim, T.S. Kim/ International Journal of Solids and Structures 37 (2000) 477493 482
D
e

_
_
_
_
1
G
0
0
1
G
_

_
: 21
As for Young's modulus, the shear modulus is assumed to vary as
G Gr
e
G
0
r
n
e

E
0
11 n
r
n
e
n 2: 22
Assembling Eq. (18) yields the nal system equation:
KFFF f, 23
where
K

N
e
e1
K
e
, 24a
FFF

N
e
e1
FFF
e
24b
and
f

N
e
e1
f
e
: 24c
The torsional rigidity D
z
is now given by
D
z
2

N
e
e1
_
A
e
N
e
FFF
e
dA 2

N
e
e1
_
A
e
N
e
dA

N
e
e1
FFF
e
: 25
Using Eqs. (24b) and (24c), the nal expression for D
z
is obtained as
D
z
f
T
FFF with y 1: 26
4. Optimization problem formulation
4.1. Objective function
In most beam section design problems, larger bending and torsional rigidities for a given mass
constraint are desired. Unless specic values for the rigidities are specied, the minimization of R
M
(half
of the dierence between the maximum and minimum principal values) and the maximization of D
mean
(the mean value of the two principal values) may be sought for (see Eq. (9)). At the same time, the
torsional rigidity D
z
needs to be maximized.
Taking the element densities as the design variables, the objective function to minimize may be written
as
Y.Y. Kim, T.S. Kim/ International Journal of Solids and Structures 37 (2000) 477493 483
f w
J
D
z
w
M
D
mean
w
I
R
M
c
p

N
e
e1
r
e
1 r
e
27
and
0<r
e
R1, e 1,2 . . . ,N
e
, 28
where the w's represent weighting factors and the last term in Eq. (27) is a penalty function with a
penalty constant c
p
. Since the optimal shape of a cross section is extracted by the density distribution, it
is always desirable to push the design variables towards the lower and upper limits 0 and 1 using the
penalty function.
Instead of using Eq. (27) as the objective function, we propose to use a simpler function for the
minimization problem:
f w
J
D
z
w
I
R
M
c
p

N
e
e1
r
e
1 r
e
: 29
The use of this function is justied because there exists an inequality relation between the mean
bending rigidity D
mean
and the torsional rigidity D
z
such that
D
mean
> D
z
for n > 0: 30
The proof of Eq. (30) is given in Appendix A.
4.2. Constraint
A typical constraint in structure optimization problems is a mass constraint. If the maximum
allowable mass is denoted by M
0
, the constraint equation is written simply as

N
e
e1
_
A
e
r
e
dA M
0
R0: 31
4.3. Sensitivity analysis
The sensitivities of the object function and the constraint equation need to be calculated during the
optimization process. Therefore, the formula needed for the sensitivity calculations are derived explicitly
below.
The sensitivity of the objective function in Eq. (29) with respect to the design variable r
e
is simply
@f
@r
e
w
J
@D
z
@r
e
w
I
@R
M
@r
e
c
p
1 2r
e
: 32
The sensitivity of the quantity R
M
in Eq. (32) is given by utilizing Eq. (11).
@R
M
@r
e

_
D
x
D
y

_
@D
x
@r
e

@D
y
@r
e
_
4D
xy
@D
xy
@r
e
_
2

D
x
D
y

2
4D
2
xy
_ , 33
Y.Y. Kim, T.S. Kim/ International Journal of Solids and Structures 37 (2000) 477493 484
where the sensitivity of the bending rigidities can be obtained by utilizing the denitions given in Eq.
(7):
@D
x
@r
e
E
0
_
A
e
y
2
dA

N
e
e1
E
_
A
e
2y
@y
@r
e
dA,
@D
y
@r
e
E
0
_
A
e
x
2
dA

N
e
e1
E
_
A
e
2x
@x
@r
e
dA
and
@D
xy
@r
e
E
0
_
A
e
xy dA

N
e
e1
E
_
A
e
_
x
@y
@r
e
y
@x
@r
e
_
dA: 34
In Eq. (34), ( )' denotes dierentiation with respect to r
e
. The sensitivities of x and y should not be
neglected in Eq. (34). Using Eqs. (4) and (8),
@x
@r
e

@X
c
@r
e

E
0
_
A
e
dA

N
e
e1
E
_
A
e
X dA E
0
_
A
e
X dA

N
e
e1
E
_
A
e
dA
_

N
e
e1
E
_
A
e
dA
_
2
and
@y
@r
e

@Y
c
@r
e

E
0
_
A
e
dA

N
e
e1
E
_
A
e
Y dA E
0
_
A
e
Y dA

N
e
e1
E
_
A
e
dA
_

N
e
e1
E
_
A
e
dA
_
2
: 35
The sensitivity of the torsional rigidity in Eq. (32) can be found by utilizing Eq. (26).
@D
z
@r
e

@f
T
FFF
@r
e
f
T
@FFF
@r
e
FFF
T
K
@FFF
@r
e
: 36
To simplify Eq. (36), Eq. (23) is dierentiated with respect to the design variable and premultiplied by
F F
T
:
FFF
T
K
@FFF
@r
e
FFF
T
@K
@r
e
FFF: 37
Substituting Eq. (37) into Eq. (36) yields
@D
z
@r
e
FFF
T
@K
@r
e
FFF FFF
T
e
@K
e
@r
e
FFF
e

G
0
G
FFF
T
e
K
e
FFF
e
: 38
Note that the sensitivity of the torsional rigidity with respect to the density of the e-th nite element
can be computed within the element level.
Y.Y. Kim, T.S. Kim/ International Journal of Solids and Structures 37 (2000) 477493 485
The sensitivity of the constraint in Eq. (31) is straightforward to obtain:
@
@r
e
_

N
e
e1
_
A
e
r
e
dA M
0
_

@
@r
e
_
A
e
r
e
dA
_
A
e
dA: 39
From the result given by Eq. (39), the sensitivity of the constraint equation with respect to the density
of the e-th element is nothing but the element area.
For the numerical analysis of the present optimization problem, the method of feasible direction (see
Haftka and Gu rdal, 1992 and Vanderplaats, 1984a) is used. (ADS (Vanderplaats, 1984b) is used for the
actual numerical work). The values of the design variables are updated utilizing the sensitivity results
given above. When converging results are obtained, the design variables take on the values close to the
limit values. This is due to the penalty term added in the denition of the objective function.
Subsequently, clearly identiable section shapes may be obtained.
5. Numerical examples
As the application examples of the present topology optimization in the beam section design, we
consider three cases. The rst one is a simple case in which a section bounded by a square prole is to
be optimized. (For all the optimization problems discussed in this section, the objective function f in Eq.
(29) will be minimized). This case serves to check the validity of the present analysis. The second case
deals with the reinforcement of a thin-walled cross section, and the resulting reinforced cross section has
the same topology as the original one. The third case also deals with thin-walled beam section
reinforcement, but the resulting cross sections can be topologically dierent from the original one. This
is the case that most other optimization techniques including the shape optimization technique cannot
handle. Perhaps, this is the example for which the present topology optimization contributes in a unique
way to the thin-walled beam section design in comparison with existing approaches.
5.1. Verication problem
To verify the validity of the present method, a simple design optimization problem is considered. Fig.
4 shows the domain for the section design. The problem is to nd the section shape to minimize the
function f in Eq. (29) subject to dierent mass constraints. In this case, we take w
J
=1 and w
I
=0 as
symmetric cross sections will be sought for. The optimal section congurations obtained from the
present analysis are shown in Fig. 5.
As expected, the strict limitation on the total mass yields a thin-walled section close to a hollow
circular cross section. To obtain the conguration shown in Fig. 5, the initial design variables are taken
to be constant with r
e
=0.1. During the iterations of the design optimization, the design variables are
pushed towards to the values close to either 0 or 1.
5.2. Thin-walled section reinforcement: resulting in variable beam thickness
As the second case, the problem of the reinforcement of a thin-walled beam section is considered. The
thin-walled beam section is shown in Fig. 6. The goal is to nd a cross section with the maximum
rigidity subject to a mass constraint. Two results obtained from the present topology optimization
technique are shown in Fig. 7. The mass constraint ratio is set equal to 20% of the design domain.
Depending on the values of the weighting factors, somewhat dierently reinforced beam sections are
Y.Y. Kim, T.S. Kim/ International Journal of Solids and Structures 37 (2000) 477493 486
Fig. 4. Section design domain (dimensionless units are used for convenience).
Fig. 5. Optimized cross sections: (a) 30% mass constraint; (b) 50% mass constraint; (c) 60% mass constraint.
Fig. 6. The design domain for a thin-walled cross section.
Y.Y. Kim, T.S. Kim/ International Journal of Solids and Structures 37 (2000) 477493 487
obtained. The convergence history is shown in Fig. 8, and the nal values of the bending and torsional
rigidities listed in Table 1 show the sectional characteristics of the optimized sections.
From the optimized section shapes shown in Fig. 7, one may estimate the optimal distribution of the
cross section wall thickness. The present observation is not only interesting but also very useful in a
wide class of thin-walled beam section design problems. As shall be seen in the next example, the
topology optimization technique also gives a reinforced beam section that is topologically dierent from
the original section. Therefore, a unied treatment of (thin-walled) section design may be carried out
within the same frame of the section topology optimization.
Fig. 7. Optimized cross sections (20% mass constraint): (a) w
J
=1.0 and w
I
=0.0; (b) w
J
=1.0 and w
I
=0.02.
Fig. 8. Convergence history for the results shown in Fig. 7: (a) w
J
=1.0 and w
I
=0.0; (b) w
J
=1.0 and w
I
=0.02.
Y.Y. Kim, T.S. Kim/ International Journal of Solids and Structures 37 (2000) 477493 488
5.3. Thin-walled section reinforcement: resulting in dierent topology
There are some instances in which the external prole of a thin-walled beam cross section cannot be
changed although the section rigidity needs to be increased. For instance, when a beam needs to be
assembled to other structural elements which are already manufactured or whose proles are dicult to
change, the beam section prole may not be changed, either. In this case, only stiening inside the beam
cross section may be allowed.
As a specic example, we consider a section shown in Fig. 9 where the beam prole marked by thick
solid lines is assumed not be altered because of its assembly requirement with adjacent structural
components. The goal of this problem is to nd the optimal location and direction of a stiener in the
design domain that lies inside the cross section prole.
To nd the optimal location and direction of a stiener, we apply the topology optimization
technique formulated in the previous section. Two sets of weighting factors are considered with the 40%
mass constraint. The section shapes obtained from the present analysis are shown in Fig. 10. The
convergence history is also shown in Fig. 11. Table 2 compares the sectional rigidities of the initial
section having a uniform density distribution with those of the nal optimal cross section. The optimal
location and the direction of a stiener can be identied from the optimal shape shown in Fig. 10. This
indeed demonstrates the usefulness of the present topology optimization technique in the thin-walled
beam section stiener design. Without this approach, optimal stiening congurations would be dicult
to nd.
In topology optimization, utilizing the articial material density model as used in this work, the
solution usually depends on the mesh-size. However, the present problems are insensitive to meshing
size: compare Fig. 10(a) and Fig. 12. The result shown in Fig. 12 is obtained with the half size of the
original mesh shown in Fig. 10(a). The solution insensitiveness to mesh-size in this problem is partly
because of the use of the penalty function introduced in the objective function.
Table 1
Change of the cross section rigidities (E = 13 kN mm
2
, n=0.3)
20% constraint
w
J
=1.0 and w
I
=0.0 w
J
=1.0 and w
I
=0.02
Initial Final Initial Final
D
z
(kN mm
2
) 2910.4 4488.4 2910.4 3805.2
D
min
(kN mm
2
) 4963.1 8327.7 4963.1 10074
D
max
(kN mm
2
) 8308.6 14870 8308.6 10095
Fig. 9. Initial design domain with a geometric constraint.
Y.Y. Kim, T.S. Kim/ International Journal of Solids and Structures 37 (2000) 477493 489
We have also considered the case with a tighter mass constraint, namely a 25% mass constraint. The
results are depicted in Fig. 13, but they appear somewhat impractical. However, it is worth noting that
the candidate location and direction of an optimal stiener may be predicted quite well even with these
results.
6. Conclusions
A new topology optimization technique of beam cross sections is proposed in this paper. It is
demonstrated that a unied treatment of optimal section prole design, wall-thickness distribution and
section topology conguration can be achieved using the present technique. Among others, optimal
Fig. 10. Optimized cross sections (40% mass constraint): (a) w
J
=1.0 and w
I
=0.0; (b) w
J
=1.0 and w
I
=0.4.
Fig. 11. Convergence history for the results shown in Fig. 10: (a) w
J
=1.0 and w
I
=0.0; (b) w
J
=1.0 and w
I
=0.4.
Y.Y. Kim, T.S. Kim/ International Journal of Solids and Structures 37 (2000) 477493 490
Fig. 12. Optimized cross section with the half size of the original mesh used for Fig. 10(a).
Fig. 13. Optimized cross sections (25% mass constraint): (a) w
J
=1.0 and w
I
=0.0; (b) w
J
=1.0 and w
I
=0.4.
Table 2
Cross sectional rigidities before and after optimization; The results before optimization are those obtained from the sections with
uniform density distribution (E = 13 kN mm
2
, n=0.3)
40% constraint
w
J
=1.0 and w
I
=0.0 w
J
=1.0 and w
I
=0.4
Initial Final Initial Final
D
z
(kN mm
2
) 86.385 3318.8 86.385 3189.5
D
min
(kN mm
2
) 198.87 6139.8 198.87 6357.6
D
max
(kN mm
2
) 360.62 9831.7 360.62 7900.6
Y.Y. Kim, T.S. Kim/ International Journal of Solids and Structures 37 (2000) 477493 491
stiening direction and location in thin-walled section can be found with the present approach; this
would be dicult to achieve with any other existing approaches.
Acknowledgements
Acknowldegements: The rst author wishes to acknowledge the nancial support of Korea Research
Foundation made in the program of 1997. This work was completed during his stay at Stanford
University and he thanks the Division of Mechanics and Computation, Stanford University.
Appendix A
In order to prove Eq. (30), it is convenient to formulate the torsion problem in terms of the warping
function c that satises the following equation and boundary condition:
@
@x
_
G
_
@c
@x
y
__

@
@y
_
G
_
@c
@y
x
__
0, A1
_
@c
@x
y
_
dy
ds

_
@c
@y
x
_
dx
ds
0 or
dc
dn
yn
x
xn
y
on C: A2
In terms of the warping function c, the torsional rigidity can be written as (see Timoshenko and
Goodier, 1970 or Sokolniko, 1955)
D
z

_
A
G
_
x
2
y
2
x
@c
@y
y
@c
@x
_
dA: A3
Now applying the divergence theorem to the integral involving c in Eq. (A3) and using Eq. (A2) gives
D
z

_
A
Gx
2
y
2
dA
_
A
c
_
y
@G
@x
x
@G
@y
_
dA
_
C
Gc
_
y
@y
@s
x
@x
@s
_
ds

_
A
Gx
2
y
2
dA
_
A
c
_
y
@G
@x
x
@G
@y
_
dA
_
C
Gc
_
y
@c
@x
dy
ds

@c
@y
dx
ds
_
ds: A4
Reapplying the divergence theorem to the last integral in Eq. (A4) and then substituting Eq. (A1)
yields the following result:
D
z

_
A
Gx
2
y
2
dA
_
A
c
_
y
@G
@x
x
@G
@y
_
dA
_
A
@
@x
_
Gc
@c
@x
_

@
@y
_
Gc
@c
@y
_
dA

_
A
Gx
2
y
2
dA
_
A
G
_
_
@c
@x
_
2

_
@c
@y
_
2
_
dA: A5
The mean bending rigidity D
mean
can be explicitly written when Eqs. (7) and (10) and the relation
E = 2E/(1+n ) are utilized:
Y.Y. Kim, T.S. Kim/ International Journal of Solids and Structures 37 (2000) 477493 492
D
mean

1
2
_
A
Ex
2
y
2
dA 1 n
_
A
Gx
2
y
2
dA: A6
It is then straightforward to show Eq. (30) from the following result:
D
mean
D
z
n
_
A
Gx
2
y
2
dA
_
A
G
_
_
@c
@x
_
2

_
@c
@y
_
2
_
dA: A7
References
Banichuk, N.V., 1976. Optimization of elastic bars in torsion. Int. J. Solids and Struct. 12, 275286.
Banichuk, N.V., Karihaloo, B.L., 1976. Minimum-weight design of multipurpose cylindrical bars. Int. J. Solids and Struct. 12,
267273.
Bendse, M.P., 1995. Optimization of Structural Topology, Shape and Material. Springer, Berlin.
Bendse, M.P., Kikuchi, N., 1988. Generating optimal topologies in structural design using a homogenization method. Comp.
Meth. Appl. Mech. Engrg. 71, 197224.
Crandall, S.H., Dahl, N.C., Lardner, T.J., 1978. An Introduction to the Mechanics of Solids, 2nd ed. McGrawHill, New York.
Dems, K., 1980. Multiparameter shape optimization of elastic bars in torsion. Int. J. Numer. Meth. Engrg. 15, 15171539.
Gracia, L., Doblare, M., 1988. Shape optimization of elastic orthotropic shafts under torsion by using boundary elements.
Comput. Struct. 30, 12811291.
Guedes, J.M., Kikuchi, N., 1990. Preprocessing and postprocessing for materials based on the homogenization method with adap-
tive nite element methods. Comp. Meth. Appl. Mech. Engrg. 83, 143198.
Haftka, R.T., Gu rdal, Z., 1992. Elements of Structural Optimization, 3rd ed. Kluwer Academic Publishers, London.
Jog, C.S., Haber, R.B., Bendse, M.P., 1994. Topology design with optimized, self-adaptive materials. Int. J. Numer. Meth. Engrg.
37, 13231350.
Kim, Y.Y., Yoon, M.S., 1997. A modied Fourier series method for the torsion analysis of bars with multiply-connected cross sec-
tions. Int. J. Solids and Struct. 34, 43274337.
Mota Soares, C., Rodrigues, H.C., Oliveira Faria, L.M., Haug, E.J., 1984. Optimization of the geometry of shafts using boundary
elements. J. of Mech. Transms. Automa. Des. 106, 199202.
Na, M.S., Kikuchi, N., Taylor, J.E., 1983. Shape optimization for elastic torsion bars. In: Optimization Methods in Structural
Design. Bibliographisches Institut AG, Zu rich, pp. 216233.
Olho, N., Bendse, M.P., Rasmussen, J., 1991. On CAD-integrated structural topology and design optimization. Comp. Meth.
Appl. Mech. Engrg. 89, 259279.
Parbery, R.D., Karihaloo, B.L., 1977. Minimum-weight design of hollow cylinders for given lower bounds on torsional and exural
rigidities. Int. J. Solids and Struct. 13, 12711280.
Schramm, U., Pilkey, W.D., 1993. Structural shape optimization for the torsional problem using direct integration and B-splines.
Comp. Meth. Appl. Mech. Engrg. 107, 251268.
Sokolniko, I.S., 1956. Mathematical Theory of Elasticity, 2nd ed. McGrawHill, New York.
Suzuki, K., Kikuchi, N., 1991. A homogenization method for shape and topology optimization. Comp. Meth. Appl. Mech. Engrg.
93, 291318.
Timoshenko, S.P., Goodier, J.N., 1970. Theory of Elasticity, 3rd ed. McGrawHill, New York.
Vanderplaats, G.N., 1984a. Numerical Optimization Techniques for Engineering Design. McGrawHill, New York.
Vanderplaats, G.N., 1984b. ADS A Fortran Program for Automated Design Synthesis NASA CR 172460.
Y.Y. Kim, T.S. Kim/ International Journal of Solids and Structures 37 (2000) 477493 493