10th World Congress on Structural and Multidisciplinary Optimization

May 19-24, 2013, Orlando, Florida, USA

A discussion about choosing an objective function and constraint conditions in structural

topology optimization
Yunkang Sui1, Guilian Yi2
1

Beijing University of Technology, Beijing, China, [email protected]

Beijing University of Technology, Beijing, China, [email protected]

1. Abstract
Based on some examples calculated using the 99-line code made by Sigmund [1], the structural topology
optimization model construction is discussed to propose a problem that whether the topology optimization model
construction could be more reasonable. In order to solve this problem, a topology optimization model according
to practical engineering is constructed to search for the minimization of structural weight with a displacement
constraint. And a corresponding 120-line code written in Matlab is made by us. Calculation results of the
examples show that the model of the 120-line topology optimization code is more reasonable.
2. Keywords: structural topology optimization; optimization model; choosing an objective function and
constraint conditions; 99-line topology optimization code; 120-line topology optimization code
3. Introduction
The purpose of structural topology optimization is to seek the optimum layout of structural components or
sub-domains within a given design space under a given set of loads and boundary conditions such that the
resulting layout meets a prescribed set of performance targets. As an important branch of structural optimization
research, topology optimization has been enriched by lots of solution methods. An exhaustively detailed
summary for the development of topology optimization was made by Bendse and Sigmund [2], and the
application and development of numerical modeling methods used in structural topology optimization were also
reviewed in details by Rozvany [3].
Looking back the research of truss structure topology optimization led by Michell [4], a great progress has
been made since then. The Michell truss theory is not only developed by Rozvany et al. [5-6], but also the
research objects of topology optimization have developed from skeleton structures such as truss and frame into
continuum structures. Although the paper [7] superficially stated to be the homogenization method, Bendse and
Kikuchi broaden the research objects of structural topology optimization, because they presented the concept of
topological optimization for the continuum structures. The ground structure approach [8], which was only used
in topological optimization of skeleton structures, has become the foundation of the homogenization method in
topological optimization of continuum structures, and also the foundation of the methods seeking the optimum
layout of sub-regions for a given domain, such as the variable thickness method [9], the artificial material
method [1,10-12], the evolutionary structural optimization method [13], the independent continuous and
mapping method [14-17], the level set method [18-20], etc. Due to limited length of the paper, we cant list all of
the research here and can only mention the methodologies above, which are a few of thousands.
Among many studies, the 99-line topology optimization code written in Matlab by Sigmund has been a
unique significant research [1]. The code has been published on web (http://www.topopt.dtu.dk/?q=node/2),
which could be downloaded. We notice that Rozvany gave a good evaluation for this work. He said it played an

important role in SIMPs general acceptance [3].

In our opinion, an important role has two senses: it is not only extracurricular materials for students to
master the knowledge of structural topology optimization, but also an introductory tool for engineers to
understand and use a method of structural topology optimization. The 99-line code is short and includes finite
element analysis and optimizer subroutines, where the users can change the definition of structural sizes, set
different loading and boundary conditions and solve problems. Beginners can get great inspiration and help from
it.
The 99-line topology optimization code is based on the SIMP (Solid Isotropic Material with Penalization)
model of the artificial material method and its formulation shows as below:

Ee = E ( e ) = ep E0

e [0,1]

where Ee and E0 are the Youngs modulus of artificial and real material respectively,

(1)

is the elementary

artificial relative density, and p is the penalty factor.

On one hand, Eq. (1) involves the artificial relative density variables on a closed interval [0, 1], and in fact
the discrete 0/1 topology variables are made to be continuous ones. On the other hand, the penalty function is
used to establish the relationship between artificial relative densities and Youngs modulus, and the explicit
relationship of element stiffness and continuous topology variables is formulated essentially. Actually, the
artificial material method has overcome two difficulties of structural topology optimization through Eq. (1). The
first difficulty is that the model can not be constructed and the optimization problem can not be solved owing to
discrete 0/1 topology design variables. The second difficulty is that the structural performance couldnt be
expressed by topology design variables.
The structural topology optimization model in paper [1] is to seek minimum compliance, or maximal
stiffness, with a volume constraint. We call it the MCWC (minimum compliance with a weight constraint)
formulation. It is presented as below:

For

min

s.t.

c ( ) = U T KU =
KU = F
V ( )
= f
V*
0 < min e 1

u eT k e u e
e =1

(2)

( e = 1, L , N )

where K, U and F are the structural global stiffness matrix, total displacement vector and total force vector,
respectively, ue is the displacement vector of the e-th element, k e = k ( e ) = E ( e ) k e is the element
0

stiffness matrix of the e-th element, k e is the element stiffness matrix for unit Youngs modulus, is the
artificial relative density vector. In order to avoid singularity of the stiffness matrix, the minimum relative density

min is usually taken as 0.001. N is the total number of elements, and the penalty factor p is usually equal to 3.
V ( ) and V* are the total volumes of the designed structure and the ground structure, respectively, f is the
pre-setting volume ratio. Eq. (2) has been solved by an optimality criteria method.
During trying to adopt the 99-line topology optimization code, a number of examples were computed. We

found that the pre-setting volume ratio had a decisive influence on the results of structural topology optimization:
different volume ratios would produce different topological configurations. This awaken us to think further on
the relationship between the pre-setting volume ratios and optimal topological configurations. Based on this, we
consider a question: whether structural topology optimization has a more reasonable model?
Following the train of thinking, this paper proposes a structural topology optimization model of minimizing
structural weight with a displacement constraint, and develops a corresponding 120-line topology optimization
code. Through the example calculations by the 120-line code and then they are compared with the calculation
results by the 99-line code. Research results show that constructing a more reasonable structural topology
optimization model is a significant mention.
4. Expression of structural topology optimization with a more reasonable model construction
The optimization problems discussed in this paper will be focused on structures that are able to bear external
forces, rather than force inverter or compliant mechanisms. Therefore, 4 simple structures are used as examples
in this paper, which are shown in Figure 1.
Example 1, shown in Figure 1 (a), is the half of MBB-beam, with a ground structure of 60mm20mm1mm,
unit force F=1, unit Youngs modulus E0=1.0 and the Poissons ratio v = 0.3 .
Example 2, shown in Figure 1 (b), is the short cantilever beam, with a ground structure of
32mm20mm1mm, unit force F=1, unit Youngs modulus E0=1.0 and the Poissons ratio v = 0.3 .
Example 3, shown in Figure 1 (c), is the short cantilever beam with a fixed hole, with a ground structure of
45mm30mm1mm, unit force F=1, unit Youngs modulus E0=1.0 and the Poissons ratio v = 0.3 . The center of
the hole is located at the intersection of 1/3 horizontal length and 1/2 vertical length from left to right, while the
radius is equal to 1/3 vertical length.
Example 4, shown in Figure 1 (d), is the cantilever beam, with a ground structure of 80mm50mm1mm,
loading force F=9kN , Youngs modulus E0=1.0106MPa and the Poissons ratio v = 0.3 .

(a)

(b)

(c)

(d)

Figure 1: Ground structures and boundary conditions for the four test examples
The first three examples are from paper [1]. There are 9 sub-examples for each example, where the
pre-setting volume ratios are 9 equally distributing numbers between 0.1 and 0.9.
A total of 36 sub-examples are all computed by the 99-line topology optimization code and the results are
shown in Table 1. Meanwhile, the optimal results of four examples with the pre-setting volume ratio of 0.1 are
examined in more detail and shown in Table 2, where in the Number of elements column, Black presents the
number of black elements with a relative density of 1.0, White presents the number of white elements with a
relative density of 0.001, and Grey presents the number of grey elements with various relative densities on the
open interval (0.001, 1.0).

Table 1: Optimum topological configurations for different pre-setting volume ratios

Volume ratio
Iterations
0.9

96

80

106

134

56

17

81

79

50

23

129

94

69

34

58

63

71

35

47

135

42

66

44

363

222

83

44

1132

756

740

1971

Opt. Config.
Iterations

0.1

29

Opt. Config.
Iterations

0.2

Opt. Config.
Iterations

0.3

44

Opt. Config.
Iterations

0.4

45

Opt. Config.
Iterations

0.5

Example 4

Opt. Config.
Iterations

0.6

Example 3

Opt. Config.
Iterations

0.7

Example 2

Opt. Config.
Iterations

0.8

Example 1

Opt. Config.

Table 2: Results of 4 examples with the pre-setting volume ratio of 0.1

Number of
Example

structure

Number of elements
Compliance

analysis

Total volume of

Average density of

Black

White

Grey

grey elements

grey elements

119.3960

0.2003

Example 1

1132

4785.2511

604

596

Example 2

756

899.0410

382

258

63.6180

0.2466

Example 3

740

787.8421

904

440

128.0961

0.2911

Example 4

1971

33266.6946

2740

1254

391.2600

0.3120

From the results of 36 sub-examples in Table 1, we observe dependency of the structural optimal
topological configurations on the pre-setting volume ratios, and summarize the following three aspects:
(1) 3 or 4 different kinds of typical structural configurations are obtained in all 4 examples.
(2) When the pre-setting volume ratio is too small, the structure will be broken off or disconnected, and it
degenerates into a mechanism. According to the data shown in Table 2, when the pre-setting volume ratio of each
example is 0.1, the number of the black elements in the structural optimal topological configurations is very
small or even zero. At the same time, the average densities of grey elements are 0.2003, 0.2466, 0.2911 and
0.3120, respectively. These show that the structures have already degenerated into mechanisms.
(3) Along with the reduction of pre-setting volume ratios, all 4 examples go through an evolution from
cumbersome structures, Michell-truss-like structures, light structures into degenerating mechanisms at the
end.
Since the optimal topological configurations are really dependent on the pre-setting volume ratios, there is a
question that how a reasonable pre-setting volume ratio can be selected in the stage of conceptual design, Could
the most appropriate volume ratio be found logically? In other words, rather than using a presumed volume ratio,
why dont we determine an optimal volume ratio at the same time with searching for an optimal topological
configuration?
5. Presentation of a topology optimization model best fitting practical engineering problems
Many examples in the last section inspire us to seek the optimal volume ratio at the same time with searching
for an optimal topological configuration. This means that there will be a new volume ratio objective function
besides the compliance objective function when we use the 99-line topology optimization code. If there are a
compliance objective function and a volume ratio objective function at the same time, the optimization problem
wouldnt be the one with a single objective function and the optimum solutions wouldnt be determined.
Could the volume ratio only be considered as a single objective function? Since the compliance is not taken
as the objective function, it should be transformed into a constraint condition. However, what should the
compliance constraint value be selected? Therefore, there is a similar difficulty like that one in the 99-line
topology optimization code, in which a constraint value of the most appropriate volume ratio is not determinate.
How can the difficulty of selecting a compliance constraint value be overcome? In fact, if we move our
focus from compliance into another one, there is a solution. Actually, rather than the compliance constraint, the
strength or stiffness constraint is considered for practical engineering problems. That is, we can use the strength
or stiffness constraint instead of the compliance constraint.
In order to try to put this idea into practice, we present a topology optimization model of minimizing
structural weight with satisfying a displacement constraint at a certain point of interest. Why is the volume ratio
objective function changed into the structural weight objective function? This is because minimizing structural

weight is equivalent to minimizing structural volume with the same material, while the minimum structural
volume is equivalent to the minimum structural volume ratio with the same ground structure. So minimizing
structural weight should be equivalent to minimizing the volume ratio. The advantage of using the structural
weight objective function is that the formulation of topology optimization is in line with formulations of the
section optimization and the shape optimization.
A minimum weight formulation for topology optimization with a displacement constraint can be presented
as follows:

For
N

min W ( ) = e we0

e =1
s.t. u ( ) = u

0 < min e 1 (e = 1,L, N )

(3)

where we is the element weight when the density of the e-th element equals to 1, W ( ) and u () are the
structural total weight and the displacement at a point of interest respectively, u is the displacement allowable
value. Similarly with the 99-line code, the minimum density is set as min = 0.001 for an avoidance of
singular stiffness matrix in seeking optimum solutions.
5.1. Derivation for the explicit displacement function at the point of interest
Using the unit virtual load method [21], the displacement at a point of interest can be expressed by calculating
the virtual work as follows:
N

u = ue = RT v dv
e =1

(4)

e =1 e

where ue is the e-th elements contribution to the displacement value at the point of interest,

R and v

denote the element stress vectors associated to the real load and the element strain vectors associated to the
virtual load, respectively.
According to the work energy theorem, the virtual work in Eq. (4) is equal to the virtual work obtained
from nodal forces and nodal displacements:
N

u = ( PeR ) T uev

(5)

e =1

where Pe

and ue are the elements nodal force vector associated to the real load and the corresponding

elements nodal displacement vector associated to the virtual load in the e-th element, respectively, and

ue = ( PeR ) T uev .
Due to the element stiffness equation, we have:

K e uev = Pev

(6)

where K e and Pe are the elements stiffness matrix and the elements nodal force vectors associated to the

virtual load in the e-th element, respectively.

Substituting Eq. (6) into Eq. (5), we can get:
N

u = ( PeR ) T ( K e ) 1 Pev

(7)

e =1

Since the element stiffness is proportional to the Youngs modulus, the elements stiffness can be written as:

K e = ke0 Ee = ke0 E0 ep = K e0 ep

(8)

where K e is the element stiffness matrix; k e is the element stiffness matrix for unit Youngs modulus; Ee
is the element Youngs modulus for linear isotropic material (artificial material); E0 is the Youngs modulus of
solid material (real material,

e = 1 ); K e0

is the element stiffness matrix when

e = 1

with the Youngs

modulus E0 .
Substituting Eq. (8) into Eq. (7), we obtain:
N

e =1

e =1

e =1

u = ue = ( PeR )T ( K e0 ) 1 Pev e p

De0

(9)

ep

where De is the constant coefficient of the displacement contribution component of the e-th element to the
0 1

point of interest. In Eq. (9), the reason of ( Pe ) ( K e ) Pe De rests with introducing an assumption of the
R T

statically determinate structure. Because the displacements of the statically determinate structure have no
relationship with the nodal force vectors and our structure is regarded as a statically determinate one in each
0

structural optimization iteration, De is a constant which is not dependent on Pe and Pe for each iteration.

5.2. Solutions of the optimization model

Substituting Eq. (9) into Eq. (3), we obtain an explicit topology optimization model as follows:

For
N

=
min
W
(
)

e we0

e =1

N D0
s.t. ep = u
e =1 e

0 < min e 1 (e = 1,L, N )

(10)

According to Lagrangian multiplier method, we can get:

N

e =1

e =1

L( , ) = e we0 + (

De0

ep

u)

The first order derivatives of the Lagrangian function are found as:

(11)

L
D0
= we0 p pe+1 = 0
e
e

(12)

L N De0
=
u = 0
e =1 ep

(13)

The design variables can be obtained from Eq. (12) as follows:

e = (

pDe0
we0

1
p +1

(14)

Substituting Eq. (14) into Eq. (13), we can get

N

( p )

1
p +1

=(

(D )
e =1

e = (

(D )
k =1

1
p +1

(w )
0
e

p
p +1

Then, the Lagrangian multiplier

are got as follows:
N

0
e

0
k

1
p +1

1
p

(15)

is eliminated by substituting Eq. (15) into Eq. (14), and the solutions

0
k

(w )

p
p +1

1
p

D0
) ( e0 ) p +1
we

(16)

In order to think of contribution of element densities to the displacement at the point of interest, the partial
derivative of Eq. (9) with respect to densities are obtained as follows:

u
pD 0
= p +e1
e
e
When De > 0 , we have
0

displacement

(17)

u
> 0 . Thats to say, if the element density e is increased, the
e

u at the point of interest will be increased; if the element density e is decreased, the

displacement at the point of interest will be decreased. So increasing element density

is not economical in

terms of design. In this case, the element density values have to be decreased to reach the minimum value,
therefore the displacement contribution and stiffness contribution of the corresponding element can reach
minimum and maximum values, respectively. We dont need to design the density of this element in this case, so
it is called as passive design variable.
When De 0 , we have
0

displacement

u
< 0 . Thats to say, if the element density e is increased, the
e

u at the point of interest will be decreased; if the element density e is decreased, the

displacement at the point of interest will be increased. In other words, the element density

is increased,

displacement contribution and stiffness contribution of the corresponding element will be decreased and
increased, respectively. This is a reasonable return on economic investment in increasing the element density e .
In this case, the density should be designed, and it is called as the active design variable.
The discussion above is summarized as following: the elements satisfied with De 0 are designable
0

elements,

and

the

corresponding

density

variables

are

the

active

variables

included

in

set

I a = e De0 0, e = 1,L , m ; the elements satisfied with De0 < 0 are not designable and the corresponding
density variables are in the passive design variable set.
The values of the density variables in the passive design variable set are to remain constant during each
iteration of seeking the optimum solutions. Therefore, the topology optimization model in Eq. (9) can be
rewritten as follows:

For
m

=
e we0 + W0

min
W
(
)

e =1

m D0
e
s.t. p + u0 = u
e =1 e

0 < min e 1 (e = 1,L, N )

where

W0 =

w
e

eI a

0
e

, and u0 =

De0

eI a

p
e

(18)

. W0 and u0 are constants. So the constant terms, which have

nothing to do with seeking optimization, can be eliminated from the objective function. Similar to the previous
deduction, we can obtain the following equation to replace Eq. (16) as:
1
p
m
0 p +1
0 p +1
1
1
( Dk ) ( wk )
De0 p +1
p
e = ( k =1
) ( 0 )
u u0
we

min or 1 (e I a )

(e I a )

(19)

The active design variables calculated from the equation above are not ensured to have a smaller value than
1 or a larger value than the minimum density min . Then an adjustment should be made as below:

where e I a (if e 1)

e* = e where e I a (if min < e < 1)

min where e I a (if e min )

(20)

Now that the active/passive design variable set is adjusted, the calculation according to Eq. (19) has to be
re-performed, which is called as a small optimization iteration. The active/passive design variable set classified
0

by the sign of De is a global pre-judgment after one structural analysis before the topology optimization model
construction for iterative solutions, but Eq. (20) is a local adjustment of reclassifying the active/passive design
variable set after a small optimization iteration. This kind of small optimization iteration should be computed
continuously according to Eq. (19) until the active/passive design variable sets are no longer changedand the
process of small optimization iterations stops. Thus, the optimal solutions of the model Eq. (18) for the
corresponding structural analysis are obtained. Here, we say that a big optimization iteration is finished. Before a

new big optimization iteration, a new structural analysis should be made. However, the whole optimization
process needs sequential reconstructions of optimization models corresponding to sequential big optimization
iterations.
To ensuring mesh-independency, a filter function expressed in Eq. (21) [17] is used in this paper. The
displacement contribution component of each element to the point of interest in Eq. (9) is filtered. Thus the new
coefficients for the filtered displacement components are obtained in Eq. (22) as follows:

u~e =

w( x ) u
w( x )

j N e

j N e

(21)

De0 = u~e ep

(22)

where x j is the spatial location of geometric center for the j-th element; N e = j x j xe rmin

} denotes the

neighborhood of the e-th element within a given filter radius rmin of the center of the e-th element;

w( x j ) = max(0, rmin x j xe ) is a weight function, also a linearly decaying weighting function, which is
linearly reduced along the distances of the neighbor elements to the center element.
In order to keep the comparability of our topology optimization, the same convergence criterion as that in
the 99-line topology optimization code is used in this paper. The optimal solutions
iteration and

(k )

( k +1)

for the (k+1)-th

for the k-th iteration should satisfy a relationship as follows:

max(max( ( k +1) ( k ) )) 0.01

(23)

6. Results comparison of two models

The solution process for the optimization model Eq. (3) is programmed into a 120-line topology optimization
code, which can be downloaded from website https://sites.google.com/site/yiguilian/link. In order to compare
with the 99-line code [1], 4 examples in Figure 1 are solved by the two codes. The point of interest is taken as the
loading point.
Since a small-volume structure has little material and small stiffness, it causes a large displacement at the
point of interest. On the contrary, a large-volume structure has a large stiffness and a small displacement at the
point of interest. Therefore, whole calculation for each example has two stages. In the first stage, the model in Eq.
(2) under a pre-setting volume ratio constraint is solved using the 99-line topology optimization code to obtain
the displacement value at the loading point and the optimum topological configuration. This displacement value
is taken as a displacement constraint value in the second stage. Then, the model in Eq. (3) under the displacement
constraint is solved using the 120-line topology optimization code. All the data and optimal topological
configurations are shown in Table 3 and Figure 2, respectively. In the 120-line code, there is a measure of
discreteness [22] which is used to distinguish whether an optimum solution converges to a discrete one. This
measure of discreteness is defined as:

10

M nd =
where

4
e =1

*
e

(1 e* )

100%

(24)

e* is the optimum relative density of the e-th element. Once the design is converged, M nd with a

value of 0% denotes all element relative densities are equal to 0 or 1, and M nd with a value of 100% denotes
all element relative densities are equal to 0.5. Therefore, a smaller value of M nd means less grey elements and
is expected to give a better design.
In Table 3, the meanings of Black, White and Gray in the Number of elements column are the same as
those in Table 2. Total weight denotes the structural weight without the white elements with a relative density
of 0.001, including the weight of all black and grey elements. Weight of grey elements denotes the total weight
of all grey elements, and Ratio of grey elements is the percentage ratio of the number of grey elements in the
number of all elements.
Table 3: Results for 4 examples

Number of
Example

structural
analysis

99-line
Example

code

120-line
code
99-line

Example

code

120-line
code
99-line

Example

code

120-line
code
99-line

Example

code

120-line
code

Total
weight

Displacement
at the point of
interest

Number of elements
Black

White

Grey

Weight of
grey
elements

Ratio of
grey

Mnd

elements

(%)

(%)

94

599.67

203.2980

447

321

432

152.67

36.00

17.55

121

599.15

203.2961

467

370

363

132.15

30.25

16.35

71

255.74

57.3525

197

257

186

58.74

29.06

13.96

70

255.81

57.3543

208

273

159

47.81

24.84

12.75

34

674.60

52.0993

544

467

339

130.60

25.11

12.91

26

673.29

52.0988

565

491

294

108.29

21.78

11.67

58

1998.54

0.3488

1718

1458

824

280.54

20.60

9.56

68

1990.17

0.3488

1764

1532

704

226.17

17.60

8.70

In Figure 2, five parameters between parentheses in each of eight captions for configurations all have their
own meaning on the left side and right side, respectively. For example on the left side, in Figure 2 a1) there is a
caption 99-line code (60,20,0.5,3.0,1.5), which represents concrete computing data of an example for using the

11

99-line code, where 60 elements grids in x-direction, 20 elements grids in y-direction, volume ratio constraint
value 0.5, penalty factor 3.0 and filter radius 1.5. For example on the right side, in Figure 2 a2), there is a caption
120-line code (60,20,203.2980,3.0,1.5), which represents concrete computing data of an example for using
120-line code, where 60 elements grids in x-direction, 20 elements grids in y-direction, displacement constraint
value 203.2980, penalty factor 3.0 and filter radius1.5.
Topology optimization model of Eq. (2)

Topology optimization model of Eq. (3)

a1) 99-line code (60,20,0.5,3.0,1.5)

a2) 120-line code (60,20,203.2980,3.0,1.5)

b1) 99-line code (32,20,0.4,3.0,1.2)

b2) 120-line code (32,20,57.3525,3.0,1.2)

c1) 99-line code (45,30,0.5,3.0,1.5)

c2) 120-line code (45,30,52.0993,3.0,1.5)

d1) 99-line code (80,50,0.5,3.0,1.5)

d2) 120-line code (80,50,0.3488,3.0,1.5)

Figure 2: Optimum topological configurations for 4 examples

The computational results show that, the topology optimization model proposed in this paper can basically
satisfy the displacement constraint, which is given by the 99-line code at the loading point. And the 120-line
code obtains smaller structural weight than the 99-line code. Most of all, the 120-line code can get smaller values
of measure of discreteness and less grey elements than the 99-line code. The method of this paper improves the
degree of discreteness of design variables in the topology optimization. In computing speed, the convergence
rates of both codes are of about the same. They indicate the topology optimization model presented in this paper
is more reasonable in the model construction and results.
7. Conclusions
In the introduction of this paper, we pointed out: the 99-line topology optimization code played an important
role in helping students and engineers to learn topology optimization. Our work here indicates that the 99-line
topology optimization code also plays an important role for researchers in structural topology optimization.
The reason is that the 99-line code inspires us to put forward a more reasonable model construction for topology

12

optimization through the examples calculated by itself.

In structural topology optimization, although there are many researches on optimization model with
objective function of structural weight subjected to displacement constraints, we have not seen a detailed
comparison on results about above model and a model with objective function of structural compliance under a
structural volume ratio constraint. Admitting that the method of structural topology optimization can be used for
references to design useful compliant mechanisms such as the force inverter, but we limited ourselves here to the
study of topological changes of structures.
(1) The essence of the structural weight or volume ratio is an economic indicator, and the compliance,
displacement, stress, buckling, frequency, fatigue etc. are mechanical performance indicators.
(2) The purpose of structural optimization is to find the optimum points representing a compromise in the
mutual restraint and mutual game between economic indicators and mechanical performance indicators. Finally,
a contradiction between construction cost and structural safety will be solved.
(3) The practical engineering problems usually seek the minimum economic indicator, which is as the
objective function, with guarantee of the constraints of mechanical performance indicators. Why can not we
reverse them? Because its enough to guarantee structural safety represented by the mechanical performance
indicators mechanical performance has no significance. Since the economic indicator belongs to fabricating cost
in essence, its beforehand setting is impossible. Thereby it is a behavior of apriorism. Conversely, seeking the
minimum economic indicator should be done along with satisfying the mechanical performance indicators, and
the reasonable cost can be obtained in accordance with its natural tendency.
(4) The mechanical performance indicators can be classified into engineering and academic. The
engineering performance indicators include the allowable displacements, the allowable stresses and so on. But
the compliance belongs to the academic performance indicators. The engineering performance indicators are the
main points that should be satisfied directly in design and they have been refined into perfection day by day by
the accumulated experience of engineers. But the academic performance indicators are sometimes the
information concerned by researchers. If the academic performance is supplied to reference in design process,
the reference is only indirect in consideration, but without direct interest.
(5) The mechanical performance indicators are related to the loading cases. If a mechanical performance is
taken as the objective function, in multi-loading cases, the optimization problem will become a multi-objective
optimization problem, which is difficult to be solved. Since the economic indicators are not the function of
loading cases, it is taken as the objective function to avoid producing the multi-objective optimization problem,
and we always obtain an optimization problem with the single objective function.
8. Acknowledgements
The authors acknowledge the support of the National Natural Science Foundation of China under grant no.
(11172013) and the financial assistance from the China Scholarship Council. And we are grateful to Dr. Nam-Ho
Kim in University of Florida for polishing English expression of the paper.
9. References
[1] O. Sigmund, A 99-line topology optimization code written in Matlab, Structural Multidisciplinary
Optimization, 21, 120-127, 2001.
[2] M. P. Bendse and O. Sigmund, Topology optimization: theory, methods and applications, Springer, Berlin,
2003.
[3] G. I. N. Rozvany, A critical review of established methods of structural topology optimization, Structural
Multidisciplinary Optimization, 37(3), 217-237, 2009.

13

[4] A.G. M. Michell, The limits of economy of material in frame structure, Philosophical Magazine, 6(8),
589-597, 1904.
[5] G. I. N. Rozvany and W. Gollub, Michell layouts for various combinations of line support, International
Journal of Mechanical Sciences, 32(12), 1021-1043, 1990.
[6] G. I. N. Rozvany, Some shortcomings in Michells truss theory, Structural Optimization, 12(4), 244-250,
1996.
[7] M. P. Bendse and N. Kikuchi, Generating optimal topologies in structural design using a homogenization
method, Computer Method in Applied Mechanics Engineering, 71(1), 197-224, 1988.
[8] W. S. Dorn, R. E. Gomory, and H. J. Greenberg, Automatic design of optimal structures, Journal de
Mecanique, 3(1), 25-52, 1964.
[9] G. D. Cheng and X. Guo, -relaxed approach in structural topology optimization, Structural Optimization,
13, 258-266, 1997.
[10] M. P. Bendse, Optimal shape design as a material distribution problem, Structural Optimization, 1,
193-202, 1989.
[11] M. Zhou and G. I. N. Rozvany, The COC algorithm, part II: topological, geometry and generalized shape
optimization, Computer Method in Applied Mechanics Engineering, 89(1-3), 309-336, 1991.
[12] H. P. Mlejnek, Some aspects of the genesis of structures, Structural Optimization, 5, 64-69, 1992.
[13] Y. M. Xie and G. P. Steven, A simple evolutionary procedure for structural optimization, Computers and
Structures, 49(5), 885-896, 1993.
[14] Y. K. Sui, Modeling, Transformation and Optimization New Developments of Structural Synthesis
Method. Dalian University of technology Press, 1996. (in Chinese)
[15] Y. K. Sui and D. Q. Yang, A new method for structural topological optimization based on the concept of
independent continuous variables and smooth model, Acta Mechanica Sinica, 14(2), 179-185, 1998.
[16] Y. K. Sui, J. Z. Du, and Y. Q. Guo, Topological optimization of frame structures under multiple loading
cases, International Conference on Computational Methods, 15-17, Singapore, 2004.
[17] Y. K. Sui and X. R. Peng, The ICM method with objective function transformed by variable discrete
condition for continuum structure, Acta Mechanica Sinica, 22(1), 68-75, 2006.
[18] J. A. Sethian and A. Wiegmann, Structural boundary design via level set and immersed interface methods,
Journal of Computational Physics, 163, 489-528, 2000.
[19] M. Wang, X. Wang, and D. Guo, A level set method for structural topology optimization, Computer
Method in Applied Mechanics Engineering, 192(1-2), 227-246, 2003.
[20] G. Allaire, F. Jouve, and A. M. Toader, Structural optimization using sensitivity analysis and a level-set
method, Journal of Computational Physics, 194(1), 363-393, 2004.
[21] R. C. Hibbeler, Structural analysis, 8th edition. Prentice Hall, Inc. 2012.
[22] O. Sigmund, Morphology-based black and white filters for topology optimization, Structural
Multidisciplinary Optimization, 33(4-5), 401-424, 2007.

14