See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/265167203

A Comparison Between Different Topology Optimization Methods

Article  in  Engineering Journal · December 2013

CITATION READS

1 3,647

3 authors:

Mohamed Fanni N. Shabara

Egypt-Japan University of Science and Technology Mansoura University
76 PUBLICATIONS   215 CITATIONS    30 PUBLICATIONS   285 CITATIONS   

SEE PROFILE SEE PROFILE

Mohamed Alkalla
Mansoura University
11 PUBLICATIONS   21 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Development of Endoscopic Manipulator System for Minimally Invasive Robotic Liver Surgery View project

Propeller-type Climbing Robot for Industrial Vessels Inspection View project

All content following this page was uploaded by Mohamed Alkalla on 11 March 2019.

The user has requested enhancement of the downloaded file.

 A Comparison between Different Topology
Optimization Methods
Fanni, M.*, Shabara, M. N.†, Alkalla, M. G.‡
* Associate Professor, Prod. Eng. & Mech. Design Dept.
† Professor, Prod. Eng. & Mech. Design Dept.
‡Demonstrator, Prod. Eng. & Mech. Design Dept.
Faculty of Eng., Mansoura University, Mansoura, Egypt.

‫ملخص البحث‬
‫ فً هزا انبحث يخى دساست انخصًيى‬.‫يعخبش انخصًيى انخىبىنىجً األيثم ين أكثش يجاالث انخصًيى انبنائً أهًيت‬
‫ هزه‬.‫انخىبىنىجً األيثم نهحصىل عهً أقم يطاوعت نألجزاء انًيكانيكيت باسخخذاو طشيقت خطىط انخقاسب انًخحشكت‬
‫ حيث انها حسخطيع انخعايم يع أي نىع ين يعادالث انهذف‬،‫انطشيقت حعخبش ين طشق انخصًيى األيثم انشايهت وانًشنت‬
‫فً هزا انبحث حى دساست حأثيش حغييش خطىط انخقاسب انذنيا وانقصىي عهً عًهيت‬.‫وأي عذد ين يعادالث انقيىد‬
‫ كًا حى حنفيز‬.‫انخقاسب نهخصًيى األيثم بهذف انحصىل عهً انخصًيى األيثم بصىسة ثابخت وفً أقم وقج يًكن‬
ً‫انخصًيى األيثم ننًارج يخخهفت يثم انعاسضت انًثبخت ين طشف واحذ وانعاسضت انًثبخت ين طشفين سىاء كانج ف‬
‫ وأيضا حى عًم يقاسنت بين طشيقت خطىط انخقاسب انًخحشكت وطشق يخخهفت‬.‫يسخىي واحذ أو يجسًت ثالثيت األبعاد‬
‫نهخصًيى انخىبىنىجً األيثم يثم طشيقت انبشيجت انخشبيعيت انًخعاقبت و طشيقت انًعاييش انًثهً وطشيقت انخاليا راحيت‬
.ً‫ طبقا نقيًت انًطاوعت نهخصًيى وانىقج انًسخههك نكم طشيقت وانشكم اننهائ‬، ‫انحشكت انهجينت‬

Abstract
Topology optimization approach is considered among the most interesting fields of structural
optimization. In this paper topology optimization for compliance minimization using method of moving
asymptotes MMA is presented. This method is considered as a general and flexible optimization method,
where it can handle any kind of objective function and any number of constraints. The effect of changing
the lower and upper asymptotes on the optimization process convergence is studied for seeking the
demanded convergence with more stability and minimum time as possible. Topology optimization of
different models such as a cantilever beam and simply supported beam for two and three dimensional
structure is accomplished. Also a comparison between Method of Moving Asymptotes (MMA) and
different methods such as Sequential Quadratic Programming (SQP), Optimality Criteria (OC), and
Hybrid Cellular Automata (HCA) is made according to the compliance value, time consumed and the
resulted topological shape.

1. Introduction Topology optimization strives to achieve

Topology optimization approach is
considered among the most interesting
fields of structural optimization. It is
considered as an important field and a
promising area that meets a great
interest from mechanical designers and
manufacturers. It is a relatively new but
rapidly expanding research field. It also
has important practical applications in
automotive and aerospace industries.
the optimal distribution of material
within finite volume design domain;
which maximizes a certain mechanical
performance under specified constraints.
Its algorithms selectively remove and
relocate the elements to achieve the
optimum performance [3]. It can
provide a good configuration concept for
the structure as a minimum compliance
or maximum stiffness design.

1
 The first paper on topology
optimization was published over a
century ago by the versatile Australian
inventor Michelle (1904) who derived
optimality criteria for the least weight
layout of trusses. Starting with the
landmark paper of Bendsøe and Kikuchi
(1988) that has presented most popular
numerical FE-based topology
optimization method. this method is
SIMP method.
Ole Sigmund (2001) developed a
Matlab code for topology optimization
based on minimizing compliance,
mainly using optimality criteria
approach that depends on the sensitivity
of the objective function [1]. Also there
are some methods that can be considered
as numerical methods such as,
sequential linear programming (SLP),
sequential quadratic programming
(SQP), and method of moving
asymptotes (MMA) by Krister Svanberg
[4] that can be used for topology
optimization.
In this paper, the MMA method is
applied to different 2D and 3D models
(with different number of elements and
boundary conditions) as it will be
discussed later in section 5&7. A
comparison between this method and
other methods such as SQP, optimality
criteria (OC) approach [1], and hybrid
cellular automata method (HCA) [3] is
presented in section 8.
2. Method of Moving Asymptotes
The ideal method for structural
optimization should be flexible, general,
and able to handle not only element size
as design variables, but also other
variables such as shape and material
orientation angles. It should be able to
handle all kinds of constraints. MMA
method can handle all of these problems
2
in addition to general non-linear
programming problems. Moreover it is
easy to implement and use. The method
of moving asymptotes (MMA) is a new
method for structural optimization that is
based on a special type of convex
approximation [4].
It is a common approach to
mathematical programming method for
non-linear optimization problems to
formulate a local model at an iteration
point. This local model approximates the
original one at the given iteration point
but is easier to solve. Classical methods
like sequential quadratic programming
(SQP) use such local models. But with
respect to the large number of design
variables, the use of SQP methods and
solving the local models is very costly if
not even impossible, due to the fact that
gathering second order information for
the approximation of the Hessian could
be an insuperable task [11].
consider a structural optimization
problem of the following form:
P : min 𝑓0 𝐱 𝐱 ∈ 𝑅𝑛 (1)
S. t. : 𝑓𝑖 𝐱 ≤ 𝑓𝑖 , 𝑓𝑜𝑟 𝑖 = 1, … , 𝑚 (2)
𝑥𝑗 ≤ 𝑥𝑗 ≤ 𝑥𝑗 𝑓𝑜𝑟 𝑗 = 1, … , 𝑛 (3)
Where 𝐱 = (𝑥1 , … … , 𝑥𝑛 )𝑇 is the vector
of design variables; 𝑓0 𝑥 is the
objective function; 𝑓𝑖 𝑥 ≤ 𝑓𝑖 , is the
behavior constraints; 𝑥𝑗 and 𝑥𝑗 are
given upper and lower bounds.
The method is interpreted in brief that
(𝑘)
each 𝑓𝑖 , where k is the current
iteration, is obtained by a linearization
of 𝑓𝑖 in variables of the type
1 (𝑥𝑗 − 𝐿𝑗 ) or 1 (𝑈𝑗 − 𝑥𝑗 ) dependent
on the signs of derivatives of 𝑓𝑖 at 𝐱 (𝑘) .
The values of the parameters 𝐿𝑗 and 𝑈𝑗
are normally changed between the Then the general equations can be
iterations, and we will sometimes refer written as, [1, 7]
to 𝐿𝑗 and 𝑈𝑗 as "Moving Asymptotes".
n

For more details on this method, see [4]. min : c( x )  U T KU   ( xi ) p u e k e u e 
T
x
i 1

V (x)  (7)
3. Topology Optimization s.t. :  v frac 
V0 
Using MMA : KU  F 

MMA algorithm that was presented : 0  x m in  x  1 
by Krister Svanberg and written with
Matlab in [6] is used in this section. At where, U and F are the global
the beginning, it is important to define displacement and force vectors,
the general equations of topology respectively; K is the global stiffness
optimization for minimizing the matrix, ue and k e are the element
compliance is conjunction with Solid displacement vector and stiffness matrix,
Isotropic Material with Penalization respectively, x is the vector of design
approach (SIMP) that was presented by variables which is relative density of
Bendsøe [12]. This approach proposed each elements, 𝐱 𝐦𝐢𝐧 is a vector of
that the material properties are assumed minimum relative densities (non-zero to
constant within each element in the avoid singularity) V(x) and V0 is the
design domain. Normally, a continuous material volume and the initial volume
relative density is used as a design respectively; where vfrac is the prescribed
variable. The modulus of elasticity for volume fraction.
each element 𝐸𝑖 is modeled as a function
of the relative density xi using a power If we have a 2D model which needs to
law: be optimized, we have to descritize it to
horizontal elements number nelx, and
𝜌𝑖 𝑥𝑖 = 𝜌0 𝑥𝑖 vertical elements number nely as in the
𝑝
initial design in Fig. 1. Then the number
𝐸𝑖 𝑥𝑖 = 𝐸0 𝑥𝑖 , (0 ≤ 𝑥𝑖 ≤ 1) (4) of design variables that will be used, n
equals to nelx*nely, also the number of
where 𝜌𝑖 is element density; 𝜌0 is the constraints equations, m equals to 2 as in
initial density; 𝐸𝑖 is element elasticity; equation 7.
𝐸0 is the elastic modulus of the base
material; and p is a penalization power.
This power penalizes intermediate
densities and drives the design to a black
and white structure. To select the proper
value of p depend on Poisson's ratio ν,
see Bendsøe and Sigmund Material
interpolation schemes in topology
optimization [9]
2 4 Fig. 1. Initial design domain of half MBB beam.
P ≥ max , (in 2D) (5)
1−ν 1−ν

1−ν 3 1−ν Using a Matlab 2D finite element

P ≥ max 15 7−5ν , 2 1−2ν
(in 3D) (6) analysis with the properties of model as

3
in Table. 1, we will obtain the optimal 4. Effect of lower and upper
topology design as seen in Fig. 2. moving asymptotes on
Table. 1. Default design parameter for optimization convergence
topology optimization of half MBB beam.
Since the method of moving
Properties Values
asymptotes is a general method, so the
Young's modulus (E) 1 N/mm2 asymptotes can be adopted to be suitable
for seeking the demanded convergence
Poisson's ratio (ν) 0.3
of specific problems. A general
Force (F) 1N (although heuristic) rule for how to
(k) (k)
SIMP factor (P) 3 change the values of Lj and Uj is the
Volume fraction (vfrac) 0.5 following:
No. of elements (nelx*nely) 30*20 a) If the process tends to oscillate, then
Initial design variables 𝐱 0 0.5 it needs to be stabilized. This
stabilization may be accomplished by
moving the asymptotes closer to the
current iteration point.
b) If, instead, the process is monotonic
and slow, it needs to be relaxed. This
may be accomplished by moving the
asymptotes away from the current
iteration point. See [4].
The default rules for updating the lower
(k)
Fig. 2. The resulted topology optimization of half asymptotes Lj and the upper
MBB 30x20 element using MMA method.
(k)
asymptotes Uj will be now explained.
We can observe that with increasing The first two iterations, when k =1 and k
the number of elements, the method will =2; will be:
consume more time and it will be
(𝑘) (𝑘)
impractical. So by changing the range of 𝐿𝑗 = 𝑥𝑗 − 𝑎𝑠𝑦𝑖𝑛𝑡 (𝑥𝑗𝑚𝑎𝑥 − 𝑥𝑗𝑚𝑖𝑛 )
lower and upper moving asymptotes 𝑈𝑗
(𝑘)
= 𝑥𝑗
(𝑘)
+ 𝑎𝑠𝑦𝑖𝑛𝑡 (𝑥𝑗𝑚𝑎𝑥 − 𝑥𝑗𝑚𝑖𝑛 ) (8)
(𝐿𝑗 , 𝑈𝑗 ) closer or far away from the
design variables 𝑥𝑗 , the convergence In later iterations, when 𝑘 ≥ 3
process can be achieved. More details on (𝑘) (𝑘)
how to choose the moving asymptotes 𝐿𝑗 = 𝑥𝑗 − 𝛾𝑗 𝑘 𝑥𝑗 𝑘−1 − 𝐿𝑗 𝑘−1
and how to generate strictly conservative 𝑈𝑗 𝑘 = 𝑥𝑗 𝑘 + 𝛾𝑗 𝑘 𝑈𝑗 𝑘 −1 − 𝑥𝑗 𝑘−1 (9)
or more linearly approximations can be
found in [4]. This problem will be Where,
treated in next section. 𝑎𝑠𝑦𝑑𝑒𝑐𝑟 , 𝑖𝑓 𝑥𝑗
𝑘
− 𝑥𝑗
𝑘−1
𝑥𝑗
𝑘−1
− 𝑥𝑗
𝑘−2
<0
𝑘 𝑘 𝑘−1 𝑘−1 𝑘−2
𝛾𝑗 = 𝑎𝑠𝑦𝑖𝑛𝑐𝑟 , 𝑖𝑓 𝑥𝑗 − 𝑥𝑗 𝑥𝑗 − 𝑥𝑗 >0
𝑘 𝑘−1 𝑘−1 𝑘−2
1, 𝑖𝑓 𝑥𝑗 − 𝑥𝑗 𝑥𝑗 − 𝑥𝑗 =0
(10)

4
Where the default value of 𝒂𝒔𝒚𝒊𝒏𝒕 equals 180
0.5, 𝒂𝒔𝒚𝒅𝒆𝒄𝒓 equals 0.7 and 𝒂𝒔𝒚𝒊𝒏𝒄𝒓 equals 160
1.2, see [5]. 140

120
It is also found that there are some rules

Compliance
100
that can be used in the sub-problem file
80
in the MMA code, see [6]. That can be
60
added to the previous asymptotes rules.
40
These rules are:
20
(𝑘) (𝑘)
𝐿𝑗 𝑚𝑖𝑛 = 𝑥𝑗 − 𝑆𝑚𝑎𝑥 (𝑥𝑗𝑚𝑎𝑥 − 𝑥𝑗𝑚𝑖𝑛 ) 0

35
56
77
98
119
140
161
182
203
224
245
266
287
308
329
350
371
392
413
434
(𝑘) (𝑘)
𝐿𝑗 𝑚𝑎𝑥 = 𝑥𝑗 − 𝑆𝑚𝑖𝑛 (𝑥𝑗𝑚𝑎𝑥 − 𝑥𝑗𝑚𝑖𝑛 ) itration no.

(𝑘) (𝑘) Fig. 3. Convergence of 20x10 elements half MBB

𝑈𝑗 𝑚𝑖𝑛 = 𝑥𝑗 + 𝑆𝑚𝑖𝑛 (𝑥𝑗𝑚𝑎𝑥 − 𝑥𝑗𝑚𝑖𝑛 )
beam with 𝑺𝒎𝒂𝒙 and 𝑺𝒎𝒊𝒏 of 100 and 0.01.
(𝑘) (𝑘)
𝑈𝑗 𝑚𝑎𝑥 = 𝑥𝑗 + 𝑆𝑚𝑎𝑥 (𝑥𝑗𝑚𝑎𝑥 − 𝑥𝑗𝑚𝑖𝑛 ) Fig. 4 shows the optimization process
(𝑘) (𝑘)
at the second case and it indicates that
𝐿𝑗 (𝐿𝑗 𝑘 , 𝐿𝑗 𝑚𝑖𝑛 )
= max⁡ there is no convergence (it finally
(𝑘) (𝑘)
oscillates between two values of
𝐿𝑗 = min⁡(𝐿𝑗 𝑘 , 𝐿𝑗 𝑚𝑎𝑥 ) compliance C=99 , C=100) and does not
(𝑘) (𝑘) introduce the optimum solution.
𝑈𝑗 (𝑈𝑗 𝑘 , 𝑈𝑗 𝑚𝑎𝑥 )
= min⁡

(𝑘) (𝑘) 180

𝑈𝑗 = max⁡(𝑈𝑗 𝑘 , 𝑈𝑗 𝑚𝑖𝑛 ) (11)
160

140
Where the default values of 𝑺𝒎𝒂𝒙 and
120
𝑺𝒎𝒊𝒏 is 10 and 0.01. These values can be
Compliance

100
changed to suit any optimization
80
problem.
60
To illustrate the difference between 40
topology optimization using different 20
ranges of lower and upper asymptotes. 0
An example of half MBB beam with
35
70
105
140
175
210
245
280
315
350
385
420
455
490
525
560
595
630
665
700
735

20x10 elements is implemented with itration no.

different values of 𝑺𝒎𝒂𝒙 and 𝑺𝒎𝒊𝒏 in
equation (11) such as (100, 0.01) and Fig. 4. Divergence of 20x10 elements half MBB
(600, 0.06) for example. Fig. 3 shows beam with 𝑺𝒎𝒂𝒙 and 𝑺𝒎𝒊𝒏 of 600 and 0.06.
the convergence of topology
optimization process at first case which 5. Topology optimization of Two
shows that the convergence is steady dimensional (2D) models
after 440 iteration number, with
compliance C = 96.8030 N.mm. 5.1. Cantilever Beam
Using initial data as in Table.1 with
30 x 20 mm and the boundary
conditions, as shown in Fig. 5, we obtain
the final topological optimum design as

5
shown in Fig. 6 with a minimum
compliance of 36.76 N.mm. The
compliance at the beginning was equal
to 154.87 N.mm and the volume fraction
vfrac was 0.5 from the total volume.
5.2. Half MBB Beam
A topology optimization of MBB
beam with different mesh size with the
same initial design as shown in Fig. 1
and initial design parameters as in
Table.1 is illustrated in Fig. 8. Same
conclusions are reached as in section
5.1.

Fig. 5. Initial design of cantilever beam 30x20

elements

Fig. 8. Topology optimization for different mesh

Fig. 6. Topology design of cantilever beam of 30x20
elements
Topology optimization of a cantilever
beam with different mesh size is
summarized in Fig. 7 which shows that
the finer mesh leads to a topological
optimum design with lesser compliance
and finer shape than courser mesh.
Although it takes longer time, it has a
finer shape with higher resolution and
easy to determine the void and material
areas.
size of half MBB (a) 20x10 elements with
compliance C=96.9919 (b) 40x20 elements
C=84.1450 (c) 60x30 elements C=81.3742 (d) 80x40
elements C=80.0763 (e) 100x50 elements C=79.6268
(f) 120x60 elements C=79.8129.
6. Improvement of Numerical
Instabilities
Using the mesh independency
filtering developed by (Petersson and
Sigmund 1998) [8], the filter modifies
the design sensitivity of a specific
element based on a weighted average of
the element sensitivities in a fixed
neighborhood.

Where the sensitivity of the objective

function is found as:
𝜕𝑐
= −𝑝(𝑥𝑖 )𝑝−1 𝑢𝑖𝑇 𝑘𝑖 𝑢𝑖 (12)
𝜕𝑥 𝑖

Modifying the element sensitivities

using the mesh-independency filter to
Fig. 7. Topology optimization for different mesh involve the effect of sensitivity of
size for cantilever (a) 20x10 elements with neighbors [1] is found as:
compliance C=88.8544 (b) 40x20 elements
C=69.0953 (c) 60x30 elements C=66.6591. (d) 80x40
𝜕𝑐 1 𝑁 𝜕𝑐
=𝑥 𝑎=1 Ĥ𝑎 𝑥𝑎
elements C=65.0711 (e) 100x50 elements C=65.1185 (13)
𝑁
(f) 120x60 elements C=64.9388. 𝜕𝑥 𝑖 𝑖 𝑎 =1 Ĥ𝑎 𝜕𝑥 𝑎

6

Ĥ𝑎 = 𝑟𝑚𝑖𝑛 − 𝑑𝑖𝑠𝑡 𝑖, 𝑎
Where Ĥ𝑎 is the convolution operator
(weight factor); and the operator
𝑑𝑖𝑠𝑡(𝑖, 𝑎) is defined as the distance
between centre of element 𝑖 and centre
of element 𝑎. The convolution operator
Ĥ𝑎 is zero outside the filter area. Effect
of using mesh independency filter is
shown in Fig. 9 which is a cantilever
with mesh of 60x30 elements.
Fig. 9. Difference between topological design (a)
without mesh filtering and C= 64.6405. (b) with
mesh filter and C= 66.6591.
7. Topology Optimization Of
Three Dimensional (3D) Models
7. 1. 3D Cantilever Beam
Applying the method of moving
asymptotes (MMA) on three
dimensional models, and using ANSYS
finite element analysis with "Solid45"
linear and isotropic solid element type
and achieve an interface between the
ANSYS program and the MMA Matlab
code.
Fig. 10. 3D cantilever beam 20x10x8 elements (a)
initial design (b) final topology optimization design
Using different mesh size (i.e. solid
model of length*width*thickness,
7
20x10x8) as shown in Fig. (10-a) where
the initial design and boundary
conditions (loads and DOF) is
illustrated, and the final topological
optimization design is shown in Fig.
(10-b). Also Fig. 11 shows topological
optimum design for the same initial
design but with different mesh size of
30*20*10.
Fig. 11. Final topology optimization of 3D
cantilever beam 30x20x10 elements.
7. 2. 3D Half MBB Beam
Applying the method of moving
asymptotes (MMA) also on a three
dimensional half MBB beam. Applying
MMA algorithm on this model using a
Matlab code written by Krister Svanberg
and the finite element analysis using
ANSYS program. With a number of
elements of 20*10*8 as shown in Fig
(12-a), where the initial design and
boundary conditions (loads and DOF) is
illustrated. The final topological
optimization design is illustrated in Fig.
(12-b), where the compliance at first
iteration is C= 63.09 N.mm and at final
optimum iteration became C= 11.57
N.mm. Also Fig. 13 shows the topology
optimization of half MBB beam but with
number of elements of 30*20*10, the
compliance of this example at the first
iteration was C= 42.91 N.mm and at
final iteration becomes C= 6.81 at the
same volume fraction vfrac = 0.5.
Fig. 12. 3D MBB beam 20x10x8 elements (a) initial
design (b) final topology optimization design
Fig. 13. Final topology optimization of 3D half
MBB beam 30x20x10 elements.
8. Comparison between MMA,
SQP, OC, and HCA methods
In this section, a comparison between
MMA method and different methods
such as, sequential quadratic
programming (SQP), optimality criteria
(OC) that earlier presented by Bendsøe
and used in a program of "A 99 line
topology optimization code written in
Matlab" by O. Sigmund, and hybrid
cellular automata (HCA) with
proportional integral derivative (PID)
control rule [3] presented by Tovar et al.
is accomplished.
This comparison is achieved according
to the compliance values, the time
consumed and the resulted topological
shape in each method.
8
All topology optimization methods
use the same initial design and the same
conditions in order to achieve a fair
comparison (i. e. the same 2D finite
element analysis using Matlab). It is
important to define the type of PC that is
used and its CPU and memory
specifications to make time readings
more realistic. HP computer with
INTEL Core 2 Duo CPU 2.01GHZ and
2 GB of RAM is used here. Table. 2
shows the difference between each of
these methods in compliance, time and
resulted shape.
Fig. 14 shows that the compliance
values in each method almost near and
equal except the HCA method which
shows that with increasing the number
of elements, the compliance increase
than other methods, which indicate that
this method will not be a practical
method. Fig. 15 shows a large difference
in time consumed from one method to
another. Where at the beginning with a
small number of elements, it is observed
that the time consumed in all methods is
almost equal. But with increasing the
number of elements, the method of SQP
results in large increase in consumed
time which makes this method totally
impractical.
However MMA method comes the
second method in consumed time, its
consumed time is still much less than
SQP method, as shown in Fig. 15. The
time that MMA method consumes with
30x30 elements is 20 minutes, while
SQP method consumes 199 minutes.
While at 50x50 elements MMA method
consumes 80 minutes, despites SQP
method that consumes more than 25
hours and does not give the final
optimum solution because it becomes
out of memory of CPU.
Table.2 Difference
in time consumed,
compliance, and
resulted shape
between MMA,
SQP, OC, and HCA
methods

9
 MMA SQP OC HCA
used for any optimization problem,
35 while the SQP and MMA methods
30
are considered as general
optimization methods and handle
25
any objective function and any
Compliance

20 number of constraints.
15
If we wish to change the objective
10 function from the compliance
5 minimization to any other objective
0
function in this case, MMA method
10x10 20x20 30x30 40x40 50x50
will be the most suitable and
general optimization method.
no. of elements

Fig. 14. Difference in compliance between MMA, 9. Conclusions

SQP, OC, and HCA.
a. This paper has investigated the
MMA SQP OC HCA topology optimization using the
250
method of moving asymptotes and
200
other methods and shows the
difference between them.
time in minutes

150
b. The method of moving asymptotes is
100 considered as a general and flexible
method for structural topology
50 optimization problems, it can handle
any type of optimization problems.
0
10x10 20x20 30x30 40x40 50x50 c. Using the method of moving
no. of elements asymptotes let the one control the
Fig. 15. Difference in time consuming between
convergence, stability and speed of
MMA, SQP, OC, and HCA. the optimization process.

Also, if we wish to compare the d. With increasing the number of

HCA with OC method, it is elements, the range of asymptotes
observed that HCA is faster than the should be increased to save much
last one in convergence and time. The recommended range of
reaching the optimum solution and asymptotes that can be make the
gives minimum compliance values convergence stable as indicated in
for the small number of elements eqn. 11 is from 100 to 1200.
but with increasing this number we
can notice from Fig. 14 that HCA Acknowledgments
gives high compliance values than
other methods. The authors are grateful to Professor
Krister Svanberg, from the Royal
The HCA and OC methods Institute of Technology, Stockholm, for
considered as non general providing us with his MMA code.
optimization methods and cannot be
10
 References: [11] Roman Stainko, "Advanced Multilevel
Techniques to Topology
[1] Sigmund, O., "A 99 line topology Optimization", PhD thesis, Johannes
optimization code written in Matlab", Kepler University, Linz, Austria,
Struct. Multidisc. Optim. 21, pp 120– 2006.
127, Springer-Verlag, 2001.
[12] Bendsøe, M. P., "Optimal shape design
[2] Arora, J. S., "Introduction to Optimum as a material distribution problem",
Design", Handbook, Second Edition, Structural Optimization 1, pp 193-202,
Elsevier Inc ,2004. 1989.

[3] Tovar et al, "Topology Optimization

Using a Hybrid Cellular Automaton
Method With Local Control Rules",
Journal of Mechanical Design, Vol.
128, November 2006.

[4] Svanberg, K., "The Method of Moving

Asymptotes - A new method for
structural optimization", International
Journal for Numerical Methods in
Engineering 24, pp 359-373, 1987.

[5] Svanberg, K., "Some modeling aspects

for the Matlab implementation of
MMA" September 2004.

[6] MATLAB MMA code written by

Krister Svanberg, received by email.

[7] Bendsøe, M. P. and Sigmund, O.,

"Topology Optimization - Theory,
Methods and Applications",
Handbook, Springer Verlag, Berlin
Heidelberg, 2004.

[8] Sigmund, O., Petersson, J., Numerical

instabilities in topology optimization.
Struct. Optim. 16, pp 68-75, 1998.

[9] Bendsøe, M. P., Sigmund, O., "Material

interpolation schemes in topology
optimization", Applied Mechanics, 69,
pp 635-654, Springer Verlag, 1999.

[10] Rozvany, G. I. N., "A critical review of

established methods of structural
topology optimization", Struct
Multidisc Optim, 37, pp 217–237,
2009.

11

View publication stats