Documente Academic
Documente Profesional
Documente Cultură
DOI 10.1007/s00158-010-0495-9
RESEARCH PAPER
Received: 5 December 2008 / Revised: 22 January 2010 / Accepted: 27 January 2010 / Published online: 28 March 2010
c Springer-Verlag 2010
efficiency, as well as ease of integration with general- exchange method (EEM) falls under the same category as
purpose finite element analysis (FEA) codes. Besides the BESO and GA due to the use of heuristic relationships, but
existence of multiple local minima, topology optimization it has certain features that are quite distinct from the other
problems can also suffer from mesh dependency and the two methods. In the remaining portion of the paper, we pro-
formation of checkerboard regions. Some approaches to vide details of the EEM and describe the element exchange
combat the latter two problems have included the use of strategy, checkerboard control procedure, convergence cri-
heuristic mesh-independent filtering (Sigmund and Peterson teria, and the algorithmic parameters used in conjunction
1988), higher-order finite elements (Jog and Haber 1996), with different operations in EEM. The convergence prop-
perimeter control (Haber et al. 1996), alternative density– erties of EEM are illustrated along with a comparison to
stiffness interpolation schemes (Guo and Gu 2004), hyper- a known Michell truss structure. Moreover, the results for
bolic sine functions for the intermediate densities (Bruns several two- and three-dimensional problems of varying
2005), and techniques for producing better topologies with complexity are presented while making comparisons with
sharper solid–void solutions having greater stiffness (Fuchs the solutions found using other methods as reported in the
et al. 2005) as well as less checkerboards (Zhou et al. 2001; literature.
Poulsen 2002; Pomezanski et al. 2005).
Research efforts in non-gradient based topology opti-
mization have led to the development and application
of such methods as simulated biological growth (SBG; 2 General principle of element exchange
Mattheck and Burkhardt 1990), particle swarm optimization
(PSO; Fourie and Groenwold 2001), evolutionary structural The general principle of element exchange for the case of
optimization (ESO; Xie and Steven 1993), bidirectional compliance minimization is explained using a simple struc-
ESO (BESO; Querin et al. 1998), and metamorphic devel- tural system that is idealized by a combination of four
opment (MD; Liu et al. 2000). Some of these methods, linearly elastic springs and associated boundary conditions
together with other algorithms mimicking biological sys- as shown in Fig. 1. The total strain energy, Et stored in
tems such as genetic algorithms (GA; Goldberg 1989) and
cellular automata (CA; Kita and Toyoda 1999), have also
been used in the solution of sizing and shape optimization
problems. The use of binary design variables enables these
methods to produce a black–white (solid–void) optimal
topology that excludes any gray (i.e., fuzzy or intermedi- (a)
ate density) regions without using penalization. Another
advantage of the stochastic direct search methods is their
non-local search algorithms that can lead to a better solution
than the local optimum in the vicinity of the initial design
point. However, due to the need for a large number of func-
tion evaluations for the multitude of candidate designs, the
direct search methods tend to be computationally inefficient
(Fourie and Groenwold 2001; Werne 2006; Jakiela et al.
2000; Mei et al. 2007). To remedy the checkerboard prob-
lem, non-gradient based methods also resort to using mostly
heuristic schemes. Zhou and Rozvany (2001) discussed (b)
some of the shortcomings of non-gradient based methods
such as ESO, and more recently, Rozvany (2009) offered
a detailed critical review of SIMP and ESO by examin-
ing their mathematical foundations and highlighting their
differences in terms of solution accuracy and computational
efficiency.
In this paper, we introduce a new non-gradient based
topology optimization method that has many of the same
advantages and some of the shortcomings of the other
stochastic direct search methods but with noticeably better
computational efficiency. Named after the principal oper-
ation in the topology optimization strategy, the element Fig. 1 Spring system a before and b after element exchange operation
Element exchange method for topology optimization 217
the system is simply the sum of energy stored in individual tion problem is one of finding the optimal distribution of
springs found as solid and void elements that would
M
M
min f (ρ) = uT Ku = uTj K j u j = Ej
2E
1
4 4
Et = Ei = K i δi2 (1) j=1 j=1
2
M (2)
i=1 i=1 s.t. ρ j V j V0
j=1
where Ei is the energy in the ith spring defined in terms of ρmin ρ 1.0
the corresponding stiffness, K i and elongation, δ i .
where f (ρ) represents the total strain energy, ρ the vec-
Assuming that only two springs can be used for min-
tor of non-dimensional element densities treated as design
imizing the strain energy of the system in this example,
variables, u the vector of global generalized nodal displace-
the problem becomes one of deciding which two springs
ments, K the global stiffness matrix, M the total number
to keep and which ones to eliminate. The two springs that
of finite elements, with u j , K j and E j as the displace-
are kept create an optimal load path between the loaded
ment vector, stiffness matrix and strain energy of the jth
and supported points of the system. For simplicity, a solid
element, respectively. With ρ j and V j representing the non-
spring is assumed to have a stiffness of K s while a void
dimensional density and volume of the jth element, the
spring has a stiffness of K v = 0.001 K s . Using the ini-
constraint in (2) imposes an upper bound on the acceptable
tial distribution of springs shown in Fig. 1a and recognizing
volume fraction of solid elements in the design domain. To
that springs 1 and 2 are under equal axial force, elonga-
avoid ill-conditioning of stiffness matrix, the void elements
tions of springs 1 through 4 can be shown to be: δ1 =
δ δ 1,000 are assumed to have a density equal to ρmin , with a very
1,001 ≈ 1,000 ; δ2 = 1,001 δ ≈ δ; δ 3 = δ 4 = δ. Thus, small positive value.
the substitution of appropriate values into (1) gives
Ks −6 + 1 × 10−3 + 1 × 100 + 1 × 10−3 δ 2
Et = In EEM, stiffness of the jth element represented by
2 1 × 10 ≈
2 E j is linearly related to its non-dimensional density (i.e.,
2 Ksδ = 12 FK s . Since spring 1 is a solid spring with the
1 2
E j = ρ j E, where E is the Young’s modulus of the solid
lowest strain energy between the two solid springs, it will material), with ρ j treated as a discrete design variable ρ j ∈
be converted into a void spring in the next iteration while {ρmin , 1.0}, where ρmin = 0.001. For a uniformly dis-
spring 4—representing a void spring with the highest strain cretized domain of identical elements, the volume fraction
energy between the two void springs—will be converted constraint in (2) becomes a strict equality that is satisfied
into a solid spring. Figure 1b shows the updated layout after in every iteration. However, for non-uniform meshes, there
the element exchange operation is performed. Now, the total can be a small fluctuation of volume around the specified
strain
Ks
energy stored in the system can be shown to be Et =
−4 + 2.5 × 10−4 + 1 × 100 + 1 × 100 δ 2 ≈
limit, V0 .
2 2.5 × 10 The EEM algorithm, as depicted by the flowchart in
2
K s δ 2 = 14 FK s . While the number of solid springs is kept Fig. 2 applies to both single- and multiple-load case prob-
constant, the total strain energy of the system is reduced by lems. Once the domain is discretized into a uniform finite-
50%, signifying greater stiffness and smaller compliance. element mesh, the number of solid elements found as
Hence, by identifying and switching the less influential Ns = MV0 is randomly distributed throughout the design
solid spring into a void spring and the more influential void domain. If the mesh is non-uniform, then Ns would need
spring into a solid spring, a better (more efficient) load path to be adjusted accordingly to satisfy the specified volume
is created. fraction. All void elements are given a non-dimensional
By extending the problem to a continuum domain repre- material density of 0.001. The EEM parameters and their
sented by a finite element mesh, it would be possible to use recommended values appear later in the paper.
the element exchange as part of a more general algorithm Because of the stochastic operations that occur at
and solution procedure for finding the optimal topology. different stages of EEM together with the fact that the initial
distribution is selected in random, EEM procedure can take
different solution paths for the same topology optimization
problem. As will be shown later, if a problem has multi-
3 EEM algorithm ple local optima with nearly equal objective function values
(Kutylowski 2002), then it is possible for the EEM solu-
Here the EEM algorithm is discussed with focus on com- tion to converge to any one of these locations, which may
pliance minimization problems. With a continuum structure have different solid–void material distributions. However,
represented by a discretized domain of finite elements and the randomly selected initial design and the stochastic oper-
associated boundary conditions, the compliance minimiza- ations that occur at different stages of the solution process
218 M. Rouhi et al.
est strain energy density are converted into solid until the
exchange volume is balanced. Due to the use of discrete
density and variation in element geometry, it is possible to
encounter a small difference in volume between the two sets
of exchanged elements, which requires the relaxation of the
volume fraction constraint. Since the type of mesh used does
not change the basic framework of EEM, henceforth, the
mesh is assumed to be uniform.
The new layout is analyzed for strain energy, and
the element exchange operation is repeated. This pro-
cedure is continued for a specified number of iterations
before the checkerboard control (detection and elimination)
operation is performed, with further details provided later in
the paper.
After completion of several iterations, a subset of the
solid elements are randomly scattered in the void regions.
This so-called random shuf f le is similar to the mutation
operation in GA and serves a similar purpose in that it
enhances the chance of finding a better solution to the topol-
ogy optimization problem by exploring other regions of
the design space. For a given design problem, the princi-
pal operations (i.e., FEA, element exchange, checkerboard
control, and random shuffle) are repeated at different inter-
vals until a convergence criterion is satisfied. As in the case
of the other stochastic methods, a limit is imposed on the
number of iterations in order to stop the program when the
selected convergence criterion is too tight.
Figure 3a illustrates the evolution in topology from the
initial to optimum (minimum compliance) design point for a
two-dimensional domain using the EEM. Here, convergence
is defined as a nearly stationary topology with changes in
Fig. 2 Flowchart of EEM the strain energy below the specified threshold. The strain
energy history plot in Fig. 3a shows the general convergence
pattern in EEM. The spikes that appear at different intervals
make it more likely for EEM to find a better solution for a are mostly due to the random shuffle operation, although
more general problem. it is also possible to see an abrupt change in strain energy
With the initial design domain and boundary conditions during a routine element exchange operation.
specified, a static FEA is performed to find the strain energy In EEM, unlike ESO (Xie and Steven 1993; Querin et al.
distribution among the elements as well as the total strain 1998), void elements can be converted into solid and vice
energy of the structure as a whole. At the next step, a sub- versa. Note that in EEM, the void elements have small but
set of solid elements with the lowest strain energy density nonzero stiffness and density. Even if the initial random
amongst the solid elements are converted into void elements distribution of solid elements gives an appearance of a dis-
while a volumetrically equivalent number of void elements continuous load path (infeasible topology), EEM gradually
with the highest strain energy density amongst the void ele- connects all the solid elements in its search for an opti-
ments are converted into solid elements such that the volume mum topology. Furthermore, in EEM the solid and void
fraction remains fixed. In the case of a uniform mesh, all elements participating in the conversion operation are not
elements are geometrically identical; hence, volume frac- limited to any specific region of the design domain as is
tion remains constant by simply setting the number of solid the case with BESO (Querin et al. 1998). These character-
elements converted into voids and vice versa equal to each istics together with the random shuffle and overall topology
other. On the other hand, if the mesh is non-uniform, then optimization scheme help distinguish EEM from both ESO
for the specified exchange volume the solid elements with and BESO. The EEM algorithm is readily applicable to any
the lowest strain energy density are converted one by one two- or three-dimensional domain and boundary conditions,
into void, and similarly the void elements with the high- irrespective of its geometric or loading complexity.
Element exchange method for topology optimization 219
where k and Nmax denote the iteration counter and the pre-
scribed maximum number of EEM iterations, respectively.
If the value of Nmax is set too low (<100), EEM may not
have sufficient opportunity to explore the design space in
search of the optimum topology. On the other hand if it set
too high (>1,000), the computational time will significantly
increase without commensurate improvement in the opti-
mal topology. For the multitude of benchmark problems
discussed later in this paper, we have found that Nmax =
500 is acceptable. Moreover, for the total number of finite
elements, M and volume fraction, V0 specified in the prob-
lem, EEM parameter values in the range of 5% MV0 ≤
MEE−max ≤ 10% MV0 and 0.2% MV0 ≤ MEE−min ≤
0.4% MV0 are found to be effective in producing con-
verged solutions. Generally speaking, these parameters have
a wide range of acceptable values. Although they can affect
the convergence property of EEM, there is practically no
significant influence on the optimum design.
Fig. 3 Results of EEM for a tip-loaded cantilevered beam with a hole Fig. 4 a Solid checkerboard. b Void checkerboard. c Topology before
showing a the evolution in topology and the strain energy convergence checkerboard control. d After solid checkerboard elimination. e After
history, and b standard deviation of element strain energy distribution void checkerboard elimination
220 M. Rouhi et al.
Checkerboard patterns are generally undesirable and Although both element exchange and checkerboard con-
depending on the topology optimization methodology trol are effective tools in helping the EEM algorithm push
used, different strategies are employed to eliminate them toward the optimum topology, they are not sufficient. There-
(Sigmund and Peterson 1988; Zhou et al. 2001; Poulsen fore, an additional operation (i.e., random shuffle) is intro-
2002; Pomezanski et al. 2005). In EEM, a solid checker- duced. A random shuffle involves the selection of a subset
board element is defined as a solid element whose edges are of solid elements and their redistribution in void regions of
shared with void elements as shown in Fig. 4a, whereas a the domain. This action is analogous to the mutation oper-
void checkerboard element is the exact opposite as illus- ation in GA (Goldberg 1989) or craziness in PSO (Fourie
trated in Fig. 4b. Whether the dashed elements shown and Groenwold 2001), and is used for the same princi-
in Fig. 4a and b are solid or void will not change the pal reason, i.e., it helps prevent premature convergence or
checkerboard condition. insufficient exploration of the design domain in search of
Since in EEM the initial topology is a random distribu- optimum design. While preserving the specified volume
tion of solid elements per the specified volume fraction, it fraction, random shuffle can also help with convergence by
is natural to immediately encounter multiple checkerboard alleviating the occasional back and forth oscillation (oscil-
regions as shown in Fig. 3a. However, at the beginning, sev- latory exchange) in a subset of elements from solid to void
eral element exchange iterations (NCI = 5–10% Nmax ) are back to solid in successive element exchange operations.
allowed to proceed before actively searching for checker- Random shuffle will change the topology of the structure
board patterns. To eliminate checkerboard regions, first by introducing a random replacement of a group of solid
the solid checkerboard elements are identified and con- elements thereby alleviating the oscillatory exchange. The
verted into void elements as shown in Fig. 4c and d, and effect of random shuffle on the solution is shown later in
then the void checkerboard elements are converted into the paper. Two things can generally happen as a result of the
solid elements (Fig. 4d, e). The checkerboard search and random shuffle operation: (1) an abrupt change in stiffness
elimination step is repeated every NCC = 1–5% Nmax (Rouhi and Rais-Rohani 2008) and total strain energy as
iterations. To maintain the specified volume fraction, the shown by the spikes in Fig. 3a, and (2) creation of checker-
difference between the numbers (volumes) of the switched board regions. However, neither one of these side effects is
solid and void elements is randomly redistributed in the detrimental as both are corrected by the element exchange
design domain. It is possible for this random redistribution and checkerboard control operations of EEM.
of the difference to result in the creation of small checker- Random shuffle is third in the sequence of operations in
board region(s). However, as EEM procedure is continued, EEM and occurs at NRS = 2–10% Nmax iterations until the
these regions tend to gradually diminish before the final optimal topology is found. The number of elements partici-
topology emerges. The checkerboard elimination procedure pating in the random shuffle, MRS varies from its maximum
in EEM is heuristic and checkerboard elements are removed value, MRS−max at the beginning (k = 0) to its minimum
regardless of their impact on the overall compliance of the value, MRS−min as the solution approaches convergence.
structure. The value of MRS is found using the expression
It is important to note the interaction between checker-
board and the mesh size. As shown in exact analytical solu- MRS−max − MRS−min
k
MRS = int MRS−max − k (4)
tions for Michell truss structures (Lewinski and Rozvany Nmax
2008; Rozvany et al. 2006; Lewinski et al. 1994a), the opti-
mal layout is one with many narrow connecting members where k and Nmax denote the iteration counter and the pre-
or branches. However, if the mesh is relatively coarse, the scribed maximum number of EEM iterations, respectively.
elements in some of these branches would have a pixilated For the benchmark problems discussed later in this paper,
appearance and, hence, marked as checkerboard elements MRS−max = MEE−max , MRS−min = MEE−min have been
to be eliminated. As will be shown in the example problems found to be effective in producing converged solutions.
later, there is a greater chance for EEM to produce a topol-
ogy that approaches the analytical solution through mesh 3.4 Passive elements
refinement.
Although the current checkerboard control algorithm is Some continuum structures may contain solid and/or void
fairly effective, it does have some limitations in that it may sub-regions whose geometry and locations cannot be altered
not recognize checkerboard patterns that do not perfectly during topology optimization. As a matter of convenience
match the models shown in Fig. 4a and b. and meshing simplicity, the permanent voids (or solids) may
Element exchange method for topology optimization 221
be included in the finite element mesh but represented by a Figure 3a shows how the strain energy in EEM converges
series of passive elements with ρ pi = ρ min (or ρ pi = 1.0) to its minimum value for a cantilevered beam with a fixed
that will not be exchanged during the EEM solution pro- hole. Strain energy starts from an extremely large value
cess. The hole region in the structure shown in Fig. 3 was because of the randomly distributed elements in the initial
modeled using passive elements. step. However, after a few iterations, it reduces to roughly
the same order of magnitude as its minimum value. The
3.5 Convergence criteria continuation of the element exchange will refine the topol-
ogy toward its minimum strain energy as shown in Fig. 3a.
Increasing the number of iterations in EEM will usually lead Although there are occasional jumps in total strain energy
to a more refined optimal topology but at the expense of (due to element exchange or random shuffle operation), the
more function calls (i.e., additional FE solutions). Besides overall trend shows a gradual convergence. It should be
imposing a limit on the maximum number of iterations, two noted that similar spikes in the strain energy convergence
additional criteria are also used to establish a two-part con- plots have also been observed when using BESO (Querin
vergence condition in EEM-based topology optimization. et al. 1998).
The first convergence criterion considers the relative Figure 3b shows the plot of the standard deviation of
difference in the element strain energy distributions in two strain energy in individual solid elements at different stages
consecutive elite topologies. Here, elite topology refers to of the solution process. The trend indicates that the strain
the topology with the lowest strain energy obtained prior to energy density field is approaching a more uniform state,
the current iteration in the EEM procedure. Since it is pos- resembling the fully stressed design in optimality criteria
sible for two distinctly different topologies to have almost (Bendsoe and Sigmund 2002; Tanskanen 2002), as it makes
equal total strain energies (as noted in the SIMP based the most efficient use of available material in the design
results in Fig. 5), it is necessary to compare the element domain.
∼
strain energy distribution, as represented by the vector E , Another point that needs to be mentioned here is that,
for two consecutive elite topologies as in its current implementation, when EEM identifies an
∼ ∼
elite topology, no check is made whether any of the solid
E ce − E pe
elements that were previously redistributed as a result of
∼
≤ εE (5)
random shuffle operation still remain in the void regions. As
E pe
such, a few floating solid elements may appear as specks in
the void regions of the final topology in different example
where subscripts “ce” and “pe” refer to the current and pre- problems. Since the few floating elements have no sig-
vious elite topologies within Nmax iterations, respectively. nificant impact on the final topology, no attempt is made
The second convergence criterion examines the density to remove them in any of the presented solutions. Due to
distribution in two consecutive elite topologies. The domain the stochastic nature of EEM, another solution to the same
∼
topology is defined by vector D whose individual terms optimization problem may show different floating elements.
have binary values depending on the solid (1) or void
(0) property of the corresponding elements. Based on this
definition, the convergence criterion is defined as 4 Results for two-dimensional problems
∼ ∼
D ce − D pe
∼
≤ εt (6) Several benchmark problems are used to evaluate the per-
formance of EEM and to compare its solutions with those
D pe
topologies are fairly similar with both EEM and BESO solu-
tions showing one extra member than that in the Level Set
solution.
Table 4 Comparison of EEM results with enhanced GA solution for
4.2 Cantilevered beams model B2
Design domain & Boundary conditions
Model B1: Single tip force applied at the bottom X
F
Y
(nx, ny, V0) = (20, 47, 0.5) (nx, ny, V0) = (40, 94, 0.5)
(n x , n y , V0 ) = (90, 90, 0.35), EEM finds a minimum com- Fig. 7 a L-shaped domain
pliance solution after 167 iterations with the strain energy with distributed tip load and
of 85.1. The optimal EEM topology in Fig. 7c is compared optimal layouts based on b
Neighborhood Search method
with that of the Neighborhood Search method (Svanberg (Svanberg and Werme 2005)
and Werme 2005) in Fig. 7b. For the same mesh size, the and c EEM
solution based on Neighborhood Search method takes 378
iterations. In both cases, the larger members show the aver-
age orientation for the more finely distributed members over
the same domain (Lewinski and Rozvany 2008).
(a)
(b)
F2=F
(F1 and F2 represent separate load cases)
(c)
226 M. Rouhi et al.
Fig. 8 a L-shaped domain with the analytical case. However, the trend captured in the EEM
single tip load and optimal solutions appears to be in reasonable agreement with the
layouts based on b exact analytical solution. Since there is some randomness in the
analytical solution (Lewinski
and Rozvany 2008). c EEM EEM procedure, the number of lumped bars and their ori-
with coarse mesh. d EEM with entation may vary from one solution to another for the same
fine mesh problem.
(a)
(d)
Z DOF fixed
The convergence parameters (Nmax , εE and εt ) control topology, it makes entrapment in a local minimum less
the solution accuracy. Generally speaking, the EEM search likely.
for optimum topology becomes more rigorous by increas- The value selected for NCI should provide sufficient
ing the value of Nmax and decreasing the values of εE and opportunity for element exchange operation to improve
εt at the expense of increased computational time. As with upon the initial topology (random distribution of solid ele-
the other stochastic methods, if Nmax is too small, EEM ments), which may include many checkerboard regions.
may not be able to find an optimum solution. On the other Similarly, NCC should not coincide directly with NRS , as
hand, if εE and εt are too small, convergence may be hard random shuffle is likely to cause the creation of checker-
to achieve. As shown in Fig. 3, EEM finds the basic lay- board elements. Choosing NCI ≈ 5–10% Nmax and NCC ≈
out of the optimum topology in about 20 iterations with the 1% Nmax are found to be sufficient to allow both the ele-
remaining iterations devoted to refinement of the topology. ment exchange and random shuffle operations to improve
However, when the design domain includes multiple local the design before the accumulated checkerboard elements
optima, then it is possible for the topology to vary widely are identified and eliminated by the checkerboard control
during the solution sequence. procedure.
Since element exchange is the main operation in EEM, All of these parameters are flexible and reasonable devia-
the value selected for MEE is important. As noted tions from the suggested values may not dramatically affect
previously, MEE cannot be treated as a constant and must the final results. The values selected for the EEM parame-
be gradually reduced from its prescribed maximum value ters in some of the example problems presented earlier are
(MEE−max ) at the very beginning to its imposed mini- shown in the Appendix.
mum (MEE−min ) at the end. Based on the multitude of
problems examined, choosing MEE−max ∼ 5–10% and
MEE−min ∼ 0.2–0.4% of the solid elements (MV0 ) would be
appropriate. Generally speaking, selecting a relatively large 8 Concluding remarks
value for MEE−max will lead to the formation of the main
load path in the early stages of the optimization process. A stochastic direct search topology optimization method
However, it may also reduce the computational efficiency called element exchange method (EEM) was presented with
due to participation of a large number of elements in application to compliance minimization problems subject to
oscillatory exchange phenomenon discussed earlier. Like- a volume fraction constraint. The non-dimensional density
wise, choosing a relatively small value for MEE−min makes of each finite element was treated as a binary design vari-
the solution easier to converge with a more refined final able (ρvoid = 0.001 and ρsolid = 1) with a linear element
topology. density-stiffness relationship. The basic premise of the pro-
Similar to element exchange, the number of elements posed method is that by systematically converting the less
(MRS ) and the interval (NRS ) selected for random shuffle critical solid elements into void and the more critical void
are crucial to the success of EEM. Equation (4) provides elements into solid, an optimum topology will emerge. The
an acceptable reduction scheme for MRS in the range proper selection of EEM parameters assures convergence.
MRS−min ≤ MRS ≤ MRS−max . While starting with MRS = However, depending on the selected mesh density and the
MRS−max widens the domain of exploration for optimum desired level of clarity in the final topology, the number of
design (similar to a larger coefficient for the particle’s veloc- iterations required for convergence may vary.
ity in PSO (Fourie and Groenwold 2001), it may reduce Through the solution of several two- and three-
the computational efficiency by exchanging a large num- dimensional example problems for compliance minimiza-
ber of elements in a random fashion. On the other hand, tion, the accuracy and efficiency of EEM were examined
small MRS−min value helps with convergence while enhanc- and compared with different gradient based (e.g., SIMP)
ing the final topology. For problems with a large number and non-gradient based (e.g., GA, EBSO) methods as
of solid elements (i.e., large volume fraction) or large MV0 , reported in the literature. The results show that EEM is
EEM is less likely to get trapped in a loop (see Section IIIC) reasonably accurate in finding an optimum topology with
and the step size for random distribution can be increased to clear solid-void layout of material in the design domain.
improve the computational efficiency. Choosing NRS ≈ 2– As for computational efficiency, EEM is shown to be infe-
5% Nmax is found to be sufficient to help EEM not to get rior to gradient based methods (e.g., SIMP) but superior to
trapped in a local optimum while reducing the number of non-gradient based methods such as GA and PSO.
elements involved in oscillatory exchange.
Acknowledgments This material is based upon work supported
As shown in (3), a large value for Nmax decreases the
by the Department of Energy under Award Number DE-FC26-
rate of reduction in MEE , which increases the number of 06NT42755. The authors are also grateful to Prof. Rozvany for
EEM iterations. Although it delays the finding of the final valuable discussions on topics related to topology optimization.
230 M. Rouhi et al.
Appendix
Table 7 Table of values selected for EEM parameters in the example problems studied in the paper
Mei YL, Wang XM, Cheng GD (2007) Binary discrete method of Sigmund O, Peterson J (1988) Numerical instabilities in topology opti-
topology optimization. Appl Math Mech 28(6):707–719 mization: a survey on procedures dealing with checkerboards,
Olhoff N, Bendsoe MP, Rasmussen J (1991) On CAD integrated struc- mesh-dependencies and local minima. Struct Optim 16(1):68–75
tural topology and design optimization. Comput Methods Appl Svanberg K, Werme M (2005) A hierarchical neighborhood search
Mech Eng 89:259–279 method for topology optimization. Struct Multidisc Optim
Olhoff N, Ronholt E, Scheel J (1998) Topology optimization of three- 29:325–340
dimensional structures using optimum microstructures. Struct Tanskanen P (2002) The evolutionary structural optimization method:
Optim 16(1):1–18 theoretical aspects. Comput Methods Appl Mech Eng 191:5485–
Pomezanski V, Querin OM, Rozvany GIN (2005) CO-SIMP: extended 5498
SIMP algorithm with direct corner contact control. Struct Wang MY, Wang X (2003) Level set models for structural topology
Multidisc Optim 30:164–168 optimization. In: Proceedings of DETC’03, ASME 2003 design
Poulsen TA (2002) A simple scheme to prevent checkerboard patterns engineering technical conferences and ASME 2003 design engi-
and one-node connected hinges in topology optimization. Struct neering technical conferences and computers and information in
Multidisc Optim 24:396–399 engineering conference, 2–6 Sep. 2003, Chicago, IL
Querin OM, Steven GP, Xie YM (1998) Evolutionary structural opti- Wang SY, Tai K, Wang MY (2006) An enhanced genetic algorithm
mization (ESO) using a bidirectional algorithm. Comput Struct for structural topology optimization. Int J Numer Methods Eng
15(8):1031–1048 65(1):18–44
Rouhi M, Rais-Rohani M (2008) Topology optimization of con- Werne M (2006) Globally optimal benchmark solutions to some small-
tinuum structures using element exchange method. In: 49th scale discretized continuum topology optimization problems.
AIAA/ASME/ASCE/AHS/ASC structures, structural dynamics, Struct Multidisc Optim 32(3):259–262
and materials, 7–10 April 2008, Schaumburg, IL Xie YM, Steven GP (1993) A simple evolutionary procedure for
Rozvany GIN (2009) A critical review of established methods in struc- structural optimization. Eng Comput 49(5):885–896
tural topology optimization. Struct Multidisc Optim 37:217–237 Zhou M, Rozvany GIN (1991) The COC algorithm, part II: topo-
Rozvany GIN, Zhou M, Birker T (1992) Generalized shape optimiza- logical, geometry and generalized shape optimization. Comput
tion without homogenization. Struct Optim 4:250–252 Methods Appl Mech Eng 89(1–3):309–336
Rozvany GIN, Lewinski T, Querin OM, Logo J (2006) Quality con- Zhou M, Rozvany GIN (2001) On the validity of ESO type
trol in topology optimization using analytically derived bench- methods in topology optimization. Struct Multidisc Optim 21:
marks. In: 11th AIAA/ISSMO multidisciplinary analysis and 80–83
optimization conference, 6–8 September 2006, Portsmouth, VA Zhou M, Shyy YK, Thomas HL (2001) Checkerboard and minimum
Sigmund O (2001) A 99 line topology optimization code written in member size control in topology optimization. Struct Multidisc
Matlab. Struct Multidisc Optim 21(2):120–127 Optim 21:152–158