20th Analysis & Computation Specialty Conference © 2012 ASCE 310

Developing Benchmark Problems for Civil Structural Applications of Continuum

Topology Optimization

Colby C. Swan

Department of Civil & Environmental Engineering, University of Iowa, Iowa City, Iowa
Downloaded from ascelibrary.org by University of Saskatchewan on 01/12/15. Copyright ASCE. For personal use only; all rights reserved.

USA 52242; Email: colby-swan@uiowa.edu; Phone: 1-319-335-5831.

ABSTRACT

In this paper, consideration is given to candidate benchmark problems that utilize

continuum topology optimization to find long-span structural forms that maximize
stability. It is posited that by beginning the structural design process with forms of
optimal stability as opposed to forms that are inherently unstable, one avoids getting
locked into inferior designs that will be problematic at all stages. This issue is examined
here by introducing some problems in which the goal is to obtain sparse lightweight
structural forms that maximize critical buckling loads computed from linearized buckling
analysis.

INTRODUCTION
Over the past four decades computer-based structural optimization has evolved into
increasingly sophisticated forms within the research community and achieved a certain
level of maturity. Nevertheless, structural optimization is not yet widely used in civil
structural engineering practice. To facilitate its adoption in structural engineering
practice, the ASCE-SEI Committee on Optimal Structural Design (OSD) is engaged in an
effort to establish a range of well-crafted benchmark structural design problems
representative of the types encountered in the structural engineering community. The
purpose of the benchmark problems is two-fold: (1) showcase the current capabilities of
structural optimization to the structural design practice community; and (2) encourage the
structural optimization research community to focus attention on issues and problems of
interest to the practicing structural design community.
In this paper, we explore benchmark problems associated with the earliest phase of
the design process where given only a set of loads and a long span that they must bridge,
one seeks to identify the most suitable structural forms. Two issues of primary concern
for long spans structures that must be addressed in the benchmark problems are economy
of material usage and stability of the overall form. Since this stage of the design process
can have the greatest impact on the gross features of the final overall structural design it
is important to successfully identify those forms that are most promising.
It is well-recognized that for short-span structures, forms that utilize compression
and/or flexure can be most suitable resulting in arch and girder structures. However as
the span length increases, buckling stability concerns become preeminent making these
forms less attractive and raising the merit of structural forms that utilize relatively short
compression members in concert with long tension members. The need for buckling

20th Analysis and Computation Specialty Conference

 20th Analysis & Computation Specialty Conference © 2012 ASCE 311

stability in long-span structural designs is heightened by the fact that such designs, in
order to be competitive, are expected to be lightweight and sparse. To quantify the
notion of sparsity it is helpful to imagine the minimum-sized rectangular box that could
completely contain long-span structure and to denote its volume as the envelope volume
of the structure. The sparsity of the structure itself is captured in the ratio of its
cumulative material volume to its envelope volume. The higher the degree of sparsity
for a structure, the smaller its material to envelop volume ratio will be. Many long-span
structures realistically have material to envelope volume ratios of on the order of 1% or
Downloaded from ascelibrary.org by University of Saskatchewan on 01/12/15. Copyright ASCE. For personal use only; all rights reserved.

less. To simultaneously achieve good stability and sparsity in long-span structures,

cable-stayed or suspension forms tend to be most natural.
The benchmark problems to be explored in this paper will utilize continuum topology
optimization to find long-span structural forms that feature both structural stability and a
realistic degree of sparsity. Before doing so, it must be noted that because continuum
topology optimization has been researched and developed quite extensively over the past
two decades a number of benchmark problems already exist for insuring that the
numerical methods themselves are sound and will yield structural forms free of numerical
artifacts (such as checkerboarding), free of mesh-dependence, and convergent with mesh
refinement. All of these problems benchmark problems for mesh-independence involve
finding structural forms that yield structures of minimal mean linear elastic compliance
and are usually solved with material to envelope volume constraints on the order of 25%.
See [Bendsoe and Sigmund, 2003] for details on some of these problems. As a starting
point for the present work, it should be assumed that any computational method to be
employed on the new benchmark problems proposed here can be successfully applied to
solve any of these prior benchmark problems to establish that it does indeed yield mesh-
independent structural forms.
One of the major challenges associated with using continuum topology optimization
to find stable structural forms featuring a high degree of sparsity is that one must
necessarily use extremely refined mesh structures. Such refined meshes are needed to
resolve the layout and associated structural performance of small amounts of structural
material that occupy only about 1% of the structure’s envelope volume. The substantial
computational cost of using highly refined meshes has led to a variety of approaches to
deal with it. Maute (1998) used adaptive mesh refinement as the form of the structure
became clearer to coarsen the void regions and refine the material regions. A similar
approach is now extensively involved in finding optimal design forms using the
evolutionary structural optimization method (Bulman, 2001). In the so-called analysis
problem size reduction technique, researchers developed algorithms to automatically
remove the massive number of void nonstructural elements temporarily from the
structural analysis problem but allowed such elements to be reintroduced if needed if
structural material re-entered these regions during the optimization process (Bruns, 2003;
Rahmatalla, 2003a). In analysis problem size reduction approaches one can very
efficiently resolve very sparse material distributions with highly refined meshes while
enjoying the low computational cost of much coarser meshes.
This article presents a methodology for solving large-size sparse systems in
continuum structural topology design framework based on sequential refinement and size
reduction strategy in a new way that is conceptually simple and theoretically sound. In

20th Analysis and Computation Specialty Conference

 20th Analysis & Computation Specialty Conference © 2012 ASCE 312

sequential refinement, the proposed methodology solved a preliminary problem involving

relatively coarse meshes and moderate material usage constraint. The resulting optimal
form from this stage, which comprises the solid structural material, is then mapped onto a
finer mesh and with realistic material usage constraint. The new problem is then solved
where a new topological form is obtained. The mesh refinement process is repeated until
the final design converges to a realistic shape and performance with minimum error. It
should be noted here that the current methodology is based on interpolation of nodal
design variables using nodal basis functions (Rahmatalla, 2004) as opposed to element-
Downloaded from ascelibrary.org by University of Saskatchewan on 01/12/15. Copyright ASCE. For personal use only; all rights reserved.

based design variables. Although node-based design variables feature C0 continuity, they
must generally be used with perimeter constraints (Haber, 1996) to achieve design
convergence with mesh refinement.
PROBLEM FORMULATIONS
Structural Model and Material Layout Description
The objective of continuum structural optimization is to find a layout of a structural
material of specified properties in a defined spatial region that provides optimum
structural performance. In order that the widest possible class of structural layouts can be
considered, the methods in question must accommodate such generality. In this work, the
spatial region that the candidate structural models can occupy is denoted Ωs . To
facilitate both description of the structural material layout in Ωs and analysis of the
performance associated with each layout considered, the domain is discretized into a
relatively fine mesh of nodes and finite elements.
It is desired that at the end of the form-finding process, the structural region Ωs will
be decomposed into a collection of regions cumulatively denoted Ω A that contain the
structural material in question, and the remaining regions Ω B = Ω S \ Ω A that are devoid
of structural material. Since solution of the form-finding problem in this way is ill-posed,
an alternative relaxed approach is usually employed, wherein it is assumed that an
amorphous “mixture” of structural material A and a void material B exists throughout the
structural region Ωs . In each region of Ωs , the nature of the mixture is characterized by
a local volumetric density φ A of structural material A . By permitting mixtures, the
structural material A and a fictitious void material B are allowed to simultaneously
occupy an infinitesimal neighborhood about each Lagrangian point X ∈ Ω s . The
volumetric density of structural material A at a fixed Lagrangian point X ∈ Ω s is
denoted by φ A ( X ) and represents the fraction of an infinitesimal region surrounding
point X occupied by material A . Natural constraints upon the volumetric densities are:
0 ≤ φ A (X) ≤ 1; 0 ≤ φB (X) ≤ 1; φ A (X) + φB (X) = 1 . (1)

Clearly, when φA ( X ) = 1 the point X contains solid structural material, and when
φ A ( X) = 0 the point X is devoid of structural material. The last physical constraint of
Eq. (1) states that the material volume fractions at X are not independent and so one
need only be concerned with the layout of structural material A . The design of a

20th Analysis and Computation Specialty Conference

 20th Analysis & Computation Specialty Conference © 2012 ASCE 313

structure is here considered to be the spatial distribution of the structural material A in

Ωs .

To describe the distribution of material A throughout Ωs using a finite number of

design parameters, the volumetric density at each of the nodal points forms a set of
NUMNP design variables. These are then interpolated over the space of all intermediate
points in the structure using the nodal shape functions:
Downloaded from ascelibrary.org by University of Saskatchewan on 01/12/15. Copyright ASCE. For personal use only; all rights reserved.

NUMNP
φ ( X) = ∑i =1
bi Ni ( X) ∀ X ∈ Ωs (2)

where bi are the nodal volumetric density values associated with the structural material
and Ni ( X ) are the nodal shape functions. This approach yields a C 0 continuous design
variable field.
Given the finite element model of the structural region Ωs and the restraints, the
structural loads acting on this region are specified as a set of design loads. For each set
of design loadings, and for each realization of the design vector b = {b1 , b2 ,K , bNUMNP } , the
response of the structure will be analyzed. From the computed response both the
performance of the structure and its sensitivity to variations in the design variables are
computed.
Constitutive Mixing Rules
In the proposed design framework, each finite element comprising the spatial domain
Ωs of the structure will generally contain a spatially varying mixture of the structural and
void materials. It is necessary to prescribe the stiffness (or elastic moduli) of such
mixtures in terms of the stiffness characteristics of the solid material Csolid , those of the
fictitious void material Cvoid , and the local volumetric density of the structural material
φ ( X) . Here, the well-known powerlaw formula (Bendsøe, 1999) is used to accomplish
this task, providing the local effective stiffness of the mixture C* as:
C* = φ p C solid + (1 − φ p ) Cvoid (3)

where typically the mixing rule parameter p ∈ [1, 4] . With p = 1 , the Voigt rule of
mixtures is obtained which does not penalize mixtures, but which does yield a convex
formulation for compliance minimization problems (Swan, 1997) so that only one
solution exists for the design problem. With p = 4 , mixtures are penalized in the final
design, so that regions of Ωs tend to be either solid or void, but the optimization problem
is not convex, and will admit a number of solutions that satisfy the first order optimality
conditions.
Structural Analysis
For each design, a structural analysis problem is solved on the continuum domain Ωs.
In general terms, the structural analysis problem solved for each realization of the design

20th Analysis and Computation Specialty Conference

 20th Analysis & Computation Specialty Conference © 2012 ASCE 314

vector b is the following: Find the displacement field u( X) : Ω s a ℜ3 such that the
variational equilibrium problem is solved:
∫ σ : δε dΩs = ∫ h ⋅ δu dΓs + ∫ ρg ⋅ δu dΩs (4)
Ωs Γs Ωs

where σ(X) is the local stress field in the structure; h is a traction vector consistent with
the design loads being applied to the structure; ρ ( X ) is the local mass density of the
Downloaded from ascelibrary.org by University of Saskatchewan on 01/12/15. Copyright ASCE. For personal use only; all rights reserved.

solid-void mixture; g is the gravitational body force vector; δ u is a kinematically

admissible variational displacement field; and δ ε is the corresponding variational strain
field. In the structural model, the material features linear elastic behavior such that
σ = C* : ε where the effective elasticity tensor is design dependent and prescribed in
accordance with Eq. (3). The matrix problem associated with variational equilibrium of
the discrete finite element structural model, for which, u ( X ) = ∑ i N i ui is

0 = K ⋅ u − f ext = f int − f ext (5)

where:
K LM L M
jk = ∫ B mj C *mn B nk dΩ s
Ωs
int
f = K ⋅ u = ∫ B T σ dΩ s (6)
Ωs

f ext = ∫ N h dΓs + ∫ N ρg dΩ s .
Γs Ωs

In all of the above, N denote the nodal shape functions and B denote the standard strain-
displacement matrices. The structural stiffness matrix K is positive definite due to the
characteristics of the effective elasticity tensor C*, and this guarantees a unique solution
to the structural analysis problem for each realization of the design b.
Once the equilibrium solution to the problem of Eq. (5) is obtained, then the
linearized geometrical stiffness matrix G can be computed based on the stress field σ in
the structure:
G LM ∫N
L
jk = ,m N ,Mn σ mn δ jk dΩ s (7)
Ωs

It is worth noting that G is not necessarily positive definite but rather depends heavily
upon the nature of the stress field in the structure. A purely tensile stress field clearly
makes G positive definite, although for any compressive stresses, G will not be positive
definite.
Structural Performance Measures
As noted previously, structural topology design problems can be formulated in a
number of alternative ways through utilization of assorted objective and constraint
functions. Generally, the objective function measures the performance of the structure,
and the constraint function limits the amount of structural material that can be used,
although the roles can be reversed equally well. The significant aspects of using CSTO to

20th Analysis and Computation Specialty Conference

 20th Analysis & Computation Specialty Conference © 2012 ASCE 315

design large-scale sparse structures can be demonstrated here using the linear elastic
structural compliance performance measure and the critical load buckling factor.
Linear elastic structural compliance
If a structure features a linear elastic response behavior, the resulting displacement
field u in response to a set of applied external loads f ext will be simply u = K −1 ⋅ f ext
where K represents the stiffness matrix of the structure. For a given set of loads, the
compliance Π (b ) of the structure is simply
Downloaded from ascelibrary.org by University of Saskatchewan on 01/12/15. Copyright ASCE. For personal use only; all rights reserved.

1
Π(b) = f ext ⋅ u . (8)
2
Structural concept designs b that are stiff with respect to the applied loads will have
small compliance Π (b ) , whereas structures that are not stiff with respect to the applied
loads will have large compliance. To facilitate usage of gradient-based optimization
solution techniques, it is necessary to compute the design derivatives of the compliance
function. It can be shown that the design gradient of structural compliance is provided by
the following expression:
dΠ 1 ⎛ ∂K ∂f ext ⎞
= − u ⋅ ⎜⎜ ⋅u − ⎟ (9)
db 2 ⎝ ∂b ∂b ⎟⎠
Linearized buckling performance measure
Linearized buckling eigenvalue analysis proceeds as follows: A prescribed force
loading fext is applied to the structure with its magnitude necessarily being less than that
required to induce geometric instability in the structure. Once the resulting linear,
elastostatic displacement solution u = {ui } ∈ R N in response to the applied loading f ext is
( )
obtained K ⋅ u = f ext , where K is the elastic material stiffness matrix, then the following
eigenvalue problem is solved
[K (b) + λG (u, b)]⋅ ψ = 0 (10)

In the preceding, b = {b e } ∈ R M is again the vector of design variables; G (u , b ) is

the linearized geometric stiffness matrix; λ = −(ψ ⋅ K ⋅ ψ ) / (ψ ⋅ G ⋅ ψ ) is an eigenvalue
denoting the magnitude by which f ext must be scaled to create instability in the structure,
and ψ is a normalized eigenvector satisfying ψ ⋅ K ⋅ ψ = 1 . To avoid numerical
difficulties in the solution of (10) stemming from the indefinite characteristics of G , it is
common (Bathe 1996) to solve a modified eigenvalue problem that deals with two
positive definite matrices.
[(K + G ) − γ K ] ⋅ ψ = 0 (11)
where
λ −1 1
γ = ⇔ λ= . (12)
λ 1−γ

20th Analysis and Computation Specialty Conference

 20th Analysis & Computation Specialty Conference © 2012 ASCE 316

In Eq. (11), the matrix K is positive definite irrespective of the loading applied to the
structure, whereas the matrix (K+G) will only be positive definite when the magnitude of
the loading applied to the structural model is less than the critical magnitude that creates
instability in accordance with linearized buckling theory.
The design problem is formulated to maximize the calculated minimum-buckling load
factor λ , and accordingly the objective function f E to be minimized for this problem
would simply be the reciprocal of the lowest eigenvalue λ as follows.
Downloaded from ascelibrary.org by University of Saskatchewan on 01/12/15. Copyright ASCE. For personal use only; all rights reserved.

1
f E (u, b) = (13)
min(λ )
The optimization problem is thus stated to minimize the reciprocal of the first (or
minimum) critical buckling load as follows
⎛1⎞ ⎛ ψ ⋅G ⋅ψ ⎞
f E = min ⎜ ⎟ = min ⎜⎜ − ⎟ (14)
ψ ⋅ K ⋅ ψ ⎟⎠
min max
b ,u λ
b ,u
⎝ ⎠ b ,u ⎝ ψ ≠0

subject to the normal bound constraints on the design variables (1), the linear structural
equilibrium state equation (5), and a constraint on material resources.
The design gradient of the objective function can be expressed as:
df E ∂f E ∂f E ∂u
= + ⋅ (15)
db ∂b ∂u ∂b
∂u
To avoid explicit computation of the term ∂b , adjoint design sensitivity analysis is
employed by augmenting the objective function f E with the equilibrium state equation as
follows
Ξ = fE + ua ⋅ r (16)
where u a is the adjoint displacement vector which functions as a matrix of Lagrange
multipliers and determined by the solution of a linear adjoint problem. The design
derivative of the augmented Lagrangian is then written as follows:
⎡ ∂f E ∂r a ⎤ ∂u ⎛ ∂u ⎞
a
dΞ ⎛ ∂f E a ∂r ⎞
=⎜ +u ⋅ ⎟+⎢ + ⋅u ⎥ ⋅ ⎜
+ ⎜r ⋅ ⎟⎟ (17)
db ⎝ ∂b ∂b ⎠ ⎣ ∂u ∂u ⎦ ∂b ⎝ ∂b ⎠
The last term of Eq. (17) vanishes due to satisfaction of the equilibrium constraint (r=0),
and the second term can be made to vanish by selecting the adjoint displacement vector
to solve the following linear adjoint equality statement
∂G
K ⋅ ua = ψ ⋅ ⋅ψ. (18)
∂u
Since it can be shown that dΞ = df E , it follows that the design gradient expression for
db db
the objective function is

20th Analysis and Computation Specialty Conference

 20th Analysis & Computation Specialty Conference © 2012 ASCE 317

df E ⎛ ∂G 1 ∂K ⎞ ⎛ ∂K ∂f ext ⎞
= −ψ ⋅ ⎜ + ⎟⋅ψ +u
a
⋅ ⎜⎜ ⋅u − ⎟⎟ . (19)
db ⎝ ∂b λ ∂b ⎠ ⎝ ∂b ∂b ⎠
The preceding expression is valid only when the minimum eigenvalue is a simple, or
non-repeated, eigenvalue. When the minimum eigenvalue is nonsimple, or repeated, the
variation of the eigenvalue in design space is non-smooth, and direct usage of the
expression in Eq. (19) is technically incorrect (Choi et al 1983; Seyranian et al 1994).
Resolution of this issue is nontrivial, although it can be ameliorated somewhat by using
Downloaded from ascelibrary.org by University of Saskatchewan on 01/12/15. Copyright ASCE. For personal use only; all rights reserved.

small and variable move limits in the design optimization process. Despite this
challenge, designs that successfully maximize the buckling stability of a structural system
can nevertheless be obtained. Further details on formulation and solution of stability
design problems using linearized buckling theory are provided in Rahmatalla and
Swan(10).
ANALYSIS PROBLEM SIZE REDUCTION TECHNIQUE
In continuum topology optimization of sparse structures with very limited structural
material usage the design optimization process will typically remove all structural
material from regions not on the load path. For sparse structures in particular, once the
structural form begins to develop, the number of empty or void elements can greatly
exceed the number of solid elements. Since these large void regions do not contribute to
the function of the structure they can be neglected during structural analysis. In fact,
neglecting these void regions during analysis greatly reduces the size of the analysis
problem to be solved resulting in very substantial computational savings each time the
structural model requires analysis. An automated algorithm for identifying such regions
and removing them from the structural analysis problem each time a new design vector b
is considered is presented below. It is worth noting that the procedure proposed and
investigated here is reversible in that it permits low-density regions of the structure to
return as high-density structural regions even after they have previously been removed
from consideration during structural analysis.
The essence of the proposed analysis problem reduction technique can be captured in
the three steps listed below:
1. All finite elements in the structural analysis model that are devoid of solid
material, or nearly so, are identified as “void” elements. (Typically, in the examples
presented below, if an element’s volume fraction of solid material is less than or equal to
.002, it is identified as “void”.)
2. All nodes that are members only of “void” elements are identified as “prime”
nodes. The degrees of freedom of such “prime” nodes are restrained, reducing the size of
the analysis problem.
3. If only “prime” nodes comprise an element, that element is then denoted as a
“prime” element. Such “prime” elements are then neglected in the structural analysis
problem so that if they undergo excessive distortion it does not create any singularities in
the system of finite element equations. It is worth noting, that “prime” elements are those
that are surrounded by “void” elements.

20th Analysis and Computation Specialty Conference

 20th Analysis & Computation Specialty Conference © 2012 ASCE 318

A graphical description
d supporting
s th
he explanation of this teechnique forr reducing thhe
analy
ysis problem is presented
d in Figure 1.
The
T current reduction teechniques have h proven to be bothh robust andd efficient iin
solvinng the bencchmarks pro oblems conssidered beloow. The teechniques aare especiallly
poweerful and efffective wheen applied in i design prroblems invvolving extremely sparsse
structtures, since highly refined meshes are
a needed w when very sttringent mateerial resourcce
consttraints are im
mposed. Wh hen a fine mesh
m is emp loyed with a very limited amount oof
Downloaded from ascelibrary.org by University of Saskatchewan on 01/12/15. Copyright ASCE. For personal use only; all rights reserved.

structtural materiaal, the propo

osed reductio
on techniquees will allow
w for dramattic savings iin
compputing effort.

Figure 1. Schematic of partial meesh to illustraate

S S S S S S S S
analysis pproblem reducttion techniquee. Nodes havinng
vanishing design variaable values arre denoted wiith
S S S S S S S S
open circcles; filled cirrcles denote nnodes associatted
S S V V V V V V with nonzzero design vvalues; nodes represented bby
open circcles with X’ss are “prime”” nodes whoose
S S V P P P P P degrees off freedom are rrestrained in thhe size reductioon
method. Elements dessignated with “S” are at leaast
S S V V V V V V partially solid and thoose with “V”” are devoid of
material. Those designnated with “P”” are prime annd
S S S S S S S S need not bbe considered dduring structurral analysis sinnce
all of theirr degrees of freeedom are resttrained.
S S S S S S S S

S S S S S S S S

BENCHMARK EXAMPLE
ES
The
T validity of o the propoosed methoddologies is nnow demonsttrated on tw wo benchmarrk
probllems. In bo oth problemss presented the structuraal material iis steel and the fictitiouus
void material is given a stifffness equal to that of sstructural maaterial scaleed down by a
factorr of 10-6.
Stabiility Design of Sparse Structure
S un
nder Fixed--Fixed End C
Conditions
In
n this probleem, a point lo
oad is applieed to the topp central porttion of a struuctural regioon
with fixed supporrts at both laateral edges of the domaain. A sparsee, stable struuctural desiggn
ught that carrries the app
is sou plied load baack to the suupports. Thiis problem iis solved herre
by optimizing
o the
t materiall layout succh that thee minimum buckling eeigenvalue is
maxim mized. The problem has also been solved
s usingg different foormulations and objectivve
functtions by Buh hl (2000), Gea (2001), and
a Rahmataalla (2003a).. In this speccific examplle
5
(Figuure 2a), a po
oint load of magnitude
m 1.0·10
1 is appplied to the domain as shown whicch
has relative
r dimeensions of 100
1 by 50. The optimizzation probllem is solveed to find thhe
consttrained material layout thhat maximizzes the minim mum bucklinng eigenvaluue (14).
The
T material layout optim mization prob blem was firrst solved onn a relativelyy coarse messh
of 10
00 by 50 billinear quadriilateral conttinuum elem ments. The m material layyout shown iin
Figurre 2b was obtained fro om a starting “design”” of structuural materiaal completelly
occuppying the enntire the strructural regiion, followeed by impossition of a constraint oon

20th Analysis and Computation Specialty Conference

 20th Analysis & Computation Specialty Conference © 2012 ASCE 319

structtural materiaal: Vmaterial ≤ 0.20Vdomain where

w Vdomainn represents the volume of the 100 bby
50 sttructural do omain. Thee design sollution show wn was obtaained using a powerlaw w
parammeter [p=1.7 75]. The maaterial layoutt in Fig. 2b iis both heavvy and somew what difficuult
to intterpret. Con nsequently, th he design shhown was theen projectedd onto a unifo formly refineed
mesh h of 200 by 100 bilineaar continuum m elements.. With thiss refine messh, a reduceed
materrial usage co onstraint of Vmaterial ≤ 0.080
0 Vdomain waas imposed iin addition tto a perimeteer
consttraint P ≤ 2 (l + h ) wheere l and h are the lat
1
ateral and veertical dimennsions of thhe
Downloaded from ascelibrary.org by University of Saskatchewan on 01/12/15. Copyright ASCE. For personal use only; all rights reserved.

structtural region in Fig. 2a. The materiaal layout wass then optim mized on the refined messh
for 10 00 SLP (seq quential lineaar programm ming) iteratioons with thee result beingg as shown iin
Fig. 2c.
2 To obtaain a very cllear, yet sparrse, structurral design, thhe design off Fig. 2c. waas
then mapped ontto a mesh of o 400 by 2002 bilinear continuum elements. The materiaal
usagee constraint was tighten ned to Vmateerial ≤ 0.030Vdomain , and a mixing ruule powerlaw w
parammeter of [p p=4] was employed along with a tighteneed perimeteer constrainnt
P ≤ 3 (l + h ) . Th
1
he final, optimal materiaal layout dessign achieved is as show wn in Fig. 2dd.
This final design n shown in Fig.F 2d. is similar to thoose obtained in previouss studies citeed
previiously involv ving this problem.

50

100 b))
a)

c) d))
Fig. 2. Stability design
d of sparrse structure fo
or fixed-fixed eend conditionss. a) shows thhe design domaain,
support conditions, and loading;; b) design obtaained on a messh of 100 x 50 bilinear elemeents with materrial
usagge constraint of 0.10; b) prreceding desig gn mapped ontto mesh of 2000 x 100 elem ments and furtther
optiimized while imposing
i a tig
ghtened materiial usage consstraint of 0.05,, and a perimeeter constraintt of
500
00; d) precedin ng design mapp ped onto mesh of 400 x 200 elements and further optimizzed with materrial
usagge constraint of
o 0.03 and periimeter constraiint of 3750.

g Span Desig
Long gn Problem
m for Complliance and S
Stability
For main
m span leengths greater than 1,00 00 meters, suuspension bbridges that uuse primarilly
tensio
on to carry both
b the design loads annd their ownn weight are generally opptimal in thaat
the primary strucctural elemennts are not subject
s to buuckling. Herre the layoutt optimizatioon

20th Analysis and Computation Specialty Conference

 20th Analysis & Computation Specialty Conference © 2012 ASCE 320

of a very long span (2,250m) bridge is considered in which the candidate structural
region is selected to lie at or above the traffic deck level as shown in Fig. 3a. The design
traffic loading on the bridge is 10kPa uniformly distributed on the deck level. The
structural material usage is limited to 12.5% of the envelope volume. The envelope
volume is meshed with 10,000 bilinear continuum finite elements, and the problem is
solved first to maximize the minimum critical buckling load (Fig. 3b) and then to
minimize the structural compliance of the structure under the traffic loading (Figs. 3c,
3d.). Since the compliance-minimizing design shown in Figure 4c is somewhat difficult
Downloaded from ascelibrary.org by University of Saskatchewan on 01/12/15. Copyright ASCE. For personal use only; all rights reserved.

to interpret, the problem was re-solved at a substantially higher mesh resolution, and is
shown in Figure 3d.
In both of the compliance-minimizing designs, the proposed methodology yields designs
that use flexure of a flying beam-like structure whose supports are cantilevered out into
the span. While these designs are in many ways quite plausible and realistic, particularly
regarding the distributed support of the deck by systems of cables suspended from the
compression cord of the beam, the very serious problem with these designs is that the top
chord of the beam-like structure is very long, slender, and in compression under the
design loading. If these compliance-minimizing concept designs were to be taken into a
secondary more detailed design stage, the long compression cord members would need to
be sized very large to avoid buckling, and the resulting design would be excessively
heavy and inefficient. It is worth noting that by utilizing this linear elastic compliance
minimizing formulation, similar topologies will result regardless the length of the span or
the magnitude of the external loads. On the other hand, the layout design of Fig. 3b
shows a conceptual design that maximizes the linearized critical buckling load factor. As
can be seen, the suspension concept design solution uses compression in the relatively
stout “towers” that elevate the cable, tension in the long suspension cable that extends
across the span, and tension in the relatively short hanger system that suspends the deck
from the suspension cable. That the proposed formulation produces a suspension type
concept design (Fig. 3b) resembling actual long span bridges in usage today is an
encouraging development. It is worth noting here that a concept design similar to that in
Fig. 3b was obtained by Oberndorfer et al (1996) with a discrete ground structure
topology optimization method that considered only local buckling instabilities.

SUMMARY
In this work, two benchmark problems have been introduced and solved to find forms
of sparse, long-span structures with optimal stability. In this work, the sparse forms have
been optimized to maximize the minimum linearized buckling eigenvalue. Structural
forms obtained in this manner use both compression and tension members, but the
compression member are necessarily short.

20th Analysis and Computation Specialty Conference

 20th Analysis & Computation Specialty Conference © 2012 ASCE 321
3000 m
200 m

375
3 m
(a) 375 m
50 elements

200 elements

50 elements (b)
Downloaded from ascelibrary.org by University of Saskatchewan on 01/12/15. Copyright ASCE. For personal use only; all rights reserved.

200 elements

60 elements (c)

900 elementts

(d)

Fig. 3.
3 (a) Design domain with loading and support
s condiitions; (b) Forrm obtained bby maximizinng
the minimum
m buck m obtained byy minimizingg the structuraal compliance;
kling eigenvaalue; (c) Form
(d) Co
ompliance miinimizing form
m obtained with
w a more reefined model.

REFE
ERENCES
Bathee, K.J.:1996: Finite Elem
ment Procedu
ures, Prenticce-Hall.

Bend
dsøe, M.P.; Kikuchi,
K N. 1988:
1 Generrating optimaal topology in structurall design usinng
a homogeenization meethod, Compu ut. Meth. Apppl. Mech. E
Engng, 71, 197-224.

Bend
dsøe, M.P.; Sigmund, O.O 1999: Material
M interrpolations inn topology optimizationn.
Archive of Applied Mechanics,
M 69
9, 635-54.

Bend
dsøe, M.P.; Sigmund, O.O 2003: Topology O
Optimizationn: Theory Methods annd
ons, Springeer-Verlag.
Applicatio
Bruns, T. E.; Torrtorelli, D. A.
A 2003: An element rem
moval and reeintroductionn strategy foor
the topolo
ogy optimization of struuctures and compliant m
mechanisms,, Int. J. Num m.
Meth. Eng gng, 57(10), 1413-1430.

Buhl,, T.; Pederseen, W.; Sigm mund, O. 20 000: Stiffnesss design off geometricaally nonlineaar
structuress using topollogy optimizzation, Structt. Multidisc. Optim., 19, 93-104.

Bulm
man, S.; Sien
nz, J.; Hintoon, E. 2001:: Comparisoon between algorithms for structuraal
topology optimizatio on using a series of bbenchmark studies, Coomputers annd
Structuress, 79(12), 12
203-18.
Choi,, K.K.; Haugg, E.J.; Seon
ng, H.G.: 198
83.An iteratiive method ffor finite dim
mensional
structural optimization problems with
w repeateed eigenvaluues.” Int. J. N Num. Meth.
Engng., 19, 93-112.

20th Analysis and Computation Specialty Conference

 20th Analysis & Computation Specialty Conference © 2012 ASCE 322

Gea, H. C.; Luo, J. 2001: Topology optimization of structures with geometrical

nonlinearities, Computers and Structures, 79, 1977-85.

Haber, R.B.; Jog, C.S.; Bendsøe, M.P. 1996: A new approach to variable-topology shape
design using a constraint on perimeter. Struct. Optim. 11, 1–12.

Kosaka, I.; Swan, C.C. 1999: A symmetry reduction method for continuum structural
topology optimization. Computers and Structures, 70(1), 47-61
Downloaded from ascelibrary.org by University of Saskatchewan on 01/12/15. Copyright ASCE. For personal use only; all rights reserved.

Maute, K.; Schwarz, S.; Ramm, E. 1998: Adaptive topology optimization of elastoplastic
structures, Struct. Optim. 15, 81-91.

Michell, A.G.M. 1904: The limits of economy in frame structures. Philosophical

Magazine Sect. 6, 8(47), 589-597.

Oberndorfer J.M.; Achtziger W.; Hörnlein, H.R.E.M: 1996. Two approaches for truss
topology optimization: a comparison for practical use. Struct. Opt., 11, 137-144.

Rahmatalla, S.; Swan, C.C. 2003a: Continuum topology optimization of buckling-

sensitive structures, AIAA J. 41(5), 1180-1189.

Rahmatalla, S.; Swan, C.C. 2003b: Form-finding of sparse structures using continuum
topology optimization. J. Struct. Engng. 129(12), 1707-16.

Rahmatalla, S.; Swan, C.C. 2004: A Q4/Q4 continuum structural topology optimization
implementation. Struct. Multidis. Optim. 27, 130-135.

Swan, C.C.; Kosaka, I. 1997: Voigt-Reuss topology optimization for structures with
nonlinear material behaviors, Int. J. Numer. Meth. Engng., 40, 3785-814.

20th Analysis and Computation Specialty Conference