You are on page 1of 12

Struct Multidisc Optim (2011) 43:561572

DOI 10.1007/s00158-010-0585-8
RESEARCH PAPER
A general formulation of structural topology optimization
for maximizing structural stiffness
Fei Niu Shengli Xu Gengdong Cheng
Received: 15 March 2010 / Revised: 27 July 2010 / Accepted: 9 October 2010 / Published online: 3 November 2010
c Springer-Verlag 2010
Abstract This paper presents a general formulation of
structural topology optimization for maximizing structure
stiffness with mixed boundary conditions, i.e. with both
external forces and prescribed non-zero displacement. In
such formulation, the objective function is equal to work
done by the given external forces minus work done by
the reaction forces on prescribed non-zero displacement.
When only one type of boundary condition is specified, it
degenerates to the formulation of minimum structural com-
pliance design (with external force) and maximumstructural
strain energy design (with prescribed non-zero displace-
ment). However, regardless of boundary condition types,
the sensitivity of such objective function with respect to
artificial element density is always proportional to the neg-
ative of average strain energy density. We show that this
formulation provides optimum design for both discrete and
continuum structures.
Keywords Topology optimization Structural stiffness
Prescribed non-zero displacement Optimality condition
F. Niu G. Cheng (B)
State Key Laboratory of Structural Analysis for Industrial
Equipments, Department of Engineering Mechanics,
Dalian University of Technology, Dalian,
Peoples Republic of China
e-mail: chenggd@dlut.edu.cn
F. Niu
e-mail: niufei@mail.dlut.edu.cn
S. Xu
School of Energy and Power Engineering, Dalian University
of Technology, Dalian, Peoples Republic of China
e-mail: shengli.xu.xu@gmail.com
1 Introduction
Topology optimization method enables designers to find a
suitable structural layout for the desired structural perfor-
mance. With the development of many important formula-
tions and efficient numerical algorithms, structural topology
optimization has found applications in many areas (Bendse
and Sigmund 2003). Various structural topology optimiza-
tion formulations with different objectives/constraints have
been studied to meet specific performance requirements
in engineering. All continuum mechanics problems are
strongly dependent on the type of boundary conditions.
Strucutral mechanics is no different, and structural topol-
ogy optimization under different boundary conditions can
be equally interesting.
To find the stiffest design of structure, minimum struc-
tural compliance has been a very popular formulation in
structural topology optimization. The structure being opti-
mized in this formulation is subjected to given external
forces and fixed structural support, i.e., the second type
boundary condition in continuum mechanics. Structural
compliance is defined as twice of the work done by the
external forces when the structure deforms. Low compli-
ance is equivalent to small displacement and high stiffness,
i.e. a minimum compliance design implies a stiffest design.
Structural compliance is also equal to twice of the strain
energy stored in the structure for a fixed load condition.
Compliance minimization is therefore equivalent to mini-
mization of the structural strain energy.
However, in some less studied yet realistic applications,
structural boundary has prescribed non-zero displacements
in addition to the given fixed supports, i.e., the first type
boundary condition. One naturally expects the stiffest struc-
ture to respond to the prescribed displacement with the
largest reaction force, or maximizing the external work done
562 F. Niu et al.
by the reaction forces in the direction of the prescribed non-
zero displacement. Reaction force here is defined as the
force acting at the structure by the support. In this case, the
strain energy in the stiffest structure is maximized.
There are also situations where a loaded structure has
fixed support and prescribed non-zero displacements, or
subjected to the mixed boundary condition. For example,
in the hierarchical approach of structural optimization the
whole structure is decomposed into several levels. Displace-
ments and forces at the interface of top level structure and
low level substructure are obtained from structural analy-
sis of the top level structure. They are used as boundary
conditions in the optimum topology design of the low level
substructures. At each boundary node of the low level sub-
structure, the boundary condition of the optimization model
may be a given external force or a prescribed nodal displace-
ment. This naturally leads to the optimization problems with
mixed boundary conditions.
This category of problems to a certain degree is related
to optimization of structural support locations for minimum
structural compliance: Mroz and Rozvany (1975), kesson
and Olhoff (1988), Hou and Chuang (1990), Bojczuk and
Mroz (1998), Won and Park (1998), Buhl (2002). In most
cases, at the support, either zero displacement is specified
or elasticity of the support is to be designed. In other
cases, an objective function of maximum rigidity is used for
structure subjected to the prescribed non-zero displacements
(Cho and Jung 2003; Huang and Xie 2008). For structures
subjected to mixed boundary conditions, the recent study
by Pedersen and Pedersen (2009) discusses three possible
objectives, i.e., minimization of structural compliance, max-
imization of structural strength, and a design of structural
uniformstrain energy density. More discussion can be found
in their recent work (Pedersen and Pedersen 2010).
Analogous optimization problems can be found in opti-
mal topological design of fluid flow channel (Olesen et al.
2006). If fluid in the channel is driven at a prescribed pres-
sure drop, an engineering design objective is to maximize
the flow rate which is equivalent to maximizing the total
dissipated power. Optimization of this objective results in
optimum design of flow channel topology with the least
hydraulic resistance when flow passes the channel. How-
ever, if the system is driven at a prescribed flow rate at the
inlets and outlets, then the objective can be described as
minimizing the pressure drop between inlet and outlet which
is equivalent to minimizing the dissipated power. When the
boundary is partially subjected to a prescribed pressure drop
and partially constrained by flow rate, an adequate formu-
lation for flow channel topology optimization needs special
consideration.
In this paper, a general structural topology optimiza-
tion formulation is proposed to maximize structural stiffness
under the mixed boundary conditions. The proposed objec-
tive function consists of the work done by the given external
forces minus the work done by the reaction forces on
the prescribed non-zero displacements. Minimum structural
compliance design is a degenerated form of this general for-
mulation if the structural boundary conditions contain only
given external forces and fixed supports, i.e., zero displace-
ment only on part of boundaries. It is also equivalent to
the formulation of maximum structural strain energy design
if the structure is only applied by the prescribed non-zero
displacements and fixed supports.
The paper is organized as follows. First, we provide
details on the background of the problem statement, which
leads up to Section 2 on the mathematical formulation of
structural topology optimization for maximizing structural
stiffness under the mixed boundary conditions. Then, the
adjoint method is used to derive the sensitivity of objective
function with respect to the element density in Section 3.
This is mainly to show an interesting property that the sen-
sitivity of the objective is proportional to the negative of
average strain energy density, which is identical to that of
minimum structural compliance with the boundary condi-
tions of external forces and fixed supports. In Section 4, two
examples of six-bar truss and MBB beam are used to verify
the general formulation. Using SIMP (Solid Isotropic Mate-
rial with Penalization) model, we can obtain the analytical
solution of the optimization problem of six-bar truss and the
optimum material distribution of MBB beam. In Section 5,
we discuss more on the distribution of specific strain energy
density and Mises stress in the stiffest design. Finally, in
Section 6, we provide a summary and further thought on
this new formulations performance.
2 Formulation of topology optimization for maximizing
structural stiffness
SIMP is the most popular structural topology optimiza-
tion approach. In SIMP approach, the structural topology
is described by material distribution in the design domain,
which is divided into finite element meshes of N elements.
The artificial density
e
(e = 1, 2, , N) of each element
is the design variable. In the optimum design, the eth ele-
ment is a solid element if its artificial density
e
= 1, and
the eth element is a void element if its artificial density

e
= 0. To reduce the computation burden, the discrete
design variable
e
{0, 1} is relaxed as a continuous
variable, i.e., 0
e
1.
Based on the discussions in Section 1, for the
stiffest structure design with SIMP topology optimization
A general formulation of structural topology optimization for maximizing structural stiffness 563
approach, the formulation of minimum structural compli-
ance design (with external force) and maximum structural
strain energy design (with prescribed non-zero displace-
ment) is described in Sections 2.1 and 2.2 respectively, and
which lead to our general formulation III.
2.1 Formulation I: minimizing structural compliance
For a structure subjected to the prescribed external forces
and the given fixed supports, the formulation for the stiffest
structure design is that of a minimum compliance design
find : = {
1
,
2
, ,
N
},
min : C = f
T
p
u
p
,
s.t. :
N

e=1

e
v
e
V

,
: 0 <
min

e
1,
(1)
where displacement vector u is the function of , and can
be obtained by solving
K()u = f. (2)
Here the design domain is meshed by N finite elements,
represents the vector of design variables, which are ele-
ment artificial material densities. f is the external force
vector. K is structural global stiffness matrix and is obtained
by assemblage of element stiffness matrix over the design
domain. Based on SIMP model, the stiffness matrix of
each element depends on its artificial density
e
through its
Youngs modulus E
e
=
n
e
E
0
, where n is the penalization
power and E
0
is the Youngs modulus of solid material. V

is the given total material volume and v


e
is the volume of
element e. The lower bound of artificial material density,

min
, is introduced to avoid numerical singularity in struc-
tural analysis. The subscript p refers to the structural nodal
degree of freedom where the external force applies. Thus,
f
p
is the vector of given external forces and u
p
is the vec-
tor of nodal displacements where f
p
applies. Since the work
done by the external forces is equal to the structural strain
energy stored in the structure when the structure deforms,
the objective in (1) can also be written as
min : C = u
T
Ku, (3)
which shows that the optimum design of Eq. 1 is a minimum
structural strain energy design.
2.2 Formulation II: maximizing structural stiffness
For a structure subjected to prescribed non-zero displace-
ments and given fixed supports, the formulation for the
stiffest structure design is
find : = {
1
,
2
, ,
N
},
max : C = f
T
w
u
w
,
s.t. :
N

e=1

e
v
e
V

,
: 0 <
min

e
1,
(4)
where displacement vector u is the function of , and can
be obtained by solving Eq. 2, the subscript w refers to the
nodal degree of freedom where the prescribed non-zero dis-
placement is imposed. u
w
is the vector of prescribed forced
displacements, f
w
is the vector of reaction forces, which
are the forces acting at the structure by the supports, in the
direction of prescribed displacements. Again, notice that
the work done by the external forces is equal to the struc-
tural strain energy stored in the structure when the structure
deforms, the objective in Eq. 4 can also be written as
max : C = u
T
Ku, (5)
which shows that the optimum design of Eq. 4 is a maxi-
mum structural strain energy design.
2.3 Formulation III: a general formulation
for stiffest structure
Although in formulations I and II one seeks the opposite
extreme in terms of structural strain energy, interestingly
they all lead to the stiffest structure topology. Now let us
consider optimal topology design for the stiffest structure
for which part of structural boundary is subjected to pre-
scribed external forces f
p
and other part has prescribed
displacements u
w
. The material property and total volume
of structural material are specified as constraints. Based on
the formulations I and II, we propose the mathematical for-
mulation for the stiffest structure under the mixed boundary
conditions as
find : = {
1
,
2
, ,
N
},
min : C = f
T
p
u
p
f
T
w
u
w
,
s.t. :
N

e=1

e
v
e
V

,
: 0 <
min

e
1,
(6)
564 F. Niu et al.
where displacement vector u is the function of , and can
be obtained by solving Eq. 2.
It is interesting to note that the objective function in Eq. 6
is not equal to the twice of the strain energy stored in the
structure. Obviously it is also not equal to the work done by
the external forces, including the given external forces and
reaction forces at the prescribed displacements. Because of
that, the objective in the present formulation is neither max-
imum structural strain energy nor minimum structural strain
energy. The iteration history of structural strain energy does
not converge to maximum or minimum (see Figs. 5 and 7 in
Section 4.2).
The general theory of objective function selection can
be facinating and deserve further clarification. However,
in the following, we would like to focus on the property
of the sensitivity and its degenerated formulations to jus-
tify the general formulation (Eq. 6), and demonstrate its
effectiveness through specific examples.
3 Sensitivity analysis
To find the optimum design of Eq. 6 by the gradient based
mathematical programming methods, we shall derive the
sensitivity of the objective function first. Such sensitivity
analysis also serves as clarification of the physical meaning
of a general formualation and provides additional insight
into the nature of the problem. As the different boundary
conditions apply on different parts of structural bound-
ary, the governing equation for displacement at each nodal
degree of freedom
Ku = f, (7)
can be written as the following partitioned form
_
_
K
i i
K
i w
K
i p
K
wi
K
ww
K
wp
K
pi
K
pw
K
pp
_
_
_
_
u
i
u
w
u
p
_
_
=
_
_
f
i
f
w
f
p
_
_
, (8)
where the subscript i refers to the free structural nodal
degree of freedom on which no prescribed displacements
or external forces are imposed. The subscript w refers to the
structural nodal degree of freedom with prescribed displace-
ments including non-zero one and given structural fixed
support. The subscript p refers to the structural nodal degree
of freedom with external forces. f
i
is a zero vector, u
w
and f
p
are known prescribed quantities, u
i
, u
p
and f
w
are
unknown quantities. u
i
and u
p
can be obtained by solving
the following equations
_
K
i i
K
i p
K
pi
K
pp
__
u
i
u
p
_
=
_
f
i
K
i w
u
w
f
p
K
pw
u
w
_
, (9)
and f
w
is obtained as
f
w
= K
wi
u
i
+K
ww
u
w
+K
wp
u
p
. (10)
Let us consider the sensitivity analysis of the objective func-
tion. We introduce a Lagrange multiplies vector
T
=
(
T
i
,
T
w
,
T
p
) and apply the adjoint method
C

e
=f
T
p
u
p

f
T
w

e
u
w
+
T
(Ku f)

e
=f
T
p
u
p

f
T
w

e
u
w
+
T
i
_
K
i i

e
u
i
+K
i i
u
i

e
+
K
i w

e
u
w
+K
i w
u
w

e
+
K
i p

e
u
p
+K
i p
u
p

f
i

e
_
+
T
w
_
K
wi

e
u
i
+K
wi
u
i

e
+
K
ww

e
u
w
+K
ww
u
w

e
+
K
wp

e
u
p
+K
wp
u
p

f
w

e
_
+
T
p
_
K
pi

e
u
i
+K
pi
u
i

e
+
K
pw

e
u
w
+K
pw
u
w

e
+
K
pp

e
u
p
+K
pp
u
p

f
p

e
_
,
(11)
where u
w
/
e
= 0 because u
w
is given. Equation 11 can
be rewritten as
C

e
=
f
T
w

e
(u
w

w
)+
u
T
i

e
_
K
i i

i
+K
i w

w
+K
i p

p
_
+
u
T
p

e
_
K
pi

i
+K
pw

w
+K
pp

p
+f
p
_
+
T
i
_
K
i i

e
u
i
+
K
i w

e
u
w
+
K
i p

e
u
p

f
i

e
_
+
T
w
_
K
wi

e
u
i
+
K
ww

e
u
w
+
K
wp

e
u
p
_
+
T
p
_
K
pi

e
u
i
+
K
pw

e
u
w
+
K
pp

e
u
p

f
p

e
_
.
(12)
By setting

w
+u
w
= 0,
K
i i

i
+K
i w

w
+K
i p

p
= 0,
K
pi

i
+K
pw

w
+K
pp

p
= f
p
,
(13)
then, we have
C

e
=
T
K

e
u
T
i
f
i

T
p
f
p

e
. (14)
A general formulation of structural topology optimization for maximizing structural stiffness 565
Comparing Eqs. 9 and 13, considering f
i
= 0, we have
= u. (15)
And the sensitivity of the objective function is
C

e
= u
T
K

e
u
T
p
f
p

e
. (16)
If the external force f
p
does not depend on the structural
topology, the sensitivity can be further simplified as
C

e
= u
T
K

e
u. (17)
Furthermore, if we consider that the stiffness matrix of each
element depends only on its artificial density
e
and assume
its Youngs modulus E
e
=
n
e
E
0
, we have
k
e
=
n
e
k
0
e
, (18)
where the subscript e refers to the eth element, k
0
e
is the
element stiffness matrix for
e
= 1. From Eqs. 17 and 18,
we have
C

e
= u
T
e
k
e

e
u
e
=
n

e
u
T
e
k
e
u
e
, (19)
where u
e
is the nodal displacement vector of the eth
element.
Considering the dilemma discussed earlier regarding
opposite strain energy extremes in different formulations,
this result is really surprising. With this adjoint method and
appropriate Langragian multiplier, Eq. 19 states that the sen-
sitivity of the objective function to topological change is
proportional to the negative of average strain energy den-
sity. This is identical to the sensitivity of compliance for
the structure topology design in formulation I, i.e., the
minimization of compliance subjected to given external
forces.
Since the general formulation degenerates to the for-
mulation of minimum structural compliance design (with
external force) and maximum structural strain energy design
(with prescribed non-zero displacement), the sensitivity
(Eq. 19) is valid for formulation II as well. In other words,
maximizing the objective function in Eq. 5 is equivalent to
minimizing the negative of structural strain energy. Such
negative sensitivity in Eq. 19 physically means that addi-
tional material in any element makes the structure stiffer. In
this case, the objective function is monotonous with respect
to the change of design variable.
With the sensitivity function, the optimization problem 6
can be solved by the method of gradient-based mathemati-
cal programming. For the second numerical example, MMA
(The Method of Moving Asymptotes) is utilized as the opti-
mizer (Svanberg 1987). To avoid the mesh-dependency,
checkerboard and other numerical instabilities during the
iterative process of optimization, the sensitivity filter (Sig-
mund and Petersson 1998) is used.
4 Numerical examples
To verify the general formulation for the topology optimiza-
tion of maximizing structural stiffness, this section studied
two well-known examples, i.e., the six-bar truss (Stolpe and
Svanberg 2001) and the MBB beam.
4.1 Example 1: optimum topology design of six-bar truss
The first example is a six-bar truss structure (see Fig. 1).
The angle between two neighboring bars is = 30

. The
Youngs modulus of truss material is E = 1.0. The six bars,
which lengths are all equal to 1, are grouped in two groups.
The cross sectional areas a
1
and a
2
of bars in the groups 1
and 2 are the design variables. The total available volume of
structural material is 3.0. The govering equation which has
a diagonal global stifness matrix is
1
2
_
a
1
+5a
2
0
0 5a
1
+a
2
__
u
x
u
y
_
=
_
f
x
f
y
_
. (20)
When we introduce the SIMP approach and set the penalty
factor n = 2, Eq. 20 can be rewritten as
1
2
_
a
2
1
+5a
2
2
0
0 5a
2
1
+a
2
2
__
u
x
u
y
_
=
_
f
x
f
y
_
. (21)
Now let us check the two formulations I and II for the design
of stiffest structure.
First, let us verify the formulation I. A horizontal con-
centrated force f
x
= 1 is applied (see Fig. 1), and then
the force vector is f
p
= (1, 0)
T
. The result of nodal dis-
placement vector which can be obtained from Eq. 21 is
Fig. 1 Six-bar truss
566 F. Niu et al.
Fig. 2 The objective functions
for six-bar truss topology
optimization, Line 1 is the
objective function of the
optimization problem 22, Line 2
is the objective function of the
optimization problem 23
u
p
= (2/(a
2
1
+5a
2
2
), 0)
T
. The formulation I of maximizing
structural stiffness can be rewritten as
find : a
1
, a
2
,
min : C = f
T
p
u
p
=
2
a
2
1
+5a
2
2
,
s.t. : a
1
+a
2
= 1,
: 0 a
1
1,
: 0 a
2
1.
(22)
which is solved by the method of Lagrange multiplier. The
result of the optimum design is a
1
= 0, a
2
= 1 and
C = 2/5, which shows that only the bars in the group 2
are retained for the stiffest design. The nodal displacement
vector of the optimal structure, i.e., u
p
= (2/5, 0)
T
, will be
used as the prescribed non-zero displacement loading in the
study of the following paragraph.
Second, let us verify the formulation II for the optimum
topology of the stiffest structure under Dirichlet boundary
conditions. The prescribed non-zero horizontal displace-
ment u
x
= 2/5 instead of the concentrated horizontal force
f
x
= 1 is applied. The formulation III of maximizing
structural stiffness can be rewritten as
find : a
1
, a
2
,
min : C = f
T
w
u
w
=
2
25
_
a
2
1
+5a
2
2
_
,
s.t. : a
1
+a
2
= 1,
: 0 a
1
1,
: 0 a
2
1.
(23)
we also use the Lagrange multiplier method to solve prob-
lem 23 and obtained the optimum design a
1
= 0, a
2
= 1
and C = 2/5. The optimal design of Eq. 23 is exactly the
same as that of Eq. 22 which shows the consistence of the
two formulations I and II under the given external force and
the corresponding prescribed non-zero displacement respec-
tively. The feasible domain and the objective functions of
the feasible solution of these two optimization problems are
shown in Fig. 2. Notice that these two objective functions
are both not monotonous function along the feasible line
a
1
+a
2
= 1 because of the material volume constraint.
4.2 Example 2: optimum topology design of MBB beam
A MBB beam having a rectangular area is considered
(see Fig. 3). A vertical force f = 1 is applied at the mid-
point A of the lower side of the design domain. The bottom
right point B is fixed in the vertical direction and either free
or subjected to prescribed displacement D in the horizontal
direction. The given material volume is 40% of the design
domain. The Youngs modulus and Poissons ratio of solid
material are E = 1.0 and = 0.3 respectively.
To verify our optimization formulations I, II and III for
the stiffest structural design, we follow the same logic of the
six-bar truss example in Section 4.1. In Section 4.2.1, the
formulations I and III were compared. In Section 4.2.2, the
Fig. 3 MBB beam
A general formulation of structural topology optimization for maximizing structural stiffness 567
(a) (b)
Fig. 4 The optimum material distributions, (a) Optimum design with the formulation I, the bottom right point B is free, (b) Optimum design with
the formulation III, the prescribed displacement of B is 15.16479175. The value of penalization power in SIMP model is n = 1
formulations II and III were compared. For the same phys-
ical problem, two different formulations were expected to
obtain the same stiffest structure. However, with the penal-
ization in SIMP model, even the originally convex problem
(e.g. minimization of compliance for given material vol-
ume) becomes non-convex, and therefore, it is theoretically
possible to converge to a local optimum (Sigmund and
Petersson 1998; Rozvany 2009). The discrepancy between
the optimum designs of different formulations is then due to
numerical reasons. To avoid such complications, the value
of the penalization power n = 1 is used to get a global
optimum below.
Many components of the numerical approach used here
are relatively standard methods in the structural optimiza-
tion community. As such, those details about implementa-
tion are omitted here. To reach a clearly black/white (1/0)
structural topology and avoid the local optimum, we used a
continuation method (Sigmund and Petersson 1998). At the
beginning of optimization process, we probably got a global
optimum design for the optimization problem with n = 1 in
SIMP model. Then n was increased with 0.05 every 20 iter-
ations until n = 3 which could decrease grey regions in the
design domain. With this continuation method, it is usually
expected that the optimal design does not move away too far
from the global optimum.
In the following numerical examples, a four-node plane
stress element is adopted and a 60 20 mesh discretiza-
tion is applied for the design domain. The sensitivity filter
technique is used in these examples. The filter radius is a
constant equalling to twice of the size of element during
optimization iteration.
4.2.1 Comparison of formulations I and III
Let us consider the following two optimization problems.
Problem (1): The minimum compliance design of MBB
beam with the given concentrated force f = 1. The
bottom right point B is free in the horizontal direction.
The optimization formulation I for the structure under
the given external force was solved.
Problem (2): The stiffest design of MBB beam with
the given concentrated force f = 1 and prescribed
non-zero displacement D at point B in the horizontal
direction. The value of D is equal to the nodal hori-
zontal displacement at point B of the optimum design
obtained in problem (1). The optimization formulation
III for the structure under mixed boundary conditions
was solved.
The optimum material distributions of problem (1) is
shown in Fig. 4a, where the horizontal displacement D at
point B is 15.16479175 (To ensure accuracy, 10 significant
digits are taken). Then, the stiffest structural design of
problem (2) was obtained with the prescribed displacement
D = 15.16479175, which is shown in Fig. 4b. Obviously,
the same optimal results were achieved although these two
optimization problems were solved with different bound-
ary conditions and different objective functions. Figure 5
Fig. 5 Iterative histories of objective values and related qualities of
problems (1) and (2), Line 1 is the objective values of problem (1),
Line 2 is the objective values of problem (2), Line 3 is the strain energy
stored in MBB beam for the problem (2), Line 4 is the horizontal dis-
placement at point B for the problem (1), Line 5 is the reaction force
at point B for the problem (2)
568 F. Niu et al.
Table 1 Material distributions
and objective values of
problems (1) and (2) at iterative
steps 1, 5 and 9
Prob. (1) Minimum compliance Prob. (2) Maximum stiffness
Iter. Material distribution Obj. value Material distribution Obj. value
1 46.5156 42.0884
5 24.4741 27.1114
9 22.9819 23.4056
shows the iterative histories of their objective values and
related qualities. At the beginning of optimization process,
the objective values are different for the two optimization
problems (see line 1 and line 2 in Fig. 5). As the iteration
goes the difference between the objective values decreases
gradually and diminishes eventually. The iterative histories
of the horizontal displacement for the problem(1) (see line 4
in Fig. 5) and the reaction force (see line 5 in Fig. 5) at
the support B for the problem (2) are also shown in Fig. 5.
As expected, the strain energy (see line 3 in Fig. 5) stored
in MBB beam for problem (2) is not monotonous with
iterations. Table 1 shows the topology evolutions and the
corresponding objective values of problems (1) and (2) at
the iterative steps 1, 5 and 9 from which we can see their
different convergence processes.
4.2.2 Comparison of formulations I and III
Let us compare these two optimization problems.
Problem (3): The maximum stiffness design of MBB
beam with the boundary conditions of given concen-
trated force f = 1 at point A and prescribed non-zero
horizontal displacement D = 20 at point B. The opti-
mization formulation III for the structure under mixed
boundary conditions was solved.
Problem (4): The maximum stiffness design of MBB
beam with the boundary conditions of prescribed dis-
placement u
p
at point A and prescribed non-zero dis-
placement D = 20 at point B. The value of the
prescribed displacement u
p
at point A is equal to the
corresponding displacement of optimum design of the
problem (3). The optimization formulation II for the
structure under the Dirichlet boundary conditions was
solved.
The optimum designs of problems (3) and (4) were
obtained by solving the formulations III and II, respectively.
Figure 6a shows the optimum material distributions of prob-
lem (3) where the value of the displacement u
p
at point A
is 25.99080873 (see line 4 in Fig. 7). Then, the stiffest
structure of problem (4) was obtained using the prescribed
non-zero displacement u
p
= 25.99080873 at point A.
(see Fig. 6b). Similar to the example in Section 4.2.1, the
same optimal results of these two problems with different
boundary conditions and different objective functions were
also achieved which were also demonstrated by the iterative
histories of optimization (see line 1 and line 2 in Fig. 7).
Along with the process of optimization iterations, the value
of the vertical reaction force (see line 5 in Fig. 7) at point
A in problem (4) converged to the external force applied
in problem (3). Thus, the maximum stiffness design of
(a) (b)
Fig. 6 The optimum material distributions, (a) Using the formulation III under the mixed boundary conditions, (b) Using the formulation II under
the Dirichlet boundary conditions. The value of penalization power in SIMP model is n = 1
A general formulation of structural topology optimization for maximizing structural stiffness 569
Fig. 7 Iterative histories of objective values and related qualities of
problems (3) and (4), Line 1 is the objective values of problem (3),
Line 2 is the objective values of problem (4), Line 3 is the strain energy
stored in MBBbeamfor the problem(3), Line 4 is the vertical displace-
ment at point A for the problem (3), Line 5 is the reaction force at point
A for the problem (4)
problem (4) and the maximum stiffness design of prob-
lem (3) are identical. It is also shown that the strain energy
stored in MBB beam (see line 3 in Fig. 7) for problem (3)
is not monotonous with iterations. However, the topology
evolutions and the objective values during the optimization
iteration (Table 2) show that the problems (3) and (4) are
two different optimization problems and as such follows
different pathes to the same final solution.
4.2.3 The stif fest designs with varying
prescribed displacement
To discuss the relative influence of the given external force
and the prescribed non-zero displacement, we let the dis-
placement value of D change from30 to 50 gradually and
study the corresponding stiffest structures.
The stiffest designs of MBB beam with varying pre-
scribed displacements are shown in Fig. 8a and b is the
partial enlargement of Fig. 8a. It can be seen fromthe results
that the optimum topology is changed drastically, and when
the prescribed displacement at point B approaches to zero
form either directions, the optimum topology is the same,
i.e., the minimum compliance design with the boundary
condition of fixed point B.
Figure 8 shows that the greater than zero the optimal
objective value is, the more obviously the optimum topol-
ogy manifests itself mainly as the load transfer path of
prescribed external force, which means that the iterative
process of optimization is mainly under the control of pre-
scribed external force f ; otherwise, the less than zero the
objective value is, the more obviously the optimum topol-
ogy manifests itself mainly as the load transfer path of
reaction force in the prescribed displacement, which means
that the iterative process of optimization is mainly under the
control of prescribed non-zero displacement D.
When the value of D is changed in the vicinity of 1520,
the objective function achieves the maximum value. Then,
the optimum topology is the same as the result of minimum
compliance design with the boundary condition of no pre-
scribed horizontal displacement at point B. That is why the
greater stiffness is usually obtained in some industrial appli-
cations through increasing constraints of statically determi-
nate structures to make it become statically indeterminate
structures.
5 Distribution of specific strain energy and Mises stress
in the stiffest design
Here we re-examine the search for the stiffest design using
optimality condition approach and focus on the distribu-
tion of specific strain energy density and Mises stress. With
the sensitivity information available, the rational optimal-
ity condition, i.e., Kuhn-Tucker condition can be easily
derived. Considering the problem of maximum structural
stiffness design under the constraint of material resource,
the material resource usually reaches the upper limit of con-
straint in order to obtain a stiffer design. For simplification,
Table 2 Material distributions
and objective values of
problems (3) and (4) at iterative
steps 1, 5 and 9
Prob. (3) Maximum stiffness Prob. (4) Maximum stiffness
Iter. Material distribution Obj. value Material distribution Obj. value
1 34.4340 19.4100
5 24.9408 28.2115
9 22.1616 30.0278
570 F. Niu et al.
Fig. 8 The optimum topology
results, (a) The stiffest designs
of MBB beam with changing
prescribed displacement D, (b)
The partial enlargement in the
vicinity of D = 0 in Fig. 8a
(a)
(b)
we restate the active volume constraint in formulation
(Eq. 6) as an equality constraint:
find : = {
1
,
2
, ,
N
},
min : C = f
T
p
u
p
f
T
w
u
w
,
s.t. :
N

e=1

e
v
e
= V

,
:
min

e
+
2
1e
= 0, e = 1, 2, , N,
:
e
1 +
2
2e
= 0, e = 1, 2, , N,
(24)
where
1e
and
2e
are slack variables, and displacement
vector u is the function of , which can be obtained by solv-
ing Eq. 2. Let us introduce Lagrange multiplies , ,
1e
and
2e
(
1e
0,
2e
0), and construct an augmented
function
L =C +
_
N

e=1

e
v
e
V
_
+
T
(Ku f)
+
N

e=1

1e
_

min

e
+
2
1e
_
+
N

e=1

2e
_

e
1+
2
2e
_
.
(25)
From the stationarity of L, we obtain the optimality condi-
tion for stiffest structure as

u
T
e
k
e
u
e

e
v
e
+

n
=
_
_
_
= 0, f or
e
,
min
<
e
< 1
0, f or
e
,
e
=
min
0, f or
e
,
e
= 1
e = 1, 2, , N,
(26)
where k
e
is the eth element stiffness matrix (see Eq. 18).
A general formulation of structural topology optimization for maximizing structural stiffness 571
The first equality of the optimality condition 26 states that
the specific strain energy density, i.e., specific strain energy
per unit mass, is constant for all elements of intermediate
density
min
<
e
< 1. In other words, the specific strain
energy density is constant for all elements of intermediate
density. It is larger than or equal to such constant in regions
with a density
e
= 1 and less than or equal to this constant
in regions with a density
e
=
min
.
This is a nice property and our numerical experiment
above also shows the specific strain energy density to be
uniform for the optimum density distribution obtained by
SIMP with the penalization factor n = 1. However it is not
a black/white design as intended by the original structural
topology optimization goal. With density filter technique
and n > 1, our final optimum design showed almost
no element of intermediate density. Some of the density
filter techniques choices are Heaviside function (Guest et al.
2004), modified density filter with a Heaviside function
(Sigmund 2007), volume preserving nonlinear density filter
based on Heaviside functions (Xu et al. 2010). However, for
solid element, i.e., element of density
e
= 1 the optimal-
ity condition does not guarantee a uniform specific strain
energy density.
Stress constraint is one of the most important constraints
in structural engineering applications. For metal structures,
Mises stress is often used to check the strength criterion of
the design. Fully stressed design is an intuitive criterion
for engineering structural optimizations. Structural topol-
ogy optimization optimizes the load path and is expected to
give a nearly uniform stress design. However, the specific
strain energy is known as,
U =
1
2

i j

i j
=
1
2E
_

2
1
+
2
2
+
2
3
2
1

2
2
1

3
2
2

3
_
,
(27)
which is equal to the sum of the volume-change strain
energy U
V
and shape-distortion strain energy U

as
U = U
V
+U

,
U
V
=
1 2
6E
(
1
+
2
+
3
)
2
=

2
m
2K
=
3
2

m
,
U

=
1 +
6E
_
(
1

2
)
2
+(
1

3
)
2
+(
2

3
)
2
_
.
(28)
And the shape-distortion energy U

is proportional to the
square of Mises stress
U

=
1 +
3E

2
, (29)
which leads to the relation between specific strain energy
and Mises stress
U = U
V
+
1 +
3E

2
. (30)
So, in general, the specific strain energy is not equal to the
Mises stress. The Mises stress is not uniform even if the
specific strain energy is uniform.
However, as was pointed out in literatures, the stiffest
design and design with uniform specific strain energy is
very similar. The Mises stress in stiffest design is almost
uniform. Nevertheless, the difference is also notable and
critical to practical application.
6 Conclusions
In this paper, a general formulation for the stiffest design
of structures under the mixed boundary conditions has been
proposed, i.e., part of boundary is subject to external forces
and another part of boundary is subject to prescribed non-
zero displacements. The optimization algorithm is based on
the SIMP model combined with analytical sensitivity analy-
sis obtained by the adjoint method. It is very interesting that
the formulation of sensitivity is the same as that of mini-
mum structural compliance with the boundary conditions of
external forces and fixed supports.
The minimum structural compliance design and the max-
imum structural stiffness design can be recovered from such
formulation, depending on the type of boundary conditions.
We demonstrate its effectiveness on a number of stiffest
structural topology designs. However, it still needs improve-
ment to apply to general cases where external force depends
on the topology design.
For the problem of maximizing stiffness design of struc-
tures with mixed boundary conditions, the optimal topology
changes dramatically under the different scales between
external forces and prescribed non-zero displacements.
These examples also support the general observation that
that greater stiffness is usually obtained in some indus-
trial applications through increasing constraints of stati-
cally determinate structures to make it become statically
indeterminate structures.
Similarly, our approach can be extended to the optimiza-
tion design of fluid flow channel, when the system is driven
at a prescribed pressure drop and a prescribed flow rate at
the inlets simultaneously, the proposed objective function
can be composed of the dissipated power caused by pre-
scribed pressure drop and the minus of the dissipated power
caused by prescribed flow rate. The optimal flow channel
design can be obtained through maximizing this objective
function.
572 F. Niu et al.
Acknowledgements This work is supported by the National Natu-
ral Science Foundation of China (90816025, 10421202). The supports
are gratefully acknowledged. We thank Prof. Pauli Pedersen and Prof.
Niels L. Pedersen for the stimulating and invaluable discussions, we
thank Doc. Jiangs comments and suggestions for improvement of an
earlier version of this manuscript.
References
kesson B, Olhoff N (1988) Minimum stiffness of optimally located
supports for maximum value of beam eigenfrequencies. J Sound
Vib 120(3):457463
Bendse MP, Sigmund O (2003) Topology optimization: theory, meth-
ods, and applications. Springer, Berlin
Bojczuk D, Mroz Z (1998) On optimal design of supports in beam and
frame structures. Struct Multidisc Optim 16(1):4757
Buhl T (2002) Simultaneous topology optimization of structure and
supports. Struct Multidisc Optim 23(5):336346
Cho S, Jung HS (2003) Design sensitivity analysis and topology opti-
mization of displacement-loaded non-linear structures. Comput
Methods Appl Mech Eng 192(2224):25392553
Guest J, Prvost J, Belytschko T (2004) Achieving minimum length
scale in topology optimization using nodal design variables and
projection functions. Int J Numer Methods Eng 61(2):238254
Hou JW, Chuang CH (1990) Design sensitivity analysis and optimiza-
tion of vibrating beams with variable support locations. In: 16th
automation conference, ASME transaction, DE-23-2. Chicago, pp
281290
Huang X, Xie YM (2008) Topology optimization of nonlinear
structures under displacement loading. Eng Struct 30(7):2057
2068
Mroz Z, Rozvany G (1975) Optimal design of structures with variable
support conditions. J Optim Theory Appl 15(1):85101
Olesen LH, Okkels F, Bruus H (2006) A high-level programming-
language implementation of topology optimization applied to
steady-state navier-stokes flow. Int J Numer Methods Eng
65(7):9751001
Pedersen P, Pedersen NL (2009) A discussion on the application of
optimality criteria for compliance with forced support displace-
ment. In: 8th world congress on structural and multidisciplinary
optimization, ISSMO. Lisbon, Portugal
Pedersen P, Pedersen NL (2010) Design objectives with forced support
displacements. Struct Multidisc Optim (online)
Rozvany G (2009) A critical review of established methods of struc-
tural topology optimization. Struct Multidisc Optim 37(3):217
237
Sigmund O (2007) Morphology-based black and white filters for
topology optimization. Struct Multidisc Optim 33(45):401424
Sigmund O, Petersson J (1998) Numerical instabilities in topology
optimization: A survey on procedures dealing with checker-
boards, mesh-dependencies and local minima. Struct Optim
16(1):6875
Stolpe M, Svanberg K (2001) On the trajectories of penalization meth-
ods for topology optimization. Struct Multidisc Optim 21(2):128
139
Svanberg K (1987) The method of moving asymptotesa new method
for structural optimization. Int J Numer Methods Eng 24(2):359
373
Won KM, Park YS (1998) Optimal support positions for a struc-
ture to maximize its fundamental natural frequency. J Sound Vib
213(5):801812
Xu S, Cai Y, Cheng G (2010) Volume preserving nonlinear den-
sity filter based on heaviside functions. Struct Multidisc Optim
41(4):495505

You might also like