You are on page 1of 12

Computers and Structures 112–113 (2012) 433–444

Contents lists available at SciVerse ScienceDirect

Computers and Structures


journal homepage: www.elsevier.com/locate/compstruc

Optimal topology design of steel–concrete composite structures


under stiffness and strength constraints
Yangjun Luo a,b,⇑, Michael Yu Wang b,⇑, Mingdong Zhou b, Zichen Deng a
a
School of Mechanics, Civil Engineering & Architecture, Northwestern Polytechnical University, Xi’an 710072, China
b
Department of Mechanical and Automation Engineering, The Chinese University of Hong Kong, Shatin, NT, Hong Kong

a r t i c l e i n f o a b s t r a c t

Article history: This study presents a three-phase topology optimization model and an effective solution procedure to
Received 6 July 2012 generate optimal material distributions for complex steel–concrete composite structures. The objective
Accepted 14 September 2012 is to minimize the total material cost (or mass) while satisfying the specified structural stiffness require-
Available online 12 October 2012
ments and concrete strength constraints. Based on the Drucker–Prager criterion for concrete yield behav-
iour, the extended power-law interpolation for material properties and a cosine-type relaxation scheme
Keywords: for Drucker–Prager stress constraints are adopted. An enhanced aggregation method is employed to effi-
Topology optimization
ciently treat the large number of stress constraints, and the optimal topology is obtained through a stan-
Steel–concrete composite structures
Drucker–Prager criterion
dard gradient-based search. Several examples are provided to demonstrate the capability of the proposed
Enhanced aggregation method optimization method in automatically finding the reasonable composite layout of steel and concrete.
Ó 2012 Elsevier Ltd. All rights reserved.

1. Introduction produced form duo-symmetrical welded I-sections, composite


trusses formed from rolled channel sections, and composite trusses
Steel–concrete composite structures, such as the I-girder con- made from cold formed hollow sections. In addition, some numer-
nected with concrete slab [1] and the steel tube filled with ical techniques, such as the genetic algorithm, the harmony search
concrete [2], have been widely used in bridges and multi-storey algorithm and the reliability-based optimization method have also
buildings. By utilizing the compressive strength of concrete in con- been incorporated into the design of steel–concrete composite
junction with high tensile strength of steel, such composite may structures by Senouci and Al-Ansari [9], Kaveh and Shakouri [10]
provide better integrated performance and more economic advan- and Luo et al. [11], respectively.
tage than pure steel structures or conventional reinforced concrete Nonetheless, previous studies for the optimization of steel–
structures. concrete composite structures mainly focused on optimal sizing
Over the past several decades, there has been a growing interest design problems. As widely recognized, the sizing optimization
in the optimization design of steel–concrete composite structures may lead to sub-optimal results as it highly depends on the pre-
for a range of applications in civil engineering constructions [3]. scribed structural topology. Hence, to find out a reasonable topol-
For example, Long et al. [4], Negrão and Simões [5] presented a ogy configuration before the sizing optimization of a composite
nonlinear programming and an analytical sensitivity analysis for structure is very important.
the optimization design of cable-stayed bridges. In order to reduce Topology optimization of continua [12], which aims at deter-
the total cost of concrete, steel beams and shear studs, Adeli and mining the optimal material distribution features such as the num-
Kim [6] proposed a mixed integer-discrete optimization for com- ber of holes and their location within a given 2D or 3D design
posite floors by using neural dynamics model. With a predefined domain, has been extensively studied by both the academic and
steel price and fixed economical parameters, Kravanja and Šilih the industrial community. In particular, much effort has been de-
[7] performed an optimization based comparison between com- voted to the multi-material topology optimization problem. The
posite welded I-beams and composite hollow-section trusses. earliest work by Thomsen [13] utilized the homogenization tech-
Furthermore, Klanšek and Kravanja [8] developed the cost optimi- nique to design two isotropic material structures. As an extension
zation of three different composite systems: composite beams of the SIMP (Solid Isotropic Microstructure with Penalization)
model, a three-phase power-law interpolation scheme has been
⇑ Corresponding authors. Address: School of Mechanics, Civil Engineering & introduced by Sigmund [14]. By using Gaussian distribution func-
Architecture, Northwestern Polytechnical University, Xi’an 710072, China. Tel.: tions as weights on each phase, Yin and Ananthasuresh [15] pro-
+852 39438487; fax: +852 26036002 (Y. Lou). posed a ‘‘peak function approach’’ to perform the topology design
E-mail addresses: yangjunluo@nwpu.edu.cn (Y. Luo), yuwang@mae.cuhk.edu.hk of multi-material compliant mechanisms. Other typical ways to
(M.Y. Wang).

0045-7949/$ - see front matter Ó 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.compstruc.2012.09.007
434 Y. Luo et al. / Computers and Structures 112–113 (2012) 433–444

address the multi-material topology optimization problem in- by using the adjoint variable method is complemented in Appendix
clude: the ‘‘color’’ level-set method by Wang and Wang [16], the A. Finally, three examples are given in Section 5 to show the valid-
generalized Cahn–Hilliard model by Zhou and Wang [17] and the ity and effectiveness of the present method.
pseudo-sensitivity based discrete-variable approach by Ramani
[18]. During the development of these methods, multi-material 2. Optimization problem formulation
topology optimization has been successfully applied in multidisci-
plinary problems for the design of multi-physics actuators [19], The design of steel–concrete composite structure is not only a
multiphase composites with extremal bulk modulus [20], energy- stiffness design, the prevention of concrete crack and crush failure
absorbing structures [21] and laminated multi-material compos- should also be included in the design process. If both the structural
ites considering buckling [22]. Recently, Hvejsel and Lund [23] stiffness and concrete strength requirements are met, the steel
presented two new interpolation schemes for unified topology material will hardly suffer a catastrophic failure due to its high
and multi-material optimization, which allow for an arbitrary yield limit, good plasticity and good toughness. With these
number of pre-defined materials in generally application. Ramani considerations, the topology design problem for steel–concrete
[24] proposed a heuristic approach to handle strength constraints composite structures is stated as to find the optimal material dis-
in multi-material problems. tribution that minimizes total material cost (or mass) with both
Although some progress has been made, the optimization de- the stiffness constraints of the structure and the strength
sign of steel–concrete composite structures in the framework of constraints of concrete members satisfied.
multi-phase topology synthesis, as well as associated numerical Usually, the concrete and the steel members are connected by
techniques that enable real-scale applications, has not been well metal shear connectors such as headed studs and hoops. For sim-
addressed in literature. In fact, when determining the united distri- plification, a perfect connection is assumed and the material cost
bution of steel–concrete phases with metallic and non-metallic of shear connectors is not considered in the topology design. A to-
mixed material properties in a design domain, there are two main tal material cost function is defined as
challenges to address.
The first is the rational treatment of different yield behaviours C t ¼ cV;s  V s þ cV;c  V c ð1Þ
for two candidate materials. Steel is a homogenous material with
where cV,s and cV,c represent the price or the mass per unit volume of
constant and high strength in all directions, while concrete exhibits
steel and concrete, respectively; Vs and Vc are the volume of steel
a remarkable tension–compression asymmetry property. The com-
and concrete, respectively.
pression stress limit of concrete is almost an order of magnitude
The whole design domain Xdes of the structure is composed of
larger than its tension limit. Therefore, the Drucker–Prager
three different material phases (steel, concrete and void). Denoting
criterion accounting for the hydrostatic pressure effect on failure
the steel region and the concrete region by the subsets Xs and Xc,
is appropriate for describing concrete yielding behaviour, other
respectively, it yields (Xs [ Xc)  Xdes. Two discrete characteristic
than the most commonly used von Mises stress criterion. As the
functions, vs and vc, can be used to defined the material distribu-
power penalized stiffness model is used, the relevant stress relax-
tion at any point z 2 Xdes as
ation formulations based on Drucker–Prager criterion must be
 
defined for multi-phase intermediate density materials. 1 if z 2 Xs 1 if z 2 Xc
The second is the large-scale nature of elemental yield stress vs ðzÞ ¼ ; vc ðzÞ ¼ ð2Þ
0 if z 2 Xdes n Xs 0 if z 2 Xdes n Xc
constraints, meaning that one should devise an efficient constraint
reduction strategy for reducing the number of constraints in an Note that the topology definition in (2) formulates a discrete-
optimization procedure. Topology optimization of steel–concrete valued 0/1 integer optimization problem. In order to solve this
composite structures may confront with computational burden problem by continuous mathematical programming algorithms,
since a direct handling of large-scale constraints by existing continuous functions qs 2 [0, 1] and qc 2 [0, 1], which can be inter-
gradient-based algorithms is very time-consuming. The ‘‘active- preted as the relative densities of the steel phase and the concrete
set method’’ [25] and the ‘‘global constraint method’’ [26–29] are phase, are constructed to replace the discrete valued function vs
two major techniques to tackle this challenge. However, on the and vc, respectively. In addition, the values of qs and qc should
one hand, the ‘‘active-set method’’ is difficult to bring the required satisfy the correlation qs + qc 6 1, which may introduce extra con-
efficiency when a fine finite element discretization is used. On the straints in the optimization problem. To this end, the relative den-
other hand, the ‘‘global constraint method’’ may introduce high sity functions are written in terms of two independent design
nonlinearity to the global constraint function and cannot ade- variables x1 2 [0, 1] and x2 2 [0, 1] as
quately control the local stress state. Therefore, developing a more
q s ¼ x1 x2
effective way to overcome the challenge arising from large-scale ð3Þ
stress constraints is practically meaningful. Recently, Bruggi and qc ¼ x1 ð1  x2 Þ
Duysinx [30] proposed an alternative approach that couples a In the implementation of the finite element (FE) technique, the
global compliance enforcement with a selected set of local con- design domain is meshed into N elements. The discretized formu-
straints to improve the topology design of problems with localized lation of the optimization problem is expressed as
stress concentrations.
This paper investigates the multi-phase topology optimization X
N X
N
min C t ¼ cV;s  V s þ cV;c  V c ¼ cV;s  xe1 xe2 V e þ cV;c  xe1 ð1  xe2 ÞV e
methodology for generating optimal material distributions of x1 ;x2
e¼1 e¼1
steel–concrete composite structures. It first presents the mathe-
s:t: Ku ¼ f
matical model for a material cost (or mass) minimization problem T
considering the stiffness requirement and concrete yield strength g u ðx1 ; x2 Þ ¼ f u  L 6 0
constraints in Section 2. In the following section, the Drucker– g r;e ðx1 ; x2 Þ 6 0 ðe ¼ 1; 2; . .. ;NÞ
Prager criterion for concrete yield behaviour is introduced and a xmin 6 xe1 6 1; 0 6 xe2 6 1 ðe ¼ 1;2; .. .; NÞ
new relaxation formulation for Drucker–Prager yield stress is gi- ð4Þ
ven. Then an enhanced aggregation algorithm is suggested to effi-
ciently treat the large-scale constrained problem in Section 4. In where the superscript e denotes the eth element; vectors
the proposed algorithm, the required design sensitivity analysis x1 ¼ ½x11 ; x21 ; . . . ; xN1 T and x2 ¼ ½x12 ; x22 ; . . . ; xN2 T are design variables;
Y. Luo et al. / Computers and Structures 112–113 (2012) 433–444 435

is exceeded. Hence, it is appropriate to define the concrete material


failure constraint gr,e(x1,x2) in (4) in terms of the yield stress
function. Note that the hydrostatic stress has a strong influence
on the shearing capacity of concrete material, as well as on the
corresponding yield surface. Thus, the Drucker–Prager criterion is
employed for characterizing the concrete yielding, and a corre-
sponding constraint relaxation for intermediate density elements
is also discussed in this section.

3.1. Drucker–Prager criterion

The Drucker–Prager criterion was suggested by Drucker and


Prager [32] in 1952 and it takes the form as
pffiffiffiffiffiffiffi
gðI1 ; J 2 Þ ¼ aI1 þ 3J 2  b ¼ 0 ð6Þ

where a > 0 and b > 0 are material constants. Moreover, a is dimen-


sionless whereas b has the dimension of force per unit area. I1 is the
first invariant of the stress tensor and J2 is the second invariant of
the stress deviator tensor.
The invariants I1 and J2 can be expressed in different forms in
terms of the principal values (r1,r2,r3) or the components of the
stress vector r as

I1 ðrÞ ¼ r1 þ r2 þ r3 ¼ rx þ ry þ rz ¼ wr
1h i
Fig. 1. Elemental Young’s modulus E vs xe1 and xe2 . J 2 ðrÞ ¼ ðr1  r2 Þ2 þ ðr2  r3 Þ2 þ ðr3  r1 Þ2
6
1h i
Ve is the element volume; K,u and f are the global stiffness matrix, ¼ ðrx  ry Þ2 þ ðry  rz Þ2 þ ðrz  rx Þ2 þ s2xy þ s2yz þ s2zx
6
the displacement vector and the load vector, respectively; Ku = f
1
defines the equilibrium state of the FE model; gu(x1,x2) is the struc- ¼ rT Vr ð7Þ
3
tural compliance constraint and L⁄ is the maximal allowable value
of the mean compliance. Though it is possible that gu can be defined where w ¼ f 1 1 1 0 0 0 g. Let A be the 3  3 identify matrix
as a more general function (e.g. the nodal displacements), this study and B the 3  3 matrix of ones, the constant matrix V is given by
mainly focuses on the mean compliance constraint. gr,e(x1,x2) is the
( )
concrete material failure constraint of eth element and will be ex- 1
ð3A BÞ 0
2
pounded in Section 3. The small positive number xmin = 103 is V¼ ð8Þ
0 3A
introduced for avoiding numerical difficulties caused by zero mate- 66
rial stiffness in FE analysis. The whole optimization problem (4) can
also be referred to as the MWCS (minimum weight formulation In the Drucker–Prager criterion, the influence of the hydrostatic
with compliance and stress constraints) by Bruggi and Duysinx [30]. pressure is accounted for by introducing the linear term aI1. The
The Young’s modulus of each element with three material phase yield surface (6) in the principal stress space is a right-circular
is related to the design variables by using an extended power-law cone, which can be described in the meridian plane and the
interpolation scheme [14] as deviatoric plane. The meridian plane and the deviatoric plane
are defined as a plane containing and a plane perpendicular to
Eðxe1 ; xe2 Þ ¼ ðxe1 Þp ððxe2 Þp Es þ ð1  ðxe2 Þp ÞEc Þ; e ¼ 1; 2; . . . ; N ð5Þ the hydrostatic axis r1 = r2 = r3, respectively. The intersection
where Es and Ec are the Young’s modulus of fully steel material and curves between the yield surface and these two planes are illus-
fully concrete material, respectively; and p is the penalization factor trated in Fig. 2. It is revealed from the figure that the meridian
for driving intermediate density values to 0 or 1 during the optimi- makes a certain slope with the hydrostatic axis and the trace in
zation process. In this study, the penalization factor p = 3 is set. the deviatoric plane is a circle.
Fig. 1 shows the monotonous mapping between the elemental In practice, concrete material characteristics, e.g. the uniaxial
Young’s modulus and the design variables based on the interpola- strength limits and the biaxial strength limits, are usually mea-
tion (5). It is well known that an interpolation model can be sured by means of standard specimen tests in laboratory. It is thus
achieved by a practical composite only if the model satisfies the worthy to identify the material constants a and b from these avail-
so-called ‘‘Hashin–Shtrikman bounds’’, which defines the lower able data. For a biaxial plane stress state, i.e. r3 = 0, the yield sur-
and upper limits of the material bulk and shear moduli [31]. face (6) reduces to
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Although the interpolation in (5) may violate the Hashin–
Shtrikman bounds for multiphysics problems, it performs very well
aðr1 þ r2 Þ þ r21 þ r22  r1 r2  b ¼ 0 ð9Þ
for the considered stiffness and strength constrained design of As shown in Fig. 3, Eq. (9) represents an off-center ellipsoid in
steel–concrete composite structures in this paper. the coordinate plane of r3 = 0. Six particular stress cases (denoted
by point 1, 2, . . ., 6), which are related to the uniaxial compressive
= =
3. Concrete local stress constraints strength fc , the uniaxial tensile strength ft , the biaxial compressive
strength fbc and the biaxial tensile strength fbt= , are also marked in
=

As a typical brittle material, concrete suffers failure with an the figure. Substituting these particular stress values into Eq. (9),
immediate loss of its load-carrying capacity when the yield surface one has
436 Y. Luo et al. / Computers and Structures 112–113 (2012) 433–444

Fig. 2. Drucker–Prager criterion. (a) The meridian plane; (b) The deviatoric plane.

material; and ee is the element average strain. Usually, the strain


at the center of the element is chosen as the average strain, thus
one has

rec ¼ Dc Bue ð13Þ

where B is the strain–displacement matrix and ue is the elemental


displacement vector.
On the other extreme, qec ¼ 0 indicates the fully steel or void
material element. In this case, the invariants I1 and J2 for concrete
stresses should be vanished and thus

g r;e ðx1 ; x2 Þjqec ¼0 ¼ 1 ð14Þ

For an intermediate value of qec , the strength constraints require


special consideration for avoiding the well-known ‘‘stress singular-
ity phenomenon’’ that arises from the discontinuous nature of the
Fig. 3. Drucker–Prager yield surface under a biaxial stress state.
elemental average stress at zero density. This numerical complica-
b b tion has been found widely in stress-constrained topology optimi-
fc= ¼ ; ft= ¼ ð10Þ zation problems, not only for truss structures [34] but also for
1a 1þa
continua [35,36]. In order to circumvent the stress singularity phe-
or nomenon, a frequent treatment is to relax the strength constraints
b b involved in the intermediate density materials, say
fbc= ¼ ; fbt= ¼ ð11Þ 
1  2a 1 þ 2a qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
g r;e ðx1 ; x2 Þ ¼ rðqec Þ  aI1 ðrec Þ þ 3J 2 ðrec Þ b1 ð15Þ
The material constants a and b may be obtained by using these
relations.
To meet the geometric requirements of the concrete failure sur- where 0 6 rðqec Þ 6 1 is the relaxation function and it should satisfy
face more accurately, several complex criteria including hydro- that rjqec ¼1 ¼ 1 and rjqec ¼0 ¼ 0.
static pressure effects, such as the Bresler–Pister three-parameter Obviously, the relaxation function has important consequences
model, the Ottosen four-parameter model and the Willam–Warnke for the yield behaviour of the intermediate density material. It
five-parameter model, have been developed [33]. However, the means that for a fully concrete phase element ðqec ¼ 1Þ, the yield
Drucker–Prager model is preferable to the topology optimization surface defined by the relaxed formulation (15) is identical to the
problems due to its simplicity and its wide application in practical original strength constraint in (12). When qec < 1, the relaxed for-
engineering. mulation allows to expand the yield surface by 1/r times, as shown
in Fig. 4. By relaxing the yield surface for intermediate density
3.2. Relaxation of Drucker–Prager yield stress constraints material, the discontinuity of the original constraints at zero

The material failure constraints gr,e(x1,x2) are imposed on the


concrete phase of elements, indicating that it is necessary to define
an appropriate stress state for an intermediate material phase.
On one extreme, if the relative density of the concrete phase
qec ¼ 1, for which the design variable xe1 is equal to 1 while xe2 is
equal to 0, the element e is fully occupied by the concrete material.
Based on the Drucker–Prager yield criterion (6), it is easy to define
the elemental failure constraint as
 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
g r;e ðx1 ; x2 Þjqec ¼1 ¼ aI1 ðrec Þ þ 3J 2 ðrec Þ b1 ð12Þ

where rec ¼ Dc ee is the elemental average stress of the concrete


phase; Dc is the constitutive elasticity tensor of the fully concrete Fig. 4. Relaxation of the Drucker–Prager yield surface in the meridian plane.
Y. Luo et al. / Computers and Structures 112–113 (2012) 433–444 437

density is avoided and this makes it applicable to find the true op- solution if only strength constraints are considered. However, the
tima by commonly-used gradient based algorithms. stiffness constraint included in our problem is very helpful to over-
A widely used relaxation function for the stress singularity phe- come this difficulty and numerical examples show that no conver-
nomenon is the ‘‘e-relaxation’’ function proposed by Cheng and gence problems would occur.
Guo [37], which is given by By taking the proposed relaxation function into account,
Drucker–Prager yield stress constraints in (15) are then reformu-
1 e
rðqec Þ ¼ ; he ¼ 1 þ e ð16Þ lated by
he qec  qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ð1  cosðp  qec ÞÞ
where e > 0 is a prescribed small positive real number and it is g r;e ðx1 ; x2 Þ ¼  aI1 ðrec Þ þ 3J 2 ðrec Þ
2b
usually set to be 0.1–0.5. This function creates a continuous
 1 ðe ¼ 1; 2; . . . ; NÞ ð19Þ
point-to-set mapping between the predefined parameter e and
the feasible design region.
Another useful relaxation function is the ‘‘qp relaxation’’ func- 4. Enhanced aggregation strategy for the optimization problem
tion introduced by Bruggi [38]. By choosing a suitable stress
penalty exponent differing from the interpolate stiffness parame- The number of strength constraints gr,e(x1,x2) 6 0 is equal to
ter, this function is defined as the number of elements N, which may be huge for complex
pffiffiffiffiffi structures when a fine discretization is used. In general, it is very
rðqec Þ ¼ qec ð17Þ
time-consuming to directly solve the optimization problem (4)
Numerical investigation by Bruggi [38] showed that these two by gradient-based algorithms. To address the computational
approaches are similar in principle but different in convergence challenge, an enhanced aggregation method is presented for
features. Although the relaxation functions (16) and (17) have been effectively solving this medium-or large-scale problem.
successfully applied in many studies for single material topology Given an iteration step k and the current values of all elemental
optimization problems with von Mises and Drucker–Prager stress ðkÞ
strength constraints g r;e ðe ¼ 1; 2; . . . ; NÞ, we first check whether a
constraints, they must be used with care for the multi-phase mate- strength constraint belongs to a so-called ‘‘potentially active set’’.
rial problem in (4). It is noted that the Drucker–Prager yield surface
The potentially active set AðkÞ is to record the indices of the con-
in the stress space is a cone. Moreover, the cone’s apex is located
straints at which equality will be hold on the optimal point. After
very close to the origin of principle stress coordinate. As a conse- ðkÞ
quence, the concrete stress state of intermediate density elements sorting the current values of g r;e in descending order, i.e.
ðkÞ ðkÞ ðkÞ ðkÞ
is rather easily to overrun the Drucker–Prager yield surface, espe- g r;1 P ... P g r;j P g r;jþ1 P ... P g r;N ; AðkÞ is defined as
cially in the tensile stress field. In the multi-phase topology optimi- n o
zation, this particularity of the Drucker–Prager yield surface can AðkÞ ¼ 1; 2;    ; NðkÞ
act ;
ðkÞ
Nact ¼ maxfjg ð20Þ
ðkÞ
g r;j Pf
cause considerable difficulty in obtaining an initial feasible
solution. where j = 1, 2, . . . , N represents the sorted order number of the
In order to obtain reasonable results in the optimization pro- strength constraints, and the real negative number f is the pre-
cess, performing a quite large relaxation for the low-density mate- scribed critical value. Herein, we set f = 0.3 in the considered
rial is necessary and thus a new relaxation function is proposed as topology optimization problem.
1  cosðp  qec Þ The strategy is to aggregate the strength constraints belonging
rðqec Þ ¼ ð18Þ to the potentially active set AðkÞ into several global constraints by
2
using a general formulation of the Kreisselmeier–Steihauser
As shown in Fig. 5, all of three relaxation functions have a sim- (K–S) function and thus greatly save the computational effort.
ilar approximation to the original constraint for the high-density The general K–S function is expressed by
material, but the proposed function exhibits more relaxation than
the other two functions for the low-density material. We remark
here that the proposed function may fail to obtain a clear 0–1

Fig. 5. Comparison of three different relaxation functions. Fig. 6. Feasible regions defined by the general K–S function under different value of h.
438 Y. Luo et al. / Computers and Structures 112–113 (2012) 433–444

8 9 8 9
1 <X = 1 <X = the original feasible region. The maximum violation value (denoted
ðkÞ ðkÞ ðkÞ
GL ðx1 ; x2 Þ ¼ ln expðg
g r;j Þ  ln ð1 þ hg r;j Þ by Dg) of the global constraint to the local constraints is equal to
g : ; g : ;
j2AðkÞ j2AðkÞ log (2)/g, that is why the conventional K–S function can not pro-
ð21Þ vide a good approximation to the original problem when g is small.
As h increased, the global feasible region shrinks and excludes par-
where g > 0 is the aggregation parameter, and 0 6 h < g is the tial region. When h tends to g, the feasible region tends to the point
ðkÞ
reduction parameter. Note that when h = 0, Eq. (21) degenerates D, at which both local constraints are active, i.e. g r;1 ¼ 0 and
ðkÞ
into the conventional K–S function as described in many studies g r;2 ¼ 0. This shows that the general K–S function is well suited
[28,39,40]. to handing the active constraints since the optimum will be
The characteristics of the general K–S aggregation function can determined exactly. The search process based on the general K–S
be observed from an illustrative example as shown in Fig. 6. There- function is most likely to find an accurate optimal solution under
ðkÞ ðkÞ
in, two local constraints g r;1 6 0 and g r;2 6 0 are considered and a small value of g.
the aggregation parameter g = 7 is set. In the figure, the shade During the optimization process, the potential active
domain denotes the original feasible region defined by two local constraint in the current iteration may become inactive in the
constraints, while the solid curves represent the feasible region subsequent iteration, and vice versa. Therefore, the reduction
ðkÞ
boundaries by the global constraint GL ðx1 ; x2 Þ 6 0 under different parameter h should be automatically adjustable to define an
value of h. If h = 0, the global feasible region completely surrounds appropriate global feasible region, and thus makes the

Fig. 7. Flowchart of the optimization process using the enhanced aggregation strategy.
Y. Luo et al. / Computers and Structures 112–113 (2012) 433–444 439

‘‘re-generation’’ and the ‘‘removal’’ of potential active constraints optimal active set is reached and the reduction parameter h ap-
possible. The value of h at each iteration is estimated using the proaches to the value of g. In this implementation, k = 0.25 is set
following procedure for a moderate step at each iteration.
ðkÞ
Xh i. X Inactive constraints g r;j < 0 ðj R AðkÞ Þ may affect the searching
h0 ¼ expðgg ðkÞ
r;j Þ  1 g ðkÞ
r;j path of the optimization process. In order to improve the numeri-
j2AðkÞ j2AðkÞ cal stability, these constraints are also treated globally by using the
8 9, 0 1
<Xh i = X ðkÞ conventional K–S function. That is
hnew ¼
ðkÞ ðkÞ
expðgg r;j Þ  1  max½expðgg r;j Þ þ 1 @ g r;j  maxg r;j A
ðkÞ
: ðkÞ j2AðkÞ ; ðkÞ j2AðkÞ 8 9 8 9
j2A j2A
1 <X = 1 <X =
h ¼ h0 þ kðhnew  h0 Þ ðkÞ ðkÞ
GKS ðx1 ; x2 Þ ¼ ln expðg
g r;j Þ  ln 1 ð23Þ
g : ; g : ðkÞ ;
ð22Þ jRAðkÞ jRA

ðkÞ
where 0 < k 6 1 is the step length. If maxj2AðkÞ g r;j < 0; hnew is greater In addition, the grouped aggregation approach [28], which di-
than h0, which indicate that the feasible region shall be expanded. vides all the strength constraints into several groups and integrates
ðkÞ
On the other hand, if maxj2AðkÞ g r;j > 0; hnew is less than h0, represent- a relatively small number (e.g. M = 50) of constraints contained in
ing the feasible region is shrank. At the final optimum, the true each group by an aggregation function (21) or (23), would be help-
ful to mitigate the numerical instability. Obviously, the reduction
of constraints proves to be very effective because only a total of
N/M global constraints are involved in computation.
The detailed optimization process using the enhanced aggrega-
tion strategy is shown in Fig. 7. With the proposed method, the
computational efficiency can be dramatically improved without
sacrificing the accuracy and introducing numerical instability.
The Matlab implementation of MMA algorithm, which is kindly
provided by Svanberg, is used as an optimizer for updating the de-
sign variables in each iteration.

5. Examples

In this section, three numerical examples for finding the opti-


Fig. 8. Design domain for a Michell type structure. mal steel and concrete composite topologies are presented. Herein,

Fig. 9. Optimal composite topologies with different concrete compressive Fig. 10. Contours of concrete strength constraint functions for different designs. (a)
strengths. (a) fc= ¼ 40MPa; (b) fc= ¼ 60MPa; (c) fc= ¼ 80MPa. fc= ¼ 40MPa; (b) fc= ¼ 60MPa; (c) fc= ¼ 80MPa.
440 Y. Luo et al. / Computers and Structures 112–113 (2012) 433–444

all the materials are assumed to be linearly elastic before yield. The concrete’’ member to share the compression. The union of steel
Young’s modulus and the Poisson’s ratio of steel and concrete are and concrete thus creates a cost-minimal composite structure to
selected from experimental values as: Es = 210,000 MPa, Ec = 36,6 satisfy the stiffness and strength requirements. This obtained opti-
00 MPa, ms = mc = 0.3. The relative (dimensionless) prices of steel mal layout undoubtedly agrees with designers’ intuition and
and concrete are cV,s = 1 and cV,c = 0.1, respectively. The uniaxial experience. As the compressive strength fc= of concrete increases,
tensile strength of concrete is taken as 1/8 of its compressive not only more concrete and less steel are used, but also the optimal
strength. In the proposed optimization process, four-node plane material distribution significantly changes (see Fig. 9(c)).
stress elements are used in the FE analysis. The initial values of The contours of the concrete strength constraint function gr,e
ð0Þ ð0Þ
the design variables are set to be x1 ¼ 1 and x2 ¼ 0:5. (x1,x2) (as expressed in (19)) for these three designs are plotted
in Fig. 10(a)–(c). The concrete crack or crush failure under the spec-
5.1. Design of a Michell type structure ified external load will not occur, because the maximum value of
the concrete strength constraints is nearly zero. Therefore, the
The first example is the design of a Michell type structure. As optimal material layouts obtained by the proposed method are fea-
shown in Fig. 8, the length of the rectangular design domain is sible in practice. Iteration histories of the total material cost and
1.6 m and the height is 0.8 m. A downward concentrated force the maximal strength constraint value (i.e. maxe = 1, 2, . . ., N{gr,e})
F = 200 kN is applied at the middle of the bottom edge. In view of are plotted in Fig. 11(a) and (b). A monotonic convergence of the
the symmetry, a mesh of 6400 (80  80) elements is used to objection function is achieved by the enhanced aggregation strat-
discretize the left-half design domain. The maximal allowable value egy after the first several iterations. Moreover, the maximal
of the mean compliance is specified as L⁄ = 300 N  m (or the vertical strength constraint value may exhibit serious violations before find
displacement constraint for the loading point is u⁄ = L⁄/F = 1.5 mm). an initial feasible solution, but it tends to remain stable at the pre-
Three types of concrete with different uniaxial compressive defined upper bound (zero) in the subsequent iterations. These
strengths, fc= ¼ 40; 60 and 80 MPa, are used. The design objective iteration histories show good stability and convergence of the pro-
is to minimize the total material cost. posed enhanced aggregation method.
The optimal layouts by using concrete with fc= ¼ 40; 60 and To demonstrate the validity of the present three-phase model as
80 MPa are presented in Fig. 9(a)–(c), respectively. In the figures, well as the efficiency of the proposed numerical techniques, opti-
the black areas denote solid steel material and the green areas indi- mal topology solutions based on two common used models, (1)
cate solid concrete material, while the white areas represent void single steel material model, (2) three-phase material model with-
phase. When fc= ¼ 40 MPa, the optimal layout in Fig. 9(a) uses pure out strength constraints, are also given in Fig. 12(a) and (b).
steel bars to withstand the tension, and an arched ‘‘steel encased In the latter two models, the same mean compliance constraint

Fig. 11. Iteration histories using the proposed method. (a) The total material cost. (b) The maximal strength constraint value.

Fig. 12. Optimal topologies using other models. (a) Single steel material solution. (b) Three-phase material solution without strength constraints.
Y. Luo et al. / Computers and Structures 112–113 (2012) 433–444 441

Table 1
Comparison of topology optimization results for the Michell type structure.

Cases Total material Steel volume Achieved mean Maximal strength Iteration Total computing
cost (103) ratio (%) compliance (Nm) constraint value number time (h)
=
Proposed method with fc ¼ 40MPa 2.896 20.39 300.22 0.03 103 0.88
=
Proposed method with fc ¼ 60MPa 2.789 17.85 300.21 0.04 119 1.02
=
Proposed method with fc ¼ 80MPa 2.640 14.66 300.31 0.01 117 1.05
Single steel material design 3.211 25.09 300.54 – 112 0.26
Three-phase material design without 2.205 9.13 300.20 – 125 0.29
strength constraints

fTu 6 300 N  m is imposed. The design objective is still the total


material cost. As compared to the designs in Fig. 9, these two solu-
tions are distinctly different in the amount and the distribution of
steel material. The single-material solution uses the most steel
material since it ignores the contribution of concrete phase. In con-
trast, the three-phase material solution uses the least steel mate-
rial. However, it should suffer from a serious crack failure under
the applied load because a large amount of concrete material is dis-
tributed in the tensile zone. This highlights the necessity of
accounting for the weak tensile strength nature of concrete in
the layout design of steel–concrete composite structures.
The corresponding details of solutions, including the total mate-
rial cost, the steel volume ratio, the achieved mean compliance, the
maximal strength constraint value, the number of iteration, and
the total computing time, are summarized in Table 1. The total
material cost of the steel–concrete composite design with
fc= ¼ 40 MPa is 2.896  103, which is about 9.8% less than the
material cost of the single steel design (3.211  103). When the
high strength concrete with fc= ¼ 80 MPa is used, the cost reduction
ratio of composite design to single steel design further increases to Fig. 13. Design domain for a deep beam with a square hole.
1  2.640  103/(3.211  103) = 17.8%. The comparison of these
results proves the significant economic advantage of the steel–
concrete composite structure over the single steel design when Fig. 13. A downward concentrated force F = 200 kN is applied at
the same structural stiffness needs to be achieved. It is also the upper edge of the design domain. The whole design domain
revealed from the last column of Table 1 that: the total computing is meshed by 14,100 (150  94) square elements, and the square
time of the proposed designs (need treating 6400 strength hole is set as a non-design void region. The uniaxial compressive
constraints) is about one hour, which is only three times the strength of concrete is fc= ¼ 60 MPa. Here, the optimization objec-
computing time spent in three-phase topology optimization tive is the same to the first example, but the maximal allowable
without strength constraints (0.29 h). The efficiency of the value of the mean compliance is set as L⁄ = 400 N  m.
enhanced aggregation strategy for treating large-scale constraints The optimal material distribution obtained from the proposed
in the topology optimization is thus illustrated. method is depicted in Fig. 14(a). The total material cost and the
steel volume fraction ratio are 2.460  103 and 13.00%,
5.2. Design of a deep beam with a square hole respectively. For comparison, the single steel material topology
optimization with the same compliance constraint is also
The second example is the design of a simply supported deep implemented and the obtained optimal solution is shown in
beam with a fixed square hole, whose dimensions are shown in Fig. 14(b). The corresponding material cost and the steel volume

Fig. 14. Comparison of optimal topologies. (a) Design using the proposed method. (b) Design based on single steel material model.
442 Y. Luo et al. / Computers and Structures 112–113 (2012) 433–444

Fig. 15. Iteration histories using the proposed method. (a) The total material cost. (b) The maximal strength constraint value and the structural mean compliance.

ratio are 2.860  103 and 20.26%, respectively. One can easily cV,s = 7850 kg/m3 and cV,c = 2700 kg/m3, respectively. Due to the
recognize the remarkable cost saving in the proposed design. symmetric characteristic of this problem, one half of the design
This illustrates again that it is more economical to adopt a domain is discretized with 9600 (120  80) elements.
steel–concrete composite structure in the civil engineering than The optimal topologies obtained by the proposed method and
use a pure steel structure. It is more interesting to point out that the three-phase topology optimization without strength con-
the proposed design (Fig. 14(a)) not merely replaces the straints are presented in Fig. 17(a) and (b). Both designs present
expensive steel compressive struts in the single steel design a composite structure consisted of a steel arch and a number of
(Fig. 14(b)) by cheap concrete columns, but suggests a more concrete columns. The total mass and the steel volume ratio for
efficient steel–concrete composite structure with a novel multi- the proposed design in Fig. 17(a) are 149.66 kg and 12.07%,
phase material topology. respectively. When the concrete strength constraints are not con-
The iteration histories of the objective function, the structural sidered, the optimal topology in Fig. 17(b) presents a less-mass
mean compliance and the maximal strength constraint value are design, with the total mass of 145.36 kg and the steel volume ra-
plotted in Fig. 15. Again, a monotonic and stable convergence is tio of 12.72%. For this particular problem, only small changes in
achieved in the topology design by using the proposed method. the optimal layout are suggested by the proposed design and
more concrete columns are used to satisfy the strength
5.3. Design of a bridge structure constraints.
The concrete strength constraint function contours for the
In the third example, a bridge structure to be optimized within two topologies are shown in Fig. 18. It is seen from Fig. 18(a)
a design domain, with a geometrical dimension of 2.4 m  0.8 m that the maximal strength constraint value of the design
and a thickness of 0.01 m, is illustrated in Fig. 16. The design do- obtained by the proposed method is close to zero. However, as
main is fixed at both bottom corners and subjected to a uniform shown in Fig. 18(b), local crush behaviours will occur in the
pressure load of 200 kN/m on the top edge. Here, the 0.04 m-height concrete region in which the concrete stress distinctly violates
domain below the top surface is set as a non-design concrete re- its strength constraint. Therefore, it may be dangerous to employ
gion. The maximal allowable value of the structural mean compli- the latter method to design steel–concrete composite structures
ance is L⁄ = 1200 N  m. High strength concrete with the uniaxial
compressive strength fc= ¼ 60 MPa is used. In this example, the
objective function is defined as the total mass of steel and concrete
to be minimized. The densities of steel and concrete are taken as

Fig. 17. Comparison of optimal topologies. (a) Design using the proposed method.
Fig. 16. Design domain for a bridge structure. (b) Design using three-phase topology optimization without strength constraints.
Y. Luo et al. / Computers and Structures 112–113 (2012) 433–444 443

Appendix A. Sensitivity analysis

Design sensitivity analysis in the optimization procedure re-


quires the calculation of the derivatives of the global constraints
(21) and (23) and the mean compliance with respect to the design
variables. These informations are derived as follows.
By applying the definition of invariants in (7) and the stress–
displacement relationship in (13), concrete strength constraints
can be reformed by

 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ð1  cosðp  qjc ÞÞ
g r;j ðx1 ; x2 Þ ¼  awrjc þ ðrjc ÞT Vðrjc Þ  1
2b
 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ð1  cosðp  qjc ÞÞ
¼  awDc Buj þ ðDc Buj ÞT VDc Buj  1
2b
 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Fig. 18. Contours of concrete strength constraint functions for different designs. (a) ð1  cosðp  qjc ÞÞ
The proposed design. (b) Three-phase material design without strength constraints.
¼  asc uj þ ðuj ÞT Mc uj ðj ¼ 1; 2;. . . ; NÞ ðA1Þ
2b

where sc = wDcB and Mc ¼ BT DTc VDc B are used for simplification.


Sensitivities of the jth concrete strength constraints with re-
even for this entirely compressive case. On the other hand, the spect to the design variable xei ði ¼ 1; 2Þ of the eth element are
proposed method with the Drucker–Prager yield constraints
0 1
ensures that the minimal mass design is attained without
@g r;j ðx1 ; x2 Þ ð1  cosðp  q j j T
concrete failure. B ðu Þ Mc C @uj
c ÞÞ
¼  @asc þ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiA e
@xei 2b @xi
ðuj ÞT Mc uj
6. Conclusions and discussions  qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
p sinðp  qjc Þ @ qjc
þ  asc uj þ ðuj ÞT Mc uj
2b @xei
Incorporating stiffness and concrete strength requirements into
j
the topology optimization problem is of practical importance for @u j @ qc
¼ ðaj ÞT þ b ði ¼ 1; 2; j ¼ 1; 2; . . . ; N;
the layout design of steel–concrete composite structures. This @xei @xei
study develops a cost (or mass) minimization topology optimiza- e ¼ 1; 2; . . . ; NÞ
tion method based on the three-phase material interpolation
ðA2Þ
model and the Drucker–Prager criterion. For improving the compu-
j
tational efficiency, the large-scale stress constrained topology opti- where a is the jth sparse vector for expanding the element dimen-
mization problem is solved by the proposed enhanced aggregation sion to the global dimension and bj is the coefficient, namely
method. The accuracy of the solution is ensured by using the gen-    T
eral K–S function to aggregate the active constraints. ð1cosðpqjc ÞÞ Þ Mc j T
aj ¼ 0    2b
asc þ pðuffiffiffiffiffiffiffiffiffiffiffiffiffiffi
j T

j
0  0
The flowchart of the overall optimization process is presented. ðu Þ Mc u
 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ðA3Þ
Three design problems under plane stress conditions are provided j p sinðp  qjc Þ
to illustrate the validity and the effectiveness of the proposed opti- b ¼  asc uj þ ðuj ÞT Mc uj
2b
mization model and numerical techniques. The obtained solutions
are compared with the designs obtained by using conventional By applying the chain rule, the derivative of the global con-
topology optimization models. It is shown that the proposed straint (21) is then given by
steel–concrete composite designs are more economical than the P n o P
@g @g r;j
@GL ðx1 ; x2 Þ j2AðkÞ
expðgg r;j Þ @xre;j h j2AðkÞ @xei
pure steel counterparts when the same stiffness needs to be ¼ P i
 P
achieved. The comparison also reveals that it may be dangerous @xei j2AðkÞ
expðgg r;j Þ g j2AðkÞ
ð1 þ hg r;j Þ
to design the steel–concrete composites without considering the ( )
X expðgg r;j Þ h @g r;j
concrete strength constraints. ¼ P  P
When the first design variable vector is kept fixed at x1 = 1, the ðkÞ
expðgg r;j Þ g j2AðkÞ ð1 þ hg r;j Þ @xei
j2AðkÞ
j2A
proposed optimization formulation reduces to a two-material
ðA4Þ
(steel and concrete) model without void. In particular, it may be
j
applicable to the design of reinforced concrete structures with high Introducing a shorthand variable x for the brace-related items
crack control requirements and the related study has been and substituting Eq. (A3) into (A4), we have
reported in the prior work of Luo and Kang [41]. !
The reported work in this paper is believed to be novel in @GL ðx1 ; x2 Þ X j @g r;j X @u @ qj
e
¼ x e ¼ xj ðaj ÞT e þ bj ec
generating innovative topologies for steel–concrete composite @xi ðkÞ
@xi ðkÞ
@xi @xi
j2A j2A
structures. The method can be readily extended to three- !
dimensional problems, and this development will be described in X @u X @ qj
future papers.
¼ ðx a Þj j T
e
þ xj bj ec ðA5Þ
ðkÞ
@xi ðkÞ
@xi
j2A j2A

  !
Acknowledgements T @f @K @u X j
j j @ qc
¼ ðkL Þ  u þ xb e
@xei @xei @xei ðkÞ
@xi
j2A
The supports from the National Natural Science Foundation of
China (Grant 51008248) and the Hong Kong Scholars Program where kL denotes the adjoint vector and is the solution to the ad-
(XJ2011029) are gratefully acknowledged. joint system
444 Y. Luo et al. / Computers and Structures 112–113 (2012) 433–444

X
KkL ¼ ðxj aj Þ ðA6Þ [15] Yin L, Ananthasuresh GK. Topology optimization of compliant mechanisms
ðkÞ
with multiple materials using a peak function material interpolation scheme.
j2A
Struct Multidiscip Optim 2001;23:49–62.
[16] Wang MY, Wang XM. ‘‘Color’’ level sets: a multi-phase method for structural
Analogously, the derivative of the global constraint (23) can be topology optimization with multiple materials. Comput Methods Appl Mech
obtained from Eqs. (A4) and (A5) if we set h = 0. Eng 2004;193:469–96.
Finally, the sensitivities of the mean compliance constraint gu [17] Zhou S, Wang MY. Multimaterial structural topology optimization with a
generalized Cahn–Hilliard model of multiphase transition. Struct Multidiscip
with respect to design variables xe1 and xe2 are Optim 2007;33:89–111.
 T [18] Ramani A. A pseudo-sensitivity based discrete-variable approach to structural
@g u @f @K topology optimization with multiple materials. Struct Multidiscip Optim
¼ 2 u  uT e u ði ¼ 1; 2; e ¼ 1; 2; . . . ; NÞ ðA7Þ
@xei @xei @xi 2010;41:913–34.
[19] Sigmund O, Torquato S. Design of materials with extreme thermal expansion
It is noted that the term @f=@xei vanishes if the variation of the using a three-phase topology optimization method. J Mech Phys Solids
1997;45:1037–67.
material distribution has no effect on the applied loads. Moreover,
[20] Gibiansky LV, Sigmund O. Multiphase composites with extremal bulk
the term @K=@xei can be obtained by analytical calculations in the modulus. J Mech Phys Solids 2000;48:461–98.
elemental level, that is [21] Jung D, Gea HC. Design of an energy-absorbing structure using topology
optimization with a multimaterial model. Struct Multidiscip Optim
@Ke 2006;32:251–7.
¼ pðxe1 Þp1 ððxe2 Þp Kes þ ð1  ðxe2 Þp ÞKec Þ [22] Lund E. Buckling topology optimization of laminated multi-material
@xe1 composite shell structures. Compos Struct 2009;91:158–67.
ðA8Þ
@Ke [23] Hvejsel CF, Lund E. Material interpolation schemes for unified topology and
¼ pðxe1 Þp ðxe2 Þp1 ðKes  Kec Þ multi-material optimization. Struct Multidiscip Optim 2011;43:811–25.
@xe2 [24] Ramani A. Multi-material topology optimization with strength constraints.
Struct Multidiscip Optim 2011;43:597–615.
where Ke is the elemental stiffness matrix, Kes and Kec are the ele- [25] Bendsøe MP, Sigmund O. Topology optimization: theory, methods, and
mental stiffness matrix computed with the elastic constants of fully applications. Berlin Heidelberg: Springer; 2003.
solid steel and fully solid concrete, respectively. [26] Duysinx P, Sigmund O. New developments in handling stress constraints in
optimal material distributions. In: 7th Symposium on multidiciplinary
analysis and optimization, AIAA/USAF/NASA/ISSMO, AIAA-98-4906; 1998. p.
References 1501–9.
[27] Le C, Norato J, Bruns T, Ha C, Tortorelli D. Stress-based topology optimization
[1] Luo YJ, Li A, Kang Z. Parametric study of bonded steel–concrete composite for continua. Struct Multidiscip Optim 2011;41:605–20.
beams by using finite element analysis. Eng Struct 2012;34:40–51. [28] Luo YJ, Kang Z. Topology optimization of continuum structures with Drucker–
[2] Ellobody E, Young B, Lam D. Behaviour of normal and high strength concrete- Prager yield stress constraints. Comput Struct 2012;90–91:65–75.
filled compact steel tube circular stub columns. J Construct Steel Res [29] Parı́s J, Navarrina F, Colominas I, Casteleiro M. Block aggregation of stress
2006;62:706–15. constraints in topology optimization of structures. Adv Eng Soft
[3] Awad ZK, Aravinthan T, Zhuge Y, Gonzalez F. A review of optimization 2010;41:433–41.
techniques used in the design of fibre composite structures for civil [30] Bruggi M, Duysinx P. Topology optimization for minimum weight with
engineering applications. Mater Des 2012;33:534–44. compliance and stress constraints. Struct Multidiscip Optim 2012;46:369–84.
[4] Long W, Troitsky MS, Zielinski ZA. Optimum design of cable stayed bridges. [31] Bendsøe MP, Sigmund O. Material interpolation schemes in topology
Struct Eng Mech 1999;7:241–57. optimization. Arch Appl Mech 1999;69:635–54.
[5] Negrão JHO, Simões LMC. Optimization of cable-stayed bridges with three- [32] Drucker DC, Prager W. Soil mechanics and plastic analysis or limit design.
dimensional modelling. Comput Struct 1997;64:741–58. Quart Appl Math 1952;10:157–65.
[6] Adeli H, Kim H. Cost optimization of welded of composite floors using neural [33] Chen WF, Han DJ. Plasticity for structural engineers. Florida: J Ross Publishing;
dynamics model. Commun Num Methods Eng 2001;17:771–87. 2007.
[7] Kravanja S, Šilih S. Optimization based comparison between composite I [34] Sved G, Ginos Z. Structural optimization under multiple loading. Int J Mech Sci
beams and composite trusses. J Construct Steel Res 2003;59:609–25. 1968;10:803–5.
[8] Klanšek U, Kravanja S. Cost estimation, optimization and competitiveness of [35] Kirsch U. On singular topologies in optimum structural design. Struct
different composite floor systems. Part 2: Optimization based competitiveness Multidiscip Optim 1990;2:133–42.
between the composite I beams, channel-section and hollow-section trusses. J [36] Cheng GD, Jiang Z. Study on topology optimization with stress constraints. Eng
Construct Steel Res 2006;62:449–62. Optim 1992;20:129–48.
[9] Senouci AB, Al-Ansari MS. Cost optimization of composite beams using genetic [37] Cheng GD, Guo X. e-Relaxed approach in structural topology optimization.
algorithms. Adv Eng Software 2009;40:1112–8. Struct Multidiscip Optim 1997;13:258–66.
[10] Kaveh A, Shakouri A. Cost optimization of a composite floor system using an [38] Bruggi M. On an alternative approach to stress constraints relaxation in
improved harmony search algorithm. J Construct Steel Res 2010;66:664–9. topology optimization. Struct Multidiscip Optim 2008;36:125–41.
[11] Luo YJ, Li A, Kang Z. Reliability-based design optimization of adhesive bonded [39] Kreisselmeier G, Steinhauser R. Systematic control design by optimizing a
steel–concrete composite beams with probabilistic and non-probabilistic vector performance index. In: International federation of active controls
uncertainties. Eng Struct 2011;33:2110–9. symposium on computer-aided design of control systems. Zurich, Switzerland;
[12] Bendsøe MP, Kikuchi N. Generating optimal topologies in structural design 1979. p. 113–7.
using a homogenization method. Comput Methods Appl Mech Eng 1988;71: [40] Parı́s J, Navarrina F, Colominas I, Casteleiro M. Topology optimization of
197–224. continuum structures with local and global stress constraints. Struct
[13] Thomsen J. Topology optimization of structures composed of one or two Multidiscip Optim 2009;39:419–37.
materials. Struct Optim 1992;5:108–15. [41] Luo YJ, Kang Z. Layout design of reinforced concrete structures using two-
[14] Sigmund O. Design of multiphysics actuators using topology optimization. Part material topology optimization with Drucker–Prager yield constraints. Struct
II: Two-material structures. Comput Methods Appl Mech Eng 2001;190: Multidiscip Optim. http://dx.doi.org/10.1007/s00158-012-0809-1.
6605–27.

You might also like