Computers and Structures 89 (2011) 702–711

Contents lists available at ScienceDirect

Computers and Structures

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

Topology optimization for the seismic design of truss-like structures

Iman Hajirasouliha a,⇑, Kypros Pilakoutas b, Hassan Moghaddam c
a
Department of Civil Engineering, The University of Nottingham, Nottingham, UK
b
Department of Civil and Structural Engineering, The University of Sheffield, Sheffield, UK
c
Civil Engineering Department, Sharif University of Technology, Tehran, Iran

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

Article history: A practical optimization method is applied to design nonlinear truss-like structures subjected to seismic
Received 15 June 2010 excitation. To achieve minimum weight design, inefficient material is gradually shifted from strong parts
Accepted 3 February 2011 to weak parts of a structure until a state of uniform deformation prevails. By considering different truss
structures, effects of seismic excitation, target ductility and buckling of the compression members on
optimum topology are investigated. It is shown that the proposed method could lead to 60% less struc-
Keywords: tural weight compared to optimization methods based on elastic behavior and equivalent static loads,
Truss structure
and is efficient at controlling performance parameters under a design earthquake.
Topology optimization
Weight minimization
Ó 2011 Elsevier Ltd. All rights reserved.
Seismic excitation
Nonlinear behavior

1. Introduction structural optimization [7], homogenization-based optimization

[8], fully stressed topology design [9], rule-based optimization
Structural optimization techniques are effective tools that can [10] and soft-kill optimization [11]. However, by utilizing these
be used to obtain lightweight, low-cost and high performance methods, in general, it is not possible to find the optimal cross-
structures. In the absence of efficient practical topology design sectional properties of truss members. This may lead to suboptimal
tools, the selection of element topology in current design practice results as the size of the structural elements has a significant effect
is usually based on the engineering judgment and past experience. on the weight and structural performance of the final design.
Optimal design of truss structures has always been a challenge as Besides, the application of these methods is limited to elastic
they present a common and complex category of engineering structures.
structures. Various techniques based on classical optimization Genetic algorithms (GAs) have been widely used in topology
methods such as feasible direction method [1], primal–dual optimization of truss structures [12–15]. These methods represent
method [2], dynamic programming [3], random search method general adaptive search techniques that simulate natural inheri-
[4] and sequence of linear programs [5] have been developed to tance by adapting appropriate models based on genetics and natu-
find optimum design of elastic truss structures. These conventional ral selection. Computational effort that is associated with GAs
methods are usually gradient-based solution strategies that re- seems to be high even for optimizing large elastic truss structures
quire the satisfaction of some specific mathematical conditions. under static loads [16]. Therefore, it is not realistic to use these
Due to the high computational effort needed to calculate deriva- algorithms for optimum seismic design of truss structures in the
tives of the objective functions at each load step, these methods practice.
are not appropriate for optimum design of non-linear truss struc- By considering the nonlinear dynamic behavior of structures,
tures subjected to dynamic loads. Kirsch [6], studied the effect of most of the available optimization techniques become very com-
compatibility conditions on optimal topology of truss structures. plicated. However, it is well known that many of the structural sys-
His studies showed that using classical numerical search algo- tems experience nonlinear deformations during medium to strong
rithms for particular geometries or loading conditions may lead earthquakes. Therefore, the employment of conventional optimiza-
to an optimal topology that represents an unstable structure. tion techniques based on the elastic behavior of structures may not
Several methods have been proposed for topology optimization lead to the optimum use of materials in the nonlinear range of
of trusses by using continuum structures, including evolutionary behavior. This paper presents a new practical method for optimum
design of nonlinear truss-like structures subjected to dynamic
⇑ Corresponding author. Tel.: +44 115 9513922; fax: +44 115 9513898. excitations. This simplified method is based on the concept of uni-
E-mail address: [email protected] (I. Hajirasouliha). form distribution of deformation demands as it is introduced in

0045-7949/$ - see front matter Ó 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compstruc.2011.02.003
 I. Hajirasouliha et al. / Computers and Structures 89 (2011) 702–711 703

Section 2. Section 3 explains the developed optimization algorithm In the performance-based design methods, design criteria are
in detail. The proposed algorithm is applied on a simple truss struc- expressed in terms of performance objectives. Recent design
ture in Section 4, and the effects of target ductility demand, seismic guidelines, such as FEMA 356 [21], place limits on acceptable val-
exaction and buckling of compressive members on the optimum ues of response parameters, implying that exceeding these accept-
solution are investigated. A practical method for optimum seismic able values represents violation of performance objectives. The
design of truss structures subjected to a group of earthquakes is proposed method can optimize the design for different types of
also introduced in this section. Section 5 presents application of performance objectives such as deformation, acceleration and
the proposed method for optimum design of more complex truss velocity. In the nonlinear range of response, member ductility ratio
structures subjected to gravity and lateral seismic loads. Conclud- is a good criterion for assessing seismic performance of truss-like
ing remarks are given in Section 6. structures. Therefore, here the design variables are chosen to sat-
isfy design constraints as follows:
2. Proposed methodology Subject to : li 6 lt ði ¼ 1; 2; . . . ; nÞ ð2Þ

Seismic design of structures is currently based on force rather where li and lt are maximum ductility ratio of ith member and tar-
than displacement as the seismic design codes generally use lateral get ductility ratio, respectively.
inertia forces to account for seismic ground motion effects. The dis- In this study, the iterative optimization procedure developed by
tribution of these static forces (and therefore, stiffness and strength) Hajirasouliha and Moghaddam [18] for optimum design of
is based implicitly on the elastic vibration modes [17]. Optimization shear-building models is extended for topology optimization of
of non-linear truss-like structures for dynamic excitations is more non-linear truss structures. The proposed optimization algorithm
challenging and associated with high computational effort and is explained in detail in the following steps:
numerical difficulties due to the complexity of the problem. 1. To design a truss-structure, certain nodes are considered to be
In severe earthquakes, as structures exceed their elastic limits, important and must exist in any feasible design, while some
the use of inertia forces corresponding to elastic modes may not optional nodes are just added for load sharing. The important
lead to the optimum distribution of structural properties [18]. nodes are usually the ones that carry external loads or support
Therefore, during strong earthquakes the deformation demand in the truss, and are usually specified by the designer. The
some parts of the structures does not utilize the allowable level proposed algorithm assumes a ground structure, which is a
of seismic capacity. If the strength of these underused elements complete truss with all possible (or reasonable) member con-
(inefficient material) is decreased incrementally, it is expected to nections among all nodes in the structure. The initial cross-sec-
eventually obtain a status of uniform deformation demand. In such tional areas of the members are allowed to have arbitrary
a condition, the dissipation of seismic energy in each structural different values. In this study, an initial ground structure is
element is maximized and the material capacity is fully exploited. assumed with all elements having uniform cross-sectional area.
Previous studies showed that the seismic performance of such a It is shown later that the initial cross-section areas do not have
structure is near optimal, and it undergoes less damage in compar- any effect on the final optimum solution.
ison with code-based designed structures having similar structural
weight [18,19]. Therefore, in general, it can be concluded that a sta- 2. The structure is subjected to design loads (e.g. gravity loads and
tus of uniform deformation demand is a direct consequence of the seismic excitations), and maximum ductility demand for each
optimum use of material [19]. It should be mentioned that the structural element (li) is calculated, and compared with the
optimum structure with minimum structural weight may not nec- target value (lt). If the calculated ductility demands are close
essarily be the one with uniform ductility demand. Nevertheless, enough to the target ductility (e.g. 5% tolerance), the truss
the proposed method has capability to decrease the required struc- structure is considered to be practically optimum. Otherwise,
tural weight by exploiting all the material capacity. the optimization algorithm is continued.
The concept of the proposed optimization method (i.e. exploit- 3. Positions where the ductility demand is less than the target val-
ing all the material capacity) is close to the classic optimization ues are identified and the inefficient material is reduced until an
methods (such as fully-stress method [9,20]) which are sometimes optimum structure is obtained. To obtain good convergence in
used in engineering practice. However, the application of conven- numerical calculations, this alteration is applied incrementally
tional methods are usually limited to structures with elastic behav- [22]. To accomplish this, the following equation is used:
ior. The optimization method developed in this study is one of the
few methods that is capable of optimizing non-linear structures ½Ai mþ1 ¼ ½Ai m ½SRi a ð3Þ
subjected to dynamic excitations. While the basic philosophy of ei l
SRi ¼ ¼ i ð4Þ
the proposed procedure is traditional, the developed optimization et lt
algorithm is capable of converging to the optimum solution in a
where [Ai]m is the cross-sectional areas of the ith member at
few steps and without any fluctuation.
mth iteration. ei and et are maximum strain of the ith member
and target strain, respectively, and a is the convergence param-
3. Optimization algorithm eter ranging from 0 to 1. The proposed equations are valid for
both elastic and inelastic structures, since they are based on
Assuming that the cost of a member is proportional to its strain rather than stress. The convergence parameter a plays
material weight, the least-cost design can be interpreted as the an important role in the convergence of the problem as will
least-weight design of the structure. Therefore, the cost objective be discussed in the following section. In this study, a was set
function f to be minimized can be formulated as: to be 0.2.
X
n 4. The need of specific structural members is determined by com-
Minimize : f ðAÞ ¼ q i l i Ai ð1Þ paring the cross-sectional area of each member with a user-
i¼1
defined minimum value of critical cross-sectional area, a0. If
where the design variables are the cross-sectional areas of truss the element area is smaller than a0, that member is removed
members (denoted as A) and qi and li are material density and from the regenerated truss structure. Subsequently, the optimi-
length of ith member, respectively. zation procedure is repeated from step 2 until the maximum
704 I. Hajirasouliha et al. / Computers and Structures 89 (2011) 702–711

ductility demand of all truss elements reaches the target ductil-

ity (i.e. SRi = 1). At this stage a status of uniform ductility
demand is achieved; and therefore, based on the concept of uni-
form distribution of deformation demands a truss structure
with minimum required structural weight is obtained.
The user-defined parameter of a0 plays an important role in the
stability of the final solution. If a0 is small enough, the optimization
algorithm always leads to stable results, since the dynamic behav-
ior and stability of the structure is controlled at each iteration. On
the other hand, to find a solution with the minimum number of
truss members, a0 should be greater than a minimum value that
is a function of the cross-sectional area of the elements. Numerous
analyzes carried out on different truss-like structures showed that,
to obtain the best results, a0 should be between 5% and 10% of the
average cross-sectional area of all remaining elements [22]. In this
case, a0 is adaptive and always results in a stable design with the
minimum number of structural elements. In this study, a0 has been
taken equal to 10% of the average cross-sectional area of all
remaining elements.
The proposed method can be a very useful tool for optimum de-
sign problems encompassing both the selection of optimal topol-
ogy and the determination of optimal sizing of structural
members. While using this method can rapidly lead to an optimum
design, the quality of the solution depends on the location of the
nodes and the connectivity of bars of the initial ground structure.
However, to obtain a design that is close to the global optimum
solution there is no need to consider a dense initial ground struc-
Fig. 1. (a) Preliminary arrangement of members in ground structure. (b) Optimum
ture [22,23].
topology reported by Liang [9]. (c) Optimum topology using the proposed
The proposed optimization method applies for members with optimization method.
continuous variation of the cross-sectional area. However, in the fi-
nal design, the cross-sectional area of truss elements could be re-
placed by the first higher available section. It should be
mentioned that the minimum weight design is not always the 6000
least-cost design as other factors such as the cost of the joints
Total Structural Weight (Kg)

could be considerable. It was explained before that the proposed 5000

optimization method removes very lightly loaded elements, and
therefore, tends to minimize the number of elements. As the cost 4000
of joints is to some extend proportional to the number of elements
and the force they carry, their cost is indirectly taken into account 3000
in the optimization process.
2000

4. Elastic design example

1000

To investigate the effectiveness and validity of the proposed

0
optimization method, a problem for which the optimal solution
is known is considered in this section. The optimization objective 0 10 20 30 40
is to find the optimum topology for a truss bridge with 30 m span Steps
(Fig. 1a) that leads to a minimum structural weight design. The de- Fig. 2. Variation of total structural weight from initial structure to the optimum
sign criterion adopted limits the compressive and tensile stress of design.
all truss elements to 60% of the yield stress (i.e. SRi = 0.6) when the
structure is subjected to a 400 kN concentrated static load at the it was capable of reaching the known optimum solution. However,
mid-span. The initial ground structure is chosen by considering unlike most of the conventional optimization methods such as the
all possible member connections, as shown in Fig. 1a. Using the one used by Liang [9], the application of the proposed method is
proposed optimization algorithm, the optimum topology is deter- not limited only to elastic structures subjected to static loads.
mined by removing all inefficient elements. Fig. 2 shows the vari-
ation of total structural weight from initial structure to the
optimum design. The result indicates that, using the proposed 5. Non-linear design example
method, the optimum solution was obtained in less than 10 steps.
It should be mentioned that the optimum topology in this example The efficiency of the proposed optimization method for non-lin-
is not dependent on the magnitude of the applied load and maxi- ear structures is demonstrated through a conceptually simple
mum allowable stress, since the structure is within the elastic example shown in Fig. 3a. The objective is to design a truss struc-
range of behavior. Fig. 1 compares the results of this study with ture for sustaining three masses M1–M3 (which are assumed to be
the optimum topology reported by Liang [9] based on the fully 10, 20 and 10 tons, respectively) by using any number of members
stressed method using continuum structures. The results confirm connecting these masses to each other and to the supports. Seismi-
the validity of the optimization method proposed in this study as cally induced inertia forces are the only forces that act on points
 I. Hajirasouliha et al. / Computers and Structures 89 (2011) 702–711 705

a M2=20 ton b 8

3m M1=10 ton M3=10 ton 6 7

9m

3m 4m 4m 3m 1 2 3 4 5

Fig. 3. (a) Position of masses and supports. (b) Joint numbers and preliminary arrangement of members in ground structure.

M1–M3. This structure should not exceed a member ductility

demand of 4 when subjected to the horizontal component of the
a b
Northridge Earthquake of 1994 (CNP196). As a starting point, a
ground structure with members having identical cross-sectional
area is chosen by considering all possible member connections,
as shown in Fig. 3b. It is assumed that the yield strength of each
member is equal in tension and compression. Nonlinear dynamic
analyzes were performed using DRAIN-2DX program [24]. A stiff-
ness and mass proportional Rayleigh damping model with a con-
stant damping ratio of 0.02 was assigned to the first and second
modes of vibration. Buckling at the global level is considered by
taking into account the P–D effects. An 18-bar truss shown in Fig. 4. Optimum topology. (a) Similar behavior in tension and compression. (b)
Fig. 3b with uniform cross-sectional area equal to 10 cm2 was con- Considering buckling of compressive members, lt = 4.
sidered as the initial ground structure. By applying the proposed
optimization method, the optimum design was obtained. Table 1
5.1. Efficiency of the proposed method
shows the cross-sectional area of truss members for the optimum
solution. The results indicate that the proposed method was able to
The preliminary seismic design of most truss-like structures is
recognize and eliminate the redundant and inefficient members.
commonly based on equivalent static forces that are distributed
Out of the 18 members in the primary arrangement in Fig. 3b, only
proportional to the weight (or mass) of different parts of the struc-
8 members remained in the final step as shown in Fig. 4a. In the
ture (Fig. 5a). The optimum topology of the above example (Fig. 3a)
optimum design structure, the maximum ductility demands for
was calculated for the equivalent static forces shown in Fig. 5a, by
all of the remaining truss elements are very close to the target duc-
utilizing the fully stress design method [20]. Fig. 5b, shows that the
tility of 4. The effect of the convergence parameter, initial cross-
optimum topology in this case is different with the optimum topol-
sectional area, target ductility demand, seismic excitation and
ogy based on the non-linear dynamic behavior of the structure
buckling of the compressive members on the optimum design is
(Fig. 4a). Table 1 compares the cross-sectional area of truss
investigated in the following sections.

Table 1
Definition of truss members and optimum answer for target ductility demand of 4.

Member First joint Second joint Cross-sectional area (cm2)

Based on the non-linear Based on the equivalent Considering
dynamic behavior static loads buckling effects
1 1 6 10.7 7.2 21.9
2 2 6 0 0 0
3 3 6 0 0 18.1
4 4 6 10.5 2.9 0
5 5 6 0 7.3 0
6 7 6 0 0 0
7 8 6 5.0 0 6.9
8 1 7 0 7.3 0
9 2 7 10.5 2.9 0
10 3 7 0 0 18.1
11 4 7 0 0 0
12 5 7 10.7 7.2 21.9
13 8 7 5.0 0 6.9
14 1 8 6.3 19.4 0
15 2 8 0 0 0
16 3 8 0 0 0
17 4 8 0 0 0
18 5 8 6.3 19.4 0
706 I. Hajirasouliha et al. / Computers and Structures 89 (2011) 702–711

a 2F b cross-sectional area on the final results, the previous example

was solved using two different initial ground structures with uni-
form cross-sectional area equal to 10 and 1000 mm2. The varia-
F F tions of total strength from each initial ground structure toward
the final solution are compared in Fig. 7. It is shown in this figure
that the optimum design is not dependent on the initial cross-
sectional area, as two different ground structures converged to
the same total structural weight level. This conclusion has been
confirmed by analysis of different structures and ground motion
records [22].
The speed of convergence also appears to be dependent on the
Fig. 5. (a) Distribution of equivalent static seismic forces. (b) Optimum topology initial material distribution. The results indicate that using a more
based on the equivalent static loads.
educated estimate for the initial cross-sectional area (i.e. relatively
members for the optimal and conventionally design structures close to the final solution) results in a faster convergence. It is
with the target ductility demand of 4. The proposed optimization shown in Fig. 7 that a ground structure with uniform cross-
method resulted in a truss structure with 535 kg structural weight, sectional area equal to 1000 mm2 practically converged to the opti-
while using the optimum topology based on the equivalent static mum solution in less than 15 steps.
loads led to a structure with 748 kg structural weight (40% more),
to obtain the same maximum member ductility demand (i.e. lt 5.4. Effect of buckling of members under compression
= 4). This implies that optimizing a structure based on the equiva-
lent static loads and elastic behavior does not necessarily lead to a In practical applications, compressive members tend to buckle
minimum weight design solution when the structure is subjected before they reach the yield point. In addition, local buckling can
to seismic excitations. This conclusion is in agreement with the significantly change the hysteresis behavior of axially loaded steel
results obtained by further analysis on different models and seis- members, thus influencing optimum design. To address this issue,
mic ground motions [22]. It should be emphasized that the pro- the hysteretic model suggested by Jain et al. [25] is utilized in this
posed optimization method leads to a more efficient seismic study. In this model, the axial compressive strength (Pc) of truss
resistance design as it is based on the nonlinear dynamic behavior members is reduced by using the following equation to consider
of the structures. the influence of buckling on their hysteretic behavior and energy
dissipation capacity:
5.2. Effect of convergence parameter
25
Pc ¼ KL Py ð5Þ
In order to study the effect of convergence parameter, a, the r
previous example was solved for different values of a. Fig. 6 shows
how the total structural weight varies from the initial ground where (KL/r) and Py are effective slenderness ratio and yield
structure (with uniform cross-sectional area) toward the optimum strength of truss members, respectively. This hysteresis model, in
design using values of 0.05, 0.1, 0.2 and 0.5 for a. It is evident that general, compares well with the inelastic behavior observed in
as a increases from 0.05 to 0.2, the convergence speed increases experimental investigations [25]. In this study, the cyclic behavior
without any fluctuation. However, for a equal to 0.5, the method of axially loaded members after buckling has been considered at
is not stable and the problem does not converge to the optimum element level, and local buckling has not been considered directly.
solution. It is concluded that an a value of 0.2 results in the best However, local buckling of individual elements could be accounted
convergence for this design problem. Numerous analyzes carried in exactly the same way, but with a different hysteretic model. To
out on different structures indicate that, for non-linear truss struc- eliminate the effect of discrete section sizes, auxiliary tube sections
tures, an acceptable convergence is usually obtained by using a have been artificially developed by assuming a continuous variation
values of 0.1–0.2. of section properties based on those of ASTM.
In order to achieve better convergence, when buckling behavior
5.3. Effect of initial cross-sectional area of compressive members is considered in the optimization proce-
dure, the criterion for selection of structural members should be
An initial cross-sectional area is always needed to begin the based on the effective slenderness ratio of truss members rather
optimization algorithm. To investigate the effect of this initial than their cross-sectional area. Therefore, in this case, the proposed

Fig. 6. Variation of total structural weight using different convergence parameters, Fig. 7. Convergence to the optimum solution from two different initial ground
lt = 4. structures with uniform cross-sectional area equal to 10 and 1000 mm2, lt = 4.
 I. Hajirasouliha et al. / Computers and Structures 89 (2011) 702–711 707

optimization algorithm eliminates truss members with relatively depends on target ductility demand; and therefore, a fixed
high slenderness ratio (e.g. over five times of the average slender- arrangement of truss members is not appropriate for different per-
ness ratio of all remaining elements). After eliminating less effi- formance levels. This confirms that conventional optimization
cient members, the buckling length of the remaining elements is methods which are usually based on elastic vibration modes may
modified automatically. This may increase the second order effects not lead to an optimum design for structures in the nonlinear
(i.e. P–D effects) for some members that remain active in the struc- range of behavior.
ture. If the new design solution requires more structural weight, it
is more cost effective to keep intersecting elements. In this case, 5.6. Effect of seismic excitation
the elimination process is reversible and a minimum cross-sec-
tional area is chosen for the intersecting elements. These intersect- Four strong ground motion records were used to evaluate the
ing elements are just required to provide lateral stability for long effect of seismic excitation on the optimum topology of non-linear
length elements, and their cross-sectional areas do not change dur- truss structures, including EQ1: the 1994 Northridge earthquake
ing the optimization process. Table 1 shows the cross-sectional CNP196 component with a PGA (peak ground acceleration) of
area of truss members for the optimum solution in this case. 0.42 g, EQ2: the 1994 Northridge earthquake STC180 component
The optimum topology with and without considering the buck- with a PGA of 0.48 g, EQ3: the 1979 Imperial Valley earthquake
ling of compressive members are compared in Fig. 4. It is shown H-E04140 component with a PGA of 0.49 g, and EQ4: the 1979
that buckling of compressive members influences the optimum Imperial Valley earthquake H-E05140 component with a PGA of
topology of the truss structure by avoiding long elements that 0.52 g. All of these excitations correspond to the sites with soil pro-
are more vulnerable to buckling. file similar to the SD type of UBC-97 [26] and were recorded in a
low to moderate distance from the epicenter (less than 45 km)
5.5. Effect of target ductility demand with rather high local magnitudes (i.e., ML > 6.7). Acceleration re-
sponse spectra of these records are relatively close to each other
To study the effect of non-linear behavior of members on the as shown in Fig. 9.
optimum topology of truss structures, the proposed optimization The normalized cross-sectional area (with respect to maximum
algorithm was applied on the previous example (without buckling) member cross-sectional area) of optimum structures determined
for different values of target ductility demands. The cross-sectional for the above-mentioned seismic excitations are compared in Table
areas of optimum design structures for target ductility demands of 3 for target ductility demands of 1 and 4. It is shown that every
1, 1.5, 2, 3, 4, 5, 6 and 8 are given in Table 2. As expected, Table 2 seismic excitation has a unique optimum distribution of structural
shows that cross-sectional area of optimum designed structures, in materials. While the optimum topologies for elastic truss struc-
general, was decreased by an increase in target ductility demand. tures (lt = 1) seem to be very similar in the four earthquake exci-
The optimum topology for different target ductility demands are tations, the results indicate that they could be very different in
compared in Fig. 8. It is shown in this figure that optimum topology the non-linear range of behavior (lt = 4).
To obtain required performance criteria (such as maximum
ductility demand) during the design earthquake, a structure should
Table 2
The optimum arrangement of truss members for different target ductility demands.
be designed based on an appropriate load level. In the conventional
force-based seismic design methods, a force reduction factor
Member Cross-sectional area (cm2) (behavior factor q) is generally considered in the calculation of
lt = 1 lt = 1.5 lt = 2 lt = 3 lt = 4 lt = 5 lt = 6 lt = 8 design seismic loads that accounts for inherent ductility of a
1 34.0 26.0 17.9 12.7 10.7 11.8 4.8 3.8 structure:
2 0 0 0 0 0 0 0 0
3 0 0 0 0 0 0 0 0 Ve
q¼ ð6Þ
4 0 0 0 4.9 10.5 10.5 0 0 Vy
5 35.5 33.0 20.9 6.2 0 0 5.9 4.7
6 0 0 0 0 0 0 0 0 where Ve and Vy are lateral strength for elastic and inelastic struc-
7 11.2 5.4 5.9 7.5 5.0 7.0 2.2 1.4
tures, respectively. For optimized truss-like structures designed
8 35.5 33.0 20.9 6.2 0 0 5.9 4.7
9 0 0 0 4.9 10.5 10.5 0 0 mainly to resist earthquake loads, total structural weight could be
10 0 0 0 0 0 0 0 0 considered proportional to the design seismic load and lateral
11 0 0 0 0 0 0 0 0 strength. Therefore, for this type of structural systems, the behavior
12 34.0 26.0 17.9 12.7 10.7 11.8 4.8 3.8 factor q could be expressed by the following equation:
13 11.2 5.4 5.9 7.5 5.0 7.0 2.2 1.4
14 63.7 59.2 31.2 4.2 6.3 0 4.7 4.6 We
15 0 0 0 0 0 0 0 0 q¼ ð7Þ
16 0 0 0 0 0 0 0 0 Wl
17 0 0 0 0 0 0 0 0
18 63.7 59.2 31.2 4.2 6.3 0 4.7 4.6 where We and Wl are total structural weight of elastic and inelastic
truss structures, respectively.

µ= 1 to 2 µ= 3 µ= 4 µ= 5

Fig. 8. Optimum topology for different target ductility demands, Northridge 1994 (CNP196).
708 I. Hajirasouliha et al. / Computers and Structures 89 (2011) 702–711

We / Wµ
Fig. 9. Acceleration response spectra of selected seismic excitations.
Target Ductility

Fig. 10. Required structural weight ratio of elastic to inelastic structure as a

For each target ductility demand, the required structural weight
function of target ductility demand in different earthquakes.
is calculated for the optimum designed truss structures subjected
to different seismic excitations. Fig. 10 compares the ratio of re-
quired structural weight for elastic (lt = 1) to inelastic structures
with the results of Eq. (7). The results indicate that for low to med- by averaging the optimum patterns corresponding to a number
ium target ductility demands (i.e. lt 6 4); the structural weight ra- of earthquakes representing a design spectrum. To verify the valid-
tio was close to the target ductility demand. This implies that the ity of this idea for seismic design of truss structures, the average of
required structural weight for optimum elastic truss structure optimum material distribution pattern corresponding to the four
was almost lt times higher than a similar inelastic structure. selected earthquakes were calculated for the previous example
Therefore, in practical applications where lt is usually less than using different target ductility demands. Subsequently, these aver-
4, the behavior factor q for truss structures resisting earthquake age patterns were used to design an example truss structure, sim-
loads could be considered equal to the target ductility demand. ilar to the one shown in Fig. 3. For each seismic excitation, the
Similar results have been obtained by using different truss struc- cross-sectional areas of the truss structure were scaled, without
tures and seismic ground motions [22]. However, to propose an changing the primary pattern, until the maximum member ductil-
appropriate behavior factor for truss-like structures, further analy- ity reached the target ductility demand. As expected, the required
sis should be done, something that is not within the scope of this structural weight for this structure was higher compared to the
study. optimum design corresponding to each specific seismic excitation.
Fig. 11 compares the ratio of required to optimum structural
weight for truss structures designed using the equivalent static
5.7. Design for a group of earthquakes loads (Fig. 5a) and the average of the optimum material distribu-
tion patterns. This figure shows the average of the results obtained
The seismic load that a structure may experience during its life- in the four selected earthquakes (Eq. (1)–(4)). It is shown that using
time is an important source of uncertainty. While each of the fu- equivalent static loads and elastic analysis resulted in truss struc-
ture seismic events will have its own signature, it is generally tures with up to 60% more structural weight compared to the opti-
accepted that they will have relatively similar characteristics. Pre- mum designed models based on non-linear dynamic behavior of
vious studies by Moghaddam and Hajirasouliha [27] showed that, structures. It should be noticed that truss structures designed
for shear building models, a better seismic design could be found using the average of optimum patterns and the equivalent static

Table 3
Comparison of normalized cross-sectional area corresponding to optimum topology in four different seismic excitations, lt = 1 and 4.

Member Scaled cross-sectional area (lt = 1) Scaled cross-sectional area (lt = 4)

EQ1 EQ2 EQ3 EQ4 EQ1 EQ2 EQ3 EQ4
1 0.53 0.49 0.44 0.60 1 0.93 1 0.87
2 0 0 0 0 0 0 0 0
3 0 0 0 0 0 0 0 0
4 0 0 0 0 0.98 0 0 0
5 0.56 0.63 0.51 0.76 0 1 0.90 0.97
6 0 0 0 0 0 0 0 0
7 0.18 0.15 0.11 0.23 0.47 0.47 0.68 0.36
8 0.56 0.63 0.51 0.76 0 1 0.90 0.97
9 0 0 0 0 0.98 0 0 0
10 0 0 0 0 0 0 0 0
11 0 0 0 0 0 0 0 0
12 0.53 0.49 0.44 0.60 1 0.93 1 0.87
13 0.18 0.15 0.11 0.23 0.47 0.47 0.68 0.36
14 1 1 1 1 0.59 0.73 0 1
15 0 0 0 0 0 0 0 0
16 0 0 0 0 0 0 0 0
17 0 0 0 0 0 0 0 0
18 1 1 1 1 0.59 0.73 0 1
 I. Hajirasouliha et al. / Computers and Structures 89 (2011) 702–711 709

a b

12

14 @ 2m = 28 m
13 13
2 2
10

9 9
11 11

3 8 8 3

4 5 6 6 5 4
1 1
7

Fig. 11. The ratio of required to optimum structural weight for truss structures
designed using equivalent static loads and average of optimum material distribu- 8 @ 2m = 16 m
tion patterns (average for 4 earthquakes).
Fig. 12. (a) Geometry and potential nodes of initial ground structure. (b) Optimum
topology and identification number of truss elements, Northridge 1994 (CNP196).

loads have similar member ductility demand (performance objec-

tive), and therefore, are of similar safety margin. This implies that
using extra structural material did not provide higher safety for the = 0.5ey). In this example the strength of truss members is consid-
conventionally designed structures. ered to be similar in tension and compression (buckling restrained
The results indicate that, for the same target ductility demand, members). The Rayleigh damping model with a constant damping
the required structural weight for truss structures designed ratio of 0.05 was used for the first mode and for any mode at which
according to the average of the optimum material distributions is the cumulative mass participation exceeded 95%. Symmetry con-
on average 20% less than those optimized based on the equivalent strains are considered by applying the design seismic excitation
static loads. Fig. 11 shows that the efficiency of using the average of in two opposite directions (two different load cases).
the optimum distributions patterns is relatively higher for elastic A highly connected initial ground structure with a reasonable
and near-elastic structures (i.e. lt 6 1.5) where optimum distribu- member distribution was initially selected using basic engineering
tion patterns for different earthquakes are almost similar. How- judgement. The cross-sectional area, axial stress ratio (due to grav-
ever, using this average pattern always leads to a reduction in ity loads) and maximum member ductility demand of the opti-
the required structural weight compared to the conventional de- mum structure are shown in Table 4. The optimum topology of
sign methods. Therefore, the average of the optimum distribution the design example is illustrated in Fig. 12b. It is shown that out
patterns corresponding to a group of similar earthquakes, repre- of more than 800 truss members in the initial ground structure,
senting a design spectrum, seems to be appropriate for seismic de- only 26 members (13 different types as shown in Fig. 12b) re-
sign of non-linear truss structures. For engineering practices, a set mained in the final solution. While maximum axial stress due to
of synthetic earthquakes having a close approximation to the code- the gravity loads is less than 0.5fy, member ductility demand for
based elastic design spectrum could be utilized. all remaining elements is equal to the target ductility of 1.5. There-
fore, the material distribution is regarded as being optimum.
Surprisingly, the optimum solution results in a rather simple
6. Optimum design for multiple load cases
structural configuration preferring triangular arrangements to car-
ry all individual masses. It is shown that all truss elements of the
The proposed optimization algorithm, can be easily extended to
optimum designed structure are only tied at the ground level
consider multiple load cases for practical design applications. To
and two more levels. Since the mass distribution of a structure is
achieve this, in the proposed optimization algorithm, Eq. (4) should
normally more complex than the one shown, the optimum topol-
be substituted with the following equation:
" # ogy found for this structure is not likely to be the same for a more
ðei Þ1 ðei Þ2 ðei Þj uniformly distributed load.
SRi ¼ Max ; ;...; ð8Þ
ðet Þ1 ðet Þ2 ðet Þj
Table 4
Cross-sectional area, axial stress ratio (due to gravity loads) and
where (ei)j and (et)j are maximum strain and target strain of the ith
maximum member ductility demand of optimum design truss
member for jth load case, respectively. Using Eq. (8), different load structure shown in Fig. 10b.
cases can be considered in the design process for both elastic and
Member type Cross-sectional rmax/fy Ductility
inelastic range of behavior. This leads to an optimum design that
area (cm2)
satisfies all the design requirements.
1 167.2 0.19 1.50
The application of the proposed optimization method for a truss
2 485.0 0.07 1.49
structure subjected to a multiple load case (seismic and gravity 3 33.4 0.37 1.50
loads) is demonstrated by considering an illustrative example 4 552.4 0.17 1.50
shown in Fig. 12. The objective is to design an optimum truss struc- 5 361.3 0.12 1.49
ture to sustain seven 100 ton-force equipment located in a 7-storey 6 176.4 0.18 1.50
7 125.8 0.30 1.50
28 m height building (Fig. 12). The maximum member ductility de- 8 13.1 0.38 1.49
mand of 1.5 (i.e. et = 1.5ey) is considered as the design criterion 9 222.6 0.17 1.49
when the truss structure is subjected to the horizontal component 10 12.7 0.41 1.50
of the Northridge Earthquake of 1994 (CNP196). At the same time 11 303.7 0.18 1.50
12 106.7 0.32 1.49
it is required that the truss elements are capable of resisting grav-
13 185.0 0.18 1.49
ity loads with maximum axial stress (rmax) less than 0.5fy (i.e. et
710 I. Hajirasouliha et al. / Computers and Structures 89 (2011) 702–711

a b discussions, the proposed optimization method seems to be reli-

able and should prove useful in practical performance-based seis-
mic design of truss like structures. However, more research needs
to be done to extend this method to include a penalty factor to ac-
count for the cost of joints and incorporate more complex multi-
criteria objectives.
14 @ 2m = 28 m

7. Conclusions

In this study a practical optimization method is presented for

optimum seismic design of non-linear truss structures. Based on
the results, the following conclusions can be drawn:

 The concept of uniform deformation can be used efficiently for

topology optimization of nonlinear truss structures subjected to
8 @ 2m = 16 m gravity loads and seismic excitations. It is demonstrated that
Fig. 13. (a) Geometry and potential nodes of initial ground structure. (b) Optimum there is a unique optimum distribution of structural properties,
topology and identification number of truss elements, Northridge 1994 (CNP196). which is independent of the initial cross-sectional area of the
ground structure.
 The results indicate that optimal topology is influenced by the
To investigate the effect of mass distribution on the optimum variation of target ductility demand; and therefore, a fixed
topology, the above mentioned example was solved by considering arrangement of truss members cannot be appropriate for differ-
two 50 ton-force masses located at each storey, as shown in Fig. 13. ent performance levels. In addition, buckling of compressive
Fig. 13b shows the optimum topology of the design example in this members influences the optimum topology by avoiding long
case. The cross-sectional area, axial stress ratio (due to gravity slender elements.
loads) and maximum member ductility demand of the optimum  It is shown that using conventional optimization methods based
structure are shown in Table 5. The results indicate that this opti- on elastic behavior and equivalent static loads could lead to a
mum structure is capable of resisting seismic and gravity loads design with up to 60% more structural weight compared to
without violating any design constraints. Although the resulting the optimum designed model. It was concluded that non-linear
topology is more complex compared to the previous structure, dynamic behavior of truss structures should be considered in
the structural configuration follows a similar trend. The masses optimum topology design of truss structures subjected to seis-
are still carried by simple triangles which are tied at the ground mic excitations.
and two more levels. It is anticipated that if the floor elements  While there is a specific optimum topology corresponding to
were maintained at each level, since they are required to carry each seismic excitation, it is shown that using the average of
the uniform dead-loads anyway, a different optimum topology will optimum distribution patterns for a group of earthquakes leads
be obtained. to acceptable results for practical applications. The results indi-
It should be mentioned that in the performance-based design cate that, for the same target ductility demand, truss structures
methods, different multiple limit states (e.g. service event, rare designed according to the average of the optimum material dis-
event, very rare event) are usually considered. Optimum design tributions have on average 20% less structural weight compared
for a specific limit state does not guarantee the optimum behavior to those designed based on the conventional elastic optimiza-
in other conditions. In this case, it is usually accepted to consider tion methods.
the very rare event as the governing criterion for preliminary de-
sign, and control the design for other events. Based on the above
References
Table 5
Cross-sectional area, axial stress ratio (due to gravity loads) and maximum member [1] Vanderplaats GN. Numerical optimization technique for engineering
design. New York: McGraw-Hill; 1984.
ductility demand of optimum design truss structure shown in Fig. 11b.
[2] Imai K, Schmit LA. Configuration optimization of trusses. J Struct Div ASCE
Member type Cross-sectional area (cm2) rmax/fy Ductility 1981;107:745–56.
[3] Topping BHV. Shape optimization of skeletal structures: a review. J Struct Eng
1 344.2 0.08 1.50 1983;11:1933–51.
2 513.8 0.14 1.50 [4] Ohsaki M. Random search method based on exact reanalysis for topology
3 132.8 0.13 1.50 optimization of trusses with discrete cross-sectional areas. Comput Struct
4 298.9 0.10 1.49 2001;79:673–9.
5 153.4 0.14 1.50 [5] Pedersen P. Optimal joint positions for space trusses. J Struct Div ASCE
6 216.8 0.05 1.49 1973;99:2459–76.
7 294.3 0.04 1.51 [6] Kirsch U. Optimal topologies for truss structures. Comput Method Appl Mech
8 436.1 0.04 1.50 Eng 1989;72:15–28.
[7] Xie YM, Steven GP. Evolutionary structural optimization. Berlin: Springer-
9 321.8 0.15 1.50
Verlag; 1997.
10 213.4 0.07 1.50
[8] Bendsoe MP, Sigmund O. Topology optimization: theory, methods, and
11 214.5 0.05 1.50
applications. Berlin: Springer-Verlag; 2003.
12 181.9 0.03 1.50 [9] Liang QQ. Performance-based optimization of structures: theory and
13 254.4 0.07 1.51 applications. New York: Spon Press; 2005.
14 98.4 0.08 1.50 [10] Seireg AA, Rodriguez J. Optimizing the shape of mechanical elements and
15 130.6 0.13 1.51 structures. New York: Marcel Dekker Inc.; 1997.
16 106.8 0.07 1.50 [11] Mattheck C. Design in nature: learning from trees. Berlin: Springer-Verlag;
17 173.5 0.09 1.49 1998.
18 41.1 0.46 1.50 [12] Ohsaki M. Genetic algorithm for topology optimization of trusses. Comput
19 160.9 0.14 1.49 Struct 1995;57:219–25.
20 285.3 0.09 1.51 [13] Soh CK, Yang J. Optimal layout of bridge trusses by genetic algorithms.
Comput-aided Civil Infrastruct Eng 1998;13:247–54.
 I. Hajirasouliha et al. / Computers and Structures 89 (2011) 702–711 711

[14] Deb K, Gulati S. Design of truss-structures for minimum weight using genetic [22] Hajirasouliha I. Optimum strength distribution for seismic design of
algorithms. Finite Elem Anal Des 2001;37:447–65. structures. PhD thesis, Civil Engineering Dept., Sharif University of
[15] Rahami H, Kaveh A, Gholipour Y. Sizing, geometry and topology optimization Technology, Tehran, Iran; 2006.
of trusses via force method and genetic algorithm. Eng Struct 2008;30:2360–9. [23] Hagishita T, Ohsaki M. Topology optimization of trusses by growing ground
[16] Ohsaki M. Simultaneous optimization of topology and geometry of a regular structure method. Struct Multidiscip Optim 2009;37:377–93.
plane truss. Comput Struct 1998;66:69–77. [24] Prakash V, Powell GH, Filippou FC. DRAIN-2DX: base program user guide. UCB/
[17] Hart GC. Earthquake forces for the lateral force code. Struct Des Tall Build SEMM-92/29, Earthquake Engineering Research Centre, University of
2000;9:49–64. California, Berkeley, 1992.
[18] Hajirasouliha I, Moghaddam H. A new lateral force distribution for seismic [25] Jain AK, Goel SC, Hanson RD. Hysteretic cycles of axially loaded steel members.
design of structures. J Struct Eng ASCE 2009;135:906–15. J Struct Div ASCE 1980;106:1777–95.
[19] Moghaddam H, Hajirasouliha I. Toward more rational criteria for [26] Uniform Building Code, UBC. International conference of building officials, vol.
determination of design earthquake forces. Int J Solid Struct 2006;43:2631–45. 2; 1997.
[20] Kirsch U. Structural optimization: fundamentals and applications. New [27] Moghaddam H, Hajirasouliha I. Optimum strength distribution for seismic
York: Springer-Verlag; 1993. design of tall buildings. Struct Des Tall Special Build 2008;17:331–49.
[21] FEMA 356. Prestandard and commentary for the seismic rehabilitation of
buildings. Washington, DC: Federal Emergency Management Agency; 2000.