Optimizing Single-Span Steel Truss Bridges With Simulated Annealing

See discussions, stats, and author profiles for this publication at: https://www.researchgate.
net/publication/260705761
OPTIMIZING SINGLE-SPAN STEEL TRUSS BRIDGES WITH SIMULATED

ANNEALING
Article in Asian Journal of Civil Engineering · January 2010
CITATION READS
1 3,989
2 authors:
O. Hasançebi E. Dogan
Middle East Technical University Manisa Celal Bayar University
60 PUBLICATIONS 2,754 CITATIONS 23 PUBLICATIONS 705 CITATIONS
SEE PROFILE SEE PROFILE
Some of the authors of this publication are also working on these related projects:
The 10th International Student Competition in Structural Optimization (ISCSO 2021) View project
All content following this page was uploaded by O. Hasançebi on 12 March 2014.
The user has requested enhancement of the downloaded file.

ASIAN JOURNAL OF CIVIL ENGINEERING (BUILDING AND HOUSING) VOL. 11, NO. 6 (2010)
PAGES 763-775
OPTIMIZING SINGLE-SPAN STEEL TRUSS BRIDGES WITH

SIMULATED ANNEALING
O. Hasançebi*a and E. Doğanb

a
Department of Civil Engineering, Middle East Technical University, Ankara, Turkey
b
Department of Engineering Sciences, Middle East Technical University, Ankara, Turkey
ABSTRACT
This study presents applications of a simulated annealing integrated solution algorithm to

the optimum design of single-span steel truss bridges subjected to gravity loadings. In the
optimum design process of a bridge the members are sized simultaneously as the coordinates
of the upper chord nodes are determined such that the least design weight is attained for the
structure. The design constraints and limitations are imposed in accordance with serviceability
and strength provisions of ASD-AISC (Allowable Stress Design Code of American Institute of
Steel Institution) specification. A numerical example is presented, where optimum designs
produced according to nine alternative topological forms of single-span truss bridges, namely
Pratt, Parker, Baltimore, Pettit, K-Truss, Warren, Subdivided Warren, Quadrangular Warren
and Whipple are compared for a selected span length of 600 ft (182.88 m) to quantify the
influence of choice of a topological form on the final design weight of the bridge.
Keywords: Structural optimization; simulated annealing; single-span steel truss bridges;

bridge topological forms; minimum weight design
1. INTRODUCTION
Over the years steel truss bridges have kept their popularity amongst bridge engineers. Apart
from architectural attractiveness, these systems exhibit certain advantages from structural and
constructional standpoints, such as speed of construction, durability, modification and repair,
recycling, versatility, etc. They are rigid skeletal structures with straight members configured
generally in some form of triangles to transfer the design loads from deck to the piers through
mainly axial forces of the members. Single span steel truss bridges refer to a particular subset
of these systems where the whole opening is crossed with a single span with generally simply
supported end conditions. Especially, they are more preferable with respect to multi-span
bridges in cases where disturbance to the stream bed needs to be avoided.
The single span truss bridges can be designed in a variety of different topological forms,
such as Pratt, Parker, Baltimore, Pettit, K-Truss, Warren, Subdivided Warren, Quadrangular
*
E-mail address of the corresponding author: oguzhan@metu.edu.tr (O. Hasançebi)
764 O. Hasançebi and E. Doğan
Warren, Whipple, etc. The Pratt truss has a topological feature such that the diagonals are all
sloped in the same direction on each side of the truss around the mid-span. In this form of
truss bridge, the upper chord and vertical members are subjected to compression whereas the
diagonals and lower chord members are under tension. The Baltimore truss has additional
bracing members in the lower section to prevent buckling in the compression members as well
as to control deflection. Both Pratt and Baltimore trusses have a constant height throughout the
span length of the bridge. The Parker and Pettit trusses have identical topological forms with
Pratt and Baltimore trusses, respectively, except that the formers have a polygonal shape with
the bridge height increasing from the ends towards the mid-span. In the Warren truss alternate
diagonals sloped in different directions frame into each others at lower chord nodes. The
Subdivided Warren truss has the topological form of Warren truss with vertical members
having sub-diagonals and sub-verticals. The Quadrangular Warren is a double intersection
truss form with alternating tension and compression diagonals. The K-truss is configured in
the form of letter “K” by the orientation of the vertical member and two oblique members in
each panel. Finally, the Whipple truss is the one having elongated and usually thin tension
members which cross two or more members.
In the present study, optimum design of single-span steel truss bridges is investigated in
conjunction with simulated annealing (SA) optimization technique. The SA employs a meta-
heuristic numerical optimization procedure, the theory of which extends to the annealing
process of physical systems in thermodynamics [1]. In this process a physical system (a
solid or a liquid) initially at a high-energy state is cooled down to reach the lowest energy
state. The idea that this process can be simulated to solve optimization problems was put
forward by Kirkpatrick et al. [2] by establishing a direct analogy between minimizing the
energy level of a physical system and lowering the cost of an objective function. The
successful applications of the technique in the fields of structural optimization and
computational structural mechanics have been reported in a number of publications in the
literature, such as Refs. [3-8].
In the solution process both size and shape variables are employed simultaneously by the
SA algorithm to minimize the design weight of a bridge. Size variables are used to
determine the required steel sections for the bridge members to satisfy strength and
serviceability requirements imposed according to provisions of ASD-AISC [9] standard.
Shape variables, on the other hand, are utilized to find the best height or shape of the upper
chord of a bridge. The application of the solution algorithm developed is illustrated with a
numerical design example, where the optimum designs produced according to
abovementioned nine different topological forms of single-span bridges are compared for a
selected span length of 600 ft (182.88 m). This way, the weight efficiency of various
topological forms is investigated and it is shown that the choice of a topological form has a
great impact on the final design weight of the bridge.
2. OPTIMUM DESIGN PROBLEM FORMULATION
In the optimum design process of a single-span steel truss bridge, it is required to select the
structural members from a standard steel section table as well as to determine the height or
OPTIMIZING SINGLE-SPAN STEEL TRUSS BRIDGES WITH... 765
shape of the upper chord of the bridge such that the structure satisfies the strength and
serviceability limitations imposed by a code of practice, while the minimum weight of the
bridge is attained. The formulation of design constraints according to ASD–AISC [9] yields
the following mixed discrete programming problem.
Find a vector of size design variables A and a vector of shape design variables S,

A T  A1 , A2 ,..., AN m  and S T

 S1 , S 2 ,..., S N s  (1)
to minimize the weight of the bridge (W),
Nm
W    m L m Am (2)
m 1
subject to a set of behavioural and performance limitations expressed as
m
gm   1  0, m  1,.., N m (3)
( m ) all
m
sm   1  0, m  1,.., N m (4)
(m ) all
d j ,k
 jk   1  0, j  1,.., N j (5)
,
(d j ,k ) all
In Eqns. (1)-(5), N m is the total number of bridge members; Am Lm ,  m are the cross-
sectional area, length and unit weight of the m-th bridge member, respectively; S j is the j-
th shape variable; N s is the total number of shape variables associated with the coordinates
of the upper chord nodes of the bridge; the functions g m , s m and  j,k are referred to as
constraints being bounds on stresses, slenderness ratios and displacements, respectively;  m
and ( m ) all are the computed and allowable axial stresses for the m-th member,
respectively; m and (m ) all are the slenderness ratio and its upper limit for m-th member,
respectively; N j is the total number of nodes in the bridge; and finally d j ,k and (d j ,k ) all are
the displacements computed in the k-th direction of the j-th node and its permissible value,
respectively.
According to ASD-AISC [9], the maximum slenderness ratio is limited to 300 and 200
for tension and compression members, respectively. Therefore, the slenderness related
design constraints can be formulated as follows:
K m Lm 300 (for tension members)

m   (6)
rm 200 (for compression members)
where K m is the effective length factor of m-th member ( K m  1 for all members), and rm is
its minimum radii of gyration.
The allowable tensile stresses for tension members are calculated as in Eqn. (7).
( t ) all  0.60 Fy and ( t ) all  0.50 Fu (7)
where Fy and Fu stand for the yield and ultimate tensile strengths, and the smaller of the
two formulas is considered to be the upper level of axial stress for a tension member.
The allowable stress limits for compression members are calculated depending on two
possible failure modes of the members known as elastic and inelastic buckling, Eqn. (8)
  (K m L m / rm ) 2 
 1  2  Fy
  2 C 
c
 5 3(K L / r ) (K L / r ) 3 ,  m  C c (inelastic buckling)
 c all   m m m
 m m3 m (8)
 3 8 C c 8C c
 12 E
2
 ,  m  C c (elastic buckling)
 23(K m L m / rm ) 2
where E is the modulus of elasticity, and C c  2 2 E / Fy is the critical slenderness ratio

between elastic and inelastic buckling.
3. SIMULATED ANNEALING ALGORITHM
The basic computational steps of simulated annealing algorithm are outlined in the
following. The further enhancement of the algorithm that results in additional performance
of the technique can be found in Hasançebi et al. [10].
Step 1. Cooling schedule: The first step is the setting of an appropriate cooling schedule.
After choosing suitable values for the starting acceptance probability ( Ps ), the final
acceptance probability ( Pf ), and the number of cooling cycles ( N c ), the cooling schedule
parameters are calculated as follows:
1 / N c 1
1 1  ln( Ps ) 
Ts   , Tf   ,   (9)
ln( Ps ) ln( Pf )  ln( Pf ) 
In Eqn. (9), Ts , T f and  are referred to as starting temperature, final temperature, and
cooling factor, respectively. The starting temperature is assigned as the current temperature
of the process, i.e., T  Ts .
Step 2. Initial Design: Next, an initial design is originated via random initialization in a
way such that size variables are set to arbitrary ready sections selected from a standard
profile list and shape variables are assigned to any values between their lower and upper
bounds. The analysis of the bridge is then performed and the force and deformation
responses are obtained under the applied loads. If all the constraints are satisfied, the
objective function value (  ) of the design is directly computed from Eqn. (2), i.e.,   W .
If, however, the design violates some of the problem constraints, it is penalized and its
objective function value is calculated from Eqn. (10).
  
  W 1     ci  (10)
  i 
In Eqn. (10), ci is the i-th problem constraint of the problem at hand, and  is the
penalty coefficient used to tune the intensity of penalization as a whole. It is set to an
appropriate static value, such as   1 .
Step 3. Generating candidate designs: A number of candidate designs are generated in

the close vicinity of the current design. This is carried out as follows: (i) either a size or
shape design variable is selected, (ii) the selected variable is given a small perturbation in a
predefined neighborhood, and (iii) finally, a candidate design is generated by assuming the
perturbed value of the variable, while keeping all others same as in the current design. It
follows that a candidate design differs from the current one in terms of one design variable
only. It is important to note that each design variable is selected only once in a random order
to originate a candidate design. Hence, the total number of candidate designs generated in a
single iteration of the cooling cycle is equal to the number of design variables.
Step 4. Evaluation of a candidate design: Each time when a candidate design is

generated, its objective function value (  ) is computed first and then it is set to compete
with the current design. If the candidate provides a better solution, it is accepted
automatically and it replaces the current design. Otherwise, the so-called Metropolis test is
employed to determine the winner, in which case the probability of accepting a poor
candidate (P) is assigned as follows:
P  exp(   / KT ) , where   a  c  0 (11)
In Eqn. (11), c and  a are the objective function values of the current and candidate
designs, respectively; T is the current temperature of the process; and K is the Boltzman
parameter which is manipulated as the working average of positive  values.
Step 5. Iterations of a Cooling Cycle: A single iteration of a cooling cycle is referred to

the case where all design variables are selected once and perturbed to generate candidate
designs. Generally, a cooling cycle is iterated a certain number of times in the same manner
to ensure that the objective function is reduced to a reasonably low value associated with the
temperature of the cooling cycle. Having selected the iterations of the starting and final
cooling cycles ( i s and i f ), the iteration of a cooling cycle ( ic ) at a given temperature T is
determined by a linear interpolation between i s and i f , as follows:
 T  Tf 
ic  i f  i f  is   (12)
T T 
 f s 
Step 6. Reducing Temperature: After the cooling cycle is iterated a number of times, the
temperature is reduced by the ratio of the cooling factor η, and the temperature of the next
cooling cycle is set as in Eqn. (13)
T k 1  T k (13)
where T k and T k 1 denote the temperature at the k and (k+1)-th cooling cycles,
respectively.
4. NUMERICAL EXAMPLE
A numerical design example is studied here, (i) to verify and demonstrate the applicability
of the solution algorithm to the optimum design of single-span steel truss bridges, (ii) to
explore the weight efficiency of various topological forms of a bridge under gravity loads,
and finally (iii) to emphasize the significance of a bridge topological form on its final design
weight.
Figure 1. A single-span steel truss bridge design example
In the example it is intended to build up a single-span steel truss bridge to cross an

opening of 600 ft (182.88 m). Only the lower chord of the bridge is geometrically defined
as shown in Figure 1, and it consists of 25 panel points equally spaced at 25 ft (7.62 m). The
design loads are calculated according to the provisions of ASCE 7-05 [11]. Live loads
resulted from traffic are combined with dead loads of the deck and floor systems, which are
later on transmitted to the lower chord, resulting in equivalent panel point loads of 60 kips
(267 kN) each in each model.
a) Pratt truss
b) Parker truss
c) Baltimore truss
d) Pettit truss
e) K-truss
f) Warren truss
g) Quadrangular Warren truss
h) Subdivided Warren Truss
i) Whipple truss
Figure 2. Bridge topological forms: a) Pratt, b) Baltimore, c) Pettit, d) Baltimore, e) K-truss, f)
Warren, g) Quadrangular Warren, h) Subdivided Warren, i) Whipple
To comprise the superstructure, nine alternative topological forms of single-span truss

bridges shown in Figure 2 are considered. In the optimum design process with each
topological form, both size and shape design variables are employed together. Size variables
are used to determine the required steel sections for bridge members and are grouped
considering the symmetry of the structures around mid-span. Hence, the number of
independent size variables used in a model is equal to half of the total number of members in
the model. The size variables are selected from a total of 83 wide-flange ready sections
ranging between W10x12 and W14x730. The shape variables are chosen to define the height
or shape of the upper chord of a bridge model, again considering a desired symmetry of the
structure about mid-span. The number of shape variables used in a model depends on the
bridge topological form. For instance, a single shape variable is used to define the height of
the truss in Pratt, Baltimore, Warren, Subdivided Warren, Quadrangular Warren, Whipple
and K-truss forms, since these forms have straight upper chords. In Parker and Pettit forms,
however, the y-coordinates of all upper chord nodes are allowed to vary, resulting in 12 and
6 independent shape variables, respectively. The ranges of shape variables are chosen
between zero and half of the span length (L/2 =150 ft = 91.44 m). In all the bridge models,
the strength and stability requirements of the designs are specified as per ASD-AISC [9]
provisions, plus the maximum displacements of panel points in any direction are restricted to
1/600 of the total span length. The following material properties of the steel are used in all
the models: Fy (yield stress) = 36 ksi (2531 kg/cm2) and E (modulus of elasticity) = 29,000
ksi (2,038,936 kg/cm2).
The optimum design of the bridge under each of the nine topological forms was sought
by running the solution algorithm five times independently due to stochastic nature of the
SA technique. The control parameters are set to the following values in these runs in line
with recommendations of the former studies [10, 12-14]: Ps  0.50 , Pf  10 3 , is  1 ,
i f  5 , N c  300 , resulting in the cooling schedule parameters Ts = 1.4427, T f = 0.1448
and  = 0.9923 from Eqn. (10). The average computing time for a single run takes about 12
min on a serial computer with Intel Quad Core Q9300 2.5GHZ LGA775 processor.
The best (optimum) designs obtained for the bridge under nine different topological
forms are tabulated in Table 1 with section designations attained for each member group and
the resulting values of shape variables in conjunction with design variable numbering shown
in Figure 2 for each model. The minimum weight design for the bridge is produced by Pettit
truss with a design weight of 624,277.4 lb (283,172.2 kg). This design is followed by Parker
truss with the corresponding design weight of 722,558.3 lb (327,752. 4 kg). It is interesting
to note that both of these truss forms have a varying bridge height. Although a polygonal
shape is not enforced by any sort of constraints, the resulting shapes of the bridges generated
in the optimum design process confirm with this particular shape naturally. As compared to
Parker model, Pettit truss results in smaller member lengths, in which intermediate panel
points are defined between upper and lower chords to keep the member lengths within
reasonable limits, and thereby increasing the buckling strength of the members significantly.
Nevertheless, both truss forms outperform all the others with straight upper chords (constant
bridge height models). Amongst the trusses with straight upper chords, the subdivided forms
(K-truss, Subdivided Warren and Baltimore) perform more efficiently with the
corresponding design weights of 830,048.1 lb (376,509.8 kg), 855,874.5 lb (388,224.7 kg)
and 869,646.4 lb (394,471.6 kg), respectively owing to the increased buckling strength of
the members in these models. The largest design weights for the bridge are produced by
Pratt and Whipple trusses with design weights of 1,379,721.2 lb (625,841.5 kg) and
1,384,147.8 lb (627,849.4 kg), respectively. The design history curve representing the
variation of the best feasible obtained thus far in the optimum design process is plotted in
Figure 3 for each model.
Table 1: Optimum designs obtained for the bridge under each topological form.
Bridge topological forms
Variable
Pratt Parker Baltimore Pettit K_truss Warren Q_Warren S_Warren Whipple
Size variables, ready sections
A1 W12x136 W14x176 W12x152 W12x152 W10x68 W14x145 W14x74 W12x190 W14x109
Bridge topological forms

Variable
Pratt Parker Baltimore Pettit K_truss Warren Q_Warren S_Warren Whipple
A48 N/A N/A N/A N/A W12x72 N/A W12x210 N/A W14x730
A56 N/A N/A N/A N/A W12x40 N/A W14x730 N/A N/A
A59 N/A N/A N/A N/A W12x152 N/A N/A N/A N/A
Shape variables (in.)
y1 739 222 802 450 867 726 733 810 638
y2 N/A 363 N/A 736 N/A N/A N/A N/A N/A
y7 N/A 769 N/A N/A N/A N/A N/A N/A N/A
Weight,lb 1379721.2 722558.3 869646.4 624277.4 830048.1 1150368.2 1090091.5 855874.5 1384147.8
(kg) (625841.5) (327752.4) (394471.6) (283172.2) (376509.8) (521807.0) (494465.5) (388224.7) (627849.4)
2500000
K_truss
Parker
2250000 Baltimore
Petit
Best Feasible Design (lb) 2000000 Warren
Quad_warren
Pratt
1750000 Sub_warren
Whipple
1500000
1250000
1000000
750000
500000
0 10000 20000 30000 40000 50000
Number of Analysis
Figure 3. The design history curve of the best solution for each topological form.
5. CONCLUSIONS
This study concerns with application of a SA integrated solution algorithm to the optimum
design of single-span steel truss bridges. In the study, a bridge with a certain span length is
first configured according to nine topological forms commonly used for single-span steel
truss bridges, and the resulting structures are sized and shaped simultaneously for the
minimum design weight subject to strength, stability and displacement provisions of ASD-
AISC [9]. The optimum design produced for the bridge is 624,277.4 lb (283,172.4 kg) with
the most favourable model (Pettit) and 1,384,147.8 lb (627,849.4 kg) with the most
unfavourable model (Whipple), indicating that an appropriate choice of the bridge
topological model may lead to savings in material as much as 62 %. However, it should be
kept in mind that the cost of construction of the bridge cannot directly be related to its
weight; rather it is influenced by many other factors, including the type and number of joints
used in the structure, etc. For the design example considered here, Pettit and Parker trusses,
which have polygonal upper chords, turn out to be the most weight-effective models,
outperforming all the other forms with straight upper chords. Additional studies are required
to investigate the cost-efficiency of the bridge models with respect to a set of parameters,
such as span length, design loads, constraints, etc.
REFERENCES
1. Metropolis N, Rosenbluth AW, Rosenbluth M, Teller A, Teller H. Equations of state

calculations by fast computing machines, J. Chem. Phys. 21(1953) 1087–91.

2. Kirkpatrick S, Gelatt CD, Vecchi MP. Optimization by simulated annealing, Science,
220(1983) 671–80.
3. Balling RJ. Optimal steel frame design by simulated annealing, Journal of Structural
Engineering, No. 6, 117(1991) 1780–95.
4. Chen GS, Bruno RJ, Salama M. Optimal placement of active/passive members in truss
structures using simulated annealing, AIAA Journal, No. 8. 29(1991) 815–27.
5. Shim PY, Manoochehri S. Generating optimal configurations in structural design using
simulated annealing, International Journal for Numerical Methods in Engineering,
40(1997) 1053–69.
6. Ceramic B, Fryer C, Baires RW. An application of simulated annealing to the optimum
design of reinforced concrete retaining structures, Computers and Structures, 79(2001)
1569–81.
7. Erdal O, Sönmez FO. Optimum design of composite laminates for maximum buckling
load capacity using simulated annealing, Composite Structures, 71(2005) 45–52.
8. Değertekin SÖ. A comparison of simulated annealing and genetic algorithm for
optimum design of nonlinear steel space frames, Structural and Multidisciplinary
Optimization, 34(2007) 347–59.
9. Manual of Steel Construction, Allowable Stress Design, 9th edition, AISC, American
Institutes of Steel Construction, Inc, Chicago, Illinois, USA, 1989.
10. Hasançebi O, Çarbaş S, Saka MP. Improving the performance of simulated annealing in
structural optimization, Journal of Structural and Multidisciplinary Optimization,
41(2010) 189–203.
11. ASCE 7-05, Minimum Design Loads for Building and Other Structures, 2005.
12. Bennage WA, Dhingra AK. Single and multi-objective structural optimization in
discrete-continuous variables using simulated annealing, International Journal in
Numerical Methods in Engineering, 38(1995) 2753–73.
13. Hasançebi O, Erbatur F. On efficient use of simulated annealing in complex structural
optimization problems, Acta Mech, 157(2002) 27–50.
14. Hasançebi O, Erbatur F. Layout optimisation of trusses using simulated annealing, Adv
Eng Softw, 33(2002) 681–96.
View publication stats

Optimizing Single-Span Steel Truss Bridges With Simulated Annealing

Uploaded by

Copyright:

Available Formats

Optimizing Single-Span Steel Truss Bridges With Simulated Annealing

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Optimizing Single-Span Steel Truss Bridges With Simulated Annealing

Uploaded by

Copyright:

Available Formats

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

OPTIMIZING SINGLE-SPAN STEEL TRUSS BRIDGES WITH SIMULATED

Article in Asian Journal of Civil Engineering · January 2010

SEE PROFILE SEE PROFILE

The user has requested enhancement of the downloaded file.

OPTIMIZING SINGLE-SPAN STEEL TRUSS BRIDGES WITH

O. Hasançebi*a and E. Doğanb

This study presents applications of a simulated annealing integrated solution algorithm to

Keywords: Structural optimization; simulated annealing; single-span steel truss bridges;

2. OPTIMUM DESIGN PROBLEM FORMULATION

to minimize the weight of the bridge (W),

subject to a set of behavioural and performance limitations expressed as

K m Lm 300 (for tension members)

( t ) all  0.60 Fy and ( t ) all  0.50 Fu (7)

where E is the modulus of elasticity, and C c  2 2 E / Fy is the critical slenderness ratio

3. SIMULATED ANNEALING ALGORITHM

Step 3. Generating candidate designs: A number of candidate designs are generated in

Step 4. Evaluation of a candidate design: Each time when a candidate design is

P  exp(   / KT ) , where   a  c  0 (11)

parameter which is manipulated as the working average of positive  values.

Step 5. Iterations of a Cooling Cycle: A single iteration of a cooling cycle is referred to

Figure 1. A single-span steel truss bridge design example

In the example it is intended to build up a single-span steel truss bridge to cross an

g) Quadrangular Warren truss

h) Subdivided Warren Truss

To comprise the superstructure, nine alternative topological forms of single-span truss

Bridge topological forms

1. Metropolis N, Rosenbluth AW, Rosenbluth M, Teller A, Teller H. Equations of state

calculations by fast computing machines, J. Chem. Phys. 21(1953) 1087–91.

View publication stats

You might also like