Professional Documents
Culture Documents
DOI 10.1007/s00158-009-0404-2
INDUSTRIAL APPLICATION
The purpose of this study is to minimize the cost of 2.2 Design variables and design parameters
precast concrete and prestressing steel for middle span
bridge girders through topological and shape optimiza- The design variables considered in this study are as
tion. The considered bridge girders are composed of follows:
pretensioned prestressed concrete adjacent simply sup-
nb number of the girders at the bridge cross-section
ported I section beams. The topological optimization
h height of the precast beam
of such a structure requires determination of the op-
bb bottom flange width
timum number of adjacent beams. Shape optimization
bw web thickness
requires the determination of the dimensions of the I
tt top flange thickness
section beam. In this study, the topological and shape
tb bottom flange thickness
optimization of the girder are performed at the same
pt top sloped thickness
time using a GA to reach an optimal solution.
pb bottom sloped thickness
A computer program coded in Visual Basic is used
nps number of the prestressing steel bars in each
to identify the optimum design, and the program is then
beam.
used for several design examples.
A typical cross-section of a prestressed concrete I-
beam section is illustrated in Fig. 1 to highlight several
of the design variables. In this study, adjacent I-beam
2 Optimum design of prestressed concrete girders are considered. Therefore, the slab thickness is
bridge girders not taken as a design variable. The design parameters
in this study are previously selected bridge cross-section
2.1 Objective function dimensions, material properties, unit costs of materials,
loads and the other relevant parameters.
In this study, the design criterion is the cost of pre-
cast beams. The objective is to minimize the cost 2.3 Constraints
without violating the constraints. The cost of precast
beams includes the cost of the precast concrete and the A total of 28 constraints are considered, consistent
cost of the prestressing steel. The cost of the precast with the AASHTO Standard Specifications (AASHTO
beams also includes the cost of the non-prestressed 2002). These constraints can be categorized into six
conventional reinforcing steel and the cost of suitable groups:
frameworks; however, these costs do not contribute
significantly to the total cost compared to the costs of 1. Flexural working stress constraints.
the precast concrete and the prestressing steel. The 2. Flexural ultimate strength constraints.
total cost of a prestressed concrete bridge girder is
formulated as
bt
tt
where Cpc and Cps are the costs of the precast con-
pt
crete and the prestressing steel, respectively. The costs
of the precast concrete and the prestressing steel are
calculated as
hw
Cpc = uppc · nb · Apc · L (2) h bw
3. Shear constraints (working stress and ultimate where σ b , σ t and σ ts are the calculated stresses at the
strength). bottom edge of the precast section, at the top edge of
4. Ductility constraints. the precast section, and at the top edge of the composite
5. Deflection constraints. section, respectively; σ̄ci and σ̄ti are the initial allowable
6. Geometrical constraints. compression and tension stresses, respectively, of the
precast concrete; σ̄cf and σ̄tf are the associated final
These constraints are explained below. allowable compression and tension stresses; and σ̄cs is
the allowable compression stress of the traditional deck
2.3.1 Flexural working stress constraints concrete.
where Vu is the factored shear force at the section; sections, respectively, and β 1 is a concrete strength fac-
and Vc and Vs are the shear strengths provided by the tor as defined in the AASHTO Standard Specifications
concrete and the web reinforcement, respectively. φ in (AASHTO 2002).
this equation is assumed to be equal to 0.90 for shear. The total amount of the prestressing steel must be
sufficient to develop an ultimate moment that is no
2.3.4 Deflection constraints less than 1.2 times the cracking moment at the critical
section. Therefore, the minimum prestressing steel con-
The deflection constraints are considered for four dif- straints can be given as
ferent load cases, as in the case of the flexural work- ∗
Mcr,c · 1.2
ing stress. The deflection constraints are listed below, g23 = − 1, (26)
φ Mn
consistent with these loads and final cases.
∗
Mcr,p · 1.2
f1 g24 = − 1, (27)
g16 = − 1, (19) φ Mn
(L/500)
∗ ∗
where Mcr,c and Mcr,p are the cracking moments of the
f2 composite and the precast sections, respectively.
g17 = − 1, (20)
(L/500)
2.3.6 Geometrical constraints
f3
g18 = − 1, (21) The geometrical constraints in this study are given
(L/500)
below.
The thickness of the top flange must be adequate to
f4
g19 = − 1, and (22) transfer the latitudinal moment. Thus, the top flange
(L/1000) thickness of the optimal design section must be greater
than the constrained thickness. The related constraint
f can be given as
g20 = − 1, (23)
(L/500)
(tt + pt )min
g25 = −1 (28)
where f1 , f2 , f3 and f4 are deflections associated with (tt + pt )
the related number of load cases, and f is the deflection
of the final case. where (tt + pt )min is the minimum value of the top
It is considered that the deflection f1 is due to the flange thickness.
beam’s self weight; f2 is from the beam and deck self The width of the bottom flange must be smaller than
weights; f3 is due to all dead loads; f4 is due to the live the width of the top flange. The related constraint is
loads; and f is due to all live and dead loads.
bb
g26 = − 1. (29)
bt
2.3.5 Ductility constraints
The width of the web must be smaller than the width of
According to the AASHTO Standard Specifications the bottom flange. The related constraint is
(AASHTO 2002), the prestressed concrete members
must be designed so that the steel yields when the bw
g27 = − 1. (30)
ultimate capacity is reached. Therefore, the maximum bb
prestressing steel constraints for the composite and
The prestressing steels do not generally fit into one
precast sections are given below, respectively:
row. Thus, the thickness of the bottom flange must
ψc be adequate to fit the prestressing steels. The related
g21 = − 1, and (24) constraint can be written as
0.36β1
ψp tb,min
g22 = − 1, (25) g28 = −1 (31)
0.36β1 tb
where ψ c and ψ p are the prestressing steel reinforce- where tb,min is the minimum bottom flange thickness
ment indices for both the composite and the precast that is sufficiently thick to place prestressing steels.
Optimum topology and shape design of prestressed concrete bridge girders using a genetic algorithm
3 Optimum design with a genetic algorithm where n is the total number of constraints, and pi is the
violation factor of the ith constraint, calculated as
A GA is an optimization method from the artificial
intelligence family that uses the principles of natural pi = gi for gi > 0
selection. According to Darwin’s natural selection the-
pi = 0 for gi ≤ 0. (33)
ory, strong individuals of a population survive, while
weak individuals do not. The surviving individuals mate Given this penalty function, the penalized objective
to produce the next generation of individuals. It is function can be written
clear that individuals in the next generation will be
better than the individuals of the previous generations; (x) = C(x) · [1 + K · P] , (34)
individuals continue to get fitter as evolution continues.
Thus, individuals ultimately become perfect (Holland where K is a coefficient that depends on the type of the
1975; Goldberg 1989). problem. K determines the activity of the penalty func-
A GA is an evolutionary optimization method that tion when calculating the penalized objective function.
uses a group of initial solutions. These solutions are It is important to determine a suitable value of K to
represented by the design variables. The GA improves reach the best solution in a short time.
these solutions by using certain genetic rules. The initial
population is randomly determined. This generation 3.2 Coding of solutions and determining the initial
of solutions is taken to be the first generation of the population
design, and each individual is considered to be a solu-
tion within a given generation. The solutions for each There are several coding strategies for GAs, including
generation are coded using binary numbers in order binary, integer and real-valued representations. Binary
that genetic operators can be readily applied. coding is chosen in this study, since the purpose of
After determining the initial population, the genetic this study is to optimize the topology and the shape of
evolution begins. At first, the fitness of each individ- the prestressed concrete bridge girders, not to explore
ual is determined according to their closeness to the different coding types.
optimum solution, and individuals are categorized ac- The GA uses a set of design variables to generate an
cording to their fitness. After that, the reproduction, optimum design. Thus, the probable value set for each
the crossover and the mutation operators are applied design variable must be chosen before commencing.
to each generation to constitute a new generation. This This set is termed the design variable values set.
evolution continues until the majority of the solutions In the case of binary-coded GA, the solutions are
are identical (Ayvaz and Aydın 2000; Daloğlu and represented as strings that represent the values of de-
Aydın 1999), i.e. the convergence has occurred. sign variables. They are obtained by coding the row
GAs are widely used in engineering—for examples, numbers of the design variables into a binary represen-
see Rajaev and Krishnamoorthy (1992, 1998), Jenkins tation. Thus, the solution strings are constituted by the
(1992), Sarma and Adeli (2000), and Kameshki and numbers “0” and “1”.
Saka (2001). For example, suppose there are sixteen different
entries in a design variable values set for a given de-
3.1 Penalty function and penalized objective function sign variable. Then “0000” indicates the first possible
value, “0001” indicates the second possible value, and
GAs are suitable only for unconstrained optimum de- following this pattern, “1111” indicates the sixteenth
sign problems. Therefore, the objective function must possible value. All the design variables are coded in
be expressed as an unconstrained function. For this this manner and the solution strings are generated by
purpose, a penalty function is calculated consistent with joining these short strings. In this context, “0” and “1”
constraint violations. The unconstrained penalized ob- are termed genes, the small strings are chromosomes
jective function can be defined by adding this penalty and the solution strings are individuals who belong to a
function to the objective function. The penalty function certain generation.
is the fully normalized value of the violated constraints. The number of solution strings of a given generation
Thus, the penalty function can be expressed as depends on the length of the strings. If the string lengths
increase, more strings are required to increase the effi-
n ciency of the genetic operators. The solutions from the
P= pi , (32) initial population are determined by randomly coding a
i=1 certain number of “0”s and “1”s.
Z. Aydın, Y. Ayvaz
101101100011 101101101000 1 0 1 1 0 1 1 0 0 0 1 1
Parent strings
0 1 1 0 1 0 1 0 1 0 0 0
011010101000 011010100011
0 0 1 1 0 0 1 0 1 0 0 1
New strings
Crossover position ( 7 ) 1 1 1 0 1 1 1 0 0 0 1 0
Table 1 Hybrid mutation String Generation before the Fitness Mutation Generation after the
operation
no mutation factor characters mutation
1 10100011100011100111 1.10 7–12 10100011100011100111
2 10100001110101001110 0.74 10100001110101001110
3 01010000111001110111 0.48 01010001111001110111
4 11000100100001111101 1.12 11000100100001111101
5 00110001010001110101 1.49 00110001010001110101
6 10101100101010011000 1.14 3–8 10101100101010011000
7 00111011111010111011 0.96 00111011111010111011
8 01101111011101110111 0.85 01101111011101110111
9 10011100000011010010 1.11 10011100000011010010
Mutation rate: 1% 10 00011100011111011011 1.01 00011100011111011011
mutation, the controlled mutation and the hybrid oped program. These examples have been previously
mutation—are used. designed by Emay International Eng. Cons. Co. (Emay
For the standard mutation, a fixed mutation rate is 1998, 2004) and by Haytek International Eng. Cons. Co.
determined at the beginning of the design. The mu- (Haytek 2004) and have been constructed in Turkey.
tation characters from each generation are randomly The discrete design variables are used in our software.
selected at this rate. The values of these characters are If continuous valued design variables had been selected
changed when the mutation operator is applied. instead, it would have been impossible to calculate the
In a controlled mutation, the mutation rate is re- dimensions of the beam cross-section as integer values.
duced stepwise as the generated solution becomes fit- In the industry, the dimensions of the beam cross-
ter. In this study, the mutation rate is reduced by 50% section are always integers—and often multiples of 5
when one-third of the solution strings in a generation or 10 cm. Also, the concrete strength is not chosen as a
are the same. The mutation stops when two-thirds of design variable in this study, because in reality concrete
the solution strings in a generation are the same. strength is normally pre-selected before the production
In a hybrid mutation, a fixed mutation rate is used, of prestressed concrete elements can begin.
as with the standard mutation. However, in the hybrid The thickness of the slab for all of the examples
mutation, instead of changing the value of the mutation given below is constant since the chosen bridge girders
character directly, it is made to match the character include pretensioned and prestressed concrete in adja-
expressed by the fittest string in the generation. In other cent, simply supported I section beams.
words, if the mutation character has the same value as
the fittest solution, then the value of the character is 4.1 Example 1
unchanged. Otherwise, it is changed. An example of a
hybrid mutation is given in Table 1. This example was optimised several times with differ-
ent optimal design parameters to demonstrate the capa-
bilities of our software. The design parameters used for
4 Numerical examples the first attempt are listed below. The design variable
set for this example is listed in Table 2.
A software package has been developed to perform
the optimization as explained above. Three examples Length of span (L) 32.00 m
are solved to demonstrate the efficiency of the devel- Width of deck (W) 10.00 m
Table 4 Values of design variables and cost function for the real life project and for the solution of this study
Design nb hb (cm) bb (cm) bw (cm) tt (cm) tb (cm) pt (cm) pb (cm) nps C
Real life project 10 125 75 20 10.7 16 7.3 8 25 158,215
This study 8 150 70 15 8 18 5 5 22 117,560
Optimum topology and shape design of prestressed concrete bridge girders using a genetic algorithm
As can be seen from Table 4, the solution obtained of the prestressed beam is 75 cm. The design variable
in this study is 25% more economical than the one of value sets for this example are listed in Table 5.
the real life project. As can be seen from Fig. 5, the
number of beams at the bridge superstructure cross- Length of span (L) 14.725 m
section obtained in this study is less than the one for Width of deck (W) 9.00 m
the real life project. The variation of the penalized ob- Width of motorway 7.00 m
jective function average value according to the iteration Distance between top flanges of girders (F) 7.5 cm
number for the first design is illustrated in Fig. 6. Unit weight of coating 2.4 t/m3
1400000 with the ones of the real life project. The cross-section
of the bridge superstructure and of the prestressed
1200000
beam are given in Fig. 7 for the real life project and
1000000 for the solution obtained in this study.
As can be seen from Table 6, the solution obtained
800000
in this study is 28% more economical than the one of
600000 the real life project. As confirmed by Fig. 7, the number
400000 of beams at the bridge superstructure cross-section in
this study is fewer than the one for the real life project.
200000
0 4.3 Example 3
100
120
140
160
180
200
220
240
260
280
20
40
60
80
0
iteration number
In this example, the chosen bridge superstructure is
Fig. 6 Variation of the penalized objective function value for the optimized twice. First, the number of prestressed beams
first design at the bridge superstructure cross-section is maintained
Z. Aydın, Y. Ayvaz
Table 6 Values of design variables and cost function for the real life project and for the solution of this study
Design nb hb (cm) bb (cm) bw (cm) tt (cm) tb (cm) pt (cm) pb (cm) nps C
Real life project 11 75 75 20 10 15 12.5 7.5 8 46,028
This study 8 75 65 15 10 15 5 5 10 33,080
Table 8 Values of design variables and cost functions for the real life project and for the solutions obtained in this study
Design nb hb (cm) bb (cm) bw (cm) tt (cm) tb (cm) pt (cm) pb (cm) nps C
Real life project 11 120 75 20 10 18 8 10 24 101,781
This study (number of beams is fixed) 11 125 40 15 10 35 5 5 21 82,907
This study (number of beams 7 140 65 18 10 25 5 5 29 75,941
is variable)
Optimum topology and shape design of prestressed concrete bridge girders using a genetic algorithm
equal to the one of the real life project in order to in- tion for the optimum solution are listed in Table 8 by
vestigate the effects of the other design variables on the comparing our results with the ones of the real life
optimal design. Secondly, the number of prestressed project. The cross-sections of the bridge superstructure
beams at the bridge superstructure cross-section is and the prestressed beam are given in Fig. 8 for the
taken as the design variable. The design parameters real life project and for both solutions obtained in this
that differ from those of the first example are given be- study.
low. The other design parameters used are the same as As can be seen from Table 8, the solution reached
the ones of the first example. The design variable values by the GA is 19% more economical than the one of
for the eight design variables are listed in Table 7. the real life project if the number of prestressed beams
at the bridge superstructure cross-section is kept to
Length of span (L) 27.70 m be equal to the one of the real life project. As can
Width of deck (W) 12.00 m also be seen from Table 8, the solution reached by
Width of motorway 9.50 m the GA is 25% more economical than the one of the
Number of traffic lanes (N L ) 3 real life project if the number of prestressed beams at
Thickness of slab 0.21 m the bridge superstructure cross-section is taken as the
Distance between top flanges of girders (F) 4.5 cm design variable. This result suggests that it is important
Distance between slab and girder 1 cm to consider the number of beams a design variable,
Unit weight of coating 2.0 t/m3 in addition to the cross-sectional shape parameters, in
order to reach a more economical solution.
The optimal design parameters for this example are:
As can be seen from Fig. 8b, the number of beams at
Number of solutions in a single 30 the bridge superstructure cross-section obtained in this
generation study is fewer than the one of the real life project if the
Length of solution strings 28 number of prestressed beams at the bridge superstruc-
Type of crossover operator Uniform ture cross-section is taken as the design variable.
Type of mutation operator Controlled
Rate of mutation 0.4%
Convergence criterion 60%
Maximum number of iterations 1,000 5 Conclusions
Violation factor 1.5.
The aim of this study was to optimize the cost of a
The required convergence is obtained by the 356th prestressed concrete I-beam bridge superstructure by
iteration. The design variable values and the cost func- using a genetic algorithm. It is concluded that our
Z. Aydın, Y. Ayvaz
GA-driven software can be used to efficiently optimize Emay International Eng.Cons.Co. (2004) Edirnekapı kavþak
the shape and the topology of prestressed concrete köprüleri Köprü-4 uygulama projesi hesapları (in Turkish).
İstanbul, Turkey
bridge girders. In addition, the following conclusions Goldberg DE (1989) Genetic algorithms in search, optimization
can be drawn from this study: and machine learning. Addison-Wesley, New York
Haytek International Eng. Cons. Co. (2004) O-1 Otoyolu
• KM:17 + 088.000 üst geçit köprüsü üst yapı uygulama projesi
The solution obtained by the proposed GA is up to
hesapları, İstanbul, Turkey
28% more economical than the real life project. Holland JH (1975) Adaptation in natural and artificial systems.
• In this study, various crossover and mutation types University of Michigan Press, Ann Arbor
are used. The best solutions are obtained by using a Jenkins WM (1992) Plane frame optimum design environ-
ment based on genetic algorithm. J Struct Eng-ASCE
uniform crossover and a controlled mutation.
118(11):3103–3112
• In addition to the beam cross-section, it is impor- Jones HL (1985) Minimum cost prestressed concrete beam de-
tant to include the number of prestressed beams sign. J Struct Eng-ASCE 111(11):2464–2478
at the bridge superstructure cross-section as a de- Kameshki ES, Saka MP (2001) Optimal design of nonlinear steel
frames with semi-rigid connections using a genetic algo-
sign variable in order to obtain a more economical
rithm. Comput Struct 79:1593–1604
solution. Lewinski T, Rozvany GIN (2007a) Exact analytical solutions for
• Constraints can easily be adapted to the problem some popular benchmark problems in topology optimization
using a GA penalty function and design variable II: three-sided polygonal supports. Struct Multidiscipl Optim
33:337–349
value sets.
Lewinski T, Rozvany GIN (2007b) Exact analytical solutions for
• It is difficult to solve topological and shape opti- some popular benchmark problems in topology optimization
mization problems at the same time by using clas- III: L-shaped domains. Struct Multidiscipl Optim 35:165–174
sical optimization methods, but they can be easily Lounis Z, Cohn MZ (1993) Optimization of precast prestressed
concrete bridge girder systems. PCI J 38(4):60–78
solved with a GA.
Lounis Z, Mirza MS, Cohn MZ (1997) Segmental and conven-
• The solutions reached by the GA are realistic and tional precast prestressed concrete I-bridge girders. J Bridge
constructible without modification because of the Eng 2(3):73–82
use of discrete design variables and a probable Naaman AE (1982) Prestressed concrete analysis and design fun-
damentals. McGraw-Hill, New York
design variable value sets.
Rajaev S, Krishnamoorthy CS (1992) Discrete optimization of
structures using genetic algorithms. J Struct Eng-ASCE
118(5):1233–1250
Rajaev S, Krishnamoorthy CS (1998) Genetic algorithm-based
methodology for design of reinforced concrete frames.
Comput-Aided Civil Infrastruct Eng 13:63–74
References Rozvany GIN (2008) A critical review of established methods of
structural topology optimization. Struct Multidiscipl Optim.
AASHTO (2002) Standard specifications for highway bridges. doi:10.1007/s00158-007-0217-0
American Association of State Highway and Transportation Rozvany GIN, Querin OM, Lógó J, Pomezanski V (2006) Exact
Officials, Washington, DC analytical theory of topology optimization with some pre-
Ayvaz Y, Aydın Z (2000) Optimum design of trusses using a ge- existing members or elements. Struct Multidiscipl Optim
netic algorithm. In: Proceedings of the second international 31:373–377
conference on engineering using metaphors from nature. Sarma KC, Adeli H (2000) Fuzzy discrete multicriteria cost
Leuven, Belgium, pp 159–168 optimization of steel structures. J Struct Eng-ASCE
Chon MZ, Lounis Z (1994) Optimal design of structural concrete 126(11):1339–1347
bridge systems. J Struct Eng-ASCE 120(9):2653–2674 Sirca GF, Adeli H (2005) Cost optimization of prestressed con-
Daloğlu A, Aydın Z (1999) Kafes sistemlerin uygulamaya yöne- crete bridges. J Struct Eng-ASCE 131(3):380–388
lik optimum tasarımı (in Turkish). Mühend Bilim Der 5(1): Totres GGB, Brotchie JF, Cornell CA (1966) A program for
951–957 the optimum design of prestressed concrete highway bridges.
Emay International Eng.Cons.Co. (1998) Uzunçayır Barajı PCI J 11(3):63–71
Varyantı yolu Dinar Köprüsü uygulama projesi hesapları Yu CH, Das GNC, Paul H (1986) Optimization of prestressed
(in Turkish). İstanbul, Turkey concrete bridge girders. Eng Optim 10(1):13–24