Professional Documents
Culture Documents
xv
) (MATLAB
) (EXCEL
) (GRG
) (GA
) (GRG.
CONTENTS
NOTATION
vii
LIST OF FIGURES
LIST OF TABLES
xii
CHAPTER 1: INTRODUCTION
1.1
1.2
1.3
1.4
1.5
Motivation
Design optimization of RC structures
Aims and objectives
Methodology
Thesis outline
1
1
2
2
2
4
6
7
7
8
Introduction
Optimization of individual RC beams and columns
Optimization of RC structures with several elements
Conclusions
10
10
13
15
Contents
iv
16
16
16
17
17
17
17
18
18
19
19
19
20
20
20
20
21
22
22
22
23
24
24
24
24
26
26
Introduction
Fixed parameters
Design variables
Objective function
Design constraints
Optimization of a simple RC beam
5.6.1 Fixed parameters
5.6.2 Design variables
5.6.3 Variables bounds
5.6.4 Design constraints
5.6.5 Objective function
5.7 Optimization model for simple PC beams
28
28
28
29
29
29
30
30
30
31
32
33
Contents
5.7.1
5.7.2
5.7.3
5.7.4
5.7.5
Fixed parameters
Design variables
Variables bounds
Design constraints
Objective function
34
34
34
35
37
39
39
40
41
41
41
42
42
42
44
44
44
44
45
45
45
45
45
45
46
47
47
47
49
50
50
53
55
57
60
60
Contents
vi
62
65
68
70
72
73
73
75
REFERENCES
76
Appendix A: Sample input data files for the developed GA computer models
81
Appendix B: Sample output data files for the developed GA computer models
85
Appendix C: Sample input data files for the developed spreadsheet classical
optimization models
89
Notation
All abbreviations and notations have been defined where they first used. A summary of
those is given below:
ACI
Ac
A ps
As'
As
Av
Cc
Cps
Cs
d'
db
dp
ds
dt
Diameter of tendons
EA
Evolutionary Algorithm
Ec
fbe
fbi
fc
f c'
fci
f c 'i
Notation
viii
fps
f py
f pu
ft
fte
fti
ftmi
fy
GA
Genetic Algorithm
Ic
Icr
Iel
Ied
Ma
M cr
Md
Mn
Ms
nt
Number of tendons
nb
PC
Prestressed Concrete
RC
Reinforced Concrete
Sc
Concrete cover
Vc
Vs
Vu
Wu
Wstr
Weight of stirrups
Wd
Notation
ix
g j (X )
hk ( X )
(X )
List of Figures
Figure 2.1.
Figure 4.1.
Figure 4.2.
Figure 4.3.
Figure 4.4.
Figure 4.5.
Figure 5.1.
Figure 5.2.
Figure 5.3.
Figure 5.4.
Figure 5.5.
Figure 6.1.
Figure 6.2.
Figure 6.3.
Figure 6.4.
Figure 6.5.
Figure 7.1.
Figure 7.2.
Figure 7.3.
Figure 7.4.
Figure 7.5.
Figure 7.6.
Figure 7.7.
Figure 7.8.
Figure 7.9.
Figure 7.10.
Figure 7.11.
Figure 7.12.
Figure 7.13.
Figure 7.14.
List of Figures
xi
Figure 7.15.
Figure 7.16.
Figure 7.17.
Figure 7.18.
Figure 7.19.
Figure 7.20.
Figure 7.21.
Figure 7.22.
Figure 7.23.
Figure 7.24.
Figure 7.25.
Figure 7.26.
List of Tables
Table 2.1.
Table 4.1.
Table 4.2.
Table 7.1.
Table 7.2.
Table 7.3.
Table 7.4.
Table 7.5.
Table 7.6.
Table 7.7.
Table 7.8.
Table 7.9.
Table 7.10.
Table 7.11.
CHAPTER 1
Introduction
1.1
Motivation
Optimum design of structures has been the topic of many studies in the field of
structural design. A designers goal is to develop an optimal solution for the structural
design under consideration. An optimal solution normally implies the most economic
structure without impairing the functional purposes the structure is supposed to serve,
(Rafiq, 1995). The total cost of the concrete structure is the sum of the costs of its
constituent materials; these constituent materials are at least: concrete, steel and
framework, (Sarma and Adeli, 1998).
A study on the design optimization of structural concrete simple beams according
to ACI 318-05 Code and using the Genetic Algorithm optimization technique is to be
carried out in this thesis.
As there are an infinite number of possible beam dimensions, reinforcement ratios
and prestressing forces that yield the same moment of resistance, it becomes difficult to
achieve the least-cost design by conventional iterative methods, (Barakat, 2004). It was
shown that even for a simple and well-defined RC structure of a small garage, the
designs proposed by experienced design engineers can be very different (Brge and
Schneider, 1994). In such a situation an optimization procedure can help designers to
find the best design or at least, a good design among different possible designs.
1.2
Introduction
for a RC structure can be larger than that for a steel structure. Also cracking and
durability requirements are two characteristic properties of RC structures; these increase
the number of design constraints of the optimization problem of RC structures. (Sahab,
2002).
1.3
The goal of this study is to develop design optimization models for structural
concrete beams using genetic algorithms as an optimization technique. The following
objectives are defined to achieve the research goal:
Develop computer models to automate the design process of reinforced and
prestressed concrete beams according to ACI 318-05 Code.
Develop genetic algorithm (GA) optimization models using MATLAB software.
Develop spreadsheets for the design optimization of reinforced and prestressed
concrete simple beams using Generalized Reduced Gradient (GRG) method as an
example of the classical optimization methods.
1.4
Methodology
Carry out a survey on the previous relevant researches.
Join the GA optimization models with the developed computer design models to
perform the design optimization process.
Validate the developed optimization models using two numerical examples, one for
reinforced concrete simple beams and the other for prestressed concrete simple
beams.
Perform a tuning process of the GA operators using the identified numerical
examples.
Perform a study on the influence of the fixed parameters of the numerical examples
on the cost functions.
Compare the design results obtained from the GA optimization process with those
obtained from the GRG method.
1.5
Thesis outline
The thesis includes 8 chapters and three appendices. A brief description of the
chapters contents is presented below:
Chapter 1 highlights the need for research in the field of design optimization of
concrete structures. The characteristic features of design optimization of structures are
studied. The aims and objectives of the project are described. At the end, the structure
of the thesis is presented.
Chapter 2 includes a review of optimization problems and techniques, including
classical and heuristic search and optimization techniques.
Chapter 3 includes a review of optimization researches done in this area,
structural optimization generally, and design optimization of reinforced concrete beams
specifically.
Introduction
CHAPTER 2
Optimization Techniques
2.1
Optimization problems
xm
x mu
fi ( X )
i =1, 2, . . ., no (2.1)
j= 1, 2, . . ., ng
(2.2)
k = 1, 2, . . ., ne
(2.3)
m = 1, 2, . . ., ns
(2.4)
where X is the vector of n design variables, fi(X) is an objective or merit function, gj(X)
and hk(X) are the inequality and the equality constraints, respectively. These constraints
represent limitations on the behaviour or performance of the system. Therefore, they are
called behavioural or functional constraints. Side constraints (2.4) restrict the acceptable
range of potential solutions of the problem based on non-behavioural constraints. In this
expression xml and xmu are the lower and upper limits on the design variable x m ,
respectively. In the above expressions no, ng, ne and ns are the number of objective
functions, number of inequality, equality and side constraints, respectively. Depending
on the specific choice of design variables, objective functions, and constraints, various
types of optimization problems may exist. Table 2.1 presents a classification of
optimization problem which has been collected from (Foulds, 1981), (Rao, 1984),
(Arora, 1989), (Haftka and Gurdal, 1992), (Kirsch, 1993), (Sarma and Adeli, 1998),
(Vanderplaats, 1999), (Sarma and Adeli, 2000).
oujy
Optimization Techniques
Number of
objective
functions
Category
Specifications
Single variable
Multi-variable
Single objective
Multi-objective
Unconstrained
Constrained
Linear
programming
(LP)
Non-linear
programming
(NLP)
Static
Presence of
constraints
Features of
constraints and
objective
functions
Nature of
design
variables
Type of design
variables
Dynamic
Discrete
Continuous
Mixed
Nature of
design
variables and
design input
data
Deterministic
Probabilistic
functions
of
other
Optimization Techniques
2.2
Traditional search and optimization methods can be classified into two distinct
groups: Direct and gradient-based methods (Deb, 1995; Reklaitis et al., 1983). In direct
methods, only objective function and constraints are used to guide the search strategy,
whereas gradient-based methods use the first and/or second-order derivatives of the
objective function and/or constraints to guide the search process. Since derivative information is not used, the direct search methods are usually slow, requiring many
function evaluations for convergence. For the same reason, they can be applied to many
problems without a major change of the algorithm. On the other hand, gradient-based
methods quickly converge to an optimal solution, but are not efficient in nondifferentiable or discontinuous problems. In addition, there are some common
difficulties with most of the traditional direct and gradient-based techniques:
Optimization Techniques
evaluated. These difficulties can be eliminated if only feasible values of the variables
are allowed during the optimization process.
The above discussion suggests that traditional methods are not good candidates
for an efficient search and optimization algorithm. In the following sections, we
describe the genetic algorithm which works according to principles of natural genetics
and evolution, and which has been demonstrated to solve various search and
optimization problems.
2.3
In the last three decades, heuristic methods have been rapidly developed to solve
optimization problems. These methods are principally intuitive and do not have
theoretical support. Heuristic methods such as genetic algorithms (GAs), simulated
annealing (SA) and tabu search (TS) provide general ways to search for a good but not
necessarily the best solution.
Manoharan and Shanmuganathan (1999) applied four different search
mechanisms of TS, SA, GA and branch and bound technique to the design optimization
Optimization Techniques
of three different truss problems and compared them. They concluded that all three
heuristic search methods, TS, SA, GA, work well and produce an acceptable solution
within a reasonable amount of time. Amongst these three methods, GA has a feature
that it does not need an initial guess to search for the optimum. Branch and bound
technique consumes an enormous amount of computing time to find optimum solution.
Also, it is not viable for structures of any reasonable size (e.g. a truss structure with
more than 6 bars).
2.3.1
Genetic algorithms
Genetic algorithms were introduced by Holland in the 1960s. With the aid of his
colleagues and students, he developed these algorithms during the 1970s in University
of Michigan. He summarized the results of these researches in the book Adaptation in
natural and artificial systems (Holland, 1975). GAs are numerical optimization
techniques inspired by the natural evolution laws. A GA starts searching design space
with a population of designs, which are initially created over the design space at
random. In the basic GA, every individual of population (design) is described by a
binary string (encoded form). GA uses four main operators, namely, selection, creation
of the mating pool, crossover and mutation to direct the population of designs towards
the optimum design. In the selection process, some designs of a population are selected
by randomized methods for GA operations, for example in creation of the mating pool,
some good designs in the population are selected and copied to form a mating pool. The
better (fitter) designs have a greater chance to be selected. Crossover allows the
characteristics of the designs to be altered. In this process different digits of binary
strings of each parent are transferred to their children (new designs produced by the
crossover operation). Mutation is an occasional random change of the value of some
randomly selected design variables. The mutation operation changes each bit of string
from 0 to 1 or vice versa in a designs binary code depending on the mutation
probability. Mutation can be considered as a factor preventing from premature
convergence.
A GA uses a discrete set of design variables in the optimization process.
However, by defining the number of decimal digits for representation of continuous
variables or step size between the sequential values of design variables this method can
be applied to continuous problems as well. Lin and Hajela (1992) implemented GAs in
the optimal design of structural systems with a mix of continuous, integer and discrete
design variables.
Optimization Techniques
Begi
Initialize
gen =
Evaluatio
Assign
Cond?
No
gen = gen + 1
ye
Reproductio
Stop
Crossove
Mutatio
Figure (2.1) Flowchart of working principle of a Genetic Algorithm
CHAPTER 3
Optimization of RC structures
3.1
Introduction
3.2
Optimization of RC Structures
11
Optimization of RC Structures
12
applied to optimize design of a RC beam with fixed support at one end and
simple support at the other end with variable depth and width along its length.
Al-Salloum and Siddiqi (1994) presented minimum cost design of singly
reinforced rectangular concrete beams based on the American Code (ACI 31889) provisions. The cost function contains the cost of concrete, bending
reinforcement and formwork of the unit length of the beam. They presented a
closed-form solution for optimal depth and steel ratio of a beam with given
width and design moment in terms of material costs and strength parameters.
Zielinski et al. (1995) investigated optimum design of RC short-tied rectangular
columns under biaxial bending based on the Canadian Code (CAN3-A23.3M84). The cost components of the objective function are the cost of concrete,
longitudinal reinforcement and formwork. Design variables are the crosssectional dimensions of the column and the area of tension and compression
bars or, alternatively, the number of reinforcement along the width and depth of
a column cross-section. Then the problem is solved by the Powells method.
Al-Gahtani, (1995). An effective formulation for optimum design of two-span
continuous partially prestressed concrete beams is described. Variable
prestressing forces along the tendon profile are considered. The imposed
constraints are on flexural stresses, ultimate flexural strength, cracking moment,
ultimate shear strength, reinforcement limits cross-section dimensions, and
cable profile geometries, the capabilities of the program to solve several
engineering problems are presented.
Barakat et al. (2003). A general approach to the single objective reliabilitybased optimum (SORBO) design of prestressed concrete beams (PCB) is
presented in this paper. Several limit states are considered, including permissible
tensile and compressive stresses at both initial and final stages, prestressing
losses, ultimate shear strength, ultimate flexural strength, cracking moment,
crack width, and the immediate deflection and the final long term deflection.
The results consist of, but not limited to, the initial and final prestressing forces,
prestressing losses, immediate and final deflections, and upper and lower bounds
on the parabolic tendon profile.
Barakat et al. (2004). A general approach to the multiobjective reliability-based
optimum (MORBO) designs of prestressed concrete beams (PCB) is presented
in this paper. The competing objectives in the multiobjective optimization of
PCB are selected from, minimization of the overall cost of the PCB,
maximization of the system reliability index, maximization of the flexural
strength reliability index, and maximization of the tensile stress reliability index
at service stage.
Kirsch, (2004). A two-level design procedure for prestressed concrete structures
is developed, where the prestressing force and the tendon coordinates are
Optimization of RC Structures
13
optimized at the first-level, while the concrete dimensions are selected at the
second-level.
Shopova et al. (2006). This paper introduces in details a genetic algorithm-called
BASIC, which is designed to take advantage of well known genetic schemes so
as to be able to deal with numerous optimization problems. A range of various
optimization problems has been solved to test its capability; the solutions
obtained are commensurable with other genetic algorithms and solution
techniques.
3.3
Optimization of RC Structures
14
Optimization of RC Structures
15
structural element covers that of material and labour for reinforcement, concrete
and formwork. Cost optimisation for three reinforced concrete flat slab
buildings is illustrated and the results of the optimum and conventional design
procedures are compared.
Govindaraj and Ramasamy (2005) presented the application of Genetic
Algorithms for the optimum detailed design of reinforced concrete continuous
beams based on Indian Standard specifications. The produced optimum design
satisfies the strength, serviceability, ductility, durability and other constraints
related to good design and detailing practice. The optimum design results are
compared with those in the available literature. An example problem is
illustrated and the results are presented. It is concluded that the proposed
optimum design model yields rational, reliable, economical and practical
designs.
3.4
Conclusions
Some of the major advances during the history of structural optimization are reviewed
in chapter 3.
The main conclusions of the literature survey can be drawn as follows:
The optimality criteria method has been applied to various field of structural
optimization. This method is more efficient for design optimization of largescale problems when the number of constraints is small compared to the
number of design variables. In this method, the optimization problem is
treated as a continuous problem.
The use of modern heuristic optimization techniques is rapidly increasing.
They have become increasingly popular for solving optimization problems in
three recent decades. These techniques have been made possible optimization
of a large-scale structure according to practical design codes. These
techniques can be used for solving continuous or discrete optimization
problems.
In RC structures, the cost of concrete, reinforcement and formwork can be
involved in the total cost of the structure. Therefore, the minimum weight for
these structures is not necessarily equivalent to the minimum cost, or in other
words, the total cost of the structure is not generally proportional to its weight.
In most structural design optimization problems, design variables are
essentially discrete. For example, the area of a reinforcing bar can take some
discrete values from the catalogue. However, in many publications in the field
of structural optimization the problem has been treated as a continuous
problem.
CHAPTER 4
4.1
The chief task of the structural engineer is the design of structures. Design is the
determination of the general shape and all specific dimensions of a particular structure
so that it will perform the function for which it is created and will safely withstand the
influences that will act on it through its useful life. These influences are primarily the
loads and other forces to which it will be subjected (Nilson et al., 2004). Beams are
structural members carrying transverse loads that can cause bending moments, shear
forces and in some cases torsion.
Design of a beam starts with proportioning its sections to resist bending moment
and choosing the required reinforcement. Once this is done, the chosen sections are
checked and designed for shear and torsion. In order to limit deflections, the depth of
the cross section is chosen to fulfil the ACI Code serviceability requirements.
4.1.1
The main stages of design of the superstructure of Reinforced Concrete beams can
be summarised as follows:
1.
2.
3.
4.
5.
6.
7.
8.
9.
Pre-assigning of parameters
Loading and structural analysis.
Determining the design bending moments.
Calculation of bending reinforcement.
Detailing of bending reinforcement.
Determining the design shear force.
Calculation of shear reinforcement
Detailing of shear reinforcement
Checking of deflections.
4.1.2
Pre-assigning of parameters
At the first stage of a structural design of beams, the designer initially assumes the
general shape of the beam, size of beam and material properties based on his previous
experiences and some rules of thumb. In this manner the self-weight of the beam can be
calculated and the model of beam can be provided for the structural analysis. In an
optimization program the process of design optimization is started from an initial
17
feasible design or an initial population of designs. During the optimization process the
initial design or designs are improved in such a way that they become compatible with
the constraints and minimize the objective function. Those specifications of the
structure such as the characteristic strength of the materials, which can be considered to
be fixed during the design optimization process, are called pre-assigned parameters. The
Code provisions of ACI 318-05 have to be considered in pre-assigning of parameters or
determining the minimum dimensions of the beam.
4.1.3
In this research it is assumed that the load which the reinforced concrete beam
will carry is only of the distributed load type, the loads are the self weight of the beam,
the dead load and the live load.
4.1.4
Wu L2
( KN .m)
8
(4.1)
Where:
Wu =factored total distributed load = 1.2 (Wd
L = Span of the beam
4.1.5
Wsd ) 1.6 Wl
Mn
4.1.6
As * f y * d
As * f y
1.7 * f c' * b
(KN.m)
(4.2)
The ACI Code defines a tension controlled member as one with a net tensile strain
greater than or equal to 0.005. The corresponding strength reduction factor is
0.9 ,
The Code additionally defines a compression controlled member as having a net tensile
strain of less than 0.002 the strength reduction factor for compression controlled
members is 0.65. Between net tensile strains of 0.002 and 0.005, the strength reduction
factor varies linearly, and the ACI Code allows a linear interpolation of based on t as
shown in figure (4.1):
18
4.1.7
As per paragraph 10.5.1 of the ACI 318-05 code, the minimum amount of flexural
reinforcement is given by the following relationship:
As, min
f c'
0.25
fy
bw d
(cm 2 )
1.4 bw d
fy
(4.3)
(4.4)
And according to paragraph 10.3.5 of the Code, the maximum reinforcement ration is:
max
4.1.8
0.85
f c'
fy
0.003
0.003 0.004
(4.5)
d b cm
S
2.5 cm
4
* max agg. size
3
(cm)
300
252
(cm)
0.6 f y
(4.6)
4.1.9
19
The factored shear force is calculated at a distance of d from the face of the
support as per paragraph 11.1.3.1 of the ACI 318-05 building Code, assuming that the
width of the support is z m, the factored shear force is calculated using the following
relationship:
Vu
Wu ( L
z /2 d)
2
( KN )
(4.7)
Where:
Wu = Factored total distributed load
L = Span of the beam
d = Depth of the beam
z = Width of the support
Av
Vc
fy d
S
( KN )
(4.9)
0.17
f c'
b d
0.29
f c' b d
(4.10)
20
Av , min
f c'
0.062
b s
fy
0.35
b s
fy
(cm2 )
(4.11)
Where:
s = Longitudinal spacing of web reinforcement
f y = Yield strength of web steel
Av = Total cross sectional area of web steel within distance s
A
v
s
yt
d
0.66
f'
c
(KN)
(4.12)
4.1.14 Deflection
The deflection is calculated at two stages, at the first stage, the initial or short term
deflection is calculated and at the second stage the long term deflection is calculated.
i d
5 Wd L4
(mm)
384 Ec Ied
(4.13)
Where:
Wd = service distributed dead load
L = Span of the beam
Ec = Modulus of elasticity of concrete
Ied = Effective moment of inertia under dead load only
The deflection due to live load only is calculated from the following relationship
i d l
5 Ws L4
(mm)
384 Ec Iel
(4.14)
21
Where:
Ws = service distributed total load
L = Span of the beam
Ec = Modulus of elasticity of concrete
Iel = Effective moment of inertia under dead and live load
i l
i d l
i d
i l
where :
i l
(4.15)
2*
i d l
i l
limit by code
(4.16)
The calculated deflections are then compared with the allowable ones as per table 9.5(b)
of the ACI 318-05 Code, which is shown here in table (4.1)
Table (4.1) Maximum permissible computed deflection (ACI 318-05)
Type of member
Deflection to be
considered
Deflection
limitations
180
360
480
240
4.2
22
4.2.1
In this research it is assumed that the load which the prestressed concrete beam will
carry is only of the distributed load type, the loads are the self weight of the beam, the
superimposed dead load and the live load.
4.2.2
Mu
Wu L2
( KN .m)
8
Where:
Wu = Factored total distributed load
L = Span of the beam
(4.17)
4.2.3
23
Calculate
a
Mn
Aps (d p
As f y (d
a
)
2
a
)
2
As' f y (
a
2
d ')
Mn
f c' bd 2p (0.36
0.08
2
1 )
4.2.4
24
Strength reduction factored was presented in section 4.1.6 for bending moment,
figure (4.1) can be used to calculate the value of the strength reduction value for shear
also.
4.2.5
The factored shear force is calculated at a distance of d from the face of the
support as, assuming that the width of the support is z , the factored shear force is
calculated using the following relationship:
Vu
Wu ( L z
2
d)
( KN )
(4.18)
Where:
Wu =factored total distributed load
L = Span of the beam
d = depth of the beam
z = width of the support
4.2.6
The design shear force, shear strength provided by concrete and minimum web
reinforcement are calculated as shown in figure (4.4).
Where:
Vc = Nominal shear strength provided be concrete.
Vu = Factored shear force.
Vs = Nominal shear strength provided be shear reinforcement.
S = Longitudinal spacing of web reinforcement
h = Total depth of the concrete section
4.2.7
Deflection
The deflection is calculated at two stages, at the first stage, the initial or short term
deflection is calculated and at the second stage the long term deflection is calculated.
It consists of two components, the deflection due to dead load and the deflection due to
superimposed dead load and live load, see figure (4.5).
25
START
Vu
No web steel
1
Vc
2
N
Vu
Enlarge section
Vc
Ye
Vs
Vu
8 f c' bd p
Vc
END
Vs
Vu
Vc
Ye
4.2.8
4 f c' bd p
26
START
M cr
Ma
Ic
M cr
Ma
I cr
Ic
4.2.9
The PCI multipliers method provides a multiplier C1 which takes account of long term
effects in prestressed concrete members. The following table provide multipliers of
immediate deflection and camber. Shaikh and Branson (1970), propose that substantial
reduction can be achieved in long term camber by the addition of nonprestressed steel.
In that case, a reduction multiplier C2 can be used given by:
27
C1
C2
1
As
As
Aps
(4.19)
Aps
1.85
1.80
Final
Deflection (downward) component-apply to the elastic deflection due
to the member weight at release of prestress
2.70
2.45
3.00
The calculated deflections are then compared with the allowable ones as per table 9.5(b)
of the ACI 318-05 Code, see table 4.2 above
CHAPTER 5
5.1
Introduction
5.2
Fixed parameters
The span of the beam, the characteristic strength, modulus of elasticity and unit
weight of concrete and reinforcement, the intensity of the dead and live loads are
assumed to be fixed parameters. It is assumed that the total cost of concrete and
reinforcement is proportional to volume and weight of each material, respectively.
Consequently, the total cost of a structure is calculated using fixed parameters of the
cost of unit volume of concrete and unit weight of reinforcement.
5.3
Design variables
29
In the present study the width, depth and the number and size of bars in a given
section of the RC beam are considered as design variables, the width, depth, the number
and size of bars, the number and size of tendons and the eccentricity of the tendons in a
given section of the PC beam are considered as design variables.
The width of the beam, number and size of bars, number of tendons variables are
taken as integer variables. Although width of the beam may theoretically take any real
number, practically it is restricted to a set of discrete values. For example, it can change
from 40 cm to 80 cm with an increment of 5 cm, also the diameter of the bars are
chosen to be changing with an increment of 2.
Shear reinforcement may need to be provided along the span of the beam either it is a
RC beam or a PC one.
5.4
Objective function
5.5
Design constraints
5.6
In this section, the model of the RC beam is described, showing the fixed
parameters, the design variables, the design variables bounds, the design constraints
and the objective function.
A typical simply supported rectangular RC beam has a span of L m and may be carrying
a uniform dead and live loads of WD KN/m and WL KN/m respectively, in addition
to its own weight. It is intended to optimize the design of the beam according to the
provisions of the ACI 318-05 Code. Figures (5.1) and (5.2) below show the geometry of
the simple RC beam.
30
W D = D .L KN /m ',
W L= L.L K N /m '
Lm
Figure (5.1): Typical simple RC beam with distributed loads
A
b
Sc
S
Section A-A
5.6.1
The fixed parameters for this RC simple beam model are taken as the span of the
beam, max. aggregate size, the cost/m3 of concrete, the cost/ton of steel, the modulus of
elasticity of concrete, the compressive strength of concrete, the yield strength of
reinforcement and the value of the distributed dead and live loads.
5.6.2
Design variables
The design variables which are considered in this RC beam model are listed below:
5.6.3
b = Beam width
d = Effective beam depth
nb = Number of flexural bars
d b = Diameter of flexural bars
(integer values)
(real values)
(integer values)
(integer values)
As it is stated earlier, some or all of the variables have bounds, these bounds
results from different issues such as the provisions of the code under consideration, the
aesthetic of the structural elements in the building, the practical issues and the
31
availability of some sizes of the material at the local market, listed below are the bounds
considered at this research project.
d min
b
db
2
hmin
bmin
Sc d s ,
b
d max
hmax
db
2
Sc d s
bmax
(5.1)
(5.2)
where bmin and bmax are chosen according to architectural and practical considerations
db
db min ,
db
db max
(5.3)
where d b min and db max are chosen according to range of reinforcement available at market
nb
nb min ,
nb
nb max
(5.4)
Sc = Concrete cover
ds = diameter of stirrups
5.6.4
Design constraints
The design constraints which are considered in this optimization model are listed below:
Design for flexure
Mu
Mn
(5.5)
S min
(5.6)
db cm
S min
max 2.5 cm
4
* max agg. size
3
ACI - Code
(5.7)
32
S max
S max
(5.8)
min 300 *
280
280
,380 *
0.67 * f y
0.67 * f y
2.5 * 50 (cm)
ACI - Code
(5.9)
max ,
min
Ast
b d
(5.10)
2
3
Vs
fc
b d
(KN)
ACI - Code
(5.11)
Deflection control
Md
i l
2*
Ms
3 M cr ,
3 M cr
(5.12)
i d l
i l
Where:
Md
Ms
M cr
i l
2*
5.6.5
i d l
Objective Function
The chief task of the optimization process is to select the values of variables in a
way that satisfies the provisions of the code regarding safety and serviceability within
the least cost possible, the function below defines the total cost of the RC simple beam
model in terms of the cost of the concrete and reinforcement used.
Cc * ( Ac
As ) * L
33
Wstr
Cs *
* As
Wstr
($)
(5.13)
Where:
Cc = Cost of concrete per cubic meter
Cs = Cost of reinforcement steel per ton
Ac = Area of concrete cross section
As = Area of longitudinal reinforcement
L = Span of the beam
Wstr = Weight of stirrups
= Unit weight of steel reinforcement
WL=L.L KN/m'
Lm
Figure (5.3): Typical simple PC beam with distributed loads
B
B
b
Sc
S
Section B-B
34
(integer values)
(real values)
(integer values)
(integer values)
(integer values)
(real values)
(real values)
dt
2
hmin
bmin
Sc d s ,
b
d p max
hmax
dt
2
Sc d s
bmax
where bmin and bmax are chosen according to Architectural and practical considerations
db
db min
db
d b max
where bmin and bmax are chosen according to range of bars available at local market
dt
d t min
dt
dt max
where d t min and d t max are chosen according to range of tendons available at local market
nb
nb min
nb
35
nb max
(5.18)
nt min
nt
nt max
(5.19)
emax
(5.20)
0.75
= Concrete cover
= diameter of stirrups
= diameter of tendons
= tendon eccentricity
Mu
Mn
(5.21)
Smin
(5.22)
db cm
S min
max 2.5 cm
4
* max agg. size
3
ACI - Code
(5.23)
S max
(5.24)
S max
min 300 *
36
280
280
,380 *
0.67 * f y
0.67 * f y
2.5 * 50 (cm)
ACI - Code
(5.25)
M n,,,,
0.36
(5.26)
f ci , f te
f c , f be
f t , f ti
f tmi
(5.27)
Where:
fci = maximum allowable compressive stress in concrete at initial prestress
fbi = Extreme fiber compressive stress in concrete at initial prestress
ftmi = maximum allowable tensile stress in concrete at initial prestress
fti = Extreme fiber tensile stress in concrete at initial prestress
fte = Extreme fiber compressive stress in concrete at service load
fc = Maximum allowable compressive stress in concrete at service load
fbe= Extreme fiber tensile stress in concrete at service load
ft = Maximum allowable tensile stress in concrete at service load
d
'
(
) = Total reinforcement index
t =
p
dp
1
(5.28)
Vs
fc
fy dp
Vu
0.75 (Vc
(KN)
fc
f y d p ) (KN)
ACI - Code
(5.29)
ACI - Code
(5.30)
Deflection control
Md
i l
2*
Ms
3 M cr ,
3 M cr
(5.31)
i d l
i l
37
Where:
Md
Ms
M cr
i l
2*
i d l
Geometry constraint
e
dt
2
ds
Sc
h
2
(5.32)
Cc * ( Ac
( As
C ps *
* A ps
Aps )) * L
L
Wstr
Cs *
* As
Wstr
( $)
Where:
Cc = Cost of concrete per cubic meter
Cs = Cost of reinforcement steel per ton
Cps = Cost of prestressing tendons per ton
Ac = Area of concrete cross section
As = Area of longitudinal reinforcement
Aps = Area of prestressing tendons
L = Span of the beam
Wstr = Weight of stirrups
= Unit weight of steel reinforcement
Figure (5.5) shows a flowchart for the developed GA optimization models.
(5.33)
38
GA Model
Design the beam for flexure, shear and deflection for each
individual population
Calculate the objective function (cost) for every individual in the
population
Check for constraint violation. Is there any violation?
Ye
Ye
END, Display
results
CHAPTER 6
6.1
Historical background
6.2
Some biological terms are often used in the context of genetic algorithms in the
spirit of analogy with biology. Being familiar with these terms enables better
understanding of the basic GA concepts.
Each cell of a living organism contains the same set of one or more chromosomes.
Chromosomes are strings of DNA (DNA stands for Deoxyribo-Nucleic_Acid, an acid
Collection of chromosomes
forming a cell (genome)
A gene of a chromosome
a
Allele
40
that carries genetic information in a cell); they serve as a model for the organism. A
chromosome can be divided into genes. Any of the alternative forms of a gene that may
occur at a given position of a gene is called an allele. Each gene is located at a particular
position (locus) in the chromosome. The complete collection of chromosomes in the cell
is called the organisms genome. Figure 6.1 illustrates these concepts.
6.3
Evaluation of the
fitness of individuals
Termination
?
Yes
Result
s
No
Choose the best individuals
and creation of a new
population using GA
Figure (6.1) Flowchart of a simple genetic algorithm
some solutions are first selected according to their fitness values to enter a mating pool.
The mating pool is then filled up by cloning the individuals (solutions), which have
been entered in the mating pool, in proportion to their fitness values (creation of the
mating pool). Creation of a new population is implemented by repeating the crossover
and mutation operations. In the crossover stage, two individuals are initially selected as
parents and then some segments of encoded strings of parents are swapped to create two
children. For the mutation, some children are selected randomly, then some alleles of
these children are altered at random. The fitness value of individuals in the new
generation is then evaluated. If the termination conditions are satisfied, the process is
terminated. Otherwise, the iterative process is repeated for a new generation.
41
6.3.1 Representation
In GAs, design variables are usually encoded and represented in a string
form. Two popular systems of coding are binary and real coding schemes. In the real
number representation, each variable is represented as a conventional floating-point
number. Binary and real codings differ mainly in the way of implementation of GA
processs. In this project the basic binary coding method is used.
Binary coding
In this method, design variables are represented by a string of bits, 0 or 1 in a
binary arithmetic system. Figure 6.3 illustrates how n design variables in a particular
problem form a binary string. In this figure, sxn is a binary number in a reserved location
for n-th design variable. Each binary number in a given location of the binary string
shows an encoded integer decimal number. This integer number corresponds to a
specific value of a certain design variable. Therefore, the length of each sub-string sxi
should be sufficient to produce all integer numbers corresponding to a given design
variable xi. If the lower and upper bounds on the i-th design variable are xil and xiu ,
respectively and the increment for the possible values for this design variable in the
specified range is i, the required binary sub-string length to represent this variable can
be computed as follows:
x u xil
li ceiling (log 2 ( i
))
(6.1)
i
where ceiling(x) is a function which delivers the closest integer number to the argument x
11101011001011010111..1010
sx 1
sx2
sx3
sxn
l st
li .
(6.2)
i 1
Every gene with a specific number of bits of the length li generates np possibilities for
the configuration of a gene (schema), where np is:
n p 2 li .
(6.3)
In the process of random creation of the initial population, all possible solutions should
have equal probability of creation.
42
2000) among many others. The use of these additional processes is optional in a
GA. Depending on the problem and the selected representation scheme some of these
processs can be applied. In the present research only the four main processs, i.e.
selection, creation of the mating pool, crossover and mutation have been used.
6.3.2.1
Selection
43
the GA literature, among them one-point, two-point, uniform, cycle and arithmetic
crossover. Below the first two most common methods, which have been used in this
study, are described.
One-point crossover is the simplest type of crossover. In this method, one location
along the bit string of each parent is chosen at random. The parts of two parents after
the crossover site are swapped to create two offspring (Figure (6.4a)). One-point
crossover has some shortcomings (Coley, 1999, Mitchell, 1998). This method
recombines the bit strings in a limited way. Sometimes only a few bits in the parents
strings need to be altered. However, this method exchanges the whole set of bits after
the crossing site. This effect is called positional bias. Such positional bias implies that
long strings are likely to be destroyed under a one-point crossover (Coley, 1999,
Mitchell 1998). The segments exchanged between the parents always include the
endpoints of the strings, which is called endpoint effect. In addition, crossover of
parents string that are identical after the crossing point, has no effect, as the children
will be identical to the parents.
In order to reduce positional bias and remove two other aforementioned problems
of the one-point crossover, many researchers use two-point crossover. In this method,
two crossing sites are selected at random and the portions of strings between them are
exchanged as shown in Figure (6.4b).
Crossover site
Parent 1
Parent 2
Child 1
Child 2
10110100010101110
10010111010101011
10110100010101011
10010111010101110
a) One-point crossover
Two crossover sites
Parent 1
Parent 2
Child 1
Child 2
10110100010101110
10010111010101011
10110111010101110
10010100010101011
b) Two-point crossover
Parent 1
Parent 2
Child 1
Child 2
10110100010101110
10010111010101011
10010110010101010
10110101010101111
C) Uniform crossover
44
As Figure (6.4c) shows in uniform crossover every bit along the parents string is
exchanged at random. Each bit of an offspring in a specific position of its string can
randomly be adopted from each parent. Uniform crossover does not have any of the
aforementioned shortcomings of the one-point crossover. However, Researches have
exclaimed that this method can be highly disruptive especially in early generations
(Coley, 1999, Mitchell 1998).
6.3.2.4
Binary-valued mutation
Before mutation
After mutation
10101011110011110
10101001111011110
6.4
6.4.1
Mutated alleles
The no of iterations identifies the number of loops that the model must run in
order to find the best solution. Using very little number of iteration may allow little
application of the basic operators of the GA on the population, so the model will not be
able to discover wide range of developed populations (solutions) and the model may get
stuck to a suboptimal or infeasible solution. Meanwhile, using very high value of the
iteration may slow down very much the model which is not proffered in most cases.
6.4.2
Population size
The population size defines the number of initial (feasible and infeasible)
solutions which are created at the beginning of the GA model work. These initial
solutions are created taking into account the variables bounds which are stated in the
GA model. Limiting the population size to very little initial solutions will prevent the
development of much better solutions since the interaction between the initial solutions
are limited to a narrow range. Meanwhile increasing the population size enables the
genetic algorithm to search more points and thereby obtain a better result, however, the
larger the population size, the longer the genetic algorithm takes to compute each
generation. So one can experiment with different settings for population size that return
good results without taking a prohibitive amount of time to run.
6.4.3
45
Crossover Type
Crossover enables the algorithm to extract the best genes from different
individuals and recombine them into potentially superior children. Three types of
crossover are found in the literature of the GA, two of them are considered in this study,
namely: single point crossover and two point crossover.
6.4.4
Mutation
Mutation adds to the diversity of a population and thereby increases the likelihood
that the algorithm will generate individuals with better fitness values. Different values
of the mutation rate are considered in the models in order to examine the best mutation
rate that will lead to the optimum solution.
6.4.5
Number of matings
In this stage a percent of the total population size is chosen in order to perform the
crossover and mutation on it. In our models, this percent is chosen randomly and varies
between (0.3 and 0.8), these different values are examined in order to get the best
solution.
6.5
Elitist strategy
The elitist strategy, introduced by DeJong (1975), transfers some number of the
fittest individuals of the current generation according to the assumed percentage of elite,
Pre, to the next generation. Therefore, the best individuals are not lost due to the
crossover and mutation operations and the maximum fitness value in a generation can
not decrease as compared to the previous generation.
6.6
Constraint handling
46
P ( x, R, r )
f ( x)
R j g j ( x)
j 1
rk hk ( x)
(6.4)
k 1
The parameters R j and rk are the penalty parameters for inequality and equality
constraints, respectively.
The success of this simple approach lies in the proper choice of these penalty
parameters. One thumb rule of choosing the penalty parameters is that they must be so
set that all penalty terms are of comparable values with themselves and with the
objective function values. This is intuitive because if the penalty corresponding to a
particular constraint is very large compared to that of other constraints, the search
algorithm emphasizes solutions that do not violate the former constraint.
This way other constraints get neglected and search process gets restricted in a
particular way. In most cases, most search methods prematurely converge to a
suboptimal feasible or infeasible solution. Since a proper choice of penalty parameters
are the key aspect of the working of such a scheme, most researchers experiment with
different values of penalty parameter values and find a set of reasonable values.
Deb, (1998) proposed a technique for handling constraint optimization problems
for Genetic Algorithms. The proposed method belongs to both second and third
categories of constraint handling methods described by Michalewicz and Schoenauer
(1996). Although a penalty term is added to the objective function to penalize infeasible
solutions, the method differs from the way the penalty term is defined in conventional
methods. He devised the following fitness function, where infeasible solutions are
compared based on only their constraint violation:
f ( x),
F ( x)
if g j ( x)
0,
f max
g i ( x) ,
otherwise
j 1,2,......., m,
(6.5)
j 1
The parameter f max is the objective function value of the worst feasible solution in the
population. Thus, the fitness of an infeasible solution not only depends on the amount of
constraint violation, but also on the population of solutions at hand. However, the
fitness of a feasible solution is always fixed and is equal to its objective function value.
The two penalty functions discussed above are used in this research.
6.7
Fitness function
The fitness function is a criterion for the evaluation of the goodness of each
individual in a population. Since the fitness function is a figure of merit, and some of
the selection methods require a positive fitness value, it must therefore take positive
values. In typical cost optimization problems, minimization of some cost function is
carried out rather than maximization of utility or profit.
CHAPTER 7
7.1
Introduction
7.2
In order to validate the developed GA models for both the RC and the PC cases,
two numerical examples were applied and the results are studied.
W d= 7.85 K N /m ',
W l= 12.5 K N /m '
7m
Figure (7.1) The RC simple beam numerical example
48
(integer values)
(real values)
(integer values)
(integer values)
30,70
d cm
70,100
nb
i.e., 30 b 70cm
4,12
d b (mm)
12,24
The Constraints:
The design constraints for the simply supported reinforced concrete beam model
were stated in section 5.6.4. These constraints were used for this beam
The Objective Function:
The objective function for the simply supported reinforced concrete beam model
as stated and explained in section 5.6.5. is:
Z
C c * ( Ac
As ) * L
Wstr
Cs *
* As
Wstr
($)
(7.1)
The model was run to optimize the design of a rectangular cross section for this beam
and loadings while satisfying the provisions of the ACI 318-05 Code.
49
W l= 3 K N / m '
1 9 .5 m
Figure (7.2) The PC simple beam numerical example
b = Beam width
e = Eccentricity of the c.g of the tendons
nb = Number of flexural bars
db = Diameter of flexural bars
d p = Effective beam depth
nt = Number of tendons
dt = Diameter of tendons
(integer values)
(real values)
(integer values)
(integer values)
(real values)
(integer values)
(real values)
50
30,70
e cm
10,50
nb
4,10
d b (mm)
12,18
d p (cm)
80,120
nt
2,20
d t (in)
0.25,0.5625
The Constraints:
The design constraints for the simply supported prestressed concrete beam model
were stated in section 5.7.4, these constraints were used for this beam
The Objective Function:
The objective function for the simply supported prestressed concrete beam model
was stated and explained in section 5.7.5. The function is stated again as:
Z
Cc * ( Ac
C ps *
( As
* A ps
Aps )) * L
L
Wstr
( $)
Cs *
* As
Wstr
(7.2)
It is intended to optimize the design of a rectangular cross section for this beam and
loadings while satisfying the provisions of the ACI 318-05 Code
51
The Population size = 100, the Mutation rate = 0.15, the Percent of population not
selected for mating = 0.5, Roulette Wheel selection function, Single and two point
crossover and Simple and Deb penalty methods.
Figures (7.3), (7.4), (7.5) and (7.6) show the results of tuning the number of
iterations. It is clear that by using a very little number of iterations (< 50), the optimizer
kept the value of the cost far away from the optimum (minimum). Therefore, the
obtained solution is not the best possible one. This means that using a very little number
of iteration may allow little application of the basic operators of the GA on the
population. Therefore, the model will not be able to explore wide range of developed
populations (solutions) and the model may get stuck to a suboptimal or infeasible
solution.
It can be said that as the number of iteration increases, the solution will enhance,
this is true until the number of iteration reaches the value of 200 for both the single
point and the two point crossover types, these results are the same when using both the
simple and Deb penalty functions, beyond this value, any increase in the number of
iteration will not significantly alter the best solution.
The number of iterations of 200 is a moderate value that could result in the
optimum solution in an acceptable time of run of the GA model for both the RC and PC
models. It can also be seen that the type of crossover and the type of the penalty
function has a small effect on the results of the GA model either on the trend of the
graph or on the minimum cost obtained.
170
165
160
155
150
145
140
135
130
0
50
100
150
200
250
No. of iterations
One point crossover
Figure (7.3) No. of iterations Vs. Min cost for the RC beam
300
Figure (7.4) No. of iterations Vs. Min cost for the RC beam
(Population Size = 100, Simple Penalty unction)
Figure (7.5) No. of iterations Vs. Min cost for the PC beam
(Population Size = 100, Deb Penalty function)
52