Computers and Concrete, Vol.13, No.

4 (2014)457-482
DOI: http://dx.doi.org/10.12989/cac.2014.13.4.457 457

Optimal design of reinforced concrete beams: A review

Ima Rahmanian1, Yves Lucet2a and Solomon Tesfamariam1b
1
School of Engineering, The University of British Columbia, Kelowna, BC, Canada
2
Computer Science, The University of British Columbia, Kelowna, BC, Canada

(Received May 31, 2013, Revised January 21, 2014, Accepted February 1 2014)

Abstract. This paper summarizes available literature on the optimization of reinforced concrete (RC)
beams. The objective of optimization (e.g. minimum cost or weight), the design variables and the constraints
considered by different studies vary widely and therefore, different optimization methods have been
employed to provide the optimal design of RC beams, whether as isolated structural components or as part
of a structural frame. The review of literature suggests that nonlinear deterministic approaches can be
efficiently employed to provide optimal design of RC beams, given the small number of variables. This
paper also presents spreadsheet implementation of cost optimization of RC beams in the familiar MS Excel
environment to illustrate the efficiency of the exhaustive enumeration method for such small discrete search
spaces and to promote its use by engineers and researchers. Furthermore, a sensitivity analysis is performed
on the contribution of various design parameters to the variability of the overall cost of RC beams.

Keywords: reinforced concrete beam; optimization; discrete search space; exhaustive enumeration;
genetic algorithm; spreadsheet implementation; sensitivity analysis

1. Introduction

Reinforced concrete (RC) is now widely used in a variety of structures owing to its versatility,
high compressive strength, durability and resistance to fire and water damage. The vast usage of
concrete structures calls for economical design, and thus, many attempts have been made to
optimize the structural design of RC structures (Structural Engineering Institute 2002), including
the fundamental design of RC beams as primary bending elements.
Conventional structural design of steel reinforced concrete beams involves iterative design and
checks for section dimensions and the amount of steel reinforcement. The process usually starts
with a trial section, where the depth of the beam is selected based on guidelines for deflection
control. The composite flexural resistance of the trial section is checked against the applied
bending moment, considering the effects of the self-weight. This is then followed by checks for
shear resistance, deflection and other code requirements. This practice usually requires many
repeats and takes considerable time until a section is found that satisfies both ultimate (strength)
and serviceability limit states prescribed by design codes. The resultant design, while complying

Corresponding author, Ph.D., E-mail: irahmanian@heroldengineering.com

a
Ph.D., E-mail: yves.lucet@ubc.ca
b
Ph.D., E-mail: Solomon.Tesfamariam@ubc.ca

Copyright © 2014 Techno-Press, Ltd.

http://www.techno-press.org/?journal=cac&subpage=8ISSN: 1598-8198(Print), 1598-818X (Online)
458 Ima Rahmanian, Yves Lucet and Solomon Tesfamariam

with all the code requirements, by no means provides an optimal solution.

Advancement in computer technology and analysis has led many researchers to develop
modern techniques for economic design of RC members and structures, yet early research studies
on the optimization of RC beams date back to 1960s (Norman 1964), when access to machine
computing was very limited. For that reason and many others, the objective of the optimization,
the restrictions applied and the methods employed to find the optimized solution has varied widely
among different research works. This paper, therefore, aims to provide a review of the literature on
the structural design optimization of steel reinforced concrete beams. The problem formulation by
different researchers and various optimization techniques applied to RC beam design are studied
and presented in the following sections. Finally, a simple Excel based optimization tool is
developed and illustrated with an example.

2. Problem formulation

The flexural design procedure for reinforced concrete beams now established in most building
codes around the world, including the Eurocode 2 (British Standards Institution, 2007), ACI 318-
05 (American Concrete Institute 2005) and CSA A23.3-04 (Canadian Standards Association 2004),
is based on the ultimate strength design method. This procedure accounts for the nonlinear
material properties, and recognizes the difference in certainty of various load types (e.g. dead and
live loads) by considering different safety factors for the load types and their combinations. Based
on this design method, RC beam sections have to satisfy ultimate (strength) and serviceability
limit state criteria, which, respectively, account for the ultimate failure of the structure and its
functionality for the intended routine use.
While the design criteria for steel reinforced concrete beams are well established, the optimal
design problem requires to be clearly defined by its objective(s), design variables and the
constraints enforced by codes or prompted by practical restrictions. A typical optimization problem
can be formulated mathematically as (Structural Engineering Institute 2002):

Minimize (or maximize) 𝑓(𝑥)

𝑔𝑗 𝑥 = 0 , 𝑗 = 1, … , 𝑝
𝑆𝑢𝑏𝑗𝑒𝑐𝑡 𝑡𝑜 (1)
𝑔𝑗 𝑥 ≤ 0 , 𝑗 = 𝑝 + 1, … , 𝑚

𝑥 = 𝑥1 , … , 𝑥𝑛
𝑥𝑖𝐿 ≤ 𝑥𝑖 ≤ 𝑥𝑖𝑈 , 𝑖 = 1, … , 𝑛
where 𝑓(𝑥) is the objective function, which may require minimization (cost, weight, etc.) or
maximization (benefits); 𝑔𝑖 (𝑥) is an equality or inequality constraint;𝑥𝑖𝐿 and 𝑥𝑖𝑈 are the lower
and upper bounds for the design variable 𝑥𝑖 ; and 𝑛, 𝑚 and 𝑝 are the number of design variables,
total constraints and equality constraints, respectively.
For the optimization problem of a RC beam design, various research studies over the years
have adopted different design variables as well as objective functions and constraints. The
following sections present a summary of such problem formulations.
 Optimal design of reinforced concrete beams: A review 459

2.1 Design variables

The design variables in optimization of RC beams are generally associated with the dimensions
of the beam and the area of steel reinforcement bars (tensile, compressive, and shear reinforcement)
as well as their arrangement (see Table 1). Although rarely adopted, the strength of the materials
can also be considered as variables (Goble and Moses 1975). The number of variables considered
in various research studies are normally limited to a handful. In a very early study, Norman (1964)
chose the depth of the beam as the only design variable for singly reinforced rectangular and T-
beams, while more than 30 years later, Rajeev and Krishnamoorthy (1998) envisaged seven design
variables that included the width and depth of the beam and the area of five reinforcement groups
to allow for a non-uniform distribution of tensile and compressive reinforcement along the length
of the beam. Nevertheless, the main variables are typically one or both dimensions of a rectangular
beam and the area of tensile reinforcement (Chakrabarty 1992b, Camp et al. 2003).The depth of
the beam is either expressed as the effective depth (d) (i.e., depth of tensile reinforcement from the
top compression edge) or as the overall height of the beam (h), as shown in Fig. 1(a). Nonetheless,
the thickness of concrete cover can be assumed fixed and dictated by durability and other
construction requirements. For T-beams, it is usually assumed that the thickness of the flange (hf)
is determined by the depth of the floor slab and therefore assumed to be fixed (Chou 1977,
Balaguru 1980b, Prakash et al. 1988).
Rectangular or T-beams are the most commonly practiced shapes; however, in a recent study
Narayan and Venkataramana (2007) attempted to optimize the shape of a RC beam. Referring to
the pioneer study in shape optimization by Michell (1904), they argued that shape should be
considered as a primary variable in design of structures. Due to concrete cracking, RC design
approaches ignore the insignificant strength of concrete in tension zone. Thus, to reduce the
concrete in tensile zone and incorporate it in the more efficient compression zone, Narayan and
Venkataramana assumed a trapezoidal shape for the RC beam (Fig. 1(b)). They adopted the depth
of the beam and α (the angle in Fig. 1(b)) as the design variables. Comparing the costs of
rectangular and trapizoidal beams, Narayan and Venkataramana concluded that trapezoidal RC
beams are more economical and their usage should be encouraged.

(b) Trapezoidal beam assumed by

(a) Rectangular and T-beams
Narayan and Venkataramana (2007)
Fig. 1 Cross-sectional dimensions of reinforced concrete beams
460 Ima Rahmanian, Yves Lucet and Solomon Tesfamariam

Table 1 Design variables adopted by various research studies

Study Design variables
Norman (1964) D
Sandhu (1971) As
Friel (1974) D, As
Goble and Moses (1975) b, D, As - also suggest provisions for fc
Chou (1977) T beams – D and As
Balaguru (1980a) b, D, As, A’s
Balaguru (1980b) T beams – bw, D and As
Colin and Macrae (1984) As and A’s for RC beams of I, T or rectangular shapes
Prakash et al. (1988) D/b ratio and As. D is assumed to be fixed.
Chakrabarty (1992a) b, D, As
Chakrabarty (1992b) b, D, As
Adamu and Karihaloo (1994b) D, As (As is allowed to change freely along the beam)
Adamu and Karihaloo (1994a) D, As (each variable is allowed to change freely along the
beam)
Adamu and Karihaloo (1994c) b, D, As (each variable is allowed to change freely along the
beam)
Chung and Sun (1994) b and As
Al-salloum and Siddiqi (1995) D and As
Coello and Hernández (1997) b, D, As
Rajeev and Krishnamoorthy (1998) b, D and 4 other variables for area of steel reinforcement
along a continuous beam: Top and bottom continuous bar
diameter, additional top reinforcement at support, and
additional bottom reinforcement at mid-span
Ceranic and Fryer (2000) D, As
Dole et al. (2000) D, As
Camp et al. (2003) b, D, As, A’s
Ferreira et al. (2003) Ratio of A’s/ As
(Although not declared as variable, considered 8 cases for the
position of neutral axis and depth of the flange of T-beam.)
Lee and Ahn (2003) b, D, As, A’s
Lepš and Šejnoha (2003) b, D, As, A’s
Guerra and Kiousis (2006) b, D, As, A’s
Narayan and Venkataramana (2007) D and angle α in Fig. 1(b).
González-Vidosa et al. (2008) b, D and other variables for area of steel reinforcement along
a continuous beam: Top and bottom continuous bar diameter,
additional top reinforcement at left and right supports, and
additional bottom reinforcement at mid-span
D: beam depth (h or d)
b: beam width
bw: T-beam web width
As: area of tensile steel reinforcement
A’s: area of compressive steel reinforcement
fc: Concrete strength
 Optimal design of reinforced concrete beams: A review 461

Table 2 Optimization objective function as adopted by various research studies

Study Optimization objective Included in the objective function
Norman (1964) Minimum cost Cs, Cc and Cf
Sandhu (1971) Minimum cost Cs, Cc and Cf
Friel (1974) Cs, Cc, Cf and the cost of increase
Minimum cost in the building height due to beam
depth
Goble and Moses (1975) Minimum cost Cs, Cc and Cf
Chou (1977) Minimum cost Cs and Cc
Balaguru (1980a) Minimum cost Cs, Cc and Cf
Balaguru (1980b) Minimum cost Cs, Cc and Cf
Colin and Macrae (1984) Minimum cost
Also suggest alternative
objectives such as minimum Cs, Cc and Cf
weight, concrete volume or
steel reinforcement
Prakash et al. (1988) Minimum cost Cs and Cc
Chakrabarty (1992a) Minimum cost Cs, Cc and Cf
Chakrabarty (1992b) Minimum cost Cs, Cc and Cf
Adamu and Karihaloo (1994b) Minimum cost Cs, Cc and Cf
Adamu and Karihaloo (1994a) Minimum cost Cs, Cc and Cf
Adamu and Karihaloo (1994c) Minimum cost Cs, Cc and Cf
Chung and Sun (1994) Minimum weight Weight of concrete and steel
Al-salloum and Siddiqi (1995) Minimum cost Cs, Cc and Cf
Coello and Hernández (1997) Minimum cost Cs, Cc and Cf
Rajeev and Krishnamoorthy (1998) Minimum cost Cs, Cc and Cf
Ceranic and Fryer (2000) Minimum cost Cs and Cc
Dole et al. (2000) Minimum cost Cs and Cc
Camp et al. (2003) Minimum cost Cs, Cc and Cf
Ferreira et al. (2003) Minimum Reinforcement Area of steel reinforcement
Lee and Ahn (2003) Minimum cost Cs, Cc and Cf
Lepš and Šejnoha (2003) Minimum cost Cs and Cc
Guerra and Kiousis (2006) Cs, Cc, cost of formworks as a
function of beam width and cost of
Minimum cost placing concrete and vibrating,
including labour and equipment as
a function of cross-sectional area
Narayan and Venkataramana
Minimum cost Cs, Cc and Cf
(2007)
González-Vidosa et al. (2008) Minimum cost Cs, Cc and Cf
Cs: Cost of steel - constant per weight
Cc: Cost of concrete - constant per weight
Cf: Cost of formworks - constant per surface area
462 Ima Rahmanian, Yves Lucet and Solomon Tesfamariam

2.2 Optimization objectives

Design optimization of structures is evidently driven by economic implications of building

construction. Yet, the objective function for design optimization of RC beams is defined differently
by various researchers (see Table 2). Chung and Sun (1994) for example, adopted minimization of
the overall weight of the beam as the objective. Acknowledging that weight does not appropriately
represent the cost of material, they performed a comparative study with three ratios for cost per
unit weight of concrete (Cc) to that of steel (Cs). The maximum cost reduction was obtained for the
case where the cost of concrete was ten times compared to steel. However, when steel was
assumed to cost ten times compared to concrete, the optimum cost showed an increase and the
upper limit of the beam width was reached, implying that, owing to the high Cs, increasing the
width of concrete beam did not sufficiently increase the resistance of the section. Nevertheless,
realistic cost ratios of concrete and steel are far from Chung and Sun’s assumptions, with Cs being
of the order of a few tens of Cc (Camp et al. 2003).
Naturally, the majority of the available literature has adopted cost minimization as the objective
of optimal design. Some research works only included the cost of concrete and steel in the analysis
(Chou 1977, Prakash et al. 1988, Ceranic and Fryer 2000, Lepš and Šejnoha 2003). Prakash et al.
(1988) argued that a weight-base optimization is better suited to high-rise buildings where the
same component recurs in all stories; nonetheless they used the minimum cost criterion as the
basis of their optimization approach. They stated that the cost of formwork (Cf) is also one of the
main factors to be considered, but they neglected it in calculations to simplify the problem. To
cover the practical variety, Prakash et al. considered the cost ratio of the unit volume of steel to
concrete to vary in the range 50 to 100.
Most other studies accounted for the cost of formworks as well as the material, assuming
constant values for the cost per unit weight of material and per unit area of formwork (Norman
1964, Sandhu 1971, Goble and Moses 1975, Balaguru 1980b, Al-salloum and Siddiqi 1995). Friel
(1974) also considered in the objective function, the cost of the increase in the building height due
to the beam depth. In a more recent study, Guerra and Kiousis (2006) incorporated the cost of
placement, labour, equipment and accessories in the construction costs in addition to the material
costs. Moreover, they defined the cost of forming and placing concrete as a function of cross-
sectional dimensions, introducing a non-linear cost coefficient in the objective function.

2.3 Constraints

Optimization constraints are the functional and structural requirements of the structure
expressed as equality or inequality equations. Design constraints for RC beams can either be
induced by practical restrictions, for example on constructability and transportability, that limit the
design variables, or be enforced by code requirements on structural response of the structure (see
Table 3). As mentioned before, the code requirements on a RC beam are of two types: ultimate
limit states (ULS) -also called strength limit states- and serviceability limit states (SLS). As
outlined below, there are three main flexural ULS requirements for a RC beam.

1. The moment resistance capacity of the cross-section should be higher than the applied
bending moment. The applied bending moment generally includes the effects of the self-weight of
the beam. The bending moment capacity imposes a nonlinear constraint to the optimization and
can be expressed as
 Optimal design of reinforced concrete beams: A review 463

Table 3 Optimization constraints as adopted by various research studies

Study Constraints
Norman (1964) As is at balanced design, b is constant
Sandhu (1971) X1
Friel (1974) X1, X2, X3
Goble and Moses (1975) Not reported
Chou (1977) X1, X2, X3
Balaguru (1980a) X1, X2, X3
Balaguru (1980b) X1, X4, reinforcement ratio is fixed at maximum allowable
Colin and Macrae (1984) X1, X2, X3, X11
Prakash et al. (1988) X1, X3, X4
Chakrabarty (1992a) X1, X5
Chakrabarty (1992b) X1, X6
Adamu and Karihaloo (1994b) X1, X2, X3, X8, X9, X11, X12
Adamu and Karihaloo (1994a) X1, X2, X3, X8, X9, X11, X12
Adamu and Karihaloo (1994c) X1, X2, X3, X6, X7, X8, X9, X11, X12
Chung and Sun (1994) X1 (expressed in terms of tensile and compressive stresses), X2,
X3, X6, X7
Al-salloum and Siddiqi (1995) X1, X2, X3, X9 and suggested but not used X8
Coello and Hernández (1997) X1, X4, X5, X6
Rajeev and Krishnamoorthy (1998) X1, X2, X3, X4, X5, X6, X7, X8, X9, X10
Ceranic and Fryer (2000) X1, X2, X3
Dole et al. (2000) X1, X2 (expressed as maximum depth of the neutral axis) and
maximum concrete cover which is checked after optimization is
performed.
Camp et al. (2003) X1, X2, X3, X6 (to allow minimum reinforcement clear
spacing), X5, X8 (to control deflection), X9, X12 and maximum
difference in bar sizes within a single row of reinforcement
Ferre ira et al. (2003) X1, X3
Lee and Ahn (2003) X1, X2, X3, X4, X5, X6, X7 and allowable arrangements of bars
in two rows
Lepš and Šejnoha (2003) X1, X11, X12
Guerra and Kiousis (2006) X1, X2, X3, X4, X5, X6 (to allow minimum reinforcement clear
spacing), X7, X8, X9 and tensile reinforcement area is greater
that the compressive reinforcement area
Narayan and Venkataramana (2007) X1, X2, X3, X9, X11 (assuming the effective moment inertia of
the section is half of the gross moment of inertia) and minimum
and maximum values for the angle of the trapezoidal beam
González-Vidosa et al. (2008) X1, X11, X12

X1: Moment resistance capacity X6: Minimum beam width

constraint X7: Maximum beam width
X2: Minimum steel reinforcement X8: Minimum beam depth
X3: Maximum steel reinforcement X9: Maximum beam depth
X4: Minimum beam width to depth X10: Maximum beam span to depth ratio
ratio X11: Maximum allowable deflection
X5: Minimum beam width to depth X12: Shear Capacity constraint
ratio
464 Ima Rahmanian, Yves Lucet and Solomon Tesfamariam

𝑀𝑓 + 𝑀𝑠𝑤 ≤ 𝑀𝑟 (2)

where 𝑀𝑓 is the factored design bending moment, 𝑀𝑠𝑤 is the bending moment due to self-
weight of the beam and 𝑀𝑟 is the factored bending moment resistance
𝑎
𝑀𝑟 = 𝜑𝑠 𝑓𝑦 𝐴𝑠 𝑑 − (3)
2

𝜑 𝑠 𝑓𝑦 𝐴𝑠
𝑎=𝛼 ′ (4)
1 𝜑 𝑐 𝑓𝑐 𝑏

where
𝜑𝑠 is the resistance factor for steel,
𝜑𝑐 is the resistance factor for concrete,
𝑓𝑦 is the yield strength of steel reinforcement,
𝑓𝑐′ is the compressive strength of concrete,
𝐴𝑠 is the area of tensile steel reinforcement,
𝑏 is the width of the beam,
𝑑 is the effective depth of the beam,
𝑎 is the depth of the equivalent rectangular stress block, and
𝛼1 , 𝛽1 are the equivalent rectangular stress distribution parameters.

2. The area of tension reinforcement (𝐴𝑠 ) should satisfy a minimum amount to ensure that the
steel reinforcement compensates the loss of tensile strength caused by cracking in the concrete.
This nonlinear constraint can be expressed as

0.2 𝑓𝑐′
𝐴𝑠 ≥ 𝑓𝑦
𝑏𝑕 (5)

0.2 𝑓𝑐′
where 𝑕 is the overall height of the beam and 𝐴𝑠,𝑚𝑖𝑛 = 𝑓𝑦
𝑏𝑕 is the minimum reinforcement
provisioned by CSA A23.3-04 (Canadian Standards Association 2004).

3. The area of tension reinforcement (𝐴𝑠 ) should satisfy a maximum amount to avoid brittle
concrete-controlled failure. This constraint is also nonlinear and provided in Eq. (6).
0.75× 700 𝛼 1 𝛽 1 𝜑 𝑐 𝑓𝑐′
700+𝑓𝑦 𝜑 𝑠 𝑓𝑦
𝑏𝑑 (6)

0.75× 700 𝛼 1 𝛽 1 𝜑 𝑐 𝑓𝑐′

where 𝐴𝑠,𝑚𝑎𝑥 = 𝑏𝑑 is the maximum reinforcement provisioned by CSA
700+𝑓𝑦 𝜑 𝑠 𝑓𝑦
A23.3-04.

Serviceability limit state requirements are generally of two types: restrictions on maximum
deflection of the beam and provisions to control the crack width in concrete. Due to formation of
cracks in concrete, calculating the deflection of RC beams can be tedious. To avoid the
 Optimal design of reinforced concrete beams: A review 465

cumbersome calculations, design codes (e.g. CSA A23.3-04) allow for an alternative indirect
approach for deflection control. This approach sets an upper limit on the span to depth ratio of the
beam and ensures sufficient stiffness in the beam to control the deflection. The indirect deflection
control is commonly adopted for applications in usual ranges of load and member sizes. The direct
approach however imposes a complex nonlinear constraint.
Practical restrictions generally limit the dimensions of a RC beam, imposing minimum and
maximum bounds for both width and depth of the beam. Moreover, detailing requirements usually
set a minimum spacing between the reinforcement bars and therefore dictate more feasibility
constraints on the design variables.
Flexural and shear optimization of a RC beam can be done separately one after another. With
the beams in regular range, flexure dominates the design and should be optimized first. For deep
beams the procedure might be reversed. Nevertheless, even when flexural design precedes shear
design, there is an ultimate limit state constraint corresponding to the shear design of the beam that
can affect the flexural optimization and should be taken into account. This constraint sets a
maximum limit for the overall factored shear resistance of the section and is prescribed to prevent
brittle shear compression failure especially in beams heavily reinforced with shear reinforcement.
This ultimate limit state essentially adds a lower bound limit to the width of the section.

2.4 Continuous and discrete search space

In analytical design of RC beams, for calculation purposes, the design variables (i.e., cross-
sectional dimensions and reinforcements) are assumed to be continuous quantities. However, the
final dimensions and reinforcements are selected from among practical discrete choices. The cross-
sectional dimensions of concrete beams are rounded to nominal sizes as a result of practical limits
on the accuracy of formwork measurements and construction. Therefore, although optimization of
concrete beams can be performed using continuous beam dimensions, the final selection of
dimensions will be from a discrete set of nominal sizes based on a pragmatic accuracy. The same
applies to selection of steel reinforcements. Flexural constraints on the RC beam limit the
minimum and maximum area of reinforcement. However, to provide the area required, the
reinforcement bars are selected from discrete nominal sizes available in the construction industry.
Moreover, the detailing requirements of the design codes restrict the clear spacing between the
reinforcement. To satisfy such requirements, the mere area of the reinforcement would not be
sufficient; the size and number of the reinforcement bars should be known.
Thus, the optimization search space for RC beams is essentially discrete. However, given the
lower complexity of some continuous optimization techniques the discrete nature of the design
variables is ignored by many early research studies and it is assumed that treating the design
variables as continuous leads to sufficiently accurate optimal results (Norman 1964, Sandhu 1971,
Friel 1974, Chou 1977, Balaguru 1980a, Prakash et al. 1988, Chakrabarty 1992a, Chung and Sun
1994, Al-salloum and Siddiqi 1995).

3. Optimization techniques

Once the optimization problem is fully defined by its objective(s) and constraints, a suitable
method can be chosen to find the optimal solution. A vast range of optimization techniques are
available that can be categorized into two main types: linear and non-linear programming
466 Ima Rahmanian, Yves Lucet and Solomon Tesfamariam

techniques.
Linear programming approaches can be applied to problems where the objective functions and
constraints can all be expressed by linear equations. The most widely used algorithm for linear
programming problems with a small number of variables is the simplex method or one of its
variant (Revelle et al. 2004). However, the RC beam design is usually neither a linear nor a convex
problem. Regardless of the different problem formulations adopted by various authors, there exists
nonlinearity in both the objective function and the constraints of the optimization of RC flexural
sections. Hence, nonlinear methods should be explored. Non-linear programming approaches can
be divided into three large categories: Enumerative, Deterministic, and Heuristic methods (Garcia
et al. 2006). The nonlinear methods are further discussed here.

3.1 Enumerative approach

The simplest optimization method is exhaustive enumeration. This method is based on

generating and evaluating all combinations of the discrete variables. The overall number of
evaluation 𝑛𝑒 is:
𝑛𝑑
𝑛𝑒 = 1 𝑝𝑖 (7)

where 𝑝𝑖 is the number of possible discrete values for each variable and 𝑛𝑑 is the number of
discrete variables. The optimal solution is obtained by examining the list of feasible solutions
against the objective function. This method is conceptually simple and guarantees the global
optimum, but the computational time can be impractically large.
To speed up the location of the global optimum, the search space is represented as a decision
tree where nodes represent discrete variables and edges represent possible values for the parent
node. With the availability of a bounding function, parts of the search space that do not contain the
global minimum can be skipped. The base technique is called branch-and-bound. In the worst case
it amounts to an exhaustive search but performs much better in practice. Several variations have
been implemented, e.g. branch-and-price, branch-and-cut, and branch-reduce-cut.
Search strategies for (mixed) integer linear programming can be found in (Linderoth and
Savelesbergh 1999) and in the reference book (Nemhauser and Wolsey 1999) (see also the more
recent book (Chen et al. 2010)). A history of integer programming including the latest
developments can be found in (Juenger et al. 2010), while a very recent survey is available in
(Burer and Letchford 2012).

3.2 Deterministic approaches

Deterministic methods use a successive search within the solution space which can be based on
the objective function, its gradient information or both. They focus on optimization of continuous
variables. The optimization process starts with a possible solution and finds the next iteration
either by finding a descent direction and computing a step in that directly (line search strategy) or
Table 4 Deterministic optimization techniques adopted by various research studies by solving a
simple optimization subproblem (trust region strategy (Nocedal and Stephen Wright 2006)).The
search continues until a stopping condition is satisfied, typically until a first order optimality
condition, usually the Karush-Kuhn-Tucker (KKT) condition is found within the admissible error.
The calculations for search direction can be done through various methods with one of the most
 Optimal design of reinforced concrete beams: A review 467

Table 4 Deterministic optimization techniques adopted by various research studies

Study Deterministic optimization technique
Norman (1964) Simple derivatives (one variable)
Sandhu (1971) Simple differentiation (one variable at a time)
Friel (1974) Lagrange multiplier technique
Goble and Moses (1975) Penalty function programming
Chou (1977) Lagrange multiplier technique
Balaguru (1980a) Lagrange multiplier technique
Balaguru (1980b) Simple differentiation (one variable at a time)
Colin and Macrae (1984) Solve a sequence of linearly constrained nonlinear
programs†
Prakash et al. (1988) Lagrange multiplier technique
Chakrabarty (1992a) Geometric or standard nonlinear programming
algorithms
Chakrabarty (1992b) Geometric programming technique - the newton-
raphson method
Adamu and Karihaloo (1994b) Discretized continuum-type optimality criteria
Adamu and Karihaloo (1994a) Discretized continuum-type optimality criteria
Adamu and Karihaloo (1994c) Augmented lagrangian technique
Chung and Sun (1994) Sequential linear programming algorithm
Al-salloum and Siddiqi (1995) Lagrange multiplier technique
Ceranic and Fryer (2000) Lagrange multiplier technique
Dole et al. (2000) Polynomial optimization technique
Dole et al. (2000) Simple differentiation (one variable)
Guerra and Kiousis (2006) Sequential quadratic programming algorithm
Narayan and Venkataramana (2007) Sequential unconstrained minimization technique
†
As specified in (Hock and Schittkowski 1983), the code FCDPAK implements (Robinson 1972)

popular methods being the BFGS (Broyden-Fletcher-Goldfarb-Shanno) quasi-Newton method

with line search and Wolfe condition that requires the function and the gradient value at each
iterate.
Evidently, the adopted technique by different researchers is based on the complexity of the
problem formulation (see Table 4). In the simplest case, the design variables are determined by
simple derivations as only one parameter (or one parameter at a time) is assumed as variable
(Norman 1964, Sandhu 1971, Balaguru 1980b). Goble and Moses (1975), however, used a penalty
method to find the optimal dimensions of a RC beam and the area of steel reinforcement, and
suggested utilizing Powell’s (1964) search method that does not depend on the derivatives of the
function (modern interior point methods are more appropriate nowadays). Nevertheless, the most
widely used methods for RC beam optimization are the gradient-based optimization techniques,
such as the geometric programming technique (Chakrabarty 1992a,b), the sequential linear and
quadratic programming algorithm (Chung and Sun 1994, Guerra and Kiousis 2006). In the simpler
case, the KKT optimality conditions can be solved and an analytic solution obtained 1explicitly
(Friel 1974, Chou 1977, Balaguru 1980a, Al-salloum and Siddiqi 1995, Ceranic and Fryer 2000) or

1
Solving the optimality conditions of a constrained optimization problem is sometimes called the Lagrangian
technique (or Lagrange multiplier technique) as it requires writing the Lagrangian, computing its gradient,
and solving the resulting system.
468 Ima Rahmanian, Yves Lucet and Solomon Tesfamariam

Table 5 Stochastic optimization techniques adopted by various research studies

Study Stochastic optimization technique
Coello and Hernández (1997) Genetic algorithm
Rajeev and Krishnamoorthy (1998) Genetic algorithm
Camp et al. (2003) Genetic algorithm
Lee and Ahn (2003) Genetic algorithm
Lepš and Šejnoha (2003) Simulated annealing
González-Vidosa et al. (2008) Augmented simulated annealing
McCluskey and McCarthy (2009) Particle swarm algorithm
Ozturk et al. (2013) Artificial bee colony (swarm-based)
Medeiros and Kripka (2013) Simulated annealing

with a computer (Prakash et al. 1988).

Gradient-based optimization techniques are more efficient. They usually enjoy a linear rate of
convergence for first-order methods and even superlinear rate for second-order methods (Newton
and quasi-Newton methods) as opposed to a sub-linear rate for the non-gradient-based
optimization methods. However, since gradient-based methods find iterations based on derivative
information, they can at best guarantee local optimality. Global optimality can only be achieved
for convex problems (Nocedal and S. Wright 2006, Theorem 2.5 p.16).Therefore, depending on
the problem formulation, deterministic optimization methods may not be the best approach for the
optimal RC beam design.

3.3 Stochastic and heuristic approaches

Stochastic and heuristic methods search for the optimal solution using probability rules and an
oriented random manner (Sivanandam and Deepa 2008). When deterministic methods fail to find the
global optimum of the objective function, or they are computationally too expensive and time
consuming, stochastic methods may be used to provide a satisfactory solution in a timelier manner.
Stochastic methods are most suited to problems of higher complexity and higher number of variables
and constraints. Therefore, to satisfactorily scan all the regions of the problem domain for the optimal
solution, stochastic methods usually require numerous computer calculations. Given the increase in
computational power, stochastic methods are attracting increasing popularity among researchers in
engineering. The main advantages of these methods over the conventional optimization techniques
include the following: they do not require gradient information; constraints do not need to be explicit;
and restrictions about the search space (e.g. continuity) do not prevent the application of these methods.
Various heuristic optimization methods have been applied in structural engineering (Hare et al. 2013).
The most commonly adopted techniques in this category are simulated annealing (Salamon et al. 2002),
evolutionary algorithms (Sivanandam and Deepa 2008) and swarm-based optimization algorithms
(Olsson 2011).
Using stochastic methods for design of RC beams are introduced just before the turn of the century
by the likes of Coello and Hernández (1997) and Rajeev and Krishnamoorthy (1998) through
implementation of genetic algorithms (see Table 5). Genetic algorithm (GA) is inspired by the process
of natural evolution (Goldberg 1989). In this approach, a population of individuals (variables) is
 Optimal design of reinforced concrete beams: A review 469

evolved to produce better solutions. The evolution progresses in generations. In each generation, the
fitness of the individuals is evaluated against a fitness function (objective function); better fit
individuals are selected and randomly modified to produce the next generation; and iterations continue
until a maximum number of generations are formed or a satisfactory level of fitness is achieved.
Coello and Hernández (1997) utilize GAs with various representation schemes such as floating
point and binary representation. They assume a continuous search space and they adjust different
parameters of the GA to obtain reasonable solutions in shorter times. Rajeev and Krishnamoorthy
(1998), however, argue that using continuous design variables requires further modification to the
solution obtained and therefore does not represent the realistic constraints and does not provide rational
solutions. In their paper, Rajeev and Krishnamoorthy include detailing and other construction related
constraints in the optimization problem and use GA-based methodologies to optimize the design of
reinforced concrete frames (including beams and columns) with discrete design variables. Camp et al.
(2003) and Lee and Ahn (2003) also use discrete variables and implement GAs to provide optimization
procedures for flexural design of simply supported beams, uniaxial columns and multistory frames.
The simulated annealing algorithm emulates the physical process of crystallization of a melted solid.
As the mass cools down slowly, higher energy configuration of crystals reduces and eventually the
solid reaches the minimum energy configuration. The iterations in this method are based on
probabilistic information and decisions as to stay in a state or to move to a neighbouring state, which
should ultimately lead the structure to lower levels of energy. González-Vidosa et al. (2008) use
simulated annealing procedures to provide optimal solutions to the design of RC walls and frames.
They reiterate that the restricting constraints in beam design are those related to flexure, shear and
deflection of the beams. Lepš and Šejnoha (2003) employ a version of the augmented simulated
annealing method to solve the optimal design of a RC beam with discrete variables, and they account
for both flexural and shear reinforcements.

4. Spreadsheet implementation and illustrative example

The review of the literature suggests that the design optimization of RC beams as isolated
members can be achieved using deterministic optimization techniques and there is no need to use
heuristic methods. Although recent advancement in computational power and consequent
popularity of stochastic and heuristic methods can be attractive, the more reliable deterministic
approaches can be sufficient for flexural design optimization of RC beams, given the small number

Fig. 2 A singly reinforced concrete beam

470 Ima Rahmanian, Yves Lucet and Solomon Tesfamariam

(a) Stress-strain diagrams for concrete and steel (b) Actual and equivalent rectangular stress
distribution
Fig. 3 Stress distribution in a reinforced concrete beam

of variables. The optimization problem can be implemented in the familiar spreadsheet

environment to promote its use by engineers. This section presents such spreadsheet
implementation of the RC beam optimization problem and demonstrates, through an example, the
efficiency of deterministic methods in finding the optimal solution.

4.1 Problem formulation

4.1.1 Design variables

Assuming the singly reinforced rectangular beam illustrated in Fig. 2, the design variables are
the beam width, depth and the area of steel reinforcement that can be optimized by applying the
structural and practical constraints. As mentioned earlier in section 2.4, the values that these
variables can assume are restricted by practical implementations and are, hence, of a discrete
nature. To perform the optimization in a discrete search space, the width and depth of the beam are
expressed as a function of a user-defined precision. This dimension precision refers to nominal
beam sizes and the practical limit for the accuracy of formwork measurement at the construction
site. For example, if dimension precision is set to 50mm, the beam dimensions are rounded to the
nearest 50mm. Therefore, each dimension is expressed as the product of an integer multiplier
times the dimension precision (see Eqs. (8)-(9)). Therefore, the discrete decision variables to be
used in optimization are the integer multipliers for width and depth of the beam (𝑏𝑝𝑚 and 𝑕𝑝𝑚
in Eqs. (8)-(9)).
𝑏 = 𝑏𝑝𝑚 ∗ 𝑝𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛 (8)
𝑏 = 𝑏𝑝𝑚 ∗ 𝑝𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛 (9)
The area of the reinforcement bars is also a function of the area of each bar. The size of the bars
used is defined by the user; hence, the area of steel can be expressed as:
𝐴𝑠 = 𝑁𝑏 ∗ 𝑎𝑟𝑒𝑎 𝑜𝑓 𝑒𝑎𝑐𝑕 𝑡𝑒𝑛𝑠𝑖𝑙𝑒 𝑏𝑎𝑟 (10)

4.1.2 Objective function

The objective of the optimization is to minimize the overall cost of the beam including both
material and construction cost. The objective function is expressed in Eq. (11)
𝑓 𝑏, 𝑕, 𝐴𝑠 = 𝑐1 𝐴𝑠 + 𝑐2 𝑏𝑕 + 𝑐3 (2𝑕 + 𝑏) (11)
 Optimal design of reinforced concrete beams: A review 471

where
𝑓(𝑏, 𝑕, 𝐴𝑠 )is the cost per unit length of the beam, and 𝑐1 , 𝑐2 and 𝑐3 are constant coefficients
that can be defined by the user:
𝑐1 is the cost coefficient due to the volume of tensile reinforcement steel
𝑐2 is the cost coefficient due to the volume of concrete
𝑐3 is the cost coefficient due to shuttering along the surfaces of the beam.

4.1.3 Constraints
The constraints of the optimization problem can be divided into three categories: ULS and SLS
constraints prescribed in CSA A23.3-04 (Canadian Standards Association 2004), and other
practical constraints resulting from architectural or construction limitations.
Flexural and shear ULSs ensure that flexural stresses in the RC beam do not exceed the
material strength. Fig. 3(a) illustrates the material properties of concrete and steel reinforcements.
The stress-strain relationship of concrete is a nonlinear curve, while the stress-strain relationship
for steel can be represented by an elastoplastic diagram. Using the actual nonlinear stress-strain
curve for concrete is not practical for design purposes. Therefore, CSA A23.3-04 allows for an
equivalent rectangular stress block to be used instead of the nonlinear stress distribution, as
demonstrated in Fig. 3(b).The depth and magnitude of the equivalent stress block are calculated
using 𝛼1 and 𝛽1 , mathematical parameters that ensure the compressive stress resultants of the
actual and the equivalent rectangular stress distribution are equal. Therefore, in quantifying the
constraints, the bending resistance of the section is calculated using the equivalent rectangular
stress distribution.
Optimization constraints resulting from ULSs can be summarized as:
1. Maximum bending moment resistance constraint
2. Minimum reinforcement constraint
3. Maximum reinforcement constraint
4. Maximum factored shear resistance constraint
The following constraints result from SLSs:
5. Maximum deflection constraint
6. Maximum crack opening constraint
Other practical constraints include:
7. Bar spacing constraint
8. Minimum and maximum beam width constraint
9. Minimum and maximum beam depth constraint
10. Minimum and maximum beam depth to width ratio constraint
The detailed equations for the constraints are presented in the appendix.

4.2 Spreadsheet implementation

Using the aforementioned objective function and constraints, design optimization of the singly
reinforced rectangular beam can be implemented in the familiar MS Excel spreadsheet
environment. A worksheet is assigned to input all the user-defined parameters that define the
material properties, design requirements and practical restrictions, as depicted in Fig. 4.
Calculations are accommodated in a separate worksheet and the optimization results are presented
in the third worksheet (Fig. 5).
The Solver Add-In embedded in MS Excel can be used to run the optimization. The decision
472 Ima Rahmanian, Yves Lucet and Solomon Tesfamariam

variables and the constraints are added to Solver through a VBA macro assigned to user buttons.
Running the solver minimizes the objective function and determines the optimal values for the
design variables.
The Solver Add-In in MS Excel 2010 offers three techniques for solving optimization problems:
Simplex method for linear problems, Generalized Reduced Gradient (GRG) algorithm (Lasdon
and Waren 1978) for optimizing smooth nonlinear problems, and Evolutionary method for non-
smooth nonlinear problems which utilized genetic algorithms. The optimal beam design is a
nonlinear problem and either the GRG Nonlinear or the Evolutionary options can be used in MS
Excel. However, given the non-convex nature of the problem, none of the two techniques
guarantee the global optimum solution. In fact, depending on the starting values of the decision
variables, Solver may fail to converge to a feasible solution at all.
A good starting point can help the optimization process. Engineering practice suggests
estimates for the dimensions of the beam that can be used as starting point. The overall depth of
the beam (𝑕) can be estimated about 10% of the beam span to make sure that the deflection criteria
is met. The width of the beam (𝑏) can then be calculated using a depth to width ratio of about 1.5.
The area of reinforcement (𝐴𝑠 ) should be selected so that ductile behaviour is ensured. If the onset
of yielding in tensile reinforcement and crushing in concrete occurs simultaneously, the area

User specified material properties

Concrete
Compressive strength f'c 25 MPa
maximum size of aggregate amax 20 mm
Unit weight force wc 23500 N/m3
Material cost 2 $/m3
Steel
Yield strength fy 400 MPa
Modulus of elasticity Es 200000 MPa
Material cost 250 $/m3
Nominal diameter of tensile bars 25 mm
Nominal diameter of stirrups 10 mm
Minimum number of tensile bars 2
Shuttering
Material and construction cost 1 $/m2

Design parameters
Uniformly Distributed Dead Load DL 15 kN/m
Uniformly Distributed Live Load LL 25 kN/m
Concrete cover provided c 30 mm
Maximum allowable deflection Ln/240

Beam dimensions and exposure condition

Exposure class N
Span of the beam L 5 m
Clear span of the beam Ln 4.6 m
Dimension precision 50 mm
Beam width
Minimum 300 mm
Maximum 1000 mm
Beam depth
Minimum 100 mm
Maximum 800 mm
Depth to width ratio
Minimum 0.5
Maximum 3

Fig. 4 User-defined input worksheet

 Optimal design of reinforced concrete beams: A review 473

Decision variables
Beam width b 300 mm Width precision multiplier 6
Beam Depth h 500 mm Depth precision multiplier 10
Area of tensile steel reinforcement As 1500 mm2 Number of tensile bars 3
Objective function Min..
Total cost of materials and construction $ 1.98
Material cost of concrete $ 0.30
Material cost of steel $ 0.38
Construction and material cost of shuttering $ 1.30
Constraints RHS
C1 Minimum factored bending moment 194.7 >= 189.6 kNm Pass
C2 Minimum reinforcement 1500 >= 375 mm2 Pass
C3 Maximum reinforcement 1500 <= 2251 mm2 Pass
C4 Maximum factored shear resistance 129 <= 489 kN Pass
C5 Minimum beam width precision multiplier 6 >= 6 mm Pass
C6 Maximum beam width precision multiplier 6 <= 20 mm Pass
C7 Minimum beam depth precision multiplier 10 >= 2 mm Pass
C8 Maximum beam depth precision multiplier 10 <= 16 mm Pass
C9 Minimum depth to width ratio 1.7 >= 0.5 Pass
C10 Maximum depth to width ratio 1.7 <= 3 Pass
C11 Maximum deflection 10.3 <= 20.8 mm Pass
C12 Maximum crack control parameter 20027 <= 30000 N/mm Pass
C13 Maximum number of tensile bars 3 <= 4 mm Pass
C14 Minimum number of tensile bars 3 >= 2 N/mm Pass

Run with recommanded initial Run with exhaustive enumeration

Run with user initial dimensions
dimensions

Fig. 5 Optimization worksheet

of reinforcement is at balanced condition. Using about 40% of the reinforcement at balanced

condition should result in ductile behaviour. Hence, the initial value of reinforcement can be
estimated. Running the optimization with the proposed starting point (usually a feasible solution)
can enhance the optimization process. The effect of the starting point is numerically studied in the
next section.

4.3 Numerical results

This section presents a numerical example to illustrate the efficiency of various optimization
methods in quantifying the optimal design of a RC beam. The assumed RC beam spans 5m, is
rectangular in cross-section and is reinforced with a single layer of tensile steel bars (see Fig. 2).
The material properties and loading of the beam are presented in Fig. 4.
Using the Evolutionary and GRG Nonlinear options in Solver, the optimization does not always
result in a feasible solution, i.e., not all constraints are satisfied. Although using the estimated
dimensions and reinforcement as the starting point can help converging to a feasible solution, it
does not guarantee such results. To compare the optimization results with the absolute global
optimum solution, a VBA code is developed that enumerates all the possible combinations of the
decision variables and compares the feasible solutions to determine the global optimal design.
The degree of convergence in Solver is set to 0.001 for both GRG Nonlinear and Evolutionary
methods. The population size in Evolutionary method is set to 100 and the mutation rate is 0.075.
The estimated starting point for this example is 𝑏 = 350 𝑚𝑚, 𝑕 = 500 𝑚𝑚, 𝐴𝑠 = 1500 𝑚𝑚2 .
Table 6 presents the result of the three optimization techniques and their processing times.
According to the exhaustive enumeration, the number of all possible combinations of decision
variables is 3,150 but only 874 combinations are feasible for this typical RC beam design and with
such a small search space, the global optimum can be found in less than a second.
The GRG Nonlinear method fails to provide a feasible solution with the recommended starting
point; however, if the optimization is run for a second time using the result of the first run as the
starting point, the program returns the absolute global optimum. The optimization takes less than a
second to complete with a typical office computer. On the other hand, the Evolutionary method
takes about 36 seconds to complete and still fails to move from the starting point (which is a
474 Ima Rahmanian, Yves Lucet and Solomon Tesfamariam

Fig.6 Probability distribution of feasible solutions from GRG Nonlinear Solver

Fig. 7 Probability of optimization constraints being met with the solver GRG Nonlinear method

Table 6 Numerical results for optimal design of a singly RC beam using various optimization methods
𝑏 𝑕 𝐴𝑠 Total cost Processing time
Optimization technique Feasible
(mm) (mm) (mm2) (relative) (sec)
GRG nonlinear 300 400 2500 1.97 No 0.16
Evolutionary 350 550 1500 2.21 Yes 36.64
Exhaustive
300 500 1500 1.98 Yes 0.04
Enumeration

feasible solution). Changing the various parameters of the algorithm such as the mutation rate or
the population size does not change the result, although requiring bounds on the variables will
reduce the processing time to less than a second.
Based on the literature reviewed in previous sections, deterministic nonlinear methods should
be sufficient for finding the optimal solution to the problem of RC beam design with relatively
limited search space. But is the GRG Nonlinear method of MS Excel suited to solving this
 Optimal design of reinforced concrete beams: A review 475

problem? To answer this question, the efficiency of the GRG Nonlinear method is examined using
various starting points for optimization. A VBA code was implemented to feed the spreadsheet
with all the possible combinations of design variables (based on upper bound and lower bound
constraints limiting the width and depth of the beam and the number of reinforcing bars) and to
calculate the results based on the GRG Nonlinear Solver. The optimization results are highly
dependent on the starting points. Using the input parameters defined in Fig. 4, there are a total of
3150 possible combinations of design variable; i.e., 3150 possible starting points for the
optimization. The GRG Nonlinear Solver returns a feasible solution with only 8% of the starting
points, but 55% of these feasible solutions are equal to the global optimum. In other words,
although the Solver rarely comes to a feasible solution, if a solution is found, there is a good
chance of it being the global optimum. Fig. 6 shows the probability distribution of the dimensions
and reinforcement of the RC beam as returned by the GRG Nonlinear Solver when feasible
solutions are returned. It is evident that the Solver feasible results are generally very close to the
global optimum values provided a feasible solution is obtained.
A closer examination of GRG Nonlinear Solver results with various starting points shows that
optimization fails to meet the maximum reinforcement constraint with 77% of the starting points.
This constraint is a critical one that ensures ductile failure of the beam. Fig. 7 demonstrates the
probability of failure of all the optimization constraints (notations for constraints refer to those
introduced in Fig. 5). Constraint C13, the maximum number of bars, is next with 51% chance of
not being satisfied. This constraint refers to the physical possibility of arranging tensile bars in one
row as defined earlier. Constraints C11 and C12 correspond to serviceability limit states (SLS) of
deflection and cracking. These constraints are usually governing design factors for long-span
beams. Hence, with the aim of improving the optimization, it is reasonable to relax these
constraints from optimization and later check the SLS on the optimal solution. The analysis is
repeated with SLS constraints relaxed, however, the results hardly improve. The rate of feasibility
of GRG Nonlinear solutions is slightly raised to 11%, and the probability of obtaining the global
optimum is 52% of the feasible solutions. The probability distribution of design variables and
failure rates of various constraints are also very similar.
In conclusion, the feasibility rate of the solutions obtained from the GRG Nonlinear method is
low, while optimality of the feasible solutions is high. The overall results suggest that although
running the nonlinear optimization method available in MS Excel is quick, it is highly dependent
on the starting point and not efficiently reliable. In contrast, the exhaustive enumeration technique
can be efficiently implemented with VBA to provide the optimal design and ensure global
optimality.

4.4 Sensitivity analysis

To investigate the effect of various input parameters on the overall cost of a RC beam, a sensitivity
analysis is performed using the developed spreadsheet. The main variable input parameters
considered are the strength and cost of concrete and steel reinforcement and also the size of
reinforcing bars. The bending moment demand on the beam is usually the governing factor in
designing the size and reinforcement of typical RC beam sections. Therefore, to perform the
sensitivity study, three levels of applied bending moment have been considered: Low, medium and
high. This is achieved by keeping a constant span and increasing the applied dead and live loads
on the input worksheet shown in Fig. 4.
A common sensitivity analysis available in MS Excel is the Pearson correlation method.
476 Ima Rahmanian, Yves Lucet and Solomon Tesfamariam

Pearson correlation coefficient measures the extent of linear correlation between normally
distributed variables (Veaux et al. 2012). However, when variables are not normally distributed or
nonlinear correlations are examined, Spearman’s rank correlation coefficient provides a
moreappropriate method by assessing the degree of monotonic (not necessary linear) correlation
between two variables. In this study, the input parameters are uniformly distributed and the
correlation between the input parameters and the overall cost of construction is not necessarily
linear. Hence, the Spearman’s rank correlation coefficient is applied for sensitivity analysis.
Table 7 shows the range of the five input parameters that are varied uniformly to produce the
data. A VBA code is developed to feed the spreadsheet with all the combinations of input variables
and calculate the respective cost of construction. The procedure is repeated for three levels of
applied bending moment. The data is then ranked to facilitate the calculation of Spearman’s rank
correlation coefficient. Fig. 8 shows the results in the form of tornado graphs for the three bending
moment demand levels (low: 187kNm, medium: 458kNm and high: 743kNm). For all three
bending moment levels, the input parameter that affects the overall cost is the cost of concrete. The
rank correlation coefficient for concrete cost is 0.52 for low bending moment, which increase to

Fig. 8 Percent of contribution of various input parameters to the variability of the overall cost of
construction for three levels of bending moment demand

Table 7 Range of uniform distribution of input parameters used for sensitivity analysis
Cost of
Concrete Steel Reinforcing bar Cost of steel Cs
Input concrete Cc
strength f’c strength fy diameter db (relative to cost of
parameters (relative to cost
(MPa) (MPa) (mm) formworks)
of formworks)
Range 25 - 40 250 - 400 10 -55 1-5 100 - 300
 Optimal design of reinforced concrete beams: A review 477

0.58 and 0.63 for medium and high bending moment levels, respectively. The positive value of the
coefficients means higher values of cost of concrete result in higher value of the overall cost.
When bending moment demand is high, the diameter of the reinforcing bars is the next most
influential input parameter on the overall cost of construction of RC beams. The negative value of
the Spearman’s rank correlation coefficient confirms the higher the diameter of reinforcing bars is,
the lower the cost of construction will be. With low and medium bending moment levels, higher
strength of steel and concrete only slightly reduce the overall cost, with a correlation just over
10%. However, when bending moment level is high, strength of steel plays a significant role in
reducing the overall cost, with a correlation close to 40%.
In conclusion, as can also be predicted by inspection, the sensitivity analysis confirms that
reducing the cost of concrete and steel can significantly decrease the overall cost of construction of
RC beam. The strength of material within the common range has the minimum effect on the
overall cost, as long as bending moment demand is considered low or medium. However, with
high bending moment demand, choosing higher strength steel and larger bar sizes in design of RC
beams can substantially reduce the overall cost of construction.

5. Conclusions

This paper presents a review of the available literature on the design optimization of reinforced
concrete beams as structural members. A comprehensive optimization of a structure ideally studies
the structure as one entity and takes into account the cost of materials, construction and
maintenance, as well as functional and structural constraints. However, given the complex nature
of many concrete structures, it is common to optimize the design of individual components of a
structure to achieve a more economical design for the whole structure. The optimal design of
concrete beams, either individually or as part of a frame, has been addressed by many research
studies using various optimization approaches depending on the problem formulation. The
objective of optimization (e.g. minimum cost, weight ...), the design variables and the constraints
considered by different studies vary widely and hence, different optimization methods have been
employed to provide the optimal design. The review suggests that cost optimization of a concrete
beam is a discrete, but nonlinear and non-convex problem in nature, yet it can be achieved using
deterministic approaches rather than heuristic ones, especially when the beam is considered as an
isolated member and the number of design variables is limited.
This paper also presents a spreadsheet implementation of the optimization of concrete beams
and a numerical example to demonstrate the efficiency of deterministic methods. The results
indicate that given the small search space of many typical RC beam designs and the instability of
nonlinear approaches and their dependency on the starting point, exhaustive enumeration is the
most efficient and reliable method that guarantees global optimality. Exhaustive enumeration
method is comfortably implemented in the MS Excel spreadsheet using VBA to promote its use by
engineers.
Using the enumerative method implemented, a sensitivity analysis is performed to evaluate the
influence of different input parameters on the optimal cost of construction. The results suggest that
cost of concrete and steel expectedly have the highest influence on the overall cost of construction
of RC beam; however, when bending moment demand is high on the beam, using higher strength
steel and larger reinforcing bar sizes can greatly reduce the overall costs.
478 Ima Rahmanian, Yves Lucet and Solomon Tesfamariam

Acknowledgements

The first author was supported by an Engage grant from the Natural Sciences and Engineering
Research Council of Canada (NSERC). The second author acknowledges support from NSERC
through the Discovery grant program, and the Canadian Foundation for Innovation (CFI) through
the Leaders Opportunity Fund. The last author was supported by an NSERC Discovery grant.

References

Adamu, A. and Karihaloo, B.L. (1994a), “Minimum cost design of RC beams using DCOC, Part I : beams
with freely-varying cross-sections”, Struct. Optim., 7(4), 237-251.
Adamu, A. and Karihaloo, B.L. (1994b), “Minimum cost design of RC beams using DCOC, Part II : beams
with uniform cross-sections”, Struct. Optim., 7(4), 252-259.
Adamu, A. and Karihaloo, B.L. (1994c), “Minimum cost design of reinforced concrete beams using
continuum-type optimality criteria”, Struct. Optim., 7(1-2), 91-102.
Al-salloum, Y.A. and Siddiqi, G.H. (1995), “Cost-optimum design of reinforced concrete beams”, ACI
Struct. J., 91(6), 647-655.
American Concrete Institute (2005), “Building code requirements for structural concrete and commentary”,
ACI 318-05.
Balaguru, P.N. (1980a), “Cost optimum design of doubly reinforced concrete beams”, Build. Environ.,
15(4), 219-222.
Balaguru, P.N. (1980b), “Optimum design of T-beam sections”, Build. Environ., 15(2), 91-94.
British Standards Institution (2007), “Eurocode 2: Design of concrete structures”, BS EN 1992.
Burer, S., Letchford, A.N. (2012), “Non-convex mixed-integer nonlinear programming: A survey”, Surv.
Oper. Res. Manage. Sci., 17(2), 97-106.
Camp, C. V, Pezeshk, S. and Hansson, H. (2003), “Flexural design of reinforced concrete frames using a
genetic algorithm”, Struct. Eng., 129(1), 105-115.
Canadian Standards Association (2004), “Design of Concrete Structures”, CSA A23.3-04.
Ceranic, B. and Fryer, C. (2000), “Sensitivity analysis and optimum design curves for the minimum cost
design of singly and doubly reinforced concrete beams”, Struct. Multidiscip. Optim., 20(4), 260-268.
Chakrabarty, B.K. (1992a), “A model for optimal design of reinforced concrete beam”, Struct. Eng.,
118(11), 3238-3242.
Chakrabarty, B.K. (1992b), “Models for optimal design of reinforced concrete beams”, Comput. Struct.,
42(3), 447-451.
Chen, D.-S., Batson, R.G. and Dang, Y. (2010), Applied Integer Programming: Modeling and Solution, John
Wiley and Sons Inc., New York.
Chou, T. (1977), “Optimum reinforced concrete T-beam sections”, J. Struct. Div., 103(8), 1605-1617.
Chung, T.T. and Sun, T.C. (1994), “Weight optimization for flexural reinforced concrete beams with static
nonlinear response”, Struct. Optim., 8(2), 174-180.
Coello, C.C. and Hernández, F.S. (1997), “Optimal design of reinforced concrete beams using genetic
algorithms”, Expert Syst. Appl., 12(1), 101-108.
Colin, B.M.Z. and Macrae, A.J. (1984), “Optimization of structural concrete beams”, J. Struct. Eng., 110(7),
1573-1588.
Dole, M.R., Ronghe, G.N. and Gupta, L.M. (2000), “Optimum design of reinforced concrete beams using
polynomial optimization technique”, Adv. Struct. Eng., 3(1), 67-79.
Ferreira, C.C., Barros, M.H.F.M. and Barros, A.F.M. (2003), “Optimal design of reinforced concrete T-
sections in bending”, Eng. Struct., 25(7), 951-964.
Friel, L.L. (1974), “Optimum singly reinforced concrete sections”, J. Am. Concrete Inst., 71(11), 556-558.
Garcia, J.S.D., Avila, S.L. and Carpes, W.P. (2006), “Introduction to optimization methods: a brief survey of
 Optimal design of reinforced concrete beams: A review 479

methods”, IEEE Multidisciplinary Engineering Education Magazine, 1(2), 1-5.

Goble, G.G. and Moses, F. (1975), “Practical applications of structural optimization”, J. Struct. Div., 101(4),
635-648.
Goldberg, D.E. (1989), Genetic ALorithms in Search, Optimization, and Machine Learning, Addison
Wesley Longman, New York, USA.
González-Vidosa, F., Yepes, V., Alcalá, J., Carrera, M., Perea, C. and Payá-Zaforteza, I. (2008),
Optimization of Reinforced Concrete Structures by Simulated Annealing, In Simulated Annealing, I-Tech
Education and Publishing, Vienna, pp. 307-320.
Guerra, A. and Kiousis, P.D. (2006), “Design optimization of reinforced concrete structures”, Comput.
Concrete, 3(5), 313-334.
Hare, W., Nutini, J. and Tesfamariam, S. (2013), “A survey of non-gradient optimization methods in
structural engineering”, Adv. Eng. Software, 59, 19-28.
Hock, W. and Schittkowski, K. (1983), “A comparative performance evaluation of 27 nonlinear
programming codes”, Computing, 30, 335-358.
Juenger, M., Liebling, T.M., Naddef, D., Nemhauser, G.L., Pulleyblank, W.R., Reinelt, G., Rinaldi, G. and
Wolsey, L.A. (Eds. 2010), 50 Years of Integer Programming 1958-2008, Springer-Verlag, Berlin
Heidelberg, Germany.
Lasdon, L.S. and Waren, A.D. (1978), “Generalized Reduced Gradient Software for Linearly and
Nonlinearly Constrained Problems”, (In H.J. Greenberg, ed.), Design and Implementation of Optimization
Software, Sitjhoff and Noordhoff.
Lee, C. and Ahn, J. (2003), “Flexural design of reinforced concrete frames by genetic algorithm”, J. Struct.
Eng., 129(6), 762-774.
Lepš, M. and Šejnoha, M. (2003), “New approach to optimization of reinforced concrete beams”, Comput.
Struct., 81(18-19), 1957-1966.
Linderoth, J.T. and Savelesbergh, M.W.P. (1999), “A computational study of search strategies for mixed
integer programming”, INFORMS J. Computing, 11(2), 173-187.
McCluskey, S. and McCarthy, T.J. (2009), “A particle swarm optimisation approach to reinforced concrete
beam design according to AS3600”, First International Conference on Soft Computing Technology in
Civil, Structural and Environmental Engineering, Civil-Comp Press, Madeira, Portugal, pp. 1-14.
Medeiros, G.F. and Kripka, M. (2013), “Structural optimization and proposition of pre-sizing parameters for
beams in reinforced concrete buildings”, Comput. Concrete, 11(3), 253-270.
Michell, A.G.M. (1904), “The Limits of Economy of Materials in Frame Structures”, Philosophical
Magazine Series 6, 8(47), 589-597.
Narayan, K.S.B. and Venkataramana, K. (2007), “Shape optimization of steel reinforced concrete beams”,
Comput. Concrete, 4(4), 317-330.
Nemhauser, G.L. and Wolsey, L.A. (1999), Integer and Combinatorial Optimization, John Wiley and Sons
Inc., New York, USA.
Nocedal, J. and Wright, S. (2006), Numerical Optimization, Springer, New York, USA.
Norman, D.G. (1964), “Economic aspects in the design of some reinforced concrete structural members”, J.
Am. Concrete Inst., 61(4), 419-440.
Olsson, A.E. (2011), Particle Swarm Optimization: Theory, Techniques and Applications, Nova Science
Pub. Inc.
Ozturk, H.T., Durmus, Ay. and Durmus, Ah. (2013), “Optimum design of a reinforced concrete beam using
artificial bee colony algorithm”, Comput. Concrete, 10(3), 295-306.
Powell, M.J.D. (1964), “An efficient method of minimizing a function of several variables without
calculating derivatives”, Comput. J., 7(2), 155-162.
Prakash, A., Agarwala, S.K. and Singh, K.K. (1988), “Optimum design of reinforced sections”, Comput.
Struct., 30(4), 1009-1011.
Rajeev, S. and Krishnamoorthy, C.S. (1998), “Genetic algorithm-based methodology for design optimization
of reinforced concrete frames”, Comput. Aided Civil Infrastruct. Eng., 13(1), 63-74.
Revelle, C.S., Whitlatch, E.E. and Wright, J.R. (2004), Civil and Environmental Systems Engineering, (2nd
480 Ima Rahmanian, Yves Lucet and Solomon Tesfamariam

ed., W.J. Hall, ed.), Pearson Education Inc., New Jersey, USA.
Robinson, S.M. (1972), “A quadratically-convergent algorithm for general nonlinear programming
problems”, Math. Program., 3(1), 145-156.
Salamon, P., Sibani, P. and Frost, R. (2002), Facts, Conjectures, and Improvements for Simulated
Annealing, Society for Industrial and Applied Mathematics (SIAM).
Sandhu, B.S. (1971), “Economical design of reinforced concrete slabs and walls”, Civil Eng. ASCE, 41(7),
68-69.
Sivanandam, S.N. and Deepa, S.N. (2008), Introduction to Genetic Algorithms, Springer, Berlin Heidelberg,
Germany.
Structural Engineering Institute Technical Committee on Optimal Structural Design (2002), Recent
Advances in Optimal Structural Design, (S.A. Burns, ed.), American Society of Civil Engineers.
Veaux, R.D.De, Velleman, P.F., Bock, D.E., Vukov, A.M. and Wong, A.C.M. (2012), Stats: Data and
Models, Pearson Canada Inc., Toronto, Ontario, Canada.
 Optimal design of reinforced concrete beams: A review 481

Appendix

Optimization constraints for a singly reinforced rectangular beam can be expressed as the
following.
1. Bending moment resistance: 𝑀𝑓 + 𝑀𝑠𝑤 ≤ 𝑀𝑟
𝑀𝑓 Factored design bending moment (constant user-defined value)
𝑀𝑠𝑤 Bending moment due to self-weight of the beam:𝑀 = 𝐷𝑤 𝑐 𝑏𝑕𝐿2
𝑠𝑤 8
𝑤𝑐 Unit weight force of concrete (constant user-defined value)
𝐷 Load factor for dead load (constant code parameter)
𝐿 Span of beam (constant user-defined value)
𝑀𝑟 𝑎
The factored bending moment resistance: 𝑀𝑟 = 𝜑𝑠 𝑓𝑦 𝐴𝑠 𝑑 − 2
𝜑𝑠 Resistance factor for steel (constant code parameter)
𝑓𝑦 Yield strength of steel reinforcement (constant user-defined value)
𝑑 𝑑
Effective depth of beam:𝑑 = 𝑕 − 𝑐 − 2𝑏 − 𝑑𝑠
𝑐 Concrete cover provided (constant user-defined value)
𝑑𝑏 Diameter of the bending reinforcement bars (constant user-defined value)
𝑑𝑠 Depth of the equivalent rectangular stress block: 𝑎 = 𝜑𝑠 𝑓𝑦 𝐴𝑠 /𝛼1 𝜑𝑐 𝑓𝑐′ 𝑏
𝑎 Depth of the equivalent rectangular stress block: 𝑎 = 𝜑𝑠 𝑓𝑦 𝐴𝑠 /𝛼1 𝜑𝑐 𝑓𝑐′ 𝑏
𝑓𝑐′ Compressive strength of concrete (constant user-defined value)
𝜑𝑐 Resistance factor for concrete (constant code parameter)
𝛼1 , 𝛽1 Equivalent rectangular stress distribution parameters (code prescribed constant
based on the compressive strength of concrete)
0.2 𝑓𝑐′
2. Minimum reinforcement: 𝐴𝑠 ≥ 𝑏𝑕
𝑓𝑦
0.75× 700 𝛼 1 𝛽 1 𝜑 𝑐 𝑓𝑐′
3. Maximum reinforcement: 𝐴𝑠 ≤ 700+𝑓𝑦 𝜑 𝑠 𝑓𝑦
𝑏𝑑
4. Maximum factored shear resistance: 0.25𝜑𝑐 𝑓′𝑐 𝑏𝑑𝑣 ≥ 𝑉𝑓
𝑉𝑓 Factored design shear (constant user-defined value)
𝑑𝑣 Effective shear depth: 𝑀𝑎𝑥[0.9𝑑, 0.72𝑕]

5. Maximum deflection: ∆𝑖 ≤ ∆𝑚𝑎𝑥

∆𝑚𝑎𝑥 Maximum allowable deflection (code prescribed constant based on beam span)
∆𝑖 5 𝑀𝑎 𝐿2
Immediate deflection: ∆ = 𝑖 48 𝐸𝑐 𝐼𝑒
𝑀𝑎 Unfactored design moment (constant user-defined value)
𝐸𝑐 Modulus of elasticity of concrete (calculated constant): 𝐸𝑐 = 4500 𝑓′𝑐
𝐼𝑒 𝑀𝑐𝑟 3
Effective moment of inertia: 𝐼𝑒 = 𝐼𝑐𝑟 + (𝐼𝑔 − 𝐼𝑐𝑟 ) 𝑀𝑎
𝑀𝑐𝑟 2𝑓𝑟 𝐼𝑔
Cracking moment: 𝑀𝑐𝑟 = 𝑕
𝑓𝑟 Modulus of rupture: 𝑓𝑟 = 0.6 𝑓′𝑐
𝐼𝑔 𝑏𝑕 3
Gross moment of inertia: 𝐼𝑔 = 12
 482 Ima Rahmanian, Yves Lucet and Solomon Tesfamariam

𝐼𝑐𝑟 𝑏𝑦 3
Moment of inertia of the cracked section: 𝐼𝑐𝑟 = 3
+ 𝑛𝐴𝑠 (𝑑 − 𝑦)2
𝑛 𝐸
Modular ratio: 𝑛 = 𝐸𝑠
𝑐
𝐸𝑠 Modulus of elasticity of steel (constant user-defined value)
𝑦 −𝑛𝐴𝑠 + 𝑛𝐴𝑠 2 +2𝑏𝑑 𝑛𝐴𝑠
Neutral axis depth of cracked section: 𝑦 = 𝑏

6. Maximum crack control parameter: 𝑧 ≤ 𝑧𝑢

𝑧𝑢 Upper limit for crack control parameter (constant code parameter)
𝑧 Crack control parameter: 𝑧 = 𝑓𝑠 3 𝑑𝑐 𝐴
𝑑𝑐 The distance from the extreme tension fibre to the centre of the longitudinal bar
𝑑
located closest thereto: 𝑑𝑐 = 𝑀𝑖𝑛 𝑐, 50𝑚𝑚 + 2𝑏 + 𝑑𝑠
𝐴 Effective tension area of concrete surrounding the flexural reinforcement:
𝐴
𝐴= 𝑒
𝑁𝑏
𝐴𝑒 Total effective tension area: 𝐴𝑒 = 𝑏(2𝑑𝑐 ) for one row of tensile bars
𝑓𝑠 Stress in steel at maximum service load: 𝑓𝑠 = 0.6𝑓𝑦

7. Bar spacing constraint: 𝑁𝑙 ≤ 𝑁𝑏 ≤ 𝑁𝑢

𝑁𝑏 Number of bending reinforcement bars
𝑁𝑙 Lower limit for number of reinforcement bars (constant user-defined value)
𝑁𝑢 Upper limit for number of bending reinforcement bars:
𝑏+𝑆𝑚𝑖𝑛 −2𝑐−2𝑑 𝑠
𝑁𝑢 = 𝐼𝑛𝑡 𝑑 +𝑆
𝑏 𝑚𝑖𝑛
𝑆𝑚𝑖𝑛 Minimum clear bar spacing: 𝑆𝑚𝑖𝑛 = 𝑀𝑎𝑥[1.4𝑑𝑏 , 1.4𝑎𝑚𝑎𝑥 , 30𝑚𝑚]
𝑎𝑚𝑎𝑥 Maximum aggregate size (constant input)

8. Beam width constraint: 𝑏𝑙 ≤ 𝑏 ≤ 𝑏𝑢

𝑏𝑙 , 𝑏𝑢 Lower and upper limits for beam width (constant user-defined values)

9. Beam depth constraint: 𝑕𝑙 ≤ 𝑕 ≤ 𝑕𝑢

𝑕𝑙 , 𝑕𝑢 Lower and upper limits for beam depth (constant user-defined values)

10. Beam depth to width ratio constraint: 𝑅𝑙 ≤ 𝑕/𝑏 ≤ 𝑅𝑢

𝑅𝑙 , 𝑅𝑢 Lower and upper limits for beam depth to width ratio (constant user-defined
values)

CC

View publication stats