Uncertainly Analysis of Flexural Overstrength for Capacity

Design of RC Beams
Carmine Galasso 1; Giuseppe Maddaloni 2; and Edoardo Cosenza 3

Abstract: To ensure an overall ductile structural behavior, the reinforcing steel used in the seismic design of reinforced concrete (RC)
Downloaded from ascelibrary.org by CASA Institution Identity on 08/30/23. Copyright ASCE. For personal use only; all rights reserved.

structures is governed by certain specific requirements given in many international codes, such as the Eurocodes and the recently released
Italian Building Code (IBC). This study’s primary focus is the statistical analysis of reinforcing steel properties, based on data from over 600
material tests. The data sets considered include a wide range of reinforcing steel bars (from 12 to 26 mm) provided by different Italian
industries and used for a large structure built in Naples (Southern Italy). The test results are analyzed to determine the appropriate cumulative
distribution function for yield and ultimate strengths, as well as the ultimate deformation, and other statistical parameters of interest as defined
in the codes. The comparison with previous tests confirms an improvement in the quality of materials, as reflected in reduced variability and
increased bias factors, consistent with other investigators’ recent findings. Finally, this study investigates the flexural overstrength of RC
beams designed according to the current IBC—consistent with Eurocodes—and the accuracy of code requirements in light of realistic
material models both for concrete and reinforcing steel and of uncertainties associated with mechanical models, structural members geometry,
and material properties (as obtained in this study for reinforcing steel). The obtained results show that code provisions do not seem
conservative and provide a basis for an improved calibration of future editions of seismic design codes for buildings. DOI: 10.1061/
(ASCE)ST.1943-541X.0001024. © 2014 American Society of Civil Engineers.
Author keywords: Reinforcing steel; Seismic design; Capacity design; Flexural overstrength; Eurocode 8; Structural safety and reliability.

Introduction other undesirable failure mechanisms: for example, concentration

As ductility is an essential property of structures responding inelas-
tically during severe shaking, the design of reinforced concrete
(RC) buildings to resist seismic action should provide the structure
with an adequate capacity to dissipate energy and an overall ductile
behavior.
The seismic design of structures aims to ensure ductility by ap-
propriately dimensioning and detailing regions intended for energy
dissipation (i.e., plastic hinges or critical regions) in response to
seismic action. Moreover, a controlled inelastic response must
be achieved by preventing brittle failure modes (with a certain
safety margin) through the capacity design of structural members:
ductile modes of failure (e.g., flexure) must precede brittle failure
modes (e.g., shear) with sufficient reliability because brittle failure
implies near-complete loss of resistance and the absence of ad-
equate warning. In other words, capacity design rules must be used
effectively to obtain the hierarchy of resistance of the various struc-
tural components and failure modes necessary for ensuring a suit-
able plastic mechanism and for avoiding brittle failure modes or
1
Lecturer, School of Civil Engineering and Geosciences, Newcastle
Univ., NE1 7RU Newcastle upon Tyne, U.K. E-mail: carmine.galasso@
ncl.ac.uk
2 shear deformation do not exhibit characteristics of energy dissipa-
Assistant Professor, Dip. di Ingegneria, Univ. degli Studi di
Napoli Parthenope, 80143 Naples, Italy (corresponding author). E-mail:
giuseppe.maddaloni@uniparthenope.it
3
Full Professor, Dip. di Strutture per l’Ingegneria e l’Architettura, Univ.
degli Studi di Napoli Federico II, 80125 Naples, Italy. E-mail: cosenza@
unina.it
Note. This manuscript was submitted on April 10, 2013; approved on
December 16, 2013; published online on March 25, 2014. Discussion per-
iod open until August 25, 2014; separate discussions must be submitted for
individual papers. This paper is part of the Journal of Structural Engineer-
ing, © ASCE, ISSN 0733-9445/04014037(9)/$25.00.
of plastic hinges in columns of a single story of a multistory build-
ing, shear failure of structural elements, or failure of beam–column
joints.
Ductility in structural members can be developed only if the
constituent materials themselves are ductile. To achieve overall
ductility in a structure, appropriate concrete and steel qualities must
be adopted to ensure local ductility first of cross sections and then
of elements.
Specifically, the ability of reinforcing steel to sustain repeated
load cycles to withstand high levels of plastic strain without sig-
nificant reduction in strength is the prime source of ductility for
RC structural members. In particular, the steel used in critical re-
gions of primary seismic elements must have high uniform ultimate
elongation (i.e., the strain at the peak stress) to ensure a minimum
curvature ductility and flexural deformation capacity; the ultimate
(i.e., the maximum stress) strength-to-yield strength ratio (i.e., the
hardening ratio) must be significantly higher than unity. However,
the steel ultimate strength must not exceed its yield strength by
more than a certain percentage (as clarified in the following); a
low variability of actual yield strength from the specified nominal
value is also a desirable characteristic of reinforcing steel.
The importance of these properties stems from the requirements
of capacity design. For example, because beams under inelastic
tion, shear strength at all sections in critical regions is designed to
be higher than the shear corresponding to the flexural strength at the
chosen plastic hinge locations (Paulay and Priestley 1992). If the
reinforcing steel exhibits early and rapid strain hardening, the steel
stress at a section with high ductility may exceed the yield stress
(typically used as design value) by an excessive margin. Similarly,
if a specific grade of reinforcing steel is subjected to a considerable
variation in yield strength, the actual flexural strength of a plastic
hinge may greatly exceed the nominal specified value or the steel

© ASCE 04014037-1 J. Struct. Eng.

J. Struct. Eng., 2014, 140(7): 04014037

 may not yield before the concrete crushes. All these cases result in a
need to adopt high overstrength factors to protect against shear fail-
ures or unexpected flexural hinging, that is, to prevent loss of con-
trol in the hierarchy of resistance. In other words, the indiscriminate
provision of excess strength, which standard design procedures
usually consider to be positive, may adversely affect the nonlinear
seismic behavior of a structural system (e.g., Magliulo et al.
2007), especially in the case of irregular buildings (e.g., Magliulo
et al. 2012).
On the basis of these preliminary remarks, this study’s first ob-
jective is the statistical analysis of reinforcing steel properties based
on data from over 600 material tests. The data sets considered in-
clude a wide range of reinforcing steel bars (from 12 to 26 mm)
Downloaded from ascelibrary.org by CASA Institution Identity on 08/30/23. Copyright ASCE. For personal use only; all rights reserved.
from a large structure built in Naples (Southern Italy). The test re-
sults are analyzed to determine the appropriate cumulative distri-
bution function (CDF) for yield and ultimate strengths, as well
as ultimate deformation and other statistical parameters that are im-
portant for seismic design of RC structures. This study also inves-
tigates and discusses the potential for satisfying the codes
requirements regarding steel overstrength.
Finally, introducing realistic material models for both concrete
and reinforcing steel and accounting for uncertainties in mechanical
models, structural members geometry, and material properties (as
obtained in this study for reinforcing steel), this study investigates
the probable flexural strength of RC beams designed according to
the current Italian Building Code (IBC08, CS.LL.PP. 2008) and the
accuracy of its requirements for capacity design, showing that code
provisions do not seem conservative.
Code Requirements for Reinforcing Steel
As discussed in the previous section, ductility in RC structures is a
function of the component materials’ inherent ductility and of the
structural members’ configuration and detailing.
Because concrete is inherently a brittle material, ductility—first
in cross sections and then in structural members—strongly depends
on the tensile properties of the reinforcing steel, in terms of ultimate
strain ðεsu Þ, yield ðfy Þ, and ultimate ðft Þ strengths. Fig. 1 shows a
typical experimental stress-strain curve for hot-rolled steel and the
relative parameters of interest; these parameters control the re-
sponse of RC structural elements subjected to earthquake loading.
Further precautions also must be taken when steel is subjected to
compression, to ensure that premature buckling does not interfere
with the development of the desired large inelastic strains in com-
pression. Building codes generally characterize reinforcement
according to its yield strength. In particular, the reinforcement
properties required for use with Eurocode are given in Annex C
of Eurocode 2 (or EC2, European Committee for Standardization
2004a); further requirements for use in seismic design are given
in EC8 (European Committee for Standardization 2004b; e.g., in
Sections 5.4.1.1 and 5.5.1.1). Similarly, IBC08 (Section 11.3.2)
gives the reinforcement properties appropriate for use with the
Italian code.
The codes provide characteristic values of material properties,
noting the maximum percentage of tests results falling below (or
above) the characteristic value for each of the material properties.
According to EC8 guidelines, in critical regions of primary seis-
mic elements, engineers may use reinforcing steel in the form of
ribbed bars of Class B or C with a characteristic (i.e., the 5th per-
centile) yield strength, fyk , between 400 and 600 MPa. The IBC08
allows the use of only B450C steel, that is, Class C reinforcing steel
(hot-rolled steel with high ductility) with a characteristic yield
strength of 450 MPa and an ultimate strength of 540 MPa.
© ASCE 04014037-2
J. Struct. Eng., 2014, 140(7): 04014037
For the seismic elements of new buildings’ lateral-load-resisting
systems in high ductility class (DCH), both EC8 and IBC08 require
Class C reinforcing steel with a characteristic value of εsu , εuk , of at
least 7.5% (the 10th percentile). Moreover, the tensile-to-yield-
strength ratio, f t =fy , should be between 1.15 and 1.35 (the 10th
and 90th percentiles, respectively), and the 95th percentile (90th
in the IBC08) of the actual yield stress should not exceed the nomi-
nal yield strength ðfy;nom Þ (i.e., the manufacturer-guaranteed
characteristic yield strength) by more than 25%.
Significant ductility in a structural member can be achieved only
if inelastic strains can be developed over a reasonable length of that
member; the lower limit on ft =fy aims to avoid the restriction of
inelastic strains to a very small length of structural members: the
higher the value of ft =fy , the longer the zone of plasticization near
the end of a member.
The upper limits on ft =fy and f y =fy;nom aim to control the
flexural overstrength, which results primarily from the strain hard-
ening of reinforcement at high ductility levels and from the actual
yield strength variability above the specified nominal value, as dis-
cussed in the previous section. The assessment of the flexural over
strength ratio of cross sections is an important issue in the capacity
design of structures and will be investigated in the following.
The IBC08 requirements are summarized by the example of
Fig. 2, which reports two possible probability density functions
(PDFs) for the ratios f t =f y and fy =f y;nom . For simplicity, a Gaussian
PDF shape is considered for both PDFs in Fig. 2. The same figure
reports the 10th and the 90th percentiles [Fig. 2(a)] and the 5th and
90th percentiles [Fig. 2(b)]. These values are identified by the areas
under the PDFs to their left (lower percentile) or right (upper per-
centile). The vertical dashed lines represent the code-based range
of acceptance for the two ratios’ characteristic values.
Statistical Analysis
This study investigated ribbed reinforcing bars with diameters from
12 to 26 mm, assuming a 450 MPa grade consistently with IBC08.
The analyzed bars are being used in a large hospital currently under
construction in Naples, the Ospedale del Mare, the largest isolated
seismic structure in Europe (Di Sarno et al. 2011). Its plan is about
150 × 150 m2 , its maximum height is about 30 m, and the structure
has a seismic mass of about 100,000 t. The building’s seismic pro-
tection system consists of 327 circular, isolating rubber bearings,
giving the structure an equivalent period of about 2 s.
The statistical analysis is performed as follows: (1) considering
all the 651 test results together; and (2) dividing the tests results
according to the producer of origin—213 samples for the first
group/factory, 168 samples for the second group, and 270 for
800
600
(ε , f )
su
t
σ [MPa]
400 (ε , f )
y y
200
0
0 0.03 0.06 0.09 0.12 0.15 0.18
ε
Fig. 1. Typical stress-strain curve for steel and parameters considered
in this study
J. Struct. Eng.
 (a) (b)

Fig. 2. IBC08 requirements for reinforcing steel: (a) f t =fy ; (b) f y =f y;nom
Downloaded from ascelibrary.org by CASA Institution Identity on 08/30/23. Copyright ASCE. For personal use only; all rights reserved.

(a) (b)

Fig. 3. Empirical CDFs: (a) fy ; (b) ft

the third group. In fact, data on reinforcing steel used in this study
were obtained from different producers, in different Italian regions,
so they include a so-called batch-to-batch variation, which is com-
parable to the within-batch variation for this study. The investigated
data also include the variation caused by different testing methods
(data come from different labs).
Because no trend is observed in the relationship between the
reinforcing bars strengths and diameters, further subsets of data
(i.e., according to bars diameters) are not considered here. The con-
sidered properties of reinforcing steel have been verified using test-
ing procedures in accordance with EN 10080 (Italian Organization
for Standardization 2005). In particular, the tensile tests were per-
formed using universal testing machines and considering three
samples of 60 cm in length for each bar, according to EN 10080.
The empirical CDFs for f y and ft and for each data set as well as
for all the data considered together are plotted in Fig. 3; the stat-
istical parameters [mean value, coefficient of variation (CoV),
skewness, the 5th percentile (x0.05 )] for fy and ft are summarized
in Tables 1 and 2, respectively.
Both the characteristic values (x0.05 ) and the CoVs (i.e., the ratio
of the standard deviation to the mean) of fy and ft obtained from
test data are rather uniform. The characteristic values are much
larger than the nominal values (equal to 450 and 540 MPa for
f y and f t , respectively). In addition, the CoVs for the different
groups assume a very low value, confirming the high level of stand-
ardization achieved today in the manufacturing process for
reinforcing steel bars.
The bias factor for fy (i.e., the ratio between the mean of the
sample to the reported nominal value) for all analyzed reinforcing
bars is 1.22, assuming a nominal value of 450 MPa while the CoV
is 0.054. For comparison, the bias factor for fy used in previous
studies was 1.125 and CoV 0.10 (Ellingwood et al. 1982): an im-
provement in the quality of the materials is reflected in reduced
variability and increased bias factors. This conclusion is consistent
with the recent findings of other investigators in other countries
[e.g., Bartlett and MacGregor (1996), for concrete strength; Nowak
and Szerszen (2003), for both concrete and steel strength]. Then,
although the limitations of reinforcement in Italy seem to restrict
the applicability of the current study’s results, data in other coun-
tries may behave similarly. However, more compressive tests are
required to generalize the conclusions here. Finally, it is worth not-
ing that the examined samples are characterized by values of the
skewness coefficient equal to about zero; this property indicates
that models (i.e., probability distributions) characterized by a sym-
metric PDF (e.g., the normal model) may be suitable for fitting the
empirical distributions resulting from the data, as discussed in the
following.

Table 1. Statistical Parameters for f y Table 2. Statistical Parameters for f t

Sample Mean (MPa) CoV Skewness x0.05 (MPa) Sample Mean (MPa) CoV Skewness x0.05 (MPa)
Group number 1 550.3 0.049 0.229 503.9 Group number 1 659.7 0.050 0.330 606.5
Group number 2 549.5 0.062 0.092 500.1 Group number 2 659.3 0.058 −0.008 593.6
Group number 3 547.5 0.054 0.182 499.0 Group number 3 658.1 0.045 0.562 613.1
All 548.9 0.054 0.158 501.7 All 658.9 0.050 0.278 606.1

© ASCE 04014037-3 J. Struct. Eng.

J. Struct. Eng., 2014, 140(7): 04014037

 1
Group no. 1 Table 4. Statistical Parameters for f t =fy
0.9 Group no. 2
0.8 Group no. 3
Sample Mean CoV x0.10 x0.90
0.7 All data Group number 1 1.20 0.026 1.16 1.24
Empirical CDF
Group number 2 1.20 0.026 1.16 1.24
0.6
Group number 3 1.20 0.029 1.16 1.24
0.5
All 1.20 0.027 1.16 1.24
0.4
0.3
0.2
0.1 linear dependence between these two variables, as confirmed by
0 the value of the correlation coefficient, to follow. As is clear from
10 15 20 25 30 35 40 45 50
ε Table 4, both codes requirements for the lower limit (10th percentile,
su
x0.10 , required for ductility) and the upper limit (90th percentile, x0.90 ,
Fig. 4. Empirical CDFs for εsu required for the hierarchy of resistance) on ft =fy are satisfied.
Downloaded from ascelibrary.org by CASA Institution Identity on 08/30/23. Copyright ASCE. For personal use only; all rights reserved.

The empirical CDFs for fy =fy;nom are plotted in Fig. 5(b),

The empirical CDFs for εsu and for each data set, as well as for
all the data considered together, are plotted in Fig. 4; the statistical
parameters (mean value, CoV, skewness, the 10th percentile) for εsu
are summarized in Table 3.
The 10th percentile, x0.10 , of the actual experimental ultimate
strain is generally equal to about 22% (i.e., about three times
the value required by IBC08), leading to the design of cross sec-
tions that are always characterized by concrete crushing (after the
steel yields) and justifying the use of a simplified elastoideal plastic
stress-strain diagram for the reinforcing steel, with a horizontal top
branch without a strain limit, as recommended by modern codes
(e.g., EC2 and IBC08). This idealization strongly simplifies the
flexural analysis of RC members (e.g., Cosenza et al. 2011).
Further, the very high values characterizing the statistics for εsu
have a highly positive impact on RC members’ ultimate deforma-
tion and subsequently on their ductility.
The empirical CDFs for f t =f y are plotted in Fig. 5(a); the differ-
ent empirical distributions for the different groups appear nearly
identical. This visual evidence is confirmed by the statistical
parameters reported in Table 4.
The CoV of f t =f y is very low, and for all groups, equal to 0.03
(i.e., about the half of the CoV values for fy and f t ; Tables 1 and 2).
The fact that the CoV for ft =f y is very small indicates a strong
whose empirical distributions clearly correspond to those
in Fig. 3(a), except for a scale factor (in this case equal to
450 MPa). The statistical parameters for f y =f y;nom are established
based on the empirical CDFs and are listed in Table 5.
As discussed in the previous sections, codes impose an upper
limit equal to 1.25 (the 90th percentile in the IBC08) for the actual
to nominal yield strength ratio, with the latter being equal to
450 MPa according to the IBC08 for the B450C steel. As shown
in Table 6, the code requirement has never been satisfied for any
group (EC8 refers to the 95th percentile; obviously, in this case as
well, the code requirement is not satisfied), and not even consid-
ering all 651 tests together. In other words, the value of the nominal
yield strength prescribed in the IBC08 seems to be too conservative
to satisfy the code requirement in terms of the upper limit for
fy =f y;nom . Considering the empirical distribution of fy [Fig. 3(a)]
for all 651 tests and with reference to the 5th and 90th percentiles
from this distribution, equal to 502 and 590 MPa, respectively, it is
possible to search for a range of values of fy;nom that will satisfy the
code requirements, per Eq. (1)
 
fy 502
> 1 ⇒ f y;nom < ¼ 502 MPa
f y;nom 0.05 1
 
fy 594
≤ 1.25 ⇒ fy;nom ≥ ¼ 475 MPa ð1Þ
f y;nom 0.90 1.25

Table 3. Statistical Parameters for εsu Table 5. Statistical Parameters for f y =f y;nom ðf y;nom ¼ 450 MPaÞ
Sample Mean CoV Skewness x0.10 Sample Mean CoV x0.05 x0.90
Group number 1 26.0 0.126 0.080 22.1 Group number 1 1.22 0.049 1.12 1.31
Group number 2 25.7 0.147 0.579 21.2 Group number 2 1.22 0.062 1.11 1.32
Group number 3 26.0 0.129 0.263 22.3 Group number 3 1.22 0.054 1.10 1.31
All 25.9 0.132 0.304 21.9 All 1.22 0.054 1.12 1.32

(a) (b)

Fig. 5. Empirical CDFs: (a) f t =fy ; (b) f y =fy;nom

© ASCE 04014037-4 J. Struct. Eng.

J. Struct. Eng., 2014, 140(7): 04014037

 Therefore, to fulfill the two code requirements, the yield
strength nominal value should be in the range of 475–502 MPa.
For example, a future version of the Italian code should refer to
ribbed reinforcing bars of 500 MPa grade.
The values of the correlation coefficient between f y and f t ; fy
and εsu , f t and εsu , and ft =fy and f y were also computed. The
correlation coefficient measures the strength of the linear relation-
ship between two variables. In particular, fy and f t are highly cor-
related with a correlation coefficient equal to 0.9; this result
confirms the assumption in Kappos et al. (1999) and Aydemir
and Zorbozan (2012). The correlation coefficient between f y and
εsu and ft and εsu is 0.1 (i.e., the correlation is negligible and the
variables can be considered uncorrelated) in both cases, while the
Downloaded from ascelibrary.org by CASA Institution Identity on 08/30/23. Copyright ASCE. For personal use only; all rights reserved.
was the basis for EC8 current provisions on steel reinforcing prop-
erties. For comparison, this study’s results are reported in the last
column of Table 6. In Table 6, the values in bold violate the
corresponding limit for the steel to be used in DCH buildings.
As a general comment, the values in Table 6 confirm a remarkable
improvement in quality of materials and reduced variability in
strength.
Selection of Probability Distribution
In the context of probability-based limit-state design and assess-
ment, an important step is to identify an appropriate probability
distribution, that is, a theoretical probability model characterizing

correlation coefficient between f t =f y and fy is equal to −0.42
(negative correlation; i.e., one variable increases as the other de-
creases). Fig. 6 shows the scatter plots for each pair of variables
being considered, providing a visual check of the degree of corre-
lation between the two considered variables in each panel.
Finally, Table 6 reports the statistical outcome of the widest
survey of ductile steels of the type used in European seismic re-
gions (Fardis 2009). That survey, carried out in the early 1990s,
the material properties from test data. Although the task of selecting
a probability distribution fitting the experimental data does not
seem directly relevant in the context of this paper and does not prac-
tically affect the seismic design of RC members, it represents the
basis for the probabilistic assessment presented in the next section
and for similar structural reliability studies to evaluate structural
safety associated with the design procedures (e.g., Iervolino and
Galasso 2012). In fact, the model chosen to represent the material

Table 6. Outcome of Surveys of Steel Used in Seismic Regions of Europe (Adapted from Fardis 2009)
Belgium. France, Germany,
Italy, Luxembourg, Netherlands,
Parameter Spain, Portugal Italy Portugal, Spain U.K. This study
f y;nom (MPa) 400 430 500 500 450
Mean yield strength (MPa) 496 478 571 552 549
ðf y =f y;nom Þ0.95 1.335 1.19 1.23 1.165 1.36a
Mean ultimate strength (MPa) 598 733 663 653 659
ðf t =f y Þ0.10 1.15 1.44 1.10 1.13 1.16
ðf t =f y Þ0.90 1.27 1.62 1.23 1.23 1.24
ðεsu Þ0.10 9.6 9.7 8.6 9.7 21.9
Note: The values in bold violate the corresponding limit for the steel to be used in DCH buildings.
a
90% Fractile according to IBC08.

(a) (b)

(c) (d)

Fig. 6. Scatter plots: (a) ft versus fy ; (b) εsu versus f y ; (c) εsu versus f t ; (d) f t =fy versus f y

© ASCE 04014037-5 J. Struct. Eng.

J. Struct. Eng., 2014, 140(7): 04014037

 property data has a large effect on the calculation of a structural
component safety.
In response to the small data samples that typically are available,
the common approach in structural engineering has been to simply
prescribe a distribution. For the development of the RC load and
resistance-factor-design specifications, a lognormal distribution
was typically used for steel properties (Ellingwood et al. 1982;
Galambos et al. 1982). Similarly, the normal distribution has been
typically used to characterize the concrete properties (MacGregor
et al. 1983), although more elaborate models have been proposed,
for example, Tumidajski et al. (2006). These distributions were
selected to best fit the test data available at the time the studies were
conducted.
Downloaded from ascelibrary.org by CASA Institution Identity on 08/30/23. Copyright ASCE. For personal use only; all rights reserved.
This section examines the normal and lognormal probability
models, with the goal to obtain a more adequate representation
of the actual probability distribution of steel strengths (in terms
of both f y and ft ) and deformation (εsu ), in light of the new data.
The normal model has been selected because it is the one most
widely used to represent random variables that are affected by dif-
ferent factors. According to the central-limit theorem (e.g., Mood
et al. 1974), under certain conditions the sum of a number of ran-
dom variables with finite means and variances approaches a normal
distribution as the number of variables increases. For this reason,
the normal distribution is commonly encountered in practice, and is
used as a simple model for complex phenomena. Many building
codes (e.g., the IBC08) refer to this distribution. The lognormal
distribution has been selected because of the positivity of its values.
It is especially useful in applications in which the values of the ran-
dom variable are known, from physical consideration, to be strictly
positive, as in the case of the strength of materials.
Various methods exist for determining the goodness of fit of dif-
ferent probability models to a set of data. In the present study, the
widely used Anderson–Darling (A–D) test has been adopted
(Anderson and Darling 1954). In particular, probability distribu-
tions that are essentially similar in the central regions of the data
can have vastly different behavior in the lower tail region, which is
significant for structural safety. As discussed in other similar stud-
ies (e.g., Zureick et al. 2006, for composite materials), the A–D test
statistic is particularly sensitive to discrepancies in the tail region
and can be effectively used to examine the experimental data avail-
able. Table 7 gives results of the A–D test in terms of the probability
associated with the A–D statistic (p-values) for each variable and
model. With respect to both f y and ft , neither the normal nor the
lognormal distributions can be rejected with a given significance
level of 0.05, as is expected for high-mean and low-variance ran-
dom variables; however, the lognormal distribution is slightly
favored (i.e., larger p-values are observed) by the considered data
sets. A similar finding is observed for εsu, although the normal
distribution is rejected in this case (p-value in bold).
Estimation of Flexural Overstrength of
IBC08-Designed RC Beams
The proportioning and detailing requirements for buildings in seis-
mic zones are intended to ensure that inelastic response is ductile.
Table 7. Results of A–D Test
Parameter p-value (normal) p-value (lognormal)
fy 0.2536 0.6283
ft 0.1600 0.5037
εsu 0.0047 0.0973
© ASCE 04014037-6
J. Struct. Eng., 2014, 140(7): 04014037
As discussed earlier, current seismic design pursues the control of
inelastic seismic response through capacity design to achieve a
strong-column/weak-beam design that spreads inelastic response
over several stories and to avoid relatively brittle shear failure,
in both beams and columns. To achieve this aim, international
building codes require that the sum of the columns strengths exceed
the sum of beams strengths at all joints between a frame’s primary
or secondary seismic beams and primary seismic columns and in
the two orthogonal directions, with an amplification (overstrength)
factor, γ Rd , equal to 1.3 for DCH (for both EC8 and IBC08), ap-
plied to the design values of the moments of resistance of the beams
framing the joints. Similarly, shear failure in beams is avoided by
calculating the design shear forces based on the beam equilibrium
under the transverse load acting on it (in a seismic design situation)
and the moments of resistance at the beam ends, again with an over-
strength factor, γ Rd , equal to 1.2 for DCH. In practice, due to in-
herent uncertainties in material properties, geometrical dimensions,
and the equations used to compute member strengths, the actual
moment of resistance of a RC structural member, MR , differs from
its design (i.e., nominal) flexural capacity, MRd , which is calculated
based on nominal values. For that reason, the basis for designing
structural members following the capacity design procedure is to
accurately assess the beams flexural overstrength in terms of
M R =M Rd ratio at an appropriate upper fractile. The estimation
of RC beams flexural overstrength (and the calibration of
overstrength factors) must necessarily be expressed in probabilistic
terms because most, if not all, factors possibly affecting the
moment of resistance are uncertain despite the values assumed
in design, as widely discussed in the previous sections in the case
of reinforcing steel. Currently, calibration of the overstrength
factors used in codes seems to be based mainly on engineering
judgment rather than scientifically sound assessments (Nofal et al.
2013).
To assess the probability distribution of flexural overstrength for
IBC08-designed RC beams, 4,320 cross sections are analyzed, rep-
resenting combinations of variations in the following:
• Concrete geometry: This study considers rectangular cross sec-
tions obtained by varying the concrete width between 20 and
40 cm and the concrete depth between 50 and 85 cm; width
and depth are samples with a step of 5 cm.
• Geometric reinforcement ratio in tension (ρ): This parameter is
varied between 0.3% and 2% (with a 0.1% step).
• Geometric reinforcement ratio in compression (ρ 0 ): Three dif-
ferent values are considered for this parameter, that is, equal
to 50% and 75% of the reinforcement ratio in tension, and
the case of symmetric reinforcement (ρ ¼ ρ 0 ).
• Design stress-strain diagrams for reinforcing steel in computing
M Rd : Two bilinear stress-strain relationships for reinforcing
steel (according to EC8 and IBC08) are used, namely, (1) with
a horizontal top branch without a strain limit (elastoideal plastic)
and (2) with an inclined, linear, top branch with a strain limit,
εud , of 6.75% (recommended value in EC8 and IBC08) and
hardening ratio, k, equal to 1.35 (maximum allowable value
based on the codes requirement, as widely discussed in this
paper); see Fig. 7.
In computing M Rd concrete and steel are characterized by a
characteristic compressive cylinder strength, fck , of 25 MPa and
a characteristic yield strength, f yk , of 450 MPa (i.e., B450C type),
respectively; following IBC08, partial safety factors of γ c ¼ 1.5
and γ s ¼ 1.15 for concrete and steel, respectively, are used. In this
way, the assessment is general and covers a large number of real-
istic design conditions that reflect IBC08 (and EC8) provisions
(e.g., Kappos 1997; Magliulo et al. 2007, 2012; Maddaloni et al.
2012).
J. Struct. Eng.
 and equal to 0.99 for the effective depth with a CoV of 4%. The
(εuk , k fy /γ s) area of reinforcing steel is treated as a practically determinist value
(bias factor equal to 1 with a CoV of 1%). All cases assume a
i) normal model. The thickness of concrete cover is assumed to be
σ [MPa] deterministic and equal to 4 cm.
ii)
(fyd/Es, f yd = f /γ s) Mechanical Models
yk
Model (or professional; Nowak and Szerszen 2003) uncertainties
characterize the heterogeneity in sectional capacity estimation,
which is caused by design equations. In fact, such uncertainties
are generally measured by comparing the flexural capacity obtained
ε in experimental tests with the corresponding values obtained by
analytical formulations. Models’ statistical properties are compre-
Fig. 7. Design stress-strain diagrams for reinforcing steel used in com-
hensively documented in Ellingwood et al. (1980). The normal dis-
Downloaded from ascelibrary.org by CASA Institution Identity on 08/30/23. Copyright ASCE. For personal use only; all rights reserved.

puting M Rd
Uncertainty Characterization
A Monte Carlo sampling procedure is applied to accomplish the
overstrength assessment. For steel properties, the results of the stat-
istical analysis discussed in this study are used in the Monte Carlo
simulation. Representative statistics and appropriate probability
distributions for the other basic resistance variables are selected
from previous related studies. In particular, a literature review
(Ellingwood et al. 1980; Galambos et al. 1982; Nowak and Szerszen
2003) was carried out to select the statistical characterization for
each random variable referring to materials (i.e., concrete strength),
geometry (i.e., cross-sectional dimensions and reinforcement area),
and models. The resulting assumptions, corresponding to average-
quality construction, are summarized in Table 8 and described in the
following subsections; the parameters given in Table 8 are the bias
and the CoV. All random variables considered were treated as sto-
chastically independent, except for the reinforcing steel properties,
for which the aforementioned correlation structure is used in the
simulation. In particular, f y and ft are assumed to be fully corre-
lated, as in Kappos et al. (1999), assuming an intercorrelated multi-
variate lognormal distribution.
Materials
The uncertainty involving concrete properties is modeled by as-
suming a normal distribution for the ultimate compressive cylinder
strength; the bias factor is assumed to be equal to 1.35 with a CoV
of 18%. The ultimate strain of concrete is assumed to be determin-
istic and equal to 0.0035. Concrete tensile strength is not consid-
ered, as its effect is negligible.
Sectional Geometry
Uncertainties in geometry (or fabrication; Nowak and Szerszen
2003) account for the heterogeneity in the dimensions of the con-
sidered structural element due to construction quality. The consid-
ered statistical parameters are based on Ellingwood et al. (1980). In
particular, for concrete beams’ dimensions in bending, the bias fac-
tor is assumed to be equal to 1.01 for the width with a CoV of 4%,
Table 8. Summary of Resistance Statistics and Distributions
Category Variable Bias CoV
Material Concrete compressive 1.35 0.18
cylinder strength
Geometry Width of beam 1.10 0.04
Effective depth of beam 0.99 0.04
Reinforcement area 1.00 0.01
Model Experimental/theoretical 1.02 0.06
flexural capacity
tribution is typically used to represent these modeling factors,
whose mean and CoV depend on the limit state considered. In par-
ticular, the mean value for the ratio of the test to predicted flexural
strength for RC beams is 1.02 with a 6% CoV.
Methodology
The probability distributions and statistics for M R are determined
using a Monte Carlo sampling procedure, using the uncertainty
characterization discussed earlier. To achieve this aim, the authors
developed an ad hoc computer script.
In particular, for each case-study cross section, defined by a set
of nominal material strengths and nominal dimensions (and the se-
lected steel constitutive model), the following steps are carried out:
• M Rd is computed based on the design material strengths, nom-
inal dimensions, and chosen design stress-strain diagrams for
reinforcing steel.
• Given the cross-sectional nominal characteristics, a set of
material strengths and dimensions is generated randomly from
the statistical distributions of each variable that affects M R as
discussed earlier. This set of strength, etc., plus a randomly gen-
erated value of the model error, is used to estimate the cross-
sectional theoretical capacity, M R . The M R is computed based
on strain compatibility, equilibrium among internal forces, and
the controlling mode of failure (i.e., concrete crushing). A bi-
linear stress-strain relationship with an inclined, linear-top
branch with a strain limit εsu , at the ultimate strength ft , is used
for reinforcing steel (for both tension and compression); a para-
bola–rectangle diagram is used for concrete under compression
(see IBC08 and EC2 for details).
• The overstrength ratio, M R =MRd , is finally calculated. This pro-
cedure is repeated 5,000 times, enabling the probability distri-
bution of M R =M Rd to be determined numerically. The mean and
the 10% and 90% fractiles from this distribution are then
evaluated.
Although the selected steel constitutive model used in the afore-
mentioned second step may look simplistic, it is fully based on the
steel properties available from the test data; a more sophisticated
model would require additional experimental parameters not avail-
able here and would not significantly improve computational
accuracy.
As in Aydemir and Zorbozan (2012), Monte Carlo simulation is
used to conduct sensitivity analyses to investigate the effect of the
Distribution sample size on the probability distribution of M R =M Rd (particularly
Normal the 90% fractile). This part of the study used different sample sizes
ranging between 500 and 10,000 for randomly generated values of
Normal the considered random variables. As an example, Fig. 8 shows the
Normal overstrength ratios (90% fractile) for two selected cross sections
Normal (40 cm × 60 cm with ρ ¼ 0.3% and 2% and ρ 0 ¼ 0.5ρ). In particu-
Normal lar, Fig. 8 shows that the overstrength ratio (90% fractile) is stable
and does not change significantly for larger sample sizes. Thus, a

© ASCE 04014037-7 J. Struct. Eng.

J. Struct. Eng., 2014, 140(7): 04014037

 2.5 1
90% fractile, ρ = 0.3% (ρ’ = 0.5ρ)
0.9
90% fractile, ρ = 2% (ρ’ = 0.5ρ )
0.8
2
0.7

M /MRd
0.6

CDF
R 1.5 0.5
0.4
1 0.3
0.2 f = 450 MPa
y,nom
0.1 fy,nom = 500 MPa
0.5
500 3,000 5,500 8,000 10,000 0
Number of simulations 1.4 1.45 1.5 1.55 1.6 1.65 1.7 1.75 1.8
M /M
R Rd
Fig. 8. Effect of sample size on the beams overstrength ratios at the
Fig. 10. Cumulative probability of M R =MRd ratio (90% fractile) for all
Downloaded from ascelibrary.org by CASA Institution Identity on 08/30/23. Copyright ASCE. For personal use only; all rights reserved.

upper 10% fractile

the considered case-study cross sections

statistical assessment was performed for an arbitrary sample size of

5,000 for the remaining part of the study, as no differences in com-
putational effort are observed for smaller sample sizes.
Results and Discussion
As an example, Fig. 9 shows the M R =MRd distributions as a func-
tion of the level of tensile reinforcement (ρ) for a selected geometry
(40 cm × 60 cm) with ρ 0 ¼ 0.5ρ and ρ 0 ¼ ρ; all the randomly gen-
erated samples are plotted (gray circles). The lower 10% fractile,
mean, and upper 10% fractile of M R =M Rd ratios, obtained from all
generated samples, are shown in the same figure (see legend)
together with the IBC08 overstrength factor value (=1.2, black
dashed line). The upper panels refer to the elastoideal plastic
stress-strain curve used in computing M Rd , while the lower
panels refer to the bilinear stress-strain relationship with hardening
used in computing M Rd . As expected, the use of a more realistic
stress-strain relationship results in lower values of the flexural
overstrength.
Fig. 9 demonstrates that randomness in materials, section prop-
erties, and model significantly affects the M R =M Rd ratio. The over-
strength factor suggested by IBC08 (and EC8), equal to 1.2, is
always lower than mean M R =M Rd ratio of the sample beams for
each tensile reinforcement ratio and for the two considered values
of the compression reinforcement ratio.
Therefore, the code provisions do not seem to be very accurate
and conservative, although a constant increase in the design flexu-
ral strength for the calculation of probable flexural strength of
RC beams is a simple and practical way to account for the flexural
overstrength. This finding is consistent with the results of similar
studies (e.g., Aydemir and Zorbozan 2012; Somja et al. 2013).
To summarize the results obtained for all the considered case-
study cross sections and to suggest an improvement to the code
provisions, Fig. 10 shows the obtained CDFs for the 90% fractile

(a) (b)

(c) (d)

Fig. 9. Flexural overstrength in terms of MR =M Rd ratios for sample cross sections: (a) elastoideal plastic diagram for steel (in MRd computation) and
ρ 0 ¼ 0.50ρ; (b) elastoideal plastic diagram for steel (in M Rd computation) and ρ 0 ¼ ρ; (c) bilinear with hardening diagram for steel (in M Rd computa-
tion) and ρ 0 ¼ 0.50ρ; (d) bilinear with hardening diagram for steel (in M Rd computation) and ρ 0 ¼ ρ

© ASCE 04014037-8 J. Struct. Eng.

J. Struct. Eng., 2014, 140(7): 04014037

 of the M R =M Rd ratio (for all cross sections) considering the actual
nominal yield strength prescribed by IBC08 (i.e., 450 MPa) and the
value suggested earlier in this study (i.e., 500 MPa). The median
overstrength ratios extracted from these curves (i.e., at a cumulative
probability of 50%) are equal to 1.65 and 1.50, respectively; these
values could be used as overstrength factors in future editions of
the code.
Conclusions
The ability of ductile structures to dissipate energy by postelastic
deformations may be the most important factor in avoiding collapse
during major earthquakes. The fundamental source of ductility is
Downloaded from ascelibrary.org by CASA Institution Identity on 08/30/23. Copyright ASCE. For personal use only; all rights reserved.
the ability of constituent materials to sustain plastic strains without
significant reduction of strength. Both the strength and ductility of
RC structures depend to a large extent on certain proprieties of
reinforcing bars, properties controlled in practice by code specifi-
cations.
This paper reports the results of a statistical analysis of all the
relevant reinforcing steel properties of interest in the seismic design
of RC structures. A large database (of over 600 data sets) on the
properties of reinforcing steel was gathered. The considered data
sets include a wide range of reinforcing steel bars (from 12 to
26 mm) used in a large structure being built in Naples (Southern
Italy).
The obtained test results were analyzed to determine the CDFs
and other statistical parameters of interest in the seismic design of
RC structures. In general, the comparison with previous tests con-
firmed that there is an improvement in quality of materials and a
reduction in strength variability, confirming the finding of other
similar studies.
Finally, the probable flexural strength of IBC08-designed RC
beams has been estimated using Monte Carlo simulation and con-
sidering randomness in material, cross-sectional geometry, and
mechanical models. It is concluded that uncertainties in material
properties and cross-sectional behavior significantly affect the
M R =M Rd ratio and that code provisions do not seem accurate
enough to address these effects.
The results obtained provide a basis for the reliability analysis
of the RC components of building structures and for an improved
calibration of future editions of seismic design codes for buildings.
References
Anderson, T. W., and Darling, D. A. (1954). “A test of goodness-of-fit.”
J. Am. Stat. Assoc., 49(268), 765–769.
Aydemir, C., and Zorbozan, M. (2012). “Uncertainty analysis of flexural
overstrength ratio for RC columns.” J. Struct. Eng., 10.1061/(ASCE)
ST.1943-541X.0000528, 1042–1053.
Bartlett, M., and MacGregor, J. G. (1996). “Statistical analysis of the
compressive strength of concrete in structures.” ACI Mater. J., 93(2),
158–168.
Cosenza, E., Galasso, C., and Maddaloni, G. (2011). “A simplified method
for flexural capacity assessment of circular RC cross sections.” Eng.
Struct., 33(3), 942–946.
CS.LL.PP. (2008). “DM 14 Gennaio 2008: Norme tecniche per le costru-
zioni.” Gazzetta Ufficiale della Repubblica Italiana, 29 (in Italian).
Di Sarno, L., Chioccarelli, E., and Cosenza, E. (2011). “Seismic response
analysis of an irregular base isolated building.” Bull. Earthquake Eng.,
9(5), 1673–1702.
© ASCE 04014037-9
J. Struct. Eng., 2014, 140(7): 04014037
Ellingwood, B., Galambos, T. V., MacGregor, J. G., and Cornell, C. A.
(1980). “Development of a probability based load criterion for
American National Standard A58: Building code requirements for
minimum design loads in buildings and other structures.” U.S. Dept. of
Commerce, National Bureau of Standards, Washington, DC.
Ellingwood, B. R., MacGregor, J. G., Galambos, T. V., and Cornell, C. A.
(1982). “Probability based load criteria: Load factors and load combi-
nation.” J. Struct. Div., 108(5), 978–997.
European Committee for Standardization. (2004a). “Eurocode 2: Design of
concrete structures, Part 1.1: General rules and rules for buildings.”
Brussels.
European Committee for Standardization. (2004b). “Eurocode 8: Design
provisions for earthquake resistance of structures, Part 1.1: General
rules, seismic actions and rules for buildings.” Brussels.
Fardis, M. N. (2009). Seismic design, assessment and retrofitting of con-
crete buildings: Based on EN-Eurocode 8, Springer, Netherlands.
Galambos, T. V., Ellingwood, B. R., MacGregor, J. G., and Cornell, C. A.
(1982). “Probability based load criteria: Assessment of current design
practice.” J. Struct. Div., 108(5), 959–977.
Iervolino, I., and Galasso, C. (2012). “Comparative assessment of load-
resistance factor design for FRP-reinforced cross sections.” Constr.
Build. Mater., 34, 151–161.
Italian Organization for Standardization. (2005). “EN 10080: Acciaio d'ar-
matura per calcestruzzo-Acciaio d'armatura saldabile-Generalità.” (in
Italian).
Kappos, A. J. (1997). “A comparative assessment of R/C structures
designed to the 1995 Eurocode 8 and the 1985 CEB seismic code.”
Struct. Des. Tall Build., 6(1), 59–83.
Kappos, A. J., Chryssanthopoulos, M. C., and Dymiotis, C. (1999).
“Uncertainty analysis of strength and ductility of confined reinforced
concrete members.” Eng. Struct., 21(3), 195–208.
MacGregor, J. G., Mirza, S. A., and Ellingwood, B. R. (1983). “Statistical
analysis of resistance of reinforced and prestressed concrete members.”
ACI Struct. J., 80(3), 167–176.
Maddaloni, G., Magliulo, G., and Cosenza, E. (2012). “Effect of the seis-
mic input on non-linear response of R/C building structures.” Adv.
Struct. Eng., 15(10), 1861–1877.
Magliulo, G., Maddaloni, G., and Cosenza, E. (2007). “Comparison be-
tween non-linear dynamic analysis performed according to EC8 and
elastic and non-linear static analyses.” Eng. Struct., 29(11), 2893–2900.
Magliulo, G., Maddaloni, G., and Cosenza, E. (2012). “Extension of N2
method to plan irregular buildings considering accidental eccentricity.”
Soil Dyn. Earthquake Eng., 43, 69–84.
Mood, M. A., Graybill, F. A., and Boes, D. C. (1974). Introduction to the
theory of statistics, 3rd Ed., McGraw-Hill, New York.
Nofal, S., Somja, H., Hjiaj, M., and Nguyen, Q. H. (2013). “Effects of
material variability on the ductility of composite beams and over-
strength coefficients.” Earthquake Eng. Struct. Dyn., 42(7), 953–972.
Nowak, A. S., and Szerszen, M. M. (2003). “Calibration of design code for
buildings (ACI 318). Part 1: Statistical models for resistance.” ACI
Struct. J., 100(3), 377–382.
Paulay, T., and Priestley, M. J. N. (1992). Seismic design of reinforced con-
crete and masonry buildings, John Wiley and Sons, New York.
Somja, H., Nofal, S., Hjiaj, M., and Degee, H. (2013). “Effect of the steel
material variability on the seismic capacity design of steel-concrete
composite structures: A parametric study.” Bull. Earthquake Eng.,
11(4), 1099–1127.
Tumidajski, P. J., Fiore, L., Khodabocus, T., Lachemi, M., and Pari, R.
(2006). “Comparison of Weibull and normal distributions for concrete
compressive strengths.” Can. J. Civ. Eng., 33(10), 1287–1292.
Zureick, A. H., Bennett, R. M., and Ellingwood, B. R. (2006). “Statistical
characterization of fiber-reinforced polymer composite material proper-
ties for structural design.” J. Struct. Eng., 10.1061/(ASCE)0733-9445
(2006)132:8(1320), 1320–1327.
J. Struct. Eng.