Engineering Structures 116 (2016) 140147

Contents lists available at ScienceDirect

Engineering Structures
journal homepage: www.elsevier.com/locate/engstruct

Assessment of epistemic uncertainties in the shear strength of slender

reinforced concrete beams
A.B. Ribeiro, J.M.F. Calixto , S.M.C. Diniz
Structural Engineering Department, Federal University of Minas Gerais, Belo Horizonte, MG, Brazil

a r t i c l e i n f o a b s t r a c t

Article history: Selection of an appropriate model for shear capacity of reinforced concrete (RC) slender beams poses a
Received 8 June 2015 number of challenges. First, different models with different levels of conservatism have been proposed
Revised 15 February 2016 in an attempt to describe shear resistance. Second, according to Reineck et al. (2014), code provisions
Accepted 24 February 2016
for shear capacity of RC beams with shear reinforcement have been primarily derived from test data with
respect to the required amount of shear reinforcement and the calculation of maximum shear capacity.
Third, current models have been developed based on databases presenting two major drawbacks: (i)
Keywords:
most data points are crowded in the small size range, and (ii) the means of the subsidiary influencing
Beams
Epistemic uncertainties
parameters are very different within different intervals of beam size (or beam depth). In this study, a fil-
Regression analysis tered database is used in such a way to circumvent the drawbacks mentioned above. A random variable
Reinforced concrete model error, i.e. ratio experimental to predicted shear strength, is associated to each of the shear mod-
Shear strength els analyzed in this work (NBR 6118, ACI 318, EUROCODE 2, and CSA A.23.3). It was observed that in some
Shear database cases, most notably for the effective depth, a trend exists for a decrease in the model error as the effec-
tive depth increases. Considering the limitations of the four analyzed models, a nonlinear regression
model was proposed. The database presented by Reineck et al. (2014) was used in the assessment of
the effectiveness and accuracy of the proposed regression model. No trend was found associated to the
most significant variables in the shear strength prediction, i.e. a uniform level of conservatism is attained
throughout the range of these variables. The regression model proposed herein and the attendant statis-
tics of the model error (mean, coefficient of variation and type of distribution) can be easily used in a reli-
ability analysis procedure to assess safety levels implicit in different design procedures.
2016 Elsevier Ltd. All rights reserved.

1. Introduction
The selection of an appropriate model for shear capacity of rein-
forced concrete (RC) slender beams poses a number of challenges.
First, different models, e.g. NBR 6118 [1], ACI 318 [2], EUROCODE 2
[3], CSA A.23.3 [4], have been proposed in an attempt to describe
beam resistance to shear, with different levels of conservatism
for each model [5]. Second, according to Reineck et al. [6], code
provisions for shear capacity of RC beams with shear reinforce-
ment have been primarily derived from test data with respect to
the required amount of shear reinforcement and the calculation
of maximum shear capacity. Third, current models have been
developed based on databases presenting, according to Bazant &
Yu [7], two major drawbacks: (i) most data points are crowded
in the small size range, and (ii) the means of the subsidiary influ-
Corresponding author at: Av. Antonio Carlos 6627, College of Engineering, Room
4532, 31270-901 Belo Horizonte, MG, Brazil. Tel.: +55 31 3409 1996.
E-mail address: jmfcalixto@gmail.com (J.M.F. Calixto).
encing parameters, such as the steel ratio and shear-span ratio
are very different within different intervals of beam size (or beam
depth). Neglecting or underestimating these uncertainties may
estimate incorrectly the corresponding probabilities of failure thus
compromising safety of RC structures subjected to shear.
Epistemic uncertainties are related to limited knowledge. As
pointed out by Melchers [8] epistemic uncertainties refer to those
that might be reduced with: (i) additional data or information, (ii)
better modeling, and (iii) better parameter estimation. All these
topics are dealt with in the research presented herein.
In this study, the database assembled by Ribeiro [9], encom-
passing experimental results for the shear resistance of slender
beams with stirrups has been enlarged. A random variable model
error (epistemic uncertainty) is associated to each of the shear
models analyzed in this work. Better estimates of the statistics of
the model error are obtained by the use of a filtered database
in such a way to circumvent the aforementioned drawbacks. A
multi-regression analysis is performed in order to derive an
expression that best describes shear resistance of RC slender

http://dx.doi.org/10.1016/j.engstruct.2016.02.045
0141-0296/ 2016 Elsevier Ltd. All rights reserved.
 A.B. Ribeiro et al. / Engineering Structures 116 (2016) 140147
beams to be used in conjunction with reliability analyses and to
identify the most significant parameters in the proposed equation
for shear resistance. To this end, a model that provides a smaller
coefficient of variation and the same level of conservatism with
respect to the main variables affecting shear capacity of RC beams
is sought. The information derived herein can be easily used in the
estimation of the shear strength capacity of RC beams and the
attendant model error, which in turn is needed in the assessment
of the corresponding implicit reliability levels of a given design
recommendation.
2. Shear models
Different procedures exist for the computation of the shear
capacity (scalc) of RC beams. These models provide shear capacity
as the sum of the concrete and steel stirrups contributions, sc,
and ssw, respectively:
scalc sc ssw
In Eq. (1), the first term, sc, is intended to account not only for the
concrete shear strength in the uncracked concrete beyond the end
of the inclined cracks but also for the complementary truss mecha-
nisms, namely aggregate interlock along a diagonal crack and dowel
action provided by the longitudinal reinforcement [6,10,11].
The contributions, sc, and ssw, according to model I of NBR 6118
[1], the simplified model from ACI 318 [2], the model in EUROCODE
2 [3] and the simplified model from CSA A.23.3 [4] are shown in
Tables 1 and 2, respectively. In these tables, fc stands for the con-
crete compressive strength, qw for stirrups ratio, fy for steel yield
strength, and h for the angle of the concrete struts.
As can be seen from those tables, these models differ in the val-
ues set for both the concrete and stirrups contributions. For the
design of new structures, object of this study, EUROCODE 2 com-
pletely neglects the concrete contribution, sc. On the other hand,
in the evaluation of the concrete contribution sc, ACI 318, CSA
A.23.3 and NBR 6118 codes rely only on the concrete strength,
ignoring the aggregate interlock along a diagonal crack and the
dowel action provided by the longitudinal reinforcement. It is
worth mentioning that each code has a limit value of the concrete
compressive strength: for CSA A.23.3 and ACI 318 the upper values
are 60 and 70 MPa respectively, while EUROCODE 2 and NBR 6118
allow values up to 90 MPa.
Taylor [12] and Kani [13] have shown that the shear strength of
concrete beams decreases as the depth of the beams increases.
Sneed and Ramirez [14] have stated that the reduction in the shear
with increasing effective depth in beams is influenced not only by
size effects but also by differences in behavior and mode of shear
transfer at failure (beam action versus arch action). More recently,
Mari et al. [15] have also presented a mechanical based shear
model that includes size effects. All the above RC design codes do
not consider these findings.
With respect to the stirrups contribution, ssw, both the CSA
A.23.3 and the EUROCODE 2 use the same expression; however,
Table 1
Shear models for concrete contribution.
sc (MPa)
NBR 6118 (2014) Model I fck 6 90 MPa h 45
sc sc0 0:126f ck 2=3
ACI 318 (2008) f0 c < 70 MPa h 45
q
sc 0:17 f 0c
EUROCODE 2 (2004) fck 6 90 MPa 21:8 6 h 6 45
sc = 0
CSA A23.3 (2003) f0 c < 60 MPa h 35 b 0:18
q
sc 0:9b f 0c
141
EUROCODE 2 allows the designer select different values of the strut
inclination h in the range between 21.8 and 45 while CSA A.23.3
sets h equals to 35. A smaller value of h allows, according to Euro-
codes equation, a reduction in the amount of shear reinforcement
ratio necessary to achieve the same ssw.
3. Shear database
3.1. Ribeiro and Calixto [5]
Ribeiro and Calixto [5] have assembled a shear database com-
prising results from 265 tests of beams with stirrups and diagonal
tension failure. The percentage of beams with effective depth lar-
ger than 60 cm was equal to 10%. The shear models considered
in that study are models I from NBR 6118 [1], the simplified model
from ACI 318 [2], and the model from EUROCODE 2 [3]. That study
also presented details of the beams including beam geometry, steel
1 area (both for longitudinal steel and stirrups), concrete compres-
sive strength, steel yield strength, and rupture stress. Ribeiro and
Calixto observe the paucity of experimental results for beams with
effective depth larger than 60 cm. It was also observed a general
trend for the models analyzed for non-conservative results for
beam depth in this range.
3.2. Enlarged database
Following the need identified by not only Ribeiro and Calixto [5]
but also by Bazant & Yu [7], in this study, a literature survey was per-
formed in order to increase the available database, especially with
respect to larger beam depths. This effort resulted in the addition
of 15 RC beams with stirrups to the Ribeiro and Calixto database,
comprising experimental results from Collins & Kuchma [16], Zar-
aris [17], Sherwood et al. [18], Ghannoum [19], and Yoshida [20].
Regarding beams with stirrups, the ones with practical interest
, after discarding few experimental data for not providing enough
required information, it resulted in 273 RC beams. This enlarged
database is shown in the Appendix. In this database, only beams
with vertical stirrups and subjected to point loads have been
selected. Results corresponding to beams with stirrups having steel
yield strength above 950 MPa have also been ignored, since this
does not correspond to cases of practical interest. All beams have
been made of normal-weight concrete and have failed due to diag-
onal tension. They have ratio shear span to effective depth, a/d,
greater than 2.5, and aspect ratio (width to effective depth), bw/d,
with an upper limit equal to 5. The statistics of the most influential
variables of this enlarged database is presented in Table 3.
3.3. Filtered database
As pointed out by Bazant & Yu [7], current databases of shear
test results present two major drawbacks: (i) most data points
are crowded in the small size range, and (ii) the means of the sub-
sidiary influencing parameters, such as the longitudinal steel ratio,
qwfy and a/d ratio, are very different within different intervals of
beam size (or beam depth). This problem has also been observed
in the enlarged database of beams with stirrups when it was split
in three ranges of the effective depth d (d < 30 cm,
30 cm 6 d < 60 cm, and 60 cm 6 d) as shown in Table 3.
In this study, a filtered database was carefully assembled in
such a way to attempt to circumvent the drawbacks mentioned
above. In this filtered database the most important variables fc, qw-
fy, ql, and a/d would have similar averages within each effective
depth interval. As all the beams with effective depth d P 60 cm
did not have ql > 4% and qwfy > 0.2 kN/cm2, beams with these val-
ues of longitudinal steel ratio and stirrup strength were at first dis-
142 A.B. Ribeiro et al. / Engineering Structures 116 (2016) 140147

Table 2 The experimental rupture stress, sExp, is equal to the ratio of

Shear models for stirrups contribution. shear force at rupture to the area corresponding to the effective
ssw (MPa) depth and the smallest width of the beam; the model predicted
NBR 6118 (2014) Model I, fck 6 90 MPa h 45 rupture stress, sCALC, is obtained for each model. Statistics of the
ssw = 0.9qwfywd model error (mean, standard deviation, coefficient of variation,
ACI 318 (2008), f0 c < 70 MPa h 45
ssw = qwfy minimum and maximum) for the 273 beams with stirrups are pre-
EUROCODE 2 (2004), fck 6 90 MPa 21:8 6 h 6 45 sented in Table 5, for the NBR 6118, EUROCODE 2, ACI 318, and CSA
ssw = 0.9qwfywd cot gh models. From Table 5, it can be observed that the mean value for all
CSA A23.3 (2003) f0 c < 60 MPa h 35 b 0:18 models is above 1.0, with NBR 6118 presenting the smallest mean
ssw = 0.9qwfy cot gh
(1.23) and the EUROCODE 2 the largest (1.68 for h = 21.8 and 4.20
for h = 45). It can also be observed that these code design equa-
tions present large values of coefficient of variation (30%
Table 3 < COV 6 50%).
Statistics of the most influential variables: enlarged database. Statistics of the model error for the 103 beams with stirrups
comprising the filtered database are presented in Table 6, for the
Enlarged database 273 beams
NBR 6118, EUROCODE 2, ACI 318, and CSA models.
qa (%) fc (MPa) a/d qwfy (kN/cm2) sExpb (kN/cm2) From Table 6, it can be observed that the mean value for all
All beams (273) models is above 1.0, and again, with NBR 6118 presenting the
Mean 2.90 54.34 3.26 0.112 0.343
smallest mean (1.01) and the EUROCODE 2 the largest (1.68 for
St. dev. 1.14 23.39 0.79 0.106 0.185
COV (%) 39.44 43.04 24.09 94.00 54.05
h = 21.8 and 4.19 for h = 45). COV values were in the range
Minimum 0.50 12.00 2.49 0.020 0.083 between 29% and 34%. To further investigate the behavior of each
Maximum 6.69 125.31 7.20 0.781 1.390 model with respect to the variables d, ql, fc, and qwfy, plots of the
d < 30 cm (131 beams) model error as a function of each of these variables are presented
Mean 3.23 55.97 3.33 0.128 0.398 in Figs. 15. It can be observed that in the case of NBR 6118, ACI
St. dev. 1.24 23.07 0.94 0.127 0.210 318, and CSA A.23.3 (Figs. 13 respectively) a trend exists for a
COV (%) 38.36 41.23 28.19 98.91 52.67
decrease in the model error as the beam effective depth increases.
Minimum 0.98 12.00 2.49 0.034 1.57
Maximum 6.69 111.80 7.20 0.781 1.390
30 6 d < 60 cm (106 beams)
Mean 2.84 49.79 3.29 0.112 0.330 5. Nonlinear regression model
St. dev. 0.79 20.50 0.67 0.085 0.142
COV (%) 27.97 41.18 20.26 76.14 43.07 Reliability analysis of RC beams subjected to shear requires an
Minimum 0.97 13.65 2.50 0.034 0.105 adequate resistance model for the beam and the corresponding
Maximum 5.68 120.14 6.00 0.643 0.881
statistics of the model error. As aforementioned, the behavior of
d P 60 cm (36 beams) the model error according to the models analyzed in this study is
Mean 1.87 61.83 2.95 0.054 0.181
St. dev. 0.87 29.72 0.20 0.022 0.065
dependent on: (i) concrete shear strength (function of fc) in the
COV (%) 46.41 48.06 6.87 40.28 36.12 uncracked area beyond the end of the inclined cracks, (ii) dowel
Minimum 0.50 21.00 2.50 0.020 0.083 action provided by the longitudinal reinforcement (q), (iii) size
Maximum 2.99 125.31 3.28 0.104 0.337
a
Longitudinal tensile reinforcement ratio q 100As =Ac .
b Table 4
Failure shear stress = (Vfailure/bw  d).
Statistics of the most influential variables: filtered database.

carded. In order to keep the similar mean values of ql and qwfy in Filtered database 103 beams
each effective depth range, the database was assembled in ascend- qa (%) fc (MPa) a/d qwfy (kN/cm2) sExp b (kN/cm2)
ing order for the values of ql and then of qwfy. Next, beams having
All beams (103)
the largest or smallest values of these two parameters were Mean 2.01 55.22 3.08 0.069 0.239
excluded as needed to fit the mean values of these variables in St. dev. 0.61 24.66 0.37 0.030 0.090
the same range for each effective beam depth interval. This proce- COV (%) 30.42 44.65 12.05 43.63 37.79
dure was repeated until similar mean values of ql and qwfy were Minimum 0.97 12.76 2.49 0.033 0.083
Maximum 2.99 125.31 4.50 0.149 0.481
reached in each effective beam depth interval. This procedure led
to a filtered database with 103 data points. The selected beams cor- d < 30 cm (38 beams)
Mean 1.94 45.53 3.18 0.072 0.275
responding to these data points are underlined in the database pre-
St. dev. 0.50 18.97 0.54 0.022 0.083
sented in the Appendix. The statistics of the most influential COV (%) 26.00 41.67 16.92 30.48 30.13
variables of this filtered database are presented in Table 4. Minimum 0.98 12.76 2.49 0.038 0.157
A comparison between the statistics of fc, qwfy, ql, and a/d, for Maximum 2.79 82.94 4.50 0.129 0.481
the enlarged database and the filtered database (Tables 3 and 4, 30 6 d < 60 cm (33 beams)
respectively) shows that this goal has been achieved. It is also Mean 2.17 58.47 3.09 0.075 0.248
important to mention, that similar number of data points resulted St. dev. 0.47 20.33 0.19 0.041 0.097
COV (%) 21.46 34.77 6.12 54.04 39.01
in each effective depth interval (38, 33, and 32, for d < 30 cm, Minimum 0.97 32.00 2.50 0.034 0.105
30 cm 6 d < 60 cm, and 60 cm 6 d, respectively). Maximum 2.60 120.14 3.68 0.149 0.471
d P 60 cm (32 beams)
Mean 1.94 63.40 2.96 0.057 0.187
4. Model error St. dev. 0.82 30.80 0.21 0.021 0.067
COV (%) 42.15 48.58 7.15 37.23 35.74
A random variable model error, n, is associated to each of the Minimum 0.99 21.00 2.50 0.033 0.083
shear models analyzed in this work. This random variable is Maximum 2.99 125.31 3.28 0.104 0.337

obtained from the ratio experimental, sExp, to model predicted, a

Longitudinal tensile reinforcement ratio q 100As Ac .
sCALC, rupture shear stress, i.e. n = sExp/sCALC. b
Failure shear stress = (Vfailure/bw  d).
 A.B. Ribeiro et al. / Engineering Structures 116 (2016) 140147 143

effects (bw, d), and (iv) amount of transverse reinforcement (qw). Table 6
While a more complex path could be used by deriving model error Statistics of the model error (filtered database: 103 beams with stirrups).

statistics as a function of some parameter, e.g. q a more effective sExp sExp sExp sExp sExp sExp
sNBR sEUROh45 sEUROh21:8 sACI sCSA sproposed equation
way is the derivation of an expression that would not present these
Mean 1.01 4.19 1.68 1.26 1.16 1.06
shortcomings. At first a linear additive form was derived then non-
St. dev. 0.34 1.42 0.57 0.38 0.34 0.24
linear forms were investigated. This nonlinear additive form was COV (%) 33.7 34.0 34.0 30.4 29.7 22.9
chosen since it provided a smaller bias (mean value the model Minimum 0.40 1.17 0.47 0.55 0.49 0.51
error closer to 1.0) and smaller COV, while not displaying a trend Maximum 2.10 7.96 3.18 2.49 2.25 1.76
with respect to with respect to bw, d, q, fc and qwfy, or in other
words, the same level of conservatism with respect to the whole
range of these variables in the database. The functional form of
the model, with bw, d, q, fc, and qwfy, as independent variables,
3
was selected in order to be, as much as possible, similar as cur-
NBR 6118 - Model I
rently used code design equations. In the proposed model, beam
shear capacity is estimated as the sum of the concrete and steel
stirrups contributions; it is given by the following equation: 2

Exp / calc
 1=3
100q f c
sCALC 0:3 2 1:4qw f y 2
bw d 1

In the above equation, variables bw and d are in millimeters, the

longitudinal steel ratio q in percentage, the calculated shear
strength sCALC, concrete compressive strength fc and the stirrup 0
0 500 1000 1500 2000
strength qwfy are in MPa.
d (mm)
The proposed equation considers almost all aforementioned
factors that affect the mode of shear transfer in RC beams; thus, 3
with respect to current code design equations analyzed herein, NBR 6118 - Model I
the above formulation is much more comprehensive.
Statistics of the model error associated to the nonlinear regres-
2
sion, nreg, (filtered database) are shown in the seventh column of
Exp / calc

Table 6. It is worth mentioning that the mean value of the model

error is 1.06 and the COV 23%. Additionally, as can be seen in
Fig. 6, which shows the nonlinear regression model error as a func- 1
tion of d, q, fc, and qwfy, no trend is observed in these results.
In order to assess the effectiveness and accuracy of the pro-
posed equation, a comparative study of was conducted with 0
respect to the database presented by Reineck et al. [6]. This data- 0 1 2 3 4
base consists of tests results of 87 slender RC beams with stirrups, l (%)
which according to Reineck et al. are appropriate for comparisons
to shear design provisions of ACI 318 and Eurocode 2. In this data- 3

base, 28 beams (32%) have effective depth equal or greater than NBR 6118 - Model I
60 cm. The test results are also compared to NBR 6118, ACI 318,
Eurocode 2 and CSA A.23.3 estimates for maximum shear capacity. 2
Exp / calc

The statistics of the most influential variables of this comparison

are shown in Table 7.
The results presented in Table 7 indicate that mean values for 1
all models are above 1.0. The proposed model has the smallest
mean (1.15) and COV (17.6%), while the EUROCODE 2 the largest
mean (3.29 for h = 45) as well as COV (31.5%). Using h = 21.8 in
0
EUROCODE 2, the mean and COV values are 1.32 and 31.5% respec-
0 20 40 60 80 100 120 140
tively. Additionally, Fig. 7 shows that no trend is observed in the
fc (MPa)
proposed nonlinear regression model error as a function of d, q,
fc, and qwfy with respect to 87 shear test results of slender RC 3
beams with stirrups presented by Reineck et al. [6]. NBR 6118 - Model I
A box-plot diagram of the shear models analyzed in this study
and that of the nonlinear regression model proposed herein with 2
Exp / calc

respect to 87 shear test results (Reineck et al. [6]) is presented in

Table 5
Statistics of the model error (273 beams with stirrups). 1

sExp sExp sExp sExp sExp

sNBR sEUROh45 sEUROh21:8 sACI sCSA

Mean 1.23 4.20 1.68 1.48 1.33 0

St. dev. 0.41 2.10 0.84 0.48 0.44 0 1 2
COV (%) 33.3 50 50 32.4 33.1 w fy (MPa)
Minimum 0.40 1.17 0.47 0.55 0.49
Maximum 3.86 22.11 8.84 5.53 5.19
Fig. 1. Model error as a function of d, ql, fc, and qwfy for NBR 6118 (2014).
144 A.B. Ribeiro et al. / Engineering Structures 116 (2016) 140147

Fig. 2. Model error as a function of d, ql, fc, and qwfy for ACI 318 (2008).

Fig. 3. Model error as a function of d, ql, fc, and qwfy for CSA A23.3 (2003).

Fig. 8. For better visualization, due to the large value of the mean
model error associated to the EUROCODE 2 model with h = 45,
these results have not been included. It can clearly be seen from
this box-plot diagram the appropriateness of the proposed nonlin-
ear regression model.
6. Summary and conclusions
In this study, the modeling of the epistemic uncertainties asso-
ciated to RC slender beams shear resistance was performed. Most
of the difficulties arise from the available shear databases present-
 A.B. Ribeiro et al. / Engineering Structures 116 (2016) 140147 145

9
EUROCODE = 45
8
7
6
Exp / calc

5
4
3
2
1
0
0 200 400 600 800 1000 1200 1400 1600 1800 2000
d (mm)

9
EUROCODE = 45
8
7
6
Exp / calc

5
4
3
2
1
0
0 1 2 3 4
l (%)

9
8 EUROCODE = 45
7
6
Exp / calc

5
4
3
2
1
0
0 20 40 60 80 100 120 140
fc (MPa)

9
8
EUROCODE = 45
7
6
Exp / calc

5
4
3
2
1
0
0 1 2
w fy (MPa)

Fig. 4. Model error as a function of d, ql, fc, and qwfy for EUROCODE 2 (2004) with Fig. 5. Model error as a function of d, ql, fc, and qwfy for EUROCODE 2 (2004) with
h = 45. h = 21.8.

ing the crowding of data points in the small size range and large
differences of the means of the subsidiary influencing parameters,
within different intervals of beam depth. Similar limitations can
also be observed in other databases for problems such as slender
RC columns [21] and concrete creep and shrinkage [22].
In order to circumvent these problems, the database assembled
by Ribeiro & Calixto [5], encompassing experimental results for the
shear resistance of beams with and without stirrups, was enlarged;
additionally, a filtered database was used. Both the enlarged and
filtered databases are presented in the Appendix.
The evaluation of the model errors associated to different shear
models has been obtained. Statistics of the model error for both the
273 (enlarged database) and the 103 beams (filtered database)
were described. The models considered were the NBR 6118, EURO-
CODE 2, ACI 318, and CSA models. All models presented mean val-
ues above 1.0, and large COVs of roughly 30%. The mean values
associated to the EUROCODE 2 (for h = 45 and h = 21.8) model
error were larger than the other investigated code design equa-
tions; this is due to the fact that in the EUROCODE model the con-
crete contribution is completely disregarded. The other three
models displayed a trend of decreased bias as the effective depth
increases, i.e. the level of conservatism is not kept constant for
all ranges of the effective depth.
A reliability analysis of RC beams subjected to shear requires
the use of an adequate resistance model for the beam and the cor-
responding statistics of the model error. Considering the limita-
146 A.B. Ribeiro et al. / Engineering Structures 116 (2016) 140147

Table 7
Statistics of the model error with respect to shear test results of 87 slender RC beams
with stirrups presented by Reineck et al. [6].
sExp sExp sExp sExp sExp sExp
sNBR sEUROh45 sEUROh21:8 sACI sCSA sproposed equation

Mean 1.33 3.29 1.32 1.53 1.29 1.15

St. dev. 0.35 1.04 0.41 0.33 0.26 0.20
COV (%) 26.3 31.5 31.5 21.8 19.9 17.6
Minimum 0.58 1.20 0.48 0.72 0.68 0.69
Maximum 2.05 5.76 2.30 2.38 2.09 1.65

Nonlinear Regression Model

2,0

1,5

exp / calc
1,0

0,5

0,0
0 200 400 600 800 1000 1200 1400
d (mm)
Nonlinear Regression Model
2,0

1,5
/calc

1,0

0,5

0,0
0,0 1,0 2,0 3,0 4,0 5,0
l (%)

Nonlinear Regression Model

2,0

1,5
exp /calc

1,0

0,5

0,0
0 20 40 60 80 100 120 140
fc (MPa)

Nonlinear Regression Model

2,0

1,5
exp /calc

1,0

Fig. 6. Nonlinear model regression error for the enlarged database as a function of
d, q, fc, and qwfy. 0,5

0,0
0,0 0,4 0,8 1,2
tions of the models analyzed in this study, a nonlinear regression
w fy (MPa)
model was proposed. This model uses bw, d, q, fc, and qwfy, as
independent variables. The effectiveness and accuracy of this Fig. 7. Nonlinear regression model error as a function of d, q, fc, and qwfy with
model was assessed by comparing model predictions to test results respect to shear test results of 87 slender RC beams with stirrups presented by
of 87 slender RC beams with stirrups (28 beams had d P 60 cm) Reineck et al. [6]
 A.B. Ribeiro et al. / Engineering Structures 116 (2016) 140147
Fig. 8. Box-plot diagram for the model errors associated to NBR 6118, ACI 318, CSA
A23.3, EUROCODE 2 (with h = 21.8) and to the proposed nonlinear regression
model with respect to shear test results of 87 slender RC beams with stirrups
presented by Reineck et al. [6].
presented by Reineck et al. [6]. No trend was found associated to
the variables d, q, fc, and qwfy, i.e. a consistent level of conser-
vatism throughout the range of these variables. The model error
associated to this regression analysis has mean equal to 1.15 and
COV equal to 17.6%, and can described by a Normal distribution.
The regression model obtained in this study and the corre-
sponding statistics can be easily used in a reliability analysis pro-
cedure to assess safety levels implicit in different design
procedures. Work is underway at Federal University of Minas Ger-
ais (Brazil) using this information in the assessment of the corre-
sponding implicit reliability levels of current design
recommendations.
It is also interesting to note that the coefficient of variation for
all shear models considered in this work are considerably higher
than those reported by other authors. For instance, Nowak et al.
[23] report COV of 10% for the model error in the prediction of
shear capacity of RC beams. As it can be expected, this variable will
have a large impact on the reliability of RC beams subjected to
shear designed according to some major design codes; addition-
ally, the underestimation of the variability of the model error
may also represent an overestimation of the corresponding impli-
cit safety levels.
Acknowledgements
The authors would like to thank: CAPES (Coordenao de Aper-
feioamento de Pessoal de Nvel Superior), FAPEMIG (Fundao de
Apoio Pesquisa do Estado de Minas Gerais), and CNPq (Conselho
Nacional de Desenvolvimento Cientfico e Tecnolgico), for the
financial support provided.
147
Appendix A. Supplementary material
Supplementary data associated with this article can be found, in
the online version, at http://dx.doi.org/10.1016/j.engstruct.2016.
02.045.
References
[1] Associao Brasileira de Normas Tcnicas (ABNT). NBR 6118: Projeto de
Estruturas de Concreto Procedimento (in Portuguese), Rio de Janeiro; 2014. p.
170.
[2] ACI Committee 318, Building Code Requirements for Structural Concrete (ACI
31811) and Commentary, Farmington Hills, Michigan; 2011.
[3] European Committee for Standardization, Eurocode 2: Design of Concrete
Structures Part 1: General Rules and Rules for Building, Brussels; 2004.
[4] Canadian Standards Association (CSA), CSA A23.3-04: Design of Concrete
Structures, Rexdale, Ontario; 2003.
[5] Ribeiro AB, Calixto JMF. Comparative analysis of design code criteria for shear
strength of reinforced concrete beams. IBRACON Struct J 2007;3(1):1961.
<www.ibracon.org.br/publicaes>.
[6] Reineck K, Bentz E, Fitik B, Kuchma DA, Bayrak O. ACI-DAfStb databases for
shear tests on slender reinforced concrete beams with stirrups. ACI Struct J
2014;111(5):114756.
[7] Bazant ZP, Yu Q. Minimizing statistical bias to identify size effect from beam
shear database. ACI Struct J 2008;105(6):68591.
[8] Melchers RE. Structural reliability analysis and prediction. John Wiley; 1999.
[9] Ribeiro AB. Projeto Semi-Probabilstico de Vigas de Concreto Armado com
Modo de Falha por Cisalhamento (in Portuguese), PhD Dissertation. Belo
Horizonte, Brazil: PROPEEs, Federal University of Minas Gerais; 2012.
[10] Joint ACI-ASCE Committee 445 Recent Approaches to Shear Design of
Structural Concrete. J Struct Div, ASCE 1999; 124(12) 13751417.
[11] Nilson A, Winter G. Design of concrete structures. 11th ed. McGraw-Hill; 1991.
[12] Taylor HPJ. Shear strength of large beams. J Struct Div, ASCE 1972;98
(ST11):247389.
[13] Kani GNJ. How safe are our large reinforced concrete beams? ACI J 1967;64
(3):12841.
[14] Sneed LH, Ramirez JA. Influence of effective depth on shear strength of
concrete beams experimental study. ACI Struct J 2010;107(5):55462.
[15] Mari A, Bairan J, Cladera A, Oller E, Ribas C. Shear flexural strength mechanical
model for the design and assessment of reinforced concrete beams. Struct
Infrastructure Eng 2014. http://dx.doi.org/10.1080/15732479.2014.964735.
Published online: 07/October.
[16] Collins MP, Kuchma D. How safe are our large, lightly reinforced concrete
beams, slabs and footings? ACI Struct J 1999;96(4):48290.
[17] Zararis PD. Shear strength and minimum shear reinforcement of reinforced
concrete slender beams. ACI Struct J 2003;100(2):20314.
[18] Sherwood EG, Bentz EC, Collins MP. Effect of aggregate size on beam-shear
strength of thick slabs. ACI Struct J 2007;104(2):18090.
[19] Ghannoum WM. Size effect on shear strength of reinforced concrete beams.
MSc thesis. Montreal, Canada: Department of Civil Engineering and Applied
Mechanics, McGill University; 1998.
[20] Yoshida Y. Shear reinforcement for large lightly reinforced concrete members,
MSc thesis. Ottawa, Canada: Department of Civil Engineering, University of
Toronto; 2000.
[21] Zhou W, Hong HP. Statistical analyses of strength of slender RC columns. J
Struct Eng 2001;127(1):217.
[22] Bazant ZP, Li G-H. Unbiased statistical comparison of creep and shrinkage
prediction models. ACI Mater J 2008;105(6):61021.
[23] Nowak AS, Rakoczy AM, Szeliga E. Revised statistical resistance models for R/C
structural components. SP-284 Andy Scanlon Symposium on Serviceability
and Safety of Concrete Structures: From Research to Practice, American
Concrete Institute; 2012.