Engineering Structures
journal homepage: www.elsevier.com/locate/engstruct
A R T I C LE I N FO A B S T R A C T
Keywords: Reinforced concrete flat slabs are structural members frequently used in building construction. Despite their
Reinforced concrete many advantages, such structural systems suffer from several drawbacks. From a safety point of view, the
Flat slab concentration of shear forces at the vicinity of local supports, e.g., columns, edges and corners of walls, can lead
Punching to possible structural failure from punching. Punching is a dangerous phenomenon due to its brittle mode of
Safety
failure and its ability to spread over a whole structure, which can be followed by a progressive collapse. Several
models for the assessment of punching capacity have been developed and calibrated using experimental results
from laboratory tests. Some models are fully empirical, e.g., EC2 (2004), while other models reflect the physical
nature of the phenomenon; however, their safety level was calibrated using experimental data, as the model
presented in Model Code 2010 or the CSCT-based model, which is expressed in a closed form. This paper deals
with a statistical evaluation of the safety level of the previously mentioned models for punching resistance
without transverse reinforcement. A database which includes the results of more than 600 experimental tests of
flat slab specimens has been used. The relations between the safety of the model and the quality of the concrete
used, the amount of the bending reinforcement, and the effective depth were found using advanced statistical
methods.
1. Introduction EC2 (2004) model from a designer′s point of view, because the
punching capacity assessed depends on the deformations of the long-
Punching is one of the most dangerous forms of structural failure in itudinal reinforcement, shear slenderness, and maximum aggregate
reinforced concrete slabs due to its brittleness. Failure at one local size. The radial bending moments at the support and their distribution
support may lead to the overloading of neighbouring areas and then over the span also play an important role in the estimation of the
may spread over the whole structure, thereby resulting in a progressive punching capacity.
collapse, see Fig. 1. Therefore, design models for the assessment of For the sake of simplicity a new model based on the CSCT theory
punching resistance should at the same time be both conservative and was developed and formulated by Muttoni and Ruiz [6] in 2017. The
reasonably accurate to avoid unnecessary increases in construction formula for punching resistance was expressed in a closed form. The
costs. The design model for punching introduced in Eurocode 2 [1] is model was originally proposed for the second generation of Eurocode 2.
based on the model originally presented in Model Code 1990. The
model is empirical since the most influential parameters of the 2. EC2 (2004) model
punching resistance were statistically evaluated using the results of
experimental tests mostly carried out in the 1960s to 1980s. The EC2 (2004) model was based on a formula proposed by Zsutty
A new model for punching based on the Critical Shear Crack Theory [8] in 1968. Zsutty statistically evaluated the relation between the
(CSCT) was developed in the 1990s by Muttoni. The principles of the shear strength and the amount of the main reinforcement, expressed by
CSCT were published by Muttoni and Schwartz [3] in 1991 for the first the reinforcement ratio ρ and the concrete compressive strength on
time. The theory was later refined by Muttoni [4] and finally became cylinders fc [MPa]. After some refinements, the final equation (1) was
the fundamental design model for the assessment of punching re- published in Model Code 1990.
sistance in Model Code 2010 [2]. The model is more complex than the
⁎
Corresponding author.
E-mail addresses: jaroslav.halvonik@stuba.sk (J. Halvoník), jana.kalicka@stuba.sk (J. Kalická), lucia.majtanova@stuba.sk (L. Majtánová),
maria.minarova@stuba.sk (M. Minárová).
https://doi.org/10.1016/j.engstruct.2019.03.055
Received 23 July 2018; Received in revised form 16 March 2019; Accepted 18 March 2019
0141-0296/ © 2019 Published by Elsevier Ltd.
J. Halvoník, et al. Engineering Structures 188 (2019) 627–636
Fig. 1. Progressive collapse of a parking garage due to the punching of the roof slab, Bratislava (2012).
CRk,c
VRd,c = k (100 ∗ ρfck )1/3u1 d
γC (1)
where
628
J. Halvoník, et al. Engineering Structures 188 (2019) 627–636
derived from a quadri-linear model, where for LoAI, the ratio mSd/mRd where
can be taken as equal to one. For LoAII the bending moment mSd can be
determined as mSd = VEd/8 for an axi-symmetric condition that refers ddg is a coefficient that takes into account the type of concrete and
to an internal column and zero unbalanced moments. In the case of its aggregate properties, i.e., 32 [mm] for concrete of a normal
LoAIII, rs and mSd can be determined using a linear elastic analysis. weight
LoAIV assesses ψ by using a non-linear analysis that accounts for av shear span, the geometric average of the shear spans in both
cracking and tension stiffening. orthogonal directions and not less than 2.5d
1.5 μ parameter accounting for the shear force and bending moment in
rs fyd ⎛ mSd ⎞ the region of the shear, for an internal column without an un-
ψ = km ⎜ ⎟
629
J. Halvoník, et al. Engineering Structures 188 (2019) 627–636
Fig. 4. Comparison of LoAII and LoAIII as well as CF CSCT (2017) and CF CSCT (2018) using box and whisker graphs with the outliers highlighted in black (outliers)
and grey points (far out data) outside the boxes and whiskers.
the models based on CSCT. In the case of the MC2010 model, the ag- of the input parameters.
gregate size dg,max is an important parameter for an evaluation of the The aim of the statistical analysis is to compare the four approaches
punching capacity, see formula (4). However, this information is with each other by a comparison of the relevance of their design for-
missing for many tests in the database, which leads to a significant loss mulae. All the calculations were carried out using Mathematica 11.1
of the data from 385 to 241. Therefore, two datasets of the tests were software. At first, the normality of the Pi data set distribution was in-
created and then analysed. The first dataset included all the relevant spected. The Kolmogorov – Smirnov test indicates the degree of nor-
tests, where if dg,max was unknown, the value of 16 mm has been used mality in all five cases. The rational histograms are provided in Fig. 5.
(MC2010 LoAIII). The second set only included tests where dg,max was Then the outstanding data were detected and excluded from any
known (MC2010 LOAIII*). further analysis. For this purpose a special outstanding data test based
on the quartile deviation was used:
7. Statistical evaluation of the models for punching resistance I = [Q1 − k (Q3 − Q1), Q3 − k (Q3 − Q1) (11)
where
A statistical evaluation of the models for the prediction of punching
resistance without shear reinforcement has been carried out for four
Q1,Q3 the lower and upper quartiles
models and five data sets, i.e., EC2 (2004), Model Code 2010, and
k coefficient: if k = 1.5 and x ∉ I, then x are outliers
closed form CSCT models (2017) and (2018) with the partial safety
if k = 3 and x ∉ I then x are far out data
factor γC = 1.0. For strength fck the mean value of the actual concrete
strength on cylinders was used. The control perimeters have been as-
The symmetry of this test ensures the existence and variability of the
sumed at a distance of 2d from the face of a column for the EC2 (2004)
extremal values. The resulting number of outstanding values are 24, 11,
model and d/2 for the other models. Accordingly, the four formulae can
19, 10 and 14 for EC2 (2004), CF CSCT (2017), CF CSCT (2018), and
be validated.
Model Code 2010, respectively.
For the sake of a comparison of the suitability of the four formulae,
A comparison of the data range with and without the outstanding
the ratio Pi = (Vtest/VRd,c)i as the statistical variable was assigned,
data is in Figs. 6 and 8 in the box and whisker graphs.
where “i” is the number of the test; Vtest is the resistance obtained from
Fig. 7 indicates the differences between the two MC2010 datasets in
the test, i.e., the so-called experimental or test value; and VRd,c is the
a more detailed way. The first plot shows a comparison of the histo-
punching resistance obtained from the particular design equation of one
grams where the LoAIII dataset contains 385 tests and the LoAIII* 241
of the four theoretical models.
tests. The second plot is a box and whisker graph with the outstanding
For this purpose a so-called box and whisker graph was used that
data. The plot shows a slightly better level of accuracy of the model if
depicts the groups of data through their quartiles. First, the median (the
only the tests with the known dg,max are included in the statistical
thin gap within the coloured boxes in Figs. 4 and 6–8) splits the data
analysis.
into two halves. Then, the median of the upper/lower half is taken as
The 5% quantile Pk,0.05 is considered as the critical value. The
the upper/lower bound of the box enfolding the nearest data to the
characteristic values of the normal distributions of the models are as-
median. Then both the upper and lower whiskers delimit the 1st and
sembled in Table 1 (see Fig. 9).
4th quartiles. Methods exist for detecting so-called outstanding data,
which are far from the others and can affect the analysis, thereby re-
sulting in wrong deductions. Accordingly, the outstanding data has to 8. Discussion of the results obtained
be treated carefully. Sometimes a reason exists, e.g., an indication of an
experimental error, for omitting these data. However this decision is The statistical analysis of the safety models concerning the results of
left up to an expert. the tests carried out on the flat slab specimens for punching proved that
“Pi greater than 1.0” means that the model is on the safe side; if it is the MC2010 model, i.e., the level of approximation III, was the most
no greater than 1.0, then it is unsafe. Therein, a match of the para- reliable one. It stands on the safety side and is sufficiently precise at the
meters input in both the theoretical and test values has to be ensured. same time. The statistical evaluation where the outstanding values of
The ratio Vtest/VRd,c – index is a one-parametric dataset. There were the ratio P = Vtest/Vmodel are omitted reveals that the data of P are
405, 385, and 241/385 pairs of data tested in the EC2 (2004), CF CSCT, normally distributed. The Kolmogorov-Smirnov and other tests of the
and MC2010 models respectively. Each theoretical value is calculated data′s normality were used. Only 10 (14) outstanding elements of data
with the same geometrical and physical parameters as the one mea- were withdrawn from the analysis; they represented 4.2% (3.6%) of the
sured in the lab test. Thus the effect of these parameters decreases and entire dataset. The mean value of the ratio Pm reached 1.177 (1.157)
even becomes negligible. Such an approach enables the validation of with a variation coefficient of 0.131 (0.137). The values in brackets are
the empirical forms of the four models addressed along the entire range valid for the dataset that also includes the tests with an unknown
630
J. Halvoník, et al. Engineering Structures 188 (2019) 627–636
Fig. 5. Ratio test/normative punching resistance histograms: (a) EC2 (2004), (b) MC 2010 LoAIII, (c) MC 2010 LoAIII*, (d) CF CSCT (2017), and (e) CF CSCT (2018).
The outstanding data are still included.
aggregate size. The values of the CoV document the very good accuracy 12.1% (13.7%). The coefficient of variation was 0.129, which was the
of the model. The characteristic value of the model’s reliability ex- best result, but the characteristic value of the factor Pk,0.05 was only
pressed by the 5% quantile Pk,0.05 exceeds 0.90 (0.88), which is an 0.76. The reason is that Muttoni and Ruiz used a less reliable failure
acceptable value for this type of structure even though according to EN criterion here than in the MC2010 model, see Fig. 3.
1990, the target value is 1.0. The acceptance of this level of safety The proposed modification of the failure criterion in the CF CSCT
follows from the favourable effects of the membrane forces and the (2018) model increases the model safety. The normal distribution of the
redistribution of the internal forces in the actual flat slabs. These effects statistical variable P was also detected. The mean value of the variable
cannot develop in isolated slab specimens. Therefore, the model can be P reached 1.036, and 38.5% of the test data fell below one, which is a
regarded as safe. much better result than in the case of the CF CSCT (2017) model with
The currently valid EC2 (2004) provides good results as well, even original failure criterion. The model was still accurate as shown by the
though the highest number of outstanding data that were recorded, i.e., CoV, which did not exceed 0.133.
a total of 24 measurements, it was 5.9% of all the values that were An additional statistical analysis focused on the relationship be-
referred. The normality tests confirm a normal distribution, too. The tween the ratio P and the cylinder strength of the concrete fck, the ratio
mean value of the ratio Pm reached 1.122, and the coefficient of var- P and the effective depth d of the slab and, finally, the ratio P and the
iation reached 0.165; thus the value of Pk,0.05 is 0.81, which is still an reinforcement ratio ρ. The most detailed analysis was carried out for the
acceptable level of reliability. concrete strength.
The CF CSCT (2017) model has the lowest level of reliability in The statistical set was split into five strength ranges, i.e., less than
comparison with the other ones. The most positive thing to note is that 20 MPa, (20–40 MPa), (40–60 MPa), (60–80 MPa), and (80–120 MPa).
only 11 outstanding elements of the data were omitted, which was 2.9% Table 2 provides the mean values, Table 3 the coefficients of variation
of all the tests. The normal distribution of the statistical variable P was CoVs, and Table 4 the 5% quantiles of the ratio P for each strength
also detected. The mean value of the ratio Pm reached 0.962, and more range, (see also Fig. 10).
than 62% of the test data fell below one, while, e.g., in the case of EC2 No significant differences were observed along the individual
(2004), it was 23.9%, and in the case of MC2010 LoAIII, it was only strength ranges of fck in the values of Pm and Pk,0.05. The best results
Vtest/VRd,c
Fig. 6. Box and whisker graphs with outstanding data as shown in black (outliers) and grey points (far out data) outside of the boxes and whiskers.
631
J. Halvoník, et al. Engineering Structures 188 (2019) 627–636
LoAIII LoAIII*
Vtest/VRd,c
LoAIII
LoAIII*
frequency frequency
Fig. 7. Histograms and box and whisker graph model with outstanding data for MC2010.
were obtained for the concrete strengths below 20 MPa and the worst tool for the safe design of flat slabs against punching.
for the strength range (80–120 MPa) in the case of EC2 (2004). Such The diagrams in Fig. 11 indicate the relation between ratio P and
results were expected because the model was calibrated for data from the cylinder strength of the concrete fck. Since the data are evenly
the most frequently used range of the concrete cylinder strength. In the distributed around 1, we can generally claim that the proposed design
case of CF CSCT (2017), the strength range (20–40 MPa) yielded the equation proves to be rather effective in fitting the test results. The
best results, while those in the range (60–80 MPa) yielded the worst effectiveness of the different equations in fitting the test results was
results. In the case of MC2010, the best results are for the strength range checked either by parametric statistical analysis in the case of the dis-
(80–120 MPa) and the worst again for (40–60 MPa). similarity of statistically non-significant parameters or by the non-
In this study, the focus was on the strength range (20–40 MPa), parametric statistical analysis in the case of insufficient similarity.
which is nowadays typical of concrete for slabs. In each model, the In the next four plots in Fig. 12, the relation between the ratio P and
characteristic values of the ratio P affiliated with this interval were very reinforcement ratio ρ is shown for each model. The EC2 (2004) was
close to the results obtained in the global range. In the case of EC2 calibrated for ρmax = 0.02; however, no limit was imposed in our ana-
(2004), they are almost the same, while in the case of the CSCT models, lysis. Nevertheless, EC2 (2004) also provides very good results for
they are slightly better in comparison with the global analysis, e.g., in ρ > ρmax. In the case of the models based on CSCT, the limit for ρ was
the case of CF CSCT (2017), the 5% quantile increased from 0.76 to extended up to 0.04. A very good global level of fitness between the
0.773, and in the case of CF CSCT (2018), from 0.793 to 0.812. In the experimental and theoretical values was within the entire range of the
case of MC2010, the values are nearly the same, i.e., 0.91 and 0.909. data. This is apparent from the plots and is proved by the tools of the
When the extended data set for the MC2010 model is assumed, some statistical factor analysis as well. However, a tendency, which is mainly
improvement is again observable, i.e., from 0.885 to 0.906. for a reinforcement ratio ranging from 0.5% to 2.5%, can be seen. The
Generally, the coefficient of variation CoV can be regarded as the safety level of the MC2010 model is slightly decreasing, while in the
best statistical parameter showing the quality of the design equations. case of CF CSCT (2018), it is increasing. The results are very similar for
The CF CSCT (2017) model has the lowest value and thus the highest the higher reinforcement ratios.
quality. Also, the other CSCT-based expressions have quite similar va- Fig. 13 Relation ratio P vs. effective depth d for the EC2 (2004),
lues of the CoV. These models mainly differ by their safety level. From MC2010 and CF CSCT models. The outliers are coloured in yellow.
this point of view, the MC2010 model has the highest safety level for The last four plots in Fig. 13 indicate the relation between the ratio
both datasets, while the CF CSCT (2017) model has the lowest. The P and the effective depth of the slab d. In the case of EC2 (2004), there
statistical evaluation of the CF CSCT (2018) model shows that the small are some doubts concerning the factor k that take into account the size
modification of the failure criterion may increase the safety level, while effect, mainly for thicker slabs. In the plot we can see that the safety
the CoV remains nearly the same. level of the model is decreasing in the interval from 50 mm to 150 mm.
The highest CoV was evaluated for the EC2 (2004) design equation. The explanation is clear as the factor k is limited by 2.0 for slabs with an
Nevertheless, the model cannot be regarded as a low-quality model. The effective depth d < 200 mm. However, the important issue is that the
acceptable safety level of the model provides designers with an efficient ratio P becomes stable and oscillates around 1.0 for slabs with an
Vtest/VRd,c
Fig. 8. Range of the data values after omitting the outstanding data.
632
J. Halvoník, et al. Engineering Structures 188 (2019) 627–636
Table 1
Statistical evaluation of the reliability of the models for the assessment of punching resistance.
Model Number of specimens with OD Number of specimens without OD Mean value [Pm] CoV [VP] Pk,0.05 % of data below one
Fig. 9. Ratio test/normative punching resistance histograms, (a) EC2 (2004), (b) MC2010 LoAIII, (c) MC2010 LoAIII*, (d) CF CSCT (2017), and (e) CF CSCT (2018)
without the outlier data.
Table 2 Table 4
Statistical evaluation of the model safety for the concrete strength ranges, mean Statistical evaluation of the model safety for the concrete strength ranges,
values. Pk,0.05.
Strength ranges EC2 (2004) CF CSCT CF CSCT MC2010 MC2010 Strength ranges EC2 (2004) CF CSCT CF CSCT MC2010 MC2010
(2017) (2018) LoAIII LoAIII* (2017) (2018) LoAIII LoAIII*
Global 1.128 0.962 1.033 1.165 1.177 global 0.810 0.760 0.793 0.885 0.910
≤20 MPa 1.166 0.987 1.050 1.191 1.165 ≤20 MPa 0.858 0.659 0.719 0.885 0.915
20–40 MPa 1.127 0.959 1.032 1.162 1.172 20–40 MPa 0.810 0.773 0.812 0.906 0.909
40–60 MPa 1.137 0.936 1.014 1.152 1.166 40–60 MPa 0.758 0.677 0.739 0.798 0.840
60–80 MPa 1.126 0.966 1.053 1.155 1.235 60–80 MPa 0.819 0.652 0.712 0.822 0.849
80–120 MPa 1.048 0.993 1.081 1.182 1.183 80–120 MPa 0.774 0.747 0.816 0.970 0.970
633
J. Halvoník, et al. Engineering Structures 188 (2019) 627–636
Vtest/VRd,c
Fig. 11. Relation ratio P vs. concrete strength fc for the EC2 (2004), MC2010 LoAIII and CF CSCT models. The outliers are coloured in yellow. (For interpretation of
the references to colour in this figure legend, the reader is referred to the web version of this article.)
provided the highest value of the CoV in comparison with the other The low value of CoV = 0.128 shows the very good degree of accuracy
models. However, the value of 0.166 is still very good. and thus the quality of the model.
The CF CSCT (2017) model has the lowest value of Pk,0.05, i.e., only If we take a look at the models analysed from the point of view easy
0.76; when nearly 66% of the test results fell below one, slightly better of use, the best model for practitioners is the EC2 model. Besides geo-
results of 0.773 were achieved in the case of the specimens cast from metrical parameters, i.e., an effective depth and column dimensions,
the concrete (20–40 MPa). On the other hand, the model has the highest the punching resistance VRd,c only depends on two variables, which are
degree of accuracy because only 11 outstanding elements of data were the reinforcement ratio and the concrete strength. The CF CSCT model
detected, which is 2.9% of all the tests. Also, the coefficient of varia- is a bit complicated, because the shear span length has to be known to
tions was the lowest among the models investigated, i.e., only 0.129. evaluate the punching resistance. The most complicated are the
Nevertheless, the lowest level of reliability is given by the mean value, MC2010 design equations because the rotation ψ has to be directly
which is only 0.962. calculated. So, the model’s ranking is inverted in comparison to the
The CF CSCT (2018) model with modified failure criterion, provides results of the statistical analysis. Therefore, our recommendation is to
a higher level of safety than the original one. The 5% quantile has been use the CF CSCT model. However, the reliability of CF CSCT (2017) has
increased to 0.793 and the mean value to 1.036; only 38.5% of the test to be increased to at least the value of the current EC2 model or
results fell below 1.0, which is half of the previous solution. Better somewhere in between EC2 and MC2010 if the model is going to be
results were also obtained for the specimens cast from concrete with a used in design codes. For example, the application of the modified
strength ranging from 20 to 40 MPa. The 5% quantile reached 0.812. failure criterion has brought about a higher level of safety of the CF
634
J. Halvoník, et al. Engineering Structures 188 (2019) 627–636
Fig. 12. Relation ratio P vs. reinforcement ratio ρ for the EC2 (2004), MC2010 LoAIII, CF CSCT models. The outliers are coloured in yellow. (For interpretation of the
references to colour in this figure legend, the reader is referred to the web version of this article.)
Fig. 13. Relation ratio P vs. effective depth d for the EC2 (2004), MC2010 and CF CSCT models. The outliers are coloured in yellow. (For interpretation of the
references to colour in this figure legend, the reader is referred to the web version of this article.)
CSCT model and, in the case of slabs cast from the most used concrete, Slovak Republic and the Slovak Academy of Sciences Contract No. 1/
the 5% quantile is greater than in the case of the current EC2 model. 0254/19. Many thanks to Carsten Siburg and Prof. Joseph Hegger from
RWTH Aachen University for providing the database of the tests.
Acknowledgement
Appendix A. Supplementary material
This work was supported by the Slovak Research and Development
Agency under Contract No. APVV-15-0658 and by the Scientific Grant Experimental database used for evaluation of the model’s reliability.
Agency of the Ministry of Education, Science, Research and Sport of the Supplementary data to this article can be found online at https://doi.
635
J. Halvoník, et al. Engineering Structures 188 (2019) 627–636
org/10.1016/j.engstruct.2019.03.055. 2008;105(2):163–72.
[5] Muttoni A, Fernández Ruiz M. The levels-of-approximation approach in MC 2010:
applications to punching shear provisions. Struct Concr Ernst Sohn, Germany
References 2012;13(1):32–41.
[6] Muttoni A, Fernández Ruiz M. The critical shear crack theory for punching design:
[1] EN1992-1-1 Design of Concrete Structures, Part 1-1 General Rules and Rules for from mechanical model to closed-form design expressions, Punching shear of struc-
Buildings, May 2004. tural concrete slabs. In: Honoring Neil M. Hawkins ACI-fib International Symposium,
[2] Fédération Internationale du Béton (fib), Model Code 2010 – Final draft, Vol. 1, April 2017. p. 237–52.
Fédération Internationale du Béton, Bulletin 65, Lausanne, Switzerland, vol. 2; 2012. [7] Muttoni A, Fernández Ruiz M, Simões JT. The theoretical principles of the critical
[3] Muttoni A, Schwartz J. Behaviour of beams and punching in slabs without shear shear crack theory for punching shear failures and derivation of consistent closed-
reinforcement, IABSE Colloquium, vol. 62, Zurich, Switzerland; 1991. p. 703–8. form design expressions. Struct Concr 2017:1–17.
[4] Muttoni A, Fernández Ruiz M. Shear strength of members without transverse re- [8] Zsutty T. Beam shear strength prediction by analysis of existing data. ACI J
inforcement as function of critical shear crack width. ACI Struct J 1968;65(11):943–51.
636