Construction and Building Materials 263 (2020) 120126

Contents lists available at ScienceDirect

Construction and Building Materials

journal homepage: www.elsevier.com/locate/conbuildmat

Revisiting the concept of characteristic compressive strength of concrete

Chi-Cong Vu a,⇑, Olivier Plé b, Jérôme Weiss c, David Amitrano c
a
National University of Civil Engineering (NUCE), 55 Giai Phong, Hanoi, Viet Nam
b
Univ. Savoie Mont Blanc, CNRS, LOCIE, 73000 Chambéry, France
c
Univ. Grenoble Alpes, CNRS, IRD, IFSTTAR, ISTerre, 38000 Grenoble, France

h i g h l i g h t s

Asymptotic strength r1 can be regarded as the genuine characteristic compressive strength of concrete.

Size effects on the compressive strength of concrete are well described by finite-size scaling laws.

Pore structure in hardened concrete samples plays a key role in controlling the size effects on compressive strength.

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

Article history: The characteristic compressive strength is one of the most important material property used in structural
Received 31 January 2020 design and quality control of concrete. Here we show that the estimation of this characteristic strength
Received in revised form 26 May 2020 following classical standard rules (e.g. EN 1992 or ACI-318) is sample size dependent, i.e. is not a charac-
Accepted 1 July 2020
teristic of the material. From an extensive experimental dataset, an analysis of published strength data,
and an interpretation of compressive failure as a critical phase transition from an intact to a failed state,
we show that the compressive strength of concrete is characterized by a non-vanishing asymptotic mean
Keywords:
strength, r1 , but a vanishing associated variability towards very large system sizes. Consequently, the
Characteristic compressive strength
Size effect
asymptotic strength r1 can be regarded as the genuine characteristic compressive strength of concrete.
Finite-size scaling Based on this, we propose a new procedure to evaluate this genuine characteristic strength as well as to
Concrete check the conformity of concrete with strength requirements.
Strength variability Ó 2020 Elsevier Ltd. All rights reserved.

1. Introduction [3] and NF EN 206-1 [4]) or on the specified compressive strength

0
f c in the American standard (ACI-318 [5]). Hereafter, both f ck
In civil engineering, concrete is mostly used under compression 0

loading configurations since its compressive strength is much lar- and f c are referred as the characteristic strength ðf ck Þ. For example,
ger than its tensile and/or flexural strengths. Generally, the tensile in the standard EN 1992 [3], the concrete strength class ranges
and flexural strengths of concrete are of the order of respectively from C12/15 to C90/105, meaning that the characteristic strength
~10% and ~15% of the compressive strength [1]. Furthermore, the of cylinders and that of cubic specimens are respectively in a range
compressive strength is often considered as a marker of the con- 12–90 MPa and 15–105 MPa. During the design stage, designers
crete quality because it is directly related to the structure of the must size the dimensions of the concrete members to resist the
hydrated cement paste [2]. For these reasons, the compressive imposed loads based on some compressive strength classes. These
strength is usually used as the basis for taking decisions regarding chosen strength grades are then used to define the concrete mix to
the strength and serviceability of concrete members and structures be used for construction. In the construction stage, the concrete
[1]. used must be in conformity with the strength requirements previ-
For the purpose of structural design, concrete is classified into ously specified by the designers from the characteristic strength.
several strength classes based on the so-called characteristic com- Therefore, the characteristic strength can be considered as a key
pressive strength ðf ck Þ in most of European standards (e.g. EN 1992 property for structural design, concrete mix design and quality
control of concrete.
⇑ Corresponding author at: Faculty of Civil and Industrial Construction, National Implicitly assuming a normal distribution for the compressive
University of Civil Engineering (NUCE), 55 Giai Phong street, Hai Ba Trung District, strength of concrete samples, in agreement with experimental data
Hanoi, Viet Nam. [1–3,5], the characteristic strength of concrete ðf ck Þ is calculated
E-mail address: [email protected] (C.-C. Vu).

https://doi.org/10.1016/j.conbuildmat.2020.120126
0950-0618/Ó 2020 Elsevier Ltd. All rights reserved.
2 C.-C. Vu et al. / Construction and Building Materials 263 (2020) 120126

from the corresponding mean value ðf cm Þ and standard deviation The statistical size effects on compressive strength of our concretes
ðsÞ as follows: are discussed in Section 4, and then compared with experimental
data of size effect on concrete strength available in the literature
f ck ¼ f cm  k:s ð1Þ
in Section 5. Results and discussion about the characteristic com-
where the constant k sets the acceptable percentage of tests that pressive strength, taking into account these statistical size effects,
will fail under a stress lower than f ck (e.g. a fractile of 5% is used are discussed in Section 6. Finally, some overall conclusions of this
in EN 1992 [3]). These two statistical parameters ðf cm andsÞ are study are given in Section 7.
obtained from an adequate number of uniaxial compression tests
(e.g.  30 tests recommended in ACI-318 [5]), carried out on stan-
2. Experimental program
dard concrete specimens at a minimum age of 28 days. Different
standard samples are used in different countries and sometimes
Here we present an extensive experimental program based on
even in the same country. For instance, both 150  300-mm (or
the physical and mechanical characterizations of 539 concrete
6  12-in.) cylinders and 150-mm (or 6-in.) cubes are used as stan-
cylindrical samples of 3 different concrete mixes and 4 different
dard specimens in the United States [1,5,6] and in most European
sizes including 527 tested in compression. While the moisture
countries [3], while 160  320-mm (i.e. 200 cm2 of normalized
and density data are presented here for the first time, the compres-
strength surface) or 110  220-mm (i.e. 100 cm2 of normalized
sive strength data have been used in [15] to analyze size effects on
strength surface) cylinders are recommended and used in France
both the mean strength ðf cm Þand the associated variability ðsÞ. Here
[4,7]. However, quasibrittle materials in general, and concrete in
these data are used to re-examine the concept of characteristic
particular, exhibit a size-dependent behavior on the nominal com-
strength, whose the classical definition directly relies on f cm and
pressive strength [8]. Precisely, the mean compressive strength
s (see Eq. (1)). This problem is then further discussed from com-
(f cm in (1)) decreases with increasing specimen size [9–15]. In addi-
pressive strength data available in the literature (Section 5). Our
tion, it has been shown recently that the strength variability of con-
dataset is however unique, as it allows a sound estimation of
crete (s in Eq. (1)) also decrease with increasing sample size [15,16].
strength variability, s, which is not the case for these previously
These size effects on both f cm and s imply, from relation (1), that the
published data.
so-called characteristic compressive strength ðf ck Þ will depend on
the size of tested concrete samples, an effect generally not taken
into account in concrete mix and structural design. Such size depen- 2.1. Materials and mix proportions
dence implies that f ck is actually not a characteristic of the material.
We proposed recently that the compressive failure of quasibrit- The cement used in all mixes was CEM I 52.5 N type Portland
tle materials can be interpreted as a critical phase transition cement, satisfying the standard NF EN 197-1 [20]. In this study,
between an intact and a failed state [16–19]. This interpretation locally available natural sand was used as the fine aggregate, while
enabled us to derive finite-size scaling laws for the mean compres- coarse aggregates were natural gravel. All the aggregates were dry
sive strength ðf cm Þ as well as the standard deviation ðsÞ which are in and clean, and their specific properties conform to the NF EN
remarkable agreement with strength data in various materials like 12,620 [21] and NF EN 1097-5 [22] regulations. The size distribu-
rocks, concrete, coal and ice [16]. In [15], from a large number tion of aggregates was investigated from a sieving analysis, follow-
(5 2 7) of uniaxial compression tests, conducted on concrete speci- ing NF EN 933-1 [23]. The gradation curves of aggregates used in
mens with three different mixtures (mean aggregate size and pro- this work are shown in Fig. 1. Ordinary potable water available
portion) and four different sizes, we demonstrated the pertinence in the laboratory was used for the mixing and curing of concrete.
of our critical interpretation as well as of the finite-size scaling Following the French standard NF EN 206-1 [4], the weight
laws to account for size effects on the compressive strength of con- method was applied to prepare three different concrete mixtures
 
crete. In particular, this experimental study indicated that (i) a very based on three different aggregate sizes dg (see Fig. 1 and
large system ðL ! þ1Þ will have a non-vanishing asymptotic Table 1): sand (3.15 mm), medium gravel (16 mm) and coarse


strength f cm ðL!þ1Þ ! r1 but a vanishing strength variability gravel (25 mm). The corresponding abbreviations for identifying
 
sðL!þ1Þ ! 0 , and (ii) the pore structure in hardened concrete
samples, rather than the concrete mix, plays a key role in control-
ling the size effects on compressive strength. These results suggest
two important points in terms of estimating the characteristic
compressive strength of concrete as follows:

1- In this framework, the asymptotic strength ðr1 Þ, which is by

construction independent of system size, appears as the gen-
uine characteristic compressive strength of concrete;
2- A laboratory concrete specimen can only be considered as a
standard specimen if its size is very large compared with the
characteristic size of its pore structure.

Following this statistical size effect analysis of the compressive

strength of concrete detailed in [15], we revisit here the concept of
characteristic strength by taking into account the two points noted
above, through a comparison with the classical expressions of
characteristic strength proposed in concrete design codes (EN
1992 [3] and ACI-318 [5]).
Section 2 details the experimental program. Section 3 deals
with the density and moisture content of our concretes, and their
variations with the sample size and microstructure characteristics. Fig. 1. Grading curves of aggregates for three concrete mixtures.
 C.-C. Vu et al. / Construction and Building Materials 263 (2020) 120126 3

Table 1
Mixture proportions for 1 m3 of concrete, density and moisture content.

Concrete Water Cement W/ Sand Medium Coarse Maximum aggregate size, Finesse Density, Moisture
mixture (kg) (kg) C (kg) aggregate (kg) aggregate (kg) dg (mm) modulus q(kg/m3) content,
wc ð%Þ
Mean SD Mean SD
F 225 450 0.50 1350 0 0 3.15 3.24 2184.0 20.5 5.3 0.4
M 195 335 0.58 800 1065 0 16 6.95 2391.4 14.7 3.2 0.2
C 195 335 0.58 800 0 1065 25 9.21 2403.8 18.2 3.2 0.3

each concrete group are: Fine aggregate (F), Medium aggregate (M) Fig. 2a). About 45 samples for each sample diameter ð/Þ and aggre-
 
and Coarse aggregate (C) (see Fig. 2b). The volume fraction of gate size dg , for a total of 539 concrete specimens, were pro-
aggregates for the coarser mixes (M- and C-concretes) was kept duced. Among these, 12 samples were selected for analyzing the
constant at 0.7 m3/m3 of concrete and the finesse modulus of initial microstructure (see Appendix A) and the 527 remaining
aggregates for the three mixtures are reported in Table 1. The samples were used for compression tests (see Section 4).
water-to-cement (W/C) ratio was set constant for all specimens All concrete samples were cast according to the procedure of
in each concrete mixture. The details of each mixture proportion normal weight concrete described in NF EN 206-1 [4]. During mix-
for 1 m3 of concrete are summarized in Table 1. While the ing, cement and aggregates were firstly blended in dry conditions,
F-concrete is a normal mortar in accordance with the standard and water was then added in the mixer. Due to the limited capacity
NF-EN 196-1 [24], the M-concrete can be actually considered as of the concrete mixing center, for a given concrete mixture, sam-
a standard concrete [1,4,21,25]. ples of 110 mm and 160 mm diameters were fabricated from dif-
ferent batches. The smaller samples (40 mm and 70 mm in
2.2. Specimen preparation diameter) were coming from a single batch, however different
from the batches used for larger samples. The different batches
In this work, all concrete specimens were cylinders with a fixed can be considered as a source of strength variability for each
height-to-diameter ratio h=/ ¼ 2 and, for each concrete mixture, material, which will be taken into account by our approach.
the diameter (/Þranged as follows: 40, 70, 110 and 160 mm (see According to the regulation NF EN 12390-2 [26], the concrete mix-

Fig. 2. Concrete specimens and testing setup: (a) Geometries of the four different sizes of concrete specimens; (b) Cross sections of the three different concrete mixtures; (c)
Uniaxial compression testing setup (Machine A); (d) Uniaxial compression testing setup (Machine B).
4 C.-C. Vu et al. / Construction and Building Materials 263 (2020) 120126

tures were poured in cylindrical cardboard molds and compacted affected by various factors [28]. According to [1], these factors
by an internal vibrator combined with an external vibrating table can be classified into four categories: (1) characteristics and pro-
to improve the consolidation of the samples. portions of materials, (2) curing conditions, (3) loading parameters,
After casting, all concrete specimens were cured initially for and (4) specimen parameters.
48 h by covering the molds with a plastic sheet in a moisture room, Within each concrete mixture, all the samples were produced
following the recommendation of the regulation NF EN 12390-2 from the same mix proportion as well as using the same casting
[26]. After demolding, the specimens were continually cured in a and curing process (Section 2.2). As an identical loading protocol
water basin at a temperature of approximately 20 °C for 2 months. was applied for all the tests (Section 2.3), we can neglect the influ-
At the age of 56 days, all the specimens were cut by diamond ence of the loading parameters on compressive strength. Regarding
grinding discs to ensure planar surfaces and prevent flexural stres- specimen parameters, this includes the sample size, and geometry,
ses when loading applies. After sawing, the concrete specimens the microstructural characteristics, the density and the moisture
were immediately immersed again in the water basin for an addi- state of hardened concrete [1,29,30]. As all our cylindrical concrete
tional month and left exposed to atmosphere in the laboratory samples were prepared with a constant of height/diameter ratio
until the testing day. According to the recommendation of the reg- ðh=/ ¼ 2Þ, the effect of specimen geometry is thus ignored in this
ulation NF EN 12390-3 [7], loading tests should be performed after study. The effect of sample size on the compressive strength will
a minimum age of 28 days. In this work, all the loading tests were be discussed in Section 4. The microstructural characteristics of
carried out five months after concrete preparation. our concretes (including porosity) have been already detailed else-
where [15]. We characterize in the appendix the pore structure
2.3. Testing procedure related to air voids, which plays a major role on size effects on
strength (see Section 6.2). In the section below, we examine the
A series of 527 uniaxial compressive tests was carried out under density and moisture state of our hardened concrete samples,
a load control protocol, following the procedure recommended by which are potentially affected by the curing conditions [1,2]. The
the regulation NF EN 12390-3 [7]. As the samples were very differ- influences of sample size and microstructure on these properties
ent in size, two load-control uniaxial compression machines (of are also described.
different stiffness and loading capacity), complying with the stan-
dard NF EN 12390-4 [27], were used. Machine A (Fig. 2c) with a 3.1. Density
capacity of 3000kN was used for the three larger sample sizes
(70  140-mm; 110  220-mm and 160  320-mm cylinders). Before each loading test, the dimensions and the weight of the
The stiffness of this machine is 2.9 times stiffer than that of the lar- sample were measured. An average diameter, /, was computed
gest samples. For the remaining size (40  80mm-samples), the from six measures (two for each end of sample and two in the mid-
uniaxial compression tests were conducted with machine B dle of the sample length). The average length, h, was obtained from
(Fig. 2d). This machine has a capacity of 300kN and its stiffness three measurements. Following the standard NF EN 12390-7 [31],
is 3.5 times larger than the stiffness of the small samples. Conse- the apparent density, q, of the hardened concrete sample is simply
quently, the compression machines used in this work are stiff given by:
enough to perform reliable strength measurements on our sam- m
ples. In addition, these two machines have also been verified and q¼ ð2Þ
V
certified by the French Accreditation Committee (COFRAC), i.e.
are suited for standardized tests. where m is the mass of the concrete sample and V the apparent vol-

2

A constant stress rate of 0.5 MPa/s, corresponding to a strain ume V ¼ h p4/ .
rate ranging from 2.4  105/s to 3.2  105/s was applied on
Fig. 3a shows the mean density and its standard deviation for all
the concrete samples placed between two steel-hardened platens
concrete samples. It is observed that: (i) for a given concrete mix-
of the compression machine until the specimen failed. Loads were
ture, the mean apparent densities and the associated standard
continuously measured by the load cell positioned at the top pla-
deviations are similar for all sample sizes. In other words, there
ten (see Fig. 2c) and directly transmitted to the data acquisition
is no significant sample size effect on density; (ii) there is a clear
system. When the load fell below 50% of the peak load, loading
correlation between the apparent density and the aggregate grada-
was automatically stopped. While the bottom steel platen was
tion (F-, M- and C-concretes) (see Fig. 3a and Table 1), with an
fixed, the top one could rotate. This enables the upper platen to
increasing density when increasing the aggregate size, as expected;
adjust to the geometrical imperfections of the specimens.
and (iii) the apparent densities range from 2180 kg/m3 to 2400 kg/
During loading, the load ðF Þ and the axial displacement ðdÞ were
m3, i.e. all our concrete samples comply with the definition of
continuously monitored and recorded at a 5 Hz frequency. The
normal-weight concrete [1–4]. Point (i) shows that the size effects
axial displacement of the bottom platen ðdÞ was measured by
on strength discussed below in Section 4 cannot result from a size
one Linear Variable Differential Transducer (LVDT) attached on
effect on density, as this characteristics is size-independent in our
the frame of the machine. However, when a concrete specimen is
samples.
loaded, both the specimen and the machine deform. As a result,
the measured displacement ðdÞ is larger than the true axial short-
ening of the specimen ðDsp Þ. By using a calibration test on an Alu- 3.2. Moisture content
minum sample of known elastic modulus, the elastic deformation
In hardened concrete, water is present in various states: (i)
of the loading frame was determined as a function of the applied
chemically bounded water within the hydration products, (ii)
load. The associated displacement was then eliminated from the
adsorbed or physically bounded water in gel pores (gel water),
axial displacement ðdÞ to obtain Dsp .
and (iii) free water in capillary pores (capillary water) [32]. These
different types of water content control the moisture state inside
3. Density and moisture content of hardened concrete hardened concrete samples and have a considerable effect on the
mechanical properties of concrete [1,2,33]. In general, the moisture
As a multiphase heterogeneous material, the response of con- content, which depends on the microstructural characteristics of
crete to applied stress is a result of complex interactions and is the material and environmental conditions such as temperature,
 C.-C. Vu et al. / Construction and Building Materials 263 (2020) 120126 5

Fig. 3. Properties of hardened concrete samples as a function of the sample diameter: (a) Apparent density; (b) Moisture content.

relative humidity, and air velocity [1], is used to quantitatively (5 months). In other words, this enables us to conclude that all
characterize the moisture state of concrete. loading tests for a given concrete mixture were performed under
In this study, the moisture content, wc , of the concrete samples similar initial conditions. Consequently, the size effects on strength
was investigated from the following procedure: discussed below in Section 4 must result from some other reasons,
and cannot be affected by the density or the moisture content of
1. Right after loading, 10 deformed concrete samples were col- our samples.
lected for each sample size and each concrete mixture, and
weighted. This weight is noted m1 ;
2. The deformed concrete samples were kept in a drying oven at 4. Size effects on compressive strength
(105 ± 5)oC for a minimum duration of 3 days, according to
the regulation NF EN 12390-7 [31]. After this 3-days period, The compressive strength of concrete (rf ) is defined as the max-
the mass of each dried specimens was measured every 2 h. imum stress that the concrete sample can withstand [1,6,7], hence
When no significant change of the mass was detected over is calculated by dividing the maximum load, F max , carried by the
three consecutive measurements, the sample was considered concrete specimen during the test by the average cross-sectional
as totally dry and its weight called m2 ; This procedure was done area ðrf ¼ 4F max =p/2 Þ. Fig. 4 shows the histograms and the corre-
on samples collected after loading, as the moisture content (and sponding fits to a normal distribution for the compressive strength
so, in particular, such drying procedure) is known to play a sig- of our concrete samples. Low values of both the skewness and the
nificant role on strength. kurtosis of the distributions (see Table 2) demonstrate that the
3. The moisture content of the hardened concrete sample is compressive strength of our concretes is distributed according to
defined as the mass ratio of water to solid phases in the sample Gaussian statistics. For each individual distribution (fixed sample
[34]: wc ¼ ðm1mm
2
2Þ
 100% size and concrete mixture), we also performed Shapiro-Wilk tests
with a ¼ 0:01 and found that the assumption of Gaussian statistics
The mean value and standard deviation of moisture contents for was never rejected.
all concrete samples are displayed in Fig. 3b. For a given concrete For a specific concrete mixture and each sample size, the mean
group, no significant sample size dependence can be observed. In value and the standard deviation of the compressive strength were
contrast with the density (see Fig. 3a), the largest moistures are calculated from about 44 tested samples. These values are summa-
observed for the F-concrete (see Fig. 3b and Table 1). The moisture rized in Table 2. We observe that both the mean compressive
 
content is generally affected by the environmental humidity, the strength, hrf i, and the associated variability, d rf , decrease with
cement content and the porosity of the hardened cement paste increasing sample diameter (/) (see Fig. 4) and increasing aggre-
[1]. In our case, all loading tests were performed on the concrete gate size. This is the signature of the size effects on strength men-
samples during a short period (an average of 50 loading tests per tioned in the introduction.
day). Hence, the environmental conditions can be considered sim- An energetic, deterministic approach has been proposed to
ilar for all samples of each concrete group. The F-concrete was pre- explain size effects on strength in quasi-brittle materials like con-
pared with the largest water to cement ratio in the mix (see crete [8,36]. It is related to stress redistribution and associated
Table 1). This is the main reason to explain why the F-concrete energy release within a fracture process zone (FPZ), before
samples show the highest moisture content. macroscopic failure. In this framework, size effects on strength
Compressive strength is known to increase with decreasing ensue when the FPZ becomes non-negligible compared with the
moisture in hardened concrete [2,35]. However, owing to the low structure size. By construction, such a deterministic approach
moisture contents (<5.5%) observed in our concrete samples (see cannot deal with statistical properties such as strength variabil-
Table 1), this factor is not expecting to play a significant role on ity, or the probability of failure under a given stress, while these
strength in this study [2,33]. In addition, the size-independence concepts are essential in the definition of the characteristic
of density and moisture content demonstrates that at the time of strength (see Eq. (1)). The statistical approach proposed here is
loading tests, all concrete samples of a given concrete mixture fundamentally different. For a more detailed comparison
were in a stable moisture state after a long period of curing between them, see [15].
6 C.-C. Vu et al. / Construction and Building Materials 263 (2020) 120126

Fig. 4. Histograms and normal distribution fits of the compressive strength of concrete samples: (a) F-concrete; (b) M-concrete and (c) C-concrete.

Table 2
Compressive strength of concrete samples.

Concrete group Specimen size, /  h (mm  mm) Compressive strength, rf (MPa)

Mean SD Skewness Kurtosis
F 40  80 56.4 7.9 0.21 0.74
70  140 49.7 5.3 0.08 0.88
110  220 44.3 2.4 0.12 0.06
160  320 41.9 2.4 0.29 0.94
M 40  80 49.6 6.2 0.31 0.73
70  140 45.2 4.2 0.58 0.64
110  220 41.9 1.9 0.04 0.62
160  320 39.2 1.8 0.14 0.66
C 40  80 39.5 4.6 0.47 0.27
70  140 40.0 6.3 0.05 0.86
110  220 37.1 2.7 0.35 0.66
160  320 36.7 1.4 0.59 0.05

The simplest statistical theory of failure is based on the weakest- depinning transition of an elastic interface moving against a field
link hypothesis [37,38], which assumes that microfracturing and of obstacles (e.g. [39,40]). In the case of concrete, the critical nature
damage events do not interact during the rupture process. Recently, of compressive failure was investigated and confirmed from an
we have demonstrated the inadequacy of this assumption [15], acoustic emission (AE) survey of damage and microfracturing [19].
hence the inability of the weakest-link theory to describe these In this framework, finite-size effects on strength arise when the
external size effects on the compressive failure and strength of correlation length between microfracturing events, which grows
heterogeneous materials such as rocks, concrete, etc. Instead, this during loading, starts interacting with the (finite) sample size as
failure process can be interpreted as a critical phase transition from approaching failure. This gives rise to the finite-size scaling laws
an intact to a failed state [16]. This is based on an analogy with a gen- for the mean value of the compressive strength, hrf i, and its stan-
 
eric physical framework for such a critical transition, the so-called dard deviation, d rf as follows [16,41]:
 C.-C. Vu et al. / Construction and Building Materials 263 (2020) 120126 7

 1=mFS
L material–independent value of r1 ¼ 36:9MPa is obtained from the
hrf i ¼ r1  þ r1 ð3Þ h i
Lm linear fit on a hrf iv s:ð/=Lm Þ1=mFS plot (see the inset graph of

 1=mFS Fig. 5a). Taking this asymptotic strength value r1 as well as the
  L values of Ld reported in Table 3, we also obtain an excellent agree-
d rf ¼ r1  ð4Þ
Ld ment between the scaling prediction for the standard deviation
(Eq. (4)) and our experimental data (all sample sizes and all con-
where mFS is the finite-size exponent. The mean-field (material- crete mixtures) (see Fig. 5b). This last point shows that our frame-
independent) prediction [39] for this exponent is mFS ¼ 1. The length work adequately takes into account various sources of strength
scales Lm and Ld are material constants, which are related to the variability, including the internal microstructure, moisture content
characteristics of the microstructural disorder [15,16]. In this con- variability (see Section 3.2), as well as the effect of different
text, r1 is a non-vanishing asymptotic strength for a system of infi- batches (see Section 2.2).
nite size ðL ! þ1Þ. Eq. (3) resembles the scaling laws for the mean Fig. 5 reveals that, when a structure is of size L much larger than
strength proposed by Bazant [8,12] and Carpinteri [42]. However, it its microstructural disorder related scale Lm , (i) the mean compres-
is noteworthy that, in the case of the Bazant’s approach, the expo- sive strength hrf i is not affected by either the concrete mix or the
nent is a free (empirical) parameter extracted by fitting the exper- pore structure; (ii) the mean strength hrf i is close to the asymp-
imental data (see reviews in [8,11,43]), while our physical  
totic strength r1 , while the associated variability d rf is low. In
approach gives a theoretical prediction for mFS . The approach of
other words, the asymptotic compressive strength ðr1 Þ can be
Bazant is based on a ‘‘deterministic” energetic size effect and that
safely estimated from the mean compressive strength of concrete
of Carpinteri is based on geometrical arguments of a fracture sur-
samples whose the size ðLÞ is large enough to neglect the effect
face at the ultimate load, while our approach, based on a statistical
of the pore structure ðLm Þ.
physics framework (the critical depinning transition), is fully differ-
ent. It is noteworthy that both previous approaches, by construc-
tion, do not give any prediction for the size effects on the 5. Extension to published datasets
associated strength variability, while our approach does (see Eq.
 
(4)). The comparison of our approach with these two previous The size effects on the compressive strength rf of our con-
works was further detailed in a recent paper [15]. cretes were found to be in remarkable agreement with finite-size
A remarkable agreement between Eq. (3) and numerous pub- scaling predictions for both the mean value (Eq. (3)) and its stan-
lished data on the mean compressive strength of various materials dard deviation (Eq. (4)). In order to further validate our approach,
such as rocks, ice, coal, and concrete, argued for this critical inter- we analyze below different experimental data coming from the lit-
pretation of compressive failure of heterogeneous materials [16]. erature. We focus (i) on strength datasets obtained from cylindrical
In order to describe the size effects on the compressive strength samples with a height-to-diameter ratio h=/ ¼ 2similar to our con-
for our concretes, this critical interpretation was also applied in ditions, and (ii) on the size effect on the mean strength hrf i, as the
[15]. Once again, a full agreement between the finite-size scaling limited number of independent tests in these previous works did
predictions (Eq. (3) and Eq. (4)) and the compressive strength data not allow an analysis of strength variability. Because the qualities
was achieved, regardless of the concrete mixture. For each concrete of the aggregates in these previous studies as well as their W=C
mixture, the asymptotic strength r1 and the length scales Lm;d ratios are different from our study, a comparison of the asymptotic
were extracted from the best fits of the scaling on the mean com- strength ðr1 Þ is not possible. In addition, as the microstructural
 
pressive strength hrf i (Eq. (3)) and the standard deviation d rf characteristics of the concretes, and particularly the pore content,
(Eq. (4)), using mFS ¼ 1 (see [15] for more details). The values of were not detailed in these references, the role of these hetero-
r1 and of the length scales Lm;d for our three concrete mixtures geneities on compressive strength is not examined.
are recalled in Table 3. While the length scales Lm;d strongly vary The first example considered refers to the tests carried out by
the concrete mix, the asymptotic strength appears almost indepen- Dehestani et al. [44] on cylindrical samples of self-consolidating
dent of the material. In [15], we also reported that the pore struc- concrete (SCC) under uniaxial compressive loading. In their work,
ture of the material, rather than the aggregate content, plays a three different concrete mixes were fabricated using three differ-
significant role on size effects on the compressive strength of con- ent W=C ratios, but with a fixed maximum aggregate size of
crete, i.e. the length scales Lm;d are linked to the characteristics of 12.5 mm and a cement proportion of 400 kg/m3. Cylindrical spec-
the pore structure in hardened concrete samples. imen diameters were 50, 75, 100, and 150 mm. The experimental
In [19], from a campaign of acoustic emission (AE) measure- results are summarized in Table 4. Using mFS ¼ 1, Eq. (3) well
ments carried out on our concrete samples during loading (30 describes the size effect on compressive strength for these three
tests), the critical nature of compressive failure process was ana- SCC mixes (see Fig. 6a, b, c). A similar agreement is obtained for
lyzed and confirmed. In particular, this AE analysis gave an exper- another strength data of normal-weight concrete provided by
imental value of mFS ¼ 1:1  0:2, very close to the theoretical value. Muciaccia et al. [13], reported in Table 4 and Fig. 6d (same range
Consequently, using mFS ¼ 1, the values of Lm reported in Table 3, of W=C ratio). The corresponding asymptotic strength ðr1 Þ and
and taking L ¼ / in Eq. (3), we obtain a remarkable fit of the mean length scale ðLm Þ for each of these four concrete mixtures were
strength scaling (Eq. (3)) for all the concrete samples (see Fig. 5a). A obtained from non-linear fits (Eq. (3)) and summarized in Table 4.

Table 3
Finite-size scaling parameters (Eqs. (3) and (4)) (after [15]), and estimated parameters of the characteristic strength expressed according to EN-1992 and ACI 318.

Concrete mixture Finite-size scaling EN 1992 ACI 318

r1 (MPa) Lm (mm) Ld (mm) Li Lmin (mm) Li Lmin (mm)
Mean SD Mean SD Mean SD Mean SD Mean SD
F 37.5 4.2 20.5 0.4 8.8 0.2 6.3 2.0 631.0 9.0 1.7 897.0
M 36.7 4.5 14.4 0.6 7.0 0.2 3.1 2.5 305.0 5.2 2.5 517.0
C 36.2 5.9 4.1 2.5 6.9 0.8 7.1 6.2 319.0 5.0 4.7 224.0
8 C.-C. Vu et al. / Construction and Building Materials 263 (2020) 120126

 
Fig. 5. Finite-size effect on the uniaxial compressive strength of concrete: (a) for the mean strength, hrf i and (b) for the standard deviation of strength, d rf . Main graphs
 
show the relationship between hrf i and d rf with the ratio of sample size ð/Þ to the length scales Lm and Ld , respectively. Black dashed-lines are the fits by Eq. (3) for the
h i h   i
mean strength (a), and Eq. (4) for the standard deviation of strength (b). Insets show the same data and fits on hrf iv s:ð/=Lm Þ1=mFS and d rf v s:ð/=Ld Þ1=mFS plots where the
fits by Eq. (3) and Eq. (4) are the straight lines.

Table 4
Concrete mixture proportions, sample sizes and mean compressive strengths for the three concrete mixtures (M1, M2, M3) of the experiments presented by Dehestani et al. [44],
and for the mixture R2 of Muciaccia et al. [13].

Concrete mix data Cylinders Mean compressive Asymptotic strength, Lm (mm)

ðh=/ ¼ 2Þ strength, hrf i (MPa) r1 (MPa)
Mixture code W/C dmax (mm) /(mm) Mean SD Mean SD

M1a 0.35 12.5 50 51.3 43.7 8.6 9.6 6.4

75 51.0
100 48.1
150 45.6
M2a 0.39 12.5 50 41.7 29.7 3.1 19.7 9.7
75 36.9
100 35.5
150 33.9
M3a 0.43 12.5 50 28.4 22.1 1.5 14.6 5.8
75 26.5
100 25.5
150 24.0
R2b 0.41 20.0 100 54.4 25.0 13.8 90.5 37.6
200 45.7
400 37.0
800 22.8
a
Self-Consolidating Concrete (SCC) mixture;
b
Normal-weight concrete mixture.

As shown in this table, the Lm value for the three SSC groups SSC mixtures were prepared with three different weight percent-
M1 ; M 2 ; M 3 varies with the W=C ratio, while the maximum aggre- ages of HRWR and three different W=C ratios while keeping con-
 
gate size dg was kept constant. This is consistent with our previ- stant the cement content (400 kg/m3) for all mixes. This could
ous observations presented in Section 4, confirming that the size lead to a different nature of the binder matrix which caused the
effect on compressive strength of concrete is not controlled by different values of asymptotic strength for the three SSC groups.
the aggregate size. As already mentioned above, the asymptotic For the results provided by Muciaccia et al. [13], a comparison of
strengths ðr1 Þ reported in Table 4 are not constant and signifi- the asymptotic strength is not possible as well because the quality
cantly different from our results. (see Table 3). This could be of aggregates as well as the W=C ratio are different.
explained by different W/C ratio and the use of high range The second example considered is the size effect on the com-
water-reducing (HRWR) admixture type superplasticizer during pressive strength of concrete observed by Sim et al. [50]. From
the casting of the SSC groups (see Table 4 and Fig. 6a, b, c). The their data, we considered here only three different concrete mix-
main task of using a superplasticizer is to increase the fluidity of tures (named A43-0, S47-100, and N55-100) for which they per-
concrete by dispersing cement particles in paste without adding formed mechanical test with the largest range of cylindrical
excess water [45–47]. For a given cement content, the use of super- sample sizes ð/ ¼ 50; 100; 150; 250; 300; 350; 400mmÞ and with
plasticizer enables to achieve a high concrete strength and durabil- an aspect ratio of h=/ ¼ 2. In accordance with ACI 213R-03 [51],
ity due to the reduced amount of free water (low W=C ratio) in the mixtures A43-0, S47-100 and N55-100 are classified following
mixture [45,48,49]. In the work of Dehestani et al. [44], the three their densities ðqÞ as lightweight concrete, mortar and normal-
 C.-C. Vu et al. / Construction and Building Materials 263 (2020) 120126 9

Fig. 6. Finite-size effect on the uniaxial compressive strength (experimental data): for three self-consolidating concrete (SSC) mixtures (a) M1, (b) M2, and (c) M3 in ref. [44];
and (d) for normal-weight concrete mixture (R2) in ref. [13]. Main graphs show the relationship between hrf i and the sample diameter ð/Þ. Black symbols are the published
h i
experimental data. Red curves are the fits by using Eq. (3) for the mean strength with mFS ¼ 1. Insets show the same data and fits on hrf iv s:ð/Þ1=mFS plots where the fits by
Eq. (3) are the straight lines. The best-fit asymptotic strength r1 and the associated length scale Lm for each concrete group are listed in Table 4. (For interpretation of the
references to color in this figure legend, the reader is referred to the web version of this article.)

weight concrete, respectively. The details of these concrete mixes tures. This confirms that (i) the asymptotic strength r1 is not
are summarized in Table 5. The A43-0 concrete mix used expanded affected by the concrete mixture if using the same nature of aggre-
clay granule as the aggregate, while the N55-100 mix used granite gates for casting; and (ii) all the three concrete mixtures were
(coarse aggregate) and natural sand (fine aggregate) for casting. As batched for the same target concrete grade. As mentioned by Sim
the nature and the quality of the aggregates in the mixture are dif- et al. [50] ‘‘The targeted compressive strength of all ready-mixed
ferent from our work and previously detailed studies, a comparison concrete batches was 35 MPa”, meaning that, according to the
of the asymptotic strength ðr1 Þ is not possible. However, the max- EN 1992 regulation [3], the characteristic compressive strength
imum coarse aggregate size was fixed at 19 mm for both mixtures. for all three considered mixtures is f ck ¼ f cm  8 ¼ 27MPa. This
Only natural sand with a maximum size of 5 mm was used to cast f ck -value is close to the asymptotic strength r1 extracted from
the S47-100 mixture. The size effect on mean compressive strength the non-linear fits of experimental data (see Fig. 7d). In other
and the corresponding best-fit following Eq. (3) with mFS ¼ 1 for the words, the asymptotic strength r1 can be regarded as the charac-
three experimental strength data are shown in Fig. 7a, b, c. The teristic strength f ck of concrete. This observation will be discussed
agreement with the scaling prediction (Eq. (3)) is remarkable. further in Section 6.
The corresponding r1  and Lm  values for these three concrete The last example is offered by the experimental campaign of
mixtures are presented in Table 5. We confirm an excellent col- Blanks and McNamara [52] on large concrete cylinders. In their
lapse of all data (all sample sizes, all concrete mixtures) on a work, different concrete mixtures obtained by varying the maxi-
 
½hrf iv s:ðL=Lm Þ1=mFS  plot (Fig. 7d), in full agreement with our previ- mum aggregate size dg were tested for various sizes with a con-
ous results (Fig. 5a). A corresponding asymptotic strength of stant aspect ratio ðh=/ ¼ 2Þ. We consider here only three concrete
r1 ¼ 28:0  0:6ðMPaÞ is obtained for these three concrete mix- mixtures named B-8, B-9 and B-10 as they were associated with
10 C.-C. Vu et al. / Construction and Building Materials 263 (2020) 120126

Table 5
Concrete mixture proportions, sample sizes and mean compressive strengths for the three concrete mixes (A43-0, S47-100, N55-100) of the experiments reported by Sim et al.
[50].

Concrete mix data Cylinders Mean compressive Asymptotic Lm (mm)

ðh=/ ¼ 2Þ strength, hrf i (MPa) strength, r1
(MPa)
Mix notationa W/C dmax (mm) Density, q(kg/cm3) /(mm) Mean SD Mean SD

A43-0 0.43 19 1661 50 50.9 27.6 4.1 46.2 23.0

100 43.6
150 39.4
250 33.1
300 30.2
350 30.0
400 28.6
S47-100 0.47 5 1920 50 50.7 28.6 2.7 41.3 14.1
100 43.4
150 37.8
250 33.0
300 32.2
350 31.4
400 30.0
N55-100 0.55 19 2314 50 49.9 27.9 1,6 40.5 8.9
100 40.5
150 35.5
250 33.8
300 30.8
350 30.9
400 29.8
a
In the concrete mix notation, the first part indicates the type of concrete (A: lightweight concrete; S: mortar; N: normal-weight concrete), while the second and third parts
give the water-to-cement ratio and replacement sand. For example, A43-0 indicates an all-lightweight concrete mix with W=C ¼ 0:43 and do not use natural sand for casting.

the widest range of sample size. Details on the mix proportions, 6.1. Asymptotic versus characteristic compressive strength
specimen sizes and the corresponding 28-days mean compressive
strength hrf i can be found in [52]. Fig. 8 shows the best-fitted Owing to the vanishing strength variability for a system of infi-
finite-size scaling (Eq. (3), taking mFS ¼ 1) for the mean compressive nite size ðL ! þ1Þ, the asymptotic strength r1 (see Eq. (3)) can be
strength of the three considered concrete mixtures. The values of considered as the genuine characteristic strength of concrete to be
the asymptotic strength r1 as well as the length scale Lm for the used to design large scale structures. However, as mentioned in
three considered concrete mixtures, obtained from non-linear fits Section 1, the characteristic strength is instead classically calcu-
of Eq. (3), are presented in Table 6. We observe that the scaling pre- lated from the mean value and the standard deviation of compres-
diction (3) is in remarkable agreement with the published experi- sive strength from the relation (1). In EN 1992 [3], the
mental data of [52]. Using v FS ¼ 1 and the values of Lm in Table 6, characteristic strength ðf ck Þ is defined as that strength value below
h i
we plot all strength data on a hrf iv s:ðL=Lm Þ1=mFS plot. As shown in which only 5% of the compression tests are expected to fall. This
corresponds to k ¼ 1:645 in Eq. (1) (see Fig. 9). In ACI 318 [5],
Fig. 8d, a good collapse of all data (all sample size, all considered
the tolerance is larger (10%), corresponding to k ¼ 1:34 (see Fig. 9).
concrete mixtures) is obtained, with an asymptotic strength
Hence, from the mean strengths hrf i and standard deviations
r1 ¼ 22:4  0:3ðMPaÞ independent of the concrete mixture. These  
d rf reported in Table 2, we can compute the characteristic
observations, are consistent with our results (see Section 4) and
strengths for each concrete mixture and sample size, as defined
validate our theoretical framework.
by the EN 1992 and the ACI 318 codes, using respectively
Overall, the analysis of these previously published datasets is
k ¼ 1:645 and k ¼ 1:34 in Eq. (1).
fully consistent with our results detailed in Section 4, validating
Fig. 10 shows that the so-defined ‘‘characteristic” strength (i)
further our theoretical framework.
differs for the three different concrete mixtures, and, more surpris-
ingly, (ii) is sample size dependent. If point (i) could be expected at
6. Characteristic compressive strength of concrete first glance, it results in our case only from different finite size
effects for our three concretes (different Lm -values), as the asymp-
Concrete design codes evaluate the characteristic compressive totic strength r1 is essentially independent of the concrete mix for
strength from subtracting the strength variability from the mean the three materials prepared for this study (see Fig. 5). It is worth
compressive strength obtained from a set of tested concrete speci- noting that this does not mean that r1 ¼ 36:9MPa would be a
mens with a unique size (see Eq. (1)). However, due to the sample characteristic strength for all types of concrete. As already dis-
size effects on both the mean value and the associated variability cussed in [15], the value of r1 will be of course different if we
of compressive strength (see Section 4), this traditional estimation change the nature of either the fresh or hardened states of concrete
necessarily leads to size effects on the characteristic compressive (e.g. changing the type of aggregate or adding some kinds of con-
strength f ck as estimated from Eq. (1). This questions the concept crete admixtures). This is illustrated by the fact that the asymptotic
of characteristic strength itself, which should be representative of strength obtained for our concretes (Table 3), for the work of Sim
the material only and independent of external size. In this section, et al. [50] (Table 5), or that of Blanks and McNamara [52] (Table 6),
we show that the asymptotic strength, deduced from the external are significantly different, while essentially independent of the
size effects towards the limit L ! þ1, represents the genuine char- aggregate content within each of these studies.
acteristic strength of concrete. We also propose some recommenda- Point (ii) strongly challenges the classical definition of the
tions to determine accurately this asymptotic strength as well as to ‘‘characteristic” strength, which should be a size-independent
check the conformity of concrete with strength requirements. material characteristic. This is in full contrast with our defini-
 C.-C. Vu et al. / Construction and Building Materials 263 (2020) 120126 11

Fig. 7. Finite-size effect on the mean compressive strength for three different concrete mixtures of ref. [50]: (a) A43-0; (b) S47-100; (c) N55-100. Main graphs show the mean
compressive strength, hrf ias a function of sample diameter ð/Þ. Black symbols are the experimental data reported by [50]. Red curves are the fitting by Eq. (3). Insets show the
h i
same data and fits in a hrf iv s:/1=mFS graph. In this smaller graph, the fitting by Eq. (3) becomes a straight line and the asymptotic strength r1 is determined. In (d), the
collapse of all data for all concrete mix with a common asymptotic strength (r1 ¼ 28:0  0:6Þis shown. (For interpretation of the references to color in this figure legend, the
reader is referred to the web version of this article.)

tion of the asymptotic strength r1 , which is, by construction, though the 160  320-mm samples provide the values of the char-
sample size independent. The differences of the f ck values for acteristic strength that are the closest to r1 , the accuracy of the
the three concretes and different sample sizes obtained by estimation of r1 depends on the confidence level chosen (i.e. the
applying Eq. (1) from our strength dataset are actually the per- constant k in Eq. (1)), as well as on the concrete mix. Taking
 
fect illustration of the limitation of the classical methodology f cm ¼ hrf i, s ¼ d rf , mFS ¼ 1, a combination of Eq. (3) and Eq. (4)
used to determine the characteristic strength. In addition, we with Eq. (1) yields the following scaling for the characteristic
note that for most of the F-concrete and M-concrete dataset strength ðf ck Þ:
(one material, one sample size) the characteristic strength
 
ðf ck Þ calculated from the structural design codes are larger than Li
f ck ¼ r1  þ r1 ð5Þ
the asymptotic strength r1 , while an opposite result is L
observed for the C-concrete. This illustrates that the use of clas-
sical structural design codes and small concrete samples can where Li ¼ Lm  kLd . Considering the corresponding strength r1 for
lead to unreliable estimations of the asymptotic strength, rele- each concrete mixture (Table 3), Eq. (5) fits well the f ck values for all
vant for large scale structures. These classical methods give concrete mixtures, and for both the EN 1992 [3] and ACI 318 [5]
sometimes unsafe estimations (e.g. F-and M-concretes) (see design codes (Fig. 10). For a given concrete mixture, the minimum
Fig. 10a and b), or occasionally too conservative ones (C- sample size ðLmin Þ that would allow a correct estimation of the
concrete) (see Fig. 10c). asymptotic strength (r1  0:01r1 ) while using classical design
In most structural design codes, 150  300mm cylindrical sam- codes are reported in Table 3. These sample sizes are always larger
ples [3,5,6] (or 160  320mm in France [7]) are used as a standard than 160 mm and depend on the concrete mixture as well as the
to determine the characteristic compressive strength. Here, even constant k, i.e. on the code. In some cases, e.g. F-concrete, to per-
12 C.-C. Vu et al. / Construction and Building Materials 263 (2020) 120126

Fig. 8. Finite-size effect on the mean compressive strength for: (a) B-8; (b) B-9; (c) B-10 concrete mixtures of ref. [52]. Main graphs show the mean compressive strength,
hrf ias a function of specimen size. Black symbols are the experimental data reported by [52]. Red curves are the fitting by Eq. (3). Insets show the same data and fits in a
h i
hrf iv s:/1=mFS graph. In this smaller graph, the fitting by Eq. (3) becomes a straight line and the asymptotic strength r1 is determined. In (d), the collapse of all data for all
concrete mix with a common asymptotic strength (r1 ¼ 22:4  0:3Þis shown. (For interpretation of the references to color in this figure legend, the reader is referred to the
web version of this article.)

form compression tests on samples large enough to give a correct 6.2. Recommendations for the determination of the asymptotic
estimate of r1 would require an unusually large loading capacity compressive strength

2

N
for laboratory testing (e.g., 37:5 mm 2  p  8974 mm2  24  106 N
As discussed above, the form of the finite-size scaling laws for
for F-concrete with the ACI 318 code).
both the mean strength (Eq. (3)), with a non-vanishing asymptotic
This raises important problems while trying to estimate a reli-
strength ðr1 Þ;and the associated variability ðdðrÞÞ vanishing at
able characteristic strength of concrete for large scales structural
large scales, implies that r1 can be considered as the genuine char-
design from classical codes and a unique series of tests with a sin-
acteristic strength ðf ck Þ of concrete. We have also stressed that
gle sample size. To avoid sample size effects on strength, the sam-
using a single concrete sample size for laboratory testing leads to
ple size L should be much larger than Lm which itself depends on
different estimations of the characteristic strength from classical
the microstructural characteristics (i.e. pore structure) of the hard-
codes. Therefore, performing a series of compression tests with dif-
ened concrete.
ferent sample sizes is highly recommended to reliably estimate the
On the other hand, the asymptotic strength ðr1 Þ could be
characteristic (asymptotic) compressive strength of concrete. We
approximately determined by the mean strength hrf i of samples
suggest below an experimental procedure:
whose the size ðLÞ is much larger than the characteristic size of
the pore structure ðLm Þ (see Fig. 5a and Section 4). Based on this
1. Prepare cylindrical concrete samples with a length-to-diameter
idea, some recommendations for predicting the genuine character-
ratio of 2 and at least four different sizes. This condition is
istic strength of concrete, r1 , are presented in the following
required to perform a good fit of the finite-size scaling laws
section.
 C.-C. Vu et al. / Construction and Building Materials 263 (2020) 120126 13

Table 6
Concrete mixture proportions, specimen sizes and mean compressive strengths for the experiments reported by Blanks and McNamara [52].

Concrete mix data Cylinders Compressive Asymptotic strength, Lm (mm)

ðh=/ ¼ 2Þ strength, hrf i (MPa) r1 (MPa)
No. W/C dmax (mm) /(mm) Mean SD Mean SD

B-8 0.53 38.10 152.4 27.92 21.8 1.3 40.1 5.2

203.2 25.72
304.8 24.75
457.2 22.89
609.6 23.44
914.4 23.30
B-9 0.55 19.05 50.8 27.99 22.5 1.4 14.5 2.1
76.2 27.72
152.4 25.86
203.2 23.92
304.8 23.30
457.2 23.17
609.6 22.13
B-10 0.54 9.52 50.8 29.03 22.9 0.8 13.8 1.3
76.2 27.10
152.4 24.48
203.2 24.89
304.8 24.34
457.2 23.03

pressive strength of concrete which is no more affected by the

sample size or the chosen k-value. However, this requires a larger
number of tests with samples of different sizes, making this proce-
dure lengthy and costly. To circumvent this problem, a simple
analysis of the pore structure of the concrete can be performed
(see Appendix A). Indeed, Eq. (3) implies that the asymptotic
strength will be correctly estimated from hrf i if LLm 0.
In [15], we have shown that the length scale Lm mainly depends
on the intrinsic characteristic of the pore structure (air voids) in the
hardened concrete. Fig. 11 shows the positive correlation between
Lm and the the maximum pore diameter ðdp;max Þtimes the porosity
ðpo inð%ÞÞ. A good linear regression ðR2 ¼ 0:85Þ is obtained:

p 
Fig. 9. Normal distribution curve for the compressive strength of concrete samples
Lm ¼ 70  o
 dp;max ð6Þ
100
(modified from [80]).
This suggests an acceptance criterion for the minimum sample
(Eq. (3) and Eq. (4)) to the data in order to estimate the length size with respect to the pore structure characteristics. Combining
scales Lm and Ld as well as the asymptotic strength r1 . The Eq. (3) with mFS ¼ 1 and Eq. (6), this condition reads:
number of test specimens in each size should be as large as pos-  
0:7  po  dp;max
sible, and not less than 30 specimens, following the standard L ð7Þ
ACI 318 [5]. k
2. The diameter of the smallest concrete cylinders should be at where kcan be seen as the acceptance constant, corresponding to
least 3 times the nominal maximum size of the coarse aggre- the expected deviation of the average strength hrf ifrom the asymp-
gate, following the regulations ASTM C31 [53], NF EN 206-1 totic strength r1 . For example, to determine the characteristic
[4] and NF EN 12390-1 [54]. This condition allows to reduce asymptotic strength of C-concrete with an acceptance constant of
the ‘‘wall effect” that might happen when the maximum aggre- k ¼ 5%, using the values of po and dp;max reported in Table A.1, the
gate size is large relatively to the size of the mold, and to ensure minimum sample size required is L ¼ 120mm. This means that
the quality of the compaction procedure and the uniformity of the mean compressive strength obtained on cylindrical samples
the aggregate distribution in the samples. with a diameter larger 120-mm can be regarded as an estimation
3. The largest sample size should be as large as possible depending of the characteristic strength of C-concrete with a confidence level
on the laboratory loading capacity. However, it should not be of 95%. On the other hand, with the same acceptance constant k,
less than 160 mm. samples with a diameter of at least 464 mm would be needed to
4. All the concrete samples must be cast, cured and tested under determine the characteristic strength of F-concrete. Although the
the same conditions. asymptotic strength r1 is similar for the three different concrete
5. Using mFS ¼ 1 in Eq. (3), the asymptotic strength ðr1 Þ is deter- mixtures (see Table 3 and Fig. 5), the sample size required for its
mined by a linear regression in a ½hrf ivs:L1  graph. The perti- estimation with the same accuracy, from a single series of tests,
nence of the finite-size scaling can then be checked by plotting strongly varies with the material. This is another illustration of
the strength variability dðrÞ as a function of L1 (see Eq. (4)). the role of the internal pore structure on the size effects on strength,
while its impact on the asymptotic strength appears limited in our
A crucial advantage of such experimental program is that it pro- low-porosity concretes [15]. Consequently, if one uses only one
vides a well-defined value of the characteristic (asymptotic) com- sample size to estimate the characteristic strength of the material
14 C.-C. Vu et al. / Construction and Building Materials 263 (2020) 120126

Fig. 10. Comparison between the asymptotic strength, r1 (dashed black line), determined form the finite-size scaling of the mean strength (Eq. (3)), and the estimated values
of the characteristic strength, f ck , following the Eurocode 2 (filled symbols) and ACI-318 (open symbols): (a) F-concrete; (b) M-concrete and (c) C-concretes. The red and blue
curves are the fits by Eq. (5) for the f ck values of all concrete mixtures according to the EN 1992 and ACI 318 design codes, respectively. (For interpretation of the references to
color in this figure legend, the reader is referred to the web version of this article.)

from laboratory testing, the sample size criterion (Eq. (7)) should be
taken into account.
Quality control during construction is required to check the
conformity of concrete to the strength requirements posed at the
design stage. Usually, the acceptance strength criterion is derived
from the characteristic strength determined from standard proce-
dures [3], without paying attention to sample size effects or to
the internal pore structure. However, based on the above discus-
sion, we argue that, besides the strength criterion, an additional
condition on the sample size for laboratory tests should be exam-
ined for the final acceptance decision. Hence, we propose the fol-
lowing step-by-step checking procedure:

1- Assign the strength grade of concrete ðf ck Þ, specified at the

design stage, to the asymptotic strength ðr1 Þ of the consid-
ered concrete.
 
2- Obtain the mean value hrf i and the standard deviation d rf
of the compressive strength of the considered concrete from
a series of at least 30 compression tests carried out on sam-
ple cylinders of diameterL.
3- Perform an image analysis of internal sections on virgin con-
crete samples to estimate the porosity ðpo Þ and the pore
   
Fig. 11. Relationship between the length scale ðLm Þ and the product po  dp;max . maximum diameter dp;max (see Appendix A). A minimum
 C.-C. Vu et al. / Construction and Building Materials 263 (2020) 120126 15

Table 7
Example of application of the proposed procedure for checking the conformity of concrete with the strength requirement for our three concrete mixtures.

Parameter F-concrete M-concrete C-concrete

Compressive strength class C35/45 r1 ¼ f ck ¼ 35MPa
Mean compressive strength (MPa) hrf i 41.9 39.2 36.7
Sample size (mm) L 160 160 160
Porosity (%) po 4.8 1.6 1.5
Maximum pore diameter (mm) dp;max 6.9 6.7 5.4

Minimum required average strength1 (MPa) 0:7ðpo dp;max Þ 40.1 36.6 36.2
f mr ¼ r1 L þ1

Acceptance condition hrf i  f mr passed passed passed

1
The porosity po expressed in %.

Table 8
Example of application of the proposed procedure for checking the conformity of concrete with the strength requirement for three assumed cases of C-concrete.

Parameter Case 1 Case 2 Case 3

Compressive strength class C35/45 r1 ¼ f ck ¼ 35MPa
Mean compressive strength (MPa) hrf i 36.7 36.7 35.7
Sample size (mm) L 160 160 160
Porosity (%) po 1.5 3.5 1.5
Maximum pore diameter (mm) dp;max 5.4 5.4 5.4

Minimum required average strength1 (MPa) 0:7ðpo dp;max Þ 36.2 37.9 36.2
f mr ¼ r1 L þ1

Acceptance condition hrf i  f mr passed failed failed

1
The porosity po expressed in %.

of three sections including one vertical and two horizontal In addition, a smaller sample size L implies a larger ‘‘correcting fac-
sections (see Fig. A.1a) for each sample are recommended. tor” k in the relation (7) as well as a larger strength variability dðrÞ
In the case of concrete samples fabricated from different at this size. We therefore recommend using cylindrical samples
batches, one sample in each batch has to be selected and with a diameter larger than 110 mm.
analyzed. Note that other methods and techniques can be Two examples illustrating the application of the above proce-
used alternatively to determine the porosity and the pore dure are presented below.
size distribution of concrete or of other cement based mate-
rials, such as the fluid displacement method [55], mercury Application examples
intrusion porosimetry (MIP) [56–61], capillary condensation
measurement [61,62], and back scattered electron micro- 1. In the first example (Table 7), we check the conformity of our
scopy [63–67]. All these methods can be used to determine three concrete mixtures with the strength class C35/45
the porosity. expressed in EN 1992 [3]. The values of parameters obtained
4- Calculate the minimum required average strength ðf mr Þ of for the 160  320-mm samples are used for this example. As
the considered concrete at the sample size L from the equa- shown in Table 7, all of the three concrete mixtures have an
tions (3) and (6) with mFS ¼ 1, based on the values of the experimental mean compressive strength hrf i larger than the
required asymptotic strength, r1 (step 1), of the porosity, minimum required average strength f mr calculated by using
po (step 3) and of the pore maximum diameter, dp;max (step Eq. (8). Therefore, we can conclude that these three mixtures
3), yielding: comply with the strength requirements for the strength class
C35/45. In this example, the C-mixture provides the closest
   mean strength value to the required value of strength class
0:7  po  dp;max
f mr ¼ r1 þ1 ð8Þ C35/45 in comparison with the other mixtures. This means that
L
a coarser concrete mixture, having a lower porosity in its hard-
ened form, will comply better the targeted characteristic
where the porosity po is expressed in %.
strength than finer mixtures.
2. For the second example (Table 8), based on the original data of
5- Compare the experimental mean strength hrf i (obtained at
the C-160  320mm samples (Case 1), two additional cases are
the step 2) with the minimum required average strength
considered after changing either the value of porosity (Case 2)
f mr (calculated by using Eq. (8)). If hrf i  f mr , we can state
or the value of the mean compressive strength (Case 3). As
that the examined concrete complies with the strength
the results presented in Table 8, the two last cases do not com-
requirements.
ply with the strength requirement for the strength class C35/45.
Although Case 2 has the same mean strength for 160  320mm
We note that the relation between the length scale Lm and the
samples as experimental Case 1, the strength requirement is not
characteristics of the pore structure (Fig. 11) has only been vali-
satisfied for Case 2, owing to a larger porosity. This results from
dated for our low-porosity concrete. It is doubtful that it would
the fact that a larger porosity means stronger size effects on
hold for large porosity concretes, for which the pore structure is
strength, hence a larger value of the minimum average strength
much more complex (non-spherical, interconnected pores) and
required at a fixed sample size. The production of normal-
the direct impact of porosity on strength is strong ([57,68–70]).
weight concrete with a porosity as small as possible is an
Therefore, the above procedure should be applied only for
important goal to achieve a good quality of concrete as well
normal-weight concretes with a total porosity of less than 10%.
16 C.-C. Vu et al. / Construction and Building Materials 263 (2020) 120126

as to be able to correctly estimate the asymptotic strength for Declaration of Competing Interest
laboratory samples. Case 3, characterized by an unchanged
porosity but a smaller mean strength than Case 1, also fails The authors declare that they have no known competing finan-
the strength requirement. This illustrates the fact that both cial interests or personal relationships that could have appeared
the strength and the sample size criteria must be examined to influence the work reported in this paper.
for the acceptance decision.
Acknowledgements
7. Conclusion
This work was supported by the program of AGIR-POLE-PAGE
In this paper, from an experimental study of the statistical size from the University of Grenoble Alpes (UGA), France. The authors
effects on the compressive strength of concrete as well as an analy- are thankful to technicians of UGA and USMB for technical
sis of published strength data: we (i) discussed the definition of the assistance.
characteristic compressive strength of concrete when size effects on
strength are taken into account; and (ii) proposed to revise the pro-
cedures to estimate this characteristic strength and to check the Appendix A. . Characterization of the pore structure
conformity of concrete to the strength requirements defined at
the design stage. The following main conclusions can be drawn: The strength of concrete can be significantly affected by the vol-
ume of all pores: capillary pores, gel pores and air voids (entrained
1. The observed size effects on both the mean compressive and entrapped air voids) [2,61]. Hence, the porosity and size distri-
strength and the standard deviation of our concretes can be bution of pores must be quantified [71]. Mehta and Monteiro [1]
rationalized from finite-size effects associated with an interpre- noted that ‘‘entrapped and entrained air voids in the hydrated cement
tation of the compressive failure of concrete as a critical transi- paste are much bigger than the capillary voids, and are capable of
tion. This implies that a very large system system ðL ! þ1Þ has adversely affecting the strength”. Actually, there is no consensus
a non-vanishing strength ðr1 > 0Þ but its strength variability on the role of the capillary pores of the hardened concrete on com-
   
vanishes d rf ! 0 . pressive strength (e.g. [72–74]). In addition, due to their small sizes
2. Owing to these vanishing intrinsic fluctuations of strength at (ranging from 0.5 nm to 1 lm), gel and capillary pores are hard to

   detect by optical methods. Thus, we just focus here on air voids (re-
large scales d rf L!þ1 ! 0 , the asymptotic strength ðr1 Þ
ferred as pores in the following). In the present work, an image
can be considered as the genuine characteristic compressive analysis procedure was used to characterize the pore structure of
strength of concrete. Therefore, an experimental study con- our undamaged concrete samples based on the procedure pre-
ducted on a series of at least four different sample sizes is sented in [75,76].
strongly recommended to allow an accurate determination of First, one sample was selected for each size and each concrete
the size effects on strength, hence of the characteristic com- mixture (for a total 12 samples used for this image analysis) and
pressive strength of concrete. then cut into four pieces (see Fig. A.1a). To remove any scratching
3. For low-porosity concretes, the pore structure of the hardened on the sample surfaces due to the sawing process, the surfaces
concrete samples plays a key role in controlling the size effects were ground flat using hand pressure on a water-cooler wheel
on strength but does not have a significant impact on the topped with more and more grit size of metal-bonded diamond
asymptotic compressive strength. A concrete with a lower plates [75]. Then, these surfaces were polished by using succes-
porosity will be less affected by size effects on compressive sively finer grit resin-bonded diamond plates to remove any stria-
strength than a concrete with a larger porosity. Another impor- tions from the grinding materials (see Fig. A.1b). After that, the
tant consequence is that, if one wants to estimate the character- sample surfaces were treated to enhance the contrast between
istic (asymptotic) strength of the material from a series of tests pores and other components (aggregates, cement paste). This
with a single sample size, a condition on this size with respect was done by filling the depressions with a calcium carbonate paste.
to the characteristic scale of the pore structure should be After this preparation, the sample surfaces were scanned by a
fulfilled. flatbed scanner at a resolution of 1200dpi (equivalent to
4. This characteristic scale of the pore structure in a hardened con- 21.17 lm/pixel). Thanks to the white color of the calcium carbon-
crete sample can be estimated from the product of the maxi- ate paste, the pores were easily distinguished from the solid phases
mum pore diameter and the total porosity. In order to (aggregates and hardened cement paste) after applying a manual
determine these two parameters, we proposed a simple image contrast enhancement (Fig. A.1c) and thresholding (Fig. A.1d).
analysis procedure on internal sections of concrete samples, Fig. A.2 illustrates a comparison of the pore structures for three dif-
but other techniques can be used to measure these parameters. ferent concrete mixtures.
This procedure was tested so far only on a limited number of Whereas gel and capillary voids are irregular in shape, air voids
different concrete mixtures. An extension of this work to other are generally spherical-like [1,2]. This is consistent with the cross-
concrete materials with various microstructures would be nec- sections of Fig. A.2. Therefore, to obtain the three-dimensional (3D)
essary to refine this estimation. pore size distribution, a stereological 2D ? 3D conversion method
for a polydispersed system of spheres [77,78] was applied in this
CRediT authorship contribution statement work. From the binary image of pores (Fig. A.2), the cross-
sectional areas of individual pores were calculated and classified
Chi-Cong Vu: Writing - original draft. Olivier Plé: Writing - into nsize (diameter) intervals. A value of n ¼14 was chosen to
review & editing. Jérôme Weiss: Writing - review & editing. David obtain a large enough number of pores in each interval. We noted
Amitrano: Writing - review & editing. ðN A Þi the number of pores per unit area in each size interval i. The
 C.-C. Vu et al. / Construction and Building Materials 263 (2020) 120126 17

Fig. A1. Image analysis procedure of the pore structure: (a) cutting sections of the concrete sample; (b) concrete sample surface after grinding and polishing; (c) scanned
image of pores after the contrast enhancement operation; (d) binary image of pores obtained after thresholding. (). adapted from [15]

corresponding diameter range for the

th
i interval is ber per unit area of pore diameter class ðdp Þi , which can be obtained
½dpm  10 0:1ði2Þ
; dpm  10 0:1ði1Þ
, where dpm is the largest 2D diam- from spheres of different diameter ðdp Þj ; ai is the coefficient corre-
eter of an equivalent circular pore. Following the Saltykov’s sponding to the area’s diameter class ðdp Þi , this coefficient can be
method [77,78], we assumed that the largest 2D pore areas corre- obtained from the probability of a cutting plane intersecting a
spond to the largest spheres. It means that the diameter of the lar- sphere of diameter ðdp Þj to yield areas of diameter ðdp Þi . Using the
gest 2D pore area is equal to that of the largest 3D spherical pore in
values of the coefficients ai given in [77–79], the 3D pore size dis-
the sample volume. Smaller 2D areas of diameter ðdp Þi can result
tribution was determined for each concrete sample.
from cutting the spheres having a diameter ðdp Þj larger than or Fig. A.3 presents the (cumulative) probability distribution of
equal to ðdp Þi . Then, from the geometric probability distribution pore sizes for different concrete samples. We observe: (i) an
of the distances from a random cutting plane to the centers of absence of sample size effect on the pore size distribution for a
spheres, we obtain the expression: given type of concrete; (ii) that the pore size distributions have a
  a
power law shape, P > dp dp , with an exponent aslightly varying
1 X k
with the material (see Fig. A.3).
ðNv Þj ¼ a i ðN A Þi ðA:1Þ
ðdp Þj i¼j From these distributions, the average diameter and the total
porosity were estimated and reported in Table A.1 for the different
0 concrete samples. We note that the number of pore per m3 , the
where the j s refer to sphere diameters and the i0 s refer to area
diameters; i and j are integer values from 1 to n, and n is equal to mean as well as the maximum pore diameter, and the total poros-
the total number of diameter class intervals; ðN v Þj is the number ity, all decrease with increasing aggregate size (see Fig. A.2 and
of spherical pores per unit volume in the diameter class j; ðdp Þi is Table A.1). In this study, the maximum pore diameter, dp;max , was
calculated by averaging the values of dp;m of all four different sec-
the diameter of the largest areas observed on the cutting plane, this
tions (Fig. A.1a) of a given size, and of all sections and samples
diameter depends on the number of size class intervals, and corre-
P (whatever the size) of a given concrete mixture.
sponds to the sphere diameter class j; j ðN A Þi;j represents the num-
18 C.-C. Vu et al. / Construction and Building Materials 263 (2020) 120126

Fig. A2. Comparison of the pore structure for three different concrete mixtures; (a) F-110  220mm; (b) M-110  220mm; (c) C-110  220mm. Top figures are the original
scanned images and bottom figures are the corresponding binary images of pores. (from [15]).

The presence of pores in hardened concrete is related to the the largest pore sizes and porosity (see Table A.1) as well as the lar-
water and cement contents of the mix. Larger cement and water gest moisture. When large enough, porosity is known to directly
contents imply a larger ability for holding the free water in the impact the compressive strength of concrete [57,68–70]. However,
hardened cement paste after the hydration process, hence more as the porosity of our samples always remained below 5%
air voids (pores) in the hardened concrete [1]. This explains why (Table A.1), it did not impact significantly the asymptotic strength
the F-concrete, which has the largest water-to-cement ratio, shows (r1 ) of our materials (see Table 3).
 C.-C. Vu et al. / Construction and Building Materials 263 (2020) 120126 19

Fig. A3. Size distribution of pores per unit volume for different concrete samples. (a) F-concrete; (b) M-concrete; (c) C-concrete and (d) All concrete samples.

Table A1
Pore structure parameters, averaged over all sections and sample sizes, for the 3 different concrete mixtures.

Concrete group Mean pore diameter, hdp i Maximum pore diameter, Porosity, po (%)
(mm) dp;max (mm)

Mean SD Mean SD Mean SD

F 0.33 0.24 6.9 1.8 4.8 1.2
M 0.31 0.21 6.7 1.2 1.6 0.4
C 0.28 0.20 5.4 1.3 1.5 0.3

References [4] NF EN 206-1, Béton - Partie 1 : Spécification, performances, production et

conformité, Association Française de Normalisation (AFNOR), 2004.
[5] ACI 318-05, Building Code Requirements for Structural Concrete and
[1] P.K. Mehta, P.J.M. Monteiro, Concrete: Microstructure, Properties, and
Commentary, American Concrete Institute, 2005.
Materials, Third, McGraw-Hill, 2006, doi: 10.1036/0071462899.
[6] ASTM C39/C39M-14, Standard Test Method for Compressive Strength of
[2] A.M. Neville, J.J. Brooks, Properties of concrete, fourth ed., Pearson Education
Cylindrical Concrete Specimens, ASTM Int. (2014) 1–7, doi: 10.1520/
Limited, 2010, doi: 10.4135/9781412975704.n88.
C0039_C0039M-14.
[3] EN 1992, Eurocode 2: Design of concrete structures - Part 1-1 : General rules
[7] NF EN 12390-3, Essais pour béton durci - Partie 3 : Résistance à la compression
and rules for buildings, European Committee for Standardization, 2004.
des éprouvettes, Association Française de Normalisation (AFNOR), 2012.
20 C.-C. Vu et al. / Construction and Building Materials 263 (2020) 120126

[8] Z.P. Bazant, J. Planas, Fracture and Size Effect in Concrete and Other [39] D. Ertas, M. Kardar, Critical dynamics of contact line depinning, Phys. Rev. E. 49
Quasibrittle Materials, CRC Press, Boca Raton, Florida, USA, 1998. (1994) 2532–2535, https://doi.org/10.1103/PhysRevE.49.R2532.
[9] S.L. Burtscher, J. Kollegger, Size-effect experiments on concrete in compression, [40] D.S. Fisher, Collective ransport in random media: from superconductors to
Struct. Concr. 4 (2003) 163–174. earthquakes, Phys. Rep. 301 (1998) 113–150.
[10] J. Ozbolt, R. Eligehausen, Size effect in concrete and reinforced concrete [41] S. Zapperi, Current challenges for statistical physics in fracture and plasticity,
structures, Fract. Mech. Concr. Struct. (1995) 665–674. Eur. Phys. J. B. 85 (2012) 329, https://doi.org/10.1140/epjb/e2012-30471-9.
[11] Z.P. Bazant, Size effect on structural strength: a review, Arch. Appl. Mech. 69 [42] A. Carpinteri, B. Chiaia, G. Ferro, Size effects on nominal tensile strength of
(1999) 703–725, https://doi.org/10.1007/s004190050252. concrete structures: multifractality of material ligaments and dimensional
[12] Z.P. Bazant, Size effect in tensile and compression fracture of concrete transition from order to disorder, Mater. Struct. 28 (1995) 311–317, https://
structures: computational modeling and design, Fract. Mech. Concr. Struct. doi.org/10.1007/BF02473145.
Proceeding (1998) 1905–1922. [43] Z.P. Bazant, Scaling theory for quasibrittle structural failure, Proc. Natl. Acad.
[13] G. Muciaccia, G. Rosati, G. Di Luzio, Compressive failure and size effect in plain Sci. U. S. A. 101 (2004) 13400–13407, https://doi.org/10.1073/
concrete cylindrical specimens, Constr. Build. Mater. 137 (2017) 185–194, pnas.0404096101.
https://doi.org/10.1016/j.conbuildmat.2017.01.057. [44] M. Dehestani, I.M. Nikbin, S. Asadollahi, Effects of specimen shape and size on
[14] J.R. del Viso, J.R. Carmona, G. Ruiz, Shape and size effects on the compressive the compressive strength of self-consolidating concrete (SCC), Constr. Build.
strength of high-strength concrete, Cem. Concr. Res. 38 (2008) 386–395, Mater. 66 (2014) 685–691, https://doi.org/10.1016/j.conbuildmat.2014.06.008.
https://doi.org/10.1016/j.cemconres.2007.09.020. [45] P.C. Aïtcin, R.J. Flatt, Science and technology of concrete admixtures, 2015, doi:
[15] C.C. Vu, J. Weiss, O. Plé, D. Amitrano, D. Vandembroucq, Revisiting statistical 10.1016/C2015-0-00150-2.
size effects on compressive failure of heterogeneous materials, with a special [46] K. Yoshioka, E.I. Tazawa, K. Kawai, T. Enohata, Adsorption characteristics of
focus on concrete, J. Mech. Phys. Solids. 121 (2018) 47–70, https://doi.org/ superplasticizers on cement component minerals, Cem. Concr. Res. 32 (2002)
10.1016/j.jmps.2018.07.022. 1507–1513, https://doi.org/10.1016/S0008-8846(02)00782-2.
[16] J. Weiss, L. Girard, F. Gimbert, D. Amitrano, D. Vandembroucq, (Finite) [47] B. Łaźniewska-Piekarczyk, The influence of selected new generation
statistical size effects on compressive strength, Proc. Natl. Acad. Sci. U. S. A. admixtures on the workability, air-voids parameters and frost-resistance of
111 (2014) 6231–6236, https://doi.org/10.1073/pnas.1403500111. self compacting concrete, Constr. Build. Mater. 31 (2012) 310–319, https://doi.
[17] L. Girard, J. Weiss, D. Amitrano, Damage-cluster distributions and size effect on org/10.1016/j.conbuildmat.2011.12.107.
strength in compressive failure, Phys. Rev. Lett. 108 (2012) 1–5, https://doi. [48] J. Gołaszewski, Influence of cement properties on new generation
org/10.1103/PhysRevLett.108.225502. superplasticizers performance, Constr. Build. Mater. 35 (2012) 586–596,
[18] L. Girard, D. Amitrano, J. Weiss, Failure as a critical phenomenon in a https://doi.org/10.1016/j.conbuildmat.2012.04.070.
progressive damage model, J. Stat. Mech. Theory Exp. 2010 (2010) P01013, [49] F. Puertas, H. Santos, M. Palacios, S. Martınez-Ramirez, Polycarboxylate
https://doi.org/10.1088/1742-5468/2010/01/P01013. superplasticiser admixtures: effect on hydration, microstructure and
[19] C. Vu, D. Amitrano, O. Pl, Compressive failure as a critical transition: rheological behaviour in cement pastes, Adv. Cem. Res. 17 (2005) 77–89.
experimental evidence and mapping onto the universality class of [50] J. Il Sim, K.H. Yang, H.Y. Kim, B.J. Choi, Size and shape effects on compressive
depinning, Phys. Rev. Lett. 122 (2019) 6, https://doi.org/10.1103/PhysRevLett. strength of lightweight concrete, Constr. Build. Mater. 38 (2013) 854–864,
122.015502. https://doi.org/10.1016/j.conbuildmat.2012.09.073.
[20] NF EN 197-1, Ciment -Partie 1: Compostition, spécifications et critères de [51] ACI 213R-03, Guide for structural lightweight aggregate concrete, Am. Concr.
conformité des ciments courants, Association Française de Normalisation Inst. (2003), doi: 10.1016/0262-5075(79)90004-6.
(AFNOR), 2012. [52] R.F. Blanks, C. McNamara, Mass concrete tests in large cylinders, J. Am. Concr.
[21] NF EN 12620, Granulats pour béton, Association Française de Normalisation Inst. 31 (1935) 280–303.
(AFNOR), 2008. [53] ASTM C31/C31M-03, Standard Practice for Making and Curing Concrete Test
[22] NF EN 1097-5, Essais pour déterminer les caractéristiques mécaniques et Specimens in the Field, ASTM Int. (2003) 1–5.
physiques des granulats - Partie 5 : détermination de la teneur en eau par [54] NF EN 12390-1, Essais pour béton durci - Partie 1 : Forme , dimensions et autres
séchage en étuve ventilée, Association Française de Normalisation (AFNOR), exigences relatives, Association Française de Normalisation (AFNOR), 2012.
2008. [55] B.B. Das, B. Kondraivendhan, Implication of pore size distribution parameters
[23] NF EN 933-1, Essais pour déterminer les caractéristiques géométriques des on compressive strength, permeability and hydraulic diffusivity of concrete,
granulats - Partie 1 : détermination de la granularité - Analyse Constr. Build. Mater. 28 (2012) 382–386, https://doi.org/10.1016/j.conbuildmat.
granulométrique par tamisage, Association Française de Normalisation 2011.08.055.
(AFNOR), 2012. [56] C. Gallé, Effect of drying on cement-based materials pore structure as
[24] NF EN 196-1, Méthodes d’essais des ciments - Partie 1 : détermination des identified by mercury intrusion porosimetry - a comparative study between
résistances, Association Française de Normalisation (AFNOR), 2016. oven-, vacuum-, and freeze-drying, Cem. Concr. Res. 31 (2001) 1467–1477,
[25] G. Dreux, F. Gorisse, Composition des bétons : méthode Dreux Gorisse, bilan de https://doi.org/10.1016/S0008-8846(01)00594-4.
cinq années d’application en Côte d’Ivoire, Annales de l’Institut Technique du [57] R. Kumar, B. Bhattacharjee, Porosity, pore size distribution and in situ strength
bâtiment et des Travaux Publiques (1983). of concrete, Cem. Concr. Res. 33 (2003) 155–164, https://doi.org/10.1016/
[26] NF EN 12390-2, Essais pour béton durci - Partie 2 : Confection et conservation S0008-8846(02)00942-0.
des éprouvettes pour essais de résistance, Association Française de [58] R. Kumar, B. Bhattacharjee, Assessment of permeation quality of concrete
Normalisation (AFNOR), 2012. through mercury intrusion porosimetry, Cem. Concr. Res. 34 (2004) 321–328,
[27] NF EN 12390-4, Essais pour béton durci - Partie 4 : Résistance à la compression https://doi.org/10.1016/j.cemconres.2003.08.013.
- Caractéristiques des machines d’essai, Association Française de [59] R.A. Cook, K.C. Hover, Mercury porosimetry of cement-based materials and
Normalisation (AFNOR), 2000, doi: 10.1017/CBO9781107415324.004. associated correction factors, Constr. Build. Mater. 7 (1993) 231–240, https://
[28] J. Mazars, G. Pijaudier-Cabot, C. Saouridis, Size effect and continuous damage doi.org/10.1016/0950-0618(93)90007-Y.
in cementitious materials, Int. J. Fract. 51 (1991) 159–173, https://doi.org/ [60] R.A. Cook, K.C. Hover, Mercury porosimetry of hardened cement pastes, Cem.
10.1007/BF00033976. Concr. Res. 29 (1999) 933–943, https://doi.org/10.1016/S0008-8846(99)
[29] D. Jeulin, M. Ostoja-Starzewski, Mechanics of Random and Multiscale 00083-6.
Microstructures, first ed., Springer-Verlag Wien, 2001, doi: 10.1007/978-3- [61] S. Diamond, A critical comparison of mercury porosimetry and capillary
7091-2780-3. condensation pore size distributions of portland cement pastes, Cem. Concr.
[30] S. Torquato, Random heterogeneous media: microstructure and improved Res. 1 (1971) 531–545, https://doi.org/10.1016/0008-8846(71)90058-5.
bounds on effective properties, Appl. Mech. Rev. 44 (1991) 37–76, https://doi. [62] R.S. Mikhail, L.E. Copeland, S. Brunauer, Pore structures and surface areas of
org/10.1115/1.3119494. hardened portland cement pastes by nitrogen adsorption, Can. J. Chem. 42
[31] NF EN 12390-7, Essais pour béton durci - Partie 7 : Masse volumique du béton (1964) 426–438. www.nrcresearchpress.com.
durci, Association Française de Normalisation (AFNOR), 2012. [63] S.I. Igarashi, A. Watanabe, M. Kawamura, Evaluation of capillary pore size
[32] H.K. Hilsdorf, A method to estimate the water content of concrete shields, characteristics in high-strength concrete at early ages, Cem. Concr. Res. 35
Nucl. Eng. Des. 6 (1967) 251–263, https://doi.org/10.1016/0029-5493(67) (2005) 513–519, https://doi.org/10.1016/j.cemconres.2004.06.036.
90022-2. [64] D.A. Lange, H.M. Jennings, S.P. Shah, Image analysis techniques for
[33] F.M. Bartlett, J.G. MacGregor, Effect of moisture condition on concrete core characterization of pore structure of cement-based materials, Cem. Concr.
strengths, ACI Mater. J. 91 (1994) 227–236. Res. 24 (1994) 841–853, https://doi.org/10.1016/0008-8846(94)90004-3.
[34] Ifsttar, Caractérisation du béton durci: Essais liés à a corrosion des armatures, [65] S. Diamond, Aspects of concrete porosity revisited, Cem. Concr. Res. 29 (1999)
(1994) 15. 1181–1188, https://doi.org/10.1016/S0008-8846(99)00122-2.
[35] S.E. Pihlajavaara, A review of some of the main results of a research on the [66] P. Soroushian, M. Elzafraney, A. Nossoni, Specimen preparation and image
ageing phenomena of concrete: Effect of moisture conditions on strength, processing and analysis techniques for automated quantification of concrete
shrinkage and creep of mature concrete, Cem. Concr. Res. 4 (1974) 761–771, microcracks and voids, Cem. Concr. Res. 33 (2003) 1949–1962, https://doi.org/
https://doi.org/10.1016/0008-8846(74)90048-9. 10.1016/S0008-8846(03)00219-9.
[36] Z.P. Bazant, Size effect in blunt fracture: concrete, rock, metal, J. Eng. Mech. 110 [67] H.S. Wong, M.K. Head, N.R. Buenfeld, Pore segmentation of cement-based
(1984) 518–535, https://doi.org/10.1061/(ASCE)0733-9399(1984)110:4(518). materials from backscattered electron images, Cem. Concr. Res. 36 (2006)
[37] W. Weibull, A statistical theory of the strength of materials, R. Swedish Inst. 1083–1090, https://doi.org/10.1016/j.cemconres.2005.10.006.
Eng. Res. 151 (1939) 45. [68] X. Chen, S. Wu, J. Zhou, Influence of porosity on compressive and tensile
[38] W. Weibull, A statistical distribution function of wide applicability, J. Appl. strength of cement mortar, Constr. Build. Mater. 40 (2013) 869–874, https://
Mech. 18 (1951) 293–297. doi.org/10.1016/j.conbuildmat.2012.11.072.
 C.-C. Vu et al. / Construction and Building Materials 263 (2020) 120126 21

[69] S. Kolias, Investigation of the possibility of estimating concrete strength by [75] ASTM C 457/C 457M-16, Standard Test Method for Microscopical
porosity measurements, Mater. Struct. 27 (1994) 265–272, https://doi.org/ Determination of Parameters of the Air-Void System in Hardened Concrete,
10.1007/BF02473043. ASTM Int. (2016). www.astm.org.
[70] C. Lian, Y. Zhuge, S. Beecham, The relationship between porosity and strength [76] U.H. Jakobsen, C. Pade, N. Thaulow, D. Brown, S. Sahu, O. Magnusson, S. De
for porous concrete, Constr. Build. Mater. 25 (2011) 4294–4298, https://doi. Buck, G. De Schutter, Automated air void analysis of hardened concrete - a
org/10.1016/j.conbuildmat.2011.05.005. Round Robin study, Cem. Concr. Res. 36 (2006) 1444–1452, https://doi.org/
[71] P.C. Fonseca, G.W. Scherer, An image analysis procedure to quantify the air 10.1016/j.cemconres.2006.03.005.
void system of mortar and concrete, Mater. Struct. 48 (2015) 3087–3098, [77] S.A. Saltykov, The determination of the size distribution of particles in an
https://doi.org/10.1617/s11527-014-0381-9. opaque material from a measurement of the size distribution of their sections,
[72] H. Zhao, Q. Xiao, D. Huang, S. Zhang, Influence of pore structure on Springer Berlin Heidelberg, 1967, doi: 10.1007/978-3-642-88260-9_31.
compressive strength of cement mortar, Sci. World J. 2014 (2014) 1–12, [78] E.E. Underwood, Quantitative stereology, Addison-Wesley Pub. Co., 1970.
https://doi.org/10.1155/2014/247058. [79] H. Shen, S.M. Oppenheimer, D.C. Dunand, L.C. Brinson, Numerical modeling of
[73] B. Kondraivendhan, B. Bhattacharjee, Effect of age and water-cement ratio on pore size and distribution in foamed titanium, Mech. Mater. 38 (2006) 933–
size and dispersion of pores in ordinary portland cement paste, ACI Mater. J. 944, https://doi.org/10.1016/j.mechmat.2005.06.027.
107 (2010) 147–154. [80] S.U. Pillai, D. Menon, Reinforced Concrete Design, third ed., Tata McGraw-Hill
[74] L. Zingg, M. Briffaut, J. Baroth, Y. Malecot, Influence of cement matrix porosity Education Pvt. Ltd., 2009.
on the triaxial behaviour of concrete, Cem. Concr. Res. 80 (2016) 52–59,
https://doi.org/10.1016/j.cemconres.2015.10.005.