You are on page 1of 9

Construction and Building Materials 229 (2019) 116922

Contents lists available at ScienceDirect

Construction and Building Materials


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

Influence of aggregate size on flexural fatigue response of concrete


Sridhar Reddy Kasu, Sutapa Deb, Nilanjan Mitra ⇑, Amarnatha Reddy Muppireddy,
Sudhakar Reddy Kusam
Department of Civil Engineering, Indian Institute of Technology Kharagpur, West Bengal, 721302, India

h i g h l i g h t s

 The effect of nominal maximum aggregate size on fatigue life was studied.
 3-paramter Weibull distribution fitting was done using A-D statistic.
 Fatigue lives found to be very sensitive to stress levels rather than frequencies.
 Distributions shows fatigue life of S10 mix had significantly higher than S20 mix.

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

Article history: Plain concrete used in pavements undergoes fatigue damage with repeated traffic and thermal loading.
Received 18 April 2019 Numerous studies were conducted previously to evaluate the effect of various parameters on the fracture
Received in revised form 3 September 2019 behavior of concrete. Nominal maximum aggregate size which is one of the parameters investigated in
Accepted 7 September 2019
the past, to evaluate the fracture characteristics was focused to influence the fracture energy of concrete
significantly. This could be due to a larger crack path and greater crack growth resistance offered by the
larger aggregates in the fracture process zone. Hence, it is worthwhile to investigate the effect of nominal
Keywords:
maximum aggregate size (NMAS) of aggregate on the fatigue life of concrete. This factor, if influential,
Aggregate size
Flexural fatigue
might be a feasible approach for field application to control the fatigue performance of concrete. In the
Constant stress amplitude present study, the effect of nominal maximum aggregate size was evaluated under different stress levels
Weibull distribution (80, 85, and 90%) and different frequencies of loading (2, 5, and 10 Hz). For this purpose, flexural fatigue
Survival plots test was conducted on concrete beam specimens. The specimens were prepared with a blend of fine
Dynamic loading aggregate and coarse aggregate with two different nominal maximum aggregate sizes (10 mm and
20 mm, designated as S10 and S20 respectively). A total of 100 specimens of
100 mm  100 mm  500 mm size were used in the experimental investigation. The results show that
the fatigue life of samples with S10 mixture is significantly higher compared to S20 mixture for 80% stress
level but are almost similar for 90% stress level irrespective of the loading frequency. The fatigue results
are in better agreement with three-parameter Weibull distribution compared to other distributions such
as lognormal as confirmed by the goodness of fit test Anderson-Darling statistic. The value of b, the
Weibull distribution characteristic shape parameter, obtained as less than 1.0 in this study, indicates
the risk of early life failures for the considered stress levels. The survival probability plots and p-value
statistics suggest that stress level influences the fatigue life estimation for both S10 and S20 samples
but not the loading frequency. Based on the present study, it can be concluded that smaller nominal max-
imum aggregate size is more beneficial for the fatigue performance of concrete.
Crown Copyright Ó 2019 Published by Elsevier Ltd. All rights reserved.

1. Introduction pavements, and offshore structures in which the structure is sub-


jected to repetitive fatigue loads by traffic in the case of bridges
Evaluation of the flexural fatigue characteristics of concrete is and pavements, and by ocean waves for offshore structures.
essential for understanding the performance of concrete bridges, Although some of these structures have reinforced concrete, many
are with plain concrete. Plain concrete is more vulnerable to failure
from repeated and static flexural loading compared to reinforced
⇑ Corresponding author.
concrete. The percentage of reinforcement in reinforced concrete
E-mail addresses: nilanjan@civil.iitkgp.ac.in (N. Mitra), manreddy@civil.iitkgp.
ac.in (A.R. Muppireddy), ksreddy@civil.iitkgp.ac.in (S.R. Kusam).
pavements is generally very small. Common examples of

https://doi.org/10.1016/j.conbuildmat.2019.116922
0950-0618/Crown Copyright Ó 2019 Published by Elsevier Ltd. All rights reserved.
2 S.R. Kasu et al. / Construction and Building Materials 229 (2019) 116922

pavements in which plain concrete is used are Jointed Plain Con- aggregate and the matrix. Thus, larger the aggregate, longer will
crete Pavement (JPCP) and interlocking concrete block pavements. be the crack meandering and higher the fracture energy. With this
These structures undergo cumulative fatigue damage over a period in perspective, the current study investigates whether the maxi-
of time due to the repeated application of traffic and thermal loads. mum size of the coarse aggregate in the mix influences the crack
During the accumulation of fatigue damage, concrete is subjected propagation of concrete under flexure fatigue loading conditions.
to crack propagation and stiffness degradation. The rate of stiffness The experimental framework detailing the materials used for
degradation or damage at the macro level is important for the pre- sample preparation, test setup, results from the experimental data
diction of fatigue life. Stiffness degradation in concrete depends on of flexural fatigue tests and their discussion are presented in the
the type of constituents, method of specimen preparation, type of following sections of the paper.
loading, type of waveform [1,2] and other attributes. The con-
stituents of a typical plain concrete include coarse aggregates, fine 2. Experimental framework
aggregates, cement, and water. It is well known that coarse aggre-
gates and fine aggregates act as filler materials in the matrix 2.1. Materials
formed due to cement hydration. When plain concrete is subjected
Concrete mixtures were prepared using Ordinary Portland Cement (OPC 53) and
to repeated loads, the cracks originate primarily at the interface of dry fly ash sourced from Ramco Cements (Kolaghat, West Bengal, India). The
aggregates and hardened cement paste and/or at the void regions detailed chemical compositions of OPC 53 and fly ash (class F) used in this study
which may be either at the interface regions or within the matrix. were evaluated with the help of X-Ray Fluorescence spectrometer and are recorded
With the change in the size of the coarse aggregates, it is expected in previous literature [16]. The physical properties of the OPC 53 conforming to IS
12269 [17] and fly ash conforming to IS 3812-Part I, [18] are presented in Table 1.
that the crack propagation path will change.
As per ACI 226-1987 [19] and IRC SP 46-2013 [20], the percentage by volume of
From the visualization of the internal random grain structure of fly ash was limited to 20% of the OPC cement in this study.
the normal and high strength concrete, change in any of the con- The fine aggregate used in this experimental program is river sand. The ratio of
stituents such as the quantity of cement, water/binder ratio, type, sand to binder material (cement + pozzolanic materials like fly ash) was kept in this
size, shape, texture, and volume of the aggregates is found to be study as 2 by weight. Two different gradations with nominal maximum aggregate
sizes of 10 mm and 20 mm were used to prepare the mixes. Concrete with a nom-
influential on its fracture behavior/parameters. With a decrease
inal maximum aggregate size of 20 mm is usually used for thick concrete pave-
in the water/binder (w/b) ratio from 0.85 to 0.34 in the mix, an ments. The work carried out in this study was associated with ongoing research
increase in fracture energy, and fracture toughness was observed work on thin concrete pavements (50–75 mm thickness). The thin concrete slabs
[3,4]. The effect of w/b ratio on fracture energy was insignificant typically use 10 mm nominal maximum aggregate size. Generally, the usual prac-
tice is to select the maximum size of aggregate in such a way that the ratio of
(in the range of 0.2–0.3) [5]. In another study in which the volume
the smallest dimension of the specimen or structural member to the maximum size
of aggregates was increased from 40 to 60%, there was an increase of aggregate is 2.5–4. Therefore, the study was intended to carry out on concrete
in the fracture energy [6] and fracture toughness as measured by made with two nominal maximum aggregate size; 10 mm and 20 mm to evaluate
acoustic emission method [7]. The shape [8] and texture [9] of the fatigue performance of concrete for conventional thick concrete pavements
aggregates were also found to be significant in enhancing the frac- and thin concrete pavements. These are among the gradations recommended by
IS 383-1970 [21] for concrete. The properties of fine and coarse aggregates (CA)
ture toughness.
are presented in Table 2.
The weak link in the concrete material (with regard to strength) The ratio of the weights of CA (10 or 20 mm nominal size) and binder material
is the hydrated mortar matrix. This matrix is primarily composed (cement + pozzolanic materials like fly ash) was kept as 2 (by weight) in this study.
of hydrated components of cement, calcium silicate hydrate (to Fig. 1 shows the gradation of coarse and fine aggregates selected in this study. Mix
design was carried out to achieve a compressive strength of 40 MPa specified in IRC
be specific). Larger aggregate size means more of these matrix
58-2015 [22] for road works. The proportion of binder content (cementitious mate-
materials near the aggregate surfaces (due to less packing) which rial): fine aggregates: coarse aggregates was selected as 1:2:2 with a water-cement
might affect the overall strength and fatigue life of the concrete (w/c) ratio of 0.45. The mixes were designed to have a characteristic compressive
mix. Besides this, larger the size of the aggregate more flat and strength of 40 MPa and a slump value of 50 mm. The proportions of different con-
elongated the particles are likely to be. Flat and elongated aggre- stituents used in the two mixtures prepared using 10 mm and 20 mm nominal

gate are more conducive to the accumulation of water film next


to the aggregate resulting in weaker transition zone. Thus, the fati-
Table 1
gue performance of concrete with larger size aggregate is likely to Properties of Ordinary Portland Cement and Fly Ash.
have inferior fatigue performance compared to the concrete having
Properties OPC 53 Fly ash (Class F)
smaller size aggregate.
There are some studies reported in the literature in which the Specific gravity 3.15 2.20
Fineness by Blaine apparatus, m2/kg 399 730
effect of maximum aggregate size on the fracture response of plain
Normal consistency, % 30 –
and fiber concrete mixtures was investigated [4,5,9–15] through Initial setting time, minutes 147 –
crack propagation experiments conducted on notched beams. The Final setting time, minutes 273 –
conclusions from these studies indicate that as the maximum Soundness by Le-Chatelier apparatus, mm 0.20 –
aggregate size is increased, the fracture energy is also increased.
Even though a proper reason for this observation has not been pre-
sented in the previous studies, it can be expected that the crack Table 2
propagation time increases with an increase in the size of the Properties of Aggregate.
aggregate. It should be realized that crack in these types of tests
Properties Coarse aggregate Fine
typically propagates along one specific path and thereby if the aggregate
crack encounters a large aggregate in its path, the crack needs to 10 mm 20 mm

meander. As a result of this, the fracture energy of the sample Arithmetic mean 5.372 mm 4.608 mm –
increases with increase in maximum aggregate size. All these roughness
Specific gravity 2.48 2.51 2.66
investigations were conducted to evaluate the fracture characteris- Water absorption 1.78% 1.50% 0.60%
tics of cracks under static loading condition using a notched beam Existing moisture 0.48% 0.50% 0.90%
arrangement. There is hardly any literature available on the effect content
of aggregate size on the fatigue performance of concrete under Grading zone (as Single sized Single sized Zone III
per IS 383– aggregate of aggregate of
dynamic loading conditions. It is known that crack propagation
1970, [21]) nominal size 10 mm nominal size 20 mm
in concrete primarily occurs along the interface of the coarse
S.R. Kasu et al. / Construction and Building Materials 229 (2019) 116922 3

mixed for another one minute. After that, half the total amount of water required
was added to the dry mixture and mixed for about two minutes. The remaining half
portion of cement and fly ash were then added to the mixture and mixed for one
minute. Lastly, the remaining half portion of the water was added and mixed until
a uniform mixture was obtained. The mix was placed in the beam moulds and
packed using a tamping rod and compacted subsequently by placing a plate vibra-
tory compactor on the mould. The moulds were covered for 24 h, demolded and
submerged in a water curing tank for 28 days.

2.3. Flexural strength determination

For each mix (S10 and S20), a set of five specimens was tested for each mix
using a 30 kN capacity, closed loop servo-hydraulic controlled machine under
four-point bending condition. The test setup consists of a bottom loading platen
with two rollers, a reaction frame and, a jig to hold two linear variable differential
transducers (LVDTs) as shown in Fig. 2. A constant loading rate of 1 MPa/min was
maintained until rupture occurs (as per ASTM C78 [23]). The flexural strength of
the specimen was estimated using the simple beam bending equation.

2.4. Flexural fatigue life measurement

The experimental setup used for measuring the modulus of rupture (flexural
Fig. 1. Aggregate grading curves for S10 and S20. strength) of the concrete beam specimens was used for fatigue testing also. The
tests were carried out in load control mode with a haversine waveform varying
between a minimum stress level of Smin to a maximum stress level of Smax as shown
in Fig. 3. An initial seating load of 0.20 kN (Smin of 0.09 MPa) was applied prior to the
maximum aggregate are given in Table 3. Mixtures with 10 mm and 20 mm nom- application of the actual loading. Since the heaviest loads consume most part of the
inal maximum aggregate size were designated as S10 and S20 mixtures fatigue life. The maximum stress level (Smax) was selected as 80, 85, and 90% of the
respectively. static flexural strength. The selection of stress levels were based on initial trials car-
ried out on some beam specimens with different stress levels (70% and 75%) indi-
2.2. Sample preparation cated that the specimens did not fail (even after 10,00,000 repetitions). The initial
stiffness of the specimen was calculated at the 50th load cycle and the variation
In the experimental investigation, 100 beam specimens of of strain and flexural stiffness with constant stress cycles was recorded. Termina-
100 mm  100 mm  500 mm size were prepared to evaluate the flexural strength tion conditions selected for the test are (i). maximum number of load cycles of
(10 specimens) and flexural fatigue life (90 specimens). The coarse aggregate and 1,000,000, (ii). stiffness reduction by 50% of the initial stiffness and (iii). complete
sand were mixed initially for 30 s to obtain a uniform dry aggregate mix. Half the failure/fracture, whichever reaches first. To investigate the effect of frequency, sep-
quantity of the cementitious material (cement and fly ash) was added next and arate sets of specimens were tested at frequencies of 2, 5, and 10 Hz.

Table 3
Mix Proportions (kg/m3).

Variable Water Cement Fly ash Sand 10 mm/20 mm Superplasticizer


S10 206.86 292.43 157.47 834.91 853.48 0.85
S20 202.86 288.01 155.33 824.20 874.98 0.77

Fig. 2. Four-point bending test setup.


4 S.R. Kasu et al. / Construction and Building Materials 229 (2019) 116922

3.2. Flexural fatigue behavior

3.2.1. S-N plots


A total of 90 prismatic specimens comprising 45 specimens
each for the two mixtures with different nominal maximum aggre-
gate sizes (10 mm and 20 mm) were used in this study. The tests
were carried out at three different frequencies; 2, 5, and 10 Hz.
These frequency levels can be considered to correspond to the
vehicular speeds on low, medium, and high traffic volume roads
and bridges. Fatigue lives of specimens of S10 and S20 mixtures
are given in Table 5. Since the observed fatigue life had large scat-
ter, geometric mean was used to estimate the average life of the
mix. Arithmetic means are generally not used for the analysis of
fatigue test results as the fatigue life data will have significant out-
liers and will vary logarithmically. The geometric means of flexural
fatigue lives (after removal of the outliers and runouts, [24,25]) are
shown in Fig. 4(a–c). The geometric mean fatigue lives and stress
level relation follows the linear trend.
Fig. 3. The waveform used in the repeated constant stress fatigue test. It can be observed from the SN plots that irrespective of the fre-
quencies at which the specimens were tested, at 90% stress level,
there is no significant difference between the fatigue lives of S10
and S20 samples. However, significant differences are observed
3. Results and discussion at 80% stress level at which irrespective of the level of frequency,
fatigue lives for S10 sample are higher than those of S20 samples.
3.1. Flexural strength This demonstrates that for low-stress levels, concrete with 10 mm
nominal maximum aggregate size performs better in flexural fati-
Flexural strength (average values of five replicates) of the S10 gue response compared to the concrete mix with 20 mm nominal
and S20 mixtures measured in four-point bending test are pre- maximum aggregate size. It can also be observed from Fig. 4(a–c)
sented in Table 4. The maximum stress (Smax) to be applied in that for samples with larger size (20 mm) aggregates there is not
the fatigue tests was selected as 80, 85, and 90% of the flexural much change in fatigue life once the stress level reaches 85% or
strength to simulate the heavy commercial traffic loading to which higher irrespective of the frequency of loading. A similar state-
the concrete pavements are subjected. ment, however, cannot be made for samples with lower coarse

Table 4
Flexural strengths of concrete mixtures.

Mixture Average failure load, N (Std. dev) Average flexural strength, MPa (Std. dev) Flexural stresses selected for different
stress levels, MPa
80% 85% 90%
S10 9452 (5 6 4) 4.254 (0.25) 3.403 3.616 3.828
S20 9309 (2 1 3) 4.189 (0.10) 3.351 3.561 3.770

Table 5
Summary on Fatigue Lives of S10 and S20 Mixtures.

90% Stress Level 85% Stress Level 80% Stress Level


Frequency, Hz S10 S20 S10 S20 S10 S20
2 440 32a 18,753 179c 98a 26,209
2 1144 2263 70,209 1563 1126c 42,273
2 1851 6281 155,009 10,801 104,872 57,185
2 3083 11,745 377,857 77,249 247,907 64,449
2 NAd 19,313 NAd 110337c NAd 146177c
5 1323 62a 276c 199c 1331c 93a
5 8289 137 1760 919 10129c 4225c
5 13,081 3777 723,969 55,649 103,233 13,097
5 88705c 12,089 832,625 184,961 211,073 70,401
5 194817c 14,369 Terminatedb 616961c 218,369 466,945
10 3175 26a 885 51a 498c 9393
10 8849 640 105,153 775c 5253c 40,193
10 12,633 2757 522,627 7861 66,113 456,449
10 20,609 158,977 675,841 187,009 787,969 468,225
10 54,337 16,545 Terminatedb Terminatedb Terminatedb Terminatedb
a
Rejected as the specimen failed at less than 100 cycles.
b
Rejected as the specimen did not show any cracks, even after 1 million cycles; considered as run out, not included in the analysis.
c
Rejected, performing abnormally from the data sets.
d
Rejected, not available due to power failure.
S.R. Kasu et al. / Construction and Building Materials 229 (2019) 116922 5

Fig. 4. S-N plots for S10 and S20 mixtures: (a, b, & c) Geometric mean fatigue lives for loading frequencies of 2 Hz, 5 Hz, and 10 Hz respectively.

aggregate sizes (10 mm nominal size). A probable reason for this probability plots show that S10 mixture has significantly higher
could be that as the size of the aggregate is reduced, better packing survival probability at 80 and 85% stress levels compared to that
of the grains (coarse aggregates) can be attained. This packing of obtained at 90%. At a survival probability of 50%, the fatigue lives
the grains eventually helps in slowing down the crack propagation. are almost similar for 80 and 85% stress levels for S10 mix. How-
Moreover, better packing also leads to fewer air voids (or pores) ever, as expected, 20% survival probability is attained at lower fati-
which are typically the source points for crack nucleation. Thereby, gue life cycles for 85% stress level compared to that obtained for
at lower stress levels, packing density and porosity play an impor- 80% stress level. No such significant effect on survival probability
tant role in crack nucleation and propagation for samples subjected could be observed for S10 samples with variation in frequencies
to flexural fatigue loads. The observation of higher size of coarse (Fig. 5b). Variation in fatigue life with frequencies is noticeable
aggregates leading to lower fatigue life irrespective of the loading only when the survival probability is less than 20%.
frequencies corroborates the above hypothesis. The K-M lines for S20 samples (Fig. 5c) show that the stress
level influences the fatigue life. However, a different type of behav-
3.2.2. Effect of stress level and frequency on fatigue life ior is being demonstrated in these S20 samples in comparison to
Two different variables were investigated in this study; stress the S10 samples. For the S20 samples, it is observed that 50% sur-
level and frequency. The effect of stress level and frequency has vival probability is almost the same for 85% and 90% stress levels
been examined using non-parametric Kaplan-Meir (K-M) survival compared to 80% stress level. For S10 mix, a similar survival prob-
analysis on rightly censored data, which eventually provides an ability of 50% can be observed for 80 and 85% stress levels. Quite
estimate on the reliability of the fatigue lives of the specimens with similar to the observations noted from S10 samples, no significant
a graphical abstract of survival probability. The non-parametric effect could be observed in S20 samples with change in frequency
distribution analysis yields better results and does not require a level (Fig. 5d).
fitting distribution for the test data. The K-M plots shown in Further, the log-rank test was used to test whether the differ-
Fig. 5 (a-d) are plots between observed life and survival probabil- ence between the survival probabilities of different sets of concrete
ity. The line in the figure appearing to be similar to a staircase in specimens tested at three different stress levels (80, 85 and 90%)
shape indicates the progression of failure occurrence. The survival are statistically different or not. A similar exercise was done to
6 S.R. Kasu et al. / Construction and Building Materials 229 (2019) 116922

Fig. 5. Kaplan-Meir survival plots: (a) & (b) effect of stress level and frequency for S10 mixtures, (c) & (d) effect of stress level and frequency for S20 mixtures.

evaluate whether the effect of the three different frequencies (2, 5 cycle data against failure probability. The hazard function for log-
and 10 Hz) is significant or not. The null hypothesis of the log-rank normal distribution decreases with increasing life which is prob-
statistic is that there is no difference between the survival curves lematic in understanding the material deterioration, whereas it
of three groups (either three stress levels or three frequency increases with increasing life for Weibull distribution which is
levels). The observed and expected failures are similar if the null practical and imperative to use [26,27]. The 3-parameter Weibull
hypothesis is true. With the involvement of variance and covari- distribution suits well when fatigue life data do have non-zero
ance of observed and expected failures, the test statistic was location parameter [28]. The probability density function (pdf)
approximated as chi-square distribution with a degree of freedom of the Weibull distribution is
2. The critical p-values for log-rank statistic are found by referring
 b1
to chi-square tabular forms. The p-values are found to be 0.000 and b tc tc b
f ðtÞ ¼ eð g Þ ð1Þ
0.015 respectively for the mixtures S10 and S20 indicating that g g
stress level affects the fatigue life significantly. A similar statistical
test yielded p-values of 0.468 and 0.422 for frequencies showing Where t is the fatigue life; b is the shape parameter; ƞ is the scale
that the effect of frequency is insignificant. parameter; c is the location parameter.
The parametric estimation can be done by using the method of
3.2.3. Fatigue life distribution maximum likelihood estimates (MLE). The log-likelihood function
The previous section dealt with the observations made based for n failures and their fatigue lives:
on the geometric mean values of fatigue life of two different
X X  b
n n
ti  c
types of mixtures tested at different stress levels and frequencies. K ¼ n lnðbÞ  n b lnðgÞ þ ðb  1Þ ln ðti  cÞ  ln
It is also necessary to evaluate the failure probability of the mix- i¼1 i¼1
g
tures (S10 and S20) with respect to the fatigue life cycles. It ð2Þ
should be noted that these data are scattered in nature and
thereby probabilistic analysis and fitting distribution are desirable The least square estimator (rank regression method) of location
to be established. Lognormal, 2-parameter and 3-parameter Wei- parameter is used and considered as fixed. The Eq. (2) will be
bull distributions are most commonly used to fit the fatigue life reduced to Eq. (3), taking xi = ti  c
S.R. Kasu et al. / Construction and Building Materials 229 (2019) 116922 7

Table 6 Table 7
Goodness of fit test results for S10 for 3-parameter Weibull distribution. Maximum Likelihood Estimates (MLE) of Distribution Parameters.
i ti F(ti) F(tn+k-i) ith-term Distribution Parameter Value
1 440 0.045734 0.947704 0.201191 S10 S20
2 885 0.078071 0.943298 0.542008
3 1144 0.091417 0.93607 0.857049 3-parameter Weibull Shape/slope, b 0.504 0.507
4 1323 0.099481 0.929788 1.158272 Scale, ƞ 97254.148 37399.477
5 1760 0.116548 0.903026 1.344828 Location/Threshold, ! 216.260 67.604
6 1851 0.119749 0.862105 1.504662 Log-normal Mean, m 10.396 9.480
7 3083 0.155732 0.798479 1.499976 Standard deviation, r 2.340 2.257
8 3175 0.158022 0.777439 1.673787
9 8289 0.248177 0.771681 1.626687
10 8849 0.255515 0.717463 1.664667
Similarly, the solution will be obtained for b; by substituting ƞ
11 12,633 0.298395 0.646214 1.573881
12 13,081 0.302862 0.645717 1.711306 (of Eq. (4)) in Eq. (3) and taking derivative with respect to b. The
13 18,753 0.351889 0.642792 1.72823 derivative of equation will be further equating to zero and solve
14 20,609 0.365591 0.571399 1.668123 for the final solution b.
15 54,337 0.524902 0.56039 1.417531
The goodness of fit measures (Anderson-Darling (AD) [29,30]
16 66,113 0.56039 0.524902 1.367468
17 70,209 0.571399 0.365591 1.116201
tests) demonstrate that the 3-parameter Weibull distribution pro-
18 103,233 0.642792 0.351889 1.021566 vides a better fit to data in comparison to the other distributions.
19 104,872 0.645717 0.302862 0.984404 The AD statistic test makes use of the specific distribution (Weibull
20 105,153 0.646214 0.298395 1.028311 or lognormal) for calculating critical values. The null hypothesis is
21 155,009 0.717463 0.255515 0.857031
that the distribution of test data follows a specific form if the test
22 211,073 0.771681 0.248177 0.780362
23 218,369 0.777439 0.158022 0.635628 statistic AD is smaller than the critical value of 0.757 at 5% signif-
24 247,907 0.798479 0.155732 0.617786 icance level. The regular form of AD statistic can be defined as
25 377,857 0.862105 0.119749 0.450679
26 522,627 0.903026 0.116548 0.384068 AD2 ¼ n  S ð5Þ
27 675,841 0.929788 0.099481 0.313729
28 723,969 0.93607 0.091417 0.296878 Xn
1
29 787,969 0.943298 0.078071 0.265354 S¼ ð2i  1Þ½ln Fðti Þ þ lnð1  Fðtnþ1i ÞÞ ð6Þ
30 832,625 0.947704 0.045734 0.197702 i¼1
n
Ʃ ith term = 30.48937
AD = 30.48937 – 30 = 0.48937
pffiffiffiffiffiffi
X
n
th
Adj. AD = 0. 48937(1 + 0.2 30Þ = 0.507236 S¼ i term ð7Þ
i¼1

 
 b 0:2
X
n X
n
xi Adjusted AD ¼ AD 1 þ pffiffiffi ð8Þ
K ¼ n lnðbÞ  n b lnðgÞ þ ðb  1Þ lnðxi Þ  ln ð3Þ n
i¼1 i¼1
g
Where n is the sample size, i is the order number of the data point, F
Taking the derivative with respect to ƞ: to obtain ƞ, the Eq. (3) (t) is the cumulative distribution function of the specific distribu-
further, be reduced by equating to zero. The MLE solution for ƞ will tion and ti must be ordered fatigue data. The results obtained by
become AD test for S10 mix for Weibull distribution are presented in Table 6.
" #b The analysis was carried out as per the procedure given by Refs. [29]
1 XN
 b and [31].
xi ð4Þ
N i¼1 Fig. 6 shows the 3-parameter Weibull distribution fitting with
the corresponding AD (Eq. (5)) and p-values. Lower AD statistic val-

Fig. 6. Three-parameter Weibull distribution fitting for fatigue life: (a) S10 and (b) S20 mixtures.
8 S.R. Kasu et al. / Construction and Building Materials 229 (2019) 116922

Fig. 7. (a) Probability density and (b) cumulative distribution plots for three-parameter Weibull distribution: S10 and S20 mixtures.

ues of 0.487 and 0.243 (3-parameter Weibull) are obtained com- the Kaplan-Meir (KM plots) survival probability plots along
pared to the values of 0.595 and 0.311 (lognormal) obtained for with p-value statistics.
S10 and S20 mixtures respectively.  3-parameter Weibull distribution fits the failure probability vs.
It can be estimated from Fig. 6 that 50,000 cycles fatigue life has fatigue life plots better for both S10 and S20 samples in compar-
about 50% failure probability for S10 mix in comparison to the fati- ison to other distributions such as lognormal. The Anderson-
gue life of 18,000 cycles obtained for S20 mix with similar failure darling statistic has been used to demonstrate the goodness of
probability. This also demonstrates that irrespective of the stress fit of the Weibull distribution.
level and or loading frequency, S10 samples have better flexural  The distribution characteristic shape parameter (b) value of less
fatigue characteristics compared to those of S20 samples. The max- than 1.0 indicates early life failures for the considered stress
imum likelihood estimates of the Weibull distribution parameters levels.
are given in Table 7. The slope, scale, and location parameters gov-  Probability density and cumulative distribution plots depict
ern the distribution characteristics such as the shape of the curve, that the useful life (minimum and maximum lives) of S10 mix-
reliability, and failure rate respectively. Slope parameter (b), which ture is relatively higher than that of S20 mixture.
is a non-dimensional parameter was found to be 0.504 and 0.507
(less than one) for both the mixtures, indicating a hazard function
which decreases sharply and monotonically and is convex in shape Declaration of Competing Interest
(Fig. 7a & b). This suggests a high probability of early failure.
It can be noted from the probability density and cumulative dis- The authors declare that they have no known competing finan-
tribution plots presented in Fig. 7a & b, that the failure probability cial interests or personal relationships that could have appeared
for S20 mix is higher than that of S10 mix. In other words, the use- to influence the work reported in this paper.
ful fatigue life of S10 mix is higher than that of S20 mix. It also
depicts that minimum and maximum lives of S10 mix are signifi-
Acknowledgments
cantly higher when compared to the S20 mix. The fatigue behavior
of the mixtures considered in this study is highly sensitive to the
The authors would like to acknowledge Late Prof. B.B. Pandey,
stress level rather than frequency. This might be due to the use
Advisor, SRIC, Indian Institute of Technology Kharagpur for his con-
of larger size aggregate which results in a lower surface area for
stant support, motivation, and suggestions during the experimen-
the development of gel bonds, which is responsible for the lower
tal study. The authors also would like to thank the reviewers for
strength of concrete and probably more heterogeneity in concrete.
their valuable suggestions.

4. Conclusions
Appendix A. Supplementary data
In this study, the normal strength concrete prismatic specimens
were tested under haversine cyclic loading and the resulting fati- Supplementary data to this article can be found online at
gue data of mixtures with two different sizes of aggregate https://doi.org/10.1016/j.conbuildmat.2019.116922.
(10 mm and 20 mm) were used for carrying out probabilistic fati-
gue life analysis. It can be concluded that References

[1] K.D. Smith, J.R. Roesler, Review of fatigue models for concrete airfield
 Fatigue lives of samples with 10 mm nominal maximum aggre-
pavement design, in: Airfield Pavements: Challenges and New Technologies,
gate size (S10) are significantly higher compared to 20 mm 2004, pp. 231–258.
nominal aggregate size (S20) for 80% stress level but are almost [2] K. Keerthana, J.C. Kishen, An experimental and analytical study on fatigue
similar for 90% stress level irrespective of the loading frequency. damage in concrete under variable amplitude loading, Int. J. Fatigue 111 (2018)
278–288.
 Stress level influences the fatigue life for both S10 and S20 sam- [3] R. Ince, K.E. Alyamaç, Determination of fracture parameters of concrete based
ples but not the loading frequency. This has been noted through on water-cement ratio, Indian J. Eng. Mater. Sci. 15 (2008) 14–22.
S.R. Kasu et al. / Construction and Building Materials 229 (2019) 116922 9

[4] M.H. Beygi, M.T. Kazemi, J.V. Amiri, I.M. Nikbin, S. Rabbanifar, E. Rahmani, [18] IS 3812-2003, Pulverized fuel ash Specification, Part I: For use as pozzolana in
Evaluation of the effect of maximum aggregate size on fracture behavior of self cement mortar and concrete (Second Revision). Bureau of Indian Standard,
compacting concrete, Constr. Build. Mater. 55 (2014) 202–211. New Delhi (India)-110002.
[5] A.P.N. Siregar, M.I. Rafiq, M. Mulheron, Experimental investigation of the [19] American Concrete Institute Committee 226, Use of Fly Ash in Concrete.
effects of aggregate size distribution on the fracture behaviour of high strength Manual of Concrete Practice, Part 1, ACI 226, 3R-1-226, 3R-29 1987, American
concrete, Constr. Build. Mater. 150 (2017) 252–259. Concrete Institute, Farmington Hills, Michigan, USA.
[6] B. Akcay, A.S. Agar-Ozbek, F. Bayramov, H.N. Atahan, C. Sengul, M.A. Tasdemir, [20] IRC SP 46-2013, Guidelines for Design and Construction of Fibre reinforced
Interpretation of aggregate volume fraction effects on fracture behavior of Concrete Pavements, Indian Roads Congress, 2013.
concrete, Constr. Build. Mater. 28 (1) (2012) 437–443. [21] IS 383-1970, Specifications for coarse and fine aggregates from natural sources
[7] B. Chen, J. Liu, Effect of aggregate on the fracture behavior of high strength for concrete. Bureau of Indian Standard, New Delhi (India)-110002.
concrete, Constr. Build. Mater. 18 (8) (2004) 585–590. [22] IRC 58-2015, Guidelines for the Design of Plain Jointed Rigid Pavements for
[8] C.G. Rocco, M. Elices, Effect of aggregate shape on the mechanical properties of Highways, Fourth Edition, Indian Roads Congress, New Delhi, India, 2015.
a simple concrete, Eng. Fract. Mech. 76 (2) (2009) 286–298. [23] ASTM, Standard Test Method for Flexural Strength of Concrete (Using Simple
[9] P. Nallathambi, B.L. Karihaloo, B.S. Heaton, Effect of specimen and crack sizes, Beam with Third-Point Loading), ASTM C78/C78M-15, West Conshohocken, PA
water/cement ratio and coarse aggregate texture upon fracture toughness of 19428-2959, USA, 2010.
concrete, Mag. Concr. Res. 36 (129) (1984) 227–236. [24] S.P. Singh, S.K. Kaushik, Fatigue strength of steel fiber reinforced concrete in
[10] P.E. Petersson, Fracture energy of concrete: practical performance and flexure, Cem. Concr. Compos. 25 (7) (2003) 779–786.
experimental results, Cem. Concr. Res. 10 (1) (1980) 91–101. [25] Y. Mohammadi, S.K. Kaushik, Flexural fatigue-life distributions of plain and
[11] G.A. Rao, B.R. Prasad, Fracture energy and softening behavior of high-strength fibrous concrete at various stress levels, J. Mater. Civ. Eng. 17 (6) (2005) 650–
concrete, Cem. Concr. Res. 32 (2) (2002) 247–252. 658.
[12] J. Zhang, Q. Liu, L. Wang, Effect of coarse aggregate size on relationship [26] E. Castillo, A. Fernendez-Canteli, A Unified Statistical Methodology for
between stress and crack opening in normal and high strength concretes, J. Modeling Fatigue Damage, Springer, Netherlands, 2009. doi:10.1007/978-1-
Mater. Sci. Technol. 21 (5) (2005) 691–700. 4020-9182-7.
[13] I.M. Nikbin, M.H.A. Beygi, M.T. Kazemi, J.V. Amiri, E. Rahmani, S. Rabbanifar, M. [27] A. Hanif, M. Usman, Z. Lu, Y. Cheng, Z. Li, Flexural fatigue behaviour of thin
Eslami, Effect of coarse aggregate volume on fracture behavior of self laminated cementitious composites incorporating cenosphere fillers, Mater.
compacting concrete, Constr. Build. Mater. 52 (2014) 137–145. Des. 140 (2018) 267–277, https://doi.org/10.1016/j.matdes.2017.12.003.
[14] M. Karamloo, M. Mazloom, G. Payganeh, Effects of maximum aggregate size on [28] A. Hanif, Y. Kim, C. Park, Numerical validation of two-parameter Weibull
fracture behaviors of self-compacting lightweight concrete, Constr. Build. model for assessing failure fatigue lives of laminated cementitious
Mater. 123 (2016) 508–515. composites—comparative assessment of modeling approaches, Materials 12
[15] M. Ghasemi, M.R. Ghasemi, S.R. Mousavi, Investigating the effects of maximum (1) (2019) 110.
aggregate size on self-compacting steel fiber reinforced concrete fracture [29] R.B. D’Agostino, Goodness-of-fit-techniques, vol. 68, CRC Press, 1986.
parameters, Constr. Build. Mater. 162 (2018) 674–682. [30] A.K. Chandrappa, K.P. Biligiri, Flexural-fatigue characteristics of pervious
[16] S. Deb, N. Mitra, S.B. Majumder, S. Maitra, Improvement in tensile and flexural concrete: statistical distributions and model development, Constr. Build.
ductility with the addition of different types of polypropylene fibers in Mater. 153 (2017) 1–15.
cementitious composites, Constr. Build. Mater. 180 (2018) 405–411. [31] K.M.A. Sohel, K. Al-Jabri, M.H. Zhang, J.R. Liew, Flexural fatigue behavior of
[17] IS 12269-1987, Specifications for 53 grade ordinary Portland cement. Bureau ultra-lightweight cement composite and high strength lightweight aggregate
of Indian Standard, New Delhi (India)-110002. concrete, Constr. Build. Mater. 173 (2018) 90–100.

You might also like