Construction and Building Materials 153 (2017) 1–15

Contents lists available at ScienceDirect

Construction and Building Materials

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

Flexural-fatigue characteristics of pervious concrete: Statistical

distributions and model development
Anush K. Chandrappa, Krishna Prapoorna Biligiri ⇑
Department of Civil Engineering, Indian Institute of Technology Kharagpur, West Bengal 721 302, India

h i g h l i g h t s

Performed flexural-fatigue test at different stress levels and frequencies.

Found fatigue life significantly dependent on stress levels and stiffness.

Observed a three-stage process for stiffness degradation and tensile strain accumulation.

Distribution fitting indicated characteristic infant mortality type of failure.

Developed laboratory fatigue models for fatigue life predictions.

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

Article history: Pervious concrete pavements that are widely used as surface layers in low volume road applications such
Received 29 March 2017 as parking lots, residential streets, etc. are subjected to repeated loads, which may undergo fatigue.
Received in revised form 4 July 2017 Although most of the past studies have focused on the compressive strength and hydraulic properties,
Accepted 8 July 2017
limited investigations have been reported in respect of fatigue in pervious concrete (PC). The objective
Available online 15 July 2017
of this study was to investigate the effect of stress levels and frequencies on the fatigue life and develop
laboratory fatigue models of PC mixtures. The S-N curves indicated that power-function relation exists
Keywords:
between stress levels and fatigue life of PC. The non-parametric and parametric methods indicated that
Pervious concrete
Prisms
the stress levels (0.70, 0.75, and 0.80) and stiffness were significant in affecting the fatigue life compared
Fatigue to the loading frequency in the range of 2–10 Hz. The fatigue relations of PC fitted with 3-Parameter
Frequency Weibull and lognormal distributions indicated the characteristic infant mortality type of failure behavior,
Distribution fitting represented by initial portion of the reliability bathtub curve. The stiffness degradation and tensile strain
Fatigue predictive models accumulation depicted a three-stage process with rapid degradation in the primary stage, stable and con-
stant degradation in secondary phase, and accelerated degradation in the tertiary stage. The laboratory
fatigue models developed had good-to-fair statistical goodness of fit parameters (0.65  R2  0.85), which
can be used to predict the fatigue lives of pervious concrete in case of non-availability of testing
capabilities.
Ó 2017 Elsevier Ltd. All rights reserved.

1. Introduction for use as surface course layers in low volume roads such as resi-
dential streets, walkways, and parking lots, etc., where the traffic
Pervious concrete pavement systems are a class of rigid pave- volume in terms of average annual daily traffic (AADT) is assumed
ments that allow storm water to percolate through them owing to be 400. From its inception, pervious concrete has been studied
to their macroporous internal structure. The main difference under the various facets of mix design, strength characteristics,
between pervious concrete and conventional concrete pavements hydraulic properties, pore structure, durability, modeling, and
lies in the fact that the amount of fines are reduced/eliminated in non-destructive testing, chiefly to understand the material behav-
gradations used in pervious concrete paving mixtures. At present, ior and its suitability as a pavement material [1–8]. The most
pervious concrete pavements are the most promising materials important property considered in designing a concrete material
is compressive strength; in pervious concrete, the magnitudes of
compressive strength vary from 3 to 25 MPa, where the porosities
⇑ Corresponding author.
in most of the mixtures lying in the range of 15–35%, and more so,
E-mail addresses: [email protected] (A.K. Chandrappa), krishnapbiligiri@
20–25% as the preferred choice [9–14].
gmail.com (K.P. Biligiri).

http://dx.doi.org/10.1016/j.conbuildmat.2017.07.081
0950-0618/Ó 2017 Elsevier Ltd. All rights reserved.
2 A.K. Chandrappa, K.P. Biligiri / Construction and Building Materials 153 (2017) 1–15

However, it is noteworthy that pervious concrete pavements lognormal distribution as these two distributions are widely used
are subjected to flexural stresses and fatigue due to vehicular loads, in analyzing fatigue life data. Additionally, there are no readily
and these aspects have not been studied extensively. In the past, available laboratory fatigue models for pervious concrete and
researchers found that the axial fatigue behavior of polymer mod- therefore, would be required for future estimation of fatigue lives
ification in pervious concrete was pronounced at higher stress for quality control assessment. Further, these fatigue models
levels [15]. But, pavement performance that is significantly influ- would as well indicate the performance of pervious concrete pave-
enced due to lower stresses, no significant differences were ments when subjected to traffic loads with suitable calibration.
observed between fatigue lives of modified and unmodified pervi- With these gaps as the research lacunae, the objective of this
ous concrete mixtures at those stress levels. Chen et al. [16] stud- study was to comprehensively investigate the fatigue behavior of
ied fatigue behavior at three different stress levels of 0.70, 0.80, pervious concrete paving mixtures by considering the effect of
and 0.90, and found that polymer modification improved the fati- both frequency and stress level. Further, the fatigue database so
gue life compared to those modified by supplementary cementi- developed was used to establish predictive equations (models).
tious materials. Recently, flexural fatigue tests on polymer- The research outline of the study included (Fig. 1):
modified pervious concrete (with single sized aggregates: 27.5,
32.5, and 37.5 mm) for base course applications revealed that with
Design and production of pervious concrete beams with two
increase in the aggregate size, the flexural fatigue life reduced. gradations, one w/c ratio and one c/a ratio
Based on the current literature, no studies are available that
Estimation of density and porosity of pervious concrete beam
have investigated the effect of frequency on the fatigue life of per- samples
vious concrete, although it is worth mentioning that few of the
Determination of flexural strength of pervious concrete beams
past studies indicate that there is no effect of frequency on the fati-
Selection of stress levels and frequencies pertinent to field
gue life of a conventional concrete in the range of 1.2–20 Hz conditions
[18,19]. But then, pervious concrete being very different from con-
Conducting fatigue tests at fixed stress level and frequency
ventional concrete in terms of structural matrix and internal pore
Understanding the effect of frequency and stress level using
structure; it is essential to investigate the effect of frequency and parametric and non-parametric methods
stress level as these both parameters have direct relations with
Fitting distribution for the fatigue data and establishment of
vehicular traffic over a wide variety of different classes of road. laboratory based fatigue models.
Further, a 2-parameter Weibull distribution is usually used to fit
fatigue data for conventional concrete or other materials. However,
any fatigue data acquired from mechanical tests possesses a mini- 2. Research significance
mum value of fatigue life, and therefore, the applicability of the 3-
parameter Weibull distribution should be considered along with Eller and Izevbekhai [20] in their first year performance report
on pervious concrete test sections identified fatigue-type cracks

Investigations on Fatigue Behavior of Pervious Concrete

Production of Pervious Concrete Beams / Prisms

Determination of Density and Porosity

(ASTM C1754)

Determination of Flexural Strength (ASTM C78)

Selection of Stress Levels and Frequencies

Conducting fatigue tests Non-parametric method

(Kaplan-Meier Analysis)

Effect of Frequency and Stress Levels on Fatigue life

Parametric method
Fitting the Statistical Distribution (Analysis of Covariance)

Development of Laboratory Fatigue Model

Fig. 1. Research Outline.

Table 1
Mix proportions for pervious concrete mixtures.

Gradation w/c ratio c/a ratio Aggregate, kg/m3 Cement kg/m3 Water, kg/m3
P2 0.35 0.25 1506.12 376.53 131.78
P6 0.35 0.25 1520.11 380.02 133
 A.K. Chandrappa, K.P. Biligiri / Construction and Building Materials 153 (2017) 1–15 3

Fig. 2. Flexural test setup: (a) schematic showing span of the beam; (b) actual experimental setup.

Table 2
Density and porosity of flexural strength specimens.
Table 3
Mixture Average density, kg/m3 Average porosity,% Average density and porosity of beam specimens of fatigue tests.
(Std. dev) (Std. dev)
P2-0.35-0.25
P2-0.35-0.25 1997.84 (39.87) 27.82 (2.24)
P6-0.35-0.25 2167.56 (27.26) 20.19 (1.63) Density Frequency
Stress level 2 Hz 5 Hz 10 Hz
0.80 2001.92 (20.80) 1966.5 (81.89) 1920.04 (31.46)
0.75 1973.9 (55.46) 1970.5 (88.53) 2007.48 (56.53)
that initiated through aggregate-cement interface or voids and 0.70 1971.46 (2.04) 1994.1 (59.33) 1937.4 (42.14)

often propagated through the coarse aggregates. These field obser- Porosity Frequency
vations indicated that fatigue was one of the distresses in pervious Stress level 2 Hz 5 Hz 10 Hz
concrete, and needs to be taken care of in order to avoid early fail- 0.80 27.06 (1.70) 28.72 (3.98) 30.33 (1.85)
ure. The results presented in this present paper are based on per- 0.75 28.74 (2.43) 28.90 (3.70) 27.12 (1.84)
vious concrete mixtures produced with 20–30% porosity. 0.70 29.16 (0.28) 28.08 (2.23) 30.1 (0.31)
Comprehensive investigations were performed in respect of fatigue P6-0.35-0.25
behavior of pervious concrete, where new insights were provided Density Frequency
about the effect of different test parameters using different meth- Stress level 2 Hz 5 Hz 10 Hz
ods. Statistical distributions were fitted to the fatigue data using
0.8 2158.88 (24.74) 2175.5 (27.74) 2134.4 (38.65)
robust methods and the effects of different distribution parameters 0.75 2165.08 (22.63) 2143.6 (13.98) 2118.16 (37.51)
on the observed fatigue lives of pervious concrete were discussed 0.7 2169.33 (24.19) 2166.95 (15.62) 2149.92 (35.17)
in detail. Porosity Frequency
The study found that the failure strain in pervious concrete was
Stress level 2 Hz 5 Hz 10 Hz
usually higher compared to that normally observed in conven-
0.8 20.98 (2.02) 19.73 (2.13) 22.33 (2.34)
tional concrete, indicating that pervious concrete is more ductile
0.75 20.25 (1.72) 21.18 (1.03) 22.69 (2.34)
and possesses higher strain carrying capacity. Laboratory scale fati- 0.7 19.89 (1.74) 20.19 (0.97) 21.52 (1.93)
gue models were developed that included the effect of stress level,
4 A.K. Chandrappa, K.P. Biligiri / Construction and Building Materials 153 (2017) 1–15

Table 4
Flexural strength of pervious concrete mixtures.

Mixture Average failure load, Average flexural strength, Flexural stress at different
N (Std. dev) MPa (Std. dev) stress levels, MPa
0.70 0.75 0.80
P2 4937.77 (569.90) 2.220 (0.25) 1.554 1.665 1.776
P6 6430.4 (182.22) 2.890 (0.082) 2.023 2.167 2.312

porosity, and flexural stiffness, which are first-of-its-kind in the 3. Materials and methods
domain of pervious concrete research. It is envisioned that these
3.1. Materials
models will certainly help researchers, designers, and practitioners
to get an estimate of fatigue lives of different pervious concrete The calcareous and crushed aggregates, which were used, included four sizes
mixtures a priori actual implementation of the product in field with maximum sieve sizes of 6.7, 9.5, 13.2, and 19 mm with specific gravities of
applications. 2.865, 2.807, 2.844, and 2.892, respectively. These aggregates were obtained from
one of the ongoing highway construction projects, in which the same material type

Table 5
Fatigue lives of P2-0.35-0.25.

Stress level = 0.80

Frequency N, cycles Log10N Ave. Log10N Std. dev Log10N Normalized Log10N Chauvenet’s criterion (T)
2 820 2.914 3.014 0.919 0.109 N = 5, T = 1.645
2 2520 3.401 0.422 N = 4, T = 1.534
2 50 1.699 1.43 N = 3, T = 1.383
2 16650 4.221 1.314
2 680 2.833 0.197
5 280 2.447 3.246 0.954 0.837
5 2160 3.334 0.093
5 36100 4.558 1.375
5 440 2.643 0.631
5 NA NA
10 2950 3.47 3.174 0.615 0.482
10 160 2.204 1.577
10 1960 3.292 0.193
10 1130 3.053 0.196
10 7060 3.849 1.098
Stress level = 0.75
Frequency N, cycles Log10N Ave. Log10N Std. dev Log10N Normalized Log10N Chauvenet’s criterion (T)
2 9300 3.968 3.767 0.787 0.256 N = 5, T = 1.645
2 33610 4.526 0.965 N = 4, T = 1.534
2 460 2.663 1.404 N = 3, T = 1.383
2 8160 3.912 0.183
2 NA NA
5 4550 3.658 3.886 1.096 0.208
5 6430 3.808 0.071
5 229410 5.361 1.346
5 520 2.716 1.068
5 NA NA
10 1220 3.086 3.37 0.416 0.682
10 2270 3.356 0.034
10 7880 3.897 1.266
10 730 2.863 1.218
10 4450 3.648 0.669
Stress level = 0.70
Frequency N, cycles Log10N Ave. Log10N Std. dev Log10N Normalized Log10N Chauvenet’s criterion (T)
2 9750 3.989 4.106 0.595 0.196 N = 5, T = 1.645
2 56280 4.75 1.083 N = 4, T = 1.534
2 3780 3.577 0.888 N = 3, T = 1.383
2 Terminated
2 Terminated
5 46450 4.667 4.542 0.662 0.189
5 9490 3.977 0.853
5 12490 4.097 0.672
5 266920 5.426 1.336
5 Terminated
10 327580 5.515 4.709 1.14 0.707
10 8010 3.904 0.707
10 Terminated
10 Terminated
10 Terminated
 A.K. Chandrappa, K.P. Biligiri / Construction and Building Materials 153 (2017) 1–15 5

was used. Ordinary Portland cement of 53-grade conforming to IS 12269 [21] was each mixture, fifty beams of dimensions 500 mm  100 mm  100 mm were pre-
used to coat and bind the aggregate particles. A commercially available ligno- pared, where five beams were used to determine the flexural strength and the
sulphate based plasticizer was used at a rate of 0.2% by weight of cement as pre- remaining forty-five were used for fatigue test evaluation.
scribed by the manufacturer. The fatigue mechanical testing program consisted of three stress levels (ratio of
flexural stress applied to average flexural strength of five beams): 0.70, 0.75, and
3.2. Experimental program 0.80, and three frequency levels: 2, 5, and 10 Hz. These frequency levels were
selected anticipating that 2, 5, and 10 Hz would correspond to speeds of vehicular
Based on the aforementioned aggregate sizes, two gradations were prepared, traffic anticipated on low-medium volume roads. More so, ASTM D7460 [22] rec-
which included: ommends conducting the fatigue tests between 5 and 10 Hz. Further, almost
always, slow moving traffic could be expected on pervious concrete pavements in

P2: single sized gradation consisting of aggregates of sizes passing 9.5 mm and parking lots, and medium-speed traffic could be expected on residential streets, ser-
retained on 6.7 mm. vice roads, etc. At the same time, high-speed traffic may be expected on pervious

P6: dense/quaternary gradation consisting of 25% of 19–13.2 mm + 25% of 13.2– concrete when it is used as a drainable base below conventional concrete
9.5 mm + 25% of 9.5–6.7 mm + 25% of 6.7–4.75 mm pavements.

These gradations were selected in order to understand the fatigue behavior of 3.3. Specimen preparation
single size graded and blended pervious concrete paving mixtures. The water-to-
cement ratio and cement-to-aggregate ratio were 0.35 and 0.25, respectively. In A drum-type concrete mixer was utilized to prepare fresh pervious concrete
short, two pervious concrete mixtures were investigated, which were designated: mixtures. The procedure consisted of first loading the mixer with dry aggregates
P2-0.35-0.25 and P6-0.35-0.25, whose mix proportions are shown in Table 1. For and a small amount of gauged water, and mixing for 1-min. On the wet aggregates,

Table 6
Fatigue lives of P6-0.35-0.25.

Stress level = 0.80

Frequency N, cycles Log10N Ave. Log10N Std. dev Log10N Normalized Log10N Chauvenet’s criterion (T)
2 20360 4.309 2.829 0.966 1.531 N = 5, T = 1.645
2 800 2.903 0.076 N = 4, T = 1.534
2 820 2.914 0.087 N = 3, T = 1.383
2 50 1.699 1.17
2 210 2.322 0.525
5 1040 3.017 2.587 0.484 0.888
5 80 1.903 1.415
5 640 2.806 0.452
5 420 2.623 0.074
5 NA NA
10 590 2.771 2.527 0.216 1.132
10 230 2.362 0.764
10 280 2.447 0.368
10 NA NA
10 NA NA
Stress level = 0.75
Frequency N, cycles Log10N Ave. Log10N Std. dev Log10N Normalized Log10N Chauvenet’s criterion (T)
2 1030 3.013 3.369 1.18 0.302 N = 5, T = 1.645
2 49870 4.698 1.126 N = 4, T = 1.534
2 40 1.602 1.497 N = 3, T = 1.383
2 12980 4.113 0.631
2 2620 3.418 0.042
5 270 2.431 2.795 0.869 0.419
5 10480 4.02 1.41
5 50 1.699 1.262
5 1560 3.193 0.458
5 430 2.633 0.186
10 5580 3.747 2.896 0.625 1.359
10 400 2.602 0.47
10 1550 3.19 0.47
10 120 2.079 1.306
10 730 2.863 0.053
Stress level = 0.70
Frequency N, cycles Log10N Ave. Log10N Std. dev Log10N Normalized Log10N Chauvenet’s criterion (T)
2 316580 5.5 4.816 0.705 0.970 N = 5, T = 1.645
2 71540 4.855 0.08 N = 4, T = 1.534
2 Terminated N = 3, T = 1.383
2 Terminated
2 12360 4.092 1.026
5 63680 4.804 4.4 0.485 0.833
5 23270 4.367 0.068
5 5350 3.728 1.385
5 50180 4.701 0.62
5 Terminated
10 13770 4.139 4.52 0.719 0.53
10 52610 4.721 0.28
10 268030 5.428 1.264
10 60770 4.784 0.367
10 3360 3.526 1.382
6 A.K. Chandrappa, K.P. Biligiri / Construction and Building Materials 153 (2017) 1–15

cement was added and mixed for 1-min until a uniform distribution of the cement PL
Flexural Strength ¼ 2
ð1Þ
material was observed. The remaining water was added to the dry mix of cement bd
and aggregate, and mixing was continued for about 2–3 min. The fresh mixture
was transferred on a tray before placing it in the mold.
The fresh mixture was compacted into beams in two layers, where the total 108  D  h
mixture was divided into approximately two halves with the bottom layers com-
Tensile strain @ bottom ¼ ð2Þ
23  L2
pacted by the application of 20 blows using a standard Proctor hammer of mass
2.5 kg. After compacting the first layer, the second layer was placed with a small
heap above the brim of the mold. The second layer was compacted using a plate where, P = Load, kN; L = Span length, mm; b = width, mm; d = depth, mm; D = deflec-
vibrator (note that vibration energy was not known) that was passed over the mixture tion, mm.
four times that rendered almost a flat surface, following which the smoothened fin-
ishing of the beam was accomplished using a trowel. The freshly compacted beams
were covered with plastic wrap and left undisturbed for 24 h. After 24 h, the spec- 3.6. Flexural fatigue tests
imens were de-molded and placed in curing tanks for 27 days.
The experimental setup utilized for flexural strength test was used in flexural
3.4. Determination of density and porosity fatigue tests as well. During testing, a seating load of 0.2 kN was applied at all stress
levels and frequencies. The fatigue data, which included tensile strain at the bottom
The density and porosity for all the 100 beams (two types of mixtures) were of the specimen, flexural stiffness, normalized modulus x cycles versus cycles were
determined as per ASTM C1754 [23]. The dry density was calculated by determining recorded for every 10th cycle. The initial flexural stiffness was recorded at the 50th
the oven-dried mass of beams conditioned at 50 °C for seven days. The submerged cycle from the start of the test. Since fatigue test is time consuming and expensive,
mass was determined using a vacuum stick since it was found that the vacuum test termination conditions are essential as there are chances that some specimens
action of rubber removed the air voids easily. may not fail during testing. The termination conditions were as under, whichever
occurred earlier:
3.5. Determination of flexural strength

Maximum number of cycles: 500,000
Flexural strength was determined using a closed-loop servo controlled hydrau-
Complete failure
lic testing system in third-point bending mode. The experimental test setup (Fig. 2)
consisted of a reaction frame, bottom-loading system, and a jig to place the linear The maximum number of cycles for terminating the test was limited to 500,000
variable differential transducers (LVDTs). The loading rate for the flexural test cycles in order to reduce the test time and based on the trial tests on several sam-
was set as 1 MPa/min per ASTM C78 [24] that is commonly used for conventional ples. The complete failure was considered as one of the other test conditions since
concrete. The flexural strength and tensile strain at the bottom of the specimen pervious concrete underwent sudden collapse during fatigue tests that was
was determined using Eqs. (1) and (2) [22]. observed in the trial tests. The specimens that underwent complete failure were
preserved for investigating the failure pattern and crack propagation.

Fig. 3. S-N relations for pervious concrete: (a) & (c) individual fatigue lives of P2-0.35-0.25 and P6-0.35-0.25 respectively & (d) geometric mean of fatigue lives of P2-0.35-0.25
and P6-0.35-0.25 Respectively.
 A.K. Chandrappa, K.P. Biligiri / Construction and Building Materials 153 (2017) 1–15 7

4. Results and analyses in a controlled manner mainly to avoid large differences in the
properties. The average densities and porosities of flexural strength
4.1. Density and porosity specimens along with standard deviations in parentheses are
shown in Table 2. The beam specimens were selected randomly
In order to maintain similar porosities of specimens for both for fatigue tests at different stress levels and frequencies, whose
flexural strength and fatigue evaluation, compaction was achieved average densities and porosities along with standard deviations

Fig. 4. Kaplan-Meier survival analysis: (a) & (b) effect of stress level and frequency in P2-0.35-0.25 mixture & (d) effect of stress level and frequency in P6-0.35-0.25 mixture.

Table 7
Main effects of different variables on 10-logarithm of fatigue life.

P2-0.35-0.25
Source Sum of squares Degree of freedom Mean square F-statistic p-value
Corrected Model 20.113 6 3.35 12.939 0.001
Intercept 0.002 1 0 0.007 0.934
Stress level 5.306 2 2.65 10.241 0.001
Frequency 1.704 2 0.85 3.288 0.052
Density 0.026 1 0.03 0.101 0.753
Stiffness 8.347 1 8.35 32.22 0.001
Error 7.513 29 0.26
Total 506.394 36
Corrected Total 27.626 35
P6-0.35-0.25
Source Sum of Squares Degree of Freedom Mean square F-statistic p-value
Corrected Model 38.89 6 6.48 38.54 0.001
Intercept 0.001 1 0 0 0.948
Stress level 7.57 2 3.78 22.5 0.001
Frequency 0.67 2 0.33 1.99 0.153
Density 0.01 1 0.01 0.04 0.848
Stiffness 11.95 1 11.95 71.04 0.001
Error 5.38 32 0.17
Total 491.03 39
Corrected Total 44.27 38
8 A.K. Chandrappa, K.P. Biligiri / Construction and Building Materials 153 (2017) 1–15

are shown in Table 3. confidence levels for frequency (p- lier if its normalized value was greater than those mentioned in the
value = 0.737) and stress levels (p-value = 0.701). In case of P6- Table corresponding to the number of data points. The summary
0.35-0.25 mixtures, the porosities of different specimens did not showed that there were no outliers in the data. S-N curves that
differ significantly with respect to stress levels (p-value = 0.473) are widely used to represent fatigue data for various materials
and with respect to frequency, the p-value was found to be 0.05, were established for pervious concrete mixtures whose summaries
where the mean difference in porosities was found to be about are shown in Tables 5 and 6. The SN curves for P2-0.35-0.25 and
only 2%. P6-0.35-0.25 mixtures are shown in Fig. 3. As seen, the relation
between the stress level and fatigue life was found to follow the
4.2. Flexural strength power function of the form S = aNb. The parameters ‘‘a” and ‘‘b”
are associated with the material property, loading mode, and spec-
Flexural strength of two mixtures along with the constant stress imen configuration as reported in [25]. Fig. 3(a) and (c) show the
that was used at different stress levels is shown in Table 4. P2- individual fatigue lives of P2-0.35-0.25 and P6-0.35-0.25 mixtures,
0.35-0.25 mix had lower flexural strength compared to P6-0.35- respectively. The geometric mean of the fatigue lives is shown in
0.25 mixture owing to higher porosity and single-sized gradation. Fig. 3(b) and (d) in order to highlight the relation between S and
The flexural strength was essential to determine the constant N parameters. The geometric mean was used instead of the arith-
stress that was later applied during fatigue test at various stress metic mean as large scatter in the fatigue life data would result
levels. in extremely high or low average fatigue lives [19,25] if the latter
was used.
4.3. Fatigue behavior The low R2 in P2 mixtures can be attributed to the fact that
porosity and/or pore sizes at the mid-span might have affected
4.3.1. Fatigue lives and S-N curves the fatigue life and hence its dispersion. However, an improved
The fatigue lives of all the specimens of P2-0.35-0.25 and P6- R2 in P6 mixture can be ascribed to the fact the mixture was dense
0.35-0.25 mixtures are shown in Tables 5 and 6. The presence of in nature with better packing that might have resulted in similar
outliers was checked using Chauvenet’s criterion on 10-logarithm kind of porosity and/or pore sizes at mid-span of the specimen. It
fatigue life scale [26]. The data point was considered to be an out- is strongly believed that larger dispersion in pervious concrete fati-

Fig. 5. Distribution fitting for fatigue life: & (b) 3-parameter Weibull and Log-normal distribution for P2-0.35-0.25 Mixture (c) & (d) 3-parameter Weibull and Log-normal
distribution for P6-0.35-0.25 mixture.
 A.K. Chandrappa, K.P. Biligiri / Construction and Building Materials 153 (2017) 1–15 9

Table 8 4.3.2. Effect of frequency and stress level on fatigue life

Distribution parameters for different mixtures. The effect of frequency and stress levels was studied using
Distributions Parameters Mixture Kaplan-Meier (K-M) analysis and Analysis of Covariance
P2-0.35-0.25 P6-0.35-0.25 (ANCOVA), which are non-parametric and parametric methods,
respectively. The Kaplan-Meier (K-M) analysis was used, as it is
3-Parameter Shape (b) 0.487 0.399
Weibull Scale (g) 12082.130 7742.290
the most widely utilized method for analyzing fatigue data without
Location/Minimum 34.780 37.290 the assumption of any distribution. Further the survival plots
life (c) obtained from K-M analysis would pictorially represent the effect
Log-normal Mean (l) 8.397 7.794 of different parameters on fatigue lives. The ANCOVA method
Standard deviation (r) 2.045 2.488
was used in order to further understand the effect of covariates
such as stiffness and density on the fatigue lives along with fixed
factors such as stress levels and frequency. The K-M survival plots
P2-0.35-0.25 and P6-0.35-0.25 mixtures are shown in Fig. 4.
gue life would be due to the local variations in density and pore
As seen from the survival probability plots of P2-0.35-0.25 mix,
structure arrangements within the specimens of same type of
the fatigue life was more dependent on the stress level, where the
mixture.
survival probabilities decreased sharply when the stress level was

Fig. 6. Cumulative distribution plots (CDP) and probability density plots (PDP) for different distributions.
10 A.K. Chandrappa, K.P. Biligiri / Construction and Building Materials 153 (2017) 1–15

increased. For example, the survival probability at 100,000 cycles which indicated no effect on fatigue life. For the P6-0.35-0.25 mix-
was 0, 7, and 22% approximately at stress levels of 0.80, 0.75, ture, p-value for the effect of stress level was found to be 0.0001 for
and 0.70, respectively as seen from Fig. 4(a). Further, with respect both the tests and frequency did not have any significant effect on
to frequency, no clear trend was observed regarding its effect on the fatigue life, and the corresponding p-values were 0.647 and
the fatigue life as the survival lines overlapped each other. The 0.839, respectively, for Log-rank and Wilcoxon tests.
K-M plots for P6-0.35-0.25 showed a more pronounced effect of At present, there are no studies that discuss the effect of fre-
stress level on the fatigue lives where the survival probabilities quency on fatigue performance of pervious concrete. However,
was too low at higher stress level (0.80) compared to lower stress previous studies on conventional concrete have shown contradic-
level (0.70). The survival plot based on the frequency showed that tory effect of frequency on fatigue life. When frequency was in
the lines overlapped at different frequencies indicating that there the range of 1.2–20 Hz, no significant effects were observed on
might be no effect of frequency on the fatigue characteristics of the fatigue life [19,27]. But, under compression fatigue, Medeiros
pervious concrete in the range of 2–10 Hz. Log-rank and Wilcoxon [28] found fatigue life was significantly influenced at frequencies
tests whose null hypothesis is that the risk/survival probabilities of 4, 1, 1/4, and 1/16 Hz while Huili and Kanliang [29] found that
are similar across all the groups that were used to check for the fatigue life reduced as the frequency increased in fiber reinforced
effect statistically. For the P2-0.35-0.25 mixture, the stress level concrete (20–60 Hz). Unfortunately, such low or high frequencies
had a p-value of 0.002 for both the tests indicating that the stress are not expected on the pavements, and hence at this point of time,
level significantly affected fatigue life. The p-value for Log-rank it can be understood that there is no effect of frequency on the fati-
and Wilcoxon test was 0.576 and 0.529, respectively for frequency, gue life of pervious concrete mixtures when subjected to the load-
ing conditions that are anticipated for the in-service pavements.
It has to be noted that as the fatigue life was a variable, the stiff-
ness and density/porosity were co-variables, and hence could be
considered as covariates. As aforementioned, ANCOVA eliminates
the effect of covariates and then checks whether there exists a sig-
nificant effect of stress level and frequency. At first, Levene’s test
was used to check the validity of the assumption of homogeneity
of variances across different groups of stress levels and frequen-
cies. The results indicated that in both cases, the assumption was
valid at 95% confidence level, where the p-values for P2-0.35-
0.25 and P6-0.35-0.25 mixtures were 0.46 and 0.34, respectively.
The main effects of different variables on 10-logarithm of fatigue
life are shown in Table 7. In both the mixtures, the stress level
and stiffness had significant effect on fatigue lives compared to
density and frequency, although there existed some effect of fre-
quency on the P2-0.35-0.25 though not significant at 95% confi-
dence level. The statistical measures: R2 and R2adj of the ANCOVA
model were found to be 0.728 and 0.672, respectively for P2-0.35-
0.25 mixture; while for P6-0.35-0.25 mixture, the ANCOVA model
had R2 and R2adj of 0.878 and 0.856, respectively. The Sidak post
hoc tests carried out to study the pairwise comparisons revealed that
there were no significant differences on the fatigue lives obtained at
0.75 and 0.80 stress levels in both the mixtures. The pairwise com-
parisons of different frequencies indicated that at all the combina-
tions; the fatigue lives did not differ significantly. A similar kind of
observation on fatigue lives was observed based on K-M analysis;
however, there was no indication of the effect of density and
stiffness.

4.3.3. Distribution fitting for fatigue life

The main purpose of fitting a distribution on the fatigue data is
to determine the survival probabilities of the material under con-
sideration and also parametrically interpret the data characteris-
tics. For pervious concrete used in the past by several other
researchers, the 2-parameter Weibull distribution was used to fit
the fatigue life and calculate survival probabilities owing to its sim-
plicity [16,17]. However, no discussions are currently available that
state the influence of the distribution parameters on the shape of
the distribution curves. Additionally, since any fatigue data pos-
sesses minimum fatigue life (before failure), it would be more
rational to check the applicability of the 3-parameter Weibull
and lognormal distributions, as both of them are widely used dis-
tributions to characterize fatigue life.
The probability plot and maximum likelihood estimation meth-
ods were used on MinitabÒ platform to perform distribution-fitting
process. The probability plots of 3-parameter Weibull and lognor-
Fig. 7. Stiffness degradation and tensile strain evolution in pervious concrete: (a) & mal distributions for P2-0.35-0.25 and P6-0.35-0.25 mixtures are
(b) P2-0.35-0.25 mixture; (c) & (d) P6-0.35-0.25 mixture. shown in Fig. 5. The Anderson-Darling (A-D) statistic was used to
 A.K. Chandrappa, K.P. Biligiri / Construction and Building Materials 153 (2017) 1–15 11

Fig. 8. Fractured face pervious concrete beam in fatigue.

check the goodness-of-fit as this statistic performs better when similar to initial curve of famous bathtub curve of reliability as
there is skewedness in the data, in which lower value of the statis- shown in Fig. 6 for different mixtures. The Figures illustrated that
tic indicates a better fit. As observed, for both the mixtures consid- the instantaneous probability of failure decreased with time or in
ered in this study, the lognormal distribution fitted well compared other words, the hazard function was decreasing with time for
to the 3-parameter Weibull distribution, which is indicated by the both the functions. Although practically this is not possible, where
lower A-D statistic, especially in case of P2-0.35-0.25 mixture. the hazard function should be increasing with time, this distinctive
The parameters of the distribution determined from maximum behavior may be indicative of the fact that these distributions may
likelihood estimation method are shown in Table 8. It can be seen not be appropriate to fit the fatigue life of pervious concrete even
that for both the mixtures, a minimum life of 35 cycles was seen. though better goodness-of-fit statistics are obtained.
The interesting part of these parameters is the shape/slope param- In contrast, this data characteristic itself may be the behavior of
eter, where the values were found to be <1.00 for both mixtures. pervious concrete under fatigue, where it was found to be very sen-
According to [30], when the shape (b) parameter is less than 1, sitive for increase in the stress levels, which resulted in very little
the failure regime of such material is characterized by decreasing useful life at higher stress levels. The probability density plots
hazard function, where such kind of failures are termed as infant depicted that the useful life of P6-0.35-0.25 mix was slightly higher
mortality. The probability density curves for these kinds of failures than the P2-0.35-0.25 mix. The shape/slope parameter values less
basically consists of hyper-exponential distribution, where it is than 1 were also obtained by Zhou et al. [17], where fatigue tests

Table 9
Parameters estimates for the models.

Model-1
Parameter Estimate Std. error 95% Confidence interval
Lower bound Upper bound
Constant 7.654 1.83 3.929 11.378
SL 12.193 2.045 16.347 8.038
Flexural stiffness 4.294*10^4 5.07*10^5 3.265*10^4 5.323*10^4
Model-2
Parameter Estimate Std. error 95% Confidence interval
Lower bound Upper bound
Constant 9.557 1.55 14.1 8.68
SL 10.028 1.81 3.78 7.84
Flexural stiffness 4.396*10^4 5.396*10^5 3.33*10^4 5.47*10^4
Porosity 0.735 0.12 0.97 0.50
12 A.K. Chandrappa, K.P. Biligiri / Construction and Building Materials 153 (2017) 1–15

precipitated down at a very high rate of reduction approaching fail-

ure, indicating that the whole process could be divided into a
three-stage process: primary, secondary, and tertiary. In addition,
it was observed from the normalized stiffness x cycles versus
cycles plot that the curve shifted from secondary to tertiary stage
when the stiffness was reduced to 73.89 and 76.29% (on average)
of the initial stiffness in P2-0.35-0.25 and P6-0.35-0.25 mixtures,
respectively. Further, stiffness reduction at complete failure was
found to be 57.75 and 61.71% for P2-0.35-0.25 and P6-0.35-0.25
mixtures, respectively. This indicated that degradation occurred
faster in pervious concrete, where approximately 25% reduction
in stiffness would drive the specimen from secondary to tertiary
stage, where the failure rate increased very sharply.
Considering the tensile strain plots, it was seen that the ini-
tial strain was about 150 mm and then increased up to about
250–300 mm before failure. Additionally, during transition from
secondary to tertiary stage, the tensile strain was found to be
244 and 249 mm (on average) for P2-0.35-0.25 and P6-0.35-
0.25 mixtures, respectively. This indicated that pervious con-
crete had higher strain carrying capacity compared to plain con-
ventional concrete, where the failure strains in flexure were
found to be 150–200 mm, exemplifying that pervious concrete
is more ductile than conventional concrete. Brake et al. [31]
observed a similar trend from the notched pervious concrete
fracture tests. The increased strain carrying capacity in pervious
concrete could be mainly attributed to the fact that the cracks
due to repeated loading mostly propagated through the aggre-
gate and cement-aggregate interface as shown in Fig. 8, and
this corroborated with the field observations made by Eller
and Izevbekhai [20].
Further, as aggregates are more homogeneous compared to
cement mortar in terms of their internal structure, this homogene-
ity could be one of the reasons for increased strain carrying capac-
ity in pervious concrete compared to conventional concrete.
Additionally, as pervious concrete depicted measurable deforma-
tion before failure that is certainly not associated with dislocation
motions usually occurring in metals [32], and did not possess
enduring stage in the tertiary region of the stiffness degradation
curve, it could be classified as quasi-brittle in nature.

4.4. Laboratory fatigue model development

*R2=indicates accuracy of model
*RMSE and Se/Sy=indicates the bias in the model The investigations on the previous sections showed that fatigue
life of the material largely depended on the stress levels and stiff-
Fig. 9. Observed and predicted values of fatigue lives: model-1; (b) model-2. ness of the specimens compared to frequency and density as also
observed from parametric and non-parametric analyses. Based on
these observations, two laboratory fatigue models were developed,
in which one model could be used when the porosity is around
were conducted on single-sized pervious concrete mixtures, but 20%; while the other model could be used when porosity is in
importantly, that study did not comment on the distribution the range of 20–30% since the model framework had porosity as
parameters at all. one of the independent variables. The model with porosity as
one of the independent terms was developed with the pooled fati-
gue test data of both P2 and P6 mixtures. Models are shown below
4.3.4. Stiffness degradation and strain evolution
in which the first model (Model-1) had a porosity range of around
In this study, the flexural stiffness was determined at the
20% while the second model (Model-2) had a range of porosity of
50th cycle from the start of the test, and the continuous degra-
20–30%.
dation of stiffness was recorded with increasing number of
cycles. Along with the stiffness degradation, the increase in ten-
sile strain at the bottom of the beam was determined until Model-1 : Log 10 N f ¼ 7:654  12:193  SL þ 4:294  104
failure. Typical stiffness degradation and increasing tensile strain
 Flexural stiffness ð3Þ
for pervious concrete beam under flexural fatigue is shown in
Fig. 7. Similar trends of curves were obtained for other Range of variables:
specimens as well. Stress level: 0.70–0.80
It was noticed that the stiffness reduced initially at a faster rate, Flexural stiffness: 7619–13937 MPa
and then the rate of reduction decreased after which the curves Porosity: 20%
 A.K. Chandrappa, K.P. Biligiri / Construction and Building Materials 153 (2017) 1–15 13

Model-2 : Log 10 Nf ¼ 9:557  10:028  SL þ 4:396  104 ably used to estimate the fatigue lives of laboratory prepared per-
  vious concrete specimens and help in quality assurance and
1
 Flexural stiffness  0:735 ð4Þ control.
/
Range of variables:
Stress level: 0.70–0.80 5. Discussion
Flexural stiffness: 7158–13937 MPa
Porosity: 0.17–0.34 The fatigue investigation in pervious concrete has seen limited
research compared to studies on other mechanical properties.
where, Nf = Fatigue life, cycles; SL = Stress level (ratio of flexural With increasing popularity of pervious concrete for its use in pave-
stress applied/flexural strength); / = porosity ments as storm water management strategy, it becomes essential
The models indicated that as the stiffness increased and stress to understand its behavior when subjected to repeated loads.
level decreased, fatigue life of pervious concrete increased. As the Although the results of this research were confined to two mix-
models were developed using regression method, the fatigue lives tures and material types, the findings on fatigue behavior of pervi-
predicted from the model had 50% reliability. The significance of ous concrete were novel, which will definitely advance the state-
the parameter estimates is shown in Table 9. As p-values of regres- of-the-art understanding of pervious concrete. The early life failure
sion estimates are not reported in nonlinear regression, the confi- observed from the probability density plots indicated that the use-
dence intervals were used to interpret the significance, where the ful life of pervious concrete under fatigue would be less, and would
regression estimate was considered to be significant, only if the further lessen when subjected to higher stress levels.
interval did not contain zero value. The plots depicting the Even though higher stress levels that are normally used in the
observed and predicted values for both the models along with laboratory fatigue tests may not occur in the field, the increased
the statistical goodness-of-fit estimates are shown in Fig. 9. Both stress levels in the laboratory would be compensated by environ-
the models had good-to-fair statistical goodness-of-fit estimates. mental conditions in the in-service pervious concrete pavements.
Additionally, the residual analyses were carried out to check the In conventional concrete, the widely accepted damage evolution
presence of heteroskedasticity, which was essential to check the theory under fatigue is that the existing micro cracks in cement
model adequacy and assumption of normality of the residuals. mortar extend and propagate to form critical sized crack, which
The plots of residual versus predicted values and the normality leads to final failure. However, in pervious concrete, it was
plots of the residuals for both models are shown in Fig. 10. There observed that cracks propagate through aggregates and cement
was no observable pattern present in the residual plots and the paste-aggregate interface, and thus, the damage evolution theory
residuals followed normal distribution validating the assumption of conventional concrete may not be truly applicable here. Addi-
of the regression analysis. Overall, these models could be comfort- tionally, pervious concrete possesses higher strain carrying ability

Fig. 10. Residual analyses: (a) & (b) residual versus predicted values for model-1 and model-2; (c) & (d) normality plots for the residuals obtained from model-1 and model-2.
14 A.K. Chandrappa, K.P. Biligiri / Construction and Building Materials 153 (2017) 1–15

than conventional concrete that may be due to the absence of 0.65 and 0.85. The residual diagnostic indicated that there
cement mortar resulting in the reduction of micro cracks. The was no heteroskedasticity in the model and that the residuals
propagation of cracks through the aggregate cautioned that using were normally distributed.
flaky or elongated aggregates would worsen the fatigue life of per-
vious concrete, and hence special attention is required while
selecting the aggregates.
Acknowledgments
6. Conclusions
The authors gratefully acknowledge the Government of India
Ministry of Human Resource Development Department of Higher
The objective of this study was to comprehensively investigate
Education for financial support vide Future of Cities research pro-
the fatigue behavior of pervious concrete paving mixtures by con-
ject grant number F. No. 4-22/2014-TS.I, dated 23 January 2014.
sidering the effect of both frequency and stress level. Further, the
The authors are also grateful to Mr. Alan Feeley, Pavetest Pty Ltd,
fatigue database so developed was used to establish predictive
Australia for helping in setting up the test module.
equations. One hundred pervious concrete specimens with fifty
specimens per mixture were used, where five specimens per mix-
ture were utilized in flexural strength determination. The test References
parameters consisted of three stress levels and three frequencies
that are anticipated in the field conditions. Based on the study find- [1] R.C. Meininger, No-fines pervious concrete for paving, Concr. Int. 10 (1988) 20–
ings, following salient conclusions were drawn: 27.
[2] B. Huang, H. Wu, X. Shu, E.G. Burdette, Laboratory evaluation of permeability
and strength of polymer-modified pervious concrete, Constr. Build. Mater. 24
1. The S-N curves indicated that the fatigue life decreased with (2009) 818–823.
increasing stress levels for a given loading frequency. The rela- [3] N. Neithalath, M.S. Sumanasooriya, O. Deo, Characterizing pore volume, sizes,
and connectivity in pervious concretes for permeability prediction, Mater.
tion between the stress level and fatigue life followed a power Charact. 61 (2010) 802–813.
function. The geometric mean was used instead of the arith- [4] Q. Dong, H. Wu, B. Huang, X. Shu, K. Wang, Investigation into laboratory
metic mean to represent S-N relationship in log-linear form abrasion test methods for pervious concrete, J. Mater. Civ. Eng. 25 (2013) 886–
892.
2. The non-parametric and parametric methods indicated that the [5] A.K. Chandrappa, K.P. Biligiri, Comprehensive Investigation of Permeability
fatigue life significantly depended on the stress levels and flex- Characteristics of Pervious Concrete: A Hydrodynamic Approach, Construction
ural stiffness compared to frequency of load repetitions, which and Building Materials, Elsevier, UK, vol. 123, pp. 627–637.
[6] A.K. Chandrappa, K.P. Biligiri, Influence of mix parameters on pore properties
was found to be insignificant when it was in the range of 2– and modulus of pervious concrete: an application of ultrasonic pulse velocity,
10 Hz. The Kaplan-Meier survival curves indicated a sudden Mater. Struct. 49 (2016) 5255–5271.
drop in the survival proportion at higher stress levels. ANCOVA [7] C. Lian, Y. Zhuge, Optimum mix design of enhanced permeable concrete – An
experimental investigation, Constr. Build. Mater. 24 (2010) 2664–2671.
was used to remove the effect of covariates (stiffness and den-
[8] M. Gesoglu, E. Guneyisi, G. Khoshnaw, S. Ipek, Abrasion and freezing–thawing
sity) and then investigate the effect of stress levels and fre- resistance of pervious concretes containing waste rubbers, Constr. Build.
quency, whose analyses were more robust compared to non- Mater. 73 (2014) 19–24.
parametric methods [9] P.D. Tennis, M.L. Leming, D.J. Akers, Pervious Concrete Pavements, EB302.02,
Portland Cement Association, Skokie, Illinois, and National Ready Mixed
3. The 3-parameter Weibull and lognormal distributions were fit- Concrete Association, Silver Spring, Maryland, USA, 2004, p. 36.
ted for the fatigue data, where lognormal distribution presented [10] X. Shu, B. Huang, H. Wu, Q. Dong, E.G. Burdette, Performance comparison of
better fit. The distribution parameters indicated that the fatigue laboratory and field produced pervious concrete mixtures, Constr. Build.
Mater. 25 (2011) 3187–3192.
failure in pervious concrete had characteristic infant mortality [11] V.R. Schaefer, J.T. Kevern, An integrated study of pervious concrete mixture
type of failure. The probability density plots of both the distri- design for wearing course applications, in: Trans project report, Institute of
butions indicated that useful life under fatigue was low. The Transportation, Paper 150, 2011.
[12] O. Deo, N. Neithalath, Compressive response of pervious concretes
main reason for this characteristic could be that the fatigue life proportioned for desired porosities, Constr. Build. Mater. 25 (2011) 4181–
was more sensitive towards stress levels 4189.
4. The stiffness degradation and tensile strain accumulation [13] A. Ibrahim, E. Mahmoud, M. Yamin, V.C. Patibandla, Experimental study on
Portland cement pervious concrete mechanical and hydrological properties,
curves indicated a three-stage process, where the primary stage Constr. Build. Mater. 50 (2013) 524–529.
was characterized by rapid accumulation of tensile strain fol- [14] J. Yang, G. Jiang, Experimental study on properties of pervious concrete
lowed by an almost constant accumulation of strain in the sec- pavement materials, Cem. Concr. Res. 33 (2003) 381–386.
[15] M.A. Pindado, A. Aguado, A. Josa, Fatigue behavior of polymer-modified porous
ondary stage. The third or tertiary stage was characterized by an
concretes, Cem. Concr. Res. 29 (1999) 1077–1083.
accelerated degradation of stiffness or accumulation of tensile [16] Y. Chen, K. Wang, X. Wang, W. Zhou, Strength, fracture and fatigue of pervious
strains followed by complete failure concrete, Constr. Build. Mater. 42 (2013) 97–104.
5. The shift from the secondary to tertiary region mostly occurred [17] J. Zhou, M. Zheng, Q. Wang, J. Yang, T. Lin, Flexural fatigue behavior of
polymer-modified pervious concrete with single sized aggregates, Constr.
when the stiffness reduction was about 70% of the initial stiff- Build. Mater. 124 (2016) 897–905.
ness, which was determined from the normalized stiffness x [18] C.E. Kesler, Effect of Speed of Testing on Flexural Fatigue Strength of Plain
cycles versus cycles plot, where the curve started curving down Concrete, Proceedings, Highway Research Board, 1953, p. 32.
[19] J.W. Galloway, K.D. Raithby, Effects of Rate of Loading on Flexural Strength and
when there was a shift. The tensile strain at the transition Fatigue Performance of Concrete TRRL Report LR 547, Transport and Road
between secondary to tertiary was found to be nearly 240 mm. Research Laboratory, Crowthorne Berkshire, 1973.
This indicated that failure strain in pervious concrete was [20] A. Eller, B.I. Izevbekhai, MnROAD Cell 64 Pervious Concrete First Year
Performance Report, Report No. MN/RC – 2007-17, Minnesota Department of
higher than that of conventional concrete. The observation of Transportation (MnDOT), 2007.
the fractured face indicated that mostly the cracks passed [21] IS 12269:2013 Indian Standard, Ordinary Portlandcement-53-grade
through the aggregates or the cement paste-aggregate interface. specification, New Delhi.
[22] ASTM, Standard Test Method for Density and Voidcontent of Hardened
Based on these observations, pervious concrete could be classi- Pervious Concrete, ASTM D7460-10, West Conshohocken, 2012.
fied as a quasi-brittle material [23] ASTM, Standard Test Method for Density and Voidcontent of Hardened
6. The laboratory fatigue models were developed based on the sig- Pervious Concrete, ASTM C1754-12, West Conshohocken, 2012.
[24] ASTM, Standard Test Method for Flexural Strength of Concrete (Using Simple
nificant parameters identified. The models were found to be
Beam with Third-Point Loading), ASTM C78/C78M-15, West Conshohocken, PA
having good-to-fair statistical parameters with R2 between 19428-2959, USA, 2010.
 A.K. Chandrappa, K.P. Biligiri / Construction and Building Materials 153 (2017) 1–15 15

[25] J.J. Xiong, R.A. Shenoi, Fatigue and Fracture Reliability Engineering, Springer, [29] Z. Huili, T. Kanliang, Properties and mechanism on flexural fatigue of
ISBN 978-0-85729-217-9, doi:10.1007/978-0-85729-218-6. polypropylene fiber reinforced concrete containing slag, J. Wuhan Univ.
[26] S.P. Singh, S.K. Kaushik, Fatigue strength of steel fibre reinforced concrete in Technol. Mater. Sci. Ed., 533–540, doi:10.1007/s11595-011-0263-8.
flexure, Cem. Concr. Compos. 25 (2003) 779–786. [30] The Use of Weibull in Defect Data Analysis, Warwick Manufacturing Group,
[27] C.E. Kesler, Effect of Speed of Testing on Flexural Fatigue Strength of Plain 2004.
Concrete, Proceedings, Highway Research Board, 1953, p. 32. [31] N. Brake, H. Allahdadi, F. Adam, Flexural strength and fracture size effects of
[28] A. Medeiros, X. Zhang, G. Ruiz, R.C. Yu, M.L. Velasco, Effect of the loading pervious concrete, Constr. Build. Mater. 113 (2016) 536–543.
frequency on the compressive fatigue behaviour of plain and fiber reinforced [32] S. Tandon K.T. Faber Fracture of brittle and quasi-brittle engineering materials
concrete, Int. J. Fatigue, 70, 342–350. Cements Research Progress American Ceramic Society Columbus, OH 209 239