International Journal of Fatigue 48 (2013) 308318

Contents lists available at SciVerse ScienceDirect

International Journal of Fatigue

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

A probabilistic fatigue model based on the initial distribution to consider

frequency effect in plain and ber reinforced concrete
Luis Saucedo a, Rena C. Yu b,, Arthur Medeiros b, Xiaoxin Zhang b, Gonzalo Ruiz b
a
Department of Materials, University of Oxford, Parks Road, OX1 3PH Oxford, UK
b
ETSI de Caminos, C. y P., Universidad de Castilla-La Mancha, Avenida Camilo Jos Cela s/n, 13071 Ciudad Real, Spain

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

Article history: The objective of this work is twofold. First, we aim to develop a new fatigue model valid for quasi-brittle
Received 8 August 2012 materials like concrete, which properties have considerably larger standard deviation than metals. Hav-
Received in revised form 19 November 2012 ing this in mind, we t the measured strength data with a three-parameter Weibull cumulative distribu-
Accepted 20 November 2012
tion function and in turn take it as the initial distribution for an asymptotic fatigue model in concrete.
Available online 29 November 2012
Second, we endeavor to take into account the observed inuence of frequency and stress ratio on the fati-
gue life in concrete, both plain and reinforced with bers. The developed model is validated against fati-
Keywords:
gue tests in compression on cubic specimens for different stress ratios and loading frequencies. All the
Weibull cumulative distribution function
Loading frequency
parameters have found physical meaning in the extensive experimental tests performed for two plain
Secondary strain rate high strength concretes and two concretes reinforced with bers. The secondary strain rate is found to
be correlational with the number of cycles to failure. Finally, a reduced test procedure is proposed for fati-
gue strength characterization.
2012 Elsevier Ltd. All rights reserved.

1. Introduction
Interest in the fatigue of concrete began more than a hundred
years ago with the development of reinforced concrete bridges.
Since then, numerous experiments have been conducted to study
the inuence of different fatigue parameters, for instance, see [1
11,13,14,17,1921,23] and the references within. These parame-
ters are either set by the fatigue test conditions, such as the mini-
mum stress rmin, the maximum stress rmax and the loading
frequency f, or determined by the material properties, for example
the static material strength rc, which can be the compressive
strength fc or the tensile strength ft, or any other critical stress de-
ned accordingly. Other parameters include the stress ratio R, de-
ned as rmin/rmax, the stress amplitude or stress range Mr,
calculated as rmax  rmin, or the stress level, dened as rmax/rc.
For metals, the stress range plays an important role, and the fa-
tigue life (the number of cycles N resisted before failure) is often
described by the Whler curve (or SN curve). For concrete, how-
ever, the inuence of the stress ratio, loading frequency and stress
level has been observed to be important [6,13,17,23]. The fatigue
equation has evolved accordingly to illustrate the role of those
parameters. For instance, Aas-Jakobsen [6] proposed to include
the effect of the stress ratio R as follows
Corresponding author.
E-mail address: rena@uclm.es (R.C. Yu).
rmax
1  1  Rb ln N 1
rc
where b is a material parameter. The same relation was conrmed
by Tepfers and Kutti [10] and Tepfers [11] for fatigue strength of
concrete in compression and in tension for splitting tests of cubes.
Even though the inuence of loading frequency (or time) has been
observed as early as 1960s by Rusch [4] and conrmed by Awad
and Hilsdorf [7], Sparks and Menzies [8] and Holmen [12] in
1970s, it was not included in the fatigue equation until Hsu [13],
Furtak [14] improved Eq. (1) by including the loading time and fre-
quency respectively. Zhang et al. [19] further improved the equa-
tion of Furtak by redening the stress ratio R when there is stress
reversal.
However, none of them considered the marked dispersion of the
static strength rc in concrete. The rst consideration of the statis-
tical distribution of concrete strength properties for fatigue tests
was by Zhao et al. [21], who considered a normal distribution as
suggested in the design codes. Recently, Przybilla et al. [22] consid-
ered the statistical feature of the characterized strength for brittle
materials and derived the primary three-parameter Weibull cumu-
lative distribution function (CDF) of fracture stress from three- and
four-point bending tests. Weibull distribution was also used to t
the fatigue life of concrete at various stress levels by Oh [17] and
to t the exural fatigue life of concrete containing nano-particles
by Li et al. [20].

0142-1123/$ - see front matter 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ijfatigue.2012.11.013
 L. Saucedo et al. / International Journal of Fatigue 48 (2013) 308318
Nomenclature
rmax maximum stress
rmin minimum stress
Mr stress range rmax  rmin
R stress ratio dened as rmin/rmax
rmax0 static equivalence of rmax
rf failure stress
rf0 intercept of the IPFC with the rf-axis
fcd dynamic compressive failure stress or the dynamic
strength under compression
fc0 static failure stress or static strength under compression
rmin0 horizontal asymptote which determines the lower
stress value
r_ 0 loading rate of the compressive test
r_ d loading rate of the fatigue test
rc critical stress, can be the compressive strength fc or the
tensile strength ft
N number of cycles to failure
Castillo and coworkers [24,25] also postulated a general proba-
bilistic model to predict the fatigue behavior for any stress level
and range based on laboratory tests for ductile materials like steel.
The nine parameters involved are dened through the physical and
compatibility considerations of the Weibull model. However, their
model does not consider the observed inuence of loading fre-
quency in concrete.
In the current work, we rst consider the entire statistical dis-
tribution given by the characterization tests and build our fatigue
model from this initial distribution. Second we take into account
the loading frequency based on the dynamic-response description
given by the Model Code 2010 [26]. Finally, from the parameters of
specic fatigue tests, e.g., stress range and stress ratio, we establish
the output Weibull CDF, which describes the fatigue life at differ-
ent levels of failure probability. The range of application of the pro-
posed model is below 10 Hz according to the experimental tests
realized between 1/16 and 4 Hz [23]. Even though the model itself
does not limit to a given range of frequency, its application beyond
10 Hz needs further experimental collaboration.
The rest of the paper is organized as follows. A fatigue model
based on an initial distribution is postulated in Section 2. The
experimental program and validation of the model are given in
Section 3. A reduced test procedure is proposed based on the devel-
oped model in Section 4. Finally we summarize the current work in
Section 5.
2. The fatigue model with an initial distribution
As mentioned before, we aim to develop a fatigue model for
concrete, taking into account the statistical distribution of the
characterized strength data and the inuence of loading frequency
and stress ratio, the following hypotheses are assumed.
 The characterized (or experimentally measured) material prop-
erty of concrete, such as the compressive or tensile strength, fol-
lows a Weibull distribution. In the current work, we focus on
the compressive strength measured from cubic specimens.
 This distribution is inuenced by the dynamic condition
through the loading frequency. In addition, the relation given
by the Model Code [26] to describe the dynamic properties of
concrete is extendable to consider the inuence of loading fre-
quency according to the experimental data of Ruiz et al. [23].
 There exists a minimum stress which is the asymptote given by
the zero probability of failure.
309
k, k scale and shape parameter of the Weibull distribution
PF probability of failure in any point of the domain rf  lnN
f loading frequency of a fatigue test
f0 reference loading frequency
a,b and c parameter that adjusts the relation between ln N and f, R
e_ secondary strain rate in a fatigue test
e_0 reference secondary strain rate calculated as r_ 0 =E
e_i intercept of the lne_  lnf  curve with the e_ -axis
a exponent that measures the amplication of dynamic
strength
c coefcient that takes into consideration of loading fre-
quency for a
g slope of the ln e_ i lnf curve
r Pearsons coefcient of correlation
b Material parameter for the fatigue equation proposed in
[6]
Given sufcient number of characterization tests carried out at
a certain reference loading rate r_0 , which is considered static,
where the dot  represents derivation with respect to time, denom-
inating the measured strength data or the failure stress at one cycle
as rf0 (f for failure, 0 for static loading), the probability of failure
(PF) corresponding to each stress level can be tted by a three-
parameter Weibull CDF as follows
  
rf  rmin0 k
PFrf0 1  exp  0 ; rf0 P rmin0 2
k
where k and k are the scale and shape parameter respectively,
whereas rmin0 is the location parameter or the threshold stress be-
low which no failure will occur, it plays the role of endurance limit.
Note that, through Eq. (2), the concept of absolute failure or damage
is replaced by the probability of failure, which ranges from 0 to 1.
We dene the distribution in Eq. (2) as the initial distribution Di,
which is a property of the material, and is determined necessarily
through experimental characterization.
2.1. The inuence of loading frequency
In order to relate the dynamic failure strength under compres-
sion fcd with its static counterpart fc0 , we start with the empirical
expression provided by the Model Code [26], which is written as
follows
 a
fc d r_ d
3
fc0 r_ 0
where r_ d and r_ 0 are the loading rate of the compressive fatigue test
and that of the compressive characterization test respectively. The
exponent a is tted as 0.014 in the Model Code [26], where the ef-
fect of loading frequency is not taken into account.
The loading rate r_ d in Eq. (3) in each cycle can be roughly re-
lated to the loading frequency f and the stress range Dr through
r_ d 2f Dr 4
Meanwhile an expression for the exponent a that takes into consid-
eration the inuence of loading frequency is obtained by tting
experimental data of Ruiz et al. [23] for frequencies below 10 Hz,
the function is
a 0:014 expcf  5
where the parameter c needs to be determined by tting experi-
mental data for different loading frequencies. The coefcient
310 L. Saucedo et al. / International Journal of Fatigue 48 (2013) 308318

0.014 for static loading conditions is recovered for a zero frequency. [14] and it is the function that best ts the experimental results,
As a result, the inuence of frequency in a fatigue test is manifest on which we will see in the following section.
both the loading rate through Eq. (4) and the exponent a through For a fatigue test, rmax, R and ln N are known parameters, mean-
Eq. (5). In addition, Eq. (3) allows us to shift the initial distribution while, the static counterpart of rmax, denominated as rmax0 , corre-
Di for fc0 to the distribution of fcd in dynamic conditions. sponds to the value given in the characterization tests. According
to Eq. (3), they are related through the following dynamic equation
2.2. Failure curves of iso-probability  a  a
r_ r_ 0
rmax0 rmax _ 0 rmax 11
rd 2f Dr
In this Section, we explore all the conditions of the failure
curves in order to obtain the specic expression By plugging in the value of rmax0 for rf to Eq. (9), we easily obtain rf0
rf f ; rmax ; rmin0 ; rf0 ; R; N, where rmin0 and rf0 are the static material as follows
properties, whereas f, rmax and R are set by specic fatigues tests.
On the one hand, each curve represents one probability of failure rf0 rmin0 rmax0  rmin0 Na1R 12
and intercepts the rf-axis at rf0 , the PF of which is determined Introducing the value of rf0 into Eq. (2), we arrive at the general
by the distribution Di dened by Eq. (2), see Fig. 1. On the other expression for the cumulative probability of failure for any point
hand, there are three limit conditions that all the failure curves of the fatigue test
of iso-probability should comply with. (  )
rmax0  rmin0 k
lim rf rmin0 6 PFN; rmax ; f ; R 1  exp  13
N!1 kN a1R
limrf rf0 7
R!1 where N is considered as the main variable, whereas the rest are
limrf rf0 8 secondary ones, which are parameters set by a given fatigue test.
N!1
Notice that, due to the shifting of the CDF from Eqs. (2)(13), the
It needs to be emphasized again that rmin0 is the threshold stress be- shape parameter k is preserved, but the scale parameter is now re-
low which no fatigue failure will occur, whereas the rf0 is the static lated to loading frequency, stress ratio and the number of cycles
strength (when stress ratio equals one or when the failure occurs suffered. Insert Eq. (11) and Eqs. (10)(13), the following explicit
after one cycle). CDF is obtained
In Fig. 1, we give a schematic plot for the failure curves of three 8 2  a 3k 9
iso-probabilities, 0.05, 0.5 and 1.0. On the left of the rf-axis is the >
< rmax 2fr_D0r  rmin0 > =
6 7
initial distribution Di determined from experimental characteriza- PFN; rmax ; f ; R 1  exp 4 5 14
>
: kN bc ln1f 1R >
;
tion, on the right is the output distribution Do, which represents
the fatigue equation we are looking for and is going to be dened
later on. Having in mind the limiting conditions given in Eqs. (6) This is the output distribution Do shown in Fig. 1. It can be observed
(8), the following expression for each failure curve is proposed that, in Fig. 1, for a fatigue test performed at a given level of rmax0 ,
the probability of failure increases with the number of cycles en-
rf rmin0 rf0  rmin0 Na1R 9 dured. The number of cycles N resisted for a given PF under given
loading conditions is also easily derived
where the parameter a is related to the loading frequency as follows
" p#a1R
1

a b c ln1 f 10 k k  ln1  PF
NPF; rmax ; R; f 15
rmax0  rmin0
where b, c and c in Eq. 5 are the only parameters which need to be
tted with the experimental data of fatigue tests. The logarithmic In addition, by writing the stress range in terms of the stress ratio
function in Eq. (10) has been inspired by earlier work of Furtak and the maximum stress as Dr = (1  R)rmax in Eq. (11), we can also

Fig. 1. Failure curves of iso-probability, where Di is the initial CDF, determined from characterization tests and tted according to Eq. (2), whereas Do is the output CDF tted
by the fatigues tests.
 L. Saucedo et al. / International Journal of Fatigue 48 (2013) 308318 311

1 predict the maximum stress for a given PF and a designed fatigue

life at a given stress ratio and loading frequency as follows
" p#11 a  a
k k  ln1  PF 2f 1  R 1a
rmax PF; N; R; f rmin0 16
0.75
Na1R r_ 0
Probability of Failure

0.0625 Hz where a is given by Eq. (5).

0.25 Hz
0.5
1 Hz 2.3. Model sensibility with respect to frequency and stress ratio
4 Hz
6 Hz The variation of the probability distribution Do given by Eq. (13)
8 Hz is demonstrated in Fig. 2 for six different frequencies. It can be ob-
0.25
served that for the same failure probability, the specimen resists
more cycles at a higher frequency. This is attributed to the dynamic
behavior of concrete, which results an increase of the dynamic
exponent a, see Zhang et al. [27]. It should be noted that this inu-
0
ence is not linear, for example, the distance between 1 Hz and 4 Hz
0 5 10 15 20
is smaller than that between 6 Hz and 8 Hz.
ln [N]
The inuence of stress ratio is shown in Fig. 3 for the three val-
Fig. 2. Variation of the Weibull distribution function Do with respect to loading ues of R. 0.1, 0.5 and 0.9. Note that the shape of the failure curves of
frequency. iso-probability changes when the stress ratio R varies. A stress ratio

200 200
PF=0.01 PF=0.01

PF=0.5 PF=0.5

PF=0.99 PF=0.99
150 150

R=0.1 R=0.5
f (MPa)

f (MPa)

100 100

50 50

0 0
0 25 50 75 100 0 25 50 75 100
ln [N] ln [N]
200 PF=0.01

PF=0.5

PF=0.99
150
Stress of Failure (MPa)

R=0.9

100

50

0
0 25 50 75 100
ln [N]

Fig. 3. The inuence of stress ratio on the failure curves of iso-probability, R = 0.1, 0.5 and 0.9.
312 L. Saucedo et al. / International Journal of Fatigue 48 (2013) 308318

of 1, i.e., a constant load will result a horizontal failure line, which Table 2
is the limit case of zero fatigue failure. Fitting parameters for the initial distribution Di given in Fig. 4 for concrete C1 and C2,
and the corresponding Pearsons coefcient of correlation r.

Material k (MPa) k rmin0 (MPa) r

3. Experimental program and model validation
C1 94.7 12.4 3.1 0.98
C2 76.1 19.8 3.1 0.98
The experimental program was designed to look into the inu-
ence of the stress ratio and loading frequency on the fatigue
strength of concrete. Two groups of tests on two different types
of concrete C1 and C2 were carried out for cubic specimens.
Twenty compressive tests on cubes (80 mm in edge length) for Table 3
C1, loaded at a rate r_ 0 of 0.25 MPa/s, and six compressive tests Compressive fatigue tests on cubes at the stress ratio 0.3 and 0.1 for concrete C1, with
for cubes (100 mm in edge length) for C2, loaded at 0.2 MPa/s, the maximum stress of 90 MPa.
were carried out for characterization purpose. The measured
R Number of cycles resisted for f = 4 Hz, rmax = 90 MPa
strength variation are given in Table 1. The Weibull distribution
0.3 150 2927 2149 667 7600 75,378 7839 20,426
Di for both materials are plotted in Fig. 4, the corresponding
38 73 17,172 11,863 9218 7288 2798
parameters, tted through Mathematica, are listed in Table 2.
0.1 2265 4276 2352 222 46 125 731 1106
Additionally given in Table 2 is the Pearsons coefcient of correla-
7153 3961 1231 1753 302 858 5988
tion r [28]. Note that for both type of concretes C1 and C2, even
though the scale and shape parameters k and k are quite different,
the minimum stress rmin0 is 3.1 MPa, which plays the role of the
endurance limit. It needs to be pointed out that, we have tted
the experimental characterization data with the three-parameter 3.1. Tests on concrete C1 for two different stress ratios
Weibull distribution, and rmin0 is one of these three parameters.
Note that rmin0 is only 34% of the material compressive strength. For the purpose of studying the inuence of stress ratio, two
In the case of traction fatigue, we have not carried out a signicant series of compressive fatigue tests on cubic specimens with an
number of tests yet so as to compute any endurance limit. Never- edge length of 80 mm, fteen each, were carried out at a loading
theless, if the obtained rmin0 is quite small compared to the corre- frequency of 4 Hz, a maximum stress rmax of 90 MPa. The cycles re-
sponding tensile strength, it could be considered as zero. This could sisted for the two stress ratios 0.3 and 0.1 are given in Table 3. Fit-
have led to the conclusion in the literature that concrete materials ting these experimental data with Eq. (13), the values for a and c
have no endurance limit. shown in Table 4, where the correlation coefcients for both stress
Next, the model is validated against concrete reinforced with ratios are also listed. The agreement, as can be seen from Fig. 5, is
steel bers CF1 and polypropylene bers CF2. Finally, after estab- remarkable.
lishing the one-to-one relation between the secondary strain rate After inserting the obtained parameters a and c to Eq. (2), we
and the fatigue life from observed experimental trend, the fatigue arrive at the following fatigue equation, which predicts the prob-
Eq. (14) is also expressed in terms of the secondary strain rate. ability of failure for any compressive fatigue test carried out at
4 Hz, with a stress ratio R and maximum stress rmax after N
cycles.
Table 1
Compressive tests on cubic specimens performed to characterize the two concretes PFN; rmax ; R 1
C1 and C2, at loading rate r_ 0 0:25 MPa=s and 0.20 MPa/s respectively. 8 " #12:40 9
< r0:9634 1 =
Concrete fc (MPa) max 0:751R
 exp  0:0366
 N
C1 85 89 108 104 85 103 102 91 90 102
: 107:41  R 30:5 ;
97 84 102 88 100 90 100 99 93 98
C2 73 82 79 83 75 77 17

1 1

0.75 0.75
Probability of Failure

Probability of Failure

0.5 0.5

0.25 0.25

Fitted model Fitted model

Experimental data
Experimental data
0 0
60 75 90 105 120 60 75 90 105 120
fc (MPa) fc (MPa)

Fig. 4. Initial distribution of the compressive strength tted with the experimental measurements on cubes for concrete C1 (left) and C2 (right).
 L. Saucedo et al. / International Journal of Fatigue 48 (2013) 308318 313

Table 4
Fitting parameters a, c and intermediate parameters rmax0 , a for the fatigue tests
loaded at a stress ratio of 0.3 and 0.1 at 4 Hz for concrete C1 with a maximum stress of
90 MPa, and a loading frequency of 4 Hz, and the corresponding Pearsons coefcient
of correlation r.
Table 6
Fitting parameters b, c, c, intermediate parameters rmax0 , a for the fatigue tests carried
out under four different frequencies on cubic specimens for concrete C2, with a
maximum stress of 66.4 MPa, a stress ratio of 0.3, and the corresponding Pearsons
coefcient of correlation r.

R c a rmax0 (MPa) a = 0.014 exp[cf] r f (Hz) c b c a [0.014ecf] rmax0 MPa r

0.3 0.24 0.06 68.2 0.0366 0.98 4 0.24 0.061 0.0105 0.0366 50.8 0.96
0.1 67.6 0.94 1 0.0178 59.8 0.99
0.25 0.0149 62.1 0.93
0.0625 0.0142 63.5 0.97

3.2. Tests on concrete C2 under four different loading frequencies

In order to pinpoint the inuence of loading frequency, four ser-
ies of compressive fatigue tests, on cubic specimens with an edge
length of 100 mm, were carried out at four different frequencies
for concrete C2. The cycles resisted at each frequency are shown
in Table 5. Fitting these data curves with Eq. (2), the adjusting
parameters b, c and c listed in Table 6, the corresponding coef-
cients of correlation are also listed. Again the agreement, see
Fig. 6, is noteworthy. As a result, all the parameters in Eqs. (14)
(16) are determined, in turn, the fatigue life or probability of failure
for concrete C2 is characterized for any given maximum stress,
stress ratio or any loading frequency (below 10 Hz). Notice that
the parameter c for the two concretes studied remains the same.
3.3. Validation against concrete reinforced with bers
In order to further validate the fatigue equation presented in
Eq. (14), we also recur to the experimental results presented by
Ruiz et al. [23] for concrete reinforced with steel bers, denoted
as CF1, and for concrete reinforced with polypropylene bers,
denominated as CF2. The parameters for the initial distribution
Di and the fatigue equation Eq. (14), all tted through the soft-
ware Mathematica, together with the Peasons coefcient r, are gi-
ven in Table 7. We also demonstrate separately the output
distributions for concrete reinforced with steel bers CF1 at four
loading frequencies together with the experimental data in Fig. 7,
the correlation coefcients, obtained from tting through Math-
ematica, are greater than 0.90 for all the cases considered. The
distribution curves, or the curves of failure probability, at four
loading frequencies are plotted together in Fig. 8(left) for C2
and CF1, in Fig. 8(right) for C2 and CF2. Notice that, due to the
effect of added steel bers, the inuence of the loading frequency
on concrete CF1 is less pronounced as the frequency increases.
This is consistent with the known fact that the fatigue behavior
of steel is not inuenced by loading frequency. In contrast, adding
polypropylene bers also alters the inuence of loading frequency
on curves of failure probability, such a tendency is not observed
in Fig. 8(right).

1 1

0.75 0.75
4 Hz 4 Hz
Probability of Failure
Probability of Failure

0.5 0.5

0.25 0.25

Fitted model Fitted model

Experimental data Experimental data

0 0
0 5 10 15 20 0 5 10 15 20
ln [N] ln [N]

Fig. 5. Distribution Do given by Eq. (13), tted through the fatigue tests with R = 0.1 (left) and R = 0.3 (right) for concrete C1.

Table 5
Results of the fatigue tests loaded at a stress ratio of 0.3, maximum stress of 66.4 MPa, and four different frequencies on cubic specimens for concrete C2.

f (Hz) Number of cycles for R = 0.3, rmax = 66.4 MPa

4 8411 821 2485 1660 13,020 22,570 9521
4192 170,256 1578 1222 133 7038
1 282 23 759 1351 85 157 479
368 833 1571
0.25 98 1242 535 157 18 30 219
650 122 400
0.0625 339 473 102 234 11 142 76
275 329 38
314 L. Saucedo et al. / International Journal of Fatigue 48 (2013) 308318

1 1

0.75 0.75
0.0625 Hz 0.25 Hz

Probability of Failure
Probability of Failure

0.5 0.5

0.25 0.25
Fitted model Fitted model

Experimental data Experimental data

0 0
0 5 10 15 20 0 5 10 15 20
ln [N] ln [N]

1 1

0.75 0.75
1 Hz 4 Hz
Probability of Failure

Probability of Failure

0.5 0.5

0.25 0.25
Fitted model Fitted model
Experimental data Experimental data
0 0
0 5 10 15 20 0 5 10 15 20
ln [N] ln [N]

Fig. 6. Model and experimental data for concrete C2 for fatigue loading at four different frequencies.

Table 7
Summary of the model parameters for the four materials considered.

b c g f0 (Hz) k (MPa) k rmin0 MPa c E (GPa)

C1 94.7 12.4 3.1 0.240 35

C2 0.061 0.0105 0.081 0.0016 76.1 19.8 3.1 34
CF1 0.049 0.0066 0.089 0.0015 76.1 31.0 12.0 0 38
CF2 0.052 0.0035 0.086 0.0019 70.2 15.8 4.8 0.086 41

3.4. Secondary strain rate versus fatigue life
In a fatigue test for concrete, when the deformation at the upper
stress level rmax is plotted as the function of the number of cycles
under testing, the resulted curve is known as the cyclic creep curve.
This curve has a middle part, denominated as the secondary branch,
as illustrated in Fig. 9, in which the increase of deformation per
load cycle is constant. The slope of this secondary branch is called
the secondary strain rate or the secondary creep rate, denominated
as e_ sec , of a specic fatigue test [15,16]. In the current work, with-
out causing confusion, we simply denote it as e_ . According to the
work of Hordijk et al. [18] and of Cornelissen [15], there appears
to be a strong relation between the secondary strain rate and the
number of cycles to failure, with diminishing secondary strain rate,
the fatigue life increases. In general, a prediction of the number of
cycles to failure based on this secondary strain rate is more accu-
rate than that based on stresses in an S-N diagram. In this section,
we attempt to nd the relation between the secondary strain rate
with the parameters in the fatigue equation Eq. (14).
The experimental results for the above four materials, plain
concrete C1 and C2, concrete reinforced with bers CF1 and CF2,
are plotted in the space of secondary strain rate and fatigue life
(ln e_ and ln N) in Fig. 10. Note that, there are 30 fatigue tests for
C1 carried out for two stress ratios and one loading frequency, 43
tests for C2, 40 for CF1 and 40 for CF2, performed for one stress ra-
tio and four different loading frequencies. It can be observed that,
there is strong correlation between the secondary strain rate and
the number of cycles resisted in the loglog scale. Denominating
the intercept of a straight line in Fig. 10 for loading frequency f
as ln e_ i , the equation for this straight line is written as
 L. Saucedo et al. / International Journal of Fatigue 48 (2013) 308318 315

1 1

0.75 0.75
0.0625 Hz 0.25 Hz

Probability of Failure
Probability of Failure

0.5 0.5

0.25 0.25
Fitted model Fitted model
Experimental data Experimental data

0 0
0 5 10 15 20 0 5 10 15 20
ln [N] ln [N]

1 1

0.75 0.75
1 Hz 4 Hz
Probability of Failure

Probability of Failure

0.5 0.5

0.25 0.25
Fitted model Fitted model
Experimental data Experimental data

0 0
0 5 10 15 20 0 5 10 15 20
ln [N] ln [N]

Fig. 7. Model and experimental data for concrete reinforced with steel bers CF1 for fatigue loading at four different frequencies.

1 1

0.75 0.75
Probability of Failure

Probability of Failure

0.5 0.5

0.25 0.25
Without fibers
Without fibers
Polypropylene fibers
Steel fibers
0 0
0 5 10 15 20 0 5 10 15 20
ln [N] ln [N]

Fig. 8. Inuence of the frequency on concrete reinforced with steel bers CF1 (left) and polypropylene bers CF2 (right) compared to that on plain concrete C2.
316 L. Saucedo et al. / International Journal of Fatigue 48 (2013) 308318

where g is the slope, whereas e_0 is the reference secondary strain

rate, corresponding to a fatigue test carried out at the reference
loading frequency f0. The frequency f0 is an upper limit, below
which, the test is considered static. This means that the following
limit conditions should be satised for the secondary strain rate
Secondary branch of a fatigue test.
Strain

lim ln e_ ln e_0 20
R!1

lim ln e_ ln e_0 21
N!1

In analog to the limiting conditions of the failure stress in Eqs. (6)

(8), taking into account the fact that ln e_ also depends on the stress
ratio R and the frequency f through Eqs. (9) and (10), we approxi-
mate the slope d as follows
Number of cycles
d d1 b c ln1 f 1  R 22
Fig. 9. The concept of secondary strain rate or secondary creep rate in a cyclic creep
curve for a fatigue test.
Fitting Eq. (18) with the experimental results shown in Fig. 10, d1 is
obtained as 1. Inserting Eqs. (19) and (22) to Eq. (18), calculating the
reference secondary strain rate according to linear elasticity, we
ln e_ ln e_i d ln N 18 have
  
where d is the slope. r_0 f
ln e_ ln 1  g ln  f1 b c ln1 f 1  Rg ln N 23
Meanwhile, the relation between ln e_ i and the loading fre- E f0
quency can also be tted as follows
  The solid lines in Fig. 10 are predictions according to the above
ln e_i f equation. The agreement with the experimental data is extraordi-
1  g ln 19
ln e_0 f0 nary. The signicance of Eq. (23) lies in the fact that, for a specic

-3 -3
10 10 Ana. 4 Hz
Ana. R=0.1 Ana. 1 Hz
Ana. 0.25 Hz
10
-4 Ana. R=0.3 -4
10 Ana. 0.0625 Hz
Test R=0.3 Test 4 Hz
-5 Test R=0.1 -5 Test 1 Hz
10 10
(s-1)

(s-1)

Test 0.25 Hz
. . Test 0.0625 Hz
-6 -6
10 10

-7 -7
10 10

-8 -8
10 10
1 10 100 1000 10000 100000 1 10 100 1000 10000 100000
Cycles (N) Cycles (N)

Ana. 4 Hz -3 Ana. 4 Hz
10-3 Ana. 1 Hz 10
Ana. 1 Hz
Ana. 0.25 Hz
Ana. 0.25 Hz
Ana. 0.0625 Hz -4
10-4 10 Ana. 0.0625 Hz
Test 4 Hz
Test 4 Hz
Test 1 Hz
Test 1 Hz
Test 0.25 Hz -5
10-5 10 Test 0.25 Hz
(s-1)

Test 0.0625 Hz
(s-1)

Test 0.0625 Hz
. .
-6
10-6 10

-7
10-7 10

-8
10-8 10
1 10 100 1000 10000 100000 1 10 100 1000 10000 100000
Cycles (N) Cycles (N)

Fig. 10. Secondary strain rate plotted against the number of cycles resisted for material C1 (top left) at two stress ratios, and for C2 (top right), CF1 (bottom left), CF2 (bottom
right) at four loading frequencies.
 L. Saucedo et al. / International Journal of Fatigue 48 (2013) 308318 317

fatigue test, once the secondary strain rate is known, the number of concrete reinforced with steel bers and a value in between for
cycles to failure ceases to be probabilistic, since there is a one-to- the concrete reinforced with polypropylene bers.
one correspondence between e_ and the fatigue life N.
Moreover, we slightly modify Eq. (23) to obtain the number of
4. Applications to design of reduced fatigue tests
cycles to failure in terms of the secondary strain rate as a function
Ne_ , and insert it to Eq. (14). The probability of failure related to
Since there is a direct correlation between the probability of
the secondary strain rate is given in the following form
failure (or the number of cycles resisted) and the conditions of
loading in a fatigue test, such as stress ratio R, loading frequency
8 2  a 3k 9 f, and the tting parameters of the corresponding characterization
>
< rmax 2fr_D0r  rmin0 > = test, k, k and rmin0 through Eqs. (11)(15), the probability of failure
6 7
PFe_ ; rmax ; f ; R 1  exp 4 5 24 can be calculated for several stages of loading with different load-
>
: kNe_ >
;
ing frequencies and stress ratios.
An example of seven stages of loading is given in Table 8 and
Comparing Eqs. (14) and (24), we can conclude that for a fatigue
Fig. 11. The test is designed to start with a null probability of fail-
test carried out at a giving load condition maximum stress level,
ure (step 1), the probability of failure at the end of the step 1 is cal-
stress ratio and loading frequency , both the number of the cycles
culated according to Eq. (13). Next the loading condition is
to failure and the secondary strain rate are probabilistic. It cannot
changed to step 2. Evaluating Eq. (15) (considering the nal value
be overemphasized that, Eq. (23) provides the possibility to deter-
of PF in the step 1), we get the equivalent number of cycles for the
mine the fatigue life N without the need of actually exhausting all
beginning of the step 2, denominated as N1. Considering the load-
the cycles.
ing cycles DN2 at step 2, we evaluate the Eq. (13) with
Finally, we summarize all the parameters and the elastic mod-
N2 = N1 + DN2 to get the nal value of PF on this step. Repeating this
ulus for the four materials in Table 7. It needs to be mentioned that,
procedure for all the steps of fatigue loading in Table 8, fatigue fail-
since fatigue tests for C1 were performed at one loading frequency
ure is achieved with a reduced number of cycles (less than 20,000
of 4 Hz, it is not possible to obtain the reference loading frequency
cycles), see Fig. 11.
f0, neither is it feasible for a separate tting of parameters b and c.
Instead, a unique value for the parameter a of 0.06 is obtained. In
addition, the constant c in the expression for the dynamic expo- 5. Conclusions
nent a, which is extended to include the inuence of loading fre-
quency, is tted as 0.24 for the two plain concretes, zero for the By taking into consideration the dynamic properties of concrete,
tting the results of material characterization tests with a Weibull
distribution and assuming it as the initial distribution, which can
Table 8 be shifted along the failure axis, we have developed a fatigue mod-
An example of a reduced procedure for fatigue characterization with seven stages of
el which is capable of dealing with different frequencies and stress
loading.
ratios for two plain concretes and two concretes reinforced with -
Step rmax rmin f (Hz) r_ d MPa=s DN PFi PFf bers. Since only two adjusting parameters are needed, the rest are
(MPa) (MPa)
related with material properties and test conditions, reduced fati-
1 65 10 1 110 2000 0 0.2137 gue test procedure can be designed for fatigue life prediction.
2 70 10 0.001 0.12 350 0.2137 0.4988
The model is validated against a total of 153 fatigue tests for
3 110 5 9 1890 3000 0.4988 0.5008
4 80 15 2 260 500 0.5008 0.6886 two plain high strength concrete and two concretes reinforced
5 100 25 7 1050 4500 0.6886 0.7086 with steel or polypropylene bers, performed at two different
6 85 20 0.6 78 250 0.7086 0.8499 stress ratios and four different loading frequencies. In addition,
7 90 5 5 850 9000 0.8499 1 we have shown the failure is probabilistic in terms of the number
of cycles N or the secondary strain rate e_ , but there is a one-to-one
relation between N and e_ . In practical terms, this provides the pos-
sibility of determining the fatigue life N without actually exhaust-
1
ing all the cycles.
7
Finally, we would like to emphasize that, even though we have
not performed either characterization tests (with more than ten
6 specimens) or fatigue tests on normal strength concrete (NSC),
0.75 5 we do expect the proposed probabilistic model developed to work
Probability of Failure

for NSC, since the model only concerns with parameters which are
4 common to both NSC and HSC.
3
0.5
Acknowledgements

2
Funding from the Subdireccin General de Proyectos de Investiga-
cin, Spain, under Grant No. MAT 2009-12023, and from INCRECYT
0.25
is gratefully acknowledged.
Damage

Change of loading step

1 References
0
0 5000 10000 15000 20000 [1] Graf O, Brenner E. Experiments for investigating the resistance of concrete
under often repeated compression loads (in German). Bulletin No. 76 and 83,
Cycles (N) Deutscher Ausschuss fr Eisenbeton; 1934.
[2] Kesler CE. Effect of speed of testing on exural fatigue strength of plain
Fig. 11. Damage accumulation for the different steps of loading: a reduced concrete. Proc High Res Board 1953;32:2518.
procedure. [3] Nordby GM. Fatigue of concrete a review of research. ACI J 1958:191220.
318 L. Saucedo et al. / International Journal of Fatigue 48 (2013) 308318
[4] Rusch H. Researches toward a general exural theory for structural concrete.
ACI J 1960;57(1):128.
[5] Mudock JW. A critical review of research on fatigue of plain concrete. Bulletin
No. 475, Engineering Experiment Station, University of Illinois, Urbana 25;
1965.
[6] Aas-jakobsen K. Fatigue of concrete beams and columns. Bulletin No. 70-1,
NTH Institutt for Betongkonstruksjoner, Trondheim, September 1970.
[7] Awad ME, Hilsdorf HK. Strength and deformation characteristics of plain
concrete subjected to high repeated and sustained loads. In: Civil engineering
studies. Structural research series, department of civil engineering, vol.
372. Urbana: University of Illinois; 1971. p. 266.
[8] Sparks PR, Menzies JB. The effect of rate of loading upon the static and fatigue
strengths of plain concrete in compression. Mag Concrete Res (London)
1973;25(83):7380.
[9] ACI Committee 215. Considerations for design of concrete structures subjected
to fatigue loading. ACI J 1974;97121.
[10] Tepfers R, Kutti T. Fatigue strength of plain, ordinary and lightweight concrete.
ACI J 1979;76(5):63552.
[11] Tepfers R. Tensile fatigue strength of plain concrete. ACI J 1979;76(8):
91933.
[12] Holmen JO. Fatigue of concrete by constant and variable amplitude loading.
PhD thesis, University of Trondheim, Norway; 1979.
[13] Hsu TTC. Fatigue of plain concrete. ACI J 1981;78:292304.
[14] Furtak K. Ein verfahren zur berechnung der betonfestigkeit unter
schwellenden belastungen. Cement Concrete Res 1984;14(6):85565.
[15] Cornelissen HAW. Fatigue failure of concrete in tension. HERON
1984;29(4):168.
[16] CEB Comite Euro-International Du Beton. Fatigue of concrete structures: state
of the art report Bulletin DInformation 1988;188:1300.
[17] Oh BH. Fatigue-life distributions of concrete for various stress levels. ACI Mater
J 1991;88(2):1228.
[18] Hordijk DA, Wolsink GM, de Vries J. Fracture and fatigue behavior of a high
strength limestone concrete as compared to gravel concrete. HERON
1995;40(2):12546.
[19] Zhang B, Philips DV, Wu K. Effect of loading frequency and stress reversal of
fatigue life of plain concrete. Mag Concrete Res 1996;48(177):36175.
[20] Li H, Zhang MH, Ou JP. Flexural fatigue performance of concrete containing
nano-particles for pavement. Int J Fatigue 2007;29:1291301.
[21] Zhao DF, Chang QY, Yang JH, Song YP. A new model for fatigue life distribution
of concrete. Key Eng Mater 2007;348349:2014.
[22] Przybilla C, Fernndez-Cantelli A, Castillo E. Deriving the primary cumulative
distributive function of fracture stress for brittle materials from 3- and 4-point
bending tests. J Euro Ceram Soc 2011;31:45160.
[23] Ruiz G, Medeiros A, Zhang XX. Experimental study of loading frequency effect
on compressive fatigue behavior of plain and ber reinforced concretes. Anales
de Mecnica de la Fractura 2011;29:53540.
[24] Castillo E, Fernndez-Cantelli A, Ruiz-Ripoll ML. A general model for fatigue
damage due to any stress history. Int J Fatigue 2008;30:15064.
[25] Castillo E, Fernndez Cantelli A, Koller R, Ruiz-Ripoll ML, Garca A. A statistical
fatigue model covering the tension and compression Wler elds. Probab Eng
Mech 2009;24:199209.
[26] FIB-Fdration Internationale du Bton. Model Code 2010 (rst complete
draft), bulletin 55, vol. 1(5) Materials. Lausanne, 2010.
[27] Zhang XX, Ruiz G, Yu RC, Poveda E, Porras R. Rate effect on the mechanical
properties of eight types of high-strength concrete and comparison with FIB
MC2010. Construct Build Mater 2012;30:3018.
[28] Rodgers JL, Nicewander WA. Thirteen ways to look at the correlation
coefcient. The Am Statist 1988;42(1):5966.