Construction and Building Materials 65 (2014) 321–329

Contents lists available at ScienceDirect

Construction and Building Materials

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

Flexural creep of steel fiber reinforced concrete in the cracked state

E. García-Taengua a,⇑, S. Arango b, J.R. Martí-Vargas b, P. Serna b
a
Queen’s University of Belfast, David Keir Bldg., Stranmillis Rd., BT9 5AG Belfast, UK
b
ICITECH – Institute of Concrete Science and Technology, Universitat Politècnica de València, 4G Bldg., Camí de Vera s/n, 46022 Valencia, Spain

h i g h l i g h t s

 Fiber slenderness and content modify effect of load ratio on SFRC flexural creep.
 Fiber length does not have a significant effect on SFRC flexural creep.
 Increasing fiber slenderness leads to reduced creep strains.
 Creep control, fibers: no high amounts required, slender fibers is the best choice.

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

Article history: This paper aims at assessing the effect of a number of variables on flexural creep of steel fiber reinforced
Received 10 December 2013 concrete in its cracked state, namely: fiber geometry (slenderness and length), fiber content, concrete com-
Received in revised form 28 April 2014 pressive strength, maximum aggregate size, and flexural load. Notched prismatic specimens have been
Accepted 30 April 2014
subjected to sustained flexural loads for 90 days following a test setup and methodology developed by
the authors. Several experimental outputs have been measured: initial crack width, crack width at 90 days,
and crack opening rates and creep coefficients at 14, 30, and 90 days. Multiple linear regression has been
Keywords:
applied to relate these creep parameters to the variables considered. Semi-empirical equations have been
Bending
Concrete
obtained for these parameters. Statistical inference has been applied to identify the variables that have a
Creep statistically significant effect on SFRC flexural creep response. Fiber slenderness and fiber content have
Cracked state been found to significantly modify the effect that load ratio has on flexural creep response of SFRC.
Steel fiber Ó 2014 Elsevier Ltd. All rights reserved.
Test

1. Introduction Steel Fiber Reinforced Concrete (SFRC hereafter) members are

The evolution of strains and crack openings through time is fun-
damental for the durability of concrete structures. Time-dependent
phenomena such as shrinkage and creep must be taken into
account besides instantaneous strains and cracking [1,2].
Creep refers to the tendency of materials to develop increasing
strains through time when they are subjected to a sustained load.
As a result, deflection or elongation values tend to increase through
time in relation to the initial strain, i.e. right after the load is
applied. Codes for structural concrete consider compressive creep
of concrete within the usual ranges in service conditions. On the
contrary, tensile creep of either concrete or reinforcing bars is
not usually considered. However, in the case of concrete structures,
their long-term performance is basically affected by the behavior
of cracked concrete [3].
⇑ Corresponding author.
E-mail addresses: e.garcia-taengua@qub.ac.uk (E. García-Taengua), samo_59@
hotmail.com (S. Arango), jrmarti@cst.upv.es (J.R. Martí-Vargas), pserna@cst.upv.es
(P. Serna).
designed in most applications to take advantage of SFRC differen-
tial features with respect to conventional concrete: when SFRC is
brought to perform in the cracked state, cracks are under control
and residual strength provides the structural member with further
load-bearing capacity. There is no reason to expect differences
between SFRC and conventional concrete regarding compressive
creep. Any difference in terms of flexural creep behavior between
SFRC and conventional concrete is related to the possibility of
creep phenomena in the cracked zone of the section. Therefore
the relationship between tensile creep and flexural creep of SFRC
has attracted attention in recent studies [4]. Tensile creep of SFRC
has been studied by some authors [5], but is not possible to easily
extend their conclusions to flexural creep behavior.
Flexural creep of SFRC in the cracked state and the role that dif-
ferent factors play in creep behavior are quite understudied topics
within the general field of SFRC mechanical properties. There are
relatively few publications directly related to flexural creep behav-
ior of pre-cracked SFRC beams, and therefore every contribution is
a step further [2]. The need of developments in the understanding
of tensile and flexural creep behavior of SFRC is motivated by

http://dx.doi.org/10.1016/j.conbuildmat.2014.04.139
0950-0618/Ó 2014 Elsevier Ltd. All rights reserved.
322 E. García-Taengua et al. / Construction and Building Materials 65 (2014) 321–329
several purposes, mainly: to improve the prediction of cracking
and the stress evaluation in prestressed members [6].
There is a considerable consensus in relation to the main
sources of SFRC time-dependent flexural strains [7]: creep in the
compression zone, time-dependent bond strains between concrete
and fibers sewing cracks, and creep of the fibers material. However,
there are some discrepancies among experimental results concern-
ing the effect of fibers on tensile and flexural creep which call for
further research, as observed by Garas et al. [8]. While several
authors have found out that fibers reduce creep and shrinkage
[1,2,8,9], other studies have concluded that hooked-end steel fibers
increase tensile creep [6].
Therefore, and considering that fibers contribution to load-
bearing capacity is mainly related to flexural response in the
cracked state, it is very important to evaluate how the material
keeps crack opening values low enough so as to guarantee rein-
forcement effectiveness [10].
However, most of the studies on FRC flexural creep have com-
pared the role of plastic or glass fibers to that of steel fibers, not
focusing on the effect of steel fibers themselves. When some of
these papers and reports [10–15] are brought together, the follow-
ing general aspects arise:
 There is a variety of test setups and methodologies. Considering
that SFRCs mechanical properties usually show considerable
scatter, the lack of a standardized methodology contributes to
uncertainty concerning SFRC creep behavior. An attempt to
develop a consistent methodology based on a creep test setup
for pre-cracked FRC specimens was needed.
 Most of the times the goal is not to characterize SFRC flexural
creep but to compare the effect of steel fibers to that of syn-
thetic fibers.
 Most studies limit their scope to one mix design, one type of
steel fiber, or certain fiber content. In these cases, it is not con-
sidered how the variation of these parameters may affect SFRC
creep behavior.
 In relation to the values considered for the stress/strength ratio,
it is usual to consider different values. However, the way they
are selected and the criteria this selection is based upon is not
usually the same.
 There is an important heterogeneity concerning testing
procedures and experimental approach. This leads to important
differences between studies concerning several aspects, for
instance: how load is applied to specimens, whether they are
notched or not.
The aforementioned general aspects motivated two major goals
for the research to be carried out. Firstly, to propose a general,
standard-like methodology to study flexural creep of concrete.
And second, to study the case of pre-cracked SFRC members in a
comprehensive fashion, analyzing the effect of several parameters
simultaneously.
In relation to the first aspect, the authors have made an effort to
develop a test setup and methodology which have been exten-
sively described elsewhere [16]. The test setup and methodology
proposed in [16] is susceptible of standardisation and can be used:
(a) to analyze creep behavior under some given conditions (for a
determined concrete mix design, load level, etc.), and (b) to charac-
terize the effect of particular fiber types and/or dosages under
standard conditions (materials, concrete mix design, fiber concrete,
pre-cracking level, load). This way creep of concrete is studied in
standard-like conditions so that future results can be easily com-
pared. This is the methodology that has been followed in the
experimental program reported herein. It is based on a structural
test, where creep occurs in bending. This creep test has two major
advantages: it is easier to perform and control than the direct
tension test, and it can be directly correlated to the bending test
as used for SFRC characterization (EN 14651). As compressive
creep and tensile creep can occur simultaneously in the section,
results may be affected by creep in the compressed zone of mid-
span section if the derived peak compressive stress is close to con-
crete compressive strength. As a result, it is not easy to dissemble
the contributions of both phenomena on flexural creep. In spite of
this, further developments in the interpretation of the phenomena
converging in flexural creep response as obtained from this test are
very interesting, as well as their possible implementation in codes,
but these aspects fall out of the scope of this paper.
2. Objectives and scope
The major purpose of this research was to analyze the effect
that different variables have on SFRC response to sustained flexural
loads in the cracked state.
These variables have been selected to represent both SFRC com-
position and the load applied. Accordingly, different types of
hooked-end steel fibers (in terms of length and slenderness), fiber
contents, and concrete mix designs have been considered. A num-
ber of prismatic specimens have been produced and subjected to
different sustained flexural loads covering usual values of these
parameters in real applications.
Several creep parameters have been analyzed. The analysis of
experimental results has followed a rigorous, statistical approach
to assess the significance of the variables considered. The result
is therefore a unified perspective on the relative contribution of
the variables considered to flexural creep response of SFRC in the
cracked state. This perspective offers a further conceptualization
of the phenomenon under study.
3. Methodology and experimental outputs
3.1. The creep test
Prismatic 150  150  600 mm specimens have been produced, notched, pre-
cracked, and then tested under flexural loads sustained for 90 days in agreement
with the creep test setup and methodology developed by the authors [16]. An over-
view of this methodology is given in Fig. 1.
In a first stage, specimens are pre-cracked: each specimen is notched and
loaded according to a four-point scheme based on the standard bending test
[17,18], with a 450 mm span between supports, until a crack mouth opening dis-
placement (CMOD hereafter) of 0.50 mm is reached. The load corresponding to this
crack width, Fw, is retained and the specimen is then totally unloaded.
Pre-cracked specimens are reloaded and subjected to sustained load conditions
according to the test setup shown in Fig. 2 (for dimensions and further details see
[16]). Specimens are tested in columns of three to rationalize the requirements of
time and space. The creep frame and all its components, in particular loading mem-
bers and supports, have been conceived to be stiff enough to avoid undesirable,
abrupt movements as well as friction in supports in order not to interfere the devel-
opment of creep strains. This, together with the gravity loading on top of the spec-
imens column (by means of a counterweight applied through a lever arm),
guarantees the application of a constant load. This way all three specimens are
loaded according to the four-point bending test and the load is kept constant for
a determined lapse of time. In the case of this research, this timespan was 90 days,
since the largest part of time-dependent strains occurs within the first 2 months [6].
Specimens
Production
Pre-Cracking Creep Test
Loading up to Unloading
Unloading Loading
CMOD=0.5 mm and Recovery
Complete 4-Point
Bending Test
Fig. 1. General testing procedure.
 E. García-Taengua et al. / Construction and Building Materials 65 (2014) 321–329 323

Fig. 2. Creep test setup.

The creep test ends after 90 days, when specimens are unloaded and elastic
deformation is recovered. Thereafter each specimen is subjected to a complete
bending test until failure to characterize the flexural response of the material in
the cracked state.
Any other conditions (namely humidity and temperature) are kept constant
throughout the testing process (for further details see [16,18]). An appropriate set
of measurement devices have been used to quantify and monitor CMOD values
regularly. Crack opening was monitored instead of deflection because CMOD values
have been reported to be more sensitive to the number of fibers than mid-span
deflection values (most recently [1]). Some examples of the evolution of CMOD
values vs time can be found in [16].
In addition to prismatic specimens for the creep test, all batches of concrete
were characterized by assessing their flexural response and compressive strength.

3.2. Definition of response parameters

According to the general testing procedure described in the previous section,
the complete process for each specimen leads to a flexural load vs CMOD curve,
as the idealized one shown in Fig. 3 for illustration purposes.
The first part of the curve corresponds to the pre-cracking stage. There is an
ascending linear branch until the first crack occurs (A). The specimen is gradually
loaded until a crack width of 0.50 mm is reached (B), and then it is totally unloaded.
Each specimen has been pre-cracked individually by being subjected to the 4-point
bending test. After that, specimens have been transferred to the test setup shown in
Fig. 2 and subjected to the creep test.
The creep test as such begins at point (C): the ascending line (CD) corresponds
to the loading process, which is followed by a horizontal branch (DE) corresponding
to the increasing deferred deformations (load sustained through time). This hori-
zontal line ends up when the specimen is unloaded 90 days afterwards (EF).
Finally, the specimen is subjected to the four-point bending test, which is rep-
resented by the third region of the plot: it begins with an ascending line (FG) and
continues with the residual performance curve of the specimen (GH).
Fig. 3. Idealized plot obtained after complete testing of a specimen.
Several parameters are obtained from the load–CMOD curve to characterize the
response of each specimen under sustained flexural load. These parameters consti-
tute the outputs of the experimental program. Therefore, the analysis of experimen-
tal results is focused on the values of these creep parameters. They can be grouped
as follows:
 Recovery ratio, r, is related to the pre-cracking stage. It measures the recovery of
strains right after pre-cracking. It is defined by Eq. (1), where: wp is the maxi-
mum CMOD reached when pre-cracking the specimen (namely 0.5 mm), and
wpr is the residual CMOD when the specimen is unloaded after pre-cracking.
wp  wpr
r¼ ð1Þ
wp
324 E. García-Taengua et al. / Construction and Building Materials 65 (2014) 321–329
 Untransformed data regarding the material’s strains: wci, the CMOD value at the
beginning of the creep test, measured 1 min after the load has been applied; and
wcd(90), the CMOD value measured 90 days after the load was applied, retained
as representative of the delayed, accumulated creep strains.
 Crack opening rates COR(t1–t2) for different time periods, evaluated as the ratio
between the increase in crack opening and the lapse of time from t1 until t2.
Three crack opening ratios have been considered: COR(0–14), COR(14–30) and
COR(30–90).
 Specific crack opening rates spCOR(t1–t2): spCOR(0–14), spCOR(14–30), and
spCOR(30–90), are defined as the corresponding crack opening rate typified by
the equivalent flexural tensile stress corresponding to the flexural load that is
applied.
 Creep coefficients u(j) are defined at different j times as the ratio between the
deferred crack opening at time j, wcd(j), and the initial crack opening at the
beginning of the creep stage, wci. In the case of this research the creep coeffi-
cients analyzed are u(14), u(30), u(90), i.e. at 14, 30, and 90 days, respectively.
 Creep coefficients referred to the origin uo(j) are evaluated as the ratio between
the deferred crack opening at time j, wcd(j), and the crack opening at the begin-
ning of the creep test in the complete curve, which is wpr + wci. Three creep coef-
ficients referred to the origin, uo(14), uo(30), and uo(90) have been considered,
at times of 14, 30, and 90 days respectively.
4. Experimental program and results
4.1. Selection of variables and levels
Table 1 summarizes the variables and the different levels con-
sidered for them, selected according to the criteria discussed in
the following paragraphs.
Several different concrete mixes have been produced. Each
particular concrete mix results from slightly modifying one of the
so-called ‘base’ mix designs, adjusted to include different fiber con-
tents. Two different base mix designs have been considered corre-
sponding to specified compressive strength (fc) values of 25 MPa
and 40 MPa, hence covering the range of low and mid-strength
concretes. The maximum aggregate size is 10 mm for all 40-MPa
mixes, and 20 mm for 25-MPa mixes with the exception of some
batches that were adjusted to a maximum aggregate size of
10 mm. Accordingly, maximum aggregate size has also been a var-
iable. This was convenient because it might have an effect on creep
strains: aggregates are known to behave elastically while the paste
fraction of concrete is the primary agent of time-dependent strains
[9].
Five different steel fibers have been considered: both fiber slen-
derness (kf ) and fiber length (Lf) have been considered as variables.
Fiber contents (Cf) used are 40 kg/m3 and 70 kg/m3, both below 1%
in volume as it is the most usual case in most of the applications
where SFRC is used.
Table 1
Variables considered.
Variables Levels
Compressive strength of concrete, fc 40 MPa
25 MPa
Maximum aggregate size (MAS) 10 mm
20 mm
80/35
Fiber slenderness, kf 80/50
Fiber length, Lf 65/40
45/50
50/30
Fiber content, Cf 40 kg/m3
70 kg/m3
Nominal load ratio, IFn 60%
80%
Position of specimen 1 (top)
2
3 (bottom)
Load ratio is the variable which takes into account the flexural
load that has been applied and sustained. The nominal load ratio
(IFn) is defined as the ratio between the load that is applied to
the specimen at the top and the load corresponding to a CMOD
of 0.50 mm in the pre-cracking stage, Fw, in percentage. However,
the applied load ratio (IFa) is not directly IFn: different specimens
from the same batch are never identical and, since they are tested
in columns of three (see Fig. 1), the specimen at the bottom bears a
slightly higher load than the one at the top. Therefore the applied
load ratio (IFa) differs from the other variables considered: this is
not a variable that could be pre-fixed at certain values. IFn has been
considered at 60% and 80%, so that the range of load ratios selected
by other authors is covered: 50% in the case of [1,6], and between
76% and 92% in the case of [3]. However, for the analysis of results
IFa values (ranging from 54.2% to 97.2%) are considered instead of
IFn, since they represent more exactly the loading applied to each
particular specimen.
Taking all that into consideration, the relative position of a
specimen in each group of three might somehow affect the results.
This is the reason why this relative position has been considered as
one more variable, so that its effect on creep parameters, whenever
present, could be detected and properly attributed instead of con-
founding the effects of other variables. As it is derived from Fig. 2,
there are three different positions: 1 for the top, 2 for the middle,
and 3 for the bottom.
Table 2 summarizes all specimens produced and tested in this
research as combinations of the variables considered. Each set of
three specimens listed consecutively in Table 2 corresponds to
the same batch.
4.2. Experimental results
The specimens tested in this research have shown residual
strength fR1 values (corresponding to CMOD of 0.5 mm) ranging
from 3.69 to 10.19 MPa. Therefore these SFRC mixes can be
referred to as normal or high performance. Accordingly, the equiv-
alent flexural stress applied in the creep test has ranged from 2.23
to 6.11 MPa. A kinematic analysis assuming plane strain state and
rectangular distribution for tensile stresses in the concrete section
reveals that peak compressive stress values are between 30% and
50% of concrete compressive strength. The aforementioned values
have been given only for informative purposes and they set the
ground for different approaches to the analysis of SFRC flexural
creep behavior, being out of the scope of this paper. The analyses
reported herein are focused on the response parameters which
have been defined in a previous section.
Values obtained for the response parameters are extensively
reported. Table 3 presents the complete dataset of experimental
results obtained from the pre-cracking stage (r, wci) and the creep
test: wcd(90), crack opening rates, specific crack opening rates,
creep coefficients and creep coefficients referred to the origin.
5. Analysis and discussion
5.1. Overview of the analysis
The effects that the variables considered (Table 1) have on each
one of the outputs of the experiment (creep parameters) have been
assessed by means of multiple linear regression (MLR hereafter)
[19]. The objective of MLR modeling is to relate each creep param-
eter to the variables considered. Then, statistical inference regard-
ing the relative importance of each variable is evaluated by means
of significance tests on the coefficients estimated in MLR modeling.
To study separately the effect of each variable on creep parameters
on the basis of one-to-one regression lines instead of MLR models
 E. García-Taengua et al. / Construction and Building Materials 65 (2014) 321–329 325

Table 2
Combinations of variables corresponding to the variables tested.

Id. fc (MPa) Max. Cf (kg/m3) kf Lf (mm) IFa (%) Pos.

aggr. size (mm)
1 40 10 40 80 35 60.9 1
2 40 10 40 80 35 54.9 2
3 40 10 40 80 35 54.2 3
4 40 10 40 80 35 97.0 1
5 40 10 40 80 35 81.9 2
6 40 10 40 80 35 70.5 3
7 40 10 70 80 35 61.9 1
8 40 10 70 80 35 59.2 2
9 40 10 70 80 35 59.2 3
10 40 10 70 80 35 81.0 1
11 40 10 70 80 35 82.2 2
12 40 10 70 80 35 81.3 3
–a 40 10 70 80 35 –a 1
13 40 10 40 80 50 79.6 2
14 40 10 40 80 50 78.8 3
15 25 20 40 80 50 88.1 1
16 25 20 40 80 50 82.5 2
17 25 20 40 80 50 82.2 3
18 25 20 40 65 40 56.2 1
19 25 20 40 65 40 60.4 2
20 25 20 40 65 40 70.8 3
21 25 20 40 45 50 97.2 1
22 25 20 40 45 50 80.2 2
23 25 20 40 45 50 78.3 3
24 25 20 40 45 50 90.9 1
25 25 20 40 45 50 84.4 2
26 25 10 40 45 50 75.1 3
27 25 10 40 50 30 76.3 1
28 25 10 40 50 30 57.7 2
29 25 10 40 50 30 54.4 3
–a 25 10 40 50 30 –a 1
30 25 10 40 50 30 72.9 2
31 25 10 40 50 30 72.4 3
a
Specimens corresponding to unavailable data due to problems with the data acquisition systems.

Table 3
Experimental results from the creep tests performed.

Id. r wci wcd(90) COR (103) spCOR (103) u(14) u(30) u(90) u0(14) u0(30) u0(90)
0–14 14–30 30–90 0–14 14–30 30–90
1 0.469 0.263 0.229 11.2 1.42 0.82 3.44 0.44 0.25 0.598 0.684 0.870 0.297 0.340 0.432
2 0.414 0.231 0.208 10.9 1.22 0.60 3.23 0.36 0.18 0.662 0.746 0.902 0.288 0.325 0.393
3 0.377 0.147 0.123 6.5 0.78 0.34 1.85 0.22 0.10 0.616 0.702 0.839 0.197 0.224 0.268
4 0.467 0.764 0.798 32.8 13.80 1.96 5.58 2.35 0.33 0.601 0.889 1.043 0.443 0.657 0.771
5 0.434 0.544 0.496 26.0 2.67 1.50 4.34 0.45 0.25 0.668 0.746 0.911 0.435 0.486 0.593
6 0.452 0.207 0.146 6.8 1.01 0.58 1.11 0.17 0.10 0.459 0.536 0.706 0.193 0.225 0.297
7 0.542 0.278 0.259 13.3 1.96 0.70 1.96 0.29 0.10 0.668 0.781 0.932 0.362 0.424 0.506
8 0.503 0.294 0.348 17.7 1.63 1.23 2.57 0.24 0.18 0.844 0.932 1.183 0.455 0.502 0.638
9 0.466 0.153 0.131 6.3 0.90 0.46 0.91 0.13 0.07 0.582 0.676 0.856 0.208 0.241 0.306
10 0.499 0.617 0.470 25.8 1.97 1.29 3.35 0.26 0.17 0.585 0.636 0.761 0.415 0.451 0.540
11 0.503 0.649 0.479 28.4 2.08 0.81 3.63 0.27 0.10 0.612 0.663 0.738 0.440 0.477 0.531
12 0.513 0.294 0.278 15.8 1.38 0.58 2.00 0.17 0.07 0.754 0.829 0.947 0.410 0.451 0.515
13 0.511 0.457 0.334 15.6 2.26 1.32 2.85 0.41 0.24 0.478 0.557 0.730 0.309 0.360 0.472
14 0.46 0.306 0.332 15.3 1.56 1.55 2.73 0.28 0.28 0.699 0.780 1.084 0.366 0.408 0.568
15 0.497 0.506 0.585 26.3 4.40 2.45 6.67 1.12 0.62 0.727 0.866 1.156 0.484 0.576 0.770
16 0.421 0.401 0.497 24.0 3.54 1.75 5.91 0.87 0.43 0.837 0.977 1.239 0.479 0.560 0.711
17 0.437 0.220 0.406 19.0 3.01 1.53 4.56 0.72 0.37 1.208 1.427 1.846 0.520 0.615 0.795
18 0.193 0.312 0.153 6.5 1.05 0.75 3.77 0.61 0.44 0.291 0.344 0.489 0.081 0.096 0.137
19 0.282 0.221 0.129 5.4 1.06 0.60 2.97 0.58 0.33 0.345 0.421 0.585 0.117 0.143 0.198
20 0.349 0.191 0.160 7.5 1.49 0.51 3.87 0.77 0.26 0.551 0.675 0.835 0.202 0.247 0.306
21 0.298 0.404 0.553 27.9 1.44 2.34 8.79 0.45 0.74 0.965 1.021 1.368 0.513 0.543 0.727
22 0.302 0.357 0.652 30.2 2.83 3.07 9.19 0.86 0.94 1.182 1.309 1.824 0.592 0.655 0.913
23 0.353 0.195 0.320 14.8 1.82 1.40 4.36 0.54 0.41 1.062 1.211 1.642 0.392 0.448 0.607
24 0.301 0.615 0.770 34.7 5.72 3.21 9.69 1.60 0.90 0.791 0.940 1.253 0.500 0.594 0.792
25 0.294 0.353 0.830 23.7 19.48 3.12 6.41 5.27 0.84 0.940 1.824 2.354 0.464 0.901 1.163
26 0.327 0.268 0.550 19.2 8.82 2.32 5.05 2.32 0.61 1.003 1.529 2.047 0.441 0.672 0.900
27 0.212 0.212 0.499 24.3 3.03 1.84 12.9 1.60 0.97 1.611 1.839 2.360 0.555 0.634 0.814
28 0.249 0.127 0.162 6.7 1.16 0.82 3.36 0.58 0.41 0.742 0.888 1.277 0.184 0.221 0.317
29 0.332 0.102 0.143 5.4 1.28 0.78 2.55 0.61 0.37 0.739 0.940 1.401 0.168 0.214 0.318
30 0.285 0.161 0.533 14.1 46.26 6.43 4.92 16.15 2.24 0.599 2.849 4.022 0.276 1.311 1.850
31 0.469 0.263 0.229 12.3 7.09 4.13 4.16 2.40 1.40 1.070 1.774 3.314 0.328 0.544 1.016
326 E. García-Taengua et al. / Construction and Building Materials 65 (2014) 321–329

would have been a defective approach: there would have been no rf ;i ¼ r0;i þ rk;i kf þ rL;i Lf ð3Þ
possibility of evaluating their relative importance.
where r0;i , rk;i , and rL;i are coefficients to be fitted; kf is fiber slen-
The aim of this research was not to arrive at a set of descriptive
derness; and Lf is fiber length (in mm).
equations but to identify the variables that have a statistically sig-
MLR models following this structure are called additive since
nificant effect on most of the creep parameters. Once identified,
they imply that the effect of the loading (given by the term ni  IFa
these key variables can be regarded as a general and therefore
in the equation) is simply added to that of the material (all other
valuable conclusion. To do that successfully, MLR models must
terms). Therefore each creep parameter is modeled as the sum of
be conceived taking into account the conceptual basis of the phe-
these two contributions.
nomenon under study.
For each creep parameter cpi, coefficients m0,i, ni, mc,i, ma,i, r0;i ,
Fig. 4 summarizes the process that has been followed to analyze
rk;i , and rL;i are estimated by least squares when the model is
the experimental results obtained for creep parameters:
fitted to experimental data. Table 4 summarizes the results of
significance tests on these coefficients. Each row in Table 4 corre-
1. The as-obtained dataset includes the values of all creep param-
sponds to a creep parameter: each row summarizes a MLR analysis,
eters for all specimens (Table 3).
R-squared values are given in the last column. Blank cells corre-
2. Deletion of outliers: anomalous results or outliers are detected
spond to effects which are not statistically significant. In the case
and cleared out from the results dataset. This has been
of significant effects, the sign of the estimated coefficient is given
performed by means of cluster analysis, namely the k-means
for interpretative purposes.
algorithm [20]. Only one out of the thirty-one specimens tested
The process followed to come to MLR models as summarized by
has been discarded: data corresponding to specimen 30 in
Table 4 has been stepwise regression [21]. The threshold consid-
Table 3 are excluded. The clean dataset is subjected to statistical
ered for p-values identifying significant effects is 0.05 in all cases,
analysis in the following steps.
which assumes a confidence level of 95%.
3. Analysis based on additive models: a first stage deals with
There are differences among the creep parameters considered
fitting MLR models which relate each creep parameter to the
with respect to the variables that have a statistically significant
simple effects of the variables considered.
effect on them. A very simple way of looking at the overall signif-
4. Analysis based on semi-empirical models: a conceptual basis is
icance of each variable is counting the number of creep parameters
searched for within the results of the previous step. New MLR
on which this particular variable has a statistically significant
models are developed for the creep parameters to include
effect. This ‘count’ is given in the last row of Table 4. It is clearly
interactions between the variables in addition to their simple
detected that fiber slenderness (kf ) and load ratio (IFa) are the
effects. These interactions are selected on the basis of previous
key parameters on flexural creep behavior.
knowledge: that is the reason why these models are called
The relative position of a specimen (Pos. in Table 4) in the creep
semi-empirical.
test setup has turned out to be relevant concerning some of the
5. Interpretation: the results of previous analyses concerning the
creep parameters. The effect of this variable on creep parameters
effects detected and their significance are put together and
has been explored by means of box-and-whisker plots, shown in
interpreted.
Figs. 5 and 6. A very similar tendency has been found for those
creep parameters on which relative position has a significant
5.2. Additive models effect. It is observed in Fig. 5, where each box stands for the inter-
quartile length and the notches represent the 95%-confidence
The so-called additive MLR models follow this general interval for the mean in each case. Significant differences between
expression: position 3 and position 1 are observed, since their corresponding
confidence intervals for the mean do not overlap. In the opinion
cpi ¼ m0;i þ ni IFa þ mc;i fc þ ma;i MAS þ rf ;i C f ð2Þ
of the authors, this is due to the different support and loading
where cpi stands for a certain creep parameter; IFa is the applied conditions under specimen 3 (at the bottom) with respect to
load ratio (in %); fc is the specified compressive strength of concrete specimens 1 and 2. For contrastive purposes, Fig. 6 shows box-
(in MPa); MAS is the maximum aggregate size (in mm); m0,i, ni, mc,i, and-whisker plots corresponding to a creep parameter on which
and ma,i are coefficients to be fitted; and Cf is the fiber content (in
kg/m3), whose effect may vary depending on fiber geometry accord-
Table 4
ing to the following expression: Results from the MLR analyses (additive models) on creep parameters.

fc MAS Cf kfCf LfCf IFa Pos. R2

As-Obtained Deletion of Clean r () () (+) (+) – – 0.84

Dataset Outliers Dataset wci (+) (+) ()3 0.80
wcd(90) (+) () (+) ()3 0.77
COR(0–14) (+) ()3 0.83
COR(14–30) () (+) 0.27
COR(30–90) () () () (+) (+) 0.63
Significant Additive spCOR(0–14) (+) () (+) ()3 0.71
Variables Model spCOR(14–30) () (+) 0.30
spCOR(30–90) () () () (+) 0.69
u(14) () () (+) 0.40
Significant Semi-Empirical u(30) () () () (+) 0.47
Interactions Model u(90) () () () (+) 0.55
MLR-based inference uo(14) (+) 0.56
uo(30) (+) () (+) 0.58
uo(90) (+) () (+) 0.55
(Count) 6 6 5 11 4 12 4
Interpretation
of Effects +/ in each case indicates the sign of the coefficient multiplying the corresponding
simple effect or interaction. In the case of Pos., the sign is that of the coefficient
Fig. 4. Steps followed to analyze experimental results. multiplying the boolean variable which equals 1 when the specimen is in position 3.
 E. García-Taengua et al. / Construction and Building Materials 65 (2014) 321–329
0.035
0.025
COR (0-14)
0.015
0.005
1 2 3
Position
Fig. 5. Box-and-whisker plot showing the effect of position on COR(0–14).
1.2
1.0
0.8
ϕo (90)
0.6
0.4
0.2
1 2 3 section can be recalled here. Fiber length does not interact with
Position
Fig. 6. Box-and-whisker plot showing the effect of position on u0(90).
position has no significant effect: all confidence intervals for the
mean overlap.
The effect of fiber content is more important than it seems at
first sight. First of all, the effect of fibers on creep behavior is some-
what covered by the effect of load ratio. It is worth recalling that
load ratio, IFa, is the ratio between the applied load and Fw, which
corresponds to wp. Since Fw clearly depends on fiber content, the
fact that load ratio is statistically significant means that fiber con-
tent is also significant, because residual load bearing capacity of
concrete (and therefore Fw) is increased with fiber content.
5.3. Semi-empirical, constitutive models
From the additive models summarized in Table 4 it follows that
fiber slenderness (kf ) and fiber content (Cf), together with load
ratio (IFa), are the capital factors influencing most of the creep
parameters.
A further step in the analysis is concerned with MLR models
which are rooted on the theoretical background of the phenome-
non under study. Since they involve putting together this
background and the experimental results, they have been called
semi-empirical. This is an effort to interpret the statistical analysis
of experimental results from a physical, mechanical perspective.
Other authors have previously underlined the convenience of
approaching creep strains by means of empirical equations after
the form of constitutive laws [6]. Considering that constitutive
equations relate strain to stress, it has been decided to formulate
327
semi-empirical MLR models where the creep parameter is related
to the load ratio, IFa, whose effect is modified by the variables
which are related to the material. As a result, MLR semi-empirical
models follow this general expression:
cpi ¼ m0;i þ ðni þ mc;i fc þ ma;i MAS þ rf ;i C f Þ  IFa ð4Þ
where the term that multiplies IFa does not only relate to IFa itself
(such standalone simple effect is represented by coefficient ni) but
is also affected by concrete compressive strength (fc), maximum
aggregate size (MAS), and fibers geometry (rf ;i , as defined by Eq.
(3)) and fiber content (Cf). It is worth noting that assuming this for-
mulation is similar to have results for all creep parameters typified
by IFa.
The process followed to come to MLR models that best fits
experimental results and considers only statistically significant
variables has been, as in the previous section, stepwise regression
[21]. The threshold assumed for p-values identifying significant
effects is 0.05 in all cases. Table 5 summarizes the significance tests
for the semi-empirical MLR models.
Load ratio, IFa, has a standalone simple effect which is statisti-
cally significant on most of the creep parameters. This is clearly
observed in Fig. 7, which shows creep coefficient u0(90) vs IFa:
u0(90) clearly increases when load ratio is increased. The coeffi-
cient associated to this effect is positive in all cases, which is coher-
ent with the fact that increasing load ratios lead to increasing crack
widths, regardless of the particular creep parameter considered.
Specified concrete compressive strength significantly interacts
with load ratio concerning some creep parameters. However, the
coefficient corresponding to this interaction is positive in some
cases and negative in some others: at this point it is not clear
whether high concrete compressive strength directly implies
higher creep strains or not.
Concerning fibers, all considerations given in the previous sub-
load ratio. On the contrary, fiber slenderness and fiber content
clearly modify the effect of load ratio on flexural creep: the effect
of their interaction with load ratio is statistically significant con-
cerning several of the creep parameters considered.
The sign of the coefficient corresponding to the interaction
between fiber content and load ratio is positive in all cases. This
is clearly due to the following fact: the greater fiber content is,
the higher the load bearing capacity. Therefore, for a given load
ratio, the applied load will be higher when higher fiber contents
are considered. As a result, this coefficient being positive does
not mean that fibers increase creep strains for a certain load ratio.
Rather, their presence makes the load applied higher for a given
load ratio and therefore creep strains are increased.
On the other hand, the sign of the coefficient corresponding to
the interaction between fiber slenderness, fiber content and load
ratio is negative in all cases. This clearly indicates that the use of
fibers with higher slenderness lessens the effect that high load
ratios have on creep strains. In consequence, it can be said that
the use of fibers is a good strategy in order to control creep strains
and the best choice is clearly fibers with high slenderness.
5.4. Comparison of additive and semi-empirical models
The experimental results obtained from the creep test have
been analyzed on the basis of two alternative perspectives in terms
of MLR modeling: additive models, and semi-empirical, constitu-
tive models. They point out to fiber slenderness (kf ) and fiber con-
tent (Cf) together with the applied load ratio (IFa) as the most
important variables to account for.
There are two key conclusions with respect to the role that
fibers play in flexural creep. The first is that fiber length does not
328 E. García-Taengua et al. / Construction and Building Materials 65 (2014) 321–329

Table 5
Results of the MLR analyses (constitutive models) on creep parameters.

IFa fcIFa MASIFa CfIFa kfCfIFa LfCfIFa PosIFa R2

r – – – – – – –
wci (+) (+) ()3 0.84
wcd(90) (+) ()3 0.72
COR(0–14) (+) ()3 0.84
COR(14–30) (+) (+) () 0.24
COR(30–90) (+) () () 0.59
spCOR(0–14) (+) () () ()3 0.76
spCOR(14–30) (+) () 0.31
spCOR(30–90) (+) () (+) () () 0.74
u(14) (+) () 0.32
u(30) (+) () 0.39
u(90) (+) () 0.39
uo(14) (+) 0.56
uo(30) (+) 0.53
uo(90) (+) (+) () 0.55
(Count) 10 5 2 6 7 1 4

+/ in each case indicates the sign of the coefficient multiplying the corresponding simple effect or interaction. In the case of Pos., the sign is that of the coefficient multiplying
the boolean variable which equals 1 when the specimen is in position 3.

0.004
Creep coefficient at origin (90 days)
1.2
1.0

0.003
0.8

COR (30-90)
0.002
0.6
0.4

0.001
0.2

0.000
0.0

0 20 40 60 80 100
Applied load ratio, IFa(%) 45 50 65 80
Fiber slenderness
Fig. 7. Scatterplot showing the effect of load ratio on u0(90).
Fig. 8. Box-and-whisker plot showing the effect of fiber slenderness on COR(30–
90).

have a significant effect on flexural creep of SFRC (Tables 4 and 5).

The second is that fiber slenderness has a statistically significant
1.2

effect on most of the creep parameters, which is clearly detected

Creep coefficient at origin (90 days)

in Tables 4 and 5. This is also observed in Figs. 8 and 9, which

are given for illustration purposes. They show the box-and-whisker
1.0

plots of COR(30–90) and u0(90) respectively vs fiber slenderness,

the notches representing the limits of 95%-confidence intervals
0.8

for the mean in each case. Both parameters tend to decrease when
fiber slenderness is increased, a trend which has been detected to
0.6

be statistically significant.
Both perspectives (additive models and constitutive models)
have led to very similar conclusions, and this underlines the con-
0.4

sistency of the conclusions reached. Furthermore they are totally

coherent with one another in terms of the signs of the coefficients
0.2

fitted for each. Therefore they constitute two alternative under-

standings of flexural behavior of SFRC:
45 50 65 80
 The additive models define each creep parameter as the result Fiber slenderness
of directly adding the so-called material’s contribution (endog-
enous sources: variables describing the fibers, the aggregates Fig. 9. Box-and-whisker plot showing the effect of fiber slenderness on u0(90).
and the cementitious matrix) to the effect of the loading (exog-
enous source).
 Differently, the semi-empirical or constitutive models are con- On the whole, the semi-empirical, constitutive MLR models do
ceived on a mechanistic basis: creep parameters are determined not improve the goodness of fit obtained with additive MLR models
basically by load ratio, and the effect of load ratio is modified by (compare Tables 4 and 5: the R-squared value is the same in 2
material properties. cases, is increased in 5 cases, and is decreased in 7 cases). This
 E. García-Taengua et al. / Construction and Building Materials 65 (2014) 321–329
means that the semi-empirical, constitutive MLR models are not
better than the additive MLR models if they are regarded as
predictive equations. However, they represent a perspective on
the phenomenon under study which is probably better justified
from a physical point of view.
It is important to bear in mind that the purpose of this research
was not to come to very accurate predictive equations, which is a
different task because of the following reasons. First, the number
of variables considered is rather high (7 variables) when compared
to the number of tests performed (31 tests): the experimental
program would have been more exhaustive if the authors had
aimed at obtaining very accurate equations for predictive
purposes. Second, SFRC flexural response shows a considerable
scatter in the experimental results [22,23], which is mainly
dependent on the geometry of the fibers used and their dosage
[1]. Third, the selection of creep parameters has been based on pre-
vious literature but they are probably susceptible of being refor-
mulated and optimized in terms of their variability. The next
step is obviously directed to obtain accurate predictive equations
for different creep parameters which take into account the contri-
bution of fibers in SFRC and towards a more stable definition of
creep parameters. Such tasks are currently in progress, involving
new experimental programs and the use of statistical multivariate
techniques.
6. Conclusions
The experimental program carried out and the analysis of the
creep test results by means of statistical techniques has made it
possible to come to the following conclusions:
 The applied load ratio, IFa, has always a statistically significant
effect on flexural creep response, and increasing the load ratio
always leads to higher creep strains, regardless of the creep
parameter considered.
 Fiber content and fiber slenderness together with load ratio
determine flexural creep of pre-cracked SFRC members.
 Fiber length does not have a statistically significant effect on the
creep parameters analyzed.
 Fiber content has a statistically significant effect on most of the
creep parameters considered. Increasing fiber content increases
load bearing capacity and higher loads can be applied without
increasing creep strains.
 Increasing fiber slenderness leads to reduced creep strains. Fur-
thermore, fiber slenderness has an important synergic effect
with load ratio concerning the creep parameters considered:
using fibers with high slenderness lessens the effect that load
ratio has on creep strains.
 As a consequence, adding fibers is a good strategy in order to
control creep strains and, though they are not required in high
amounts, the best choice is to use fibers with high slenderness.
329
References
[1] Buratti N, Mazzotti C. Effects of different types and dosages of fibers on the
long-term behavior of fiber-reinforced self-compacting concrete. In: 8th RILEM
international symposium on fiber reinforced concrete BEFIB 2012, Guimaraes,
Portugal; 2012. p. 177–8.
[2] Nakov D, Markovski G. Time dependant behavior of SFRC elements under
sustained loads. In: 8th RILEM international symposium on fiber reinforced
concrete BEFIB 2012, Guimaraes, Portugal; 2012. p. 189–90.
[3] Barpi F, Valente S. Creep and fracture in concrete: a fractional order rate
approach. Eng Fract Mech 2002;70(5):611–23.
[4] Mouton CJ, Boshoff WP. Initial study on the tensile creep of cracked steel fibre
reinforced concrete. In: 8th RILEM international symposium on fiber
reinforced concrete BEFIB 2012, Guimaraes, Portugal; 2012. p. 326–37.
[5] Zhao G, Di Prisco M, Vandewalle L. Experimental research on uni-axial tensile
creep behaviour of pre-cracked steel fiber reinfored concrete. In: 8th RILEM
international symposium on fiber reinforced concrete BEFIB 2012, Guimaraes,
Portugal; 2012. p. 183–5.
[6] Bissonnette B, Pigeon M, Vaysburd M. Tensile creep of concrete: study of its
sensitivity to basic parameters. ACI Mater J 2007;104(4):360–8.
[7] Kanstad T, Zirgulis G. Long-time creep testing of pre-cracked fiber reinforced
concrete beams. In: 8th RILEM international symposium on fiber reinforced
concrete BEFIB 2012, Guimaraes, Portugal; 2012. p. 195–6.
[8] Garas VY, Kahn LF, Kurtis KE. Short-term tensile creep and shrinkage of ultra-
high performance concrete. Cem Concr Compos 2009;31(3):147–52.
[9] Rouse JM, Billington SL. Creep and shrinkage of high-performance fiber-
reinforced cementitious composites. ACI Mater J 2007;104(2):129–36.
[10] Barragán BE, Zerbino RL. Creep behavior of cracked steel fiber reinforced
concrete beams. In: Proceedings of the 7th international RILEM symp on fiber
reinforced concrete: design and applications (BEFIB 2008), Chennai, RILEM
Publications; 2008. p. 577–86.
[11] Chanvillard G, Roque O. Behavior of fiber reinforced concrete cracked section
under sustained load. In: Proceedings of the 3rd international workshop on
high performance fiber reinforced cement composites, Mainz (Germany),
RILEM Publications; 1999. p. 239–50.
[12] Granju JL, Rossi P, Chanvillard G, et al. Delayed behavior of cracked SFRC beams.
In: Proceedings of the fifth international RILEM symp on fiber reinforced
concrete (BEFIB 2000), Lyon, France, RILEM Publications; 2000. p. 511–20.
[13] Mackay J. Behavior of steel and synthetic fiber reinforced concrete under
flexural creep loading. MSc thesis, Dalhousie University, Canada; 2002.
[14] Bast T, Eder A, Kusterle W. Kriechversuche an Kunststoffmakrofaserbetonen
Untersuchungen zum Langzeitverhalten von Faserbetonen unter
Biegezugbeanspruchung – ein Zwischenbericht, Faserbeton Beiträge zum qq.
Vilser Baustofftag; 2007. p. 32–5.
[15] Zerbino RL, Barragán BE. Long-term behavior of cracked steel fiber-reinforced
concrete beams under sustained loading. ACI Mater J 2012;109(2):215–24.
[16] Arango SE, Serna P, Martí-Vargas J, García-Taengua E. A test method to
characterize flexural creep behaviour of pre-cracked FRC specimens. Exp Mech
2012;52(8):1067–78.
[17] European Committee for Standardization. European Standard EN 14651:2007
‘‘Test method for metallic fiber concrete – measuring the flexural tensile
strength (limit of proportionality (LOP), residual)’’; 2007.
[18] Arango SE. Fluencia a flexión del hormigón reforzado con fibras de acero
(SFRC) en estado fisurado. PhD dissertation, Universitat Politècnica de
València, Valencia, Spain; 2010 [in Spanish].
[19] Montgomery D. Design and analysis of experiments. New York: John Wiley &
Sons Inc.; 2005. 643pp.
[20] Hair F, Black W, Babin B, Anderson R. Multivariate data analysis. Prentice Hall;
2009. 816pp.
[21] Draper N, Smith H. Applied regression analysis. New York: John Wiley & Sons;
1981.
[22] Di Prisco M, Plizzari G, Vandewalle L. Fiber reinforced concrete: new design
perspectives. Mater Struct 2009;42(9):1261–81.
[23] Torrijos MC, Barragán BE, Zerbino RL. Placing conditions, mesostructural
characteristics and post-cracking response of fiber reinforced self-compacting
concretes. Constr Build Mater 2010;24(6):1078–85.