Cement and Concrete Composites 86 (2018) 160e170

Contents lists available at ScienceDirect

Cement and Concrete Composites

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

ASR expansions in concrete under triaxial confinement

Joaquín Liaudat a, *, Ignacio Carol a, Carlos M. Lo

pez a, Victor E. Saouma b
a
Department of Environmental and Civil Engineering, Technical University of Catalonia, 08034 Barcelona, Spain
b
Department of Civil, Environmental and Architectural Engineering, University of Colorado, Boulder, CO 80309, USA

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

Article history: It is well known that the Alkali-Silica Reaction (ASR) expansion in concrete is influenced by the stress
Received 21 June 2017 state of the material. However, the extent and nature of this influence is not completely understood due
Received in revised form to the scarcity of experimental investigation under triaxial stress states. This paper presents experi-
22 September 2017
mental ASR expansion curves obtained from cubical concrete specimens subjected to three different
Accepted 25 October 2017
Available online 31 October 2017
triaxial stress states, by means of a new testing machine especially designed for this purpose. The results
seem to confirm that the volumetric ASR expansion rate is reduced as the applied volumetric
compressive stress is increased. Additionally, there seems to be an increase of the expansion rate in the
Keywords:
Alkali-silica reaction
less compressed direction in detriment of the expansion rates in the most compressed ones.
Multiaxial stress © 2017 Elsevier Ltd. All rights reserved.
Durability

1. Introduction load on cylindrical specimens (Fig. 1b), usually via creep frames, e.g.
Refs. [7e9]. To isolate ASR expansions, creep deformations due to
Since the early investigations on ASR in concrete, it was noticed the applied load are deducted from the overall expansion curve
that the stress state has an influence on the magnitude and dis- obtained. Since, for obvious reasons, it is not possible to directly
tribution of ASR-induced cracking and expansions [1]. Stresses may measure creep on reactive concretes, creep strains had to be
be due to external loads applied to the structure or the reaction to assessed on concretes with similar characteristic but with non-
constrained expansion due to internal (rebars) or external reactive aggregates or by means of analytical expressions from
restraints. the literature. A similar approach is followed to deduct shrinkage
In the last two decades, a number of experimental studies on the strains when they are significant.
effect of mechanical restraints on the development of ASR expan- In the work of Multon and Toutlemonde [8], some specimens
sions have been proposed. A first group of experimental studies to were also tested with passive radial restraint, in order to induce
deal with this problem consisted in impeding the longitudinal triaxial stress in the material. The radial restraint to expansion was
expansion of mortar or concrete bars via very stiff loading frames, added by means of steel rings 3- or 5-mm-thick, which were not
e.g. Refs. [2e5]. This setup makes it possible to measure what is connected to each other so that they would not exert any longitu-
called “expansive” or “swelling pressure”, i.e. the stress developed dinal restriction (Fig. 1c). This setup has the drawback that the
in the specimen in the restrained direction. Kagimoto et al. [6], radial stresses are not constant but linearly dependent on the
introduced a variation to this scheme by performing tests with expansion in this direction, making the interpretation of the results
loading frames of different stiffnesses. Refer to Fig. 1a for a sche- troublesome.
matic representation of these kind of tests. The results obtained in Finally, in a recent work, Gautam and Panesar [10] have pro-
this setup are difficult to interpret since the evolution of the posed an ingenious test setup for measuring ASR expansions in
swelling pressure will not only depend on the ASR mechanisms, but concrete under constant triaxial stress states. For these tests, cubic
also on the rheological behavior of concrete due to the develop- specimens of 254
254
254 mm were loaded by means of post-
ment of internal pressures. tensioning bolts (Fig. 1d). The bolts were stressed close to the yield
A second group of studies consisted in applying a constant axial limit of steel, in such a way that any additional elongation of the
bolt induced by ASR expansion in the concrete will not induce any
significant increase of the reaction exerted by the bolts on the
* Corresponding author. specimen. The main drawback of this setup is that the stress dis-
E-mail addresses: [email protected] (J. Liaudat), [email protected] tribution in the specimens is highly nonuniform, since the stresses
(I. Carol).

https://doi.org/10.1016/j.cemconcomp.2017.10.010
0958-9465/© 2017 Elsevier Ltd. All rights reserved.
 J. Liaudat et al. / Cement and Concrete Composites 86 (2018) 160e170 161

of the loading plates in contact with the specimens are crossed by

grooves for a solution to circulate in order to keep the specimen wet
and supply reactants (alkalis). A LabVIEW-based computer control
and data acquisition system maintains the desired confining stress
and specimen temperature while logging all relevant data points.

2.1. Loading frame

A three-piece steel assembly is used to provide reaction forces

for the three hydraulic actuators. The frame is constructed of W
12
24 I-beams and designed to deflect less than 0.25 mm under
the full 220 kN confinement load. Each section of the frame is
bolted together at the vertical axis to form a monolithic structure.
Refer to Fig. 2 for a general view of the load frame assembly.

2.2. Triaxial confinement and measurement systems

The AARTM is equipped with three 220 kN hydraulic actuators

and six 138:3
138:3 mm stainless steel loading plates used to
apply the confining stress to the test specimen. Actuators are
connected to the three “active” loading plates by means of cylin-
drical steel rods. Three additional steel rods of similar length con-
nect the “passive” plates to the loading frame. Each rod was drilled
to accept a 32 mm pin used to transfer the confining load to the
loading plates. Active rods were machined to thread into the hy-
draulic actuators, while the passive rods were threaded to accept a
large nut. The nut provides an adjustment mechanism for aligning

Fig. 1. Schematic representation of ASR expansion tests under different mechanical

restraints.

concentrate under the bolt anchors. Therefore, the interpretation of

the results requires estimation of the real distribution of stresses in
the specimen via numerical modeling.
In this paper, a new testing machine for performing ASR
expansion tests with cubic concrete specimens under constant
triaxial stress is presented. The equipment has the advantage that
the stresses are applied by means of six loading plates that cover
most of the surface of the specimen, creating, in that way, a quasi-
uniform stress state inside the material (Fig. 1e). The machine has
been used to obtain ASR expansion curves for three different stress
states which are also presented and discussed in this paper.
Complementarily, a number of mechanical tests were performed to
characterize the concrete tested, both before and after the ASR
expansion tests.

2. AAR triaxial machine

With the aim of measuring ASR expansions in concrete under

true triaxial confinement, i.e. under different constant stress along
each of the main directions, an ad hoc testing machine, called
“Alkali-Aggregate Reaction Triaxial Machine” (AARTM), was origi-
nally designed and constructed by the fourth author at the U. of
Colorado - Boulder. Later, in 2010, it was transferred to UPC (Bar-
celona) where a number of modifications were introduced to the
original set-up.
The AARTM consists of a triaxial load frame which can deliver
0e9 MPa confining stress on each axis of a 150
150
150 mm
cubical specimen. Specimen deformation is measured using three
displacement sensors on each of the three axes. The apparatus is
designed to be capable of raising and maintaining the specimen
temperature at a pre-set value between 30 and 70  C, and the faces Fig. 2. General view of the AARTM.
162 J. Liaudat et al. / Cement and Concrete Composites 86 (2018) 160e170

the non-active loading plates. Refer to Fig. 3 for a view of the load pressure relief valve is set to a pressure slightly higher than the
application system. desired confining pressure and supplements the computer
The nominal applied stress is calculated as the ratio between the controlled hydraulic pump to provide tighter load regulation.
applied load and the nominal surface area of a specimen
(150
150 mm). It must be noted that the surface area of the plates 2.5. Liquid circulation system
is slightly smaller than that of the face of the specimens and,
therefore, the stresses within the specimen are not completely As already mentioned, the face of the loading plates in contact
uniform. However, numerical simulations of a specimen subjected with the specimen are crossed by grooves to allow the circulation of
to the same load in each direction indicate that 91% of the volume alkaline solution. Liquid leakage is prevented by means of Viton™
of the specimen has stresses within ±20% of the nominal ones. O-rings in between the plates and the specimen. Refer to Fig. 4 for
Each of the three active loading plates connected to a hydraulic detail views of a loading plate. The sealing offered by the O-ring is
actuator is equipped with three Linearly Variable Differential not completely effective and needs to be complemented by the
Transformers (LVDTs), from Macro Sensors (model CD-375, range application of silicon sealant covering the specimen-plate contours.
±2.54 mm, linearity ±0.5%). Each LVDT senses the distance to the The plates are connected to each other in a series circuit by means
opposite plate by means of a threaded extension rod which is of polyamide 4
2:7 mm hoses thermally insulated with self-
attached at the opposite plate end. Refer to Fig. 4 for the relative adhesive tape of EPDM rubber foam. The alkaline solution circu-
position of the LVDTs on the active loading plates. Strain in a given lates by gravity through the loading plates series circuit, from a
direction is then calculated as the ratio between the average rela- stainless steel tank of 12 L, located at the top of the loading frame,
tive displacements measured with the i- and j-LVDTs (see Fig. 4, to a polyethylene tank of 20 L located at floor level (Fig. 2). The
left), and the initial specimen dimension in that direction (approx. liquid flow is regulated manually by means of a stainless steel
150 mm). The k-LVDT is used for monitoring the relative rotation needle valve. Finally, a diaphragm pump commanded by a float
between opposite plates. switch in the upper tank, recirculates the liquid from the lower tank
to the upper one. In order to prevent that the circulating liquid
2.3. Temperature control system cools down the specimen, the upper tank is equipped with a 400 W
electric cartridge heater commanded by a mechanical thermostat
A closed-loop temperature control system is employed on the which keeps the liquid temperature close to the test temperature.
AARTM to accelerate the AAR. Each stainless steel loading plate is
equipped with a 150 Watt flush-mounted heating element (Figs. 3 3. Experimental procedures
and 4). The three active loading plates are equipped with precision
temperature sensors (Fig. 4, left). Averaged temperature feedback 3.1. Concrete specimens
from the loading plates is used for the closed-loop control. This
system has proven to be stable and effective, reliably holding the Two different kinds of concrete were investigated. The first one,
specimen temperature constant within ±1  C. called “control concrete”, was made only with non-reactive
(crushed limestone) aggregates. The second one, called “reactive
concrete”, in contrast, was made using the same mixture but
2.4. Hydraulic load systems replacing the coarser fraction of aggregates with crushed colorless
soda-lime glass as reactive aggregate. Control specimens were
Three independent high pressure hydraulic loading systems, tested in the same conditions as the reactive specimens in order to
one for each axis direction, are employed to provide specimen assess creep and/or shrinkage deformations. The cement used was
confining stress via the stainless steel loading plates. These systems CEM I 42,5 N-SR 5 [11]. Both mixtures had a cement content of
utilize pneumatic energy delivered via the laboratory compressed 420 kg/m3 and a water/cement weight ratio of 0.45. The alkali level
air distribution system to drive three hydraulic pumps. The air flow of the concrete was raised to 5.25 kg/m3 Na2Oeq by adding NaOH to
to the hydraulic pumps is controlled by the computer using direct the mixing water. The non-reactive aggregate was well graded and
acting electric solenoid valves. Varying the duration and interval in passed a 12.70 mm sieve. For the reactive concrete, the coarser 64%
which the solenoid valves are open allows for proportional control of the non-reactive aggregate volume (particles between 2.50 and
of the pressure rise in the system. Each hydraulic circuit includes a 12.70 mm) was replaced by crushed soda-lime glass particles.
hydraulic manifold where a pressure transducer, an analog pres- The specimens were cast at four different times. Control and
sure gauge and an adjustable pressure relief valve are installed. This reactive specimens cast on the same date constitute a “casting set”,
which is indicated for each tested specimen in Table 1. After casting,
the specimens remained in molds for 24 h at room temperature and
with the exposed surface covered with a plastic film in order to
prevent moisture loss. Once demolded, specimens were put in 1 m
NaOH solution (1 mol of NaOH per kg of water) at room tempera-
ture (21 ± 2  C) for at least three months. Before performing the
expansion tests, the pouring face of each specimen was polished to
remove excessive roughness.
The curing solution, which was the same to be used for the
following expansion tests, was chosen to have a total alkali con-
centration similar to that of the specimen pore solution, preventing
in that way the reduction of the pore solution alkalinity due to
alkali leaching towards the curing bath. The minimum extent of the
curing period has been adopted based on previous tests (not pub-
lished) which indicated that for shorter curing periods the expan-
sion rate measured in the testing conditions described below is
Fig. 3. Detail view of the AARTM. dependent on the age of the specimen.
 J. Liaudat et al. / Cement and Concrete Composites 86 (2018) 160e170 163

Fig. 4. Detail views of an active loading plate of the AARTM.

Table 1 NaOH solution at 60  C).

List of tested specimens. Note: Id. ¼ Identifier. One week before starting the free expansion test, each specimen
Type Id. Casting set # Curing period Test duration was removed from the alkaline curing bath and let air dry for 24 h.
(weeks) (weeks) Then, stainless-steel datum discs were glued with epoxy resin to
Load case 0-0-0 the specimen surface in the configuration indicated in Fig. 5. In all
cases, the Z-direction was matched with the specimen casting di-
TM01 1 39 30
TM10 2 14 36 rection. Each pair of datum discs allowed measuring the length
Reactive
TM12 2 14 36 change at the surface of the specimen in one main direction by
TM13 2 14 36 means of a 100 mm DEMEC strain gage (Fig. 5). In part of the
TM05 2 14 36 specimens, only one measurement in each direction (X1, Y1, Z1)
Control TM06 2 14 36 was performed. After 24 h, once the epoxy resin was cured, each
TM07 2 14 36 specimen was placed in a 19 L airtight plastic container filled with
Load case 1-1-1
1 m NaOH solution and kept at room temperature for another 5
Reactive
TM02 1 42 3 days for rewetting. Afterwards, the containers with the specimens
TM21 3 39 5 were introduced in an oven at 60  C. After 24 h, the containers were
Control
TM03 1 39 2 removed from the oven in order to perform the first reading (zero-
TM19 3 35 3 reading) in each pair of datum discs. Then, the containers with the
Load case 9-9-1
specimens were returned to the oven and kept there at 60  C until
TM11 2 44 3 the next reading. The average temperature in the oven during the
Reactive
TM20 3 49 3
test was 59.95  C with a standard deviation of 0.33  C. Measure-
Control
TM08 2 34 3 ments were made for total test duration of 30 weeks. For clarifi-
TM15 3 53 3 cation, a schematic timeline of the testing procedure is given in
Load case 9-9-9
Table 2.
TM14 2 56 5
Reactive
TM22 3 67 3

Control TM09 2 68 4
TM16 3 63 3
Mechanical tests

TM24 3 47 none
Reactive TM30 4 15 none
TM31 4 15 none

TM18 3 47 none
Control TM25 4 15 none
TM26 4 15 none

3.2. Free expansion test setup

To assess the effect of the applied stress state on the ASR

expansion rate, it is essential to perform additional free expansion
measurements. These tests cannot be carried out with the AARTM
since a minimum pressure is needed in each direction to ensure full
contact of the loading plates with the specimen, in order to guar-
antee the sealing effect of the O-rings. Therefore, a different pro-
cedure described below was applied to perform a number of free
expansion tests with specimens of identical characteristics to those Fig. 5. Specimen with datum discs and DEMEC strain gage used for strain measure-
used in the AARTM, and under the same exposure conditions (1 m ment in free expansion tests. Z-direction was matched with the casting direction.
164 J. Liaudat et al. / Cement and Concrete Composites 86 (2018) 160e170

Table 2 temperature were kept constant for at least 21 days. After that, in
Schematic timeline of the free expansion tests. “x” denotes the age (in days) of the some cases, the load state was changed and the test continued for
specimen when first removed from the alkaline bath.
some additional time. The average temperatures during the test
day 0  Concrete casting periods were slightly below the nominal, with values ranging be-
day 1  Unmolding tween 59.88 and 59.98  C, and with standard deviation always
/ Curing in alkaline solution at room temperature (between 14 and
39 weeks)
below 0.45  C. The average applied stresses during the test periods
were within 5% of the nominal values, with standard deviations
day x  Removal of the specimen from the alkaline bath under 3%. The total volume of circulating alkaline solution was over
day x þ 1  Gluing of the datum discs 25 L and the flow rate over 2 L/h.
day x þ 2  Returning specimen to the alkaline bath
/ Curing in alkaline solution at room temperature (5 days)
3.4. Mechanical test setups
day x þ 7  Heating up to 60  C
day x þ 8  Zero-reading In order to assess the mechanical properties of the control and
/ Free ASR expansion (30 weeks) reactive concretes, a number of uniaxial compression tests (UCTs)
day  End of test
x þ 217
and wedge splitting tests (WSTs) were performed. The WST pro-
cedure is described in detail in Ref. [12].
For the UCTs, the original cubic specimens of 150
150
150
mm were cut with a diamond saw into four prismatic specimens of
3.3. Confined expansion test setup
approximately 75
75
150 mm. The prismatic specimens were
cut in such a way that two of them were aligned with the casting
The steps followed to perform a confined expansion test using
direction and the other two were normal to it. To one specimen of
the AARTM machine are schematically summarized in Table 3. First,
each pair, two LVDTs (RDP GT2500, range ±2.5 mm, linearity error
the specimen was removed from the alkaline bath and let dry
< ±0:1 % F.S.) were attached in the loading direction for measuring
exposed to air for 24 h. Then, a thin layer of silicone sealant was
the material strain. The UCTs were performed with a controlled
applied at the rims of each face of the specimen (20 mm inwards)
displacement rate of 0.25 mm/min. On the specimens with
and was let cure for 48 h. This silicone layer improves the sealing
attached LVDTs, two load/unload cycles up to 1/4 of the estimated
function of the loading plates O-rings by providing them a smooth
compression strength were performed before the final loading until
settlement surface. Afterwards, the specimen was placed in be-
failure. The other two specimens were directly loaded until failure.
tween of the loading plates of the AARTM and isotropically
The compression strength (fc ) of the original cubical specimen is
confined with a small pressure (0.30 MPa, approx.). Immediately, a
obtained as the average of the peak compressive stress of the four
thick layer of silicone sealant was applied covering the specimen-
prismatic specimens. The initial tangent modulus (E) is obtained as
plate contours and was let cure for 72 h, while keeping the
the average of the values obtained from the two prismatic spec-
confining pressure. Once the silicone was cured, the LVDTs cores
imen with attached LVDTs.
and the hoses for liquid circulation were put in position. The 1 m
For the WSTs, the original cubic specimens were cut into two
NaOH solution was allowed to circulate for 48 h at room temper-
prismatic specimens of approximately 75
150
150 mm. Then
ature to re-saturate the specimen. In order to guarantee full contact
the grove and the notch were cut in such a way that the fracture
between the specimen and the plates, previously to heating, the
path of one specimen was aligned with the casting direction, and in
pressure in each direction was raised to 5 MPa and then lowered
the other specimen was normal to it. The crack opening displace-
again to 0.30 MPa, while maintaining the other two at 1 MPa. Next,
ment (COD) was measured along the axis of the horizontal splitting
the pressure was set again to isotropic 0.30 MPa and the heating
force with two LVDTs (RDP GT2500, range ±2:5 mm, linearity error
systems of the plates and alkaline solution tank were turned on.
< ±0:1 % F.S.), one on each side of the specimen. The applied ver-
The target temperature of 60  C was usually reached within an
tical load was measured with a 25 kN load cell (UtilCell 610, line-
hour, moment in which the specimen was loaded to the desired
arity error < ±0:25 % F.S.). The COD rate was 0.125 mm/min. The
stress state and the data acquisition started. These stress state and
specific fracture energy of the original cubical specimen in mode I
(GfI ) is obtained as the average of the values obtained from the two
prismatic specimens tested. Additionally, the tensile strength (ft ) is
Table 3 estimated via numerical modeling according with the procedure
Schematic timeline of the confined expansion tests. “x” denotes the age (in days) of
described in Ref. [13].
the specimen when removed from the alkaline bath.

day 0  Concrete casting 4. Results and discussion

day 1  Unmolding
/ Curing in alkaline solution at room temperature (between 34 and
68 weeks) 4.1. Free expansion tests

day x  Removal of the specimen from the alkaline bath Three control specimens and four reactive specimens were
 Application of the silicone sealant layer at the rim of each face of the tested in free-expansion conditions. Refer to Table 1 for a list of the
specimen.
tested specimens with their respective curing and testing times.
day x þ 2  Placing of the specimen in the AARTM The strain measured for each specimen is presented in Fig. 6 as the
 Sealing of the contours of plate-specimen contacts average of the strain measured in each of the three directions of the
day x þ 4  Beginning of the circulation of the alkaline solution specimen. The dashed lines indicate the envelope of the measured
day x þ 7  Loading/unloading to 5 MPa in each direction
strain curves.
 Heating up to 60  C
 Application of the desired loading state As it can be seen in the figure, the control specimens developed
 Starting of data acquisition a slight contraction of around 0.05% over the first 75 days and then
/ Confined ASR expansion (21 days) seemed to stabilize. Reactive specimens exhibited a roughly con-
day  End of test stant expansion rate for strains below 1%. Beyond that value, the
x þ 28
expansion rate decreased until reaching an approximately constant
 J. Liaudat et al. / Cement and Concrete Composites 86 (2018) 160e170 165

value of about one fourth of the initial rate. Note that specimen simultaneous action of loading and ASR, are folded into the “ASR
TM01, which was 39 weeks-old when tested, showed a very similar strain”, which will be also dependent on stress as shown in the
behavior to that of the remaining reactive specimens of only 14 following sections. Note that this approach has already been used
weeks. This seems to indicate that, at least within 14e39 weeks, the in the literature [7e9] and it is conceptually similar for instance to
extent of the curing period does not affect the subsequent devel- the standardized procedure for the separation of creep and
opment of the ASR expansions. However, a systematic study would shrinkage strains.
be needed to confirm this particular aspect. This issue is of interest
since the specimens tested with the AARTM were of different ages, 4.2.1. Strain curves of control specimens
as it is indicated in Table 1. The axial strain curves obtained from control specimens are
given in Fig. 7aec. All the specimens experienced a monotonous
contraction with decreasing rate throughout the test. Test on
4.2. Confined expansion results specimen TM03 (Fig. 7a) was ended prematurely due to a leak of
alkaline solution.
Three nominal load cases were studied with the AARTM: In these plots we may distinguish two different kinds of scatter.
On one hand, there are differences between curves of different axes
 “Case 1-1-1”, sx ¼ sy ¼ sz ¼ 1 MPa of the same specimen that should, in principle, be equal since they
 “Case 9-9-9”, sx ¼ sy ¼ sz ¼ 9 MPa were subject to the same load. See for instance the curves of
 “Case 9-9-1”, sx ¼ sy ¼ 9 MPa, sz ¼ 1 MPa specimen TM19 (Fig. 7a), which was subject to the 1-1-1 loading
and, nevertheless, the strain curves exhibit non-negligible differ-
In all cases the casting direction of the specimens was matched ences. This scatter seems to be random and intrinsic to the creep
with the Z-direction of the AARTM. Per each load case, two control phenomenon itself and/or to the testing procedure.
and two reactive specimens were tested. The list of the tested On the other hand, there are differences within the curves of the
specimens with their respective curing and testing times is given in two specimens tested with the same load case. For instance,
Table 1. specimens from the casting set #3 (TM19 and TM15) experienced
The reason for the control specimens is because concrete subject higher contractions than specimens from casting sets #1 (TM03)
to sustained load will inevitably show creep (and possibly cracking) and #2 (TM08). This seems to indicate that concrete form casting
deformations, and therefore the strains measured for reactive set #3 had different creep properties than the ones from casting
specimens under loading will be the combined effect of both ac- sets #1 and #2. Based on this assumption, it was decided that the
tions. Additionally, one can expect that the principle of super- control specimen used for creep deduction of a given reactive
position will not be strictly satisfied, i.e. the effect of simultaneous specimen had to be from the same casting set. For instance, creep of
loading and ASR will not be equal to the sum of ASR strains of reactive specimen TM02 was deducted considering the strain
unloaded specimens plus mechanical deformations of similar non- curves of control specimen TM03, because both belonged to the
reactive specimens. In this context, the approach followed is to same casting set (see Table 1).
accept the following conventional definitions:
4.2.2. Strain curves of reactive specimens
 Basic creep: time-dependent deformations measured on control The axial strain curves of reactive specimens, after deducting
(non-reactive) specimens under load. creep, are plotted in Fig. 7def. As already explained, the creep
 ASR strain: difference between total deformations measured on deduction consisted in directly subtracting, for each load case, the
reactive specimens under load and basic creep. strains measured in the control specimens from those measured in
the reactive ones. To do so, firstly the control specimen strain
With this convention, possible interaction effects of the curves were fitted with a proposed logarithmic function. Then, the
fitted creep function was used to estimate the creep strains to be
deducted from the measured total strains in the reactive speci-
3.5 mens. The creep function adopted and the procedure followed to fit
TM01
TM10 the function parameters are described in detail in Appendix A. It
3.0 TM12 must be noted that within the term creep we are denoting all the
TM13 ReacƟve specimens time dependent strains measured on control specimens, including
2.5 TM05 concrete creep, but probably also the deformation of the silicone
TM06
layers and the O-rings in between the specimen faces and the
TM07
2.0 machine loading plates.
As shown in Fig. 7def, the differences between curves of spec-
Strain [ ]

1.5 imens tested with the same load case are within the expected
scatter for ASR expansion tests. In all cases, the maximum expan-
sion rates needed between 2 and 4 days to be reached. For applied
1.0
stress of 9 MPa, the strain curves in the corresponding direction,
exhibited a slight contraction during the first day that is eventually
0.5 overcome by expansions (Fig. 7e and f). This seems to indicate that
Control specimens the creep strains measured on control specimens were smaller than
0.0 those occurring in the corresponding reactive ones. However, this
0 50 100 150 200 250 difference cannot be attributed to ASR since it is observed at the
0.5 very beginning of the tests when the effect of ASR may be regarded
Time [days] as insignificant. Then, it seems more likely to be the result of dif-
Fig. 6. Averaged axial strain curves in free-expansion conditions of both reactive and
ferences in the mechanical/creep parameters of control and reac-
control specimens. Dashed lines indicate the envelope of all the measured strain tive concrete, as it is discussed in Section 4.4.
curves in the three directions. In Fig. 8, the average of the volumetric strain curves obtained for
166 J. Liaudat et al. / Cement and Concrete Composites 86 (2018) 160e170

(a) 0.02 (d) 0.60

X (TM02)
X (TM21)
0.00 0.50 Y (TM02)
Y (TM21)
Z (TM02)
−0.02 0.40 Z (TM21)

strain [%]
strain [%]

−0.04 0.30

−0.06 0.20

−0.08 X (TM03) 0.10

X (TM19)
Y (TM03)
−0.10 Y (TM19) 0.00
Z (TM03)
Z (TM19)

−0.12 −0.10
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
time [days] time [days]

(b) 0.02 (e) 0.60

X (TM09) X (TM14)
X (TM16) X (TM22)
0.00 Y (TM09) 0.50 Y (TM14)
Y (TM16) Y (TM22)
Z (TM09) Z (TM14)
−0.02 Z (TM16) 0.40 Z (TM22)
strain [%]
strain [%]

−0.04 0.30

−0.06 0.20

−0.08 0.10

−0.10 0.00

−0.12 −0.10
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
time [days] time [days]

(c) 0.02 (f) 0.60

X (TM11)
X (TM20)
0.00 0.50 Y (TM11)
Y (TM20)
Z (TM11)
−0.02 0.40 Z (TM20)
strain [%]

strain [%]

−0.04 0.30

−0.06 0.20

−0.08 X (TM08) 0.10

X (TM15)
Y (TM08)
−0.10 Y (TM15) 0.00
Z (TM08)
Z (TM15)

−0.12 −0.10
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
time [days] time [days]

Fig. 7. Axial strain curves of control specimens under (a) 1-1-1, (b) 9-9-9, and (c) 9-9-1 load cases; axial strain curves, after deducting creep, of reactive specimens under (d) 1-1-1,
(e) 9-9-9, and (f) 9-9-1 load cases.

each load case, after deducting creep, are plotted together, measured only three times in the first 21 days.
including the free expansion case (0-0-0). The bi-linear shape of the For a quantitative analysis of the effect of stresses, axial and
free expansion curve is because the specimen deformations were volumetric expansion rates obtained from the curves plotted in
 J. Liaudat et al. / Cement and Concrete Composites 86 (2018) 160e170 167

Fig. 7d to e are summarized in Table 4. The expansion rates were Table 4

obtained by fitting the expansion curve points ðt; εasr Axial and volumetric ASR expansion rates expressed in ½106 h1 . Creep has been
i Þ, with a deducted for primary load cases but not for secondary load cases (marked with “*”).
straight line between t ¼ 4 and t ¼ 21 days for the specimens Values calculated with the proposed model are given in between brackets.
tested in the AARTM, and between t ¼ 0 and t ¼ 21 days for the free
expansion tests. The volumetric expansion rate was obtained as the Id. ε_ asr
x ε_ asr
y ε_ asr
z ε_ asr
v ε_ asr asr
x =_εv ε_ asr asr
y =_εv ε_ asr asr
z =_εv

sum of the three axial expansion rates. In Fig. 9 (primary load Load Case 0-0-0
states), volumetric expansion rates are plotted against the applied TM01 5.48 7.21 6.61 19.29 0.28 0.37 0.34
volumetric stress. TM10 9.17 6.99 11.10 27.26 0.34 0.26 0.41
The volumetric expansion rate was practically the same for load TM12 7.42 5.88 9.52 22.82 0.33 0.26 0.42
cases 0-0-0 and 1-1-1. However, for load cases 9-9-1 and 9-9-9 the TM13 9.45 9.94 9.93 29.32 0.32 0.34 0.34
Average 7.88 7.50 9.29 24.67 0.32 0.31 0.38
volumetric expansions rate was reduced significantly as the mean
Model (8.22) (8.22) (8.22) (24.67) (0.33) (0.33) (0.33)
compression stress was increased. Load Case 1-1-1
Under isotropic stress states (0-0-0, 1-1-1 and 9-9-9) the axial
TM02 7.27 7.54 8.61 23.41 0.31 0.32 0.37
expansion rate in the casting direction (Z-direction) was greater TM21 8.47 8.43 9.09 26.00 0.33 0.32 0.35
than that of the other two directions of the same specimen, Average 7.87 7.99 8.85 24.71 0.32 0.32 0.36
therefore possibly revealing a constitutive anisotropy of the spec- Model (8.14) (8.14) (8.14) (24.41) (0.33) (0.33) (0.33)
imens. This phenomenon has been observed by many researchers, Load Case 9-9-1
e.g. in Refs. [7,8]. The rate of ASR expansion in Z-direction measured TM11 1.88 1.16 12.29 15.33 0.12 0.08 0.80
for the load case 9-9-1 was greater than the one measured for load TM20 1.27 0.93 10.80 13.00 0.10 0.07 0.83
case 1-1-1, while in the other two directions the expansion rates Average 1.58 1.04 11.55 14.17 0.11 0.07 0.82
Model (1.13) (1.13) (11.89) (14.15) (0.08) (0.08) (0.84)
were similar to those obtained for load case 9-9-9. Somehow, the Load Case 9-9-9
confinement applied on X- and Y-directions increased the expan-
TM14 1.04 1.06 1.32 3.41 0.30 0.31 0.39
sion rate in the less stressed Z-direction.
TM22 1.00 0.95 1.26 3.20 0.31 0.30 0.39
Based on the same concepts as proposed in Saouma and Perroti Average 1.02 1.00 1.29 3.31 0.31 0.30 0.39
[14], a new formula is proposed to predict the observed ASR Model (1.14) (1.14) (1.14) (3.43) (0.33) (0.33) (0.33)
volumetric expansion rate under compressive stresses (_εasr v ):
Load Case 1-1-1 after 9-9-9 (*)

TM14 10.70 9.94 11.10 31.74 0.34 0.31 0.35

v ¼ Gc ðsv Þ_εv;free
ε_ asr asr Model (8.06) (8.06) (8.06) (24.41) (0.33) (0.33) (0.33)
(1)
Load Case 1-9-1 after 1-1-1 (*)

where ε_ asr TM21 7.72 0.42 7.18 15.32 0.50 0.03 0.47
v;free is the rate of volumetric ASR expansion without Model (10.15) (0.85) (10.15) (21.15) (0.48) (0.04) (0.48)
applied stresses, and Gc is a dimensionless factor that accounts for
the reduction of ε_ asr
v under compressive stresses, which is assumed
to be a function of the volumetric stress sv as follows
35
8 Model
< 1 if sv  0 Primary load state
Gc ¼ 1  ðsv =sv Þ 2
if 0 > sv  sv (2) 30 Secondary load state
:
0 if sv < sv
vol. strain rate [ 10 6/h]

25
where sv is a parameter that stands for the volumetric stress under
which ASR expansion would be totally suppressed. This value is 20
taken as 9:7 MPa (in contrast with Saouma and Perotti expression
in which sv was made equal to the characteristic compression 15

1.20 10
0−0−0
1−1−1
1.00 9−9−1
5
9−9−9

0.80 0
10 9 8 7 6 5 4 3 2 1 0
vol. strain [%]

v [MPa]
0.60
Fig. 9. Volumetric strain rate of tested reactive specimen versus applied volumetric
stress. Creep effects have only been deducted from the specimens under primary load
0.40
states.

0.20
strength of concrete). The volumetric stress is defined as sv ¼ ðs1 þ
0.00 s2 þ s3 Þ=3 where ðs1 ; s2 ; s3 Þ are the principal stresses.
Then, the reduced volumetric rate is distributed in the three
principal directions by means of the formula given in Saouma and
−0.20
0 2 4 6 8 10 12 14 16 18 20 Perotti [14] as follows:
time [days]
ε_ asr
i ¼ Wi ðs1 ; s2 ; s3 Þ_εasr
v (3)
Fig. 8. Comparison of average volumetric strain curves, after deducting creep, of
reactive specimens under different triaxial confinement.
168 J. Liaudat et al. / Cement and Concrete Composites 86 (2018) 160e170

where ε_ asr
i is the ASR expansion rate in the principal axis i, and Wi is (a) 0.4
a weight factor for the principal axis i, which is a function of the 9−9−9 1−1−1
three principal stresses and is calculated following the procedure
0.02
proposed by Saouma and Perotti [14]. For comparison, the ASR 0.3
expansion rates values obtained using these formulae are informed
0.00
in between brackets in Table 4. As can be seen, a fairly good
agreement is obtained. It is remarkable that the weight factors
0.2 −0.02

strain [%]
proposed by Saouma and Perotti based on the experimental results
of Multon and Toutlemonde [8] are in very good agreement with −0.04
the ones resulting from our tests, despite the fact that the consti-
0.1 19 20 21 22 23
tutive anisotropy of the specimens in the casting direction (Z-di-
rection) is not reproduced.
It has to be recognized that the model represented by Eqs.
0.0
(1)e(3) is a drastic simplification of a complex phenomenon X
involving cracking, creep and chemo-transport processes at various Y
Z
levels of observation. Therefore, the curves and parameters values
obtained have a phenomenological character and may change for
−0.1
0 5 10 15 20 25 30 35
different concretes, exposure conditions, etc. Nevertheless, it is
time [days]
remarkable that, as said before, this simple approach can fit data
from at least two independent test series.
(b) 0.7
4.3. Effect of changing the stress state X 1−1−1 1−9−1
Y
0.6
Z
In order to have some insight into the effect of changing the
stress states on the ASR expansion rates, after the standard testing 0.5
time (21 days) had been completed, the stress states of two spec-
imens (TM14 and TM21) were changed and the tests continued for 0.4
strain [%]

an additional period of approximately 12 days.

After the 21 days under 9-9-9 load case, the stresses on spec- 0.3
imen TM14 were reduced to 1-1-1. As a result, the expansion rate 0.44
0.2
increases significantly as can be appreciated in Fig. 10a. Since the
counterpart control specimen was not subject to the same change 0.40
0.1
of the stress state, the curves presented in Fig. 10a are the raw ones, 0.36
i.e. without any creep deduction. The expansion rates were 0.0
maximum immediately after the stress change, decreasing 0.32
20 22 24 26
asymptotically to constant rate values in the following days. −0.1
Interestingly, no instantaneous (elastic) expansions associated with 0 5 10 15 20 25 30 35
the reduction of the compression stresses can be observed in the time [days]
curves. The asymptotic expansion rates, presented in Table 4 and
Fig. 9 (secondary load states), were obtained by fitting the expan- Fig. 10. Axial strain curves (without deducting any creep effects) of: (a) reactive
specimen TM14, first subjected to load case 9-9-9 for 21 days and then to load case 1-
sion curve points ðt; εi Þ with a straight line between t ¼ 25 and t ¼
1-1 until the end of the test; (b) reactive specimen TM21, first subjected to load case 1-
32 days. These values are 30% greater than the ASR expansion rates 1-1 for 22 days and then to load case 1-9-1 until the end of the test.
obtained for load case 1-1-1 (Table 4). Note that the expansion rates
obtained for the secondary load cases are not only attributable to
ASR since strains due to other causes have not been deducted. In control specimen with the same stress history makes it impossible
this case, the decompression of the specimen may have originated a to deduct creep and therefore obtain the net ASR expansion rates.
differed elastic creep expansion which could explain, at least in However, one can estimate that the deduction of the creep strains
part, the initial variation of the expansion rates. Other causes in this case would result in a slight increase of the expansion rate in
related with the ASR expansion mechanisms may also be consid- the Y-direction and an even smaller, if any, reduction in the X- and
ered. For instance, the greater initial expansion rate may be Z-direction rates, due to Poisson's effect on creep strains. Therefore,
attributed to the swelling of ASR gel formed under the 9-9-9 stress the increase of the Y-stress seems not only to have reduced the Y-
state due to water absorption when the confining stresses reduce to direction expansion rate but also the expansions on the other two
1-1-1. In any case, additional testing would be necessary to explain axes. Again, additional testing would be needed to confirm these
in more detail the observed behavior. observations.
For specimen TM21, the stress state was changed after 22 days Finally, the expressions proposed in Eqs. (1)e(3) were used to
from 1-1-1 to 1-9-1. The resulting raw expansion curves are pre- estimate the ASR expansion rates under the secondary load cases.
sented in Fig. 10b. In this case, in contrast with the previous one, The values obtained are reported in between brackets in Table 4. As
instantaneous strains are clearly appreciated: a negative strain it can be seen, the weight factors are in good agreement with the
jump in the Y-direction and smaller positive strain jumps, attrib- experimental ones. However, the volumetric expansion rates are
utable to Poisson's effect, in the X- and Z-directions. After that, the not very well predicted.
expansion in the Y-direction practically stops, while in the X- and Z-
directions it continues but with a rate 10e20% smaller than for the
previous 1-1-1 stress state. In the same way as for the previous 4.4. Mechanical test results
specimen, the expansion rates for the 1-9-1 stress state has been
estimated and reported in Table 4. Again, the lack of a counterpart The results obtained are summarized in Table 5. The initial
 J. Liaudat et al. / Cement and Concrete Composites 86 (2018) 160e170 169

elastic modulus (E) and the compression strength (fc ) were under the same conditions, one reactive, with crushed soda-lime
measured on control and reactive specimens subjected to four glass aggregate, and one non-reactive.
different histories of curing and ASR testing. The specific fracture For the triaxially-confined tests, an ad hoc testing machine for
energy in mode I (GfI ) and the tensile strength (ft ) were obtained AAR confined expansion tests was designed and constructed. This
only from control and reactive specimens which had been cured for machine has shown to be capable of accurately apply and maintain
15 weeks but had not been subject to ASR testing. targeted true-triaxial compressive stress states and temperature on
Lets first consider the results from the specimens cured for 15 the tested specimen, while keeping it in contact with a highly
weeks and without ASR testing. The E and fc measured for control alkaline solution.
concrete are significantly higher than those measured on reactive The results seem to indicate that the volumetric ASR expansion
concrete. In contrast, the GfI measured on reactive concrete is rate is reduced as the applied volumetric compressive stress is
practically twice the one measured on control concrete, while the increased. Additionally, there seems to be an increase of the
estimated ft values are similar. These differences can be attributed expansion rate in the less compressed direction in detriment of the
to the flaky shape and to the smoothness of the crushed glass expansion rates in the most compressed ones. These observations
particles. Since concrete creep in absolute terms is known to are in good agreement with the model previously proposed by the
decrease with increasing elastic modulus, the creep strain experi- fourth author [14], although a correction is proposed for the de-
enced by control specimens during the confined expansion tests pendency of total volumetric expansion with volumetric stress.
may have been lower than the ones experienced by reactive Future work is intended to use the presented testing method-
specimens. This may explain the initial contraction observed in the ology for studying concrete made with natural reactive aggregates.
ASR expansion curves in the direction under 9 MPa stress (Fig. 7e
and f), and indicate that the actual ASR expansion rates may have
been slightly higher than those reported in Table 4. An effort shall Acknowledgements
be made in future tests in order to obtain similar mechanical pa-
rameters in control and reactive concretes. This research is supported by grants BIA2016-76543-R from
When considering the remaining results, it seems that curing MEC (Madrid), which includes FEDER funds, and 2014SGR-1523
and testing history only have an influence on the measured values from AGAUR-Generalitat de Catalunya (Barcelona). The fourth
of fc and E on reactive concrete but not on control concrete. Com- author was funded by the US Nuclear Regulatory Commission
parison of results from specimens TM30 and TM24 (curing times of through Grant NRC-HQ-60-14-G-0010 to the University of Colorado
15 and 47 weeks, respectively) suggests that ASR degradation (Boulder) to perform some of the studies. Initial funding for the
occurring at curing conditions was not significant, even when some AARTM was provided by the Tokyo Electric Power Company to the
signals of occurring ASR (ASR gel spots, slight cracking) were fourth author. The first author thanks the scholarship FPI (BES-
detected on the specimen cured for 47 weeks. In contrast, E and fc 2010-030515) from MEC (Madrid).
values from specimen TM13, which was previously tested in free
expansion conditions for 36 weeks and experienced ASR expan-
sions of about 2.5% (Fig. 6), showed important reductions in com- Appendix A. Creep deduction
parison with those measured on TM30. Finally, specimen TM22,
which had a longer curing period and was tested for 3 weeks The simplest possible creep deduction procedure was based in
experiencing some ASR expansions (0.05%, Fig. 7e), not only did not three assumptions. The first one was that the measured strains of
 
degrade its mechanical parameters but actually improved them. the reactive specimens erea ¼ εreax εrea
y εrea
z were the addition
of two independent terms, on one hand the ASR expansion term
5. Concluding remarks (easr ) and on the other the “creep” strain term ecr , i.e.

Accelerated ASR expansion tests performed on cubical concrete erea ¼ easr þ ecr (A.1)
specimens under free and triaxially-confined expansion conditions The second assumption was that the creep strains in the reactive
have been presented. Two different concrete mixtures were tested specimens were the same as those measured in the control speci-
mens (econ ) under the same conditions, i.e.
Table 5
Results of the performed mechanical tests: initial tangent modulus (E), compression
ecr ¼ econ (A.2)
strength (fc ), tensile strength (ft ), and specific fracture energy in mode I (GfI ).
Finally, the third assumption was that the strains developed by
Type Id. E [GPa] fc ½MPa ft ½MPa f
GI ½J=m2  the control specimens subject to a constant stress state can be
properly approximated with a function Ci ðtÞ which is fitted inde-
Cured for 15 weeks
pendently for each axis of loading i,
Control TM25 41.5 42.5 e e

Reactive TM30 21.5 26.8 e e 0 if t  ti
Control TM26 e e 2.8 74.6 Ci ðtÞ ¼ (A.3)
mi lnt þ bi if t > ti
Reactive TM31 e e 2.4 144.1
Cured for 47 weeks
where t is the time elapsed since the moment when the specimen is
Control TM18 40.0 60.5 e e
Reactive TM24 19.0 24.9 e e
loaded (beginning of the test), ti ¼ expðbi =mi Þ, bi and mi are
Cured for 63e67 weeks and tested 9-9-9 for 3 weeks fitting parameters. Best fitting values of mi and bi were obtained for
each tested control specimen by applying the least squares method.
Control TM16 40.3 54.3 e e
Reactive TM22 25.5 34.4 e e Under these three assumptions, the ASR expansion curves for a
Cured for 14 weeks and tested 0-0-0 for 36 weeks given stress state was readily obtained as:
Control TM07 41.7 52.6 e e
Reactive TM13 5.6 8.5 e e
easr ¼ erea  CðtÞ (A.4)
170 J. Liaudat et al. / Cement and Concrete Composites 86 (2018) 160e170

References

PhD thesis, in French, Ecole Nationale des Ponts et Chausse
es, 1998.
[8] S. Multon, F. Toutlemonde, Effect of applied stresses on alkali-silica reaction-
induced expansions, Cem. Concr. Res. 36 (2006) 912e920.
[1] J.K. McGowan, H.E. Vivian, The effect of superincumbent load on mortar bar
[9] C.F. Dunant, K.L. Scrivener, Effects of uniaxial stress on alkali-silica reaction
expansion, Aust. J. Appl. Sci. 6 (1955) 94e99.
induced expansion of concrete, Cem. Concr. Res. 42 (2012) 567e576.
[2] C.F. Ferraris, J.R. Clifton, E.J. Garboczi, F.L. Davis, Stress due to alkali-silica re-
[10] B.P. Gautam, D.K. Panesar, A new method of applying long-term multiaxial
action in mortars, in: K.L. Scrivener, J.F. Young (Eds.), Mechanisms of Chemical
stresses in concrete specimens undergoing ASR, and their triaxial expansions,
Degradation of Cement-Based Systems, E. & F.N. Spon, London, 1995,
Mater. Struct. 49 (2016) 3495e3508.
pp. 75e82.
[11] AENOR, Cement-part 1: Composition, Specifications and Conformity Criteria
[3] C.F. Ferraris, E.J. Garboczi, F.L. Davis, J.R. Clifton, The effect of stress relaxation,
for Common Cements UNE-EN 197-1, AENOR, Madrid, 2011.
self-desiccation, and water absorption on the alkali-silica reaction in low
[12] E. Brühwiler, F. Wittmann, The wedge splitting test, a new method of per-
water/cement ratio mortars, Cem. Concr. Res. 27 (1997) 1553e1560.
forming stable fracture mechanics tests, Eng. Fract. Mech. 35 (1990) 117e125.
[4] M. Kawamura, K. Iwahori, ASR gel composition and expansive pressure in
[13] J. Liaudat, D. Garolera, A. Martínez, I. Carol, M.R. Lakshmikantha, J. Alvarellos,
mortars under restraint, Cem. Concr. Compos 26 (2004) 47e56.
Numerical modelling of the Wedge Splitting Test in rock specimens, using
[5] A. Binal, The determination of gel swelling pressure of reactive aggregates by
fracture-based zero-thickness interface elements, in: E. On ~ ate, D.R.J. Owen,
ASGPM device and a new reactive-innocuous aggregate decision chart, Constr.
D. Peric, M. Chiumeti (Eds.), XIII International Conference on Computational
Build. Mater 22 (2008) 1e13.
Plasticity - Fundamentals and Applications, CIMNE, Barcelona, 2015,
[6] H. Kagimoto, Y. Yasuda, M. Kawamura, ASR expansion, expansive pressure
pp. 974e981.
and cracking in concrete prisms under various degrees of restraint, Cem.
[14] V.E. Saouma, L. Perotti, Constitutive model for alkali-aggregate reactions, ACI
Concr. Res. 59 (2014) 1e15.
Mater. J. 103 (2006) 194e202.
[7] C. Larive, Combined Contribution of Experiments and Modelling to the Un-
derstanding of Alkali-aggregate Reaction and its Mechanical Consequences,