ACI MATERIALS JOURNAL TECHNICAL PAPER

Title no. 103-M22

Constitutive Model for Alkali-Aggregate Reactions

by Victor Saouma and Luigi Perotti

A new constitutive model for alkali-aggregate reaction (AAR) either through large compressive stresses, and/or more
expansion is presented. This thermo-chemo-mechanical model is dramatically through the formation of structural cracks or the
rooted in the chemistry, physics, and mechanics of concrete. The sliding across critical joints. Hence, the structural integrity of
major premises of the model are the assumption of a volumetric the structure can certainly be seriously jeopardized by the
expansion of the gel and redistribution on the basis of weights pernicious and slow evolution of the reaction.
related to the stress tensor (hence induced anisotropy). This three-
component model is, for the most part, loosely coupled, with the
exception of the interdependency between the mechanical and the RESEARCH SIGNIFICANCE
chemical parts through the kinetics of the reaction. The model has Many concrete dams worldwide are affected by AARs,
been used, in conjunction with a formal parameter identification and despite much research, there is still a dichotomy between
paradigm, to analyze laboratory tests on triaxially confined models and applications. Models tend to be too narrowly
concrete cylinders. Finally, a detailed two-dimensional analysis of defined and are seldom applied to actual structures where all
an arch gravity dam is presented.
the complexity of the load is accounted for.
The proposed model is comprehensive; it is rooted in the
Keywords: alkali-aggregate reaction; alkali-silica reaction; dams; model.
chemistry, physics, and mechanics of AAR and derives much of
its parameters from recent experimental tests performed at the
INTRODUCTION Laboratoire Centrale des Ponts et Chausses (LCPC), France.
Alkali-aggregate reaction (AAR), which includes alkali-
A peculiarity of AAR in dams is that there are field
silica reaction (ASR), is the leading cause of dam concrete
measurements of the irreversible crest displacement that in
deterioration. This slow-evolving internal concrete damage is
theory should be matched by the numerical model. So far,
causing millions of dollars of damage worldwide, and whereas
this has been an ad-hoc process through manual fine-
there is no (economically) feasible method to stop the reac-
tuning. A more rational approach is hereby presented, one in
tion, it can be mitigated to some extent. This has been accom-
which parameter identification (AAR expansion properties)
plished primarily through an expensive slicing of the dam to
is the result of a formal minimization procedure.
relieve the reaction-induced compressive stresses. Hence,
given the need to plan this complex mitigation procedure, and
keeping in mind that in some drastic cases the dam may have LITERATURE SURVEY
to be decommissioned, there is an urgent need to provide the AAR was first identified by Stanton (1940) as a cause for
engineering profession with solid, sound, and practical predic- concrete deterioration. However, there were few initial related
tive tools for the dam structural response evolution. papers. Probably triggered by an ever-increasing manifestation
ASR in concrete is a chemical reaction involving alkali of the reaction in major structures, there has been recently
cations and hydroxyl ions from concrete pore solutions, and numerous investigations on AAR. In the context of the
certain metastable or strained forms of silica present within presented work, only few related works will be examined.
aggregate particles. This chemical reaction will produce More information can be found in Saouma and Xi (2004).
ASR gel that swells with the absorption of moisture. Hence, One of the most extensive and rigorous investigation of
in a simplified manner, ASR can be described as a two-step AAR has been conducted by Larive (1998) who tested more
reaction between alkalis (sodium and potassium) in concrete than 600 specimens with various mixtures and ambient and
and silica reactive aggregates. The first step is the chemical mechanical conditions. Not only did the author conduct this
reaction between the reactive silica in the aggregate with the extensive experimental investigation, but a numerical model
alkali present in concrete to produce alkali-silica gel has also been proposed for the time expansion of the
concrete. In particular, a thermodynamically based model
Reactive silica in aggregate + alkali in concrete (1) for the expansion evolution was developed, and was then
alkali-silica gel calibrated with the experimental data (Fig. 1).
[xSiO2] + [yNa(K)OH] [Na(K)ySixOzaq]
t
--------------
C ( )
The second step is the expansion of the alkali-silica gel 1e
( t, ) = ---------------------------------------
t ( , I , f )
- (3)
when it comes in contact with moisture L c
-----------------------------------
-
C ( )
1+e
Alkali-silica gel + moisture expanded alkali-silica gel
[Na(K)ySixOzaq] + [H2O] [Na(K)ySixOzwH2O] (2)
ACI Materials Journal, V. 103, No. 3, May-June 2006.
MS No. 04-411 received July 11, 2005, and reviewed under Institute publication
It is precisely this second reaction that causes the well- policies. Copyright 2006, American Concrete Institute. All rights reserved, including
known swelling of the concrete, resulting in major internal the making of copies unless permission is obtained from the copyright proprietors.
Pertinent discussion including authors closure, if any, will be published in the March-
stress redistribution inside the dam that manifests itself April 2007 ACI Materials Journal if the discussion is received by December 1, 2006.

194 ACI Materials Journal/May-June 2006

Victor Saouma is a Professor of Civil Engineering at the University of Colorado,
Boulder, Colo., and is Director of the NEES-NSF Fast Hybrid Testing Laboratory. His
research interests include dam engineering, earthquake engineering, concrete
deterioration modeling, and fracture mechanics.

Luigi Perotti is a PhD Candidate at the California Institute of Technology, Pasadena, Calif.
He received his MS from the Department of Structural Engineering of the Technical
Institute of Milan. His research interests include theoretical and computational mechanics.

where L and C are the latency and characteristic times,

respectively. The first corresponds to the inflexion point, and
the second is defined in terms of the intersection of the
tangent at L with the asymptotic unit value of . In a subsequent
work, Ulmet al. (2000) have shown the thermal dependency
AAR
of those two coefficients Fig. 1Normalized expansion curve ( ( t ) = F, Vol( t ) AAR ) .

C ( ) = C ( 0 ) exp U C --- -----
1 1
0
(4)
1 1-
L ( , I , f c ) = f ( I , f c ) L ( 0 ) exp U L --- ----
0

expressed in terms of the absolute temperature ( K = 273

+ T C) and the corresponding activation energies. The vari-
ables UL and UC are the activation energies minimum energy
required to trigger the reaction for the latency and characteristic
times, respectively, and were determined (for Larives test) to be

UL = 9400 500K (5) Fig. 2Effects of temperature on AAR expansion.

UC = 5400 500K (6) degradation observed in laboratory specimens is often much

higher than the one recorded in the field.
To the best of the authors knowledge, the only other tests Whereas a good model for AAR should start with the gel-
for these values were performed by Scrivener (2005) who induced pressure, this is a notoriously complex problem (due
obtained values within 20% of Larives values, and the to scale) and, in that context, the work of Struble and
dependency on types of aggregates and alkali content of the Diamond (1981a,b) remains most pertinent.
cement has not been investigated. Hence, in the absence of Modeling of AAR expansion has been undertaken by
other tests, those values can also be reasonably considered as various researchers. Broadly speaking, this modeling falls
representative of dam concrete. The temperature dependence is into one of three categories:
highlighted by Fig. 2, where the expansion curve determined 1. Micro models: in which aggregate and cement paste are
in the laboratory at 38 C is compared with the corresponding separately modeled and the transport equation is used to
one at a dam average temperature of 7 C. model gel formation through a two-stage process (Suwito et
Beside temperature, other parameters strongly affecting al. 2002; Lemarchand and Dormieux 2000). While essential
AAR expansion are humidity and confining stresses. to properly understand the underlying phenomenon causing
Most recently, Multon (2003) tested AAR expansion AAR, this level of modeling is of little relevance to the
under triaxial constraint. Axial traction was applied along structural analysis of AAR-affected structures, as emphasis
one direction of concrete cylinders constrained in the radial is on the transport equation for the reactants.
directions by steel cylinders. As reported first by Larive 2. Meso models: where emphasis is on the determination
(1998) (uniaxial confinement) and later confirmed by of pessimum size effect (Furusawa et al. 1994; Baant and
Multon et al. (2004) (for triaxial confinement), there is Steffens 2000).
strong evidence of an expansion transfer such that the total 3. Macro models: where one stays clear from the transport
volumetric AAR-induced strain is almost constant, irrespective modeling, and emphasis is on a global numerical model for
of the confinement. In other words, the expansion is largest the analysis of a structure. Some of the models fully
in the direction of least resistance. In uniaxially or biaxially decouple structural modeling from the reaction kinetics, and
loaded cylinders, this results in substantially reduced expansion others couple those two effects (and some ignore the kinetics
in the loaded directions and increased expansion in the all together). One of the earliest models is the one of
unconstrained ones. On the other hand, under compressive Charlwood et al. (1992) and Thompson et al. (1994), who
triaxial confinement, there is nearly equal expansion in all identified critical issues related to AAR, namely, the stress
three directions; however, the total volumetric expansion is dependencythat is, there is no AAR expansion under a
slightly reduced. Finally, there are strong indications that compressive stress of approximately 8 MPa, and that the
high compressive hydrostatic stresses retard the reaction. expansion is akin to a thermal one. Subsequently, more
Accompanying AAR expansion, there is often degradation refined models have been proposed by Lger et al. (1996)
in tensile strength and elastic modulus (Swamy and Al-Asali and Huang and Pietruszczak (1999), which focus on the
1988). One should exercise some caution, however, as the kinetics of the reaction, albeit through empirical models.

ACI Materials Journal/May-June 2006 195

 will be accomplished by assigning weights to each of the
three principal directions;
5. Relatively high compressive or tensile stresses inhibit
AAR expansion due to the formation of microcracks or
macrocracks that absorb the expanding gel;
6. High compressive hydrostatic stresses slow down the
reaction;
7. Triaxial compressive state of stress reduces but does not
eliminate expansion; and
8. Accompanying AAR expansion is a reduction in tensile
strength and elastic modulus.
Fig. 3Stress-induced cracks with potential gel absorption Hence, the general (uncoupled) equation for the incremental
(Scrivener 2003). free volumetric AAR strain is given by
AAR
Models that address both the kinetics and the mechanical vol ( t ) = t ( f t, 1 COD ) c ( , f c )f ( h ) ( t, ) = 0 (7)
model of AAR have been proposed by Bournazel and
Moranville (1997), Capra and Bournazel (1998), Capra and where COD is the crack opening displacement; (t, ) is a
Sellier (2003), Ulm et al. (2000) and Li and Coussy (2002). sigmoid curve expressing the volumetric expansion in time
It is worth noting that the kinetics model (built into a coupled as a function of temperature and is given by Eq. (3); is the
thermo-chemo-mechanical one) of Ulm et al. (2000) (based laboratory-determined (or predicted) maximum free volumetric
on the work of Larive [1998]) departs from other empirical expansion at the reference temperature 0 (Fig. 1).
models and is probably the most scientifically correct one. It The retardation effect of the hydrostatic compressive
is the one adopted in this work. Bangert and Meschken stress manifests itself through L. Hence, Eq. (4) is expanded
(2004) recently proposed a coupled model applied to reinforced as follows
concrete, and finally, Farage et al. (2004) seem to have
finally bridged the gap between scientific rigor and practical
L ( , I , f c ) = f ( I , f c ) L ( 0 ) exp U L --- -----
1 1
(8)
applicability to real structures. 0
Numerous dams worldwide have suffered from AAR, in
particular, as reported by Wagner and Newell (1995)
(Fontana Dam, U.S.), Gilks and Curtis (2003) (Mactaquac where
Dam, Canada), Shayan et al. (2000) (Canning Dam, Australia),
Peyras et al. (2003) (Chambon dam, France), Jabarooti and 1 if I > 0

Golabtoonchi (2003) (Iran), and Bon et al. (2001) (Pian f ( I , f c ) = ;I = I + II + III (9)
Telessio Dam, Italy), Portugese National Committee on I
1 + -------- if I 0
Large Dams (2003) (Pracana Dam, Portugal), and Malla and 3f c
Wieland (1999) (a Swiss dam). A comprehensive list of
dams suffering from AAR can be found in Acres (2004). and I is the first invariant of the stress tensor, and fc is the
It is worth noting that, in general, dams built in a (relatively) compressive strength. Based on a careful analysis of Multon
hot climate appear to suffer from AAR at an earlier age than (2003), it was determined that = 4/3. It should be noted that
those built at high altitudes and in colder temperatures. the stress dependency (through I) of the kinetic parameter L
Furthermore, when dam rehabilitation did occur, it included makes the model a truly coupled one between the chemical and
one or more of the following: cutting (to relieve the mechanical phases. Coupling with the thermal component is a
compressive stresses, though accelerating the expansion loose one (hence a thermal analysis can be separately run), 0 <
rate), post-tensioning, or placing an impermeable membrane g(h) 1 is a reduction function to account for humidity given by
(whose benefits are not yet well proven).
g(h) = hm (10)
MODEL
Two different aspects of mathematical modeling of ASR where h is the relative humidity (Capra and Bournazel 1998).
in concrete may be distinguished: 1) the kinetics of the However, one can reasonably assume that (contrary to
chemical reactions and diffusion processes involved, and 2) bridges) inside a dam, g(h) = 1 for all temperatures.
the mechanics of fracture that affects volume expansion and t(ft|wc,1|CODmax) accounts for AAR reduction due to
causes loss of strength, with possible disintegration of the tensile cracking (in which case gel is absorbed by macro-
material (Baant et al. 2000). cracks) (Fig. 3). A hyperbolic decay with a non-zero residual
The proposed model (Saouma and Perotti 2004c) is driven value is adopted (Fig. 4)
by the following considerations:
1. AAR is a volumetric expansion and, as such, cannot be
addressed individually along a principal direction without due 1 if ( l t f t)

regard to what may occur along the other two orthogonal ones; Linear elastic t ft
fictitious

2. Kinetics component is taken from the work of Larive r + ( 1 r ) --------------------- if t f t < l

I
(1998) and Ulm et al. (2000); t = (11)

3. AAR is sufficiently influenced by temperature to 1 if COD max t w c
account for its temporal variation in an analysis; Smeared crack
w
4. AAR expansion is constrained by compression, and is r + ( 1 ) r ------------------- if t w c < COD max
t c

COD max
redirected in other less-constrained principal directions. This

196 ACI Materials Journal/May-June 2006

where t is the fraction of the tensile strength beyond which gel
is absorbed by the crack, and r is a residual AAR retention
factor for AAR under tension. If an elastic model is used, then
ft is the tensile strength, I is the maximum principal tensile
stress. On the other hand, if a smeared crack model is adopted,
then CODmax is the maximum crack opening displacement at
the current Gauss point, and wc is the maximum crack opening
displacement in the tensile softening curve (Wittman et al.
1988). Concrete pores being seldom interconnected, and the
gel viscosity relatively high, gel absorption by the pores is not
explicitly accounted for. Furthermore, gel absorption by the
pores is accounted for by the kinetic equation through the
latency time, which depends on concrete porosity. The higher
the porosity, the larger the latency time is.
In turn, c accounts for the reduction in AAR volumetric
expansion under compressive stresses (in which case gel is
absorbed by diffused microcracks) (Multon 2003)

1 if 0 tension

c = e
(12)
- if > 0 compression
1 ------------------------------- Fig. 4Graphical representation of c and t.

1 + ( e 1 )

I + II + III
= --------------------------------
- (13)
3f c

whereas this expression will also reduce expansion under

uniaxial or biaxial confinement (Fig. 4), these conditions are
more directly accounted for, as follows, through the assignment
of weights.
The third major premise of the model is that the volumetric
AAR strain must be redistributed to the three principal
directions according to their relative propensity for
expansion on the basis of a weight, which is a function of the
Fig. 5Weight of volumetric AAR redistribution in selected
respective stresses. Whereas the determination of the weight is
cases.
relatively straightforward for triaxial AAR expansion
under uniaxial confinement (for which some experimental
data is available), it is more problematic for biaxially or 2. Compression greater than u in one direction, then the
triaxially confined concrete. corresponding weight must be less than 1/2, and the
remaining weight is assigned to the third direction; and
Given the principal stress vector defined by k, l, and m,
3. Compression less to u, then the corresponding weight
weight to each of those three principal directions needs to be
is again zero, and a unit weight is assigned to the third direction.
assigned. These weights will control AAR volumetric
expansion distribution. For instance, with reference to Fig. 5, Triaxial state of stressin which u acts on two of the three
three scenarios are considered. principle directions. The following five cases are identified:
1. Tension along direction k, then all the expansion is
Uniaxial state of stresswhere the following three cases
along k;
are distinguished:
2. Compressive stress greater than u, then we have a
1. In the first case, we have uniaxial tension, and hence, the triaxial state of compressive stress, and the corresponding
volumetric AAR strain is equally redistributed in all three weight will be between 1 and 1/3. The remaining complement
directions; of the weight is equally distributed in the other two directions;
2. Under a compressive stress greater than the limiting on 3. Compression equal to u, hence we have a perfect
(u), the weight in the corresponding (k) direction should triaxial state of compressive stress. In this case, there are
be less than 1/3. The remaining AAR has to be equally equal weights of 1/3. It should be noted that the overall
redistributed in the other two directions; and expansion is reduced through c;
3. If the compressive stress is lower than u, then AAR 4. Compression less than u but greater than the compressive
expansion in the corresponding direction is prevented strength. In this case, the weight along k should be less than
(weight equal to zero), and thus the other two weights must 1/3, and the remaining equally distributed along the other
be equal to 1/2. two directions; and
Biaxial state of stresswhere there is a compressive stress 5. Compression equal to the compressive strength. In this
equal to u in one of the three principal directions. In this case, the corresponding weight is reduced to zero, and the
case, the corresponding weight will always be equal to zero. other two weights are equal to 1/2 each.
As to the possible three combinations: Based on the preceding discussion, we generalize this
1. Tension in one direction, equal weights of 1/2; weight allocation scheme along direction k as follows:

ACI Materials Journal/May-June 2006 197

Fig. 6Weight regions.
Fig. 7Relative weights.
1. Given k, identify the quadrant encompassing l and m
(Fig. 6).* Weight will be determined through a bilinear
interpolation for those four neighboring nodes; Table 1Triaxial weights
2. Determine the weights of the neighboring nodes from
Weight direction k
Table 1 through proper linear interpolation of k.
Point l m k 0 k = u k = f c
3. Compute the weight from
1 0 0 1/3 0 0
W k ( k, l, m ) =
4
i = 1 N i ( l, m )W i ( k ) (14) 2 u 0 1/2 0 0
3 u u 1 1/3 0
where Ni is the usual two bilinear shape function used in 4 0 u 1/2 0 0
finite element and is given by 5 fc 0 1/2 0 0
6 fc u 1 1/2 0
1- ( a ) ( b ) ( b ) ( a )
N ( l m ) = ----- l m l m l m l m (15)
ab 7 fc fc 1 1 1/3
8 u fc 1 1/2 0
T
W(k) = W1 ( k ) W2 ( k ) W3 ( k ) W4 ( k ) (16) 9 0 fc 1/2 0 0
10 ft fc 1/2 0 0
a = ( a1 a2 a3 ) b = ( b1 b2 b3 ) (17) 11 ft u 1/2 0 0
12 ft 0 1/3 0 0
l = ( l f c l ) m = ( m f c m ) (18) 13 ft ft 1/3 0 0
14 0 ft 1/3 0 0
The i-j stress space is decomposed into nine distinct
15 u ft 1/2 0 0
regions (Fig. 6), where u is the upper (signed) compressive
stress below which no AAR expansion can occur along the 16 fc ft 1/2 0 0
corresponding direction (except in triaxially loaded cases).
Hence, a and b are the dimensions of the quadrant inside
Individual strain is given by
which i and j reside.
Weights of the individual nodes are in turn interpolated AAR
according to the principal stress component in the third i = W i V (19)
direction k (Table 1). It should be noted that those weights
are for the most part based on the work of Larive (1998) and and the resulting relative weights are shown in Fig. 7.
Multon (2003); but in some cases, due to lack of sufficient It should be noted that the proposed model will indeed result
experimental data, they are based on simple engineering in an anisotropic AAR expansion. While not explicitly
common sense. A simple example for the evaluation of the expressed in tensorial form, the anisotropy stems from the
weight is shown in the Appendix. different weights assigned to each of the three principal direc-
Based on the earlier work of Struble and Diamond tions. This deterioration being time-dependent, the following
(1981a), in which it was reported that no gel expansion can time-dependent nonlinear model is considered (Fig. 8)
occur at pressures above 11 MPa (though for a synthetic gel),
u is taken as 10 MPa. This value was also confirmed by E ( t, ) = E 0 [ 1 ( 1 E ) ( t, ) ] (20)
Larive (1998). The concrete tensile and compressive strengths
are ft and fc, respectively. f t ( t, ) = f t, 0 [ 1 ( 1 f ) ( t, ) ] (21)

*
Because compressive stresses are quite low compared with compressive strength, where E0 and ft,0 are the original elastic modulus and tensile
the strength gained through the biaxiality or triaxiality of the stress tensor is ignored
(Kupfer and Gerstle 1973). Furthermore, the strength gain is only approximately 14%
strength, respectively; and E and f are the corresponding
for equibiaxial compressive stresses (CEB 1983). residual fractional values when AAR tends to AAR.

198 ACI Materials Journal/May-June 2006

Fig. 8Degradation of E and fc .

Fig. 10Comparison between experimental results of Multon

and numerical calculations (after parameter identification).

Table 2System identification for Multons tests

Time, days
Characteristic Latency Iterations
Longitudinal expansion
Initial 100.0 100.0 0.001
Final 82.9 146.5 0.00363 8
Transversal expansion
Initial 100.0 100.0 0.001
Final 68.9 111.0 0.00262 7

expansion for l, c, and . The starting and final parameters

are also shown in Table 2. Having determined this initial set
Fig. 9Multons test parameter identification results for of kinetic parameters, another parameter identification for
free expansion; longitudinal and corresponding transversal the parameter in Eq. (12) for the constrained specimens
strains. Initial curve corresponds to initial guess in system yielded a value of 0.5 (Fig. 10). Finally, the parameter was
identification. used in the subsequent dam analysis. Other kinetic parameters
were determined through laboratory experiments of concrete
Finally, the possible decrease in compressive strength specimens recovered from the dam.
with AAR was ignored. Most of the literature regarding the
mechanical properties of concrete subjected to AAR show APPLICATION
little evidence of a decrease in compressive strength (as one Dam analysis data preparation
would expect because the stresses will be essentially closing Finally, a typical application to a two-dimensional analysis
the AAR-induced cracks). Furthermore, in dams (gravity and of an arch gravity dam is presented. The model has been used
arch), compressive stresses are well below the compressive in the three-dimensional nonlinear predictive analysis of an
strength, which is quite different from the tensile stresses. actual arch gravity dam, and it was shown that 50 years after
dam construction, the reaction will be exhausted (Saouma
VALIDATION and Perotti 2004b).
Validation and parameter identification was accomplished The comprehensive incremental AAR analysis of a concrete
by analyzing tests of Multon (2003). In those tests, 130 x dam is relatively complex, irrespective of the selected AAR
240 mm concrete was cast inside a steel cylinder with 3 or model, as data preparation for the load can be cumbersome.
5 mm thickness and subjected to 0, 10, or 20 MPa First, the seasonal pool elevation variation (for both
compressive stresses. thermal and stress analysis), and the stress free temperature
Figure 9 shows the three-dimensional finite element mesh Tref (typically either the grouting temperature or the average
adopted (in addition to an axisymmetric one) along with the yearly temperature) must be identified, along with the
results of the parameter identification study under free external temperature (Fig. 11).

ACI Materials Journal/May-June 2006 199

 Fig. 13Principal stress field comparison between proposed
and state of practice model (without joints).

radiation are approximated through an additional temperature

(Malla and Wieland 1999).
The selected incremental time was 2 weeks, and the initial
reference temperature set to zero. Given the external air
temperature, the pool elevation and the water temperature
boundary conditions were set to this initial boundary value
problem. Analysis was performed with Merlin and temperature
fields were examined. It was determined that, after 4 years,
the temperature field was harmonic with a 1-year frequency. At
that point, the analysis was interrupted and Tthermal (x, y, t) saved.
Following the thermal analysis, Tthermal (x, y, t) must be
Fig. 11Yearly variation of hydrostatic and thermal load. transferred to Tstress (x, y, t) as, in general, we do not have the
same finite element mesh (foundations, joints, and cracks are
typically not modeled in the thermal analysis). Following
this, a comprehensive input data file must be prepared for the
stress analysis. It includes:
1. Gravity load (first increment only);
2. T(x, y, t) = Tstress(x, y, t) Tref in an incremental format.
This is a delicate step that cannot be overlooked. In partic-
ular, the stress analysis is based on the difference between
actual and stress-free temperature. In addition, an incre-
mental analysis requires this set of data to be given in an
incremental form;
3. Stress-free referenced temperature, which will be added
to the temperature data to determine the total absolute
temperature needed for AAR;
4. Cantilever and dam/foundation joint characteristics.
The first must be accounted for in an arch dam, as the
Fig. 12Yearly variation of vertical crest displacement: expansion may lead to upstream joint opening. The second
upper curve based on Charlwoods model and lower curve must be accounted for as the AAR-induced swelling may
based on proposed model. result in separation of the dam from the foundation in the
central portion of the foundation;
Then, a transient thermal analysis is performed because 5. Uplift load characteristics (typically in accordance with
the reaction is thermodynamically activated, and the total the upstream hydrostatic load); and
temperature is hence part of the constitutive model. Heat 6. AAR data as described as follows. It should be noted
transfer by conduction only is accounted for. Convection and that a first-order approximation of the AAR kinetics parameters

200 ACI Materials Journal/May-June 2006

 Fig. 14Dam/foundation interface joint characteristics; uplift, tangential and
normal stresses, and cracking and opening displacements along joint.

may be recovered from laboratory tests of dam cores or Finally, no attempt is made to correlate computed crest
through an inverse analysis of the dam crest displacement. displacements with the (available) field measurements. The
Finally, the assembled set of data must be looped over at two-dimensional plane strain analysis conducted precludes
least 50 years to provide a complete and correct set of natural such a realistic comparison, which is described in a separate,
and essential boundary conditions. For a two-dimensional yet-to-be published, study. Furthermore, it should be noted
problem, this will result in files approximately 45 MB. that any model, irrespective of its scientific merits, can be
calibrated with field measurements. Only those models
Dam analysis results solidly based on the chemistry, physics, and mechanics of
For this preliminary plane strain analysis, a two-dimensional AAR, however, are likely to yield realistic stress field, which
central section of an arch gravity dam is selected. Results is what ultimately concerns engineers.
based on the proposed model will be contrasted with those
obtained using current state of the practice model (Charlwood CONCLUSIONS
et al. 1992) with a linear kinetics expansion. In this analysis, New constitutive model for AAR expansion
This thermo-chemo-mechanical model is rooted in the
creep is not accounted for, and the laboratory-determined
chemistry (kinetic of the reaction), physics (crack gel
Youngs modulus is retained throughout both analyses (whereas
absorption, effect of compression), and mechanics of
Charlwood tends to substantially reduce E to account for the
concrete. The major premises of the model is the assumption
creep, which in turn may yield potentially lower stresses).
of a volumetric expansion, redistribution on the basis of
To compare both analyses, final volumetric expansion has weights related to the stress tensor, and contrary to previous
been calibrated to yield identical vertical crest displacement models, the stress field affects reaction kinetics, which is a
after 50 years (Fig. 12), where the proposed model nonlinearity slight modification of the Larive (1998) model.
in the crest displacement is caused by the kinetics model and The model has been used in conjunction with a formal
its latency time in particular. parameter identification paradigm to analyze the three
Despite equal final crest displacements, internal field dimensional tests of Multon (2003). A detailed two-
stresses are drastically different as those determined from dimensional analysis of an arch gravity dam is presented.
Charlwoods model are substantially lower than those
predicted by the proposed model (Fig. 13). It should be noted ACKNOWLEDGMENTS
that the large discrepancy in stresses is partially caused by The development of the proposed model was made possible through the
the plane strain (which inhibits redistribution in the third financial support of the Swiss federal office for water and geology (FOWG).
direction) assumption of the authors model. Undoubtedly Comments and reviews provided by G. Darbre were particularly constructive.
Furthermore, the first author would like to acknowledge the support of
the lack of stress redistribution in Charlwoods model, the Italian Ministry for University and Research (MIUR), which enabled his
however, will lead to an underestimation of the stress field. visit to the Politecnico di Milano.
Furthermore, due to the influence of the thermal load, the
proposed model causes tensile stresses inside the concrete REFERENCES
dam, and a lift off along the central portion of the dam- Acres, 2004, http://www.acres.com/AcresGroup/Services/ServHydro
foundation interface (Fig. 14). These internal tensile stresses ASR/indprj.htm.
Bangert, F. D. K., and Meschken, G., 2004, Chemo-Hygro-Mechanical
can possibly explain the formation of the crack observed Modeling and Numerical Simulation of Concrete Deterioration Caused by
inside the gallery in the analyzed dam. More details can be Alkali Silica Reaction, International Journal of Numerical Analytical
found in Saouma and Perotti (2004a). Methods and Geomechanics, V. 28, pp. 689-714.

ACI Materials Journal/May-June 2006 201

 Baant, Z. P., and Steffens, A., 2000, Mathematical Model for Kinetics Saouma, V., and Perotti, L., 2004b, AAR Analysis Procedure in Kumo/
of Alkali-Silica Reaction in Concrete, Cement and Concrete Research, Merlin, Technical Report No. 3, submitted to the Swiss Federal Agency of
V. 30, pp. 419-428. Water and Geology, Bienne, Switzerland.
Baant, Z. P.; Zi, G.; and Meyer, C., 2000, Fracture Mechanics of AAR Saouma, V., and Perotti, L., 2004c, Constitutive Model for Alkali
in Concretes with Waste Glass Particles of Different Sizes, Journal of Aggregate Reaction, Technical Report No. 2, submitted to the Swiss
Engineering Mechanics, ASCE, V. 126, No. 3, pp. 226-232. Federal Agency of Water and Geology, Bienne, Switzerland.
Bon, E.; Chille, F.; Masarati, P.; and Massaro, C., 2001, Analysis of the Saouma, V., and Xi, Y., 2004, State of the Art Survey of Alkali-Aggregate
Effects Induced by Alkali-Aggregate Reaction (AAR) on the Structural Reactions in Dams, Technical Report No. 1, submitted to the Swiss
Behavior of Pian Telessio Dam, Sixth International Benchmark Workshop Federal Agency of Water and Geology, Bienne, Switzerland.
on Numerical Analysis of Dams (Theme A), Salzburg, Austria. Scrivener, 2003, personal communication.
Bournazel, J., and Moranville, M., 1997, Durability of Concrete: The Scrivener, 2005, personal communication.
Crossroad between Chemistry and Mechanics, Cement and Concrete Shayan, A.; Wark, R.; and Moulds, A., 2000, Diagnosis of AAR in
Research, V. 2, No. 10, pp. 1543-1552. Canning Dam: Characterization of the Affected Concrete and Rehabilitation
Capra, B., and Bournazel, J., 1998, Modeling of Induced Mechanical of the Structure, Proceedings of the 11th International Conference on
Effects of Alkali Aggregate Reactions, Cement and Concrete Research, AAR, Quebec, Canada, pp. 1383-1392.
V. 28, No. 2, pp. 251-260. Stanton, T., 1940, Expansion of Concrete through Reaction between
Capra, B., and Sellier, A., 2003, Orthotropic Modeling of Alkali-Aggregate Cement and Aggregate, Proceedings of ASCE, V. 66, pp. 1781-1811.
Reaction in Concrete Structures: Numerical Simulations, Mechanics of Struble, L., and Diamond, S., 1981a, Swelling Properties of Synthetic
Materials, V. 35, pp. 817-830. Alkali Silica Gels, Journal of the American Ceramic Society, V. 64, No. 11,
CEB, 1983, Concrete Under Multiaxial States of Stress Constitutive pp. 652-655.
Equations for Practical Design, Technical Report Bulletin dInformation Struble, L., and Diamond, S., 1981b, Unstable Swelling Behavior of
156, Comit Euro-International du Bton. Alkali Silica Gels, Cement and Concrete Research, V. 11, pp. 611-617.
Charlwood, R. G.; Solymar, S. V.; and Curtis, D. D., 1992, A Review of Suwito, A.; Jin, W.; Xi, Y.; and Meyer, C., 2002, A Mathematical
Alkali Aggregate Reactions in Hydroelectric Plants and Dams, Proceedings Model for the Pessimum Effect of ASR in Concrete, Concrete Science and
of the International Conference of Alkali-Aggregate Reactions in Engineering, RILEM, V. 4, pp. 23-34.
Hydroelectric Plants and Dams, Fredericton, Canada, 129 pp. Swamy, R., and Al-Asali, M., 1988, Engineering Properties of Concrete
Farage, M.; Alves, J.; and Fairbairn, E., 2004, Macroscopic Model of Affected by Alkali- Silica Reaction, ACI Materials Journal, V. 85, No. 5,
Concrete Subjected to Alkali Aggregate Reaction, Cement and Concrete Sept.-Oct., pp. 367-374.
Research, V. 34, pp. 495-505. Thompson, G.; Charlwood, R.; Steele, R.; and Curtis, D., 1994,
Furusawa, Y.; Ohga, H.; and Uomoto, T., 1994, An Analytical Study Mactaquac Generating Station Intake and Spillway Remedial Measures,
Concerning Prediction of Concrete Expansion Due to Alkali-Silica Proceedings for the Eighteenth International Congress on Large Dams,
Reaction, Durability of Concrete, Proceedings of the Third International V. 1, Q-68, R.24, Durban, South Africa, pp. 347-368.
Conference, SP-145, V. M. Malhotra, ed., American Concrete Institute, Ulm, F.; Coussy, O.; Kefei, L.; and Larive, C., 2000, Thermo-Chemo-
Farmington Hills, Mich., pp. 757-780. Mechanics of ASR Expansion in Concrete Structures, Journal of Engineering
Gilks, P., and Curtis, D., 2003, Dealing with the Effects of AAR on the Mechanics, V. 126, No. 3, pp. 233-242.
Water-Retaining Structures at Mactaquac, Proceedings of the 21st Wagner, C., and Newell, V., 1995, A Review of the History of AAR at
Congress on Large Dams, Montreal, Quebec, Canada, pp. 681-703. Three of the TVAs Dam, Proceedings of the Second International
Huang, M., and Pietruszczak, S., 1999, Alkali-Silica Reaction: Modeling Conference on AAR in Hydroelectric Plants and Dams, Tenn., pp. 57-66.
of Thermomechanical Effects, Journal of Engineering Mechanics, ASCE, Wittmann, F.; Rokugo, K.; Brhwiler, E.; Mihashi, H.; and Simonin, P.,
V. 125, No. 4, pp. 476-487. 1988, Fracture Energy and Strain Softening of Concrete as Determined by
Jabarooti, M., and Golabtoonchi, I., 2003, Alkali-Aggregate Reactivity Means of Compact Tension Specimens, Materials and Structures, V. 21,
in South-East of Iran, Proceedings of the 21st Congress on Large Dams, pp. 21-32.
Montreal, Quebec, Canada, pp. 53-62.
Kupfer, B., and Gerstle, K., 1973, Behavior of Concrete under Biaxial
Stresses, Journal of the Engineering Mechanics Division, ASCE, V. 99,
APPENDIX
No. 4, pp. 853-866. A simple example for weight determination is shown here.
Larive, C., 1998, Apports Combin'es de lExperimentation et de la Assuming that the principal stresses are given by
Modlisation la Comprehension delAlcali-Raction et de ses Effets l m k = 5.0 8.0 5.0 MPa, and that fc, ft, and
Mcaniques, PhD thesis, Thse de Doctorat, Laboratoire Central des u are equal to 30.0, 2.0, and 10.0 MPa, respectively, one
Ponts et Chausses, Paris.
Lger, P.; Cte, P.; and Tinawi, R., 1996, Finite Element Analysis of
seeks to determine Wk. The stress tensors places us inside
Concrete Swelling due to Alkali-Aggregate Reactions in Dams, Computers the quadrant defined by Nodes 1-2-3-4, whose respective
& Structures, V. 60, No. 4, pp. 601-611. weights are equal to
Lemarchand, E., and Dormieux, L., 2000, A Micromechanical
Approach to the Modeling of Swelling due to Alkali-Silica Reaction,
W 1 = --- --- = --- ; W 2 = 1--- 1--- = --- ; W 3 = --- + 1--- 1.0 --- = 2--- ;
Proceedings of the 14th Engineering Mechanics Conference, Austin, Tex. 1 1 1 1 1 1
Li, K., and Coussy, O., 2002, Concrete ASR Degradation: From 2 3 6 2 2 4 3 2 3 3
Material Modeling to Structure Assessment, Journal of Concrete Science
and Engineering, V. 4, pp. 35-46.
W 4 = --- --- = --- ;
Malla, S., and Wieland, M., 1999, Analysis of an Arch-Gravity Dam 1 1 1
with a Horizontal Crack, Computers & Structures, V. 72, No. 1, pp. 267-278. 2 2 4
Multon, S., 2003, Evaluation Exprimentale et Thorique des Effets
Mcaniques de lAlcali Raction sur des Structures Modles, PhD thesis,
Universit de Marne la Valle, France. a and b are both equal to 10 MPa, and the shape factors will be
Multon, S.; Leclainche, G.; Bourdarot, E.; and Toutlemonde, F., 2004,
Alkali-Silica Reaction in Specimens under Multi-Axial Mechanical 1 - [ ( 10 + 5 ) ( 10 + 8 ) ] 1 1 - [ 5 ( 10 + 8 ) ] = -----
1-
Stresses, Proceedings of CONSEC 4 (Concrete Under Severe Conditions), N 1 = -------- = ------ ; N 2 = -------- ;
100 10 100 10
Seoul, pp. 2004-2011.
Peyras, L.; Royet, P.; and Laleu, V., 2003, Functional Modeling of Dam
Performance Loss: Application to the Alkali-Aggregate Reaction 1 4 1 4
N 3 = --------- [ ( 5 ) ( 8 ) ] = ------ ; N 4 = --------- [ 8 ( 10 + 5 ) ] = ------ ;
MechanismChambon Dam, Proceedings of the 21st Congress on Large 100 10 100 10
Dams, Montreal, Quebec, Canada, pp. 853-872.
Portugese National Committee on Large Dams, 2003, Ageing Process
and Rehabilitation of Pracana Dam, Proceedings of the 21st Congress on and finally
Large Dams, Montreal, Quebec, Canada, pp. 121-138.
Saouma, V., and Perotti, L., 2004a, 2D AAR Parametric Investigation 1- 1--- + -----
1- 1--- + -----
4- 2--- + -----
4- 1--- = 0.40833
of a Dam, Technical Report No. 4, submitted to the Swiss Federal Agency W k = ----- .
of Water and Geology, Bienne, Switzerland. 10 6 10 4 10 3 10 4

202 ACI Materials Journal/May-June 2006