Vous êtes sur la page 1sur 15

Concerning the Use of Expansion Data from ASR Testing

Marc-Andre Berube Symposium on Alkali-Aggregate Reactivity in Concrete


(Part of the proceedings of the 8th CANMET/ACI International Conference on Recent Advances in
Concrete Technology, Montreal, May 31 June 3, 2006 pp 93-109)

David Stokes
FMC Lithium, Highway 161, Box 795, Bessemer City, NC 28016

Synopsis:
Many standard tests for alkali-silica reaction (ASR) potential of aggregates and/or supplementary
cementitious materials (SCMs) involve casting of mortar or concrete specimens and subjecting the cured
results of this casting to various environments, and then measuring %linear expansion with time. Plots of
the %expansion versus time for properly run experiments of this type always result in curves that fit a
sigmoidal growth curve to a great degree. The standard use of the results, however, is to simply take a
single data point at a predetermined period of time and use that as the sole source of information about the
potential development of ASR in concretes made with materials used in the experiment. Since all
expansions are calculated relative to the starting point, the starting point is by definition (0,0) (i.e., zero %
expansion at time zero). And since there are only two points used, (0,0) and the other data point taken
from the predetermined time interval, all of the data resulting from the test is represented by a straight
line. And since that straight line, by definition, passes through the point (0,0), this is equivalent to
meaning that the %expansion is simply proportional to the time elapsed that is, the data is utilized as if
the specimens expanded at a constant rate of expansion.
This work will discuss using the data from these types of experiments in a more rational way. Instead of
modeling the data in a function that uses a constant rate of expansion (i.e., a straight line), the data will be
modeled by a simple function that uses an expansion rate constant. Specifically, it will use the function
first suggested by Dan Johnston (Johnston et al, 2000) which is a form of the Kolmogorov-Avrami-MehlJohnson equation (KAMJ). One of the terms in the KAMJ is t0, which in this work is estimated
algorithmically from a data set obtained by first fitting the experimental data to a sigmoidal growth curve
in order to interpolate between the actual time values available from the recorded measurements.
Advantages to this approach which are discussed in the paper include estimating the potential reactivity of
materials more robustly, and better estimation of relative age until onset of significant development of
ASR in concrete made with the materials being evaluated. The method can be applied to existing test
data, but in general, may require longer test durations than the current standard in the case of accelerated
mortar bar testing (AMBT) (e.g., ASTM C1260 or ASTM C1567).

Keywords: alkali-silica reaction, ASR testing, modeling


ACI member David Stokes is Manager of Concrete Technology at FMC Corporations Lithium Division
in Charlotte, North Carolina, where he develops lithium-based materials for the prevention and treatment
of ASR in concrete. Previously he was Chief Chemist at the Materials and Research Bureau of the
Delaware Department of Transportation.

INTRODUCTION
General behavior of expansion data from standard ASR-related testing
Many standard tests for alkali-silica reaction (ASR) potential of aggregates and/or supplementary
cementitious materials (SCMs) involve casting of mortar or concrete specimens and subjecting the cured
results of this casting to various environments, and then measuring %linear expansion with time.
Examples of these are ASTM C1260, ASTM C1567, ASTM C1293, ASTM C441, ASTM C227, and
similar tests specified by other agencies such as CSA, AASHTO and RILEM. Plots of the %expansion
versus time for properly run experiments of this type always result in curves that fit a sigmoidal growth
curve to a great degree. A sigmoidal growth curve (an S shaped curve) is one in which the function
increases monotonically with increasing slope until passing through an inflection point, and then with
decreasing slope the values progress asymptotically. The particular function used in this work was first
suggested by Morgan, Mercer and Flodin (MMF) (Morgan, et al, 1975) as a general model for nutritional
responses of higher organisms. It is being used in this work in the following form:

a b c t
% Expansiont
b t
d

Here t is time, and a, b, c, and d are parameters to be fitted.


For example, Figure 1 shows fitted data from an accelerated mortar bar test with a combination of fly ash
and a reactive aggregate. The correlation coefficient demonstrates how well the model fits the data. The
coefficient is typically in excess of 0.95, so the model captures the behavior of the specimens very well.
When considering the results of ASR tests, it is not always the case that all the behaviors listed above for
this sigmoidal growth curve occur in the data set, because the actual data set is for a limited period of
time. That is, sometimes the data set will not exhibit an inflection point, because the specimen may begin
to expand very early in the process, or the inflection point may not be reached during the period of the
test. The model will still fit the data very well. In the first case, the data will simply represent that part of
the curve that occurs after the inflection point, where the slope is simply decreasing with time, and the
values proceeding to an asymptote. This part of the curve is shown in Figure 2. In the second, the data
are continually rising through the period of the test, and this is shown in Figure 3. Again, the complete
curve is shown as a reminder in Figure 4, and is similar to the example shown in Figure1.
Another function that has been shown to capture the essential features of these types of data sets is a form
of the Kolmogorov-Avrami-Mehl-Johnson (KAMJ) equation (Johnston, et al, 2000):

% Expansion(t ) 1 % Expansion(t 0 ) e k (t t0 )

Here t is time, and k, M, and t0 are parameters to be fitted. In previous work using this equation, t0 has not
been estimated, but has been generally taken to be the earliest data point available in the data set being
modeled. In many instances, this may represent considerable error as there can be considerable portions
of time during a test when very little is taking place (in terms of expansion), and these values should not
be used to fit the KAMJ equation. Different values for t0 will affect the values obtained for k (the
expansion rate constant) and M (the Avrami exponent). What has typically been done in the past with this
model has been to rearrange it in such a way as to enable the parameters k and M to be obtained by a

linear regression, while simply using the earliest value of t0 available in the data set. Thus the fitted
parameters are arbitrarily influenced by whatever the earliest time available is, in the data set, and in any
event, the earliest time may not generally be the correct time to use (Johnston, et al, 2004).
However, all other things being equal, and assuming that t0 is known or can be estimated in a nonsubjective fashion, the KAMJ has an advantage over the MMF in that it is easier to interpret, since, along
with other reasons (Capela, et al, 2001), it has only two parameters (besides t0) compared with four for the
MMF.
Note that both of these models are being explored here for empirical reasons; any mechanistic
implications or underpinnings of these models are not under consideration in the scope of work being
discussed in the paper.
The standard use of the data
The standard use of the results, however, is to simply take a single data point at a predetermined period of
time, and use that as the sole source of information about the potential development of ASR in concretes
made with materials used in the test. Since all expansions are calculated relative to the starting point, the
starting point is by definition (0,0) (zero %expansion at time zero). And since there are only two points
used, (0,0) and the other data point taken from the predetermined time interval, all of the data resulting
from the test is implicitly represented by a straight line. And since that straight line, by definition, passes
through the point (0,0), this is equivalent to meaning that the %expansion is simply proportional to the
time elapsed that is, the data is utilized as if the specimens expanded at a constant rate of expansion.
That this is a very poor way to represent the data available from these tests can be illustrated graphically,
as in Figure 5. Figure 5 shows these straight line models drawn through all the data points in this example
(but actual) set of results. Two important things stand out in this figure. One is that no line drawn
through any of the data points in the set represents the entire set very well. If each point is used to
calculate the constant expansion rate, every point gives a different value. It is not simply scatter or noise,
but constantly changing values that are obtained by considering these values. They are not simply
scattered around some central mean, as for example would be obtained by considering a collection of
repeated measurements, e.g., the weight of a sample. Each subsequent data point indicates a lower
constant expansion rate (in general). Secondly, and just as important, is that no one of these lines stands
out as being the best line to represent the data. That is, there is in general no single time that could be
picked to always give the best representation of the entire data set when modeling, or representing, the
data this way. This means that any single point that is picked as being the special time to use as a
representational value from the test results is totally arbitrary.
A rational approach to utilizing expansion data
All the standard uses of ASR testing-generated %expansion data, implicitly use the straight line as a
model of the data, since only a single point in time is used, and the other point is (0,0) by definition. This
is inherently a very poor representation of how the specimens behave in the various storage conditions
over time.
Using one of the other models discussed here leads to a greatly improved mathematical description of the
results of the test. In Figure 6, a real data set consisting of nine data points is fitted by the MMF equation,
but using only the first five data points. Note that the line which represents the fit also fits the later points
very well. Figure 7 shows the result of using seven of the nine data points, and Figure 8 shows the result
of using all of the data points. Note that all three fits of the model produce a very similar curve, even

though different numbers of points in the set were used. Thus, one thing that this type of approach yields
is a great lessening of dependence on any one specific, and otherwise arbitrary, data point in time. This is
true only when sufficient data points are collected to capture data later than the inflection point in the data
set. This data set is similar to the hypothetical example shown in Figure 2. In Figure 9, another real set of
data is used, where there is a significant period early on in the data where the expansion has not begun in
earnest. This would be an example of where using an arbitrary t0 that was too early would give a
significant deviation in the determined expansion rate constant k in the KAMJ model (Johnston, et al,
2004). In Figure 9, only the first five data points are used, and the projection of the model shows an
underestimation of the magnitudes of the expansions in the data set at later ages. Figure 10 show the
result of fitting the data using seven of the data points, and here the projection of the fitted equation shows
an overestimation of the later data points. This is because the seven data points do not progress
sufficiently beyond the inflection point in the data in this particular subset of the data. Figure 11 shows
the result when all nine data points are used. Note also, however, that the model does fit very well the
range that is covered by the points that are actually used. That is, if the data set extends sufficiently
beyond the point where an inflection point would occur, than the fit of the MMF curve will predict future
behavior of the test well. If the data set ends before the inflection point occurs in the data, the prediction
for future points will generally deviate significantly. In either case, however, for the data set that is used
to fit the curve, the curve will give a good representation of that set of data, supplying a model which can
be used to interpolate the behavior for time intervals between the existing data points.
Of the two non-linear models presented here, the KAMJ is simpler to interpret than the MMF (Capela, et
al, 2001). However, in published work to date relating to ASR testing, the constant in the equation, t0, has
been assigned a value arbitrarily, usually by using the earliest data point in the set. One non-arbitrary way
to determine t0 is to first fit the data set using the MMF equation, and then use this function to interpolate
between the existing data points, and use an algorithmic approach to determine t0. In particular, results
from the following methodology will be discussed in this work:

First, fit the data set to the MMF equation

Second, fill an array with points from the fitted MMF equation for the range covered by the actual
data (no extrapolated data is used, and in this work, the data sets were divided into 100 equal time
intervals)

Second, calculate the KAMJ for continuous subsets of the data set starting from the beginning until
the end of the data set (i.e., using later and later values of t0 )

Use the value of t0 that will give the maximum expansion rate constant k in the KAMJ -- this value of
k then becomes the estimate to use to evaluate the results of the test

RESULTS
Application of the method to data taken from the CANMET/ Industry Joint Research Program on
Alkali-Aggregate Reactions
A large and relevant research program is underway at the Canada Centre for Mineral and Energy
Technology (CANMET) in Ottawa, which is comparing the results of various laboratory testing with
corresponding large specimens stored outdoors, and which is under long-term monitoring. Details of this
program have been described elsewhere (Fournier, et al, 2004).
When supplementary cementitious materials (SCMs) are incorporated into a concrete mix for the purpose
of controlling deleterious expansions from ASR, one of the effects when testing is to delay the onset of

expansion. To illustrate this, Figure 12 shows the results of applying the methodology described above to
three accelerated mortar bar tests, where a reactive aggregate is tested with three different levels of silica
fume (0% - the control, 7.5%, 10%, and 12.5%). Note the increase of the calculated value of t0 with
increasing silica fume content found by the method with the exception of the highest silica fume level.
Table 1 shows expansion values at 14 days from the accelerated mortar bar test, expansion values for
concrete prisms at 2 years, the latest expansion values for the corresponding slabs at 10 years, the age (in
weeks) that it took the slabs to reach the failure threshold of 0.04% expansion, and the to and ln k values
from applying the KAMJ equation to the mortar bar tests according to the method described above. (The
natural logarithm of the rate constant, ln k, was suggested in earlier work as a criterion to judge potential
reactivity (Johnston, et al, 2000).) The corresponding slabs used the same high alkali cement as the
concrete prism test, but some slabs did not have the alkali level boosted to 1.25% (unboosted) as was
done in the concrete prism test and some slabs did (boosted).
Note that all of the silica fume mixes passed the 14 day limit of 0.10% expansion in the mortar bar
testing. Likewise for the 2 year concrete prism testing (although the 7.5% silica fume mix is marginally
failing). However, all of the slabs from the outdoor exposure site are exceeding the failure criterion of
0.04% by 10 years of age, and are continuing to expand.
The ln k values are taken from an analysis of the accelerated mortar bar data (including the data out to 28
days), and they correspond to the value that is obtained when then the KAMJ is fitted to the subset of the
data that begins at the indicated t0, which was found by the method outlined above. Using ln k values in
excess of -6 as a criterion for potential reactivity (Johnson, et al, 2000), all of the mixes are placed in the
category of potentially reactive, which coincides with the evidence from the slabs. A surprising result is
that while t0 is increasing with increasing silica fume content, as would be expected in general, there is a
reversal between 10% silica fume and 12.5% silica fume. While this is unexpected, in general, note that
this has also happened in the slabs as well, both boosted and unboosted.
It is also instructive to consider that all three ways of interpreting the test results, 14 day expansion in the
accelerated mortar bar tests, 2 year expansion in the concrete prism test, and the KAMJ analysis of the
mortar bar testing indicate decreasing reactivity with increasing levels of silica fume. However the
current expansion values for the slabs show higher values for the 12.5% compared with the 10% silica
fume. There may not actually be a contradiction here, however, as it must be noted that the expansion
going on in the slabs may not totally be from progressing ASR. Once cracking starts in the slabs, they
can also be affected by ongoing freeze thaw cycles in the environment in which they exist. Since the
12.5% silica fume began expanding sooner that the 10% silica fume, it could be that this effect has been
operating longer in the slabs as a result. As a reminder, the t0 that was found by the KAMJ analysis
indicated a tendency of quicker onset of reaction for this mix.
Note also the expected result regarding the effect of increased alkali level on both the higher amount of
expansion and the quicker rate at which the expansion has taken place in the slabs. This point will be
revisited in the discussion section below.
Correlation of laboratory results with failure in the outdoor storage specimens
In this section, a comparison will be made between the 2 year concrete prism test result and the rate
constant k from the application of the KAMJ to the concrete prism data. Both will be used to correlate to
the number of weeks until 0.04% expansion was reached in the outdoor slabs. Figure 13 shows the fit
between the number of weeks until 0.04% expansion was reached in the slabs and the corresponding 2
year expansion of the concrete prism. The correlation coefficient suggests that 63% of the variance is
explained by the 2 year expansion value. Note that the graph makes physical sense; the lower the
expansion value at 2 years, the longer the slab takes to reach the failure criterion.

Figure 14 shows the fit between the number of weeks until 0.04% expansion was reached in the slabs and
the corresponding rate constant k from application of the KAMJ to the concrete prism data. Here the
much higher correlation coefficient suggests that 83% of the variance is explained by the rate constant.
The same relationship is shown in a different way in Figure 15, this time using the natural log of the rate
constant, ln k.
DISCUSSION
Clearly the non-linear models discussed here do a thorough job of mathematically describing the results
obtained from doing the standard ASR-related tests that involve storing specimens in an unchanging
environment and measuring expansion over time. By modeling the entire behavior of the specimens in
time instead of merely picking an arbitrary data point from among the set yields parameters that are more
robust in estimating the properties of interest relating to potential reactivity of materials used to make
concrete.
In the first example with the silica fume mixes, use of the rate constant estimated from the mortar bar data
proved more successful than either of the standard interpretations of the mortar bar test or the concrete
prism test in regards to the performance of the outdoor storage specimens.
And the rate constant is also better able to correlate with the time until the onset of significant reaction in
the outdoor storage specimens. This is not surprising as most of the information in the standard
interpretation of the tests is simply not used, and the way it is used bears little relation to the actual
behavior of the specimens in the test procedures.
The procedure demonstrated in this work is not the only way this information could be gleaned from these
tests. It is simply one suggestion. It has the benefit of being straightforward, and doesnt require any
changes to existing methods, with the exception in the case of the standard accelerated mortar bar test.
Here it will generally be important to run the test out for longer periods of time. In order to capture the
entire behavior exhibited by a sigmoidal growth curve, it is conceivable that some mixes would need to be
run out to 56 days or longer. But the point of the exercise is not to capture the entire behavior of the
specimens, but rather to make an empirical decision regarding the potential reactivity of the materials at
hand. For this purpose, it is unlikely that the test will need to be run more than 28 days, as anything that
requires longer to expand significantly will end up with a rate constant that is low enough to demonstrate
a safe level of reactivity. Practically any siliceous mineral can be made to react if subjected to enough of
a driving force in terms of alkali level and temperature. That doesnt mean that it wouldnt be safe to use.
A more thorough cataloging of different aggregates and mixes should answer this question.
It would appear also that the rate constant could be used to compare different mix options in a rational
way to judge which would be longer lasting in the field, as shown by the second example, where the rate
constant outperformed the 2 year expansion limit in correlating to the time to reach 0.04% in the slabs.
As the data in Table 1 demonstrated, however, it will never be possible to predict the actual time to failure
by this method, or any other method that is simply an analysis of these various test regimens. This is
simply because the standard tests that exist today do not evaluate some of the most relevant variables that
would influence the actual behavior of the concrete.
First, the storage conditions are constant; unless a concrete is intended to exist in one of these conditions,
it will never experience these constant conditions. In the depths of a large mass like a dam it may come
close, but the thinner the cross section, the less that will be true in the field. Another variable of extreme
relevance that is not evaluated at all in these tests is the effect of the alkali content. Unless a small matrix

of testing is done with varying alkali contents, it is utterly impossible to predict how a given combination
of materials will react at different alkali levels. This is true both in regard to the extent of the reaction and
the rate of the reaction. Of course there are many other factors as well, but these are the main ones that
are not addressed by the standard tests.
No attempt was made in this work to evaluate how the other parameters in these models could be used to
better correlate field performance. While that work will be undertaken as well, the discussion just
preceding should make it clear that there are severe limits to what can be deduced from the information
contained in these tests. Nonetheless, the examples discussed in this work should make it just as clear
that much better use can be made of the information contained in the test results than is being utilized by
the standard methods of interpretation.

ACKNOWLEDGMENT
Thanks to CANMET for the use of data from their excellent research program in examples discussed in
the paper.

REFERENCES
Capela, M.V., Magnani, R., Adorno, A., Capela, J., Comparison of the fit of the Morgan-Mercer-Flodin
and Johnson-Mehl-Avrami models for the precipitation reaction in the Cu-3wt.%Al-5wt.%Ag alloy,
Ecltica Qumica (So Paulo), v.26, 2001.
Fournier, B., Nkinamubanzi, P., Chevrier, R., Comparative Field and Laboratory Investigations on the
Use of Supplementary Cementing Materials to Control Alkali-Silica Reaction in Concrete, Proc. Of the
12th International Conference on Alkali-Aggregate Reaction in Concrete, Beijing (China), October 2004,
Tang & Deng, Editors, pp. 528-537.
Johnston, D., Stokes, D., and Surdahl, R., A Kinetic-Based Method for Interpreting ASTM C1260,
Cement, Concrete and Aggregates, CCAGDP, Vol.22, No.2, December 2000, pp. 142-149.
Johnston, D., Stokes, D., Fournier, B., and Surdahl, R., Kinetic Characteristics of ASTM C1260 Testing
and ASR-Induced Concrete Damage, Proc. Of the 12th International Conference on Alkali-Aggregate
Reaction in Concrete, Beijing (China), October 2004, Tang & Deng, Editors, pp. 338-346.
Morgan, P., Mercer, L., and Flodin, N., General Model for Nutritional Responses of Higher Organisms,
Proc. Nat. Acad. Sci., USA, Vol. 72, No. 11, pp. 4327-4331, November 1975, Biochemistry.

Table 1: Results of testing and analysis of a reactive aggregate and various levels of silica fume
%
Silica
Fume

0
7.5
10
12.5

14 day
AMBT

2 yr
CPT
boosted

10 yr slab
unboosted

% exp

% exp

% exp

0.173
0.089
0.083
0.066

0.087
0.036
0.030
0.024

0.136
0.044
0.051
0.054

Weeks
to
0.04%
178
480
488
387

10 yr slab
boosted

%
exp
0.171
0.103
0.075
0.085

to

ln k

Weeks
to
days
0.04%
174
4.76 -3.82
254
10.64 -4.74
292
24.08 -4.90
221
13.16 -5.02

0.35
0.3

%Expansion

0.25

r = 0.9997
0.2
0.15
0.1
0.05
0
0

10

15

20

25

Age (days)
Fig. 1: Example fit of MMF equation to ASTM C1260 data.

30

0.7

0.6

%Expansion

0.5

0.4

0.3

0.2

0.1

Time
Fig. 2:

Data that would be described by the later part of the MMF equation after the inflection point.
0.12

%Expansion

0.1

0.08

0.06

0.04

0.02

Fig. 3:

Time

Data that would be described by the early part of the MMF equation before the inflection
point.

0.8
0.7

%Expansion

0.6
0.5
0.4
0.3
0.2
0.1
0

Time
Fig. 4:

Data that represents all the features that can be captured by the MMF equation.

2
1.8
1.6

%Expansion

1.4
1.2
1
0.8
0.6
0.4
0.2
0
Time
Fig. 5:

Lines of constant expansion rate drawn through each data point and the origin. Which line best
represents all the data?

2
1.8
1.6

%Expansion

1.4
1.2
1
0.8
0.6
0.4
0.2
0
0

10

15

20

25

30

Age (days)
Fig. 6:

Only the first 5 data points were used to fit the MMF equation to this data set.

2
1.8
1.6

%Expansion

1.4
1.2
1
0.8
0.6
0.4
0.2
0
0

10

15

20

25

30

Age (days)
Fig. 7:

Only the first 7 data points were used to fit the MMF equation to this data set.

2
1.8
1.6

%Expansion

1.4
1.2
1
0.8
0.6
0.4
0.2
0
0

10

15

20

25

30

Age (days)
Fig. 8:

All 9 data points were used to fit the MMF equation to this data set.

0.25

%Expansion

0.2

0.15

0.1

0.05

0
0

10

15

20

25

30

Age (days)
Fig. 9:

Only the first 5 data points were used to fit the MMF equation to this data set.

0.25

%Expansion

0.2

0.15

0.1

0.05

0
0

10

15

20

25

30

Age (days)
Fig. 10: Only the first 7 data points were used to fit the MMF equation to this data set.
0.25

%Expansion

0.2

0.15

0.1

0.05

0
0

10

15

20

25

Age (days)
Fig. 11: All 9 data points were used to fit the MMF equation to this data set.

30

0
-1

0% (Control)
-2

7.5%
-3

ln k

10%
-4
-5
-6

12.5%
-7
-8
-9
0

10

15

20

25

30

t 0 (days)
Fig. 12: Plot of ln k versus t0 for silica fume mixes. The t0 is selected that yields the maximum ln k value.

600

Weeks to reach 0.04%

500
400

R2 = 0.634
300
200
100
0
0

0.05

0.1

0.15

0.2

0.25

0.3

% Expansion @ 2 years
Fig. 13: Weeks to reach 0.04% expansion in the slabs versus the 2 year concrete prism expansions.

600

Weeks to reach 0.04%

500
400

R2 = 0.829
300
200
100
0
0

0.005

0.01

0.015

0.02

Rate Constant k
Fig. 14: Weeks to reach 0.04% expansion in the slabs versus the rate constant k from the application of
the KAMJ to the concrete prism expansions.

600

Weeks to reach 0.04%

500
400

R2 = 0.829
300
200
100
0
-10

-8

-6

-4

-2

ln k
Fig. 15: Weeks to reach 0.04% expansion in the slabs versus the natural logarithm of the rate constant k
from the application of the KAMJ to the concrete prism expansions.

Vous aimerez peut-être aussi