Cement and Concrete Research 109 (2018) 10–18

Contents lists available at ScienceDirect

Cement and Concrete Research

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

Residual service life of carbonated structures based on site non-destructive T

tests
⁎
R. Nevesa, , R. Torrentb, K. Imamotoc
a
ESTBarreiro, Polytechnic Institute of Setúbal, Barreiro, Portugal
b
Materials Advanced Services Ltd., Buenos Aires, Argentina
c
Tokyo University of Science, Tokyo, Japan

A R T I C LE I N FO A B S T R A C T

Keywords: Assessing the residual service life of carbonated reinforced concrete structures is typically conducted by mea-
Carbonation (C) suring cover depths and carbonation of drilled cores, assuming the “square root law” for the progress of the
Concrete (E) carbonation front. Given the large spatial variability of carbonation depths, the required number of cores be-
Permeability (C) comes prohibitive.
Service life
A model is presented, based on numerous experimental data, by which the carbonation rate can be estimated
by means of non-destructive site measurements of the air-permeability kT. A Weibull distribution of the car-
bonation rate Kc is fitted to the results, which is a function of kT.
Hence, for each kT value measured on site coupled with the cover depth xd, expected, optimistic and pes-
simistic values of the corrosion initiation time can be obtained without the need of drilling cores. The method is
validated on 111 paired results of kT and xd, obtained on the emblematic Tokyo's Museum of Western Art.

1. Introduction concrete structures. Indeed, measurements of the carbonation depth xc

Although less conspicuous than chloride-induced corrosion, corro-
sion damage due to carbonation constitutes a matter of concern. Its
incidence may be aggravated in the future by the gradual rise in CO2
concentration in the air, especially in industrial, motorcar and urban
environments, and by the reduction in clinker content in concrete that
is taking place in the cement and concrete industry.
The carbonation progress is generally assumed as:
xc = K c × t n (1)
xc is the carbonation depth (mm); Kc is the carbonation coefficient
(mm/yearn), t is exposure time/age (year) and n is an exponent, usually
taken as 1/2 (the so-called “square-root law”).
Hence, applying the “square-root law” (see discussion in Section
2.2), Eq. (1) becomes:
xc = K c × t (2)
Kc in Eq. (2) depends on several factors such as the “penetrability”
of the concrete cover, the amount of carbonatable material in the ma-
trix, the concentration of CO2 in the atmosphere, the exposure condi-
tions (temperature, RH, rain, solar exposure), etc.
Eqs. (1) or (2) are often used to estimate the service life of existing
obtained destructively on drilled cores or fragments removed from the
surface, at time t0 allow, by a simple application of Eq. (1) to know Kc
and, therefore, to predict the time t at which the carbonation front will
reach the steel and depassivate it (when carbonation depth xc equals
cover depth xd). xc is typically measured by spraying a pH indicator
(usually phenolphthalein) on freshly broken concrete surfaces [1].
This simple approach faces two drawbacks: the destructive or da-
maging nature of the measurements and the high variability of xc en-
countered in old structures. For instance, a range of up to 65 mm in
carbonation depths was measured on structures almost 100 years old
investigated in Chile [2], which reached a value of 140 mm in a
25 years old bridge investigated in China [3]. The high scatter of xc
results is confirmed by other results presented later (Fig. 1). Hence, to
have a representative picture of the xc values in a structure, a prohi-
bitive number of cores or broken concrete samples is required.
As discussed above, the “penetrability” of the cover concrete is one
of the main factors governing the carbonation rate Kc. Air-permeability
is one of the most accepted properties to evaluate the “penetrability” of
the cover concrete and is one of the easier tests to perform on site.
Already in the 90's, a method to estimate the progress of carbonation in
concrete, based on intrusive site measurements of air-permeability, was
proposed [4]. Non-destructive air-permeability measurements were

⁎
Corresponding author at: ESTBarreiro, Polytechnic Institute of Setúbal, R. Américo da Silva Marinho, 2839-001 Barreiro, Portugal.
E-mail address: rui.neves@estbarreiro.ips.pt (R. Neves).

https://doi.org/10.1016/j.cemconres.2018.04.002
Received 23 October 2017; Received in revised form 31 March 2018; Accepted 3 April 2018
0008-8846/ © 2018 Elsevier Ltd. All rights reserved.
R. Neves et al. Cement and Concrete Research 109 (2018) 10–18

XC3
7
XC4
Tokyo
6 Osaka
Tochigi
Kc (mm/year1/2) 5 Chiba
Sendai E
Sendai W
4
Underpass
Bridge
3 Overpass
Building
2

0
0.001 0.01 0.1 1 10 100 1000
kT (×10-16 m2)
Fig. 1. Relation between Kc and kT for all cases studied.

used to predict the service life of the emblematic Museum of Western
Art in Tokyo [5] and of precast concrete segments of the Port of Miami
Tunnel [6]. In these two investigations the coefficient of air-perme-
ability kT was measured applying the non-destructive Swiss Standard
Method SIA 262/1-E, 2013 [7], described in Section 3.
The purpose of this paper is to present a methodology for assessing
the residual service life of reinforced concrete structures at risk of
carbonation-induced corrosion, based exclusively on the non-destruc-
tive determination of the air-permeability and of the cover depth, both
measured on site.
The authors followed a very original and promising approach pro-
posed by Teruzzi [8], who based his investigation on test results of
carbonation depth and air-permeability obtained on a single building in
the South of Switzerland (Canton Ticino, known for its sunny and mild
Mediterranean climate). Teruzzi based his analysis on assessing the
carbonation rate Kc on the basis of kT measurements. Data from just one
building is not enough to develop a model for general use.
The contribution of this paper is to add robustness to Teruzzi's ap-
proach by trying it on a large number of different types of structures,
located in Japan, Portugal and Switzerland (North and South), thus
covering a wide range of climates and construction and exposure con-
ditions. In addition, the model is extended to the prediction of the
corrosion initiation time Ti, in the case where the cover thickness xd is
measured in parallel with kT, with its predictions validated on a real
case.
2. Assumptions adopted for the proposed model
2.1. Definition of service life
which involves a condition that is necessary but not sufficient to have
corrosion damage. The propagation period may be very short for sur-
faces exposed to wetting and drying conditions or very long for surfaces
exposed to permanently dry conditions. Therefore, the Service Life
predicted by the proposed model is reasonably consistent with the ap-
pearance of corrosion symptoms for external surfaces exposed to
normal outdoor conditions (e.g. XC4 according to EN 206 [10]). If the
elements are located indoors or to extremely dry outdoor conditions,
the predicted service life will be too conservative and the corrosion
propagation period must be taken into consideration. An approach for
empirical estimation of the propagation time for carbonation-induced
corrosion, proposed by Torrent and Luco, can be found in [11].
2.2. Rate of carbonation progress
The penetration of CO2 into concrete can be considered as a purely
diffusive process, obeying Fick's 2nd Law, which predicts a carbonation
depth proportional to the square root of time (Eq. (2)). The matter is
further complicated by the “carbonation” chemical reactions and by the
partial and variable blocking effect of moisture in the concrete pores.
Most models developed to predict carbonation are based on the “square
root” law [12–14], including a coefficient that reduces the predicted
carbonation depth for structures exposed to wetting, a fact observed
experimentally by Wierig [15].
In the proposed model, the “square-root law” is adopted to convert
carbonation depths xc, measured in old structures, into carbonation
rates Kc, applying Eq. (2) (see Fig. 1, later discussed). Consistently, the
“square-root law” is also applied to predict carbonation depths, based
on the Kc computed with the same “law”.

The corrosion process of steel embedded in concrete is characterized
by two well defined stages [9], namely the initiation period (time
needed for the carbonation front to reach the position of the affected
steel bar, depassivating it) and the propagation period (time needed for
the corrosion in the bar to develop sufficiently so as to generate some
visible damage). This investigation deals just with the initiation period,
2.3. Age of the structures
The carbonation depths and air-permeability values used to support
the model were measured on structures with ages predominantly in the
range 10–60 years. Therefore, the proposed model is applicable to
structures with ages within the above-mentioned range.

11
R. Neves et al. Cement and Concrete Research 109 (2018) 10–18

Table 1
Identification and description of the initial data set.
Case study Number of pairs Age (y) Location Country Tmin (C) Tmax (C) Precip. (mm) Source

Building 55 35 Lugano Switzerland 2.5 20.3 1733 [8]

Underpass 28 30 Oensingen −1.2 16.7 1053 [19]
Bridge 7 60 Basel 0.2 18.1 791
Overpass 12 30 Brugg 0.5 18.5 997 [20]
Tokyo 3 50 Tokyo Japan 5.1 27.3 1435 [21]
Osaka 7 49 Osaka 4.9 28.2 1444
Tochigi 13 12–31 Tochigi 2.9 26.5 1325
Chiba 12 42 Chiba 5.1 26.6 1435
Sendai (E) 4 49 Sendai 0.3 24.5 1286
Sendai (W) 3 49
XC3 44 11–31 Setubal Portugal 11.7 23.3 641 [22]
XC4 25 4–32 Lisbon 11.7 22.7 691

Tmin and Tmax are minimum and maximum average monthly temperature.

3. Assessing carbonation rate through air-permeability tests 3.2. Data trial

3.1. Conditional density function of carbonation rate for given air-
permeability
To express the time dependency of carbonation depth a mathema-
tical model is required. As discussed in Section 2.2, the simple and fairly
accurate model described by Eq. (2) is adopted.
An integration by substitution of a new variable, allows getting the
probability density function of xc from the density function of the time
independent variable Kc
1 and air-permeability (at the moment of the measurement) and, hence,
fXC (x c ; t ) = f (K c )
t KC
Furthermore, as the aim of this research is to propose a metho-
dology for service life design of reinforced concrete structures that has
as input parameters the reinforcement cover and the air-permeability of
concrete, the density function of Kc shall be obtained from a conditional
density function of Kc for a given air-permeability kT
+∞
fKC (K c ) = ∫kT fKC kT (kc ; kT ) fkT (kT ) d (kT )
There is the need to know the conditional density function of car-
bonation rate Kc, for a given air-permeability kT, not trivial, that will be
addressed next. A probability distribution function of carbonation rate
depending upon air-permeability is derived, applying likelihood in-
ference to data collected on several old structures. These data consist of
parallel measurements of air-permeability and carbonation depth.
Knowing the age of the structures and the carbonation depth,
measured after [1], a carbonation rate (Kc) is computed, applying Eq.
(2). The coefficient of air-permeability kT is determined applying a non-
destructive method, developed by one of the authors [16], which is
considered by the DuraCrete project consortium [17] and which is
currently adopted by Swiss Standard SIA 262/1 [7]. In brief, this
method features a double-chamber testing cell, where vacuum is cre-
ated. The vacuum and a pair of concentric soft rings, provide sealing
onto the concrete surface and create two separate chambers inside the
cell. The air in the concrete pores flows through into the inner chamber,
raising its pressure. The rate of pressure rise is directly linked to the
coefficient of air-permeability of the cover concrete. A regulator
maintains the pressure of the external chamber permanently balanced
with that of the inner chamber, ensuring a unidirectional flow into the
inner chamber and thus allowing calculating the coefficient of perme-
ability (kT) on the basis of a theoretical model. More details about the
method can be found elsewhere [16,18].
In the following, kT is considered as the independent variable,
whereas Kc is the dependent variable.
Unlike in Teruzzi's investigation [8], where data from a single
source – facades of a 35 years old building – was used, data from dif-
ferent sources will be merged. The initial data set comprises 12 case
studies, collated from 5 publications [8,19–22], as summarized in
Table 1.
Fig. 1 shows the results of kT measured at the spot where the car-
bonation depth xc was to be measured and of Kc (calculated as xc/age1/
2
), for all cases in Table 1.
The scatter of the results shown in Fig. 1 can be attributed to many
factors affecting both carbonation (during the many years of exposure)
(3) the correlation linking them. For instance, it is known that the re-
lationship between carbonation rate and air-permeability is affected by
the binder type [4,23] (as far as the authors could ascertain, the cement
used for the structures investigated in [19–22]was OPC). The en-
vironment humidity regime (pore blocking effect of moisture) is known
to affect carbonation, and possibly cracking, during the whole life of the
structure [24]; in the case of air-permeability what matters is the
moisture condition at the particular moment of the test and provisions
(4) are taken to keep it under control [7]. Microcracking may affect dif-
ferently the measured carbonation and the air-permeability [18].
Although there are several tests to compare bivariate samples
[25,26], their purpose is to test sample location, not if the relationship
between the two variables of one sample is similar to that of the other
sample. To assess if the relationships between kT and Kc are comparable
in different sets, the following procedure is adopted.
1. A relationship between kT and Kc is established for a reference set.
2. The previous relationship is applied to estimate the Kc values of
another sample.
3. The Student test for paired samples [27] is applied to test the hy-
pothesis that there is a difference between the estimated and the
observed Kc values of the sample.
To select the reference sample, each of the 12 case studies was
subjected to a regression analysis, and F-statistic test was applied to test
the significance of each model/relationship between kT and Kc, through
testing the (null) hypothesis that there is no relationship between Kc
and kT. The case study with the lowest proof-value for the F-statistic
test, i.e. the one where the confidence on rejecting the null hypothesis is
higher, is selected as the reference sample.
The regression analysis was developed upon a logarithmic re-
lationship between Kc and kT, also considered in other studies [19,21].
A log-transformation of kT was performed and a linear regression was
carried out. It shall be mentioned that a theoretical relationship be-
tween Kc and kT can be derived, leading to a power model, as presented
in [23,28]. Although such relationship was also considered through a

12
R. Neves et al. Cement and Concrete Research 109 (2018) 10–18

7
XC3
6 XC4
Bridge

Kc (mm/year1/2)
5
Overpasses
4
Tokyo
3 Sendai (W)
Prediction model
2
Conf Low
1 Conf High
Pred Low
0
Pred High
-3 -2 -1 1 2 3
Pred Teruzzi[8]
Log10(kT×1016)
Fig. 2. Carbonation rate versus air-permeability for the selected data.

log-log transformation of Kc and kT, the resulting model failed to prove
its validity according to the criteria applied in 3.4 and thus is not
presented.
The lowest proof-value obtained was 1 × 10−6, substantially lower
than the usual significance level (0.05), for the case study XC3, which is
adopted as the reference sample.
Afterwards, the regression model obtained from XC3 was applied to
estimate Kc values for the 11 remaining case studies. These estimated
values were compared with the observed ones for each case study, by
means of the Student test for paired samples with the null hypothesis
that the mean of the estimated Kc is equal to the mean of the observed
Kc. This hypothesis is rejected in 6 of the 11 tested pairs of samples and
the corresponding case studies are discarded from the data set. Thus,
the after-trial data set comprises the following case studies: Bridge;
Overpass; Tokyo; Sendai (W); XC3 and XC4. By merging all these case
studies, a sample with 94 pairs (kT, Kc) is formed.
To evaluate the propriety of the trial process, a regression analysis is
applied to the resulting sample. The relationship between Kc and kT, its
confidence and prediction intervals as well as the position of data,
identified by case study are shown in Fig. 2.
A generally uniform distribution of points around the model line is
observed, without any particular indication of skewness for any case
study, suggesting a fairness of the trial process. The corresponding
analytical model to estimate carbonation rate through air-permeability
results is:
K c = 0.805 × log10 (kT ) + 2.66
where Kc is the carbonation coefficient (mm/year1/2) and kT is the air-
permeability coefficient (10−16 m2).
This regression analysis returns a coefficient of determination
r2 = 0.341 and an F statistic F = 47.67, for which corresponds a proof-
value of 6 × 10−10, lending support to the existence of a relationship
between Kc and kT in the analysed sample. The low coefficient of de-
termination is due to the large scatter of results presented in Fig. 1 and
discussed earlier. The scatter is an essential part of the model proposed,
as the shape of the adopted Weibull distribution depends on the kT
value measured, showing a wider distribution for higher kT values.
3.3. Statistical distribution and its parameters
It is accepted that carbonation related variables, such as carbona-
tion rate, follow a normal distribution [22,29,30]. However, facing the
strong dispersion of Kc in the present data set (Fig. 1), it is anticipated
that a normal distribution would return negative carbonation rates,
which has no physical meaning. Two statistical distributions with
skewness and non-zero values were tried to fit the data: Log-Normal
and Weibull. Weibull was preferred because of a better fit to the results,
a more elegant way of dealing with the data and a useful explicit so-
lution of the cumulative probability function. Therefore, by adopting
the Weibull distribution, found to fit the distribution of carbonation
depths [31], the issue of negative Kc is overcome. Anyhow, the shape
parameter of the Weibull distribution allows it, if appropriate, to ap-
proach a Normal or a Log-Normal distribution.
If Kc follows a Weibull distribution, the probability density function
will be:
d K c d − 1 −(K c / c )d
f (K c ) = ⎛ ⎞ e
c⎝ c ⎠ (6)
where Kc is any value of the carbonation coefficient population (≥0); c
is the distribution's scale parameter (> 0); and d is the distribution's
shape parameter (> 0).
Introducing the assumed relationship between Kc and kT, i.e. the
effect of kT on Kc, as Kc = a · log10(kT) + b and the expected value of Kc
given by E(Kc) = c · Γ(1/d + 1), where Γ(.) is the gamma function, Eq.
(5)
(6) becomes
d
d (
⎛ Kc ∙Γ 1 +1 ⎞ )
( )
1 d
⎛ Γ +1 ⎞ −⎜ ⎟
d d − 1 ⎜ a ∙ log10 (kT ) + b ⎟
f (K c ) = d∙ ⎜ ⎟ ∙ (K c ) ∙e ⎝ ⎠
⎜ a∙log10 (kT ) + b⎟
⎝ ⎠ (7)
There are three parameters in Eq. (7): a, b and d, that shall be such
that Eq. (7) provides the best estimation of the (kT, Kc) pairs of the data
set. Then, the maximum likelihood method [32] is applied to estimate
the mentioned parameters. The likelihood function will be
n

Fig. 2 presents also the model fitted by Teruzzi to the results of the L (a, b, d; K c , kT ) = ∏ f (K ci , kTi | a, b, d )
i=1
building he investigated [8], which is shifted down by about 1 mm/y1/2
n∙d n d−1
with respect to that of Eq. (5), falling outside of its confidence interval. 1 (∏i = 1 K ci )
= dn∙ ⎛Γ ⎛ + 1⎞ ⎞ ∙ n d
The difference has to be attributed to the larger number of cases con- ⎝ ⎝d ⎠⎠ (∏i = 1 a∙log10 (kTi ) + b)
sidered in the present investigation (six structures instead of just one) d
and the resulting variety of exposure and testing conditions. From the
n ⎛ Kc ∙Γ 1 +1 ⎞
− ∑⎜
i (
d )⎟
⎜ a ∙ log10 (kTi) + b ⎟
viewpoint of the practical application of the model, Eq. (5) will predict ∙e i = 1 ⎝ ⎠ (8)
higher carbonation rates Kc for the same air-permeability kT and,
therefore, more pessimistic service life estimations. where n is number of (kTi, Kci) pairs in the data set.
Now, the values of a, b and d are determined by maximizing the

13
R. Neves et al. Cement and Concrete Research 109 (2018) 10–18

likelihood function. However, finding the maximum likelihood esti-

mators demands the solution of a non-linear system of equations (null
2 2.2 2.4 2.6 2.8 3
partial derivatives of L) which is not easy [33]. Therefore, an optimi- 0
zation algorithm [34] is applied. Furthermore, the logarithm of L, i.e.
the log-likelihood function (Eq. (9)) is maximized instead, as it is
equivalent and avoids yielding an underflow due to a run out of com- -0.5
puter floating point precision.
n -1
ln L (a, b, d; K c , kT ) = ∑ ln f (K ci , kTi | a, b, d ) = n∙ln d
i=1

1 ⎛ ⎞
n -1.5
+ n∙d∙ln ⎛Γ ⎛ + 1⎞ ⎞ + (d − 1) ∙ ⎜∑ ln K ci⎟

l (d)
⎝ ⎝d ⎠⎠ ⎝ i = 1 ⎠
n -2
− d∙ ∑ ln(a∙log10 (kTi ) + b)
i=1
d -2.5
− ∑
n
(
⎛ K ci ∙Γ d + 1
1
) ⎞
⎜ ⎟
i=1 ⎜ a∙log10 (kTi ) + b⎟
-3
⎝ ⎠ (9)

For the considered data set, a maximum of ln L = −137.16 is at-

tained and the corresponding maximum likelihood estimators are -3.5
a ̂ = 0.848, b ̂ = 2.636 and d = 2.444.
Confidence intervals for the model parameters are defined based on -4
the likelihood ratio test [35]. A log-likelihood ratio is computed as

ℓ = ln(L (θ∗)) − ln(L (θ ))
where θ stands for the set of maximum likelihood estimators and θ⁎
represents any other set of the model parameters.
Assuming that −2 ℓ follows a chi-squared distribution with degrees
of freedom equal to the number of restrictions imposed (varying
parameters) [35], the confidence interval may be
2
θ∗: −2ℓ ≤ χGL ,α
where GL is the number of degrees of freedom and (1-α) is the ampli-
tude of the confidence interval.
Keeping all parameters fixed at their optimum values except the one
for which the confidence interval will be estimated, GL = 1. For a
confidence interval of 95%, α = 0.05. Then, recalling Eqs. (2) and (3),
the 95% confidence interval will be the range of θ⁎ values satisfying
ln(L (θ∗)) ≤ ln(L (θ )) − 1.92
(10) Fig. 4. Log-likelihood ratio for model parameter d about its maximum like-
lihood estimator.
by Teruzzi in a similar investigation [8], it was found that the 95%
confidence interval for a overlaps with the corresponding computed by
Teruzzi. For parameter b, the values are not directly comparable, re-
quiring a previous variable transformation, after which appears that the
(11) 95% confidence intervals are disjoint, providing further support to the
plausibility of the trial process (Section 3.2), where the data considered
by Teruzzi were discarded. Concerning parameter d, the 95% con-
fidence intervals are again overlapping, meaning that the Weibull dis-
tributions will have similar shapes, where the difference is believed to
lie on different scatter of the corresponding data sets.
3.4. Goodness of fit
(12)

The resulting 95% confidence intervals are: [0.595, 0.959] for a ,̂
[2.480, 2.848] for b ̂ and [2.140, 2.864] for d . The limits of the con-
fidence intervals correspond to the values of the varying model para-
meter for which ℓ = − 1.92, as illustrated in Figs. 3 and 4.
Comparing the obtained confidence intervals with those presented
To ascertain the goodness of fit of the derived model, P-P and Q-Q
plots are constructed. Assuming that the derived model is accurate, the
∼ 
standardized variables of the Weibull distributions K ci = (K ci/ ci ̂)d , with
̂
ci ̂ = (a ∙̂ log10 (kTi ) + b )/Γ(1/ d + 1) , will follow an exponential distribu-
tion, Exp(λ) with distribution parameter λ = 1 [36]. Therefore, the

0.5 0.6 0.7 0.8 0.9 1 2.4 2.5 2.6 2.7 2.8 2.9
0 0
-0.5 -0.5
-1 -1
-1.5 -1.5
l (a)

l (b)

-2 -2
-2.5 -2.5
-3 -3
-3.5 -3.5
-4 -4
a b
Fig. 3. Log-likelihood ratio for model parameters a and b about their maximum likelihood estimators.

14
R. Neves et al. Cement and Concrete Research 109 (2018) 10–18

1 1
0.9 kT=0.01 (10-16 m2)
Model probability

0.8 0.8
0.7 kT=0.1 (10-16 m2)
0.6 0.6

f(Kc)
0.5 kT=1 (10-16 m2)
0.4 0.4
0.3 kT=10 (10-16 m2)

0.2 0.2
0.1
0
0
0 1 2 3 4 5 6
0 0.2 0.4 0.6 0.8 1
Empirical probability Kc (mm/year1/2)

Fig. 5. P-P plot of the derived model for the carbonation rate. Fig. 7. Density functions of Kc for particular values of kT.

∼ Some plots of Eq. (15) are presented in Fig. 7 for particular values of
values of K ci can be computed and compared with Exp(1) through
∼ kT.
probability and quantile plots. Ranking the computed K ci values in in-
creasing order, the i-th pair of n (sample size) pairs of the probability
plot (Fig. 5) will be 4. Computation of service life

⎛ i , 1 − e−1 ∙ K ci⎞
∼
A simple approach will be described here, assuming that the exact
⎝n + 1 ⎠ (13)
value of the cover thickness xd is known, be it by direct exposure of the
For the quantile plot (Fig. 6), the i-th pair of n pairs will be steel or by means of covermeters (electromagnetic or Ground
Penetrating Radar). A more elaborated approach is being developed,
⎛−ln ⎛ i ⎞ , ∼
⎜ K ci⎞ ⎟ considering xd as a random variable.
⎝ ⎝n + 1⎠ ⎠ (14) We assume that the service life corresponds to the corrosion in-
Both plots lend support to the fairness of the derived model, whilst itiation time Ti; i.e. the time when the carbonation front xc has reached
the two points away from the equality line on the Q-Q plot are pre- the outer surface of the steel xd, where xd is the cover depth. As dis-
sumably outliers. cussed in Section 2.1, this is a rather conservative approach, especially
for elements exposed to relatively dry environments, due to their very
3.5. Conditional density function of the carbonation rate low corrosion propagation rate.
Let us assume that, at a given location within a structure, values of
The former analysis suggests that carbonation rate Kc is a random kT = 1.0 × 10−16 m2 and xd = 16 mm have been measured. Fig. 8 re-
variable that follows a Weibull distribution. Since, for the derived produces in green colour the corresponding density function, already
model, the distribution's scale parameter c depends upon air-perme- shown in Fig. 7.
ability coefficient kT, Eq. (7) may be written as a conditional prob- Three scenarios can be assumed: optimistic, expected and pessi-
ability density function of Kc given kT. Recapping the maximum like- mistic, by taking the Kc fractiles corresponding to P = 25%, 50% and
lihood estimators of the model parameters, the conditional density 75% probability, respectively.
function f (Kc|kT) is given by From Eq. (15), parameters c and d of Weibull distribution (Eq. (6))
can be computed:
2.444
1.823 ⎛ Kc ⎞⎟ (16)
f (K c | kT ) = ∙⎜ d = 2.444
K c ⎝ 0.848∙log10 (kT ) + 2.636 ⎠
2.444 0.848. log10 (kT ) + 2.636
Kc
−0.746 ∙ ⎛
⎜ ⎟
⎞
c=
∙e ⎝ 0.848 ∙ log10 (kT ) + 2.636 ⎠ (15) 1
(17)
0.746 d
1/2 −16 2
with Kc in mm/year and kT in 10 m .
0.4
O25% E50% P75%
6
0.3
Model quantile

f(Kc)

4 0.2

2 0.1

0
0 0 1 2 3 4 5 6 7 8
0 2 4 6 Kc (mm/y½)
Empirical quantile
Fig. 8. Probability density function corresponding to kT = 1.0 10−16 m2 and
Fig. 6. Q-Q plot of the derived model for the carbonation rate. fractiles.

15
R. Neves et al. Cement and Concrete Research 109 (2018) 10–18

100000 1000000
O25% O25%

Initiation Time Ti (years)

Initiation Time Ti (years)

10000 xd = 10 mm E50% 100000 xd = 40 mm E50%

P75% P75%
10000
1000
1000
100
100
10
10

1 1
0.001 0.01 0.1 1 10 100 1000 0.001 0.01 0.1 1 10 100 1000
kT (10-16 m²) kT (10-16 m²)
Fig. 9. Initiation time as function of kT for three scenarios and cover Fig. 11. Initiation time as function of kT for three scenarios and cover
depth = 10 mm. depth = 40 mm.

for kT = 1 Eq. (17) yields c = 2.9718. 5. Application to a real case: Tokyo's Museum of Western Art
The fractile of Weibull distribution corresponding to a P (%) prob-
ability is: Being the only building in Japan designed by Le Corbusier, Tokyo's
Museum of Western Art is a “World Heritage” Building. Due to some
1
P ⎤d symptoms of steel corrosion, a condition survey was conducted when
K c (P ) = c. ⎡−ln ⎛1 − ⎞
the building reached 50 years of age [5]. The Japanese recommenda-
⎣ ⎝ 100 ⎠⎦ (18)
tions for durability design and construction practice of reinforced
In our case, is Kc (25) = 1.78 mm/y1/2, Kc (50) = 2.56 mm/y1/2 and concrete [37] indicate that interventions are to be undertaken when the
Kc (75) = 3.40 mm/y1/2. These values are marked with blue, black and probability of the carbonation depth being equal or greater than the
red vertical lines in Fig. 10, respectively. cover depth reaches 20%.
Now, from Eq. (2), the calculated service life Ti is: Due to the historical value of the building, the drilling of just three
small cores (Ø20 mm) was allowed, on which the carbonation depth xc
2 was measured. An extensive investigation of the building was con-
x
Ti = ⎛ d ⎞ ⎜ ⎟
ducted applying non-destructive techniques: air-permeability kT
K
⎝ c⎠ (19)
(PermeaTORR instrument) and cover thickness xd (GPR instrument).
Three possible values of Ti for the investigated point can be calcu- The combined measurement of xc and kT (where the cores were later
lated: expected, optimistic and pessimistic. Expected value, Ti (50), drilled) allowed a correlation to be established. Imamoto et al. [5] as-
2 sumed a linear relation between Kc = xc/√50 of the three cores and log
corresponding to Kc (50) is Ti (50) = 2.56 = 39 years. ( )
16
(kT), passing through the point kT = 0.1 10−16 m2; Kc = 0 [5], as de-
Similarly, Ti for the optimistic and pessimistic scenarios can be
picted in Fig. 12. It assumed that no carbonation occurs for kT < 0.1
computed, introducing Kc (25) = 1.78 mm/y1/2 and Kc
10−16 m2. This type of relation was established on the basis of research
(75) = 3.40 mm/y1/2, respectively, in Eq. (19). The resulting service
conducted in Japan [21].
lives are Ti (25) = 81 years and Ti (75) = 22 years for the optimistic and
There are 111 pairs of values (kT; xd) measured on the north-side
pessimistic scenarios, respectively.
exterior wall of the building, framed in red in Fig. 13.
Figs. 9 to 11 show the service life Ti for concretes of different air-
To test the approach proposed in Section 4, for each pair of values
permeabilities kT, for cover depths of 10, 25 and 40 mm, respectively.
(kT, xd), the following was done:
The three scenarios: optimistic, expected and pessimistic are plotted for
each case.
1. With each kT, parameter c of the Weibull distribution was com-
puted, applying Eq. (17), with d = 2.444
2. Carbonation coefficient Kc, for P = 25, 50 and 75% was computed,
1000000 applying Eq. (18).
O25%
Initiation Time Ti (years)

100000 xd = 25 mm E50% 6
P75%
10000 5 Test data Kc=0.63ln(kTx1016)+1.40
Prediction model
Kc (mm/year1/2)

1000 4

100 3
Mortar finishing
10 2

1 1
0.001 0.01 0.1 1 10 100 1000
kT (10-16 m²) 0
0.001 0.01 0.1 1 10 100 1000
Fig. 10. Initiation time as function of kT for three scenarios and cover kT (×10-16 m2)
depth = 25 mm. (For interpretation of the references to colour in this figure,
the reader is referred to the web version of this article.) Fig. 12. Correlation between Kc and kT established by Imamoto et al. [5].

16
R. Neves et al.
Fig. 13. kT testing on N-side external wall of Tokyo's Museum of Western Art.
Imamoto et al.[5]
Fig. 14. Probability of Ti for different scenarios, compared to that presented in
[5].
3. With xd and each Kc, the initiation time Ti is computed for the op-
timistic, expected and pessimistic scenarios, applying Eq. (19).
4. The 111 values of Ti, for each scenario, are sorted and the prob-
ability of occurrence is calculated for each of them.
Fig. 16 shows the probability curves of Ti obtained for each sce-
nario: pessimistic (P75%), expected (E50%) and optimistic (O25%).
Fig. 14 also shows (white dots on green curve) the probability curve
computed by Imamoto et al. [5]. The agreement with the values com-
puted with the method proposed in this paper, for the “expected” sce-
nario (E50%), is remarkable, considering that the calculations were
made following completely different assumptions and procedures. It is
worth noticing that both curves in Fig. 14 predict a small probability of
occurrence of corrosion initiation at the time of the investigation
(50 years), which is compatible with the localized corrosion damage
detected during the survey of the building.
Indeed, Imamoto et al. [5] based their calculation on a simple re-
lation, based on the results of just 3 cores taken from the Museum
(Fig. 12), whilst the model proposed in this paper is based on a statis-
tical treatment of a large number of measurements obtained on many
different structures (Fig. 1).
17
Cement and Concrete Research 109 (2018) 10–18
The assessment for the “pessimistic” scenario (P75%) looks rea-
sonable, whilst that for the “optimistic” scenario seems too optimistic
and requires a revision.
6. Conclusions
A method to assess the residual service life of structures exposed, for
relatively long periods, to environments that facilitate the penetration
of CO2, is presented.
The method is based on the determination, on site, of the air-per-
meability and cover depth, in an entirely non-destructive manner. In
this way, the damaging and costly drilling of cores to assess the car-
bonation depth is entirely avoided or strongly reduced.
The method is based on a large data base of paired measurements of
carbonation depth and air-permeability conducted on several concrete
structures (predominantly 10–60 years old) located in Japan, Portugal
and Switzerland.
The uncertainty of the assessment grows with the permeability of
the concrete measured. A statistical analysis was made, to obtain a
Weibull distribution for the carbonation coefficient Kc (mm/y1/2),
having a constant shape parameter and a scale parameter that is
function of the measured air-permeability kT.
The probability distribution of Kc allows the estimation of the cor-
rosion initiation time Ti for different scenarios, e.g. optimistic, expected
and pessimistic.
The approach was applied to the test results obtained on the em-
blematic Tokyo's Museum of Western Art, with the “expected” scenario
assessment in very close agreement to that made by an entirely different
method. The prediction for the “optimistic” scenario looks too opti-
mistic and should be revised.
The proposed approach is expected to provide reasonable assess-
ments of the corrosion initiation time of concrete surfaces exposed to
outdoors climate (as were the ones on which the experimental work
was conducted). Assessing the residual service life involves assessing
also the corrosion propagation time, the modelling of which is still not
fully developed. Assimilating the corrosion initiation time with service
life is reasonable for outdoors' exposed surfaces (e.g. Exposure Class
XC4 of EN 206), where available moisture and oxygen leads to
R. Neves et al. Cement and Concrete Research 109 (2018) 10–18

relatively short propagation times. For exposure conditions deprived of Resilience of Bridges and Bridge Networks - Proceedings of the 8th International
either of them, that assimilation may become too conservative. Conference on Bridge Maintenance, Safety and Management, IABMAS 2016,
(2016).
[19] R. Torrent, G. Frenzer, Methoden zur messung und beurteilung der kennwerte des
References uberdeckungsbetons auf der baustelle - Teil 2, Bern, (1995).
[20] F. Jacobs, Beton zerstörungsfrei untersuchen, 3 Der Bauingenieur, 2008, pp. 24–27.
[21] K. Imamoto, K. Shimozawa, M. Nagayama, J. Yamasaki, A. Tanaka, Relationship
[1] CPC-18 measurement of hardened concrete carbonation depth, Mater. Struct. 21 (6)
Between Air-permeability and Carbonation Progress of Concrete in Japan, RILEM
(Nov. 1988) 453–455.
International workshop on performance-based specification and control of concrete
[2] L.A. Rojas, Estudio de la durabilidad de estructuras antiguas de hormigón armado,
durability, (2014), pp. 325–332 , June.
con énfasis en la corrosión de las armaduras, Universidad de Chile Facultad de
[22] R. Neves, Concrete Air Permeability and Carbonation in Structures (in Portuguese),
Ciencias Físicas y Matemáticas, 2006.
Technical University of Lisbon - Instituto Superior Técnico, 2012.
[3] M. Te Liang, R. Huang, S.A. Fang, Carbonation service life prediction of existing
[23] R. Neves, B.S. Da Fonseca, F. Branco, J. De Brito, A. Castela, M.F. Montemor,
concrete viaduct/bridge using time-dependent reliability analysis, J. Mar. Sci.
Assessing concrete carbonation resistance through air permeability measurements,
Technol. 21 (1) (2013) 94–104.
Constr. Build. Mater. 82 (2015).
[4] P.J. Parrott, Design for avoiding damage due to carbonation-induced corrosion,
[24] R. Neves, F. Branco, J. De Brito, Field assessment of the relationship between
Spec. Publ. 145 (1994) 283–298.
natural and accelerated concrete carbonation resistance, Cem. Concr. Compos. 41
[5] K. Imamoto, A. Tanaka, M. Kanematsu, Non-destructive Assessment of Concrete
(2013) 9–15.
Durability of the National Museum of Western Art in Japan, The Second
[25] H. Tabesh, S. Ayatollahi, M. Towhidi, A simple powerful bivariate test for two
International Conference of Microstructural-related Durability of Cementitious
sample location problems in experimental and observational studies, Theor. Biol.
Composites, (2012), pp. 752–759.
Med. Model. 7 (1) (2010) 13.
[6] R.J. Torrent, Y. Taibi, J. Armaghani, Evaluation of port of Miami tunnel segments,
[26] G.A. Marcoulides, S.L. Hershberger, Multivariate Statistical Methods: A First
Concr. Int. 35 (5) (2013).
Course, Psychology Press, 1997.
[7] SIA, SN 505 262/1 Construction en béton - Spécifications complémentaires. Annexe
[27] J.L. Devore, Probability and Statistics for Engineering and the Sciences, Brooks/
E: Perméabilité à l'air dans les structures. Switzerland, (2013).
Cole Cengage Learning, 2012.
[8] T. Teruzzi, Estimating the Service-life of Concrete Structures Subjected to
[28] L. Nilsson, T. Luping, Relations between different transport parameters, in:
Carbonation on the Basis of the Air Permeability of the Concrete Cover,
J. Kropp, H.K. Hilsdorf (Eds.), Performance Criteria for Concrete Durability, E and
EUROINFRA International ECCE Conference, (2009).
FN Spon, 1995, pp. 15–32.
[9] K. Tuutti, Corrosion of Steel in Concrete, Swedish Cem. Concr. Res. Inst. (1982), p.
[29] International Federation for Structural Concrete Task Group 5.6 Model Code for
469.
Service Life Design of Concrete Structures, Model Code for Service Life Design, 34
[10] Comité Européen de Normalisation, Concrete - Part 1: Specification, Performance,
International Federation for Structural Concrete (fib), 2006.
Production and Conformity, Eur. Stand. EN 206-1, (2000), pp. 1–72.
[30] P.F. Marques, A. Costa, Service life of RC structures: carbonation induced corrosion.
[11] R. Torrent, L. Fernandez Luco, Service Life Assessment of Concrete Structures Based
Prescriptive vs. performance-based methodologies, Constr. Build. Mater. 24 (3)
on Site Testing, Proceedings of the XIII DBMC International Conference on
(2010) 258–265.
Durability of Building Materials and Components, (2014), pp. 801–809.
[31] F. Lollini, E. Redaelli, L. Bertolini, Analysis of the parameters affecting probabilistic
[12] J. Walvaren, Model Code 2010, final drafts, FIB Bull. 1 & 2 (65 & 56) (2012) 105.
predictions of initiation time for carbonation-induced corrosion of reinforced con-
[13] B. E. III, E. Union, DuraCrete, Modelling of Degradation: DuraCrete, Probabilistic
crete structures, Mater. Corros. 63 (12) (2012) 1059–1068.
Performance Based Durability Design of Concrete Structures, European Union,
[32] J. Aldrich, R.A. Fisher and the making of maximum likelihood 1912–1922, Stat. Sci.
Netherlands, 1998.
12 (3) (Sep. 1997) 162–176.
[14] M. Fomento, Instrucción de Hormigón Estructural (EHE-08), (2008), pp.
[33] G. Tzavelas, A study of the number of solutions of the system of the log-likelihood
35176–35178.
equations for the 3-parameter Weibull distribution, Appl. Math. 57 (5) (Oct. 2012)
[15] H.J. Wierig, Longtime Studies on the Carbonation of Concrete Under Normal
531–542.
Outdoor Exposure, Proceedings of the Rilem Seminar on the Durability of Concrete
[34] L.S. Lasdon, A.D. Waren, A. Jain, M. Ratner, Design and testing of a generalized
Structures Under Normal Outdoor Exposure, (1984), pp. 239–249.
reduced gradient code for nonlinear programming, ACM Trans. Math. Softw. 4 (1)
[16] R.J. Torrent, A two-chamber vacuum cell for measuring the coefficient of perme-
(Mar. 1978) 34–50.
ability to air of the concrete cover on site, Mater. Struct. 25 (6) (Jul. 1992)
[35] W.H. Greene, Econometric Analysis, 97 (2003), p. 457.
358–365.
[36] N.L. Johnson, S. Kotz, N. Balakrishnan, Continuous Univariate Distributions, 2nd
[17] B. E. III, E. Union, DuraCrete, Compliance Testing for Probabilistic Design Purposes,
ed, Vol 1 Wiley, 1994.
(1999).
[37] Architectural Institute of Japan, Recommendations for Durability Design and
[18] K.-I. Imamoto, R. Neves, R. Torrent, Carbonation Rate in Old Structures Assessed
Construction Practice of Reinforced Concrete, (2004).
With Air-permeability Site NDT, Maintenance, Monitoring, Safety, Risk and

18