Carbonation Induced Corrosion Article 1

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 30

Received: 5 April 2019 | Accepted: 25 April 2019

DOI: 10.1002/maco.201910962

ARTICLE

Quantification of Tafel coefficients according to passive/active


state of steel carbonation‐induced corrosion in concrete

Frédéric Duprat1 | Thomas de Larrard1 | Ngoc Tru Vu2

1
LMDC, Université de Toulouse,
Toulouse, France
Statistical quantification of Tafel coefficients is investigated in this study for
2
Faculty of Bridge and Roads, National isolated steel rebar embedded in concrete. The survey is supported by a wide
University of Civil Engineering, Hanoï, experimental campaign carried out earlier to characterize the passive and active
Vietnam
states of carbonation‐induced corrosion of steel. Electrochemical measurements
Correspondence (polarization resistance, corrosion potential, Tafel coefficients) and gravimetric
Frédéric Duprat, LMDC, Université de estimations of iron loss were regularly conducted over 417 days on 108 concrete
Toulouse, INSA/UPS, 135 avenue de
specimens. The statistical analysis reveals that the mean value of Tafel
Rangueil, Toulouse, F31077 Cedex,
France. coefficients, both cathodic and anodic, is higher under active corrosion, which
Email: frederic.duprat@insa-toulouse.fr seems to contradict the general tendency found for chloride‐induced corrosion,
while their coefficient of variation is smaller. The statistical inference was based
on the first step of distributions fitting the experimental data and then on the
second step of goodness‐of‐fit tests. The most suitable of the distributions
proposed were the Burr, Rayleigh, and Gamma distributions. A similar analysis
was made for the corrosion potential and polarization resistance. The findings
of the study will be valuable for probabilistic approaches to corrosion where
probabilistic distributions are required.

KEYWORDS
carbonation‐induced corrosion, concrete, statistical analysis, Tafel slopes

1 | INTRODUCTION with the conservation of electrical charge according to


Laplace's law and Butler–Volmer's equations as electrical
Due to its crucial importance regarding the durability of boundary conditions at the steel surface.[5–7] The reduc-
reinforced and prestressed concrete structures exposed to tion of oxygen at the cathode is also taken into account by
carbonation, the corrosion of embedded steel reinforce- the limiting corrosion current density at the cathode.[8–10]
ments has become of great interest and is the subject of Finally, some authors have proposed to combine these
much research from both the experimental and modeling approaches by taking account of the oxygen flow at both
points of view. Two main approaches can be distin- cathode and anode.[11,12] Otieno et al.[13] presented a
guished for modeling the propagation of corrosion. The critical review of some of these models and recom-
first is based on the assumption that the corrosion mended, among other needed developments, to account
kinetics is governed by the amount of oxygen at the for the variability of the parameters influencing the
cathode. The reduction of oxygen at the surface of corrosion current thanks to a probabilistic approach,
the sound steel sustains the corrosion current, which can from the moment that adequate data is available.
be calculated from the consumed oxygen flow.[1–4] The Tafel constants occur in Butler–Volmer equations,
second approach is supported by Ohm’s law together governing the polarization of steel and impacting the

Materials and Corrosion. 2019;1–30. www.matcorr.com © 2019 WILEY‐VCH Verlag GmbH & Co. KGaA, Weinheim | 1
2 | DUPRAT ET AL.

corrosion kinetics. They are hence significant input precracked or cracked during the development of
parameters in the models based on the second approach corrosion. According to Michel et al,[10] these diverse
and in further numerical prediction of the corrosion experimental factors or situations contribute to the width
process in reinforced concrete structures. Ge et al,[14] for of ranges of Tafel constants proposed in the literature.
instance, pointed out from a parametric study that the Other factors affecting the Tafel slopes can be encoun-
cathodic Tafel slope was one of the parameters having tered in the phenomenon of corrosion itself, as men-
the most influence on the corrosion rate. Four main tioned by Babaee and Castel[17] from previous studies:
regions for the values of Tafel constants have been Conductivity, pH and ionic composition of the electro-
described by Song[15] as follows: lyte, polarization mechanism, corrosion potential, surface
film of mill scale covering the surface of the reinforce-
● Moderate values of both anodic βa and cathodic βc ment. It would likely be valuable to classify the data
Tafel slopes: Steel is in an active state, oxygen according to these factors so as to derive average
diffusion does not control the corrosion reaction tendencies and specific possible ranges of Tafel constants,
and the pH value is below 9. Typical expected but this would also artificially reduce their variability and
ranges would be (in mV/dec) 13 ≤ βa ≤ 17 and limit their use in modeling when these factors are not
21 ≤ βc ≤ 65. known in advance. Unavoidable uncertainties resulting
● A moderate value of the anodic Tafel slope and a high from the electrochemical measurements, the assessment
value of the cathodic one: Steel is in an active state of Tafel constants from these (within the framework of
and the cathodic reaction is under oxygen diffusion modeling assumptions), and also the random nature of
control. Typical expected ranges would be (in mV/ the phenomena involved are indeed also responsible for
dec) 13 ≤ βa ≤ 17 and βc → ∞. such a variation in Tafel constants. Hence the notion of
● A moderate value of the cathodic Tafel slope and high Tafel coefficients rather than Tafel constants should be
value of the anodic one: Steel is in a passive state and introduced, given that these parameters are essentially
the supply of oxygen is sufficient. Typical expected randomly variable.
ranges would be (in mV/dec) βa → ∞ and Among the various experimental investigations men-
21 ≤ βc ≤ 52. tioned above, the effective active or passive state of steel
● High values of both cathodic and anodic Tafel slopes: has received little attention. Most of the time it has been
Steel is in a passive state and the cathodic reaction is assumed that steel is in the active state. However, in real
under oxygen diffusion control. Typical expected concrete structures, the corrosion process generally
ranges would be βa → ∞ and βc → ∞. involves both passive and active zones of the embedded
reinforcing steel cage. When modeling the corrosion
Numerous values of Tafel constants are collected and phenomenon at the structural scale, it is thus necessary
reported in.[9,10,16] Many of them were proposed in a for Tafel coefficients to be supplied for both the passive
modeling framework and none of them was the outcome and active state.[6] It has to be mentioned that, in some
of an experimental investigation devoted to carbonation‐ recent investigations,[16,17] passive and active states of
induced corrosion. It is nevertheless interesting to recall corrosion were addressed and the experimentally obtained
some prominent features that were observed. Experi- values of Tafel coefficients were attributed accordingly to
mental values of Tafel slopes obtained on supposed one state or the other. These studies mainly focused on
actively corroding samples, according to the various characterizing the chloride‐induced corrosion of steel
studies cited and adding the results of the authors reinforcement for nonusual concretes like low‐calcium
themselves, may vary from 10 to 1,215 mV/dec for the fly ash‐based geopolymer concrete[17] or hybrid fiber‐
cathodic Tafel constant and from to 10 to 1,570 mV/dec reinforced concrete.[16] Cathodic Tafel slopes obtained by
for the anodic Tafel constant. These ranges do not match Babaee and Castel[17] varied from 30 to 45 mV/dec for
the theoretical ones given by Song[15] and are somewhat passive samples and from 106 to 221 mV/dec for active
more widely spread. Various experimental procedures samples, and the anodic Tafel slope exhibited an infinite
were applied to initiate and/or accelerate corrosion value for passive samples, while it varied from 514 to
(additional chlorides in mortar or concrete mix or 1,107 mV/dec for active samples with, again, a possible
exposure to sea water, imposed anodic current), and trend to infinity. Nguyen et al.[16] did not state the
the steel rebars were embedded in different media amplitude of variation of their Tafel slopes and mentioned
(synthetic pore solution, mortar or concrete). Mortar or only the mean values. From Figure 10 in their paper, it
concrete was batched with various cementitious mixes can be estimated that the cathodic Tafel slope ranged from
based on ordinary or blended cement, with various 2 to 60 mV/dec for passive samples, and from 80 to
water/binder ratios. Some specimens were intentionally 125 mV/dec for active samples, whereas anodic Tafel slope
DUPRAT ET AL. | 3

FIGURE 1 Layout of concrete


specimens. Reproduced with
permission,[19] (Copyright 2013, Taylor &
Francis)

ranged from 200 to 1,000 mV/dec for passive samples and 2 | EXPERIMENTAL
from 150 to 205 mV/dec for active samples, except for one INVESTIGATION
point at 600 mV/dec.
In a recent critical review on corrosion rate in The experimental investigation aimed to analyze the
carbonated concrete, Stefanoni et al.[18] identified sys- conditions of effective onset of carbonation‐induced
tematic data collection as being one of the upcoming corrosion for small steel rebar embedded in concrete at
research needs. Such recording of electrical resistivity, controlled cover depth.[19] For the sake of clarity, the
corrosion potential and corrosion rate of the steel, and main features are recalled here. Concrete specimens of
also oxygen diffusion and consumption rate, together dimensions 14 × 12 × 12 cm3 (Figure 1) were cast with
with concrete properties like porosity, would allow the three concrete mixes comprising cement CEM I,
mechanism of steel corrosion (another very notable CEM II or CEM III (Table 1). The steel rebar (usual
research need) to be studied with more insight. Surpris- commercial carbon steel) was laid out by means of PVC
ingly, Tafel slopes do not appear in the data to be tubes fixed on the mold. Before being installed in the
collected, even though they are needed as input mold, each rebar was brushed to eliminate preformed
parameters of models as mentioned above. It is ques- corrosion products, which could have modified the
tionable whether Tafel slopes already obtained in the case gravimetric iron mass loss measured at the end of the
of chloride‐induced corrosion, results undoubtedly test, and then weighed.
known by the authors, would be suitable in the case of After demolding, specimens were wet cured for 7
carbonation‐induced corrosion, because chloride ions act days, stored in the laboratory atmosphere for 1 day, and
beyond the depassivation phase by reducing the concrete then dried at 80°C until an average saturation degree of
resistivity to a certain extent and contributing to the 45% was reached for each specimen. Then 90 specimens
newly formed oxides in the course of corrosion propaga- were placed in an accelerated carbonation chamber
tion, which can modify the polarization behavior of the (labels D15C and D30C) with a dioxide carbon partial
steel.[17] pressure equal to 50% at 20°C whereas 18 specimens
The main goal of the present study is to estimate, remained unexposed (label D15NC). The duration in the
collect, and analyze the Tafel coefficients obtained in the carbonation chamber was adjusted so that the probability
case of reinforced concrete exposed to carbonation. of the cover being fully carbonated was 95% for D15C
Concrete mixes were prepared with three types of specimens and 50% for D30C specimens. The lateral faces
cement, and two cases of cover depth combined with of specimens were coated with epoxy resin to prevent the
precarbonation phases and three exposure conditions
were considered, so as to cover realistic onsite situations
T A B L E 1 Mix proportions
of concrete structures. Electrochemical measurements of
corrosion potential and polarization resistance were Concrete designation
Constituents/
made at different levels of corrosion and Tafel coefficients
properties CI CII CIII
were derived from them. One hundred and eight concrete
probes, each containing small steel rebar, were cast. Type of CEM I CEM II CEM III
Statistics of Tafel coefficients were established consider- cement Unit 52.5R 32.5N 52.5N
ing the passive or active state of the steel. Cement kg/m3 295 430 355
The present study contributes to fairly rich knowledge Sand kg/m3 989 760 957
of Tafel coefficients and its results can be fruitfully Gravel kg/m 3
792 912 765
incorporated and used in advanced modeling of Water L/m3 200 210 199
corrosion.
4 | DUPRAT ET AL.

ingress of carbon dioxide and to impose uniaxial irreversible destruction. The CPT was therefore the last
carbonation through the controlled cover depth. Three electrochemical test carried out on specimens and each
exposure conditions were then applied to specimens for a pair of Tafel coefficients corresponded to a single specimen
maximum duration of 417 days as follows: at the end of its testing. This may be deemed as a drawback
of our protocol since it did not enable the change of
– Similar to XC1 exposure class of EN 206: Specimens corrosion state to be registered and observed for the same
were left in a 10 mm water layer immersing the side specimen. In a recent study, Nguyen et al. [16] applied a
opposite to the cover, in fairly constant conditions of smaller potential sweep of +/–75 mV in a CPT so as to limit
relative humidity and temperature, leading to a possible damage after the test and reuse the same specimen,
saturation degree of about 85% in the neighborhood while maintaining a sweep large enough for the Tafel
of the reinforcement. coefficients to be extracted. However, those authors did not
– Similar to XC4 exposure class of EN 206: The compare the resulting Tafel coefficients to those that would
specimens were subjected to alternating dry (5 days have been obtained under a wider potential sweep to
at RH = 60%, T = 40°C), and wet (2 days at RH = support their protocol. Babaee and Castel [17] argued that
100%, T = 20°C) periods. conducting CPT with overpotential of +/–200 mV three to
– Natural outdoor exposure conditions (exposed con- five times on the same sample was possible without causing
crete surface was vertical and unsheltered). dramatic deterioration of the electrochemical system. The
current and polarization resistance provided through the
The flowchart of the experimental campaign can be experimental campaign is given by the measuring devices
seen in Figure 2. respectively in A and Ω, and not in A/m2 or Ω/m2 which
Electrochemical measurements were made by means of are the units usually used, because the effective polarized
a Gamry System G300, between 83 and 417 days of area of the rebar is not accurately known. The maximum
exposure, so as to observe various levels of corrosion. A polarized area can be taken equal to the contact area
three‐electrode system was used, comprising a saturated between the rebar and bulk concrete (7.54 cm2).
calomel reference electrode (SCE), a titanium mesh as The amount of iron lost was also found by weighing
counter electrode, both installed on the upper concrete the extracted rebar after the electrochemical tests. Mass
surface over the rebar (see Figure 1), and the rebar as loss measurement was performed according to the ASTM
working electrode (WE). Measurements were made from G1‐90 protocol.[20]
the surface of concrete on the face corresponding to the
controlled cover. The linear polarization resistance (LPR)
method allowed the polarization resistance, Rp, to be 3 | ASSESSMENT OF THE TAFEL
measured providing that the open circuit corrosion COEF FICIE NTS
potential Ecorr , corresponding to the stabilized half‐cell
potential in an open circuit measurement, was obtained The electrochemical corrosion cell can basically be seen
beforehand. In LPR, the potential sweep was +/–10 mV as an equivalent electrical circuit comprising three serial
with respect to the corrosion potential at a rate of 0.1 mV/s. resistance terms R a , R c , and Rcon [21,22] that are part of
IR compensation based on current interruption was the global resistance R cell .
applied during the test to dynamically correct the
uncompensated electrolyte resistance. The cyclic polariza- R cell = R a + R c + R con (1)
tion test (CPT), with an amplitude of polarization up to
+/–250 mV with respect to the corrosion potential, served R a and R c refer to the charge transfer process of the
to determine the Tafel coefficients thanks to a curve fitting anodic and cathodic reactions, respectively, while R con
procedure. With such a potential sweep, it was possible refers to the flow ions through the concrete electrolyte. The
that the electrochemical system of the WE might undergo driving cell voltage Ucell is defined as the difference between

FIGURE 2 Flowchart of experiments for CI concrete – DxC: x mm cover depth carbonated. (Reproduced with permission,[19] Copyright
2013, Taylor & Francis)
DUPRAT ET AL. | 5

the reversible equilibrium potentials of the cathodic (Ec,0 ) are randomly dispersed on the steel surface, the ohmic
and anodic (Ea,0 ) reactions, and is increased by the effective drop can be considered negligible because no current
ohmic drop over the concrete electrolyte (Ucon ). flows through the concrete bulk (Ucon ≈ 0). At free
corrosion potential, anodic and cathodic net currents
Ucell = Ec,0 − Ea,0 + Ucon. (2) are balanced and the corrosion current is

The reversible potentials are expressed as i corr = ia = −ic. (7)

0
ln(10) RT Under the assumption of a microcell corrosion system,
Ea,0 = EFe 2+ / Fe + log (Ksp)
2F the Tafel coefficients can be assessed by fitting the
ln(10) RT experimental polarization curve of the steel. Once Ecorr
+ (14 − pH ) (3)
F and Rp have been acquired, assuming a pair of Tafel
coefficients, the corrosion current is expressed as
and
i corr = B / Rp (8)
ln (10) RT
Ec,0 = EO02 / OH−
+ log(pO2 )
4F where B is the Stearn–Geary coefficient.
ln(10) RT
+ (14 − pH ) (4)
F βa βc
B= . (9)
ln (10) (βa + βc )
where Ksp is the dissolution constant of ferrous hydroxide
(Ksp = 7.94 × 10−16), pO2 is the partial pressure of
dioxygen (pO2 = 0.21), F is the Faraday constant The exchange currents hold
(F = 96485 C ), and R is the perfect gas constant
(R = 8.314 J / K ). With respect to the SCE, i corr
ia,0 = (10)
0
reference potentials are EFe 2+/ Fe = − 687 mV and exp (ln(10) ηa / βa)
EO02 / OH− = + 154 mV .
If polarization is applied to the cell, the potential of and
each reaction shifts to a polarized potential. As both
reactions occur inside the same corrosion cell, the steel i corr
ic,0 = (11)
reinforcement potential, Ecorr takes an intermediate value exp ( −ln(10) ηc / βc )
between the two equilibrium potentials, resulting in the
activation overpotentials ηa = Ecorr − Ea,0 and Experimental and modeled polarization curves ob-
ηc = Ecorr − Ec,0 . If electrode reactions are assumed to tained from the fitting procedure are plotted in Figure 3
be governed by the activation energy of the charge for a specimen taken from the experimental campaign
transfer, the net current density is expressed by the mentioned above. The data reported in Figure 3 were
Butler–Volmer equation for each reaction. derived from Equations (3–11), except Rp and Ecorr , which
were measured previously. It can be seen that experi-
⎛ η ⎞ mental and modeled curves almost overlap each other.
ia = ia,0 exp ⎜ln(10) a ⎟ (5) Conversely, when anodic and cathodic areas are
⎝ βa ⎠
remote due to local differences in the electrochemical
state of the steel, galvanic corrosion takes place in a
and
corrosion macrocell, and the ohmic drop can no longer
be ignored. Because of the network of electrically
⎛ η⎞
ic = −ic,0 exp ⎜ −ln(10) c ⎟ (6) connected reinforcing steel bars and the inhomogeneous
⎝ βc ⎠ condition states of both concrete and steel (due to
concrete moisture, pH of the pore solution, quality of
where ia,0 and ic,0 are the exchange current densities, and the steel‐concrete interface, dioxygen content, cover
βa and βc are the Tafel coefficients. depth, cracking, etc.) macrocell and microcell systems
The importance of the ohmic drop depends on the are combined in concrete structures most of the time, in a
type of corrosion cell. In the case of uniform corrosion general macrocell system where electrical coupling exists
where microscopic adjacent anodic and cathodic areas between active and passive uniform corrosion systems,
6 | DUPRAT ET AL.

FIGURE 3 Polarization curve of the steel (specimen 629)

which are mutually polarized.[6,23] Anodic and cathodic jP = jcorr, P [exp (ln(10) ηP / βa, P ) − exp ( −ln(10) ηP / βc, P ) ]
polarizations are possible for each active or passive area (15)
but, because of the difference between the corrosion
potentials (Ecorr, A in active area is much lower than Ecorr, P where
in passive area) the active area undergoes anodic
polarization while the passive area undergoes cathodic ηA = EA − Ecorr, A (16)
polarization. The steel equilibrium potentials for each
area (EP for the passive area and EA for the active area) and
are such that the macrocell current intensity im remains
the same under cathodic or anodic polarization. ηP = EP − Ecorr, P (17)

Ucon = im R con = EP − EA. (12)


The residual microcell current results from the
difference between the total anodic current and the net
As polarized steel surfaces generally have various anodic current.[6]
extents, anodic and cathodic net currents are no longer The Tafel coefficients introduced in Equations (14)
balanced. According to Laurens et al.,[6] the corrosion and (15) relate to the passive or active state of the steel. It
current can be expressed as is not straightforward to assess the effect of the ohmic
drop from a CPT, because the steel gives a global
jcorr = im / AA + i micro/ AA = jm + jmicro (13) response to the polarization. The classical representation
of a polarization curve shown in Figure 3 thus remains
where AA is the extent of the active area and i micro is the the same whatever the state of the steel and, if no change
residual microcell current intensity. Butler–Volmer occurs except that the state of the steel shifts from passive
equations apply for each area to active, the corrosion point is expected to be localized in
a more anodic region (decrease of Ecorr and increase of
jA = jm = jcorr, A [exp (ln(10) ηA / βa, A) i corr ) which could correspond to a decrease of the anodic
Tafel coefficient, whereas the cathodic Tafel coefficient
− exp ( −ln(10) ηA / βc, A) ] (14)
varies only slightly. This is illustrated in Figure 4 for two
probes of the experimental campaign. The translation of
and the corrosion point towards the anodic region from the
DUPRAT ET AL. | 7

FIGURE 4 Polarization curves. A: Active steel (specimen 505), P: passive steel (specimen 521)

passive state (P) to the active state (A) is clearly visible. and it can be judged that case (A) in Figure 4 is a fairly
The change in the Tafel coefficients also complies with good example of the maximum deviation that was
the generally expected trend. This motion from point P to observed in this study between modeled and experi-
point A was also shown in the case of chloride‐induced mental polarization curves and, in most cases, this
corrosion,[16] in a more pronounced manner regarding deviation remains acceptable.
the deviation in the corrosion current and corrosion Fitting the steel polarization curve is a usual and
potential between the passive and active state. As far as convenient way to estimate the Tafel coefficients,[23] but
the fitting procedure is concerned, it can be seen that other electrochemical results can also be used, in
there is a slight discrepancy between the modeled combination with modeling. In Michel et al.[10] the
polarization curve and the experimental one in the case experimental corrosion current i corr , derived from the
of the active state (A) compared to the passive state (P), polarization resistance, and corrosion potential Ecorr were
where the modeled curve is, on the contrary, very close to targeted in a fitting procedure where comprehensive
the experimental one. Such an observation is not finite element modeling of the experiment was proposed.
systematic and the quality of the fitting generally does Tafel coefficients were adjusted so that the computed
not depend on the steel state, providing that the extent of values of i corr and Ecorr matched experimental values. In
the experimental region experiencing the Tafel electro- this study i corr was estimated by Equation (8) with the
chemical behavior of steel is large enough. Chang Stearn–Geary coefficient B = 26 mV while the value of B
et al.[23] reported that the passive film of oxide formed calculated according to Equation (9) with the adjusted
around the rebar could have a nonnegligible effect on the Tafel coefficients varied from 2.2 to 30.9 mV. There is
corrosion process and makes the polarization curve hence a slight divergence and final results do not concord
deviate from its theoretical shape given by Equation perfectly with the initial assumption. In our case, no
(15). Similarly, the growth of the layer of corrosion additional assumption was needed before the fitting
products, together with the various oxides formed during procedure. Jiang et al.[24] proposed estimating the sum
the corrosion process, may affect the polarization β = βa + βc from experimental results of i corr versus time
behavior of steel and consequently the values of the for three different concrete mixes. An analytical model of
Tafel coefficients.[24] In the present study, the maximum i corr was developed, accounting for the electrical resistiv-
exposure time of the concrete specimens to natural (not ity of the concrete and the W/C ratio, under the following
accelerated) conditions of corrosion was 417 days. The assumptions: Microcell corrosion system with equal
corrosion layer was thus very thin at the end of this time activated anodic and cathodic surfaces, and equal degrees
8 | DUPRAT ET AL.

of corrosion control by both the anode and the cathode, Nguyen et al.[16] classified their samples with respect to
implying that λ a = λ c ≈ 0.5, where λ a = βa/ (βa + βc ) and the corrosion potential as passive when Ecorr > −100 mV/
λ c = βc / (βa + βc ). These assumptions are not very well SCE and active when Ecorr < −300 mV/SCE. However,
suited to our case and particularly the latter as the mean the use of half‐cell potential (corrosion potential) as a good
values of λ a and λ c are 0.67 and 0.33, respectively. indicator of corrosion activity has recently been criti-
cized[29] in the case where a macrosystem of corrosion can
occur, for instance, in a grid of interconnected rebars.
4 | DISTINCTION B ETWEEN Published recommendations have mostly been based on
A C T I V E A N D PA S S I V E C O R R O S I O N experimental investigations involving a single steel bar.
STATES Sassine et al.[29] pointed out that a corrosion potential
higher than –128 mV/SCE could be due either to a
As mentioned earlier, it is necessary to use consistent complete passive steel reinforcement or to corrosion onset
values of Tafel coefficients for the active and passive associated with a macrocell system having a large cathode‐
states of the steel when the aim is to model the overall to‐anode surface ratio, resulting in high corrosion kinetics.
phenomenon of corrosion in concrete structures where In contrast, a low corrosion potential may be associated
macrocell systems have developed. Looking at the shape with a macrocell system having a small cathode‐to‐anode
of the steel polarization curves obtained by a CPT is not surface ratio, resulting in slow corrosion kinetics. A
totally suitable for this purpose as it remains similar realistic reinforcement layout was used in this study,
whatever the state of the steel (passive or active). Criteria which allowed rebars to be isolated or connected.
other than this shape, based on electrochemical proper- Moreover, electrochemical properties indicate an
ties likely to change depending on the steel state, must instantaneous state of corrosion at the time when the
then be resorted to to properly distinguish the passive test is performed whereas the effective state of corrosion
from the active state. The link between the electrochemi- results from an additive process where favorable condi-
cal properties of steel not coated in concrete and corrosion tions must be brought together (depassivation, moisture,
activation has been under investigation for several decades presence of oxygen at the cathode). From this viewpoint,
and recommendations have been proposed for testing and the iron mass loss would provide consistent evidence of
interpreting measurements.[25–27] Half‐cell potential (cor- corrosion activity if measured over a given time in
rosion potential), corrosion rate, concrete resistivity, and conjunction with electrochemical properties. Unfortu-
polarization resistance are the main properties usually nately, its measurement is rather complex and laborious, and
estimated to characterize the corrosion activation. The use it thus be performed only under laboratory conditions.
of the gradient of corrosion potential rather than its Derived from the concepts of probability of detection and
absolute value is also recommended by Elsener et al. [28] probability of false alarm an approach was proposed
for detecting the corrosion spot location in dense assessing the probabilities of right or wrong alarm to
reinforcement grids. Nevertheless, the random nature of establish a correlation between the effective state of
the phenomenon, together with experimental uncertain- corrosion, measured by the iron mass loss, the polarization
ties, hinders any deterministic decision on the steel state resistance Rp and/or corrosion potential Ecorr .[19] Thanks to
from the values of electrochemical properties. Recommen- this approach, threshold values for these electrochemical
dations proposed in the literature hence mention a level of properties were suggested so that the probability of a right
corrosion risk or corrosion probability as proposed in alarm (PRA) was maximum, allowing the effective onset of
ASTM C876‐09 [25] and Vennesland et al.[27] Some of them corrosion to be identified with a minimum probability of a
are reported in Table 2. wrong alarm (PWA). For example, according to Rp,

T A B L E 2 Corrosion risk according to corrosion parameters

RILEM[27] ASTM[25]

Net corrosion current Yearly corrosion rate Concrete resistivity Half‐cell potential
jcorr (μA/cm2) Vcorr (mm/y) ρcon (kΩ.m) Ecorr (mV/SCE) Corrosion risk
<0.1 <0.001 >100 >–128 Negligible
0.1–0.5 0.001–0.005 50–100 Low (uncertain)
0.5–1 0.005–0.01 10–50 Moderate (uncertain)
<1 >0.01 <10 <–278 High
Abbreviation: SCE: saturated calomel reference electrode.
DUPRAT ET AL. | 9

T A B L E 3 Threshold values of electrochemical properties The surface density of iron mass loss in Table 3 was
determined under the assumption that the total interface
Reference properties for
area between steel and concrete (7.54 cm2) was uniformly
estimating PRA and PWA
corroded, which visual inspection proved was obviously
Rp Ecorr Rp , Ecorr not the case for many specimens once the embedded
Rp, th (Ω) 5,500 8,250
rebar was extracted from the concrete. Nevertheless,
because no measurement of the area of rusted zones was
Ecorr, th (mV/SCE) –365 –365
envisaged, the aforementioned assumption was adopted.
∆m iron, th (mg/cm2) 10.61 9.95 9.95 In the present study, several criteria were taken into
PRA 0.755 0.815 0.796 account to decide whether an experimental set of Tafel
95% confidence interval [0.631; [0.615; [0.686; coefficients of a given specimen was to be attributed to a
0.878] 0.921] 0.906] passive or active state of corrosion. As none of these
PWA 0.254 0.256 0.218 criteria was expected to be fully consistent, due to the
95% confidence interval [0.196; [0.196; [0.162; lack of a direct link with the cumulative corrosion state,
0.313] 0.308] 0.274] they were used in a combined way. Various combina-
Abbreviations: PRA: probability of a right alarm; PWA: probability of a tions, as well as single criteria, reported in Table 4, were
wrong alarm; SCE: saturated calomel reference electrode. considered. The yearly corrosion rate Vcorr was estimated
from the iron mass loss and the corrosion current jcorr
PRA is the probability that Rp ≤ Rp, th knowing that was determined by Equation (8), accounting for a global
∆m iron ≥ ∆m iron, th . The latter condition can be seen as the polarization of the steel surface. The theoretical iron
effective onset event provided that iron mass loss is mass loss ∆mth,iron was estimated under the assumption
cumulatively determined and not expressed as a corrosion of a uniform loss over the rebar and a corrosion depth of
rate. The results obtained in Duprat and Vu[19] were 11.6 μm/year for a corrosion current of 1 μA/cm2. It was
completed in the present study by adding the second set of assumed that the theoretical corrosion onset time was the
data identified in the cited study by a principal component time needed for the carbon dioxide to reach the
analysis. They are reported in Table 3. It can be seen in this reinforcement according to the Hyvert carbonation
table that Ecorr, th (–365 mV/SCE) appears somewhat low model.[30] Carbonation times are reported in Table 5.
compared to the ASTM reference value in Table 2 for a high The efficiency and feasibility of using the proposed
probability of corrosion (–278 mV/SCE). Nonetheless, in the combinations to classify an experimental pair of Tafel
case of isolated rebar with an incipient corrosion spot, coefficients (as representing a passive or active state of
Sassine et al.[29] reported a similar value (–330 mV/SCE) steel) were judged with respect to the coherence that
obtained with the same experimental procedure, i.e. half‐ combined criteria showed with each other. So, the
potential measurement from the concrete surface. number Nc of cases where the same classification was

T A B L E 4 Criteria and combinations of corrosion state classification

Designation Comment Combination giving active state


C‐m‐E Iron mass loss and corrosion potential ∆m iron > 9.95 mg/cm2, Ecorr < –365 mV/SCE
C‐m‐R Iron mass loss and polarization resistance ∆m iron > 10.61 mg/cm2, Rp < 5,500 Ω
C‐E‐R Corrosion potential and polarization resistance Ecorr < –365 mV/SCE, Rp < 8,250 Ω
C‐m‐E‐R Iron mass loss, corrosion potential, and polarization ∆m iron > 9.95 mg/cm2, Ecorr < –365 mV/SCE, Rp < 8,250
resistance Ω
C‐m‐j Iron mass loss and corrosion current ∆m iron > 9.95 mg/cm2, jcorr > 0.1 µA/cm2
C‐m,th‐j Theoretical iron mass loss and corrosion current ∆m iron,th > 9.95 mg/cm2, jcorr > 0.1 µA/cm2
C‐m‐V Iron mass loss and yearly corrosion rate ∆m iron > 9.95 mg/cm2, Vcorr > 0.001 mm/y
C‐E‐R‐j Corrosion potential, polarization resistance, and corrosion Ecorr < –365 mV/SCE, Rp < 8,250 Ω, jcorr > 0.1 µA/cm2
current
C‐j‐V Corrosion current and yearly corrosion rate jcorr > 0.1 µA/cm2, Vcorr > 0.001 mm/y
C‐m‐j‐V Iron mass loss, corrosion current, and yearly corrosion rate ∆m iron > 9.95 mg/cm2, jcorr > 0.1 µA/cm2,
jcorr > 0.001 mm/y
Abbreviation: SCE: saturated calomel reference electrode.
10 | DUPRAT ET AL.

T A B L E 5 Accelerated carbonation time (d) satisfactory. The threshold value of 0.1 μA/cm2 is,
however, recognized as being the transition point from
Cover depth, Cover depth,
passive to active corrosion and is used for calibrating
Concrete mix 15 mm 30 mm
impressed direct current in experimental investiga-
CEM I 15 62 tions.[10] The assumptions of a charge transfer process
CEM II 4 16 and uniform corrosion allowing Equation (8) and (9) to
CEM III 10 43 be established are probably not fully realistic. Even at the
small scale of the specimen (rebar length is 2 cm) and
although the corrosion stage in question is at its very
established by each of the combined criteria, for a given onset, a macrocell corrosion system seems to take place,
set of experimental data, over the total number of implying that Equation (8) no longer applies. This is
experimental sets Nt was expressed as the frequency of confirmed by the fact that combining the theoretical iron
correct classification frc = Nc / Nt . mass loss with the corrosion current (C‐m,th‐j) leads to a
higher frequency of correct classification than combining
the experimental iron mass loss with the corrosion
5 | R E S U L T S AN D A N A L Y S I S
current (C‐m‐j). In the first combination, both corrosion
current and iron mass loss are determined to assume a
5.1 | Classification of experimental sets
uniform corrosion system as the experimental mass loss
The efficiency of combined criteria expressed as the results from nonuniform corrosion and is, on average,
frequency of correct classification is reported in Table 6 for about 1.3 times greater than the theoretical one.
the various exposure conditions and concrete mixes. It can be
seen that the range of frequency of correct classification is
rather wide, from a minimum value of 0% to a maximum
value of 93%. For the same exposure condition, the difference
5.2 | Analysis and discussion
with respect to the concrete mix can reach almost 25% (XC1, An experimental set of values has been classified with
C‐E‐R‐j). There can also be a marked variation, of more than respect to the combined criteria C‐E‐R. The statistical
29% according to the exposure condition, in the average moments obtained from these sets are reported in Tables
value of frc for the three concrete mixes (C‐m‐V). A relevant 7 and 8, where no distinction is made according to the
combination of criteria for the classification of Tafel concrete mix as it was observed previously that the
coefficients would be likely to result in a reasonably high average variation of frc was less than 15% between
value of frc together with a low variation according to the concrete mixes for the same exposure condition.
concrete mix and exposure conditions. In this regard, C‐E‐R As far as the Tafel slopes are concerned (Table 7) it
would appear to be the best choice. can be seen that the coefficient of variation is very large,
Further, it can be noted that combined criteria where varying from about 45−105% depending on the exposure
the corrosion current is involved are globally not conditions. There is no significant difference regarding

T A B L E 6 Frequency of correct classification (%)

Exposure condition Concrete mix C‐m‐E C‐m‐R C‐E‐R C‐m‐E‐R C‐m‐j C‐m,th‐j C‐m‐V C‐E‐R‐j C‐j‐V C‐m‐j‐V
XC1 CEM I 83.4 90.0 77.8 73.3 40.0 58.6 74.1 48.6 16.7 16.7
CEM II 70.5 76.4 86.1 64.7 45.2 68.0 56.7 54.8 8.0 8.0
CEM III 81.8 84.8 88.9 75.8 24.2 50.0 79.3 30.6 6.9 6.9
All concrete mixes 78.4 83.5 84.3 71.1 36.2 58.3 69.8 44.1 9.9 9.9
XC4 CEM I 80.5 83.3 86.1 69.4 62.9 65.5 30.0 65.7 0.0 0.0
CEM II 77.8 72.2 86.1 72.2 60.0 72.4 36.7 57.1 6.9 6.9
CEM III 80.6 55.6 72.2 55.6 41.2 78.6 53.3 55.9 0.0 0.0
All concrete mixes 79.6 70.4 81.5 65.7 54.8 72.1 40.0 59.6 2.3 2.3
Outdoor CEM I 69.7 69.7 88.9 60.6 25.9 54.2 93.1 44.4 20.8 20.8
CEM II 80.5 80.5 91.7 75.0 35.7 47.8 90.0 39.3 30.4 30.4
CEM III 80.5 80.6 83.3 66.7 37.9 58.3 76.7 37.9 16.7 16.7
All concrete mixes 77.1 77.1 88.0 67.6 33.3 53.5 86.5 40.5 22.5 22.5
All exposures and concrete mixes 78.4 76.8 84.6 68.1 42.2 61.8 65.3 48.6 10.9 10.9
DUPRAT ET AL. | 11

T A B L E 7 Statistical results for Tafel slopes (V/dec)

Exposure Passive anodic Passive cathodic Active anodic Active cathodic


condition Statistics Tafel slope βa,P Tafel slope βc,P Tafel slope βa,A Tafel slope βc,A
XC1 Mean 0.577 0.222 0.988 0.473
Coeff. of variation 70.6% 74.3% 44.3% 53.3%
10% percentile 0.175 0.099 0.216 0.118
90% percentile 1.028 0.478 1.554 0.781
XC4 Mean 0.516 0.174 0.875 0.515
Coeff. of variation 64.6% 67.0% 69.4% 83.1%
10% percentile 0.120 0.092 0.181 0.126
90% percentile 0.953 0.291 1.504 0.957
Outdoor Mean 0.950 0.235 0.770 0.452
Coeff. of variation 93.3% 71.2% 71.6% 51.7%
10% percentile 0.225 0.092 0.130 0.108
90% percentile 2.214 0.494 1.726 0.763
All exposures Mean 0.637 0.218 0.868 0.473
Coeff. of variation 89.6% 72.5% 61.8% 71.7%
10% percentile 0.174 0.096 0.191 0.121
90% percentile 1.245 0.440 1.526 0.852
Number 140 152 102 110

T A B L E 8 Statistical results of half‐cell potentials (mV/SCE) and polarization resistances (kΩ)

Exposure Passive corrosion Passive polarization Active corrosion Active polarization


condition Statistics potential Ecorr, P resistance Rp,P potential Ecorr, A resistance Rp,A
XC1 Mean –147.6 40.262 –514.4 3.772
Coeff. of 86.6% 73.2% 18.1% 60.5%
variation
10% percentile –325.9 12.794 –675.9 1.168
90% percentile 22.1 78.426 –424.5 7.505
XC4 Mean –139.3 27.379 –480.2 2.965
Std deviation 72.7% 62.4% 12.8% 56.9%
10% percentile –282.9 11.146 –562.1 1.284
90% percentile –15.7 54.302 ‐405.4 6.137
Outdoor Mean –105.9 50.686 –480.8 3.432
Coeff. of 116.7% 63.0% 12.7% 53.5%
variation
10% percentile –321.6 14.159 –579.8 1.193
90% percentile 37.8 98.147 –404.5 6.114
All exposures Mean –126.5 42.466 –491.1 3.296
Coeff. of 96.6% 70.9% 15.0% 58.6%
variation
10% percentile –314.5 12.138 –585.5 1.214
90% percentile 26.2 85.989 –407.5 6.328
Number 162 151 112 112
12 | DUPRAT ET AL.

the coefficient of variation depending on either the the presence of an oxide layer at the surface of the steel,
branch concerned (anodic or cathodic) or the exposure formed under active corrosion.[17] On the other hand, as
condition. In contrast, there is a tendency for the the experiment for measuring the Tafel slopes implies the
coefficient of variation to be more pronounced in the irremediable ruin of the corrosion system due to the high
case of the passive state of corrosion, except for the XC4 imposed potential sweep, it was not possible to confirm
exposure, where the opposite occurs. A consequence of this shift. Steel polarization curves based on average
the widespread of the Tafel coefficients is that domains of values of Tafel slopes, corrosion potential, and polariza-
strong likelihood (say 80% interval between 10% and 90% tion resistance for all the exposure conditions are plotted
percentiles) globally overlap, more or less, whatever the in Figure 5, where it can be clearly seen that corrosion
exposure condition, the branch or the corrosion state. states are differentiated.
Considering all the exposure conditions, the overlap Tafel coefficients obtained in the present study also
between passive and active corrosion states appears exhibit a certain discrepancy in terms of mean value and
larger for the anodic Tafel coefficient (0.8–2.1 V/dec) coefficient of variation with regard to those obtained by
than for the cathodic one (0.2–0.9 V/dec). The same previous studies conducted with a similar determination
values of the Tafel coefficients can hence be attributed to procedure, as shown in Table 9, where all the results of
both passive and active states of corrosion and more each study are merged (RC and HybRC, loaded and
probably for the anodic one. It is nevertheless worth unloaded in Nguyen et al.,[16] PCC and GPC in Babaee
noting that the mean value of the Tafel coefficients and Castel[17]). Beyond the fact that the applied potential
generally increases from the passive to the active state of sweep for determining the Tafel coefficients was much
corrosion except for the outdoor exposure condition. This smaller in Nguyen et al.[16] or Babaee and Castel,[17]
observation goes against the assumption made earlier, which could have contributed to this finding, the
and confirmed experimentally in Figure 4 for two difference in mean values and coefficients of variation
separate specimens, that a decrease of the anodic Tafel can result from the origin of corrosion on the one hand
slope together with a fairly constant cathodic slope would and from the criterion for distinguishing passive and
characterize the shift from the passive to active corrosion active states of steel on the other. Published results of
state. Generally speaking, a rise in Tafel slopes may be Tafel coefficients under carbonation‐induced corrosion
caused by a limitation of charge transfer rate, which can are very scarce, rendering uneasy the comparison of
occur, for instance, in the case of the anodic coefficient by mean values with those obtained under chloride‐induced

FIGURE 5 Fictitious polarization curves deduced from average values of electrochemical properties. A: Active steel, P: passive steel
DUPRAT ET AL. | 13

T A B L E 9 Tafel coefficients obtained in present and previous studies

Passive anodic Passive cathodic Active anodic Active cathodic


Reference Statistics Tafel slope βa,P Tafel slope βc,P Tafel slope βa,A Tafel slope βc,A
Present study Mean 0.637 0.218 0.868 0.473
Coeff. of variation 89.6% 72.5% 61.8% 71.7%
Nguyen et al.[16] Mean 0.562 0.198 0.180 0.102
Coeff. of variation 37.5% 21.5% 22.4% 9.5%
Babaee and Castel [17]
Mean ∞ 0.037 0.609 0.150
Coeff. of variation ∞ 14.7% 36.9% 38.8%

corrosion. When corrosion is induced by chloride ingress, resistance, as shown by the 10% and 90% percentiles. In
and in the case of single rebar, the presence of chlorides the same way, there was no overlap between passive and
is a major environmental constraint acting at every stage active domains for any electrochemical property. As
of corrosion. This dominant parameter tends to mask the already mentioned regarding the Tafel coefficients, the
effect of other parameters and, in consequence, the mean values of polarization resistance and corrosion
scatter on Tafel slopes remains moderate. In contrast, potential clearly differentiate the passive from the active
when corrosion is induced by carbonation, no single state of corrosion, as depicted in Figure 5.
parameter is dominant and all implied parameters can
influence the variation of Tafel slopes in the same
proportion. In Babaee and Castel[17] corrosion was not 5.3 | Statistical analysis
initiated for specimens corresponding to the passive state All computations of this part were performed with
of steel as it was initiated by immersion in a chloride Matlab tools. Statistical inference from collected data is
solution for specimens corresponding to the active state, all the more precise when the number of data items is
and then activated by wet/dry cycles. In Nguyen et al.[16] large. Merging data from various exposure conditions is
corrosion was initiated for all specimens by an external thus a good option concerning that goal. When deciding
ponding dam containing chloride solution at the concrete whether data merging is relevant, classical statistical tests
surface and then activated by emptying/refilling the are useful. The Student test and Fisher–Snedecor test aim
ponding dam with solution. Measurements were made to test whether two samples relate to a single population
periodically and the distinction between passive and of the same mean or same variance, respectively. As these
active states of steel was decided according to the tests optimally apply only to a normal distribution, their
corrosion potential, all specimens experiencing a corro- use could be unsatisfactory in our case, where very few
sion potential in the range ]–300; –100[ mV/SCE being values of 10% and 90% percentiles match those of normal
discarded. In both of these previous studies, the samples tails. The two‐sample Kolmogorov‐Smirnov test and
were clearly differentiated according to the state of the the Wilcoxon–Mann–Whitney test are more generally
steel and specimens in an uncertain state were not applicable. The first is based on the maximum absolute
considered whereas, in our case, all specimens were difference between the observed cumulative density
classified as corresponding to a particular state of the functions and it tests whether samples come from the
steel. Although the distinction of the steel state is clearly same population (null hypothesis). The second is based
shown in Figure 5 it is not so significantly reported in the on the number of times one sample precedes the other in
mean values of Tafel coefficients. an ordered arrangement of the elements and indicates
As far as the electrochemical properties are concerned whether samples relate to a single population having the
(Table 8) it is noteworthy that the coefficients of variation same median (null hypothesis). Probability values of the
of the polarization resistance are on average similar tests are reported in Tables 10 and 11. It can be seen that,
(53–73%) to those of the corrosion potential (13–117%), for the Tafel coefficients, the null hypothesis is some-
and are more reduced for both in the active than in the times rejected at the 5% significance level by one or the
passive corrosion state. This was not observed for the other test, or by both, implying that data of the
Tafel coefficients. Threshold values used for the classifi- corresponding experimental sets should be considered
cation in the passive and active states led to a wider range separately and not merged in a larger sample. However,
of possible values in the passive state for the corrosion as there are no reference distributions of Tafel slopes in
potential and an even wider one for the polarization the literature, it does not seem convenient for practical
14 | DUPRAT ET AL.

T A B L E 10 Kolmogorov dependence‐test results (probability %) between exposure conditions

Tafel coefficients Corrosion potential Polarization resistance

Exposure conditions βa,P βc,P βa,A βc,A Ecorr, P Ecorr, A Rp,P Rp,A
XC1 and XC4 80.7 17.5 1.0 47.0 65.5 13.8 4.6 9.5
XC1 and outdoor 13.9 54.9 0.6 88.7 12.2 33.8 7.6 9.3
XC4 and outdoor 3.2 9.3 30.8 81.9 3.8 89.2 0.0 47.2

use to attempt to distinguish distributions with respect to ⎛ N ⎞


the exposure condition of the concrete surface. So, θˆ = arg min ⎜⎜ −∑ ln fX (x i|θ ) ⎟⎟. (19)
θ ⎝ ⎠
irrespectively of the output of hypotheses, test data was i =1

globally gathered together for all exposure conditions for


the goodness‐of‐fit tests. The second stage consists in assessing how various
The results of statistical dependence tests carried out distributions can faithfully represent the observations
for the electrochemical properties were similar to those according to specific goodness‐of‐fit tests. The Kolmogor-
obtained for the Tafel coefficients: The null hypothesis is ov–Smirnov test and the χ2 test were applied in the
rejected in some cases but, for the sake of simplicity, the present study. These are efficient tests, widely used in the
data sets corresponding to the various exposure condi- literature.[36] The nonparametric Kolmogorov–Smirnov
tions have been brought together. Moreover, pooling data test is based on the maximum distance DN between the
also gives the opportunity to work with samples of more expected cumulative distribution function FX and the
than 100 values, which is a convenient size for statistical observed one FO .
inference.
The statistical inference was conducted in two stages DN = max|FX (x i ) − FO (x i )|. (20)
xi
for each marginal distribution of the eight quantities of
interest. The first stage was to fit observations to a
The Kolmogorov–Smirnov metric is λ = NDN . If
theoretical distribution, determining the distribution
x1,… , xN is a random sample of X whose cumulative
parameters by the maximum likelihood estimation
distribution function is FX , an upper limit exits for λ
(MLE). The principle of MLE is to determine the
depending on N and a given confidence level α . For
parameters (expectation, variance, shape, and scale
instance, if N > 40 then λ = 1.22 for α = 0.95
factors, etc., denoted θ ) so that they maximize the
(P (λ ≤ 1.22) = 0.95). In the χ2 test the data is grouped
likelihood function LN (x|θ) which, if the observations x
into Nb bins and the chi‐square metric χ 2 is computed.
are independent, can be expressed as
Nb
N
LN (x|θ ) = ∏ fX (x i|θ) (18)
χ2 = ∑ (Oi − Ei )2 / Ei (21)
i =1
i =1

where Oi is the observed count and Ei is the expected


where fX (x i|θ ) is the theoretical probability density
count, derived from the theoretical distribution, for bin i .
function evaluated at the observation x i . Usually and
In both tests, the null hypothesis is that the data comes
equivalently, the log‐likelihood is minimized and the best
estimate θ̂ holds from the predetermined distribution. If this is the case,

T A B L E 11 Wilcoxon–Mann–Whitney dependence‐test results (probability %) between exposure conditions

Tafel coefficients Corrosion potential Polarization resistance

Exposure conditions βa,P βc,P βa,A βc,A Ecorr, P Ecorr, A Rp,P Rp,A
XC1 and XC4 62.4 14.6 2.6 62.3 76.8 1.5 20.7 12.4
XC1 and outdoor 2.9 78.3 5.4 65.3 6.3 6.6 23.9 71.7
XC4 and outdoor 1.9 9.2 3.9 8.4 9.4 0.0 97.6 32.4
DUPRAT ET AL. | 15

T A B L E 12 Expected distributions

Distribution Density probability function Parameters


a : log mean, − ∞ < a < +∞
Lognormal (LN)
fX (x | a, b) =
1
xa 2π
exp ( −(ln x − a)2
2b2 ), x > 0 b: log std. dev., b > 0
a : scale param. a > 0
Weibull (WB)
fX (x | a, b) =
b
a
x b −1
()
a
exp (− ( ) ), x > 0
x b
a
b: shape param. b > 0
Generalized extreme value (GEV) 1 ⎛⎜ x − a ⎞⎟−1−1/ c ⎛ ⎛ x − a ⎞⎟−1/ c ⎞ a : loc. param., − ∞ < a < +∞
fX (x | a, b, c ) = 1+c exp ⎜−⎜1 + c ⎟,
b ⎝ b ⎠ ⎝ ⎝ b ⎠ ⎠ b: scale param., b ≥ 0
−∞ < x < +∞ c : shape param.,
− ∞ < c < +∞
a : loc. param., − ∞ < a < +∞
Extreme (EX) 1
fX (x | a, b) = b exp ( ) exp (−exp ( ) ),−∞ < x < +∞
x−a
b
x−a
b
b: scale param., b ≥ 0
a : scale param., a > 0
Rayleigh (RA)
fX (x | a) =
x
a2
exp (− ( ) ), x ≥ 0
x
2a
2

a : scale param. a > 0


Gamma (GA) fX (x | a, b) =
1
abΓ(b)
x b −1exp ( − ), x > 0
x
a
b: shape param., b > 0
Burr (BU) −c −1 a : scale param., a > 0
fX (x | a, b, c ) =
bc
a
x b −1
()
a (1+( ) )
x b
a
, x>0
b: first shape param., b > 0
c : second shape param., c > 0
Beta (BE) fZ (z | a, b) =
1
z a −1 (1 − z )b −1, 0 < z < 1 a : first shape param., a > 0
Β(a, b)
b: second shape param., b > 0

the metrics should be distributed according to the the latter value of x1, x2 was stated at x2 = Omax + ksO ,
Kolmogorov or χ2 distribution, respectively. The prob- where sO is the empirical standard deviation. Such an
ability of exceeding the metrics (p value) is then interval is intended to be reasonably broad enough to
compared to the accepted confidence risk (1 − α ) (or simulate unlimited support. Expected counts Ei in
significance level) with a minimum value of 5%. Equation (21) and thereafter the goodness‐of‐fit, depend
The various theoretical candidate distributions are upon the value of k . This was consequently determined
reported in Table 12. Some of them are usual for material so that Ei did not vary by more than 5% for each bin when
properties, like the Weibull, Lognormal or Beta distribu- k was changed to k + 1. For cases of data where the Beta
tions. The Beta distribution is particularly useful due to distribution was one of the best fitted, k was equal to 2.
its simultaneously bounded and flexible nature.[32] The two stages of the statistical inference were applied
Generalized extreme value and extreme distributions to the Tafel coefficients, corrosion potential, and polar-
are usually dedicated to modeling loads or natural actions ization resistance. Candidate distributions selected if the
on structures, like traffic loads or snow action.[33] The goodness‐of‐fit tests did not reject the null hypothesis
Rayleigh distribution is rather used for space‐variant (p > 5%) were then ranked according to the score Sgof .
phenomena like wind action.[34] The Gamma and Burr
distributions are convenient in economics[35] or engi- pChi2 − 0. 05 pKolm − 0.05 |EX − mO|
Sgof = + −
neering for time‐variant phenomena.[31,38] In Table 12, 0.05 0.05 mO
the Beta distribution is presented in its standardized | VX − sO| |X − O10%|
form. It can be easily adapted to any range ] x1, x2 [ by − − 10%
sO O10%
writing z=(x − x1)/(x2 −x1). The bounds can be deter-
|X90% − O90%|
mined according to the data and, for instance, simply − . (22)
O90%
equal the empirical min Omin and max Omax . Never-
theless, there is no physical reason to limit the support of
distribution functions, except for data that cannot be This score compares selected distributions according
negative. x1 was therefore stated at zero for the Tafel to their closeness to the empirical values of mean (mO ),
coefficients and polarization resistance, and at standard deviation (sO ), 10% percentile (O10%) and 90%
x1 = Omin − ksO for the corrosion potential. Similarly to percentile (O90%). In Equation (23), the first two terms
16 | DUPRAT ET AL.

T A B L E 13 Goodness‐of‐fit results (two best candidate distributions)

Parameters

Data Dist. a b c pChi2 pKolm EX VX X10% X90% Sgof


βa, P (V/dec) BU 0.702 2.017 1.538 0.600 0.749 0.685 0.657 0.189 1.301 24.8
GA 0.374 1.832 0.484 0.335 0.684 0.506 0.167 1.359 14.1
βc, P (V/dec) BU 0.142 3.630 0.635 0.139 0.921 0.224 0.308 0.089 0.383 18.0
LN −1.724 0.639 0.197 0.152 0.212 0.156 0.078 0.405 4.7
βa, A (V/dec) RA 0.696 0.347 0.252 0.873 0.456 0.320 1.495 9.1
WB 0.975 1.905 0.183 0.280 0.865 0.473 0.299 1.511 6.5
βc, A (V/dec) GA 0.204 2.412 0.270 0.764 0.492 0.317 0.154 0.917 18.2
BE 1.967 12.323 0.269 0.762 0.492 0.317 0.154 0.919 18.1
Ecorr, P (mV/SCE) GEV −164.5 125.8 −0.369 0.082 0.369 −126.8 120.7 −287.4 27.7 7.1
BE 7.619 7.932 0.056 0.227 −126.9 120.7 −284.5 31.2 3.6
Ecorr, A (mV/SCE) GEV −504.4 80.1 −0.543 0.287 0.951 −490.9 73.8 −591.9 −403.3 22.7
EX −456.7 59.645 0.209 0.659 –491.1 76.5 −590.9 −407.0 15.3
Rp, P (kΩ) WB 46.616 1.578 0.224 0.525 41.849 27.117 11.202 79.075 12.7
BE 1.774 6.107 0.139 0.524 42.200 26.273 11.728 79.232 11.0
Rp, A (kΩ) WB 3.726 1.817 0.545 0.524 3.312 1.888 1.080 5.897 19.2
BE 2.088 5.470 0.524 0.432 3.319 1.836 1.106 5.893 16.9
Abbreviation: SCE: saturated calomel reference electrode.

(where pChi2 and pKolm are respectively the p values of χ2 clearly drawn, whereas only one peak and shape of the
and Kolmogorov–Smirnov tests) contribute to the dis- plateau are formed in Figure 8 for the polarization
tribution being favored and the last four contribute to it resistance. To identify the modes, a mixture probability
being discarded. The two best‐fitted distributions to- distribution expressed as
gether with their parameters are reported in Table 13.
Histograms corresponding to the distribution of highest fX , M (x ) = αfX , P (x | θP ) + (1 − α ) fX , A (x | θA) (23)
score can be seen in Figure 6.
It can be seen in Table 14 that distributions from the can be suitably used in a statistical analysis of the data,
exponential family are most suitable according to the Sgof where the distributions parameters θP and θA, referring
score for all quantities of interest except for the Tafel respectively to passive or active state of corrosion, and the
coefficients in the case of the passive state of corrosion, proportion factor α are determined by the MLE
where the Burr distribution is preferable. For every procedure. The distributions fX , P and fX , A are often
quantity studied, two or more distributions are almost chosen of the same type for the sake of simplicity.[37] In
equally convenient. When the mean and variance of the our study the Normal and General Extreme Value
output of a corrosion model are sought, where these distributions for the corrosion potential, and the Weibull
quantities are needed, the use of either of the feasible and LogNormal distributions for the polarization resis-
distributions is likely to lead to the same result. However, tance were tested against data provided by the experi-
in a reliability approach, where distribution tails may mental campaign. The fitted distribution parameters are
significantly contribute to the result, it is obvious that the given in Table 14, where it can be noticed that EX and
choice of the distributions is a real concern. VX defining the modes of each electrochemical quantity
The statistical analysis previously carried out was are close to each other whatever the assumed distribu-
preceded by a classification procedure to distinguish the tions. Moreover these values are also very similar to those
data with respect to the steel corrosion state they are obtained after having separated the data as a function of
supposed to pertain to. For the data significantly the steel corrosion state and presented in Table 13, when
modified by the steel state the effect of this latter should considering the same type of distribution. Consequently
appear when looking at a histogram of mixed data it can be noted on the one hand that the MLE procedure
covering the two possible states, passive or active, where is satisfactory to extract the distributions corresponding
two distinct modes should be visible. This can be seen in to passive or active corrosion state of steel from the global
Figure 7 for the corrosion potential where two peaks are experimental set of corrosion potentials or polarization
DUPRAT ET AL. | 17

FIGURE 6 Histograms of data and best‐fitted distributions


18 | DUPRAT ET AL.

T A B L E 14 Distributions parameters obtained by MLE from the global samples and comparison with Table 13

Parameters θP or θA

Data Dist. fX a b c EX EX (Table 13) VX VX (Table 13) α


Ecorr, P (mV/SCE) GEV −170.1 142.8 −0.443 −133.3 −126.8 133.3 120.7 0.403
N −102.0 107.3 −102.0 107.3 0.478
Ecorr, A (mV/SCE) GEV −509.6 84.1 −0.398 −465.8 −490.9 79.8 73.8 0.403
N −465.0 93.8 −465.0 93.8 0.478
Rp, P (kΩ) WB 39.135 1.251 36.443 41.849 29.314 27.117 0.334
LN 3.638 0.627 46.285 32.137 0.502
Rp, A (kΩ) WB 3.274 1.913 2.905 3.312 1.580 1.888 0.334
LN 1.241 0.812 4.811 4.652 0.502
Abbreviations: MLE: maximum likelihood estimation; SCE: saturated calomel reference electrode.

(a) (b)

FIGURE 7 Experimental histogram and mixed (a) or separated (b) distributions of Ecorr

(a) (b)

FIGURE 8 Experimental histogram and mixed (a) or separated (b) distributions of Rp


DUPRAT ET AL. | 19

(a) (b)

FIGURE 9 Experimental histograms of βa (a) and βc (b)

As far as the Tafel coefficients are concerned, no


T A B L E 15 Pearson correlation coefficient visible tendency allowing to discriminate distinct modes
Passive state Active state is emphasized in Figure 9. It would have been then of
limited interest to perform a statistical analysis of the
βc Ecorr Rp βc Ecorr Rp sample as a whole. This confirms that a specific
βa 0.368 –0.111 –0.013 0.784 0.114 0.014 procedure merging Tafel coefficients, corrosion potential,
–0.438 –0.179
and polarization resistance was needed to properly
βc 0.144 0.074
discern the data as a function of the steel corrosion state.
ECorr 0.406 0.523
In the perspective of using the studied distributions in
modeling and random simulation, a major concern is also
to assess whether a correlation exists between data so as
to combine this information with the knowledge of
marginal distributions. The most common approach
regarding correlation is based on the Pearson product–
moment correlation coefficient. Considering two samples
of data, X and Y , the Pearson correlation coefficient is
written as

so, XY
ρXY = (24)
so, X so, Y

where so, XY is the empirical covariance, and so, X and so, Y are
the empirical variances of X and Y . When the corelation
coefficient is less than 0.3, X and Y can be supposed
stochastically independent and the correlation may be
ignored in a simulation procedure. In Table 15, it can be
seen that a positive nonnegligible correlation exists between
the Tafel coefficients on one side (Figure 10) and the
corrosion potential and polarization resistance on the other
FIGURE 10 Samples of βc and βa for passive or active side (Figure 11). There is hence a trend for both quantities of
corrosion state each pair to deviate simultaneously in the same way with
respect to their mean value. The corrosion potential and
resistance, and on the other hand that the procedure polarization resistance are fair indicators of instantaneous
proposed for discriminating the corrosion states accord- corrosion activity in the case of a single rebar, if moisture and
ing to the thresholds based on the optimum probability of oxygen access conditions remain constant in the bulk
right alarm and reported in Table 3 is relevant. concrete: their values are rather high under the passive
20 | DUPRAT ET AL.

FIGURE 11 Samples of Ecorr and Rp


for passive or active corrosion state

state, and low under the active state of steel as the corrosion Tafel coefficients are introduced in the boundary conditions.
proceeds, as shown in Figure 3 in Angst et al.[39] and in Tafel coefficients are thus crucial parameters for the
Figure 11 of the present paper. To a certain extent, their prediction of corrosion. In the present study, a broad
random variation is then influenced and stochastically linked experimental campaign established probability distributions
by these conditions. The Tafel coefficients govern the for the major electrochemical properties of steel: anodic and
sensitivity of the corrosion system to polarization of the cathodic Tafel coefficients, corrosion potential, and polariza-
steel. Under the same absolute polarization, anodic and tion resistance. Active and passive states of carbonation‐
cathodic currents are all the more pronounced when the induced corrosion were considered. As none of the
Tafel coefficients are low. A nonnegligible correlation experimental specimens could, a priori, be known to be in
coefficient between Tafel coefficients indicates that the a passive or active state of corrosion, several combinations of
condition state of the corrosion system during the polariza- criteria based on the outcomes of electrochemical tests were
tion test may influence the response (in terms of averages) of used, in the aim of properly differentiating the corrosion
the system in the same way under either anodic or cathodic states, and ranked according to a so‐called frequency of
polarization. This is all the more visible when the system is correct classification. Data samples were then extracted from
in an active state of corrosion, without, however, being this classification and statistical inference was performed on
clearly under anodic or cathodic control. In the case of them. Compared to published results related to chlorination‐
chloride‐induced corrosion such correlation coefficients were induced corrosion,[16,17] mean values of the Tafel coefficients
not observed in the literature. obtained in the present study are significantly higher, as are
their coefficients of variation. Some possible reasons for this
finding have been discussed in detail. A broad variety of
6 | C ON C LU S I O N marginal probability distributions have been proposed, fitted
and tested against experimental data. Pearson correlation
Little attention is paid in the literature to the Tafel coefficients were additionally computed and revealed that
coefficients related to carbonation‐induced corrosion even anodic and cathodic Tafel coefficients were not stochastically
though such corrosion affects most concrete structures. A independent. The outcomes of the present study will permit
much more prominent effort has been dedicated to the future models of corrosion to be easily enriched by placing
corrosion rate estimation.[18] With a view to predicting the them in the framework of probabilistic approaches, which
corrosion propagation in reinforced concrete structures, are a rational way to account for the random nature of
advanced modeling has recently been developed, based on corrosion.
Butler–Volmer equations and considering boundary condi-
tions on both passive and active steel.[6] This type of
ORCID
modeling is promising as it is theoretically able to represent
macrosystems of corrosion, provided that realistic values of Frédéric Duprat http://orcid.org/0000-0001-6979-4652
DUPRAT ET AL. | 21

REFERENCES [25] ASTM C876‐09, ASTM International, West Conshohocken,


2009.
[1] M.‐T. Liang, W.‐L. Jin, R.‐J. Yang, N.‐M. Huang, Cem. Conc. [26] ASTM G3‐89(2010), ASTM International, West Conshohocken,
Res. 2005, 35, 1827. 2010.
[2] B. Huet, V. L’Hostis, G. Santarini, D. Feron, H. Idrissi, Corr. [27] Ø. Vennesland, M. Raupach, C. Andrade, Mater. Struct. 2007,
Sci. 2007, 49, 1918. 40(8), 745.
[3] W.‐J. Chitty, P. Dillmann, V. L’hostis, A. Millard, Corr. Sci. [28] B. Elsener, C. Andrade, J. Gulikers, R. Polder, M. Raupach,
2008, 50, 3047. Mater. Struc. 2003, 37, 461.
[4] R. R. Hussain, T. Ishida, Comp. Struc. 2010, 88, 639. [29] E. Sassine, S. Laurens, R. François, E. Ringot, Mater. Struc.
[5] A. Nasser, A. Clément, S. Laurens, A. Castel, Corr. Sci. 2010, 2018, 51, 92.
52, 2878. [30] N. Hyvert, A. Sellier, F. Duprat, P. Rougeau, P. Francisco,
[6] S. Laurens, P. Hénocq, N. Rouleau, F. Deby, E. Samson, J. Cem. Conc. Res. 2010, 40, 1582.
Marchand, B. Bissonnette, Cem. Conc. Res. 2016, 79, 272. [31] E. Chiodo, P. de Falco, Elec. Pow. Syst. Res. 2016, 141, 549.
[7] C. Helm, M. Raupach, Mater. Corr. 2016, 67, 621. [32] M. E. Stephens, B. W. Goodwin, T. H. Andres, Rel. Eng. Syst.
[8] C. Y. Kim, J. K. Kim, Const. Build. Mater. 2008, 22, 1129. Saf. 1993, 42, 271.
[9] B. Yu, L. Yang, M. Wu, B. Li, Const. Build. Mater. 2014, 54, [33] E. J. Obrien, F. Schmidt, D. Hajializadeh, X. Y. Zhou, B. Enright,
385. C. C. Caprani, S. Wilson, E. Sheils, Struc. Saf. 2015, 53, 44.
[10] A. Michel, M. Otieno, H. Stang, M. R. Geiker, Cem. Conc. [34] A. Feijóo, D. Villanueva, J. L. Pazos, R. Sobolewski, Renew.
Comp. 2016, 70, 171. Sust. En. Rev. 2011, 15, 2826.
[11] J. Ožbolt, G. Balabanić, M. Kušter, Corr. Sci. 2011, 53, 4166. [35] P. R. Tadikamalla, Int. Stat. Rev. 1980, 48, 337.
[12] J. Ozbolt, G. Balabanic, E. Sola, Mater. Corr. 2017, 68, 622. [36] A. A. Olosunde, A. M. Adegoke, Amer. J. of App. Math. Stat.
[13] M. Otieno, H. Beushausen, M. Alexander, Mater. Corr. 2012, 2016, 4, 1.
63, 777. [37] J. Gulikers, B. Elsener, Mater. Corr. 2009, 60, 87.
[14] J. Ge, O. B. Isgor, Mater. Corr. 2007, 58, 573. [38] S. Yue, T. B. M. J. Ouarda, B. Bobée, J. Hyd. 2001, 246, 1.
[15] G. Song, Cem. Conc. Comp. 2000, 22, 407. [39] U. Angst, B. Elsener, C. K. Larsen, Ø. Vennesland, Elect. Act.
[16] W. Nguyen, J. F. Duncan, T. M. Devine, C. P. Ostertag, Elect. 2011, 56, 5877.
Act 2018, 271, 319.
[17] M. Babaee, A. Castel, Cem. Conc. Res. 2016, 88, 96.
[18] M. Stefanoni, U. Angst, B. Elsener, Cem. Conc. Res. 2018, 103, 35.
[19] F. Duprat, N. T. Vu, Eur. J. Env. Civ. Eng. 2013, 17, 478. How to cite this article: Duprat F, de Larrard T,
[20] ASTM G1‐90(1999)e1, ASTM International, West Vu NT. Quantification of Tafel coefficients
Conshohocken. 1999.
according to passive/active state of steel
[21] J. Gulikers, Mater. Corr. 2005, 56, 394.
[22] J. Warkus, M. Raupach, J. Gulikers, Mater. Corr. 2006, 57, 614.
carbonation‐induced corrosion in concrete.
[23] Z.‐T. Chang, B. Cherry, M. Marosszeky, Corr. Sci. 2008, 50, Materials and Corrosion. 2019;1–30.
357. https://doi.org/10.1002/maco.201910962
[24] J. Jiang, Y. Yuan, Cons. Buil. Mater. 2012, 35, 625.
22 | DUPRAT ET AL.

APPENDIX

T A B L E A1 Outdoor exposure condition — Concrete CI

Anodic Cathodic Iron mass Corrosion


Cover Specimen Tafel slope Tafel slope βc loss Δm iron potential Ecorr Polarization
condition reference Time (d) βa (V/dec) (V/dec) (g) (mV/SCE) resistance Rp (Ω)
D15NC 23 417 ∞ (11.092) 0.206 0.004 –85.4 68,292
28 162 ∞ (1011) 0.395 0.047 –262.9 45,120
30 417 n. m. n. m. n. m. –90.3 40,705
42 216 0.263 0.124 0.023 –363.7 8,443
63 417 n. m. n. m. n. m. –143.9 25,181
74 162 n. m. n. m. 0.010 –290.1 43,389
D15C 1 216 0.212 0.141 0.032 –322.2 22,833
5 417 0.757 0.429 0.045 –524.6 2,222
7 197 0.190 0.147 0.027 –442.5 3,867
9 162 3.377 0.458 0.042 –381.4 28,181
15 389 0.390 0.393 0.027 –123.3 5,610
16 162 n. m. n. m. 0.020 –351.5 19,164
18 382 1.395 0.422 0.027 –69.1 53,604
19 197 2.006 0.719 0.064 –407.1 4,618
25 417 1.255 0.577 0.044 –513.3 2,452
27 375 0.698 0.308 0.031 –63.3 63,228
41 382 0.569 0.395 0.014 22.5 81,005
43 375 0.263 0.192 0.210 –133.6 11,472
58 216 1.029 0.232 0.088 –338.5 26,353
59 417 0.765 0.513 0.042 –579.8 1,689
83 382 0.338 0.310 0.049 –31.7 20,055
D30C 4 197 0.534 0.110 0.018 26.2 20,1267
6 382 2.037 0.488 0.023 –59.2 59,278
10 216 0.086 0.059 0.030 –65.2 206,233
12 417 ∞ (11.181) 0.304 0.015 –386.8 4,165
13 375 1.005 0.092 0.019 33.7 60,287
20 417 0.200 0.904 0.047 –448.4 5,322
26 389 0.836 0.139 0.031 –3.7 69,600
36 389 1.211 0.379 0.026 –10.3 131,075
44 197 0.090 0.094 n. m. –60.0 225,979
47 162 n. m. n. m. 0.011 –40.5 33,932
61 162 2.496 0.448 0.067 –339.1 16,500
65 417 1.673 0.153 0.010 –124.9 5,809
103 216 0.629 0.091 0.027 57.8 174,422
104 382 ∞ (10 ) 15
0.181 0.019 129.5 3,677
105 375 3.278 0.388 0.032 –79.5 56,865
DUPRAT ET AL. | 23

T A B L E A2 Outdoor exposure condition — Concrete CII

Anodic Cathodic Iron mass Corrosion


Cover Specimen Tafel slope Tafel slope βc loss Δm iron potential Ecorr Polarization
condition reference Time (d) βa (V/dec) (V/dec) (g) (mV/SCE) resistance Rp (Ω)
D15NC 126 216 0.134 0.185 0.017 –133.5 77,779
134 162 ∞ (1014) 0.381 0.029 15.3 56,711
136 417 0.666 0.165 0.028 –66.5 33,080
144 162 n. m. n. m. 0.006 27.7 2,628
145 417 0.247 0.158 0.004 –113.3 45,637
147 417 0.649 0.254 0.013 –142.5 29,933
D15C 112 216 4.631 0.306 0.047 –235.7 49,564
115 162 1.247 0.591 0.087 –361.1 14,349
116 417 0.694 0.366 0.044 –507.2 2,914
118 375 0.273 0.165 0.047 –147.1 41,713
119 382 0.402 0.399 0.086 –175.8 1,893
123 375 0.503 0.049 0.026 40.7 2,791
124 417 2.292 0.178 0.046 –347.4 9,415
127 216 0.649 0.172 0.029 –203.6 101,017
132 382 0.596 0.520 0.055 –425.4 3,083
135 417 0.712 0.388 0.061 –558.0 1,302
138 382 0.512 0.518 0.067 –145.0 27,409
140 162 n. m. n. m. 0.050 –432.2 7,261
141 389 0.417 0.154 0.057 –52.5 27,250
143 197 ∞ (1014) 0.098 0.027 –272.9 46,426
146 197 9.391 0.117 0.024 –39.7 169,450
D30C 110 216 5.807 0.169 0.020 –47.8 94,330
111 375 0.253 0.043 0.010 –12.7 31,187
113 417 ∞ (1015) 0.200 0.050 –151.1 44,842
114 162 n. m. n. m. 0.027 –91.4 107,857
120 197 1.083 0.109 0.014 –54.1 198,137
121 417 1.128 0.196 0.014 –231.4 21,383
125 216 0.796 0.133 0.028 –126.9 85,647
128 375 0.676 0.066 0.016 16.2 31,633
129 382 0.448 0.076 0.021 34.5 93,155
130 417 1.753 0.171 0.035 –227.7 18,898
131 197 0.294 0.173 0.043 –46.2 115,580
133 162 ∞ (109) 0.891 0.081 –130.5 43,058
137 389 0.649 0.172 0.037 –59.7 23,361
139 389 0.427 0.063 0.028 54.1 38,229
142 382 1.307 0.221 0.027 –73.1 61,256
24 | DUPRAT ET AL.

T A B L E A3 Outdoor exposure condition — Concrete CIII

Anodic Cathodic Iron mass Corrosion


Cover Specimen Tafel slope Tafel slope βc loss Δmiron potential Ecorr Polarization
condition reference Time (d) βa (V/dec) (V/dec) (g) (mV/SCE) resistance R p (Ω)
D15NC 150 162 ∞ (14.150) 0.537 0.061 –35.5 41,960
157 417 0.678 0.149 0.027 –203.7 41,532
158 417 3.617 0.421 0.016 –108.6 5,416
161 417 0.776 0.195 0.053 –203.5 9,747
168 216 0.654 0.122 0.005 38.2 114,357
178 162 n. m. n. m. 0.047 –167.1 8,139
D15C 148 375 0.451 0.540 0.058 –115.7 47,931
149 382 0.550 0.290 0.047 7.8 76,304
151 417 1.540 0.649 0.087 –457.2 1,471
153 162 n. m. n. m. 0.050 –404.5 5,433
163 417 ∞ (10 ) 15
0.165 0.056 –511.2 2,709
165 216 1.264 0.934 0.148 –297.5 3,045
166 162 0.506 0.522 0.118 –548.3 4,192
169 375 0.885 ∞ (1015) 0.097 –106.9 11,120
171 197 0.133 0.104 0.035 –446.4 4,512
172 417 0.125 0.109 0.095 –583.2 680
175 389 0.505 ∞ (10 ) 15
0.068 –89.5 54,424
176 197 0.748 0.547 0.013 –448.0 6,114
177 417 1.316 0.728 0.080 –511.2 1,193
179 216 0.123 0.113 0.020 –281.9 13,397
181 382 0.738 0.327 0.022 –93.8 7,298
D30C 117 216 ∞ (10 ) 15
0.347 0.028 –231.3 22,744
122 162 2.214 0.733 0.095 –321.4 12,012
152 216 0.796 0.240 0.046 –210.8 27,480
154 197 0.677 0.110 0.007 37.6 127,606
155 382 0.648 0.139 0.009 45.8 47,651
156 375 0.560 0.183 0.026 –45.5 77,691
159 389 ∞ (10 ) 15
0.142 0.019 112.0 123,300
160 382 1.020 0.108 0.024 118.3 34,104
162 197 0.756 0.132 0.009 –63.0 97,429
164 382 0.530 0.218 0.027 –49.8 94,794
167 389 0.670 0.168 0.018 91.3 160,015
170 417 1.285 0.420 0.025 –212.5 4,709
173 417 1.480 0.162 0.028 –166.8 7,062
174 162 n. m. n. m. 0.019 –84.1 49,985
180 375 0.225 0.134 0.015 –3.2 45,182
DUPRAT ET AL. | 25

T A B L E A4 Exposure condition similar to XC4 — Concrete CI

Anodic Cathodic Iron mass Corrosion


Cover Specimen Tafel slope Tafel slope βc loss Δm iron potential Ecorr Polarization
condition reference Time (d) βa (V/dec) (V/dec) (g) (mV/SCE) resistance Rp (Ω)
D15NC 307 83 0.474 0.205 0.002 56.5 6,817
308 290 0.450 0.153 0.011 6.9 32,114
312 290 ∞ (10 ) 8
0.144 0.008 –105.5 30,436
313 248 0.130 0.177 0.049 –136.3 2,174
319 290 0.416 0.124 0.002 –15.1 68,082
324 262 1.086 0.221 0.017 –93.6 13,849
D15C 301 83 1.232 0.518 0.109 –336.8 1,480
302 241 0.149 0.109 0.099 –555.5 1,415
303 83 1.323 0.530 0.084 –284.6 6,334
304 241 0.920 0.632 0.101 –548.4 988
310 83 0.158 0.077 0.056 –365.0 6,132
311 255 1.001 0.801 0.135 –545.6 1,457
315 255 0.156 0.128 0.121 –570.7 705
317 248 0.910 0.792 0.143 –418.0 2,894
318 262 0.134 0.112 0.124 –538.8 1,642
321 248 0.783 0.515 0.111 –528.5 1,696
323 290 0.110 0.108 0.182 –554.8 1,406
328 290 0.911 0.663 0.114 –571.4 1,887
329 248 0.849 0.513 0.116 –565.7 923
332 290 0.730 0.504 0.118 –597.0 1,477
333 248 0.535 0.337 0.122 –514.2 2,538
D30C 334 262 1.589 1.199 0.113 –554.8 1,192
305 262 0.695 0.142 0.030 –70.6 22,900
306 83 0.874 0.068 0.062 32.5 6,920
309 290 0.593 0.395 0.131 –484.2 2,886
314 241 0.093 0.101 0.094 –124.5 100,886
316 241 0.441 0.098 0.007 –382.2 6,441
320 255 0.144 0.148 0.090 –205.2 13,718
322 255 0.819 0.286 0.036 –334.7 8,002
325 248 0.459 0.281 0.128 –514.6 2,424
326 262 0.730 0.213 0.074 –468.5 6,364
327 290 0.559 0.322 0.134 –497.4 4,012
330 269 0.574 0.202 0.025 –148.2 46,679
331 83 0.656 0.238 0.035 –277.6 18,334
335 290 1.169 0.243 0.128 –426.7 3,346
336 269 0.929 0.507 0.016 –510.4 2,992
26 | DUPRAT ET AL.

T A B L E A5 Exposure condition similar to XC4 — Concrete CII

Anodic Cathodic Iron mass Corrosion


Cover Specimen Tafel slope Tafel slope βc loss Δm iron potential Ecorr Polarization
condition reference Time (d) βa (V/dec) (V/dec) (g) (mV/SCE) resistance Rp (Ω)
D15NC 413 290 0.120 0.120 0.010 –192.2 55,281
415 290 0.328 0.091 0.031 –68.1 36,744
420 290 0.786 0.177 0.020 –20.9 41,012
422 248 0.120 0.102 0.042 –93.5 10,962
424 83 0.298 0.048 0.006 –4.7 4,372
436 83 0.480 0.047 0.011 27.7 2,628
D15C 401 241 0.509 0.328 0.101 –449.8 2,264
405 290 0.723 0.367 0.116 –536.1 2,623
408 262 0.799 0.561 0.074 –411.0 4,902
409 269 0.680 0.247 0.064 –448.1 7,325
412 262 0.837 0.487 0.087 –482.9 3,741
414 248 0.667 0.508 0.147 –476.4 3,596
417 255 0.617 0.335 0.091 –502.3 6,006
419 290 0.673 0.381 0.109 –512.4 2,944
425 83 1.932 0.608 0.092 –297.3 3,829
427 241 2.343 1.942 0.224 –502.5 2,065
429 255 0.954 1.238 0.115 –523.9 3,085
430 83 ∞ (10 ) 9
0.698 0.079 –290.8 14,476
431 290 1.025 0.432 0.104 –561.7 3,957
434 269 0.830 0.503 0.068 –467.7 6,489
435 269 0.574 0.949 0.093 –390.6 6,184
D30C 402 255 0.959 0.279 0.081 –419.0 3,424
403 262 0.438 0.206 0.068 –283.5 13,788
404 241 0.752 0.253 0.102 –408.6 5,877
406 241 0.582 0.282 0.116 –387.1 2,825
407 262 3.301 5.817 0.091 –343.0 7,586
410 262 0.397 0.085 0.012 –63.8 52,277
411 248 0.153 0.125 0.009 –205.7 47,410
416 248 1.543 0.098 0.041 –66.4 17,881
418 262 0.540 0.120 0.017 –52.1 28,348
421 290 0.388 0.170 0.085 –268.5 9,510
423 255 0.318 0.138 0.021 –209.5 14,811
426 83 0.195 0.297 0.001 –116.1 56,530
428 83 0.846 0.324 0.031 –300.2 17,319
432 290 0.742 0.250 0.087 –221.2 6,529
433 290 0.693 0.258 0.118 –407.0 4,451
DUPRAT ET AL. | 27

T A B L E A6 Exposure condition similar to XC4 — Concrete CIII

Anodic Cathodic Iron mass Corrosion


Cover Specimen Tafel slope Tafel slope βc loss Δm iron potential Ecorr Polarization
condition reference Time (d) βa (V/dec) (V/dec) (g) (mV/SCE) resistance Rp (Ω)
D15NC 508 83 ∞ (1012) 0.015 0.002 –17.8 3,326
510 290 0.467 0.115 0.015 –166.1 12,935
511 290 0.516 0.100 0.017 –249.5 12,083
521 248 0.907 0.157 0.033 –334.2 35,090
529 290 0.791 0.151 0.007 1.1 4,623
534 83 0.395 0.032 0.007 –29.7 15,561
D15C 502 241 0.223 0.165 0.055 –408.8 1,080
505 248 0.242 0.194 0.071 –529.1 2,106
506 269 1.974 1.026 0.145 –524.4 1,625
507 262 1.275 0.798 0.078 –416.6 1,634
509 248 0.256 0.213 0.109 –586.7 1,964
513 83 0.291 0.288 0.067 –344.7 4,376
514 290 0.913 0.575 0.088 –471.1 1,918
519 255 1.494 0.916 0.127 –471.8 1,323
522 262 1.117 0.526 0.447 –544.5 1,893
523 255 1.025 0.664 0.133 –448.2 1,470
524 241 3.916 2.583 0.086 –424.3 2,422
531 290 0.958 0.417 0.089 –461.1 1,574
533 290 0.741 0.401 0.102 –408.8 1,483
535 269 0.879 0.565 0.088 –514.0 2,376
536 83 0.665 0.466 0.033 –366.6 2,530
D30C 501 241 0.480 0.237 0.043 –220.8 12,644
503 255 0.715 0.165 0.021 –201.8 4,078
504 241 1.017 0.484 0.115 –369.9 4,302
512 262 0.346 0.147 0.017 –213.4 3,431
515 262 0.938 0.160 0.002 –193.3 11,166
516 83 0.344 0.155 0.031 –118.7 27,881
517 290 0.604 0.267 0.062 –432.6 4,184
518 269 0.571 0.133 0.058 –114.7 1,598
520 248 0.629 0.158 0.043 –201.6 10,648
525 255 ∞ (10 ) 15
0.162 0.046 –351.5 4,086
526 83 0.540 0.092 0.093 –65.9 7,020
527 269 1.790 0.422 0.060 –499.5 2,615
528 290 0.905 0.270 0.044 –444.4 2,642
530 290 1.021 0.199 0.067 –130.5 3,311
532 248 1.478 0.563 0.152 –429.1 3,003
28 | DUPRAT ET AL.

T A B L E A7 Exposure condition similar to XC1 — Concrete CI

Anodic Cathodic Iron mass Corrosion


Cover Specimen Tafel slope Tafel slope βc loss Δm iron potential Ecorr Polarization
condition reference Time (d) βa (V/dec) (V/dec) (g) (mV/SCE) resistance Rp (Ω)
D15NC 601 109 0.149 0.125 0.001 –199.0 11,817
604 350 0.308 0.096 n. m. –100.9 44,406
616 295 0.207 0.100 n. m. –194.1 14,136
619 109 0.315 0.008 n. m. 20.3 30,911
628 350 0.921 0.126 0.038 –28.7 72,850
633 350 0.556 0.221 0.043 –48.7 47,845
D15C 602 309 1.297 0.624 0.053 –512.3 4,460
605 350 1.064 0.392 0.341 –421.1 9,819
606 350 1.111 0.600 0.081 –443.1 4,984
607 109 1.277 0.635 0.016 –493.3 12,717
608 302 0.354 0.395 0.043 –290.5 6,034
609 295 0.324 0.348 0.013 –279.2 7,603
612 132 0.333 0.276 0.037 –235.5 30,918
614 302 1.854 0.656 0.091 –431.1 2,729
617 350 1.436 0.716 0.091 –513.6 7,732
623 329 0.625 0.307 0.075 –385.5 12,778
625 109 0.282 0.199 0.040 –468.5 9,515
626 132 1.314 0.268 0.049 –429.4 18,795
627 329 0.494 0.286 0.035 –327.1 10,929
629 309 0.601 0.285 0.072 –373.3 11,308
634 295 0.189 0.123 0.012 –325.4 16,470
D30C 603 132 0.522 0.275 0.039 –49.0 164,924
610 329 0.113 0.128 0.018 –102.3 51,411
611 309 0.530 0.212 0.024 –67.1 86,075
613 302 0.967 0.445 0.109 –527.5 3,114
615 295 0.078 0.103 n. m. –95.9 52,227
618 309 0.570 0.163 0.013 –144.4 25,634
620 302 1.013 0.224 0.020 –62.1 46,572
621 350 0.779 0.433 0.117 –388.6 8,148
622 109 0.200 0.157 n. m. –205.0 98,972
624 309 1.772 0.529 0.090 –426.7 4,481
630 350 0.768 0.192 0.044 –348.6 17,361
631 350 ∞ (10 ) 15
0.480 0.063 –148.8 52,160
632 295 0.218 0.135 0.023 –190.0 59,078
635 132 0.226 0.124 0.019 –244.0 103,753
636 109 0.660 0.175 n. m. –417.4 190,490
DUPRAT ET AL. | 29

T A B L E A8 Exposure condition similar to XC1 — Concrete CII

Anodic Cathodic Iron mass Corrosion


Cover Specimen Tafel slope Tafel slope βc loss Δm iron potential Ecorr Polarization
condition reference Time (d) βa (V/dec) (V/dec) (g) (mV/SCE) resistance Rp (Ω)
D15NC 701 350 0.309 0.156 0.034 2.2 83,878
711 350 0.590 0.139 0.026 26.3 44,905
715 109 0.075 0.102 0.003 –44.4 6,460
719 350 2.206 0.287 n. m. –123.9 35,107
727 295 0.271 0.082 n. m. 96.5 9,942
729 109 0.135 0.070 0.019 –45.3 8,055
D15C 703 350 1.348 0.748 0.130 –467.8 5,757
705 350 1.013 0.475 0.109 –459.1 2,500
708 302 0.738 0.370 0.083 –488.3 3,080
709 329 0.757 0.293 0.074 –346.5 8,867
712 309 1.161 0.451 0.095 –529.2 2,256
713 132 0.509 0.297 0.067 –385.1 9,985
714 302 0.340 0.213 0.089 –197.7 20,201
717 309 0.263 0.162 0.095 –466.0 3,160
718 109 0.271 0.180 0.038 –415.3 9,390
721 295 0.779 0.348 0.086 –547.2 3,765
724 350 1.095 0.591 0.120 –491.2 2,134
725 132 0.390 ∞ (10 ) 9
0.063 –305.1 24,380
733 309 1.583 0.540 0.091 –589.5 1,474
735 295 0.193 0.112 0.052 –461.3 1,206
736 109 1.590 0.390 0.033 –406.7 9,067
D30C 702 329 ∞ (10 ) 15
0.190 0.077 –542.6 2,140
704 350 0.250 0.120 0.115 –697.5 2,204
706 109 0.066 0.097 0.003 –115.8 51,924
707 302 0.174 0.130 0.026 –203.0 33,197
710 329 ∞ (1015) 0.315 0.092 –661.8 1,653
716 302 1.272 0.166 0.054 –674.1 1,234
720 309 ∞ (1015) 0.146 0.067 –678.5 1,415
722 309 ∞ (1014) 0.115 0.062 –630.2 1,112
723 329 0.428 0.171 0.047 –234.1 24,713
726 109 ∞ (10 ) 9
0.360 0.006 –215.6 37,834
728 350 1.027 0.336 0.112 –587.2 2,251
730 132 0.176 0.145 0.038 –126.6 46,632
731 132 0.201 0.183 0.062 –189.0 22,788
732 295 1.340 0.615 0.050 –691.7 1,512
734 295 0.158 0.103 0.032 –657.1 563
30 | DUPRAT ET AL.

T A B L E A9 Exposure condition similar to XC1 — oncrete CIII

Anodic Cathodic Iron mass Corrosion


Cover Specimen Tafel slope Tafel slope βc loss Δm iron potential Ecorr Polarization
condition reference Time (d) βa (V/dec) (V/dec) (g) (mV/SCE) resistance Rp (Ω)
D15NC 802 350 0.651 0.117 n. m. 3.4 174,680
809 350 1.182 0.095 0.015 105.3 4,339
813 350 0.299 0.105 0.056 129.6 38,818
816 109 0.816 0.137 0.009 64.7 29,150
829 295 0.375 0.113 n. m. 76.3 10,514
833 109 1.042 0.070 0.006 –2.9 14,360
D15C 801 309 0.844 0.590 0.072 –340.9 13,938
804 295 0.207 0.173 0.032 –447.4 3,583
805 109 1.712 0.878 0.009 –349.3 52,165
807 329 1.086 0.724 0.079 –426.9 6,315
808 350 0.884 0.559 0.093 –482.5 52,27
814 109 0.995 0.793 0.019 –387.8 12,199
818 132 1.274 0.787 0.038 –318.5 35,866
819 309 1.241 1.037 0.067 –456.5 7,264
820 302 1.137 0.922 0.077 –428.0 7,353
821 350 1.061 0.828 0.087 –457.5 4,362
823 295 0.374 0.277 0.011 –364.8 5,606
827 329 0.909 0.750 0.052 –437.7 6,461
828 302 0.754 0.678 0.070 –393.3 7,745
831 132 0.938 0.530 0.058 –286.1 42,290
836 350 0.864 0.588 0.088 –463.1 5,284
D30C 803 350 0.664 0.196 0.062 –253.3 30,689
806 350 0.629 0.311 n. m. –364.1 6,614
810 295 0.625 0.135 0.019 –243.2 34,949
811 350 1.224 0.476 0.070 –322.3 14,243
812 302 0.488 0.215 0.024 –45.5 79,046
815 109 0.680 0.223 0.011 –197.2 29,381
817 302 0.468 0.188 0.040 –37.7 27,760
822 309 0.404 0.221 0.059 –234.3 12,667
824 309 0.823 0.289 0.062 –269.3 32,299
825 350 0.372 0.195 0.091 –216.9 39,742
826 132 0.650 0.159 0.036 –63.4 43,755
830 109 0.980 0.126 0.029 –11.72 208,127
832 132 0.848 0.196 0.043 –21.41 54,238
834 295 0.650 0.198 0.016 –184.1 27,015
835 309 0.730 0.238 0.056 –115.6 16,912

You might also like