Download as pdf or txt
Download as pdf or txt
You are on page 1of 20

Seminar 1: Pathological study in indexed and refereed journals.

CORROSION-INDUCED CRACKING FRAGILITY OF RC BRIDGE WITH IMPROVED


CONCRETE CARBONATION AND STEEL REINFORCEMENT CORROSION
MODELS

Student name: Carlos Alberto Jaramillo Garcia


Teacher's name: Ary Alain Hoyos Montilla
Course: Patología 1
Date: 01/09/2022

Universidad Nacional de Colombia Sede Medellín


Sede Medellín
2022 - 2
CONTENTS
pág.
INTRODUCTION ………………………………………………………………………… 3
SUMMARY ……………………………………………………………………………….. 4
CASE ANALYSIS ………………………………………………………………………… 5
CONCLUSIONS………………………………………………………………………….. 6
BIBLIOGRAPHY …………………………………………………………………………. 7
ARTICLE…………………………………………………………………………………… 8
INTRODUCTION

In this first seminar, the main objective will be to analyze the presence of the
phenomenon of carbonation in concrete structures, more specifically in
vehicular bridges, carrying out a comparative analysis regarding the effect of
CO2 in the environment that reacts with humidity inside the concrete pores,
converting calcium hydroxide to calcium carbonate, reducing the pH to a more
neutral state.
To evaluate the content of Co2 present in different points of the structures,
tests were carried out that, from a visual analysis, deduce the presence of the
pathology already mentioned above, in order to generate the results that the
article shows us, taking into account the concentration of vehicular gases
(CO2) or other external factors that may affect or influence the development of
carbonation in said structures.
The article will be included with its respective bibliography where the background and
research that addresses the issue of carbonation types worldwide will be shown, taking
into account new strategies to mitigate the adverse effects on the structures of RC
reinforced concrete bridges.

3
SUMMARY

The corrosion process of reinforcing steel can be defined as the electrochemical reaction
that leads to obtaining an iron oxide that has a volume of 3 to 4 times greater than that of
the initial steel, this increase in volume in the reinforcement generates pressures internal
that lead to the detriment of the durability of the concrete, producing fissures, cracks and
delaminations, damage that can even lead to the structure's collapse. For corrosion to
occur, the presence of 3 fundamental factors is re quired: the depassivation of the
reinforcing steel, the entry of water andoxygen.
Corrosion of steel reinforcement is a long-lasting threat to bridges in general from their
construction to the time of their useful life.
There are two phases in the corrosion of steel reinforcement: The initiation phase and the
propagation phase: The first phase consists of the carbonation of the concrete and is a
phenomenon that normally occurs in urban areas due to the existing atmospheric
conditions and to the rapid growth of CO2 concentration due to industrial prosperity.
Carbonation of concrete is the loss of pH that occurs when atmospheric carbon dioxide
reacts with moisture within the pores of the concrete and converts calcium hydroxide (with
a high pH) to calcium carbonate, which has a lower pH. bass. neutral. The second phase
consists of the transmission of corrosion and the accumulation of oxide products, which
leads to cracking of the concrete.
Probabilistic concrete durability studies related to carbonation of RC bridges under urban
atmospheric conditions are limited. Most of these researchers focus on chloride-related
corrosion in the marine environment when performing probabilistic analyses. But it is
important to note that the number of CR bridges in urban cities is much higher than in
marine conditions.

4
CASE ANALYSIS

In this article, we can observe in a general way how is the carbonation process in the
concrete of RC bridges, which generally occurs because carbon dioxide (CO2) from the
atmosphere is going to penetrate the concrete deck. and react with the carbonateable
substance of the concrete.
There are different models for the study of this pathology or these injuries: Empirical
models, Formation of probabilistic models, Deterministic model, Calibration and selection
of model parameters and Summary of the model, these models are an abstract,
conceptual, graphic, visual or physics of carbonation phenomena whose purpose is to
analyse, describe, explain, simulate (in general, explore, control and predict) these
phenomena or these processes. All these models allow to determine a final result from
some input data. The creation of a model is considered an essential part of all scientific
activity.
At corrosion initiation, the high alkalinity of newly constructed concrete provides a so-
called passive layer (concrete cover) protection to the steel reinforcement against
environmental corrosion. The carbonation process lowers the alkalinity of the concrete
and thus destroys the protective effect of the concrete. That is, corrosion of steel occurs
when the carbonation limit reaches the surface of the steel reinforcement.
Corrosion product (rust) from reinforcing steel will accumulate around the corroded steel
bar. The oxide products have more volume than the original steel, causing a volumetric
expansion effect and ultimately leading to cracking of the concrete. The rate of oxide
accumulation is inversely related to the amount of oxide products due to the protective
effect of increasing oxide on ion diffusion.
For the RC bridge structure, the brittleness of corrosion-induced cracking is defined as
the conditional probability of reaching the limit state of cracking caused by expansion of
the oxide product.
Carbonation is not only due to the action of environmental CO2, but also from the moment
of construction, which affects in the first instance the resistance of the structure to
withstand this effect, such as porosity, the quality and quantity of cement, the level of
compaction, the type and times of curing and the relationship
water-cement, it is necessary to have a good mixture and placement of the concrete in
the construction stage, because if cracks are generated during the setting and are not
corrected, they can cause carbonation to be much faster and earlier.

5
CONCLUSIONS

- It is important to generate new practices and studies to determine the rate of


carbonation, thus determining solutions that help improve the durability of structures.
- Identifying the issue of durability in structures has been an achievement in construction,
due to the fact that the repercussions that can be generated and that reduce the quality of
the structures can be mitigated, however, it is still a challenge to deal with the issue of
carbonation since the lack of information is incipient in this article, so it is necessary to
continue with the advances that are required.
- According to the methods of the studies carried out and the tests, it was possible to
observe and determine that the re-alkalinized structure must be protected with an
adequate surface coating. The idea of these re-alkalinization processes is to maintain the
passivity of the steel, that is, that the reactions generated in the steel stop acting so that
the properties of the steel are not affected by corrosion.

6
BIBLIOGRAPHY
- (Sun et al., 2020)

- Tuutti K. Corrosion of steel in concrete. Stockholm: Swedish Cement and Concrete


Research Institute; 1982. Report nr 0346-6906.
Google Scholar

- S.J. Kwon, U.J. Na. Prediction of durability for RC columns with crack and joint under
carbonation based on probabilistic approach

Int J Concr Struct Mater, 5 (1) (2011), pp. 11-18

7
Engineering Structures 208 (2020) 110313

Contents lists available at ScienceDirect

Engineering Structures
journal homepage: www.elsevier.com/locate/engstruct

Corrosion-induced cracking fragility of RC bridge with improved concrete T


carbonation and steel reinforcement corrosion models

Bo Suna,b, Ru-cheng Xiaob, , Wei-dong Ruana, Peng-bo Wangc
a
Department of Civil Engineering, Zhejiang University of Technology, D515 Structural Engineering Building, 288 Liuhe Road, Hangzhou, Zhejiang 310000, China
b
Department of Bridge Engineering, Tongji University, 619 Bridge Building, 1239 Siping Road, Shanghai 200092, China
c
Beijing Urban Construction Design & Development Group Co., Limited, No. 5, Fuchengmen Beidajie, Xicheng District, Beijing 100034, China

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

Keywords: Corrosion-induced cracking is one important limit state in the durability performance analysis of the re-
Reinforced concrete bridge inforcement concrete (RC) structures. A comprehensive probabilistic approach is established for the corrosion-
Concrete carbonation induced cracking fragility analysis of the RC bridge in the urban atmospheric condition with improved concrete
Steel reinforcement corrosion carbonation and steel reinforcement corrosion models. The improved deterioration models have a probabilistic
Probabilistic model
formation by adding correction and error terms to the existing deterministic models. The deterministic parts are
Cracking fragility
selected from reviews and discussions on existing models. The correction terms are sets of explanatory functions
representing the influencing factors related to the potential bias in the deterministic models. The statistical
distributions of unknown model parameters are calibrated and the optimum collections of the explanatory
functions are selected through the Bayesian rule based on long-term data from onsite tests or natural exposure
experiments. The probabilistic analysis approach for the corrosion-induced cracking fragility are generated
based on the improved deterioration models. Fragility curves, parameter sensitivities and random variable
importance are achieved for the example RC bridge. The results show that increases on the concrete strength,
cover depth and steel bar diameter, or decrease on the CO2 density, are efficient countermeasures to improve the
durability performance of the RC bridge against corrosion-induced cover cracking and the uncertainty of the
problem mainly comes from the concrete carbonation model.

1. Introduction probabilistic studies on the carbonation-related durability performance


of RC bridge in the urban atmospheric condition are in limited num-
Steel reinforcement corrosion is a long-lasting threat to RC bridges bers. Most researchers focus on chloride-related corrosion in the marine
during life time and includes two stages: initiation stage and propaga- environment when doing probabilistic analysis of the RC bridge [6–8].
tion stage [1]. For the first stage, concrete carbonation is considered in However, the amount of RC bridges in the urban cities is much larger
this paper since it is a commonly occurred phenomenon in the urban than that in the marine conditions and carbonation is a parallel phe-
area (atmospheric condition) with the fast growth of CO2 concentration nomenon to the chloride penetration even in the marine area. Thus a
due to the industrial prosperity [2]. Concrete carbonation is a chemical comprehensive probabilistic evaluation on the carbonation-related
process that weakens the protection provided by the concrete to the durability performance for the RC bridges with respect to the corrosion-
reinforcement and initializes the corrosion of the reinforcement. For the induced cracking limit state should attract more attentions.
second stage, corrosion propagation and accumulation of the rust pro- Some researchers set the depassivation of the steel reinforcement as
ducts lead to cracking of the concrete cover. Moreover, corrosion-in- the limit state and study the durability performance of RC structures in
duced cracking is one important limit state in the durability perfor- the first stage. Kwon and Na [2] obtained the carbonation durability
mance analysis of the RC bridge. The deterioration process from failure through Monte Carlo simulation based on concrete carbonation
concrete carbonation, corrosion initiation, rust accumulation to cover velocities measured from filed investigations. Na et al. [9] proposed a
cracking contains uncertainties from environmental, material and stochastic approach for predicting service life of RC structure subjected
structural properties. While there are abundant researches in the de- to carbonation with the spatial variation of material, geometric prop-
terministic sense [3–5] (experimentally or theoretically), the erties of RC structures, and environmental factors. Teplý et al. [10]


Corresponding author.
E-mail addresses: sunbo2017@zjut.edu.cn (B. Sun), xiaorc@tongji.edu.cn (R.-c. Xiao), wdruan@zju.edu.cn (W.-d. Ruan), wangpengbo@bjucd.com (P.-b. Wang).

https://doi.org/10.1016/j.engstruct.2020.110313
Received 11 August 2019; Received in revised form 29 December 2019; Accepted 29 January 2020
Available online 12 February 2020
0141-0296/ © 2020 Elsevier Ltd. All rights reserved.
B. Sun, et al. Engineering Structures 208 (2020) 110313

Nomenclature z∗, u∗ design point

Roman case letters Greek case letters

Cl Cl− content σ standard variation


cCO2 CO2 density around concrete surface ε standard normal random variable
DCO2 CO2 diffusion coefficient in the concrete αrust rust type coefficient
dc concrete cover depth ρrust mass density of the rust product
db0 uncorroded steel bar diameter φcr creep coefficient of concrete
d1 free pore band thickness at steel/concrete interface νc Poisson’s ratio of concrete
Ec concrete original elastic modulus β reliability index
Eef concrete effective elastic modulus λ importance measure
fcuk concrete characteristic compression strength Θ, θ model parameter
fct concrete tensile strength θg deterministic design parameter
fc28 concrete 28-day compression strength θz distribution parameter
Icorr corrosion rate
kp rust accumulation coefficient Function case letters
mCO2 amount of CO2 absorbed by unit concrete
RH relative humidity F (∙) fragility
s given boundary condition ∼
F (∙) predictive fragility estimate
T temperature f (∙) posterior distribution
t service time g (∙) limit state function
ti corrosion initiation time hi (∙) explanatory function
tp corrosion propagation time J(∙) Jacobian matrix
Wrust rust product amount L (∙) likelihood function
Wcrit critical rust product amount p (∙) prior distribution
w c water to cement ratio X (∙), Ico rr (∙) improved probabilistic model
X concrete carbonation depth  (∙), Icô rr (∙) deterministic model
X
x design parameter σε model error
xd deterministic design parameter γ (∙) correction term
xr random design parameter ∇ (∙) gradient vector
z random variable

studied the durability design and provide comments on the assessment are generated based on the improved models of concrete carbonation
of RC structures in relation to existing codes. Some researchers move a and steel reinforcement corrosion. Fragility curves with both the pre-
further step to include the steel reinforcement corrosion stage in the dictive estimates and confidence bounds are obtained for the example
durability probability analysis. Teplý [11] studied the effect of and bridge based on the proposed probabilistic approach and improved
compromise between the service life and the time-varying reliability deterioration models. Parameter sensitivity and random variable im-
level of deteriorating RC structures. Niu [12] established efficient cor- portance are also analyzed to provide guidance on the design and
rosion models and studied the corrosion initiation and cracking prob- model optimization for the durability performance of the RC bridge.
abilities of RC structures. Marques and Costa [13] conducted both The rest of the paper is organized as follows. Section 2 and Section 3
safety factor and probabilistic approaches to estimate the service life review existing models and generate the improved models for the
periods of RC structures with carbonation performance properties de- concrete carbonation and steel reinforcement corrosion. Section 4 dis-
termined from experimental projects. However, those valuable attempts cusses the fragility analysis approach based on the improved dete-
have limitations: (1) The concrete carbonation and steel reinforcement rioration models; fragility curves, parameter sensitivities and random
corrosion models are simply chosen from the existing models without variable importance are obtained for the example bridge to illustrate
detailed discussions and remain deterministic. (2) The uncertainties this process. Section 5 draws some general conclusions.
accounted in those works mainly focus on aleatory aspect on material
and structural properties. (3) The epistemic uncertainties arise from the 2. Concrete carbonation
modeling and analysis process (assumptions and simplifications) are
commonly ignored, which are actually important part in the probabil- 2.1. Theoretical model
istic analysis.
In this paper we develop improved deterioration models for the The carbon dioxide (CO2) in the atmosphere will penetrate into the
concrete carbonation and steel reinforcement corrosion by adding concrete cover and react with the carbonatable substance in the con-
correction terms to the existing deterministic models. The deterministic crete [14]. The carbonation depth X is used to measure the carbonation
parts are selected based on reviews/discussions of the existing models process. The widely used theoretical model for X is derived by adopting
and the correction terms are sets of explanatory functions representing the assumption that the diffusion process obeys the Fick’s first law [1]
the influencing factors widely concerned in the current research to
maximize the scientificity and adoptability of the improved models. The 2DCO2 cCO2
X= t
statistical distributions of the unknown model parameters are cali- mCO2 (1)
brated and the optimum collections of the explanatory functions are
selected through the Bayesian rule based on long-term data from the where X = distance from the carbonation frontier to the concrete sur-
onsite tests or natural exposure experiments. A probabilistic analysis face at service time t , DCO2= diffusion coefficient of CO2 in the concrete
approach for the corrosion-induced cracking fragility of the RC bridge and cCO2= CO2 density around the concrete surface. mCO2 is the amount
of CO2 absorbed by unit volume of concrete and a theoretical solution is

2
B. Sun, et al. Engineering Structures 208 (2020) 110313

given by Papadakis [15] according to the mass-balance condition of the 2.3. Probabilistic model formation
carbonatable substance
Following the general probabilistic model development process
2DCO2 cCO2 proposed by Sun et al. [22,23] and Gardoni et al. [24,25], we define the
X= t
[Ca (OH )2 ]0 + 3[CSH ]0 + 3[C3 S ]0 + 2[C2 S ]0 (2) probabilistic model formula for the concrete carbonation depth by
adding correction and error term to the existing deterministic model
where [Ca (OH )2]0 , [CSH ]0 , [C3 S ]0 and [C2 S ]0 are the initial densities of
the carbonatable substance in the concrete.  (x) + γ (x, θ) + σε
X (x, Θ) = X (3)
X

where X (x, Θ)= improved probabilistic model for the carbonation


2.2. Empirical models depth, x= design parameters accounting for the aleatory uncertainties
in the carbonation analysis, Θ = (θ, σ )= unknown model parameters
The exact values of the model parameters DCO2 , [Ca (OH )2]0 , [CSH ]0 , that need detailed calibration and selection, X (x)= existing determi-
[C3 S ]0 and [C2 S ]0 in the theoretical models are usually hard to achieve. nistic model, γX (x, θ)= correction term and σε is the error term which
Thus engineers have to turn to efficient empirical models in the en- accounts for the epistemic uncertainties together with Θ.
gineering practice. The existing empirical models can be divided into Three assumptions are applied to the above probabilistic model
three types with respect to the key factors used to predict the carbo- formation:
nation depth: (1) test-based model, (2) w/c-based model and (3)
strength-based model. Some of the existing empirical models for the (1) Homoscedasticity assumption – the model variance σ 2 (a quantita-
concrete carbonation are listed in Table 1 for better understanding of tive measure of the model error) is independent of the design
the discussions below.  (x) that is related to specific
parameter x . All the potential bias in X
Table 1 shows that the test-based models require specific char- design properties are considered and corrected by the correction
acteristics achieved from the laboratory test to conduct the carbonation term γX (x, θ) .
prediction, e.g., the carbonation resistance in the FIB model [16]/CEB (2) Normality assumption – the model prediction error of the improved
model [17]. The dependency of the laboratory test restricts the appli- probabilistic model has a normal distribution (ε ~N (0, 1) ). The
cation of the test-based models in the design practice and the high cost prediction results for a given set of design value x are unbiased
for the repeatability makes them unsuitable for a probabilistic design (uniformly distributed around the median prediction).
scheme. The other two types of models choose the commonly used (3) Additivity assumption – the deterministic model X  (x) , correction
characteristics of the concrete (water to cement ratio or compression term γX (x, θ) and error term σε are in an additive form. The linear
strength) to construct the prediction models, together with other pos- relationship of the model components is convenient for the for-
sible influencing factors including the material composition, curing mulation of the parameter calibration and selection process.
condition and environmental condition. However, more discussions are
needed on those model parameters: some of them are not convenient to Proper transformation of model input x and output X is required in
achieve in the design practice; some of them may have minor influence order to satisfy the three assumptions. In this paper we adopt the nat-
on the prediction results of the carbonation depth. Moreover, all the ural logarithm as the variance-stabilizing transformation and Eq. (3) is
existing models are in a deterministic sense, which is inconsistent with rewritten as
the probabilistic requirement of the modern design.
The discussions above show that improvements are required on the  (x)] + γ (x, θ) + σε
ln[X (x, Θ)] = ln[X (4)
X
existing models for the prediction of the concrete carbonation depth.
The improved model should be probabilistic and the model parameters The proposed probabilistic model for the concrete carbonation
should be well calibrated and selected to reflect the most sensitive depth is unbiased and accounts for all types of uncertainties.
characteristics influencing the concrete carbonation process.

Table 1
Existing empirical models for the concrete carbonation.
Authors Type Model Description

FIB (2010) [16] test-based X= −1


2k e k c RNAC,0 cCO2 W (t ) t k e= environmental function, k c= execution transfer parameter, W (t )= weather function and
RNAC,0 is defined as the carbonation resistance obtained from the laboratory test
CEB 238 (1997) [17] test-based 2cCO2 k 0= testing method factor, k1= relative humidity factor, k2= curing factor and RC65 is the
X= k 0 k1 k2 t
RC 65 tested carbonation resistance in CEB 238
Teply (2010) [10] w/c-based 1.09DCO2 cCO2 10−6
c= cement content, w= water content, P= concrete admixture content, k= concrete admixture
X= t factor, ρc = mass density of cement and f (RH )= relative humidity factor
0.218(c + kP )
3
⎛ ⎞
w − 0.267(c + kP )
DCO2 = 6.1 × 10−6 ⎜ c + kP ⎟
⎜ 1000 +w ⎟
⎝ ρc ⎠
× f (RH )2.2
Japan model (1963) w/c-based w c − 0.25 w c= water cement ratio, rc= cement type factor, ra= aggregate type factor and rs= concrete
w c > 0.6 X = rc ra rs t
[18] 0.3(1.15 + 3w c ) admixture factor
4.6w c − 1.76
w c ⩽ 0.6 X = rc ra rs t
7.2
Zhu (1992) [19] w/c-based γ1= cement type factor, γ2= fly ash factor and γ3= environment factor
X= (
γ1 γ2 γ3 12.1
w
c
− 3.2 ) t
CECS (2007) [20] strength- X = 3K CO2 K kl K kt K ks KF see Appendix A.
based 58
× T1 4RH1.5 (1 − RH ) ⎛ − 0.76⎞ t
⎝ fcuk ⎠
Smolczyk (1962) [21] strength- 1 1 Fc= concrete compression strength and Fg= critical un-carbonized strength of concrete
based X = 250 ⎛ ⎜ − ⎞ t

Fc Fg
⎝ ⎠

3
B. Sun, et al. Engineering Structures 208 (2020) 110313

Table 2 environmental condition, we select h4 (x) = cCO2 c0 to capture the pos-


Data ranges collected for the concrete carbonation depth. sible influence of CO2 density, h5 (x) = T T0 to reflect the potential in-
Variable Symbol Range fluence of temperature and h6 (x) = RH RH0 to represent the influence
of relative humidity, where T0 = 13oC and RH0 = 0.7 are the reference
Service time (year) t 2–50 annual temperature and relative humidity respectively.
Characteristic strength (MPa) fcuk 15–55
The data {Xi , x i} used to calibrate the model parameter Θ for the
CO2 density (%) cCO2 0.03–0.0432
probabilistic carbonation depth model are all collected from the onsite
Annual temperature (oC) T 1.2–21.4
Annual relative humidity RH 0.5–0.81
carbonation tests on the real bridges in service [29–35]. The onsite data
reflect the deteriorative properties of the concrete in the exact service
conditions. The test results of the concrete carbonation depth at dif-
2.4. Deterministic model ferent service times are collected with the corresponding concrete
properties and environmental conditions. The data ranges collected are
The deterministic model X (x) is adopted to account for the existing listed in Table 2.
knowledge and experience on the concrete carbonation process. Thus
 (x) should be selected or constructed based on the existing models
X 2.6. Model parameter calibration and selection
discussed in Sections 2.1 and 2.2 to reflect the physical rules in the
carbonation process and maximize the acceptance of the improved The unknown model parameters Θ = (θ, σ ) are defined as set of
probabilistic model. The theoretical and empirical models clearly show random variables to capture the epistemic uncertainties. With the se-
a linear relationship between the carbonation depth and the square root  (x) , {hi (x)} and the onsite carbonation data {Xi , x i} , the un-
lected X
of time, i.e., known statistical properties of Θ are estimated through the Bayesian
 (x) = K (x) t approach [36]
X (5)
f (Θ) = κL (Θ) p (Θ) (8)
where K (x) is the carbonation coefficient function describing the in-
fluences of different factors on the concrete carbonation, e.g., where f (Θ) is the posterior distribution of Θ and
K (x) = 2DCO2 cCO2 mCO2 in Eq. (1). The CO2 density cCO2 is easy to κ = [ ∫ L (Θ) p (Θ) dΘ]−1 is a normalizing factor. f (Θ) incorporates two
quantify in the engineering practice and is an important influencing aspects of available information: (1) the prior knowledge on Θ before
factor that widely been used in the theoretical and empirical models; the consideration of the onsite test data, which is represented by the
thus we adopt it in the deterministic model. For other factors, con- prior distribution p (Θ) , and (2) the objective information from the
sidering that under the natural logarithm transformation, they could be onsite carbonation tests, which is represented by the likelihood function
transferred into the correction term γX (x, θ) and well calibrated/se- L (Θ) . Since the measured data set {Xi , x i} (the ith observation) is the
lected during the model parameter calibration and selection process. exact carbonation depth Xi with respect to the ith set of design value x i ,
The probabilistic carbonation depth model in Eq. (4) is rewritten as L (Θ) has the following form
cCO2
ln[X (x, Θ)] = ln
c0
t + γX (x, θ) + σε L (Θ) ∝ ∏ P [σεi = Xi − Xi (xi) − γX (xi, θ)] (9)
⏟
X (x) (6) i (x i) − γ (x i, θ) is the prediction bias of the deterministic
where Xi − X X
model X (x) for the ith observation. More details on the Bayesian model
where c0 = 0.03% is the standard CO2 density in the air.
parameter calibration process can be found in Sun et al. [22,23] and
Gardoni et al. [24,25].
2.5. Correction term and test data
The posterior distribution f (Θ) is computed by the Markov Chain
Monte Carlo (MCMC) simulation [37] based on the Bayesian approach
The correction term γX (x, θ) is adopted to capture the carbonation
 (x) . A described above. For p (Θ) , a non-informative prior is selected in order
coefficient function K (x) and correct the potential bias in X
to minimize the influence of p (Θ) since our prior knowledge on it is
linear form for γX (x, θ) is adopted in this paper
unavailable [36]. A stepwise model selection process is applied to select
k the most sensitive explanatory functions {hi (x)} (the optimum combi-
γX (x, θ) = ∑ θi hi (x) nation) step by step for the improved probabilistic model:
i=1 (7)

where Σ[θi hi (x)], i = 1, ...,k , can be treated as a transformation of K (x) (1) For the first step, calibrate the posterior statistics of the model
and {hi (x)} is a set of “explanatory” functions capturing specific factors parameter Θ for the whole bunch of explanatory functions {hi (x)} ;
that may influence the prediction accuracy of the concrete carbonation (2) In the second step (first deletion step), delete the most insensitive
depth. explanatory function hi (x) (least informative) with the largest
The explanatory function hi (x) should be selected to reflect the coefficient of variation for θi and recalibrate the model parameter
possible influencing factors indicated by the existing models in Table 1 for the rest explanatory functions;
and based on the available test data. Firstly, h1 (x) = 1 is selected to (3) Repeat the deletion process step by step. The optimum combination
represent the constant part in K (x) and capture the potential bias that is for the explanatory functions arises when there is an unacceptable
independent of x . Secondly, some literatures [13,26] pointed out that increase on the model error (indicated by the posterior mean of σ )
the relationship between the carbonation depth X and service time t in the next step;
may not strictly following the square root order. We select h2 (x) = ln t (4) The selected model keeps a balance between model accuracy and
to evaluate this hypothesis. Other explanatory functions account for the model conciseness.
two main aspects of factors influencing the concrete carbonation:
concrete quality and environmental condition. For the concrete quality, The stepwise calibration and selection process for X (x, Θ) is shown
since the characteristic compression strength is a comprehensive index in Fig. 1. The mean value for σ (representing the model accuracy) and
reflecting the concrete quality and is more convenient to measure/re- coefficient of variation (COV) for θi (representing the sensitivity of
cord in the engineering practice compared with the material composi- θi hi (x) ) obtained are shown in Fig. 1 for each step. For Step 1, the mean
tion, we select h3 (x) = fcuk fC30 , where fC30 = 30 Mpa is the character- value of σ is 0.370 and the largest COV = 2.07 appears for θ6, showing
istic compression strength for concrete grade C30 [27,28]. For the that h6 (x) is the most insensitive explanatory function related to the

4
B. Sun, et al. Engineering Structures 208 (2020) 110313

beside the 1:1 line and two dashed lines are drawn to shown the ± 1 SD
of the median predictions. The improved model of concrete carbonation
depth is unbiased and more accurate. Moreover, the formation of the
improved model is probabilistic in nature and considers all types of
uncertainties.

3. Steel reinforcement corrosion

3.1. Corrosion initiation

The high alkalinity of newly built concrete provide a protective


passive layer (concrete cover) to the steel reinforcement against en-
vironmental corrosion. The carbonation process decreases the alkalinity
of the concrete and thus destroys the protective effect of the concrete
cover [11]. The steel corrosion happens when the carbonation frontier
reaches the surface of the steel reinforcement, i.e.,
X (xX |t = ti , ΘX ) = dc (12)
where ti is the corrosion initiation time and dc is the distance from the
surface of the concrete to the steel reinforcement. Based on the im-
Fig. 1. Stepwise model selection process for X (x, Θ) where a superposed cross
(×) indicates term to be removed. proved model for X (xX , ΘX ) in Eq. (11), ti is obtained as

prediction bias. Thus θ6 h6 (x) is removed from γX (x, θ) and the re- ln ti (x ti, Θti) =
(
ln(dc c0 cCO2 ) − θX 1 + θX 3
fcuk
f30
+ σX εX )
calculated mean value of σ remains 0.370, i.e., the same model accu- 0.5 + θX 2 (13)
racy is obtained with a more parsimonious model form (fewer number where ti (x ti, Θti) is the probabilistic model for the corrosion initiation
of model parameters). Repeating this stepwise deletion process until at time of the steel reinforcement, in which x ti = (cCO2, fcuk , dc ) and
Step 4, the mean value of σ (0.366) is very close to the largest COV of θ2 Θti = ΘX .
(0.414) and a further improvement on γX (x, θ) (Step 5) will bring an
increase on the model error. So the optimum γX (x, θ) is obtained at Step
3.2. Corrosion propagation
4
fcuk The corrosion of the steel reinforcement will propagate after the
γX (x, θ) = θ1 + θ2 ln t + θ3 steel is depassivated (t > ti ). During this phase, the corrosion depth is
f30 (10)
used to describe the propagation process [38,39], i.e.,
The posterior statistics (mean, standard deviation and correlation t
coefficient) of model parameters Θ = (θ1, θ2 , θ3 , σ ) for the selected p (t ) = ∫t
i
0.0116Icorr dt
(14)
γX (x, θ) are listed in Table 3. The selection of θ1 indicates that there is a
constant value in the carbonation coefficient K (x) that is independent where t − ti = tp is the corrosion propagation time and Icorr is the cor-
of any possible influencing factors. The selection of θ2 proves that the rosion rate measured as a current density of the corroded steel bar. Icorr
carbonation depth X dose not strictly follow a linear relationship with is an important quantity in describing the corrosion propagation of the
respect to the square root of service time t . The corrected exponent steel reinforcement.
value of t is 0.5 + θ2 based on the onsite data in this paper. The survival The existing models for the corrosion rate of the steel reinforcement
of θ3 shows that the compression strength is an important property are listed in Table 4. The existing models can be divided into two main
influencing the carbonation performance of the concrete and capturing types: constant models and time-variant models. The various influen-
the potential bias in the deterministic model. cing factors considered in the existing models mainly reflect the sur-
rounding conditions of the steel reinforcement including the concrete
condition and environmental condition. However, those existing
2.7. Model summary
models remain deterministic in nature and the influencing factors are
not discussed and selected in a systematical and comprehensive
The final form of the improved model for the concrete carbonation
manner. Improvements should be applied on the corrosion rate model
depth is
following the general process in Section 2.
cCO2 ⎤ f
ln[X (xX , ΘX )] = ln ⎡ t + θX 1 + θX 2 ln t + θX 3 cuk + σX εX 3.3. Improved corrosion propagation model
⎢ c0 ⎥ f30 (11)
⎣ ⎦
where xX = (t , cCO2, fcuk ) and ΘX = (θX 1, θX 2 , θX 3 , σX ) . We write the improved model for the corrosion rate of the steel
A comparison between the improved/probabilistic model and the reinforcement in a probabilistic form with the natural logarithm
existing empirical/deterministic model (CECS model [20]) based on the
predicted and measured value of concrete carbonation depth is shown Table 3
in Fig. 2. For an ideal model, the solid dots should perfectly stand along Posterior statistics of model parameters for X (x, Θ)
the 1:1 line. The prediction results of the CECS model in Fig. 2a clearly Parameter Mean Standard deviation Correlation coefficient
show large bias: most solid dots line upon the 1:1 line. The CECS model θ1 θ2 θ3 σ
tends to overestimate the carbonation depth due to a conservative
θ1 2.30 0.405 1.0
consideration, which is actually not suitable in a probabilistic design
θ2 –0.217 0.090 –0.93 1.0
scheme. The prediction results of the improved model in Fig. 2b is –1.41 0.189 –0.86 0.64 1.0
θ3
obtained with ε = 0 (median prediction). The improved model clearly σ 0.366 0.038 0.02 –0.02 –0.01 1.0
corrects the bias of the empirical model: the solid dots line more evenly

5
B. Sun, et al. Engineering Structures 208 (2020) 110313

Fig. 2. Comparison between (a) CECS model and (b) improved model based on measured and predicted concrete carbonation depth.

transformation (1 − w c )−1.64
Icorr ,0 =
dc (17)
̂ (x)] + γI (x, θ) + σε
ln[Icorr (x, Θ)] = ln[Icorr (15)
Moreover, the influencing factors f1 , f2 , ...,fk can be transferred into
where the indexes are defined earlier in Section 2. γI (x, θ) under the logarithm transformation. Thus Eq. (15) can be re-
The review of existing models suggests a multiply structure for the written as
̂ (x)
deterministic part Icorr (1 − w c )−1.64
ln[Icorr (x, Θ)] = ln + γI (x, θ) + σε
dc
̂ (x) = f1 f2 . ..fk Icorr ,0
Icorr (16) ̂ ⏟(x)
Icorr (18)

where Icorr,0 is the initial corrosion rate and f1 , f2 , ...,fk are k influencing For the correction term γI (x, θ) , h1 (x) = 1 is selected to capture the
factors. For Icorr,0 , since the water to cement ratio w c is an important potential bias that is independent of x and other explanatory functions
property influencing the porous structure of the concrete, which has are from the possible influencing factors f1 , f2 , ...,fk suggested by the
decisive effect on the diffusion of O2, the equation suggested by Guo existing models and available data. We select h2 (x) = ln tp to capture
and Trejo [38]/Vu and Stewart [44] is adopted the potential influence of the corrosion propagation time,
h3 (x) = Cl ClTh and h4 (x) = (Cl ClTh)2 to represent the influencing

Table 4
Existing empirical models for the corrosion rate
Authors Type Model Description

2
Andrade&Alonso (2004) [40] constant < 0.1/0.1 ~ 0.5/0.5 ~ 1.0/ > 1.0 uA/cm For negligible/low/moderate/high corrosion zone respectively
Ahmad&Bhattacharjee (2000) constant Icorr = 37.726 + 6.12 × 2.231CA2 B + 2.722B 2C2 %CaCl2 is the percentage of CaCl2 in unit weight of cement
[41] A = (c − 300) 50, B = (w c − 0.65) 0.075
C = (%CaCl2 − 2.5) 1.25
Guo&Trejo (2014) [38] time-variant 6 Cl + ClTh Cl= Cl− content, mc= concrete humidity, ClTh= Cl− threshold for the corrosion
Icorr = [e−6000(mc − 0.75) ] ⎛ ⎞
⎝ 2ClTh ⎠ initiation, Thigh= average high temperature, Tlow= average low temperature and
( 284.15
1 −1 as= adjusting factor of seasoning effect
T )⎤⎡
2283 (1 − w c )−1.64
×⎡e ⎤

⎣ ⎥
⎦⎣ dc ⎦
(T high −Tlow ) sin[2π (t − a )]
p s
×⎡ + 7.6⎤
8.6(t p − as )
⎣ ⎦
Zhou et al. (2010) [42] time-variant ln Icorr = 8.617 + 0.618 ln C (dc , t ) − 3034 T C (dc , t )= Cl− content at the surface of the steel reinforcement and Ct = Cl−
− 0.000105R c content of the concrete cover
R c = exp[8.03 − 0.549 ln(1 + 1.69Ct )]
CECS (2007) [20] time-variant ln Icorr = 8.617 + 0.618 ln C (dc , t ) see Appendix A.
−3034 (T + 273) − 5 × 10−3R c + ln mcl
Li (2004) [43] time-variant Icorr = 0.3683 ln tp + 1.1305 NA
Vu&Stewart (2000) [44] time-variant Icorr = Icorr ,0 α (t − ti ) β α = 0.85, β=–0.3 and A= 37.8/27 for RH = 0.75/0.8
A (1 − w c )−1.64
Icorr ,0 =
dc
Liu&Weyers (1998) [45] time-variant ln 1.08Icorr = 7.89 + 0.7771 ln 1.69Cl NA
−3006 T − 0.000116R c + 2.24tp−0.215
Yalcyn&Ergun (1996) [46] time-variant Icorr (t ) = Icorr ,0 exp(−1.1 × 10−3tp) NA
Icorr ,0 = 0.53

6
B. Sun, et al. Engineering Structures 208 (2020) 110313

effect of chloride content in concrete, h5 (x) = T T0 to reflect the pos- where db0 is the diameter of the uncorroded steel bar and αrust is the rust
sible influence of environmental temperature and h6 (x) = RH RH0 to type coefficient which is equal to the molecular mass ratio between the
capture the potential bias related to relative humidity. It worth mention steel (Fe) and rust (Fe(OH)2 or Fe(OH)3). Based on Eq. (20)–(22), the
that since the ohmic resistance of the concrete R c is hard to measure prediction model for the rust product expansion Wrust (xwr , Θwr ) is
and can be reflected by water to cement ratio and chloride/moisture t
content [38], it is not included in {hi (x)} in this paper. Wrust (xwr , Θwr ) = 2 ∫t (x ,Θ ) [0.098πdb0 Icorr (xI , ΘI ) αrust ] dt
i ti ti (23)
The data {Icorr , x i} used to estimate the statistical distributions of the
model parameters for the corrosion rate are from the long-term tests where xwr = (db0, xI ) and Θwr = ΘI .
conducted by Liu and Weyers [45]. The 5-year natural exposure ex-
periment on sets of RC slab samples obtained the corrosion current 4. Cracking fragility analysis
densities of the steel reinforcement under different influencing factors
and the data ranges collected are listed in Table 5. 4.1. Limit state function
The stepwise calibration and selection process for Icorr (x, Θ) is
shown in Fig. 3, based on the selected Icorr ̂ (x) , {hi (x)} and collected data For the RC bridge structure, the corrosion-induced cracking fragility
{Icorr , x i} . The optimum combination of {hi (x)} is obtained at Step 2 is defined as the conditional probability of attaining the cracking limit
(with σ = 0.281) and a further deletion of θ4 h4 (x) will cause an un- state caused by the rust product expansion for a given set of boundary
acceptable raise in σ . The selected γI (x, θ) is shown in Eq. (19) and the values
posterior statistics of model parameters Θ = (θ1, θ2 , θ3 , θ4 , θ5 , σ ) are
F (s, Θ) = P [g (x, Θ) ⩽ 0 |s, Θ] (24)
listed in Table 6. The most sensitive influencing factors are identified
and the following observations are noteworthy: (1) The selection of θ1 where F (s, Θ)= conditional probability of cover cracking, s= boundary
indicates that a constant bias exists in the selected Icorr ̂ (x) , which is in values and g (x, Θ) is the limit state function defined as the amount of
accordance with the parameter A in Vu&Stewart model [44]. (2) The the rust product reaches a certain threshold
survival of θ2 proves that a time-variant model is more suitable for the
g (x, Θ) = Wcrit (xwc) − Wrust (xwr , Θwr ) (25)
corrosion rate. (3) The selection of both θ3 and θ4 shows that the pre-
diction bias in Icorr ̂ (x) has a nonlinear relationship with the chloride where x = xwc ∪ xwr and Θ = Θwr . Wcrit (xwc) is the critical amount of the
content in the concrete. (4) The survival of θ5 indicates that the en- rust product for cover cracking and the model suggest by Liu [48] is
vironmental temperature is an important/sensitive factor influencing adopted in this paper
the prediction results of the corrosion rate of the steel reinforcement.
( )
d f
b +a 2 2
2
πρrust db0 ⎡d1 + Ec ct 2 2 + νc ⎤
⎣ ef b −a ⎦
Cl Cl ⎞ T
γI (x, θ) = θ1 + θ2 ln tp + θ3 + θ4 ⎛ + θ5
⎜ ⎟
Wcrit (xwc) =
1 − αrust ρrust ρst
ClTh ⎝ ClTh ⎠ T0 (19) (26)
where ρrust = mass density of the rust product, d1= thickness of the free
3.4. Model summary pore band at the steel/concrete interface, fct = 0.5 fc28 = tensile
strength of the concrete, in which fc28 is the 28-day compression
The final form of the improved model for the corrosion rate is strength, Eef = Ec (1 + φcr )= effective elastic modulus of the concrete,
in which Ec , φcr are the original elastic modulus and creep coefficient of
(1 − w c )−1.64 the concrete respectively, a = (db0 + 2d1) 2 , b = a + dc , ρst = mass
ln[Icorr (xI , ΘI )] = ln ⎡ dc
⎤ + θI 1 + θI 2 ln[t − ti (x ti, Θti)]
⎣ ⎦ density of the steel reinforcement and νc= Poisson’s ratio of the con-
Cl 2
Cl
+θI 3 Cl
Th
+ θI 4 ( ) ClTh
T
+ θI 5 T + σI εI
0 (20)
crete.

where xI = (t , w c, Cl, T , x ti) and ΘI = (θI 1, θI 2 , θI 3 , θI 4 , θI 5 , σI , Θti) . 4.2. Example structure


A comparison between the improved model and the existing CECS
model [20] is shown in Fig. 4. Fig. 4a shows that the prediction results The example structure used to conduct the corrosion-induced
of the CECS model tends to underestimate the corrosion rate; most of cracking fragility analysis is the typical simple-supported RC bridge in
the solid dots are under the 1:1 line. This is unacceptable in the life- Shanghai, China. The concrete grade of the RC beam is C30 and the
cycle design practice for the RC structures. The proposed model in this distance from the concrete surface to the middle of the steel re-
paper (Fig. 4b) clearly corrects this potential bias and properly reflects inforcement is 50 mm. The grade for the steel reinforcement is HRB335
the uncertainties in the corrosion propagation stage of the steel re- and the steel bar diameter is 18 mm. The design parameters x can be
inforcement. divided into random design parameters x r and deterministic design
parameters x d , i.e., x = (x r , x d) . The selection and statistical distribu-
3.5. Rust product expansion tions (shown in Table 7) of the random design parameters
x r = (fc 28 , dc , db0, d1) are based on the Unified Standard for Reliability
The corrosion product (rust) of the steel reinforcement will accu- Design of Highway Engineering Structures (GB/T 50283-1999) [49]
mulate around the corroded steel bar. The rust products have more and the existing literatures [38,48]. Other design parameters are all set
volume than the original steel, resulting a volumetric expansion effect to be deterministic and their values are listed in Table 8. The
and finally lead to the cracking of the concrete [47]. The accumulation
speed of the rust has an inverse relationship to the amount of the rust Table 5
product Wrust due to the protective effect of the growing rust to the ionic Data ranges collected for the corrosion rate
diffusion [48], i.e., Variable Symbol Range

dWrust kp Propagation time (year) tp 0–5


=
dt Wrust (21) Water to cement ratio w c 0.41–0.45
Cl− content (weight ratio to cement) (%) Cl 0.082–1.46
where kp= rust accumulation coefficient is a function of the corrosion Annual temperature (F) T 28.3–103.2
rate Annual relative humidity RH 0.76–0.98
Concrete cover depth (mm) dc 50.8–76.2
kp = 0.098πdb0 Icorr αrust (22)

7
B. Sun, et al. Engineering Structures 208 (2020) 110313

Table 7
Random design parameters
Name Symbol Distribution Mean COV

Concrete 28-day compression strength fc28 Normal 39.1 0.173


(Mpa)
Concrete cover depth (mm) dc Normal 41.7 0.050
Uncorroded steel bar diameter (mm) db0 Normal 18 0.050
Free pore band thickness (mm) d1 Normal 0.0125 0.050

Table 8
Deterministic design parameters
Name Symbol Value

Characteristic strength (MPa) fcuk 30


Water to cement ratio w c 0.4
Rust mass density (mg/mm3) ρrust 3.6
Steel mass density (mg/mm3) ρst 7.85
Concrete elastic modulus (MPa) Ec 30,000
Poisson’s ratio νc 0.2
Creep coefficient φcr 2.0
Fig. 3. Stepwise model selection process for Icorr (x, Θ) where a superposed
Annual temperature (°C) T 15.7
cross (×) indicates term to be removed. CO2 density (%) cCO2 0.0507
Rust type coefficient αrust 0.57
Table 6
Posterior statistics of model parameters for Icorr (x, Θ)
4.3. Fragility estimate and bound
Parameter Mean Standard Correlation coefficient
deviation
θ1 θ2 θ3 θ4 θ5 σ The consideration of epistemic uncertainties represented by Θ in the
improved models leads to a random nature in F (s, Θ) , i.e., the condi-
θ1 0.844 0.075 1.00 tional probability F (s, Θ) for the given boundary condition s is a
θ2 –0.515 0.024 –0.10 1.00 random quantity with respect to Θ instead of a deterministic value. The
θ3 1.329 0.068 –0.42 0.03 1.00 ∼
predictive fragility estimate F (s) [22,24] is used to represent the
θ4 –0.177 0.017 0.38 0.00 –0.98 1.00
1.121 0.052 –0.80 –0.20 –0.05 0.05 1.00
average cracking probability level by calculating the expected value of
θ5
σ 0.281 0.013 0.06 0.06 –0.03 0.03 –0.06 1.00 F (s, Θ) with respect to the posterior distribution of Θ

F (s) = ∫ F (s, Θ) f (Θ) dΘ (27)
environmental parameters are taken from the typical outdoor condition
in Shanghai, China and the parameters related to the corrosion-induced and the ± 1 SD confidence bounds (approximately corresponding to
cracking analysis are referred to the existing literatures 15% and 85% probability levels) are used to represent the uncertainty
[38,47,48,50,51]. It worth mention that since this paper mainly focus levels caused by the epistemic uncertainties
on carbonation-induced corrosion in the urban atmospheric condition, ∼ ∼
{Φ[−β (s) − σβ (s)], Φ[−β (s) + σβ (s)]} (28)
the chloride ion content in the concrete is ideally assumed to be zero
(Cl = 0%) and not further discussed in the following paper. ∼ ∼
where β (s) = Φ−1 [1 − F (s)]= reliability index, Φ[∙]= cumulative

Fig. 4. Comparison between (a) CECS model and (b) improved model based on measured and predicted corrosion rate.

8
B. Sun, et al. Engineering Structures 208 (2020) 110313

density function of the standard normal distribution and σβ (s) is the where α = −∇G (u∗) ∥∇G (u∗)∥= unit vector of ∇G (u∗) , ∇G (u∗)= gra-

standard deviation of β (s) . A detail calculation of σβ (s) from the first dient vector of the limit state function G (u) (in the standard normal
order reliability analysis process [52] can be referred to Gardoni et al. space) at the design point u∗, Ju, θz (z∗)= Jacobian matrix of the random
[24]. variables u (in the standard normal space) to their distribution para-
The fragility estimates and bounds of the corrosion-induced meters θz at z∗ and ∇θg g (z∗)= gradient vector of the limit state function
cracking for the example structure are obtained through the Finite g (z) (in the original design space) with respect to the deterministic
Element Reliability Using MATLAB (FERUM package) [53] based on the design parameters θg at z∗. Thus parameter sensitivity ∇θ F (z∗) can be
improved models and general process described above. Time effect and obtained by applying firs-order derivation on F (z∗) = Φ[−β (z∗)]
harmful substance in the environment are important generalized
“loading effects” for the durability analysis of the RC bridge. Thus the ∇θ F (z∗) = −φ (β ) ∇θ β (z∗) (31)
service time t and CO2 density cCO2 are selected for the boundary value
where φ (∙) is the standard normal probability density function.
s . Fig. 5 shows the fragility curves for the corrosion-induced cracking of
The sensitivity curves ∇Θ β (t ) of the corrosion-induced cracking for
the example RC bridge with respect to the service time t . The solid line
∼ the example bridge with respect to the model parameters Θ are shown
is the curve for the predictive estimates F (t ) and the two dashed lines
in Fig. 8. The fragility analysis results in Fig. 5 show that the cracking
are the 15% and 85% confidence bounds. Fig. 5a shows the fragility
probability under 10 year is close to zero. Thus the service period from
curves in the normal service life of 0 to 100 years for the RC bridge and
0 to 10 year is not included in the sensitivity analysis for computation
an expanded time scope up to 500 years is shown in Fig. 5b. The pre-
efficiency. Fig. 8 shows that during the service life, the sensitivity
dictive estimates and bounds with respect to the CO2 density cCO2 are
(absolute value discussed here and in the following) for θX 2 increases
shown in Fig. 6 at the service time t = 100 (year). Fig. 7 shows the
sharply, the sensitivities for θI2 and σI have a decreasing trend. For σX ,
contour plot of the fragility surface F (t , cCO2) with respect to both t and
the sensitivity increases in the early age and slowly recovers and de-
cCO2 . Every point in each contour (solid line) connects pair of values on t
creases in the service period.
and cCO2 that gives rise to a specific level of predictive fragility. Obvious
Fig. 9 shows the sensitivity curves ∇x β (t ) with respect to the design
interaction between the two “load inputs” is observed, especially at low
parameters x . The analysis results suggest that increases on the con-
boundary value levels.
crete strength fcuk , cover depth dc and steel bar diameter db0 , or decrease
on the CO2 density cCO2 , are efficient countermeasures to improve the
4.4. Parameter sensitivity durability performance of the RC bridge against corrosion-induced
cover cracking during the service period. For the evolution laws, the
Sensitivity analysis shows the changing trend of fragility with re- sensitivities for dc and db0 decrease slowly during life time, and the
spect to the parameters and derives the most sensitive parameters that changes for fcuk and cCO2 are relatively moderate.
are important for the design optimization and management decision in
the engineering practice [54]. Parameter sensitivity is defined as the
gradient vector ∇θ F (z∗) of the fragility F (s) with respect to the para- 4.5. Random variable importance
meters θ at the design point z∗. The parameters θ for the probabilistic
model formation in this paper include the deterministic design para- The uncertainty of the corrosion-induced cover cracking problem
meters θg = (x d, s) and the distribution parameters θz for the random comes from the accumulation and propagation of the uncertainties
variables z = (x r , Θ, ε ) , i.e., θ = (θg , θz) . The gradient vector of the re- presented in the random variables z = (x r , Θ, ε ) in the limit state
liability index ∇θ β (z∗) can be obtained through the first order reliability function g (x, Θ). However, the effects of different random variables
process [55] with respect to different types of the parameters may be different. Some of them may be important; they account for
most part of the uncertainties of the problem. Others may be less im-
∇θz β (z∗) = αT Ju, θz (z∗) (29)
portant; the uncertainties presented in them can be ignored to improve
the computation efficiency of the probability analysis. An importance
1
∇θg β (z∗) = − ∇θ g (z∗) measure λ is defined to evaluate this phenomenon [56]
∥∇G (u∗) ∥ g (30)

Fig. 5. Fragility curves for the corrosion-induced cracking of example bridge with respect to service time.

9
B. Sun, et al. Engineering Structures 208 (2020) 110313

Fig. 6. Fragility curves for the corrosion-induced cracking of example bridge with respect to CO2 density.

αT Ju∗, z∗SD′
λT =
∥αT Ju∗, z∗SD′∥ (32)

where Ju∗, z∗ is the Jacobian matrix of u∗ to z∗ (probability transforma-


tion from the original space to the standard normal space) and SD′ is
the standard deviation of the equivalent normal variables z′, in which
z′ = z∗ + Jz∗, u∗ (u − u∗)= linearized inverse transformation at z∗.
In the random variable importance analysis for the example bridge,
the random design parameters x r = (fc 28 , dc , db0, d1) are assumed to be
independent. For the random model parameters ΘX and ΘI , their cross-
correlation statistics are listed in Tables 3 and 6; εX and εI are in-
dependent standard normal random variables in the error terms. Fig. 10
shows the importance curves of the random variables z for the corro-
sion-induced cracking fragility in the service period of the example RC
bridge. The results indicates that the uncertainty of the cover cracking
problem mainly comes from the concrete carbonation model
X (xX , ΘX ) : model parameters θX1, θX 2 , θX 3 and error term σX εX . It worth
mention that except for θX 2 , the changing trends for those important
random model parameters are relatively moderate during the life time
of the RC bridge. The cover depth dc and the model error εI for the steel
corrosion rate model Icorr (xI , ΘI ) also account for part of the uncertainty
Fig. 7. Contour plot of the fragility surface F (t , cCO2) . sources. For the other random variables, their importance and con-
tributions on the uncertainty level of the engineering problem are

Fig. 8. Sensitivity curves for the model parameters Θ .

10
B. Sun, et al. Engineering Structures 208 (2020) 110313

Fig. 9. Sensitivity curves for the design parameters x .

Fig. 10. Importance curves for the random variables z .

11
B. Sun, et al. Engineering Structures 208 (2020) 110313

relatively small. against corrosion-induced cover cracking;


(3) The uncertainty of the cover cracking problem mainly comes from
5. Conclusion the concrete carbonation model. The cover depth and the model
error for the steel corrosion rate model also account for part of the
This paper proposes a comprehensive probabilistic approach for the uncertainty sources. For the other random variables, their im-
corrosion-induced cracking fragility analysis of the RC bridge in the portance and contributions on the uncertainty level of the en-
urban area with improved concrete carbonation and steel reinforce- gineering problem are relatively small.
ment corrosion models. Reviews and discussions are conducted on the
existing deterioration models. Improved deterioration models have a CRediT authorship contribution statement
probabilistic formation with deterministic part, correction term and
error term. The explanatory functions in the correction term come from Sun Bo: Conceptualization, Methodology, Software, Formal ana-
the influencing factors suggested by the existing models and the most lysis, Writing - original draft. Xiao Ru-cheng: Resources, Data curation,
sensitive ones (optimum combination) are selected by a stepwise model Writing - review & editing. Ruan Wei-dong: Investigation, Validation.
calibration and selection process. Comparisons and discussions are Wang Peng-bo: Visualization.
conducted between the improved deterioration models and the existing
models based on the prediction results and measured long-term data. Declaration of Competing Interest
Results show that the improved models are unbiased, more accurate
and properly account for all types of uncertainties, satisfying the The authors declare that they have no known competing financial
probabilistic requirements of the modern design. interests or personal relationships that could have appeared to influ-
Fragility curves, parameter sensitivities and random variable im- ence the work reported in this paper.
portance are achieved for the example RC bridge based on the proposed
approach and improved concrete carbonation and steel reinforcement Acknowledgements
corrosion models. The following findings in the analysis results are
noteworthy: The authors wish to thank all the members of the Long-span Bridge
Center at the Tongji University, for helpful discussions and suggestions.
(1) Fragility analysis results with respect to service time and CO2 The research work was supported in part by National Natural Science
density indicate significant interaction between the two “load in- Foundation of China (Grant 51908503 and 51909236), Natural Science
puts”, especially at low boundary value levels; Foundation of Zhejiang Province (Grant LQ19E090009) and Innovative
(2) Increases on the concrete strength, cover depth and steel bar dia- Experiment Project of ZJUT (Grant SYXM1723). Opinions and findings
meter, or decrease on the CO2 density, are efficient counter- presented are those of the authors and do not necessarily reflect the
measures to improve the durability performance of the RC bridge views of the sponsors.

Appendix A. Corrosion models in CECS (2007)

A.1. Concrete carbonation depth

The concrete carbonation depth X at service time t is calculated by the following equation:
X=k t (A.1)
where k is the carbonation coefficient considering a set of influencing factors:

58
k = 3K CO2 Kkl Kkt Kks KF T1 4RH1.5 (1 − RH ) ⎜⎛ − 0.76⎞⎟
⎝ fcuk ⎠ (A.2)
where

K CO2 – CO2 density factor, K CO2 = cCO2 0.03 ;


Kkl – location factor, Kkl = 1.4 for corner of the component and 1.0 for other area;
Kkt – curing factor, Kkt = 1.2 ;
Kks – stress factor, Kks = 1.0 for compression condition and 1.1 for tension condition;
KF – fly ash factor, KF = 1.0 + 13.34F 3.3 , F is the fly ash content (weight ratio).

A.2. Corrosion rate

The corrosion rate Icorr at service time t for a chloride penetration environment is calculated by the following equation:
ln Icorr = 8.617 + 0.618 ln C (dc , t ) − 3034 (T + 273) − 5 × 10−3R c + ln mcl (A.3)

d
C (dc , t ) = C0 + (Cs − C0 ) ⎡1 − erf ⎛ c ⎞ ⎤
⎜ ⎟

⎢ ⎝ Dt ⎠ ⎥
2 (A.4)
⎣ ⎦
R c = kR (1.8 − Ct ) + 10(RH − 1)2 + 4 (A.5)
where

C (dc , t ) – Cl− content at the surface of the steel reinforcement;


C0 – original Cl− content in the concrete;
Cs – Cl− content at the concrete surface;

12
B. Sun, et al. Engineering Structures 208 (2020) 110313

D – Cl− diffusion coefficient;


R c – ohmic resistance of the concrete;
kR – factor reflecting concrete quality, kR = 11.1 when w c = 0.3~0.4 or for C40 ~ C50 concrete, kR = 5.6 when w c = 0. 5~0.6 or for C20 ~ C30
concrete;
Ct – Cl− content in the concrete cover (mean value);
mcl – local environment factor, mcl = 4.0~4.5 for outdoor environment in the marine atmospheric zone and 2.0 ~ 2.5 for indoor environment.

Appendix B. Supplementary material

Supplementary data to this article can be found online at https://doi.org/10.1016/j.engstruct.2020.110313.

References Buildings. London: British Standards Institution; 2004.


[28] JTG D62-2004. Code for design of highway reinforced concrete and prestressed
concrete bridge and culverts. Beijing: China PR. Ministry of Communications; 2004
[1] Tuutti K. Corrosion of steel in concrete. Stockholm: Swedish Cement and Concrete [in Chinese].
Research Institute; 1982. Report nr 0346-6906. [29] Bao QW. Durability of reinforced concrete bridge in Beijing. Beijing Univ Technol
[2] Kwon SJ, Na UJ. Prediction of durability for RC columns with crack and joint under 2003. [in Chinese].
carbonation based on probabilistic approach. Int J Concr Struct Mater [30] Yuan CF, Niu DT, Gai QS, Sun CT. Durability testing and carbonation life prediction
2011;5(1):11–8. of Songhua river bridge. Bridge Const 2010;2:21–4. [in Chinese].
[3] Ekolu SO. Model for practical prediction of natural carbonation in reinforced con- [31] Wang CF, Niu DT. Life forecast for carbonation of reinforced concrete railway
crete: Part 1-formulation. Cem Concr Compos 2018;86:40–56. bridges. Sichuan Build Sci 2006;3(61–63):71. [in Chinese].
[4] Cui H, Tang W, Liu W, Dong Z, Xing F. Experimental study on effects of CO2 con- [32] Wang CF. Evaluation for durability and safety of railway concrete bridge in arid
centrations on concrete carbonation and diffusion mechanisms. Constr Build Mater region. Xi'an Univ Arch Technol 2011. [in Chinese].
2015;93:522–7. [33] Niu DT, Yuan CF, Wang CF, Xu SH. Carbonation service life prediction of reinforced
[5] Guiglia M, Taliano M. Comparison of carbonation depths measured on in-field ex- concrete railway bridge based on durability test. J Xi'an Univ Arch Tech (Natural
posed existing r.c. structures with predictions made using fib-Model Code 2010. Science Edition) 2011;02:160–5. [in Chinese].
Cem Concr Compos 2013;38(2):92–108. [34] Gan HL, Xie XL. Carbonation life prediction of service reinforced concrete bridge
[6] Stewart MG, Al-Harthy A. Pitting corrosion and structural reliability of corroding based on reliability theory of durability. Concrete 2013;03:48–51. [in Chinese].
RC structures: experimental data and probabilistic analysis. Reliab Eng Syst Saf [35] Liu MH, Wang YF. Modeling of carbonation process in reinforced concrete bridges.
2008;93(3):373–82. J Beijing Jiaotong Univ 2010;04:77–83. [in Chinese].
[7] Marsh PS, Frangopol DM. Reinforced concrete bridge deck reliability model in- [36] Box GEP, Tiao GC. Bayesian inference in statistical analysis. Reading, Mass:
corporating temporal and spatial variations of probabilistic corrosion rate sensor Addison-Wesley; 1992.
data. Reliab Eng Syst Saf 2008;93(3):394–409. [37] Laine M. Adaptive MCMC methods with applications in environmental and geo-
[8] Choe D, Gardoni P, Rosowsky D, Haukaas T. Probabilistic capacity models and physical models. Lappeenranta, Finland: Lappeenranta University of Technology;
seismic fragility estimates for RC columns subject to corrosion. Reliab Eng Syst Saf 2008.
2008;93(3):383–93. [38] Guo Y, Trejo D, Yim S. New model for estimating the time-variant seismic perfor-
[9] Na UJ, Kwon S, Chaudhuri SR, Shinozuka M. Stochastic model for service life mance of corroding RC bridge columns. J Struct Eng 2014;141(6):04014158.
prediction of RC structures exposed to carbonation using random field simulation. [39] Stewart MG. Mechanical behaviour of pitting corrosion of flexural and shear re-
KSCE J Civ Eng 2012;16(1):133–43. inforcement and its effect on structural reliability of corroding RC beams. Struct Saf
[10] Teplý B, Chroma M, Rovnanik P. Durability assessment of concrete structures: re- 2009;31(1):19–30.
inforcement depassivation due to carbonation. Struct Infrastruct Eng [40] Andrade C, Alonso C. Test methods for on-site corrosion rate measurement of steel
2010;6(3):317–27. reinforcement in concrete by means of the polarization resistance method. Mater
[11] Teplý B. Interrelation between service life, reliability index, and costs of concrete Struct 2004;37(9):623–43.
structures subjected to aggressive exposure. J Perform Constr Facilit 2013;28(4). [41] Ahmad S, Bhattacharjee B. Empirical modeling of indicators of chloride-induced
[12] Niu DT. Durability and life forecast of reinforced concrete structure. Beijing: rebar corrosion. J Struct Eng 2000;27(3):195–207.
Science Press; 2003. [in Chinese]. [42] Zhou BB, Gu XL, Zhang WP, Jin XY, Huang QH. Time-dependent reliability analysis
[13] Marques PF, Costa A. Service life of RC structures: Carbonation induced corrosion. of reinforced concrete beams subjected to bending and marine atmospheric en-
Prescriptive vs. performance-based methodologies. Constr Build Mater vironment. China Civ Eng J 2010;S2:15–21. [in Chinese].
2010;24(3):258–65. [43] Li QC. Reliability based service life prediction of corrosion affected concrete
[14] Sun B, Xiao RC, Cheng J. Fragility of the reinforced concrete bridge subjected to structures. J Struct Eng 2004;130(10):1570–7.
carbonation induced corrosion. IABMAS-2016 on maintenance, monitoring, safety, [44] Vu KAT, Stewart MG. Structural reliability of concrete bridges including improved
risk and resilience of bridges and bridge networks. CRC Press; 2016. chloride-induced corrosion models. Struct Saf 2000;22(4):313–33.
[15] Papadakis VG, Vayenas CG, Fardis MN. Fundamental modeling and experimental [45] Liu T, Weyers RW. Modeling the dynamic corrosion process in chloride con-
investigation of concrete carbonation. ACI Mater J 1991;88(4):363–73. taminated concrete structures. Cem Concr Res 1998;28(3):365–79.
[16] FIB Bulletin 34. Model code for service life design. Lausanne; 2006. [46] Yalcyn H, Ergun M. The prediction of corrosion rates of reinforcing steels in con-
[17] CEB Bulletin 238. New approach to durability design. An example for carbonation crete. Cem Concr Res 1996;26(10):1593–9.
induced corrosion. Lausanne; 1997. [47] Zhong JQ, Gardoni P, Rosowsky D. Stiffness degradation and time to cracking of
[18] Kim JH, Oh KC, Park SB. A study on carbonation velocity for concrete structures. cover concrete in reinforced concrete structures subject to corrosion. J Eng Mech
Korea Inst Struct Maintenance Inspection 2008;12(2):163–70. 2009;136(2):209–19.
[19] Zhu AM. Concrete carbonation and durability of reinforced concrete. Concrete [48] Liu Y. Modeling the time-to-corrosion cracking of the cover concrete in chloride
1992;06:18–22. [in Chinese]. contaminated reinforced concrete structures. Blacksburg: Virginia Polytechnic
[20] CECS220: 2007. Standard for durability assessment of concrete structures. Beijing: Institute and State University; 1996.
China Architecture and Building Press; 2007 [in Chinese]. [49] GB/T 50283-1999. Unified standard for reliability design of highway engineering
[21] Smolczyk HG. Testing of concrete. Proceedings of RILEM Symposium. Washington, structures. Beijing: Ministry of Transport of the P. R. of China; 1999 [in Chinese].
DC; 1962. p. 489. [50] Pantazopoulou SJ, Papoulia KD. Modeling cover-cracking due to reinforcement
[22] Sun B, Gardoni P, Xiao RC. Probabilistic aerostability capacity models and fragility corrosion in RC structures. J Eng Mech 2001;127(4):342–51.
estimates for cable-stayed bridge decks based on wind tunnel test data. Eng Struct [51] Mullard JA, Stewart MG. Corrosion-induced cover cracking: new test data and
2016;126:106–20. predictive models. ACI Struct J 2011;108(1).
[23] Sun B, Gardoni P. Directional search algorithm for hierarchical model development [52] Ditlevsen O, Madsen HO. Structural reliability methods. New York: Wiley; 1996.
and selection. Reliab Eng Syst Saf 2019;182:194–207. [53] Haukaas T, Hahnel A, Sudret B, Song J, Franchin P. Department of civil and en-
[24] Gardoni P, Der Kiureghian A, Mosalam KM. Probabilistic capacity models and vironmental engineering. Berkeley, CA: University of California; 2003.
fragility estimates for reinforced concrete columns based on experimental ob- [54] Choe D, Gardoni P, Rosowsky D. Closed-form fragility estimates, parameter sensi-
servations. J Eng Mech 2002;128(10):1024–38. tivity, and Bayesian updating for RC columns. J Eng Mech 2007;133(7):833–43.
[25] Gardoni P, Mosalam KM, der Kiureghian A. Probabilistic seismic demand models [55] Hohenbichler M, Rackwitz R. First-order concepts in system reliability. Struct Saf
and fragility estimates for RC bridges. J Earthquake Eng 2003;7(spec01):79–106. 1983;1(3):177–88.
[26] Jin WL, Yan F, Zhang L. A predeterminate model of steel bar corrosion ratio con- [56] Der Kiureghian A, Ke JB. Finite-element based reliability analysis of frame struc-
sidered of concrete carbonation model. J Zhejiang Univ (Eng Sci) tures. Proceedings of ICOSSAR ‘85 4th international conference on structural safety
2000;34(2):158–63. [in Chinese]. and reliability. Japan; 1995. p. 395–404.
[27] Eurocode 2. Design of Concrete Structure, Part 1: General Rules and Rules for

13

You might also like