Journal of Structural Engineering No.

31-8
Vol. 31, No.1, April–June 2004 pp. 1–15

Risk-based remaining life assessment of corrosion affected reinforced

concrete structural members

K. Balaji Rao∗ , M. B. Anoop∗ , N. Lakshmanan∗∗ , S. Gopalakrishnan$ and T. V. S. R. Appa Rao#

Remaining life assessment of corrosion affected reinforced concrete structural members is a topic of current R&D
worldwide. There is an urgent need to develop more scientific and rational methodologies for this purpose. Number
of investigations have already been carried out to assess the damage/distress due to corrosion, that would help
in remaining life assessment. While most of these studies are deterministic in nature, efforts are also being made
to use probabilistic methods for damage assessment. However, an important aspect in remaining life estimation
is the interpretation of the data related to damage/distress and making expert judgement about damage/distress
level. Due consideration needs to be given to the quality of the data and the expert interpreting the data. In
this paper, expert judgement regarding corrosion damage level is integrated with the structural risk, expressed
in terms of probability of attaining a particular damage level, for the remaining life assessment of corrosion
affected reinforced concrete structural members. In the proposed methodology, the thinking process of the expert,
in corrosion damage assessment, is modelled within a probabilistic framework using Brunswikian theory. The
performance of each expert is determined by computing the achievement index. The damage assessment procedure
is integrated with Markov Chain model for risk-based remaining life assessment. To illustrate the usefulness of
the proposed methodology in determining more rationally the remaining life, an example problem of remaining
life assessment of a reinforced concrete bridge girder is considered.

The corrosion of reinforcement in concrete is an issue of regarding inspection/maintenance activities for the
major concern, as it affects the safety (due to the reduc- structure.
tion in area of reinforcement) and serviceability (due to A reliable method for service life estimation of the
the formation of rust stains, cracking and spalling) of structure is a pre-requisite for remaining life assess-
the structure. The premature deterioration of reinforced ment. Systematic approaches/methodologies for ser-
concrete (rc) structures has necessitated the need for vice life prediction have been proposed by various
continual structural health monitoring (SHM) to deter- researchers/codes of practice3–5 . Expert judgement is iden-
mine the existence, location and extent (or degree) of tified as an important part of evaluation in most of these
corrosion damage, if any, on the structure. SHM is the methodologies. For instance, the methodology of RILEM
process of establishing some knowledge of the current TC 31-PCM emphasizes expert judgement as an essen-
condition of the structure or its components1 , which is tial part of evaluation, and points out that ‘the complex-
required for the performance evaluation of the structure. ity of evaluating the interaction of materials or systems
While SHM is a part of the value chain as proposed by with their environment requires expertise that cannot be
Wong and Yao2 , the information obtained from SHM fully replaced by existing test methods’3 . It also identi-
needs to be processed further for quantifying the risk, fies the use of methods of social-, health-, psychological-,
which is an important link in the value chain. As pointed natural- and technical sciences for the expert judge-
out by Wong and Yao2 , there exists a gap between ment.
where SHM currently stops and where financial deci- CIBW80/RILEM 71-PSL3 pointed out that there is a
sion (remaining useful life) begins, and this gap needs requirement for an effective mechanism for obtaining and
to be filled. This requires the rational assessment of the reporting data on in-service performance of structures.
current damage state based on the data from SHM, and Also, it may be difficult to handle the large amount of data
the remaining life estimation using structural risk as the acquired from continuous SHM manually. A suitably cre-
norm. A rational estimation of the current condition and ated database of KBS would help in handling this amount
remaining life will help in making engineering decisions of data through proper acquisition, representation and

∗
Scientist, ∗∗ Director, $ Advisor [Management] and Director Grade Scientist, # Emeritus Scientist [CSIR] and Formerly Director, Structural Engineering
Research Centre, CSIR Campus, Taramani, Chennai - 600 113

JOURNAL OF STRUCTURAL ENGINEERING VOL. 31, NO.1, APRIL–JUNE 2004 1

retrieval. Different knowledge-based systems (KBS) have CORROSION DAMAGE ASSESSMENT
been developed by various agencies involved in the inspec-
tion / maintenance of infrastructural facilities (for e.g., Chloride-induced corrosion of the reinforcement is con-
Bridge Management Systems such as PONTIS in United sidered as one of the major mechanisms of degradation
States, COSMOS in UK and HiSMIS in Europe). KBS of resistance for rc structures located in marine and other
have been developed for damage assessment of the struc- aggressive environments. In addition to the reduction in
tures subjected to natural hazards such as earthquakes, cross-sectional area of steel (resulting in the reduction of
cyclones (IRAS, WINDRITE, HAZUS, SPERIL). Some of resistance), corrosion of reinforcement leads to the crack-
the KBS developed also help in assessing the rate of dete- ing of cover concrete due to the tensile stresses induced in
rioration, which will be useful for remaining life estima- the cover concrete by the formation of corrosion products.
tion. For instance, many of the BMS, including PONTIS, The service life of a rc structural member with respect to
uses a Markov deterioration model to determine the proba- corrosion of reinforcement can be divided into two stages:
bility of change between consecutive condition states, and i) corrosion initiation, and ii) corrosion propagation.
thus to assess the degree or rate of deterioration. But the
main aim of KBS is to assist the engineer by manipulat- Time for corrosion initiation
ing large amounts of data and to produce reports to aid the
engineer in decision making, the final decisions are left to The reinforcement in concrete is normally protected against
the engineer6 . Hence, the recommendations given by the corrosion by a microscopically thin oxide layer formed
BMS, such as PONTIS, has to be integrated with engi- on the surface of the reinforcement due to the high alka-
neering judgement for making rational decisions. Thus, linity of the surrounding concrete10 . Chloride ions diffuse
human judgement plays an important role in the dam- from the surface of the concrete through the cover concrete.
age assessment and decision making and in completing When the chloride concentration around the reinforcement
the value chain as indicated by Wong and Yao2 . Human exceeds a critical value (critical chloride concentration),
judgmental research is an upcoming field in psychologi- the protective oxide layer dissolves and corrosion initiates.
cal sciences, and has found applications in different areas The diffusion of chlorides through the cover concrete is
such as settlement of disputes and forensic engineering. generally modelled using Fick’s second law of diffusion.
A promising theory which unifies the preferential choice The time for corrosion initiation can be determined from
and judgement is Brunswikian theory, the application of Fick’s second law of diffusion as
which is researched upon very recently7,8 . For instance,   −2
c2 −1 cs − ccr
using the concepts of Brunswikian theory, Gigerenzer et ti = erf (1)
al7 proposed probabilistic mental models (PMM) for mod- 4D cs
elling the human mental process in making decisions. To where
the authors’ knowledge, none of the existing KBS takes c clear cover to reinforcement
into consideration the uncertainties associated with the D diffusion coefficient for chlorides in concrete
human mental process in decision making. There is a need cs surface chloride concentration
to bridge this gap by using suitable models of the human ccr critical chloride concentration
mental process in the damage assessment and decision
making. Corrosion propagation
The authors are involved in carrying out some of
the investigations related to the CSIR-IISc collaborative Due to corrosion, cross-sectional area of steel reduces.
project ‘Structural damage detection using vibration data The remaining diameter of the reinforcing bar at any time
and probabilistic health assessment’, from 2001. Towards t, 8(t), can be obtained as
this, a methodology has been developed at SERC for
8(t) = 8(0) − rcorr (t − ti ) (2)
health assessment of bridges using Markov Chain (MC)
modelling9 . The proposed methodology integrates the ear- where
lier developed MC model with the Brunswikian theory 8(0) initial diameter of the reinforcing bar, in mm
for damage assessment and risk estimation of corrosion ti time required for corrosion initiation, years
affected rc structural members which will help in the
Researchers have proposed different models for deter-
remaining life assessment.
mining the rate of corrosion of reinforcing bar. From a
The article is organised as follows. The proposed proce-
brief review of these models11 , it is found that the model
dure for corrosion damage assessment using Brunswikian
proposed by Andrade et al (given in Rodriguez et al12 ) is
theory is presented in the next section, followed by the
widely accepted. Using this model, the rate of corrosion
Markov Chain model for remaining life assessment of rc
can be determined as,
structural members subjected to chloride-induced corro-
sion of reinforcement by including the results of inspection rcorr = 0.0115∗ Icorr ∗ α (3)
and the judgement of experts regarding the corrosion dam-
age state. An example problem is provided to illustrate the where
proposed methodology for corrosion damage assessment rcorr rate of corrosion, in mm/year
and remaining life estimation. The concluding remarks are Icorr average value of corrosion current density, in
given in the last section. µA/cm2

2 JOURNAL OF STRUCTURAL ENGINEERING VOL. 31, NO.1, APRIL–JUNE 2004

 α factor for including the effect of highly • Visual inspection
localised pitting normally associated with • Concrete cover depth survey
chloride-induced corrosion (varies from 4 • Delamination survey
to 8) • Half-cell potential survey
0.0115 factor which converts µA/cm2 to mm/year • Corrosion current measurement

Modelling of corrosion damage Visual inspection

Damage is defined as the physical disruption or change in A visual inspection of rust stains, the severity, type and
the condition of a structure or its components brought about location of cracks and concrete spalls gives a first indication
by external actions and influences, such that some aspect of the extent of corrosion damage in the structure. This is
of either the current or future functionality of the structure the simplest, least expensive and the most common method
or its components are impaired13 . The loss in the area of in conducting a concrete structure survey, and provides pri-
steel due to corrosion reduces the load carrying capacity mary information on the integrity of the structure and cor-
of the rc member. This degradation can be modelled by rosion state of the reinforcement. But the interpretation of
calculating the ‘capacity ratio’, ν(t), of the member at time the results depends upon the experience of the inspector.
t as (Andrade et al14 ), Some of the guidelines available in literature for classify-
R(t) ing the defects based on visual examination are given in
ν(t) = (4) Table 1.
S
where R(t) is the load carrying capacity of the member TABLE 1
at any time t, and S is the required capacity for the struc-
tural member according to relevant design standards. ν(t) GUIDELINES FOR CLASSIFYING DEFECTS
is considered as the measure of corrosion damage to the BASED ON VISUAL EXAMINATION16
structural member at time ‘t’. Rating Appearance
Damage assessment
Cracks (diagonal, longitudinal and transverse)
Damage assessment is the process of collecting and evalu-
ating the information about the current condition of a struc- 1 (very slight) < 1 mm in width
ture or its components. A comprehensive assessment of
damage also helps in predicting the future performance of 2 (slight) 1–10 mm in width
the structure. The need for damage assessment of a struc-
ture may arise from any of the following13 : 3 (moderate) 10–20 mm in width
- The structure has deteriorated through the effects of 4 (severe) 20–25 mm in width
external environment and loads
- The structure has suffered damage from external envi- 5 (very severe) > 25 mm in width
ronmental aspects (such as seismic attacks, floods)
- The structure has suffered damage from internal dete- Spalling
riorating effects
- The structure is being used or is planned to be used 1 (very slight) Barely noticeable
for purposes significantly different from the original,
such as application of greater live loads 2 (slight) Clearly noticeable
- The structure is under consideration for redesign or
structural alterations 3 (moderate) Holes larger than popout of coarse
- The structure has been exposed to a number of the aggregate
above detrimental effects and is under consideration
for major changes in purpose and/or structural alter- 4 (severe) Holes 150 mm (6 inch) in dia. and at
ations. least 150 mm (6 inch) deep
The assessment of damage to the structure includes the
evaluation of the cause(s) of damage, degree and amount 5 (very severe) Holes larger than 150 mm
of damage, expected progress of damage with time, and
the effect of damage on structural behaviour and service-
ability. The damage assessment of rc structural compo- Concrete cover depth survey
nents/structures subjected to chloride-induced corrosion of
reinforcement has received considerable attention in recent Covermeter, profometer or pachometer are used to deter-
years. Some of the common methods and procedures that mine the concrete cover depth. These devices work on the
are used for the assessment of corrosion damage in existing principle of applying a magnetic field into the concrete
rc structures are15 : surface and recording a change in response due to the

JOURNAL OF STRUCTURAL ENGINEERING VOL. 31, NO.1, APRIL–JUNE 2004 3

 TABLE 2
FEATURES OF THE MOST WIDELY-USED METHODS FOR CORROSION DAMAGE STATE ASSESSMENT
OF REINFORCED CONCRETE STRUCTURES17

Methods Electro- Electrochemical

Charac- Gravimetric Potential LPM Guard Coulostatic chemical Impedance
teristics test mapping (Rp ) ring method Noise Spectroscopy Harmonics

Speed for
individual • ◦ ◦ ◦ ◦ ⊗ ⊗ ⊗
measurements

Speed of
response to • ◦ ◦ ◦ ◦ ◦ ◦ ◦
changes

Quantitative
◦ • ◦ ◦ ◦ ⊗ ⊗ ◦
information

Non-destructive • ◦ ◦ ◦ ◦ ◦ ◦ ◦

Non-disturbing • ◦ ⊗ • ⊗ ◦ ⊗ •

Measurement Icorr Ecorr Icorr Icorr Icorr Icorr ? Icorr Icorr

parameter average mechanisms
Note : Meaning of the symbols:
◦ The method possesses the listed characteristics in an optimal degree, e.g., the individual measurements is
instantaneous.
⊗ The method possesses the listed characteristics in a less than fully-satisfactory degree, e.g., the individual measure-
ment is relatively slow.
• The method does not possess at all the listed characteristics, e.g., the individual measurement is very slow.

rainforcement. Radar scanning method is also being used most frequently used in the field because of its simplic-
for the determination of location of reinforcement and the ity. In this method, an indication of the relative probability
depth of concrete cover. of corrosion activity was obtained empirically through the
measurement of the potential difference between a stan-
Delamination survey dard portable half-cell placed on the surface of concrete
and the reinforcement below. The results of the potential
Delamination is the separation of a portion of concrete survey can be evaluated on the basis of ASTM guidelines
along a plane parallel to the outer surface of the concrete. or the JSCE guidelines (Table 3).
It is normally located at the level of reinforcement, and is
due to the corrosion of reinforcement which will lead ulti- Corrosion current measurement
mately to the spalling of the concrete cover. Chain drag
and hammer sounding methods are commonly used for The commonly used electrochemical technique for field
detecting and mapping delamination. While the procedure assessment of corrosion currents in concrete is the linear
is simple and offers information relating to the corrosion of polarisation. The 3LP and the Gecor are two devices which
reinforcement prior to the occurrence of concrete cracking employ linear polarisation technique to measure the cor-
and spalling, the accuracy of the results depends upon the rosion current, Icorr . The main difference between these
experience of the person performing the survey. devices is that the Gecor device has a guard ring electrode
which is used to confine the influence area of the counter
Half-cell potential survey electrode by actively confining the polarisation current dur-
ing the measurement process. The general guidelines for
Since corrosion of reinforcement is an electrochemical interpreting the results of 3LP and Gecor, as supplied by
process, the main techniques used for the inspection of the instrument manufacturers, are given in Table 4. The
corrosion damage are electrochemical in nature. Some of Icorr value can be transformed into the rate of corrosion for
the commonly used electrochemical methods for corrosion determining the loss in reinforcement area using Eq. 3.
state assessment are given in Table 2. The corrosion poten- Corrosion durability of a rc structural member depends
tial (Ecorr ) measurement usin a half-cell is the method on the environment in which the member is located. Thus,

4 JOURNAL OF STRUCTURAL ENGINEERING VOL. 31, NO.1, APRIL–JUNE 2004

 TABLE 3
EVALUATION OF HALF-CELL POTENTIAL DATA18

Half-cell potential reading Corrosion

(Ecorr ) level Remarks

−250 mV ≤ Ecorr I Black surface with no corrosion

−350 mV ≤ Ecorr ≤ −250 mV II Only spots of rust on rebar surface

−450 mV ≤ Ecorr ≤ −350 mV III Thin rust layer with corrosion products adhering to concrete

IV Formation of expansive corrosion products, with little loss in

cross-section
Ecorr ≤ −450 mV
V Formation of expansive corrosion products, with loss in
cross-section

TABLE 4
GUIDELINES FOR INTERPRETING RESULTS OF 3LP AND GECOR15

3LP device Gecor device

Icorr (µ A/cm2 ) Corrosion Damage Icorr (µ A/cm2 ) Corrosion State

< 0.2 No damage expected < 0.1 Passive

0.2–1.0 Damage possible within 10 to 15 years 0.1–0.5 Low corrosion

1.0–10.0 Damage possible within 2 to 10 years 0.5–1.0 Moderate

> 10.0 Damage is expected within 2 years > 1.0 High corrosion

it is important to characterise the environment in which the condition. Using the information regarding the envi-
member is located while carrying out the corrosion dam- ronmental aggressiveness together with the information
age assessment. In most of the codes and standards, the obtained from visual inspection and field measurements,
exposure conditions are classified in a general and quali- the corrosion damage assessment of the structural member
tative manner, which leads to ambiguities in the selection can be carried out.
of exposure condition for a structure/structural member. The information from the field surveys are passed on
The importance of specifying the exposure condition by to an expert or a group of experts for making judgement
a parameter, in the context of prediction of service life of regarding the corrosion state of the structure. It has to be
reinforced concrete members, is brought out by Masters noted that the correctness of the data from visual inspection
and Brandt3 . Anoop et al19 proposed a methodology for depends on the evaluation ability of the person carrying
quantification of environmental aggressiveness taking into out the inspection. Hence, this data has to be corrected for
consideration the uncertainties associated with the speci- the expertise (or lack of it) of the observer before pass-
fication of different environmental parameters (which are ing it to expert(s). Also, there can be uncertainties in data
in general linguistic, viz., high humidity, medium temper- measured using instruments in the form of random mea-
ature). In this method, the environment is characterised surement error, systematic errors from imperfectly cali-
by an environmental aggressiveness factor (EAF), on the brated instruments, and recording and other transmission
basis of environmental variables, namely, temperature, rel- errors20 . The measurement data should first be filtered and
ative humidity and degree of wetting and drying which are processed to account for these errors before passing it on to
obtained from the field. To account for the uncertainties in the expert(s). Thus, there is a need to validate the observer
the values of these variables, they are represented by fuzzy and the measurement equipments.
sets. The environmental aggressiveness factor, represented Once the data has been passed on to the expert, he has to
by fuzzy sets, is defined on an arbitrary scale of 0 to 6. The make a judgement regarding the corrosion state of the struc-
defuzzification of the output fuzzy set for EAF will give tural component/structure based on this data. The expert
the crisp value of EAF quantifying the given exposure judgement is an essential part of evaluation, and there will

JOURNAL OF STRUCTURAL ENGINEERING VOL. 31, NO.1, APRIL–JUNE 2004 5

be uncertainties associated with the human mental pro- conceptualise the general research process in psychology23 .
cess in making judgement. A comprehensive framework These concepts helps to understand the level of calibration
for damage assessment should take into account the uncer- of the expert to the environment. Using these concepts, the
tainties associated with human mental process. The human correlation between the judgement and the actual environ-
mental process can best be desribed in a probabilistic basis ment (represented as the achievement of the expert, ra ) can
and the Brunswikian theory provides a rational framework be determined as follows.
for taking into account the uncertainties associated with
human mental process in judgement and decision making.
A brief overview of the relevant Brunswikian and related
concepts pertinent to the present study are given below.

Brunswikian Theory

Brunswik21 pointed out that one’s knowledge of a dis-

tal ‘initial focal variable’ is mediated by more proximal
‘cues’ (or information) that one has about it. According to
Brunswik22 , individuals are generally and in many situa-
tions competent, but within the framework of such an opti-
mistic estimation of the individual’s scope, the levels of
capability (the sharpness of their ‘lenses’) differ drastically
among various people. The central and typical feature of
the organisms coming into grips with his/her environment
can be seen as a ’lens’. Brunswik preferred this concise FIG. 1. SCHEMATIC REPRESENTATION OF LENS MODEL
term to express the multifaceted human competence of an
individual to reach focal goals in judgements or actions22 .
The lens model proposed by Brunswik21 conceptually rep- A simple lens model is shown in Fig. 1. In this figure, Ye
resented the situation wherein one individual has to make denotes the distal variable (criterion), Xi ’s, the cues and Ys ,
a judgement about the true state of the distal variable using the judgement made by the expert. Data (Xi s) have been
multiple pieces of information. collected for a number of base line cases with known crite-
rion values, Ye s and the expert’s judgement (Ys s) have been
Bruswik lens model obtained for these cases. Now, linear regression equations
can be formulated on the judgmental and environment side
According to Brunswik lens model, the environmental as
structure and the organismic system are joined with a
X
k
linking function that incorporates the subject’s beliefs or Ys0 = a + bi Xi (5)
assumptions about the causal texture of the environment23 . i=1
The environment is defined by its inherent objects, and at
the same time, by the subjective transformations or inter- and
pretations of these objects. In the necessary process of X
k
perception, Brunswikian aspects of distality and proximal- Ye0 = c + di Xi (6)
ity are components formed of objects. Objects are facts i=1
which may be relevant in their autonomous qualities for respectively, using multivariate linear regression.
psychological analysis. Therefore, the processes within the Let Rs = γYs ,Ys0 and Re = γYe ,Ye0 be the values of coeffi-
organism are undoubtedly subjective, but a person’s ori- cient of determination on the judgement side and the envi-
entation within such processes is partially directed to or ronment side, respectively. Rs is the correlation between
determined by objects mostly independent of subjective the actual judgement and the predictive judgement, which
decisions. is a measure of the cognitive consistency of the expert. Re
The paradigm of the ‘lens’ can be regarded as a gener- is the correlation between the actual criterion value and the
alised process characteristic for the construction of human predicted criterion value, which is an index of how pre-
judgements and action – individually and in social contexts. dictable the environment is. The correlation between the
On the other hand, the lens shows openness, flexibility and predicted values Ys0 and Ye0 is given by the matching index,
variety, facilitates communication like a network and solves G, i.e., G = γYs0 ,Ye0 . The value of G denotes how well the
problems of unexpected matters22 . Its combination of goal model of expert corresponds to the model of the environ-
direction and virtuosity in perception and action leads to a ment. The achievement index, ra , which is the correlation
concentration of essentials. between the judgement and the criterion is given by
q q
Generalised linear model ra = GRs Re + C 1 − Rs2 1 − Re2 (7)
Brunswik methodology of regression together with the gen- where C is the configurality index, defined as the correla-
eralised linear model constitute a very useful framework to tion between the residuals of the two regression equations,

6 JOURNAL OF STRUCTURAL ENGINEERING VOL. 31, NO.1, APRIL–JUNE 2004

i.e., C = γ[Ys −Ys0 ],[Ye −Ye0 ] . The configurality index accounts corrosion damage state. The information recorded by the
for nonlinearities in the environment (criterion) side and observer/instrument (cues) are corrected for the evaluation
the judgement side. ability/human error (in the case of human observer) and
The achievement index, ra , can be regarded as a measure for the detection capability and correctness of detection (in
of the accuracy of the judgements made by the expert. the case of instrument). The corrected data is the proximal
stimuli for the expert who makes a judgement regarding
Confidence limits the distal stimuli, namely, the corrosion state of the rein-
forcement.
Based on the assumption that people are good judges of the The following cues are used in this study for the corro-
reliability of their knowledge, Gigerenzer et al7 proposed sion damage assessment
probabilistic mental models (PMM) for cognitive processes
(i) Rust stains (visual inspection)
in judgement. Two important aspects of PMM are that prob-
(ii) Cracking (visual inspection)
abilistic inference is part of the cognitive process and that
(iii) Spalling (visual inspection)
uncertainty is part of the outcome. The PMM models are
(iv) Delamination (chain drag or hammer sounding)
consistent with human thinking process in the way that the
(v) Ecorr (Half-cell potential mapping)
expert would assign a confidence level for his judgement.
(vi) Icorr (linear polarisation)
This treatment would enable to characterise the thinking
(vii) Remaining area of reinforcement (radiography)
process with respect to various confidence levels. The over-
(viii) Cover thickness (cover meter)
or under-confidence limits associated with an expert for
(ix) Temperature
the different confidence levels are determined based on the
(x) Relative humidity
judgements made on a number of baseline cases. The over-
(xi) Degree of wetting and drying
or under-confidence limit for a given confidence level, ‘i’,
(The cues ix–xi are used for determining the EAF
for a given expert ‘j ’, can be determined as
quantifying the environment19 ).
ni
over - or under - confidence, ouci = (pi − fi ) (8) In this study, five corrosion damage states have been
n identified for rc structural members subjected to corro-
where sion of reinforcement based on the guidelines available in
n - total number of decisions made by the expert literature14 . The definition of these damage states together
‘j ’ with the associated range of capacity ratio for rc struc-
pi - confidence judgement (mean value of confi- tural members in these states are given in Table 5. These
dence level) ranges of capacity ratio are based on the guidelines given
ni - number of times the confidence judgement pi by CEB25 .
was used by the expert ‘j ’ By integrating the information required for corrosion
fi - relative frequency of correct answers for all damage assessment (Tables 1-5), and supplying the same
decisions for which confidence pi was assigned along with cues, the aim would be to rationally capture the
by the expert ‘j ’ thinking process of an expert in arriving at the judgement
regarding the damage state. It is known that the mental
The over- or under-confidence takes into account the
process can best be described in the probabilistic basis. A
relative bias of the expert through the term ni .
24 number of experts are asked to make judgement regarding
Brehmer and Hagafors expanded the Brunwikian lens
the corrosion damage state independently using the same
model to a multilevel lens model to study the use of set of cues. The expert is asked to identify the corrosion
experts in complex judgement making. Such a multilevel damage state(s) in which he believes the member is in,
lens model is used in the present study to model the cor- and to attach confidence level(s) for his judgement from
rosion damage state assessment of rc structural members. a confidence scale. The confidence scale consists of seven
The procedure for corrosion damage assessment using categories I to VII representing 0%, 1%–20%, 21%–40%,
Brunswikian theory is given below. 41%–60%, 61%–80%, 81%–99% and 100% confidence,
respectively. The corresponding confidence judgement pi
Corrosion Damage Assessment Using Brunswikian (mean value of the confidence level) for different confi-
Theory dence levels are 0%, 10%, 30%, 50%, 70%, 90% and 100%,
respectively. Consistent with probabilistic mental thinking,
The distal stimuli of the multilevel lens model, used in cor- the experts would judge the probable damage states of cor-
rosion damage assessment of rc structural members, is the rosion affected rc structural members, along with respective
corrosion of reinforcement, which gives rise to the proximal confidence levels. Once this information is obtained from
stimuli to the observer/instrument in the form of appear- different experts for a given case, it remains how to finally
ance and corrosion current/potential. The information on judge the damage state. The final decision should take into
proximal stimuli (such as rust stains, amount of crack- account in some form the damage information provided by
ing and spalling, corrosion current density) are recorded different experts. Instead of classifying judges as experts or
by the observer/instrument (cues). These cues, together non-experts, it is better to consider them rational to differ-
with corrosion state of reinforcement are the distal stim- ent degrees26 . This requires that the achievement of experts
uli for the expert, who is making a decision regarding the to be quantified.

JOURNAL OF STRUCTURAL ENGINEERING VOL. 31, NO.1, APRIL–JUNE 2004 7

 TABLE 5
CORROSION DAMAGE STATES FOR THE REINFORCED CONCRETE STRUCTURAL MEMBERS

Corrosion
Damage State Description Capacity ratio

No No visible signs of corrosion 1.0

Ecorr > −250 mV

Slight Rust stains barely noticeable; some cracking; no spalling; 0.95

−350 mV ≤ Ecorr ≤ −250 mV

Rust stains; several longitudinal cracks; some cracks

in stirrup direction; clearly noticeable spalling in some
Medium places; 0.80–0.95
−450 mV ≤ Ecorr ≤ −350 mV
5% loss in area of reinforcement

Highly noticeable rust stains; extensive cracking and

spalling;
Severe Ecorr ≤ −450 mV 0.60–0.80
10% loss in area of reinforcement

Highly noticeable rust stains; extensive cracking and

spalling; steel is no more in contact with concrete at some
Very Severe places; 0.35–0.60
Ecorr ≤ −450 mV
25% loss in area of reinforcement

The achievement (ra ) of each expert is determined using X

n
the generalised linear model based on a number of base- {Pc } = wj {P }j (11)
line cases. Previous data on corrosion damage assessment j =1
of distressed rc structural members or data from labora- where {P }j is the corrosion damage state vector based
tory experiments are used for this purpose. Similarly, the on the judgement of the j th expert, and wj is the weight
values of over- or under-confidence of each expert for attributed to the judgement of the j th expert. The weights
different confidence levels are also determined using the reflect the accuracy of the expert in making the judgement,
PMM theory. and can be obtained from the values of achievement (ra )
Suppose an expert has identified corrosion damage for the experts as
state(s) and assigned confidence level(s) to these corrosion
m
damage state(s) based on the given data. Then the state raj
wj = n
m ; m≥0 (12)
probabilities (probability that the structure is in damage P
state k) can be determined as raj
j =1
pk
Pk = (9) where raj is the achievement of the j th expert and m is a
P
5
value reflecting the degree of importance. When m = 0,
pk
k=1 all the experts have been attributed the same weight. As
the value of m increases, the degree of importance attached
where pk is the confidence judgement for the confidence with the achievement of the expert increases.
level assigned to the damage state k. For the corrosion dam- The over- or under-confidence associated with the judge-
age states not identified by the expert, the value of pk is ment of the j th expert is given by
taken as 0%. The state vector for the corrosion damage state
based on the judgement of the expert ‘j ’ is X
5
ECj = ouck (13)
k=1
{P }j = {P1 , P2 , P3 , P4 , P5 } (10)
where ouck is the over- or under-confidence of the expert
If there are n experts who have been asked to make judge- for the confidence level associated with damage state k.
ments independently using the same set of cues, then the The value of ECj defines the bounds for the damage state
state vector for the corrosion damage state combining the probabilities based on the expert’s judgements. This esti-
judgements of all the experts can be obtained as mate gives a rational way of determining the confidence to

8 JOURNAL OF STRUCTURAL ENGINEERING VOL. 31, NO.1, APRIL–JUNE 2004

be attached with the judgement of an expert, which is only where
possible using the PMM by modelling human mental pro-
cess on a probabilistic basis. The over- or under-confidence pij (n) = Probability {system state will by ‘j ’ after n-
associated with the corrosion damage state obtained by transitions, starting from state ‘i’}
combining the judgements of all the experts is given by
If the index space is discretised into 1 year intervals, then
X
n
pij (n) gives the probability of the system being in corro-
ECc = wj ECj (14)
sion damage state ‘j ’ at the end of n years given that the
j =1
corrosion damage state at the beginning was ‘i’.
The value of ECc will be useful as a measure of the confi-
dence that can be put on the final corrosion damage state by Life Assessment of a New Structure
processing the judgements of all the experts. A lower value
of ECc denotes tighter bounds on the damage state assessed For a new structure, the time to corrosion initiation and
and vice versa. The evaluation of the corrosion damage the rate of corrosion can be determined using Eq. 1 and
state is useful for predicting the future performance of the Eq. 3, respectively, if the values of D, cs , ccr , Icorr and α are
structure, and hence in the remaining life assessment. The known. These values are selected based on type and grade
procedure proposed in the present study for remaining life of concrete, water-cement ratio and exposure condition29 .
assessment of corrosion affected rc structural members is The capacity ratio at any time ‘t’, determined using Eq. 4,
presented in the next section. will be a random variable due to the uncertainties associ-
ated with the workmanship and the aggressiveness of the
environment to which the structural member is exposed to.
REMAINING LIFE ASSESSMENT
By computing the values of ν(t) for two consecutive years,
A rational estimation of remaining life of corrosion affected the 1-step TPM, P, can be computed. The n-step TPM,
rc structural members is required for making engineering Pn , can be computed for determining the corrosion dam-
decisions regarding the inspection/maintenance activities. age state of the structural member at the end of ‘n’ years
In this paper, the Markov Chain (MC) model that has been using Eq. 15. The corrosion damage state probabilities at
developed at SERC for health assessment of rc bridges9,27 any time can be determined from the n-step TPM for that
is used for determining the remaining life of the structural time period, using the methodology given by Balaji Rao
member. The corrosion damage state (represented by the and Appa Rao27 .
capacity ratio) of the rc structural member at a given time is By comparing the capacity ratio at any time with a target
a random variable. Also, due to the corrosion propagation, value, the service life of the structure with respect to safety
the corrosion damage state changes with time. Thus, the can be determined. The target value for capacity ratio (min-
evolution of the corrosion damage of the structural mem- imum safety level) is fixed based on engineering judgement
ber with time has to be modelled as a random process. MC and/or experience gained based on performance of simi-
is a stochastic process model in which both state space and lar structures. Different options for determining the min-
index space are discrete28 . While some of existing BMS, imum safety level have been discussed by Das30 . CEB25
such as PONTIS, uses a Markov deterioration model to gives guidelines for classifying the urgency of repairing
determine the probability of change between consecutive or strengthening structural elements undergoing damage.
condition states, and thus to assess the degree or rate of According to CEB, capacity ratio values lower than about
deterioration, in the present study, the MC model is inte- 0.5 would require immediate repair action. The probability
grated with the Brunwikian theory to take into account the of capacity ratio being less than or equal to 0.5 at different
probabilistic nature of human mental process. In this case, time instances can be determined from the n-step TPM27 .
the state space is the corrosion damage state (Table 5) of This information would help in determining the service life
the member and the index space is the time. The stochas- of the rc structural member; the service life being the time
tic evolution of the system, modelled by homogeneous MC at which the probability of capacity ratio being less than or
can be completely described by the Transition Probabil- equal to 0.5 becomes equal to a prefixed value, which in
ity Matrix (TPM), P. Typically, element pij represents the the present study is assumed to be 10−2 .
probability that the corrosion damage state of the structural
member will be i in the next year given that the corrosion Remaining Life Assessment Including the Inspection
damage state at present is j . P is a one-step transition prob- Results
ability matrix the elements of which can be evaluated ana- The MC model can also be used for determining the
lytically once the rate of corrosion propagation is known. remaining service life of existing rc structural members,
The n-step TPM is given by for which inspections have been carried out and expert
Pn = P × P × . . . × P (n-times) (15) judgement information regarding damage state are avail-
able. For existing structural members, the corrosion rates
where can be determined by measuring corrosion currents in the
field with linear polarization instruments. The state vector
P = {pij }i∈1,N ;j ∈1,N, N = number of states
for the corrosion damage of the structural member at the
Pn = {pij (n)}i∈1,N ;j ∈1,N time of inspection is obtained from the judgements given

JOURNAL OF STRUCTURAL ENGINEERING VOL. 31, NO.1, APRIL–JUNE 2004 9

by experts using Brunswikian theory. Let {Pc } be the corro- Fig. 3 . The mean values for D, cs and ccr are selected as
sion damage state vector obtained using Brunswikian the- 5 × 10−8 cm2 /s, 0.30% by weight of concrete and 0.125%
ory, and Pm be the 1-step TPM for the structural member by weight of concrete, respectively, which are represen-
for one year determined using the measured corrosion rate. tative of the values reported for similar environment for
The corrosion damage state vector after n-years from the concrete with similar properties31,32 . To account for the
time of inspection is given by variations in the values of these variables due to uncer-
tainties arising due the workmanship and aggressiveness of
{Pc (n)} = {Pc }T (Pm )n (16)
environment, they are considered to be random. The values
The probability of capacity ratio being less than or equal of coefficient of variation (COV) chosen for these variables
to 0.5 at different time instances can be determined from are given in Table 633 . Using these values, the mean and
the corrosion damage state vector corresponding to that COV of time for corrosion initiation are determined using
time using the methodology proposed by Balaji Rao and Eq. 1 as 8 years and 0.20, respectively.
Appa Rao27 , and the remaining service life of the structural
member can be determined by comparing with the target
probability. A schematic representation of remaining life
assessment without inspection and including the informa-
tion from inspection is shown in Fig. 2. The practical appli-
cation of the proposed methodology is illustrated through
an example problem below.

FIG. 3. CROSS SECTION OF T-BEAM CONSIDERED

The value of Icorr and α are taken as 8.75 µA/cm2 and

5.75 based on water-cement ratio and exposure condition29 .
The rate of corrosion is determined as 0.58 mm/year using
Eq. 3. The rate of corrosion is considered as a random vari-
able with the mean as 0.58 and COV as 0.30 to account for
the uncertainties arising due the workmanship and aggres-
siveness of environment. The cover to reinforcement (d),
FIG. 2. PROBABILITY OF THE STRUCTURAL MEMBER compressive strength of concrete (fck ) and yield strength
BEING IN A STATE REQUIRING IMMEDIATE REPAIR of steel (fy ) are also considered as random variables with
ACTION ACCORDING TO CEB [23] (SCHEMATIC mean and COV values as given in Table 6 to account for the
REPRESENTATION)
uncertainties in the workmanship. The 1-step TPM, P, is
computed and the corrosion damage state probabilities are
determined. The values of corrosion damage state proba-
EXAMPLE bilities plotted against the age of thestructural member are
shown in Fig. 4, and the probability{capacity ratio ≤ 0.5}
Service life of a reinforced concrete T -beam for a bridge with age of the structural member is shown in Fig. 5.
is estimated using the proposed methodology. The beam From Fig. 5, it is noted that at 19 years of age, the prob-
is to be designed for a bending moment of 1252.6 kNm ability of capacity ratio being less than or equal to 0.5
to carry the loads. The beam is exposed to the environ- becomes 0.01. An inspection is carried out at this time. The
mental conditions characterized by: temperature of around information (cues) obtained from the inspection is given in
35◦ C, RH of about 70% and degree of wetting and dry- Table 7. It is assumed that these information have already
ing around 0.6. The given environment can be defined as been corrected for the evaluation ability/human error (in
exposure class 4a, i.e., sea-water environment (due to high the case of human observer) and for the detection capabil-
value of degree of wetting and drying), according to Euro ity and correctness of detection (in the case of instrument).
Code 2 classifications of exposure. The minimum grade of This information (cues) are passed on to five experts, who
concrete and maximum water-cement ratio for this class of have been asked to make judgements regarding the corro-
exposure is 30 MPa and 0.55 respectively and the minimum sion damage state and to assign confidence levels for their
required cover thickness is 40 mm. The value of environ- judgements from the confidence scale of I-VII.
mental aggressiveness factor (EAF) for a temperature of
around 35◦ C, RH of about 70% and degree of wetting and Evaluation of the Experts
drying around 0.6 is obtained as 4.784 using the method-
ology proposed by Anoop et al19 . The experts have been evaluated based on their perfor-
The beam is designed satisfying the safety and ser- mance in making judgement regarding corrosion damage
viceability requirements specified in the code of prac- state for rc structural members for which actual corrosion
tice. The cross sectional details of the beam are given in damage states are known. One hundred baseline cases are

10 JOURNAL OF STRUCTURAL ENGINEERING VOL. 31, NO.1, APRIL–JUNE 2004

 TABLE 6
RANDOM VARIABLES CONSIDERED IN THE EXAMPLE PROBLEM

Variable Mean COV∗

Diffusion coefficient, D 5 × 10−8 cm2 /s 0.10

Surface chloride concentration, cs 0.30% by wt. of concrete 0.10

Critical chloride concentration, ccr 0.125% by wt. of concrete 0.05

Cover thickness, d 40 mm 0.05

Rate of corrosion, rcorr 0.58 mm/year 0.30

Compressive strength of concrete, fck 30 MPa 0.18

Yield strength of steel, fy 415 Mpa 0.12

(Note- ∗ The values of COV are from Enright and Frangopol33 )

FIG. 4. CORROSION DAMAGE STATE PROBABILITIES FOR THE RC STRUCTURAL MEMBER IN THE EXAMPLE PROBLEM
CONSIDERED

were in state 4 and 11 were in corrosion damage state

5. Five experts are considered in this study. All these
five experts are considered to be equally informed and
are equally capable in terms of qualification to assess
the damage. The information (cues) on the 100 struc-
tural members (baseline cases) were passed on to these
experts, who have been asked to make judgements regard-
ing the corrosion damage state and to assign confidence
levels for their judgements from the confidence scale of I-
VII.
The over- or under-confidence (ouci ) for the confidence
FIG. 5. PROBABILITY OF THE STRUCTURAL MEMBER BE- judgement pi for the different experts can be determined
ING IN A STATE REQUIRING IMMEDIATE REPAIR using Eq. 8. Detailed calculations for determining ouci
ACTION ACCORDING TO CEB [23]
have been carried out typically for expert 1 (Table 8). It
is assumed that similar computations have been made for
assumed for this purpose, out of which 8 were in corrosion the other four experts, and the results are plotted in Fig. 6.
damage state 1, 21 were in corrosion damage state 2, 36 The values of ouci reflect the probabilistic nature of mental
corrosion damage state 3, 24 were in corrosion damage process in judgemental decision making.

JOURNAL OF STRUCTURAL ENGINEERING VOL. 31, NO.1, APRIL–JUNE 2004 11

 TABLE 7
DATA FROM INSPECTION FOR THE EXAMPLE
PROBLEM

From Visual Inspection:

Rust stains Highly noticeable rust stains

Cracking Several longitudinal cracks;

some cracks in stirrup direction

Spalling clearly noticeable spalling FIG. 6. OVER- OR UNDER-CONFIDENCE ASSOCIATED WITH

DIFFERENT CONFIDENCE LEVELS FOR THE EX-
PORTS CONSIDERED
From Field Measurements:

Icorr (3LP) 6.0 A/cm2

TABLE 9
Ecorr −450 mV ACHIEVEMENT INDEX AND WEIGHTS FOR THE
EXPERTS
Cover depth 38 mm
Expert No. (j) raj wj
Remaining diameter
32.0 mm 1 0.93 0.215
of reinforcement

2 0.90 0.208
TABLE 8 3 0.85 0.197
RESULTS OF ASSESSMENT MADE BY EXPERT 1
(CHARACTERISED IN TERMS OF ouci ) 4 0.81 0.188

Confidence 5 0.83 0.192

Level (i) pi ni fi ouci
the weights for the five experts considered are given in
I 0.0 0 0 0 Table 9.
II 0.1 30 0 0.88
Corrosion Damage Assessment
III 0.3 19 0.105 2.52 The judgements regarding the corrosion damage state
and corresponding confidence levels for these judgements
IV 0.5 16 0.438 0.68 given by all the five experts are presented in Table 10. Using
these values, the corrosion damage state probabilities are
V 0.7 22 0.682 0.27 determined using Eq. 9, and the state vector for the cor-
rosion damage state combining the judgements of all the
VI 0.9 24 0.958 −0.95 experts is obtained using Eq. 11. These probabilities are
given in Table 11. The over- or under-confidence associ-
VII 1.0 53 1.0 0 ated with the corrosion damage state obtained by combin-
ing the judgements of all the experts is also determined
using Eq. 14 as 1.12% (the positive sign denoting over-
The achievement of the expert (defined by the achieve- confidence).
ment index, ra ), reflecting the correlation between the
judgement and the environment, can be determined using Remaining Life Assessment Including the Inspection
the linear regression model (Eq. 7) for each expert. In the Results
present study, it is assumed that such an analysis has been
carried out and the values of ra have been obtained for all The 1-step TPM, P, for the structural member for one year
the five experts. It is assumed that the experts have similar is determined using the measured corrosion rate. Using the
tendencies in viewing the environment and making judge- corrosion damage state vector obtained using the judge-
ments (viz. all having positive correlations). The values of ments of the experts and P, the corrosion damage state vec-
weight attributed to the judgement of each expert are deter- tors at different time instances are determined using Eq. 16.
mined using Eq. 12. The values of achievement index and The values of corrosion damage state probabilities plotted

12 JOURNAL OF STRUCTURAL ENGINEERING VOL. 31, NO.1, APRIL–JUNE 2004

 TABLE 10
EXPERTS’ JUDGEMENT ON CORROSION DAMAGE STATE AND ASSOCIATED CONFIDENCE LEVEL FOR
THE EXAMPLE PROBLEM

Confidence Level
Corrosion
Damage State Expert 1 Expert 2 Expert 3 Expert 4 Expert 5

1 - - - - -

2 - - - - -

3 III IV VI III V

4 VI VI V V IV

5 - - - - -
(Note: ‘-’ indicates not applicable)

TABLE 11
CORROSION DAMAGE STATE PROBABILITIES BASED ON EXPERTS’ JUDGEMENT FOR THE EXAMPLE
PROBLEM

Corrosion Damage state probabilities

Damage
State Expert 1 Expert 2 Expert 3 Expert 4 Expert 5 Combined

1 0 0 0 0 0 0

2 0 0 0 0 0 0

3 0.25 0.357 0.563 0.30 0.58 0.407

4 0.75 0.643 0.437 0.70 0.42 0.593

5 0 0 0 0 0 0

against the age of structural member are shown in Fig. 4, CONCLUSIONS

and the probability{capacity ratio ≤ 0.5} with age of the
structural member is shown in Fig. 5. It can be noted from This paper presents a methodology for damage assess-
Fig. 5 that the probability{capacity ratio ≤ 0.5} = 0.01 ment and risk-based remaining life estimation of corrosion
when the structural member is 22 years of age. Thus the affected rc structural members, by integrating research on
remaining life of the structural member can be consid- modelling thinking process with expert judgement. The
ered to be 3 years from the time of inspection against the proposed methodology tries to capture the thinking pro-
limit state of probability{capacity ratio ≤ 0.5} = 0.01. cess of the expert on a probabilistic basis which has been
It is noted that the remaining life, estimated using the embedded through the expert’s confidence in his judge-
corrosion inspection results and expert judgement, has an ment. One of the highlights of the methodology is that those
over-confidence of about 1.21%. Thus, by carrying out an involved in making judgement are considered to be rational,
inspection, the engineer has now the option to postpone the rather than classifying them as experts and non-experts, and
repair activities upto a period of three years for the prob- the performance of each expert is determined by comput-
lem considered. This type of information can be generated ing the achievement index. This is in line with the current
using the proposed methodology, which will be useful for thinking in risk perception and risk communication26 , and
making decisions regarding repair, and hence the decisions would help in creating a more effective KBS for damage
taken not only completes the value chain as proposed by assessment. The damage assessment methodology is inte-
Wong and Yao2 , but also is an improvement over the engi- grated with Markov Chain model, which is currently being
neering decisions obtained using PONTIS. researched upon at SERC for risk analysis. An example

JOURNAL OF STRUCTURAL ENGINEERING VOL. 31, NO.1, APRIL–JUNE 2004 13

problem of a rc bridge girder is presented, which illustrates 11. Balaji Rao, K., Anoop, M.B. and Appa Rao T.V.S.R.,
the usefulness of the proposed methodology in facilitating “A critical review of corrosion propagation model
decision making regarding repair. and results of probabilistic analysis of resistance
degradation of reinforced concrete flexural mem-
bers due to corrosion”, Project Report No. SS-OLP
ACKNOWLEDGEMENT 07741-RR-2000-1, Structural Engineering Research
Centre, Chennai, May 2000.
This paper is published with the kind permission of the
Director, Structural Engineering Research Centre, Chen- 12. Rodriguez J., Ortega L.M., Casal J. and Diez J.M.,
nai. The research work presented in this paper is carried “Assessing structural conditions of concrete struc-
out under the CSIR-IISc Collaborative project “Structural tures with corroded reinforcement”, in ’Concrete
damage detection using vibration data and probabilistic repair, rehabilitation and protection’, R.K. Dhir and
health assessment” (D.O. No. 70(0039)/00/EMR-II dated M.R. Jones (Eds.), E&FN Spon, London, 1996,
11.01.2002). pp. 65–78.
13. fib, “Structural Concrete: Textbook on behaviour,
design and performance Updated knowledge of the
REFERENCES CEB/FIP Model Code 1990 - Volume 3”, Interna-
tional Federation for Structural Concrete, 1999.
1. Beck, J.L, Au, S.-K and Vanik, M.W., “Monitor-
ing structural health using a probabilistic measure”, 14. Andrade, C., Alonso, M.C. and Gonzalez, J.A., “An
Computer-Aided Civil and Infrastructure Engineer- initial effort to use the corrosion rate measurements
ing, Vol. 16, No. 1, 2001, pp. 1–11. for estimating rebar durability”, in ’Corrosion rates
of steel in concrete’, N.S. Berke, V. Chaker, and D.
2. Wong, F.S. and Yao, J.T.P., “Health monitoring and Whiting (Eds.), ASTM STP 1065, American Soci-
structural reliability as a value chain”, Computer- ety for Testing and Materials, Philadelphia, 1990,
Aided Civil and Infrastructure Engineering, Vol. 16, pp. 29–37.
No. 1, 2001, pp. 71–78.
15. Gu, P., Ramachandran, V.S. and Beaudoin, J.J, “Cor-
3. Masters, L.W. and Brandt, E., “Prediction of ser- rosion of steel in concrete: assessment techniques”,
vice life of building materials and components - CIB in ’Testing and Quality Control in Cement Industry:
W80/RILEM 71-PSL-Final Report”, Materials and Progress in Cement and Concrete Vol. 3’, Akademia
Structures, RILEM, Vol. 20, 1987, pp. 55–77. Books International, New Delhi, 1997.
4. ACI 365.1R-00, “Service life prediction - State-of- 16. RILEM Draft Recommendations, “Draft recommen-
the-art report”, ACI Manual of Concrete Practice, dations for damage classification of concrete struc-
2000. tures”, Materials and Structures, RILEM, Vol. 27,
5. ASTM E632-81, “Standard practice for developing No. 170, 1994, pp. 362–369.
accelerated tests to aid prediction of service life of 17. Gonzalez, J.A., Feliu, S., Rodriguez, P., Lopez, W.,
building components and materials”, American Soci- Ramirez, E., Alonso, C. and Andrade, C., “Some
ety for Testing of Materials, 1981. questions on the corrosion of steel in concrete. Part
6. Ryall, M.J., “Bridge management”, Butterworth- II: corrosion mechanism and monitoring, service life
Heineman, 2001. prediction and protection methods, Materials and
Structures, Vol. 29, No. 186, 1996, pp. 97–104.
7. Gigerenzer, G., Hoffrage, U. and Kleinbolting, H.,
“Probabilistic mental models: A Brunswikian theory 18. Japanese Society of Civil Engineers, “Present status
of confidence”, Psychological Review, Vol. 98, No. and future trend in research related to reinforcement
4, 1991, pp. 506–528. corrosion, corrosion protection and repair”, Con-
crete Engineering Series 26, 1997.
8. Adelman, L., Miller, S.L., Henderson, D.V. and
Scholles, M., “Using Brunswikian theory and a lon- 19. Anoop, M.B., Balaji Rao, K., and Appa Rao,
gitudinal design to study how hierarchical teams T.V.S.R., “Application of fuzzy sets for estimat-
adapt to increasing levels of time pressure”, Acto Psy- ing the service life of reinforced concrete structural
chologica, Vol. 112, 2003, pp. 181–206. members in corrosive environments”, Engineering
Structures, Vol. 24, No. 9, 2002, pp. 1229–1242.
9. Balaji Rao, K. and Appa Rao, T.V.S.R., “A method-
ology for condition assessment of rc bridge girders 20. fib, “Monitoring and safety evaluation of exist-
with limited inspection data”, The Bridge & Struc- ing concrete structures: state-of-the-art report”,
tural Engineer, Ing-IABSE, Vol. 29, No. 4, 1999, International Federation for Structural Concrete,
pp. 13–26. 2003.
10. Kropp, J. and Hilsdorf, H.K., “Criteria for concrete 21. Brunswik, E., “The conceptual framework of psy-
durability”, E & FN Spon, London, 1995. chology”, University of Chicago, 1952.

14 JOURNAL OF STRUCTURAL ENGINEERING VOL. 31, NO.1, APRIL–JUNE 2004

22. Wolf, B., “Processes of constructing judgments 28. Benjamin, J.R. and Cornell, A.C., “Probability,
and actions by competent individuals with respect statistics and decision for civil engineers”, Mc-Graw
to object orientation: Programmatic ideas in Hill, 1970.
the tradition of Brunswikian thoughts”, Essay
29. Anoop, M.B., Balaji Rao, K. and Appa Rao, T.V.S.R.,
#7, The Brunswik Society, http://brunswik.org,
“A methodology for durability-based service life
2000.
design of reinforced concrete flexural members”,
23. Alvarez, M.M.R., “Contingency judgments”, Magazine of Concrete Research, Vol. 55, No. 3, 2003,
Article #31, The Brunswik Society Newsletter, pp. 289–303.
http://brunswik.org, 1999.
30. Das, P.C., “Application of reliability analysis
24. Brehmer, B. and Hagafors, R., “Use of experts in in bridge management”, Engineering Structures,
complex judgment decision making: A paradigm for Vol. 20, No. 11, 1998, pp. 957–959.
the study of staff work”, Organizational Behaviour
31. Bamforth, P.B., “The derivation of input data for
and Human Decision Processes, Vol. 38, 1986,
modelling chloride ingress from eight-year UK
pp. 181–195.
coastal exposure trials”, Magazine of Concrete
25. CEB, “Assessment of concrete structures and design Research, Vol. 51, No. 2, 1999, pp. 87–96.
procedures for upgrading”, CEB Bulletin No. 162,
32. Hobbs, D.W., “Chloride ingress and chloride-
1983, pp. 87–90.
induced corrosion in reinforced concrete members”,
26. Reid, S.G., “Perception and communication of risk Proc. of 4th International Symposium on Corro-
and the importance of dependability”, Structural sion of reinforcement in concrete construction, Cam-
Safety, Vol. 21, No. 4, 1999, pp. 373–384. bridge, 1996, pp. 124–135.
27. Balaji Rao, K. and Appa Rao, T.V.S.R., “Stochas- 33. Enright, M.P. and Frangopol, D.M., “Probabilistic
tic modelling of crackwidth in reinforced concrete analysis of resistance degradation of reinforced con-
beams subjected to fatigue loading”, Engineering crete bridge beams under corrosion”, Engineering
Structures, Vol. 26, No. 5, 2004, pp. 665–673. structures, Vol. 20, No. 11, 1998, pp. 960–971.

JOURNAL OF STRUCTURAL ENGINEERING VOL. 31, NO.1, APRIL–JUNE 2004 15