2010 JOSE Mba

Download as pdf or txt
Download as pdf or txt
You are on page 1of 9

Journal of Structural Engineering No.

37-11
Vol. 37, No. 03, April - May 2010, pp. 66–74

Change point detection in corrosion health monitoring using statistical


techniques

M.B. Anoop *, ∗ , K. Balaji Rao∗ , B.K. Raghuprasad∗∗ , K. Ravisankar∗ and Nagesh R. Iyer∗

CSIR-Structural Engineering Research Centre, CSIR Campus, Taramani, Chennai - 600 113, India.
∗∗
Department of Civil Engineering, Indian Institute of Science, Bangalore 560 012, India.
Received 02 February 2010; Accepted 26 March 2010

With the development of more comprehensive strategies for on-line monitoring and the developments in smart sensor
technology and digital data acquisition, there is a need to develop change point detection algorithms which can be used
on-line for automated diagnosis of structural health. This is especially required in the case of bridge stocks consisting
of large number of bridges which are monitored simultaneously. For structures affected by chloride-induced corrosion
of reinforcement, the ability to detect corrosion of reinforcement in its early stages is critical in directing repairs to
the most at-risk structures and will help in optimizing the use of limited funds. In this study, the identification of time
of corrosion initiation is modeled as a problem of change point detection in online monitored electrochemical current
noise data. For change point detection, an algorithm based on maximum likelihood approach is considered. Formu-
lations are made for the determination of window size for the online monitored data points to be considered and the
threshold value for the decision function. The usefulness of the algorithm is studied by considering an example pro-
blem of identification of time of corrosion initiation in a reinforced concrete bridge girder, in a Monte Carlo simulation
framework. The studies presented in this paper are towards realizing sustainable infrastructure considering service
life planning and declaration and assessment of sustainability aspects.

KEYWORDS: Reinforced concrete; chloride-induced corrosion; corrosion initiation; change point detection; structural he-
alth monitoring.

At present, the design and construction of infrastructural The ISO standards (ISO 15392, ISO 15686) 1,2 aim to im-
systems is guided by functional performance and conven- plement the concept of sustainability and to bring conside-
tional financial costing. However, the need to consider the ration of sustainability to an internationally established com-
sustainability of the built environment in the infrastructural mon ground 3. These standards are performance-based rather
design decisions is an important issue warranting attention. than prescriptive, and embrace the concept of performance-
Sustainability implies that the needs of the present generation based building (The term building here refers to the activity
are met without wasting, polluting or damaging/destroying of building). This concept is based on the clear and unambi-
the environment and without compromising the ability of the guous identification of verifiable performance requirements.
future generations to meet their needs. A sustainable infra- The concept of service life planning allows for the conside-
structure should take into consideration the three dimensions ration of how the performance of the building develops over
of sustainability, namely, social, environmental and economi- time. A design option is considered reasonable when it meets
cal. According to the concept of sustainability, the entire life or exceeds the performance requirements over time. The me-
cycle of a structure or an infrastructure can be divided into fi- thodological context of performance-based building, service
ve phases, ’from cradle to grave’, including planning, design, life planning, and declaration and assessment of building re-
construction, operation/maintenance and removal. The con- lated sustainability is shown in Fig. 1.
cept of sustainability brings to the fore the importance of: (a) Performance-based building (PBB): Considering
life-cycle management, (b) rational scheduling of in-service performance-based building as a starting point helps in esta-
inspection, (c) repair and rehabilitation, retrofitting, (d) use blishing clear and quantitative performance requirements,
of energy efficient/non-conventional materials of constructi- which can be used to identify a building design that suc-
on at various stages of life of the structure, and, (e) appli- cessfully meets the clients’ real demands. This requires
cation of advanced research tools/techniques to ascertain the transforming the user requirements (containing aspects of
efficiency of existing structures and/or extend the life of the functionality, quality, comfort, efficiency, etc.) into technical
existing structures. performance requirements. With the identification of perfor-
* E-mail: anoop@sercm.org
(Discussion on this article must reach the editor before July 31, 2010)

66 JOURNAL OF STRUCTURAL ENGINEERING, VOL. 37, NO. 01, APRIL - MAY 2010
mance requirements, the decision process can relate to and of in-service structures on a continuous basis with mini-
restrict itself to these performance requirements. mum labor requirement. Structural health monitoring is a
PBB is, in short, all about handling aspects related to de- monitoring methodology intended to continuously assess a
mand of performance requirements to the supply of perfor- structural system for identification of damage. Long-term
mance functionality. It may be noted that the supply of per- continuous monitoring of major bridges (where long-term
formance provided by a building at its design level is only designates years-to-decades and desirably the entire life cy-
considered here. cle) using permanent monitoring systems, is a very recent
In this stage, the demand of performance requirements concept, enabled by recent advances in sensing, data acqui-
and the supply of performance by different design alternati- sition, computing, communication, and data and informati-
ves are identified (scenario identification). on management 4. One of the major shortcomings of perma-
Service Life Planning (SLP): Performance of a building nent monitoring systems is the extensive lengths of coaxial
usually decreases with time. The concept of SLP allows for wires required for transfer of sensor measurements, which
the consideration of how the performance of the building de- drives up installation and maintenance costs. Straser and
velops over time. SLP establishes a rationale for how the Kiremidjian5 proposed integration of wireless radio with sen-
long-term performance of a design alternative can be model- sor to reduce the cost of structural health monitoring systems.
led for a specific building design with a defined use pattern Lynch6 has extended the functionality of wireless sensors by
and under the exposure to an identified environment. integrating sophisticated microcontroller with them to enable
sensor-based execution of embedded engineering algorithms
Performence-based building for data interrogation. Performing data interrogation at the
Demand of Supply of
wireless sensor is prudent from an energy standpoint, as wi-
performaence performence reless radios consume the most power in the wireless sensing
(requirments) (functionality) unit. Wireless communication of raw time history records
would be an inefficient use of limited battery resources. In-
Scenario stead of transmitting raw time history data, the wireless sen-
identifica- sing unit is used to first interrogate the data to distil a small
tion number of indicators that would then be wirelessly transmit-
Service Life Planning ted. For instance, damage detection algorithms could be used
to determine if damage is present and wireless radio is used
Performence over
Initial Performence for transmitting data only if damage was found 7. The com-
time
mon feature of damage detection is the fact that the problem
Process
of interest is the detection of one or several changes in some
identifica- characteristic properties of the considered system. Thus, ma-
tion ny damage detection problems can be stated as the problem
of detecting a change in the parameters of a static or dyna-
Declaration and Assessment
mic stochastic system. The time instant at which the change
Description Quantification Assessment of interest occurs is called the change point 8 .
Aspects Impacts Performence From the above discussion, it is noted that there is a need
to develop algorithms for change point detection in on-line
Economic monitoring data recorded/cached in a central location and
Environment analyzed in real-time for automated diagnosis of structural
health. In civil structures, the identification of damage be-
Social
fore critical failure is of extreme importance. For structures
Fig. 1 The sequential relationship of the concepts of performance-based affected by chloride-induced corrosion of reinforcement, the
building, service life planning and declaration, and assessment of ability to detect corrosion of reinforcement in its early stages
sustainability aspects of buildings3 . is critical in directing repairs to the most at-risk structures
and will help in optimizing the use of limited funds. With the
Modelling the performance over time allows identifying development of more comprehensive strategies for on-line
maintenance processes as well as the timing for replacement monitoring and the developments in smart sensor technolo-
of components (process identification). Remote health moni- gy and digital data acquisition 9 , there is a need to develop
toring is an emerging paradigm which needs to be considered change point detection algorithms which can be used on-line
at this stage, especially for important infrastructural facilities for automated diagnosis of structural health. This is especi-
like bridges. ally required in the case of bridge stocks consisting of large
Declaration and Assessment: The process identification hel- number of bridges which are monitored simultaneously.
ps in having a more detailed description and a subsequent From laboratory experimental investigations, it has been
quantification of all processes and activities taking place du- noted that electrochemical noise can indicate the current level
ring the life cycle of the building (declaration). Using this of corrosion activity of steel in concrete, especially transition
information, the quantification and assessment of impacts; from passive state to active corrosion. There is a need to de-
whether these are economic, environmental, or social, is car- velop automated procedures, which can be used to identify
ried out. Finally, the performances of different design alterna- the time of corrosion initiation from the online monitored
tives are assessed, based on which the selection of a particular electrochemical noise data. Towards this, the identification
design alternative is made. of time of corrosion initiation is modeled as a problem of
Sustainability of the infrastructure performance can be change point detection in online monitored electrochemical
assessed by performing continuous structural health monito- current noise data. For change point detection, an algorithm
ring. The essence of structural health monitoring can be con- based on maximum likelihood approach is considered, and
sidered to involve measurement, inspection, and assessment the performance of the algorithm is studied in a Monte Car-

JOURNAL OF STRUCTURAL ENGINEERING, VOL. 37, NO. 01, APRIL - MAY 2010 67
lo simulation framework. The studies presented in this pa- iii. Zero resistance ammeter (ZRA) mode (WE-WE-RE) in
per are towards the service life planning and declaration and which a zero resistance ammeter between two WEs mea-
assessment of sustainability aspects for realizing sustainable sure the current and a voltmeter between WE and RE
infrastructure. The details of the studies are given in the fol- measures the potential. This arrangement is used for EN
lowing sections. measurements under freely corroding conditions, and is
useful for on-line monitoring of corrosion in reinforced
CORROSION MONITORING USING ELECTROCHE- concrete structures.
MICAL NOISE TECHNIQUE
IDENTIFICATION OF TIME OF CORROSION IN-
Studies by various researchers 10-14 suggested that localized ITIATION AS A CHANGE POINT DETECTION PRO-
corrosion processes (such as that associated with chloride- BLEM
induced corrosion of reinforcement in concrete) give parti-
cularly strong electrochemical noise response. The type of Consider a reinforced concrete member, wherein the corrosi-
corrosion is indicated by the coefficient of variation of cur- on currents are monitored by recording current flow between
rent noise, ranging from 10 −3 for general corrosion to 1.0 two identical, electronically isolated, rebar probes, embedded
for localized corrosion 15. Before initiation of corrosion, the in concrete, and coupled through a ZRA. At t i (when depas-
reinforcement in concrete is in the passive state (corrosion sivation of steel occurs), there is a shift in mean corrosion
currents are negligible, i.e., < 1 mA/m 2 ) and hence the mean current, indicating initiation of active corrosion. The actual
corrosion current can be taken as zero. When depassivation shift in mean value of corrosion current depends on diffe-
of steel occurs, there is a shift in the mean corrosion current, rent factors (viz. humidity content in concrete, temperature,
indicating initiation of active corrosion. Thus, by detecting etc.) Andrade et al 17 presented typical ranges for corrosion
the shift in the mean corrosion current, time of corrosion in- current for different exposure conditions, based on measure-
itiation can be identified. In this study, identification of ti- ments made on laboratory specimens and on real structures.
me of corrosion initiation is posed as a problem of detection These ranges of values of corrosion current for different ex-
of single change point in on-line monitored electrochemical posures can be further subdivided 18 using typical trend of va-
current noise data. riation of rate of corrosion with water-cement ratio 19 . Thus,
knowing the exposure condition and water-cement ratio used,
Electrochemical noise the range of values of corrosion current that can be expected
in the girder after depassivation can be identified, which will
give an idea about amplitude of shift in mean corrosion cur-
Electrochemical noise (EN) is a general term for the ’ran-
rent. Thus, the identification of t i can be viewed as a problem
dom’ fluctuations in current or potential which occurs as an
of identifying the time of shift in mean of the monitored cor-
electrochemical process proceeds. EN technique is an emer-
rosion current data, i.e., a change point detection problem.
ging technique for monitoring corrosion of reinforcement in
The development of algorithms for change point detec-
concrete11, 12, 16 . While the corrosion current is related to the
tion in signals is an active area of research with applicati-
rate (kinetics) of the reaction, the electrochemical potential is
ons in various disciplines 8, 20-24 . The different approaches for
related to the driving force (thermodynamics) of the reaction.
solving change point detection problem include maximum
The advantages of EN technique are 14 :
likelihood, Bayesian, Bayes-type, nonparametric as well as
i. Lack of intrusiveness (its application does not involve ex- decision-theoretic procedures 25. While most of these approa-
ternal perturbation of the corroding system) ches are based on time-domain analysis, in some cases, it
ii. Instruments required to make the measurements are re- may be required to analyze data in both time and frequency
asonably simple, particularly with modern computer- domains. For instance, a step-shift in the mean of the signal
based data acquisition techniques, and, is localized in the time domain, whereas a change in variance
iii. Localized corrosion processes, which are difficult to mo- is more localized in the frequency domain 26 . Thus, if the in-
nitor with other techniques, tend to give particularly terest is to detect changes in both mean and variance, the data
strong EN signals should be analyzed using a time-frequency approach. Since,
in the present study, the interest is in the identification of a
step-shift in the mean of the corrosion current data, a time-
Measurement of electrochemical noise
domain based algorithm is used.
Arrangement for measurement of electrochemical noise typi- ALGORITHM BASED ON MAXIMUM LIKELIHOOD
cally consists of three electrodes arranged as WE-RE-CE or APPROACH FOR IDENTIFICATION OF TIME OF
WE-WE-RE or WE-WE-WE, where WE is a working elec- CORROSION INITIATION
trode (made of the same type of metal that is being monito-
red), RE is a reference electrode (which maintains a constant This is one of the oldest and most well known approaches to
potential in the environment), and CE is a counter electrode identify the change detection. The approach is based on the
(made of a noble material like platinum). The measurement log-likelihood ratio of the observations 8.
arrangements can be classified as: It is assumed that the monitored corrosion current data
i. Potentiostatic (WE-RE-CE) in which constant potential can be represented using a Gaussian white noise (GWN) pro-
is maintained between WE and RE and the current bet- cess. A sample of N observations (N points of the monito-
ween WE and CE is monitored and recorded. This ar- red data) is considered, and a decision function is computed
rangement is used in laboratory electrochemical studies to test between the two following hypotheses (H 0 and H1 )
under polarized conditions. about the parameter of interest (mean, μ, inthepresentstudy).
ii. Galvanostatic (WE-RE-CE) in which a constant current H0 : μ = μ0 ; no change in occured
is passed through WE and CE, and the potential of WE
is monitored and recorded against RE, and, H1 : μ = μ1 ; change has occured (1)

68 JOURNAL OF STRUCTURAL ENGINEERING, VOL. 37, NO. 01, APRIL - MAY 2010
The change detection is commonly carried out by com- accept H0 accept H1
puting a decision function, S N , using the observed data, and
comparing it with a threshold value Define the decision func- H0 is true
tion, S N , as: No error Type I error

N

SN = Si (2)
i=1 H1 is true Type II error No error

where si is the log-likelihood ratio for the observations


yi , i = 1, N . If pY (y) is the probability density function for Fig. 2 Schematic of decision possibilities in hypothesis testing.
the observed process, then s i is defined as:
  Type I error occurs when there is actually no change in
pY (Yi ; μ = μ1 )
si = ln (3) mean but the algorithm detects a change in mean, while Ty-
pY (yi ; μ = μ0 ) pe II error occurs when there is actually a change in mean
For the Gaussian process (as considered in the present but the algorithm does not detect the change. The probabili-
study), Eq. (3) can be written as: ty of Type I error is called the significance level (α) of the
⎛  2 ⎞ hypothesis test, and the probability of Type II error is cal-
1 y − μ1 led the operating characteristic (β) of the hypothesis test 27 .
⎜ 1 − ⎟ The complementary probability (1 − β) is called the power
⎜ √ e 2 σ ⎟
⎜ σ 2π ⎟ of the test. In change detection, Type I- and Type II- errors
si = ln ⎜⎜  2 ⎟ (4) are called false detection and non-detection, respectively. The

⎜ 1 y − μ2 ⎟ probability of false detection (P F D ) and the probability of
⎝ 1 − ⎠
√ e 2 σ non-detection (P N D ) are two of the different performance
σ 2π indices used for designing and evaluating change detection
algorithms, and these two values should be as low as pos-
where μ0 is the original mean (equal to zero in the present sible. There is a trade-off between the probability of false
case) and μ1 is the changed value of mean of the Gaussian detection and the probability of non-detection, and as one is
distribution, and σ is the standard deviation of the Gaussian reduced, the other one increases. The probability of false de-
distribution. Simplifying Eq. (4), one get: tection need to be kept to a minimum based on economic
  considerations since a detailed inspection need to be carried
μ1 − μ0 μ1 + μ0
si = yi − (5) out every time the algorithm detects a change. However, in
σ2 2 the case of chloride-induced corrosion of reinforcement in
The above equation can be written as (Basseville and Ni- reinforced concrete structural members, timely identification
kiforov, 1993): of corrosion initiation is important for controlling the dama-
  ge due to corrosion and for cost-effective maintenance and
b A repair. Therefore, the probability of non-detection should be
si = yi − μ0 − (6)
σ 2 kept as a minimum, based on economic and social considera-
tions. This can be achieved by suitably choosing the sample
where size (N ) and the threshold (h). A procedure is given in the
following section for determining the values of N and h, ba-
A = μ1 − μ0 (7)
sed on the allowable values of probability of false detection
is the amplitude of shift and and the probability of non-detection.
μ1 − μ0
b= (8) Determination of sample size and threshold for the deci-
σ
sion function - proposed method
is the signal-to-noise ratio (since the signal to be detected is
the change in mean from μ 0 to μ1 and σ is the characteristic The determination of probability of false detection and the
of the noise in the observations about the mean). probability of non-detection requires the estimation of the
Substituing Eq. (6) in Eq. (2), the decision function beco- mean and standard deviation (SD) of S N . The same can be
mes: determined as follows.
N   Since in the present study, μ 0 = 0, the amplitude of shift
N b A A = μ1 (from Eq. (7)). Accordingly, the decision function
S = yi − μ0 − (9)
σ i=1 2 given by Eq. (9) can be written as:
N   N
The stopping rule (decision rule) for the change detection
N b A b  N μ1
algorithm is given by: S = yi − = yi − (11)
σ i=1 2 σ i=1
2
N
0 if S < h; H0 is chosen
d= (10)
1 if S N ≥ h; H0 is chosen Suppose there are r sample points corresponding to the
Gaussian process with changed value of mean (μ 1 ). Then
where h is a conveniently chosen threshold. The threshold Eq. (11) can be written as:
value can be chosen based on the error probabilities in hy-
pothesis testing. When a decision regarding a hypothesis is N
 −r N

N b N μ1
made, four possibilities exist and two of them lead to error S = yi + yi − (12)
(Fig. 2). σ i=1
2
i=N −r+1

JOURNAL OF STRUCTURAL ENGINEERING, VOL. 37, NO. 01, APRIL - MAY 2010 69

Since the yi s are Gaussian, S N follows a Gaussian distri- where Φ(.) represents the cumulative density function of
bution. The mean and variance of S N are given by: standard normal distribution and SD[.]
 represents the stan-
dard deviation. Substituting for S0N and SD[S0N ] using
N
 −r N

  b N μ1 Eq. (15) in Eq. (7),
SN = yi  + yi  −
σ 2

i=1 i=N −r+1
 h + b2 N2
b N μ1 α=1−Φ √ (19)
= (N − r)μ0 + rμ1 − (13) b N
σ 2
 
b N The non-detection occurs when H 0 is chosen (no change
= μ1 r − is detected) when H 1 is actually true (there is actually a
σ 2
  change in mean). This happens when the value of the decisi-
N on function, S rN , remains within the threshold, h. Thus, the
= b2 r −
2 probability of non-detection (P N D ) is given by:
 2 N −r N
  b 
var S N = var[yi ] + var[yi ] PN D = β = P r SrN ≤ h) (20)
σ i=1 i=N −r+1
 2 Since SrN follows a Gaussian distribution, Eq. (20) can
b  
= (N − r)σ 2 + rσ 2 = b2 N (14) be written as
σ
 
It is noted from these equations [Eqs. (13) and (14)] that h − SrN
β= Φ (21)
the mean and variance of S N depend upon the signal-to-noise SD[SrN ]
ratio (b) and the sample size (N ). If there are no change
  2  
points, then r = 0 and S N = − b 2N , that is the mean Substituting for SrN and SD[SrN ] using Eq. (16) in the
value of the decision function is negative when there are no above equation,
change points in the data. When all the N data points are with
  2
a change in mean, then r = N and S N = b 2N . This is in h − b2 N2
agreement with the typical behaviour of the decision func- β= Φ (22)
bN
tion corresponding to a change in the mean of a Gaussian
sequence with constant variance 8 .
Let S0N and SrN represent S N corresponding to the hy- From Eqs. (19) and (22), it is noted that α and β are func-
potheses H0 and H1 , respectively. For hypothesis H 0 , r = 0 tions of both N and h. For the specified values of α and β,
and for hypothesis H 1 , r can be any value between 1 to N . Eqs. (19) and (22) can be solved simultaneously to obtain
Therefore, the values of mean and variance of S 0N and SrN the required values of N and h to be used in change point
can be determined as: detection.
  The usefulness of the proposed algorithm for identificati-
 N 2 N N on of corrosion initiation is illustrated through an application.
S0 = b 0− = −b2
2 2
var[SoN ] = b2 N (15) APPLICATION
N N  
 N  N  2 N
Sr = S = b r−
r=1 r=1
2 A reinforced concrete bridge girder, located in a severe en-
  vironment (as per the definitions of exposure conditions in
N (N + 1) N 2 N IS 456-2000 28 with cross-sectional details as shown in Fig.
= b2 − = b2 (16)
2 2 2 3 is considered. For studying the efficiency of the proposed
N
 N
 algorithms for change point detection, an ensemble of y(t)
var[SrN ] = var[S ] =N
b2 N is generated which is assumed to represent the electroche-
r=1 r=1 mical current noise data obtained from on-line monitoring
2 2 and stochasticity in time of occurrence of change point event
= b N
(initiation of chloride-induced corrosion) is taken into con-
As explained above, false detection occurs when H 1 is sideration. The entire problem has been formulated within
chosen (a change is detected) when H 0 is actually true (the- the framework of Monte Carlo simulation 29 , and is depicted
re is actually no change in mean). This happens when the schematically in Fig. 4.
value of the decision function S 0N becomes larger than the
threshold, h. Thus, the probability of false detection (P F D ) 2000 mm
is given by: 205 mm
PF D = α = P r(S0N > h) = 1 − P r(S0N ≤ h) (17) Ast = 6 Nos. 30 mm dia. bars
Ast
990 mm fck = 37.5 MPa
Since S0N
follows a Gaussian distribution, Eq. (17) can Water cement ratio = 0.45
be written as
 
h − S0N 350 mm
α=1−Φ (18)
SD[S0N ] Fig. 3 Cross-sectional details of reinforced concrete bridge girder.

70 JOURNAL OF STRUCTURAL ENGINEERING, VOL. 37, NO. 01, APRIL - MAY 2010
ξ1 (t) realizations of GWN process are generated representing the
possible realizations of monitored electrochemical noise for a
period of 100 years at an interval of 0.01 years. One thousand
ξ2 (t) lognormal random variables, representing time-to-corrosion
y1(t)
S . initiation, one for each realization of the observed process,
.
. are generated. Typical realizations of the observed process
y2(t) (electrochemical noise) without and with shift (corrosion in-
ξ100 (t)
. itiation) are shown in Fig. 5.
.
.
λ1
y1000(t)
λ2
H
.
.
.
λ1000

S - System (reinforced concrete structural member)


ξi (t), i = 1,2, …, 1000 – Different realizations of response of system
(electrochemical current noise) when there were no active corrosion
H – Initiation of chloride-induced corrosion due to due to ingress of
chlorides from environment (hazard)
λi, i = 1,2, …, 1000 – Different realizations of time ofcorrosion initiation
yi (t), i = 1,2, …, 1000 – Different realizations of response of system with
the initiation of chloride-induced corrosion at time λi

Fig. 4 Schematic representation of problem considered.

Assuming ingress of chlorides into cover concrete as a


diffusion process, time-to-corrosion initiation (t i ) can be de-
termined from Fick’s second law of diffusion as
  −2
d2 −1 cs − ccr
ti = erf (23)
4D cs
where d is the clear cover to reinforcement, D is the diffusion
coefficient for chlorides in concrete, c s is the surface chlori-
de concentration and c cr is the critical chloride concentrati-
on. To account for variations in workmanship and exposure
conditions, d, D, cs and ccr are treated as random variables.
The values of mean and standard deviation of these random Fig. 5 Typical realizations of the observed process (simulated in the
variables are given in Table 1. All the random variables are present study) without- and with- shift (time of corrosion
assumed to be statistically uncorrelated with each other. The initiation = 16.1 years).
mean and standard deviation of time-to-corrosion initiation
are determined using first order approximation as 14.11 years
RESULTS AND DISCUSSION
and 9.42 years, respectively. It is assumed that t i follows a lo-
gnormal distribution 30.
For the problem considered, the signal-to-noise ratio (b) is
3. The variation of decision function, S N , with time for a
TABLE 1
typical realization of the observed process, for sample size
VALUES OF MEAN AND STANDARD DEVIATION (SD) N = 100, is shown in Fig. 6. It is noted from this figure
FOR THE RANDOM VARIABLES CONSIDERED
that, as expected, the value of S N is negative in the begin-
variable mean SD Remarks ning (when there isno shift), with a magnitude around the
d (mm)

45 2.25 Assumed cov of 0.05 mean value of S 0N ( S0N = −b2 N2 = −450 using Eq. 15)
D (cm2 /s) 5 × 10−8 1 × 10−8 cov = 0.20
corresponding to the hypotheses H 0 . After the shift has oc-
(Balaji Rao et al, 2004)31
cs 0.25 0.05 cov = 0.20
curred (corrosion has initiated), the value of S N increases till
(% by weight of concrete) (Balaji Rao et al, 2004)31 the number of changed points in the sample becomes N , i.e.,
ccr 0.125 0.025 Assumed cov of 0.20 r = N . After this, the value of S N is around the mean value
(% by weight of concrete) = b2 N2 = 450 (using Eq. 16), as is noted from Fig. 6.
Considering a 1% level of significance (probability of fal-
The amplitude of shift in mean corrosion current is taken se alarm = 0.01), the variation of sample size (N ) and thres-
as 0.15μA, with a standard deviation of 0.05μA, which is hold (h) required for different values of probability of non-
consistent with exposure condition considered for the girder. detection are shown in Fig. 7.
In the present study, simulated electrochemical noise data, From Fig. 7, it is noted that as the probability of non-
representing the monitored corrosion currents, is used. Cottis detection decreases, N increases, and h decreases. This sug-
et al32 . used a shot noise model to simulate electrochemical gests that a large value of N should be chosen for minimizing
noise data. It is assumed that monitored electrochemical noi- the probability of non-detection. But, as the sample size in-
se data can be represented by a GWN process. One thousand creases, the delay in detection also increases. This is because

JOURNAL OF STRUCTURAL ENGINEERING, VOL. 37, NO. 01, APRIL - MAY 2010 71
one has to wait till the window is completely filled for com- time of corrosion initiation is shown in Fig. 9, and frequency
puting the value of decision function. Thus, an optimal value distributions of the same are shown in Fig. 10. From these
of probability of non-detection should be selected conside- figures, it is noted that the predicted times of corrosion initia-
ring the delay in detection also. The variation in mean and tion are in good agreement with the actual times of corrosion
standard deviation of delay in detection for different values initiation.
of probability of non-detection are shown in Fig. 8.
500
400
300
200
100 ti
N

0
S

0 10 20 30 40 50 60 70 80 90 100
--100 time (years)
--200
--300
--400
--500
Fig. 6 Variation of decision function with time (N = 100, time of
corrosion initiation = 16.4 years).

Fig. 9 Comparison of actual and detected times of corrosion initiation.

Fig. 7 Variation in sample size and threshold with probability of


non-detection (probability of false detection, α = 0.01). Fig. 10 Frequencies of actual and detected times of corrosion initiation.

The values of mean and standard deviation of delay in


detection are 0.06 years and 0.025 years, respectively. The
small values of mean and standard deviation of delay in de-
tection indicate the usefulness of the proposed algorithm.

CONCLUSIONS

An algorithm is proposed for identifying time of corrosi-


on initiation in reinforced concrete structures using on-line
electrochemical noise data measured using ZRA technique.
The algorithm is based on the maximum likelihood approach.
Formulations are also made for the determination of window
Fig. 8 Variation in statistical properties of delay in detection with size and the threshold value for the decision function based
probability of non-detection. on the signal-to-noise ratio. The usefulness of the algorithm
is studied by using an example problem of identification of
From Fig. 8, it is noted as probability of non-detection time of corrosion initiation in a reinforced concrete bridge
decreases (accordingly, the sample size N required increa- girder, using simulated corrosion current data. The results in-
ses), the mean and standard deviation of delay in detection dicate that proposed algorithm has desirable properties of on-
increases. However, upto a value of 0.0035 of probability of line change point detection algorithms. The window size and
non-detection, the increase in the mean and standard devia- the threshold value for the decision function determined in
tion of delay in detection is marginal. If the probability of the study are specific to the problem considered. More stu-
non-detection is decreased further, there is a sudden increase dies are required to develop general guidelines on window
in the mean and standard deviation of delay in detection. The- size and the threshold value for the decision function to be
refore, the optimal value of probability of non-detection for used for reinforced concrete structural members in different
the problem considered is about 0.0035 (0.35%). The corre- exposure conditions. The studies presented in this paper will
sponding values of N and h (from Fig. 7) are 59 and -265.5. be useful while carrying out service life planning (by hel-
These values are used in the present study for change point ping in planning maintenance/repair) and declaration and as-
detection. The comparison of actual and predicted values of sessment of sustainability aspects (economic and social con-

72 JOURNAL OF STRUCTURAL ENGINEERING, VOL. 37, NO. 01, APRIL - MAY 2010
siderations by controlling the damage due to corrosion and 13. Smulko, J.M., Darowicki, K. and Zieli?ski, A., 2006.
helping in cost-effective maintenance and repair). Evaluation of reinforcement corrosion rate in concrete
structures by electrochemical noise measurements. Rus-
ACKNOWLEDGEMENT sian Jl. of Electrochemistry, 42: 546-550.
14. Cottis, R.A. and Llewellyn, A., 2007. Elec-
This paper is being published with the kind permission of the trochemistry for corrosion. http://www.cp. um-
Director, CSIR-SERC, CSIR Campus, Taramani, Chennai. ist.ac.uk/lecturenotes/Echem/Noisefile.html (last acces-
sed on 06.06.2007).
REFERENCES
15. Gowers, K.R. and Millard, S.G., 1999. Electrochemi-
cal techniques for corrosion assessment of reinforced
1. ISO 15686-1:2000 (E), Buildings and Constructed As- concrete structures. Proc. of Institution of Civil Engrs.,
sets - Service Life Planning-Part 1: General Principles, Structs. & Buildings, 134:129-137.
International Organization of Standardization, Geneva.
16. Legat, A., Leban, M. and Bajt, Z., 2004. Corrosion pro-
2. ISO DIS 15392, Building Construction - Sustainability cesses of steel in concrete characterized by means of
in Building Construction, International Organization of electrochemical noise. Electrochimica Acta, 49:2741-
Standardization, Geneva. 2751.
3. Trinius, W. and Sjostrom, C., 2007. Sustainability in 17. Andrade, C., Alonso, M.C. and Gonzalez, J.A. An initial
building construction - international standards in pro- effort to use the corrosion rate measurements for estima-
gress. Jl. of ASTM International. Vol. 4, No. 7. ting rebar durability. in Berke, N.S., Chaker, V. and Whi-
4. Aktan, A.E., Catbas, F. N., Grimmelsman, K. A. and ting, D., Ed., Corrosion rates of steel in concrete, ASTM
Pervizpour, M., 2002. Development of a model health STP 1065, American Society for Testing and Materials,
monitoring guide for major bridges, Report submitted Philadelphia, 1990, pp 29–37.
to Federal Highway Administration Research and De- 18. Anoop, M.B., Balaji Rao, K. and Appa Rao, T.V.S.R.
velopment, CON99TRACT/ORDER NO. DTFH61-01- (2003). A methodology for durability-based service life
P-00347. design of reinforced concrete flexural members. Mag. of
5. Straser, E.G. and Kiremidjian, A.S., 1998. A modu- Con. Res., 55(3):289-303.
lar wireless damage monitoring system for structures, 19. Baweja, D., Roper, H. and Sirivivatnanon, V., 1999.
Report No. 128, John A. Blume Earthquake Enginee- Chloride-induced steel corrosion in concrete: part 2 -
ring Center, Department of Civil and Environmental En- gravimetric and electrochemical comparisons. ACI Mat.
gineering, Stanford University, Stanford, CA. Jl., 96(3):306-313.
6. Lynch, J.P., 2002. Decentralisation of wireless monito- 20. Kailath, T., 1998. Detection of stochastic processes. IE-
ring and control technologies for smart civil structures, EE Trans. on Information Theory, 44(6):2230-2259.
Report No. 140, John A. Blume Earthquake Engineering
Center, Department of Civil and Environmental Engi- 21. Perreault, L., Haché, M., Slivitzky, M. and Bobée, B.,
neering, Stanford University, Stanford, CA. 1999. “Detection of changes in precipitation and runoff
over eastern Canada and U.S. using a Bayesian ap-
7. Lynch, J.P. Overview of wireless sensors for real-time proach”. Stochastic Environmental Research and Risk
health monitoring of civil structures. Proc. of the 4th Intl. Assessment, 13: 201-216.
Workshop on Structural Control and Monitoring, New
York City, NY, USA, June 10-11, 2004. 22. Vellekoop, M.H. and Clark, J.M.C., 2003. A nonlinear
filtering approach to change point detection problems:
8. Basseville, M. and Nikiforov, I.V., 1993. Detection of Direct and differential geometric methods. SIAM J. Con-
abrupt changes: theory and applications. Prentice-Hall, trl Optim., 42(2): 469-494.
Englewood Cliffs, NJ.
23. Jann, A., 2000. Multiple change-point detection with a
9. Pines, D. and Aktan, A.E., 2002. Status of structural genetic algorithm. Soft Computing, 4:68-75.
health monitoring of long-span bridges in the United
States. Pro. in Struct. Engg. and Mat., 4:372-380. 24. Staudacher, M., Telser, S., Amann, A., Hinterhuber, H.
and Ritsch-Marte, M., 2005. A new method for change-
10. Hardon, R.G., Lambert, P. and Page, C.L., 1988. Relati- point detection developed for on-line analysis of the he-
onship between electrochemical noise and corrosion rate art beat variability during sleep. Physica A, 349:582-
of steel in salt contaminated concrete. British Corrosion 596.
Jl., 23:225-228.
25. Jandhyala, V.K., Zacks, S. and El-Shaarawi, A.H., 1999.
11. Katwan, M.J., Hodgkiess, T. and Arthur, P.D., Change-point methods and their applications: contribu-
1996.Electrochemical noise technique for the prediction tions of Ian Macneill. Environmetrics, 10:657-676.
of corrosion rate of steel in con. Mat. and Structures,
29(189):286-294. 26. Ganesan, R., Das, T.K. and Venkataraman, V., 2004.
Wavelet-based multiscale statistical process monitoring:
12. Mariaca, L., Bautista, L., Rodriguez, P. and Gonzalez, a literature review. IIE Transactions, 36:787-806.
J.A., 1997. Use of electrochemical noise for studying
the rate of corrosion of reinforcement embedded in con., 27. Bury, K.V., 1975. Statistical models in applied science.
Materials and Structures, 30(204):613-617. John Wiley Sons, Inc., New York.

JOURNAL OF STRUCTURAL ENGINEERING, VOL. 37, NO. 01, APRIL - MAY 2010 73
28. BIS 2000, Indian standard code of practice for plain and 31. Balaji Rao, K., Anoop, M.B., Lakshmanan, N., Gopala-
reinforced conc.: IS 456-2000. Bureau of Indian Stan- krishnan, S., Appa Rao, T.V.S.R. Risk-based Remaining
dards, New Delhi. Life Assessment of Corrosion Affected Reinforced Con-
crete Structural Members, Jl. of Struct. Engg., Vol. 31,
29. Balaji Rao, K., Anoop, M.B. and Lakshmanan, N. No. 1, 2004, pp 1–15.
An algorithm for detection of change point in on-
line monitoring data. SERC Project Report No. SS-
OLP11641/COR12-RR-2006-4, Structural Engineering 32. Cottis, R.A., Al-Awadhi, M.A.A., Al-Mazeedi, H. and
research Centre, Chennai. Turgoose, S., 2001. Measures for the detection of locali-
zed corrosion with electrochemical noise. Electrochimi-
30. Balaji Rao, K., Anoop, M. B. and Appa Rao T.V.S.R. ca Acta, 46:3665-3674.
A methodology for reliability-based design of concrete
cover thickness with reference to chloride induced cor-
rosion of reinforcement. Proc. of Intl. Conference on 33. Aktan, A.E., Farhey, D.N., Brown, D.L., Dalal, V., Hel-
Civil Engg., ICCE - 2001, Department of Civil Engi- micki, A.J., Hunt, V. and Shelley, S.J., 1996. Condition
neering, Indian Institute of Science, Bangalore, 2001, assessment for bridge management. Jl. of Infrastructure
pp 215–221. Systems, ASCE, 2(3):108-117.

74 JOURNAL OF STRUCTURAL ENGINEERING, VOL. 37, NO. 01, APRIL - MAY 2010

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy