ARTICLE IN PRESS

Reliability Engineering and System Safety 93 (2008) 410–422

www.elsevier.com/locate/ress

Probabilistic models for the extent of damage in degrading

reinforced concrete structures
B. Sudret
Electricité de France, R&D Division, Site des Renardières, F-77818 Moret-sur-Loing, France
Accepted 11 December 2006
Available online 23 January 2007

Abstract

Describing accurately damage in degrading reinforced concrete structures is of major interest in the context of durability analysis and
maintenance. Due to numerous sources of uncertainty in the degradation models, a probabilistic approach is suitable. The probabilistic
description of the extent of damage requires introducing random fields for modelling the spatial variability of the various parameters. In
this paper, a general formulation for the spatial extent of damage is set up. This formulation allows to derive closed-form expressions for
the mean value and standard deviation of the latter. Accordingly, practical computations can be carried out without discretizing the input
fields. In order to check the accuracy of the proposed implementation, Monte Carlo simulation (MCS) of the extent of damage is also
carried out, using an efficient random field discretization technique known as EOLE. Both approaches are compared to study the extent
of rebars corrosion in a RC beam subjected to concrete carbonation. Furthermore, the Monte Carlo approach allows to compute the full
probabilistic content on the extent of damage, e.g. histograms. It was shown that these histograms have a non-trivial shape, in the sense
that probability spikes exist for the bound values (case of undamaged and fully damaged structures). The influence of the autocorrelation
function of the various input random fields and that of their scale of fluctuation is finally studied.
r 2007 Elsevier Ltd. All rights reserved.

Keywords: Extent of damage; Space-variant reliability; Time-variant reliability; Degradation models; Concrete carbonation; Rebars corrosion; Random
fields; EOLE method

1. Introduction [5–7]. This extent of damage is the natural variable that

characterizes the global state of damage of the structure,
Probabilistic models of concrete degradation have been and that may be used in optimizing maintenance policies
intensively studied in the past 10 years. The most important [5,8–10].
degradation mechanism considered in the literature is the In this paper, a general formulation for spatially variable
corrosion of the rebars due to chloride ingress in the degradation models is proposed. The so-called point-in-
concrete mass or concrete carbonation. This mechanism is space and space-variant reliability problems are recalled in
of utmost importance in the ageing of bridge structures Section 2 [11]. Then the extent of damage is given a proper
that are submitted to deicing salts, or any structure in a definition, from which analytical derivations are carried
marine environment [1–4]. Authors have focused on the out in order to compute its first two statistical moments
prediction of the initiation time for corrosion and/or the (Section 3). Efficient implementations of these formulae
estimation of the residual strength of structures. (based on the first order reliability method (FORM) and
Recent advances in this field have pointed out the Monte Carlo simulation (MCS)) are proposed in Section 4.
necessity of modelling the spatial variability of the model In order to evaluate the accuracy of the analytical
parameters in order to be able to characterize, not only the approach, an alternative framework for the direct estima-
probability of degradation, but also the extent of damage tion of the extent of damage by MCS is proposed. This
requires the use of random field discretization techniques
Tel.: +33 1 60 73 6559; fax: +33 1 60 73 7748. and the post-processing of the simulation results. Both
E-mail address: [email protected]. approaches (called ‘‘analytical’’ and ‘‘field discretization’’

0951-8320/$ - see front matter r 2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ress.2006.12.019
 ARTICLE IN PRESS
B. Sudret / Reliability Engineering and System Safety 93 (2008) 410–422 411

in the sequel) are compared on an application example, Denoting by f Z ðzÞ the joint probability density function of
which considers carbonation-induced corrosion. Finally, random vector Z, the time-dependent probability of failure
the results of MCS allow to plot histograms of the extent of of the structure reads
damage, whose specific shape is studied. Z
Pf ðtÞ ¼ f Z ðzÞ dz (2)
gðz;tÞp0
2. Spatially varying probabilistic degradation models
or equivalently
2.1. A class of degradation models Z
Pf ðtÞ ¼ 1fgðz;tÞp0g ðz; tÞ f Z ðzÞ dz ¼ E½1fgðz;tÞp0g ðz; tÞ, (3)
RM
The degradation of structures in time may be defined in a
broad sense as the loss of certain properties as the result of where 1fgðz;tÞp0g ðz; tÞ is the indicator function of the failure
chemical, physical or mechanical processes, or combina- domain in the space of parameters.
tions thereof. Concrete structures are submitted to many In the context of degradation models, the limit state function
degradation mechanisms, including rebars corrosion due to shall be referred to as the damage criterion in the sequel. As an
chloride ingress or concrete carbonation. example, if it is defined by a fixed threshold D̄ on the damage
The deterministic models for these degradation mechan- measure, the limit state function may be of the form:
isms are usually based on semi-empirical equations that gðZ; tÞ ¼ D̄  DðtÞ ¼ D̄  MðZ; tÞ. (4)
yield a so-called damage measure D (considered here as a
scalar quantity) as a function of parameters z and time: We further suppose that the input parameters Z are not
time-dependent, i.e. that they modelled by random vari-
DðtÞ ¼ Mðz; tÞ. (1) ables and not by random processes. As a consequence, the
Examples of damages measures are: reliability problem (2), which appears time-dependent, is
actually equivalent to a series of time-invariant problems,
 crack width, which may be modelled as a function of the where t is a dummy parameter. Indeed, due to the above
corrosion rate of the rebars, the concrete cover, the assumption and the fact that the damage measure is
rebars diameter, etc. [5], increasing in time, any trajectory gðz0 ; tÞ computed from a
 loss of rebars diameter, which depends on the corrosion realization z0 of the input random vector is monotonically
rate and the time for initiation of corrosion, the latter decreasing with time. The time-dependent aspect of the
being modelled specifically in case of chloride or problem over a time interval ½0; t is taken care of by
carbonation-induced corrosion [12], solving the time-invariant problem at the final instant t.
 fatigue damage due to repeated application of stress Note that the latter assertion may not hold anymore if
cycles onto the structure [13]. some input parameters (e.g. environmental conditions)
were modelled by random processes.
In order to assess the durability of the structure with respect In all but academic cases, the integral in Eq. (2) cannot be
to a given category of damage, a limit value D̄ is usually computed analytically. Thus, numerical methods have to be
prescribed (e.g. maximal acceptable crack width, etc.). Note employed. MCS is a versatile tool that takes advantage of the
that the damage measure in Eq. (1) is an increasing function formulation in Eq. (3), where the probability of failure is
of time. Indeed, the degradation phenomena considered in viewed as the expectation of a function of the parameters. A
this paper are supposed to be irreversible. sample set of input parameters fz1 ; . . . ; zN Sim g is generated
according to f Z ðzÞ and the empirical mean of 1fgðz;tÞp0g ðzj ; tÞ
2.2. Local reliability problem is computed. This approach allows to derive confidence
intervals for the probability of failure.
The model parameters in Eq. (1) are in practice uncertain The main drawback of MCS is that the number of
and should be modelled by random variables with samples N Sim required for a fair prediction becomes
prescribed joint probability density function. In this case, untractable when the probability of failure is small. Thus,
the damage measure becomes random. Assessing the state approximate methods such as FORM/SORM have been
of the structure becomes a reliability problem. developed, see e.g. [14] for a detailed presentation.
Let us denote by Z ¼ fZ 1 ; Z 2 ; . . . ; Z M g the set of
random variables describing the randomness in the 2.3. Space-variant reliability problems
degradation model. The failure criterion under considera-
tion is mathematically represented by a limit state function The above probabilistic degradation model is referred to
gðZ; tÞ defined in the space of parameters at time instant as zero-dimensional, in the sense that it does not involve
t 2 ½0; T in a way such that: any spatial coordinate system attached to the structure. It
thus implicitly assumes a complete homogeneity of the
 gðZ; tÞ40 defines the safe state; degradation all over the structure. In other words, for a
 gðZ; tÞp0 defines the failure state; given realization z0 of the input random vector, the full
 gðZ; tÞ ¼ 0 defines the limit state surface. structure will be either in the safe state (undamaged) or in
 ARTICLE IN PRESS
412 B. Sudret / Reliability Engineering and System Safety 93 (2008) 410–422

the failure state (fully damaged). This is of course a coarse and not directly related to the collapse of the structure, none
simplification of the real world. Moreover, this does not of the above quantities are sufficient to characterize the global
allow to characterize the extent of damage. To address this state of ageing of the structure. Indeed, Eq. (6) is by definition
issue, additional notation shall be introduced. a local quantity (at point x). Eq. (7) refers to the probability
Suppose that the structure under consideration occupies that there is at least one point in subdomain H where the
a volume D  Rd , where d ¼ 1; 2 or 3. The case d ¼ 1 local damage criterion is attained. This probability is likely to
corresponds to modelling beam or arch structures, the case be close to one, without meaning that the structure is close to
d ¼ 2 to plate or shell structures. In order to address the structural failure.
problem of spatial variability, the input random vector in In contrast, the extent of damage is of major interest,
Eq. (2) should be replaced by M multivariate scalar especially for the comparison of maintenance policies, see
random fields gathered in a vector ZðxÞ, where x 2 D is e.g. [5,8].
the spatial coordinate. The probabilistic description of
these fields is yet to be specified. Note that in practice the 3. Extent of damage
following assumptions usually apply:
3.1. Definition
 The spatial variability of certain components of Z is
negligible. They are accordingly modelled as random The extent of damage is defined at each time instant t as
variables. As a consequence, only a small number of the measure of the subdomain of D in which the local
scalar random fields have to be specified in practice. failure criterion is attained:
However, for the sake of simplicity, the most general Z
notation ZðxÞ is kept in this section. EðD; tÞ ¼ 1fgðZðxÞ;tÞp0g ðxÞ dx. (8)
 The random field components are homogeneous fields. D

This is due to the fact that the size of the structure is Note that EðD; tÞ is a scalar random variable since the
usually small compared to the scale of fluctuation of the integral over x is defined for each realization of the input
parameters driving the degradation (e.g. environmental random field, say zðxÞ. It is positive-valued and is by
parameters such as surface chloride or carbon dioxide definition bounded by the volume of the structure in Rd
concentration, etc.). denoted by jDj. Again, due to the monotony of degrada-
tion phenomena, each realization of EðD; tÞ, say eðD; tÞ is a
In the space-variant context, the limit state function in continuously increasing function of time.
Eq. (4) should be replaced by
3.2. Mean and variance
gðZðxÞ; tÞ ¼ D̄  MðZðxÞ; tÞ. (5)
The point-in-space probability of failure is defined in each By taking the expectation of Eq. (8) (i.e. with respect to
x 2 Rd as follows: Z), one gets the following expression for the mean value of
Z the extent of damage:
Pf ðx; tÞ ¼ f ZðxÞ ðzÞ dz ¼ E½1fgðZðxÞ;tÞp0g ðz; tÞ. (6) Z
gðZðxÞ;tÞp0 EðD; tÞ  E½EðD; tÞ ¼ E½1fgðZðxÞ;tÞp0g ðx; tÞ dx. (9)
D
It is computed by freezing x (i.e. replacing the random field
ZðxÞ by the corresponding random vector) and t, and by By comparing the integrand of the above equation with Eq.
applying standard time-invariant reliability methods (6), one gets
(MCS, FORM/SORM, etc.). Note that if the random field Z
ZðxÞ is homogeneous, then the same reliability problem is EðD; tÞ ¼ Pf ðx; tÞ dx. (10)
D
posed at whatever the position of the point x under
consideration. Thus, the point-in-space probability of In case of homogeneous input random field, this integrand
failure is independent of x in this case. is independent of x, as explained above. Thus
The space-variant probability of failure is defined, for EðD; tÞ ¼ Pf ðx0 ; tÞ  jDj ðHomogeneous caseÞ, (11)
any subdomain H  D by [11]:
where the point-in-space probability of failure is computed
Pf ðH; tÞ ¼ Pð9x 2 H; gðZðxÞ; tÞp0Þ at any point x0 2 D.
!
[ The above equation has the following interpretation: the
¼P gðZðxÞ; tÞp0 . ð7Þ proportion of the structure where the damage criterion is
x2H attained (i.e. EðD; tÞ=jDj) is, in the mean, equal to the
This quantity is the ‘‘spatial’’ counterpart of the so-called point-in-space probability of failure. This remark has two
cumulative probability of failure in time-variant reliability important consequences:
problems [15].
When the damage measure is related to serviceability of the  It is not necessary to introduce the complex formalism
structure (e.g. apparition of cracks or rebars loss of diameter) of random fields when one is interested only in the mean
 ARTICLE IN PRESS
B. Sudret / Reliability Engineering and System Safety 93 (2008) 410–422 413

value EðD; tÞ. Only the description of the input random  For a rectangular plate of dimensions ðL1 ; L2 Þ, the
variables gathered in vector Z is required. variance of the extent of damage is
 The mean proportion of the structure that is damaged Z 1Z 1
is independent of the correlation structure of the Var½EðD; tÞ ¼ L21 L22 PðgðZð0; 0Þ; tÞ
input random field ZðxÞ, if the spatial variability is 0 0

modelled. This is a valuable result, since the deter- p0 \ gðZðL1 u; L2 vÞ; tÞp0Þ
mination of the correlation structure is difficult and 2
. . . ð2  2 uÞð2  2 vÞ du dv  EðD; tÞ . ð17Þ
hardly done in practice, due to the lack of data
(the autocorrelation functions and their parameters
being often chosen from ‘‘expert judgment’’, see e.g. The integrals in Eqs. (16), (17) are rather easy to evaluate
[5,8]). since both the integration domain and the integrands are
bounded. A typical Gaussian quadrature rule [17] can be
In order to better capture the probabilistic content of applied, as shown in the next section.
EðD; tÞ, it is useful to study the variance of the extent of
damage. By definition, this quantity reads
3.3. Conclusion
2
Var½EðD; tÞ ¼ E½E2 ðD; tÞ  EðD; tÞ . (12)
In this section, the mean and standard deviation of the
From the definition in Eq. (8) one can write extent of damage have been derived in a closed form. The
Z  Z 
obtained formulae do not require that the random fields
E2 ðD; tÞ ¼ 1fgðZðxÞ;tÞp0g ðx; tÞ dx  1fgðZðxÞ;tÞp0g ðx; tÞ dx
describing the spatial variability of the problem are
Z DZ D
discretized. It has been shown that the mean value of the
¼ 1fgðZðx1 Þ;tÞp0g ðx1 ; tÞ  1fgðZðx2 Þ;tÞp0g ðx2 ; tÞ dx1 dx2 .
extent of damage does not depend on the correlation
D D
ð13Þ structure of the input random fields. In case of homo-
geneous input, it may be computed from a single point-in-
The integrand is equal to one if and only if the limit state space analysis (Eq. (11)).
function takes negative values at both locations x1 and x2 . In case of one-dimensional or two-dimensional-rectan-
Thus (13) may be rewritten as gular structures, the computation of the variance further
Z Z reduces to a single integral over D instead of a double
2
E ðD; tÞ ¼ 1fgðZðx1 Þ;tÞp0\gðZðx2 Þ;tÞp0g ðx1 ; x2 Þdx1 dx2 . integral. These cases are of great practical importance,
D D
since the majority of civil engineering structures may be
ð14Þ
decomposed into beams and plates. Eq. (17) is also
Hence applicable to shell structures (e.g. cooling towers) as soon
Z Z as the geometry of these structures is parametrized by a
E½E2 ðD; tÞ ¼ PðgðZðx1 Þ; tÞ rectangular domain, e.g. fðy; zÞ; y 2 ½0; 2p; z 2 ½0; zmax g.
D D
p0 \ gðZðx2 Þ; tÞp0Þ dx1 dx2 . ð15Þ
4. Numerical implementation
This equation is similar to the results obtained by Koo
and Der Kiureghian [16] for the excursion time in the
As explained above, the mean and standard deviation of
context of first-passage problem in time-variant reliability
the extent of damage may be computed from Eqs. (11),
analysis.
(16), (17) without discretizing the input random fields. The
Here again, the assumption of homogeneity allows to
practical implementation of these equations is described in
simplify the result. Indeed, the integrand in Eq. (15)
the next subsection.
depends only on jx1  x2 j in this case, meaning that it is an
In order to assess the accuracy of the proposed
even function of ðxj1  xj2 Þ; j ¼ 1; . . . ; d. One can prove that
approach, the results have to be compared with the MCS
the above double integral may be reduced to a single
of the extent of damage. For this purpose, the input
integral (see details in Appendix A) and further simplified.
random fields shall be discretized and simulated. This is
For the sake of clarity, results are reported here separately
described in the second next subsection.
for d ¼ 1 and 2.

 For a beam of length L (d ¼ 1, jDj ¼ L), the variance of 4.1. Analytical approach
the extent of damage is
We consider here the computation of mean and standard
Z 1 deviation of the extent of damage (also called damage
Var½EðD; tÞ ¼ L2 PðgðZð0Þ; tÞp0 \ gðZðLuÞ; tÞp0Þ length in the one-dimensional case) by the analytical
0
2
equations Eqs. (11), (16). The extension to (17) is
 ð2  2uÞ du  EðD; tÞ . ð16Þ straightforward.
 ARTICLE IN PRESS
414 B. Sudret / Reliability Engineering and System Safety 93 (2008) 410–422

As shown in the previous section, Eq. (11) reduces to 4.2. EOLE method for random field discretization
solving a time-invariant point-in-space reliability problem.
The FORM may be used for this purpose. In order to compute the mean and standard deviation of
In order to evaluate Eq. (16), note that the probability the extent of damage by MCS, the simulation of spatial
under the integral is nothing but the parallel system failure realizations of the damage criterion is required. This can be
probability associated to the events fgðZð0Þ; tÞp0g and done after a proper discretization of the input random
fgðZðLuÞ; tÞp0g. This probability may be computed using fields. The expansion optimal linear estimation (EOLE)
the FORM method applied to parallel systems: method [18] is used for this purpose in case of Gaussian
random fields (it is assumed that the components of the
PðgðZð0Þ; tÞp0 \ gðZðLuÞ; tÞÞ vector input random field are independent). It may then be
F2 ðbð0; tÞ; bðLu; tÞ; rg ðLu; tÞÞ. ð18Þ also applied to discretize fields obtained by translation, e.g.
lognormal fields. The method is based on the pointwise
In this equation, the following notation is used: regression of the original random field with respect to
selected values of the field, and a compaction of the data by
spectral analysis. The method is now recalled.
 F2 ðx; y; rÞ is the binormal CDF; Let us consider a scalar Gaussian random field HðxÞ
 bð0; tÞ (resp. bðLu; tÞ) is the point-in-space reliability defined by its mean mðxÞ, its standard deviation sðxÞ and its
index at abscissa x ¼ 0 (resp. abscissa x ¼ Lu). Note autocorrelation coefficient function rðx1 ; x2 Þ. Let us con-
that both values are equal when homogeneous fields are sider a grid of points fx1 ; . . . ; xN g in D and denote by v the
considered. random vector fHðx1 Þ; . . . ; HðxN Þg. By construction, v is a
 rg ðLu; tÞ ¼ að0; tÞ  aðLu; tÞ is the dot product of the a- Gaussian vector whose mean value lw and covariance
vectors obtained by FORM. Remember that the a- matrix Rww read
vector is the unit vector in the direction of the design
point in the standard normal space. liw ¼ mðxi Þ, (20)

Eq. (18) is practically evaluated as follows: ðRww Þi;j ¼ Cov½Hðxi Þ; Hðxj Þ ¼ sðxi Þsðxj Þrðxi ; xj Þ. (21)

The optimal linear estimation (OLE) of random variable

 A point-in-space FORM analysis is carried out at x ¼ 0, HðxÞ onto the random vector v reads
replacing the random fields by a set of random variables
HðxÞ ^
HðxÞ ¼ mðxÞ þ R0Hw ðxÞ  R1 (22)
having the same PDF, say Z ð1Þ . ww  ðv  lw Þ,
 Another point-in-space FORM analysis is carried out at
where ð:Þ0 denotes the transposed matrix and RHw ðxÞ is a
x ¼ Lu, replacing the random fields by another set of
vector whose components are given by
random variables having the same PDF, say Z ð2Þ . If
some components Z j are actually random variables (in RjHw ðxÞ ¼ Cov½HðxÞ; wj  ¼ Cov½HðxÞ; Hðxj Þ
other words, fully correlated fields), the same variables
are used in both analysis (Z ð1Þ ð2Þ ¼ sðxÞsðxj Þrðx; xj Þ. ð23Þ
j ¼ Z j ). Otherwise, the
ð2Þ
components of Z corresponding to ‘‘true’’ random
Let us now consider the spectral decomposition of the
fields are correlated to the corresponding components of
covariance matrix Sww :
Z ð1Þ using the random field autocorrelation matrix.
 The reliability indices and a-vectors obtained by both Rww /i ¼ li /i ; i ¼ 1; . . . ; N. (24)
analysis are reported in Eq. (18).
This allows to linearly transform the original vector v:

As a conclusion, the evaluation of Eq. (18) requires two X

N pffiffiffiffi
v ¼ lw þ li xi / i , (25)
FORM analysis. Finally, the integral in Eq. (16) may be
i¼1
evaluated using a Gaussian quadrature formula. The
principle of Gaussian integration is to approximate the where fxi ; i ¼ 1; . . . ; Ng are independent standard normal
integral by a weighted summation: variables. Substituting for (25) in (22) and using (24) yields
    the OLE representation of the field:
Z 1 Z 1 XK
1 þ s ds oi 1 þ si
hðuÞ du ¼ h h , (19) XN
xi 0
0 1 2 2 i¼1
2 2 ^
HðxÞ ¼ mðxÞ þ pffiffiffiffi /  RHðxÞw ðxÞ ðOLE expansionÞ.
i¼1 li i
where fðoi ; si Þ; i ¼ 1; . . . ; Kg are the integration weights and (26)
points [17]. Again, the extension of the latter equation to
two-dimensional problems (d ¼ 2, see Eq. (17)) is straight- The series can be further truncated after rpN terms, the
forward. eigenvalues li being sorted first in descending order. This
 ARTICLE IN PRESS
B. Sudret / Reliability Engineering and System Safety 93 (2008) 410–422 415

yields the EOLE expansion: Type A

Type B
Xr
xi 0 Type C
^
HðxÞ ¼ mðxÞ þ pffiffiffiffi /  RHðxÞw ðxÞ ðEOLE expansionÞ.
i¼1 li i
(27) 0.5
The variance of the error for the EOLE discretization reads

ρ(x)
X
r
1
^
Var½HðxÞ  HðxÞ ¼ s2 ðxÞ  ð/0i  RHðxÞw ðxÞÞ2 . (28)
i¼1
li
0
The above equation allows to check that the grid density
(i.e. number of points N) and the number of terms r are
large enough to attain a prescribed accuracy in the
discretization. Details on how choosing these parameters
are given in [19]. -0.5
-5 0 5
Example. Let us consider a homogeneous univariate ðd ¼ x
1Þ standard normal random field (with zero mean and unit
Fig. 1. Autocorrelation coefficient functions (scale of fluctuation y ¼ 2).
standard deviation) having one of the following auto-
correlation coefficient function:
of x 2 ½0; 10 for the three functions. Obviously the
discretization scheme is not accurate enough for Type A
 Type A: correlation (maximal error of 16.7%), accurate enough for
   Type B (maximal error of 0.1%) and almost exact for Type
x1  x2 

rA ðx1 ; x2 Þ ¼ exp   . (29)
‘A  C (maximal error of 1013 %).

 Type B: 4.3. MCS of the extent of damage

  !
x1  x2 2 Once an input random field, say Z j ðxÞ has been
rB ðx1 ; x2 Þ ¼ exp  . (30)
‘B discretized using EOLE, a realization of the field Z 0j ðxÞ
(i.e. a usual function of x 2 D) is obtained by simulating a
set of r standard normal random variables, say fx01 ; . . . ; x0r g,
 Type C: and substituting for them in Eq. (27):
sinððx1  x2 Þ=‘C Þ Xr
x0i 0
rC ðx1 ; x2 Þ ¼ . (31) Z0j ðxÞ ¼ mðxÞ þ pffiffiffiffi /  RZj ðxÞw ðxÞ. (34)
ðx1  x2 Þ=‘C li i
i¼1
In order to study the influence of the shape of the
The above spatial realizations can then be used to get
autocorrelation function, it is desirable to find an
spatial realizations of the limit state function gðZ 0 ðxÞ; tÞ at
equivalence between the various parameters. The usual
time instant t. Each realization shall be post-processed in
measure is the scale of fluctuation proposed by Vanmarcke
order to compute the realization of the extent of damage
[20]:
Z 1 Eo ðD; tÞ.
In the recent literature on space-variant probabilistic
y¼2 rð0; xÞ dx. (32)
0 damage models, see e.g. [5,6,8], the so-called midpoint
approach (MP), originally proposed in [21], has been used
For the three types of autocorrelation coefficient
to discretize the random fields and post-process the extent
functions given above, the results are
of damage. It is important to recall that this crude
yA ¼ 2‘A , approach is by far less accurate than the EOLE decom-
pffiffiffi position, even when using a large number N MP of elements
yB ¼ p‘B , in the MP (the comparison of accuracy can be found in
[18,19]).
yC ¼ p‘C . (33)
Moreover, the MP also provides a rough estimation of
The three autocorrelation coefficient functions are plotted the extent of damage for each trajectory. Indeed, if N MP
in Fig. 1 for y ¼ 2. The random field is now discretized elements have been used for the discretization, the extent of
over the interval x 2 ½0; 10 using N ¼ 81 points (regular damage for a given realization of the input fields is
grid with stepsize equal to 1080 ¼ 0:125). A number r ¼ 16 estimated by N D =N MP jDj, where N D is the number of
terms is retained in the EOLE expansion. Fig. 2 presents elements where the damage criterion gðZ 0 ðxÞ; tÞ is attained.
the evolution of the variance error (Eq. (28)) as a function As N MP is limited in practice, since it is also equal to the
 ARTICLE IN PRESS
416 B. Sudret / Reliability Engineering and System Safety 93 (2008) 410–422

a b x 10-3
0.18 1.5
Variance error (N = 81 , r = 16)
0.16 Min : 5.1561e-005
Mean : 0.00030012
0.14
Max : 0.0014597

Var [H(x)−Hhat(x)]
Var [H(x)−Hhat(x)]

0.12 1

0.1
Variance error (N = 81 , r = 16)
0.08
Min : 0.096994
0.06 0.5
Mean : 0.13302
0.04 Max : 0.16707

0.02

0 0
0 2 4 6 8 10 0 2 4 6 8 10
x x

c x 10-15
2

1.5

1
Var [H(x)−Hhat(x)]

0.5

-0.5

-1
Variance error (N = 81 , r = 16)
-1.5 Min : -2.2362e-015
Mean : -2.8652e-016
-2
Max : 1.2913e-015
-2.5
0 2 4 6 8 10
x

Fig. 2. Variance of the error of discretization for various autocorrelation coefficient functions ðx 2 ½0; 10; y ¼ 2; N ¼ 81; r ¼ 16Þ.

number of random variables required for discretizing each

field, the above estimate may be crude.
In contrast, the EOLE method, which provides smooth
realizations of the field may be post-processed efficiently.
In the following, the one-dimensional (d ¼ 1; D ¼ ½0; L)
case is presented, the extension to larger dimensions being
straightforward, although quite technical. Each realization
of the damage criterion is stored in an array of size Q þ 1
corresponding to the evaluation of gðZ 0 ðxÞ; tÞ at selected
points:
fgj ¼ gðZ 0 ðjL=QÞ; tÞ; j ¼ 0; . . . ; Qg. (35) Fig. 3. Algorithm for computing the extent of damage by Monte Carlo
The damage length is then obtained by screening this array simulation.
according to the algorithm summarized in Fig. 3. If, for a
given j 2 ½0; Q  1, gj p0 and gjþ1 p0 it is supposed that
the failure criterion is negative for any x 2 ½jL=Q; ðj þ 5. Application example: corrosion induced by concrete
1ÞL=Q and the damage length is increased by L=Q. If g carbonation
changes its sign between jL=Q and ðj þ 1ÞL=Q, a linear
interpolation allows to compute the portion of the latter 5.1. Point-in-space model of carbonation
interval where g is negative. Of course, Q has to be selected
in such a way that the trajectories do not change sign twice Concrete carbonation is a complex physico-chemical
within an interval of the form ½jL=Q; ðj þ 1ÞL=Q, i.e. process that includes the diffusion of CO2 into the gas
consistently with the scales of fluctuation of the input phase of the concrete pores and its reaction with
fields. the calcium hydroxyl Ca(OH)2. The latter can be
 ARTICLE IN PRESS
B. Sudret / Reliability Engineering and System Safety 93 (2008) 410–422 417

simplified into
CaðOHÞ2 þ CO2 !CaCO3 þ H2 O. (36)
As the high pH of uncarbonated concrete is mainly due
to the presence of CaðOHÞ2 , it is clear that the consumption
of this species will lead to a pH drop, which can attain a
value of 9 when the reaction is completed. In this
environment, the oxide layer that protected the reinforce-
ment bars is attacked and corrosion starts. The corrosion
products tend to expand into the pores of concrete,
developing tensile stresses which eventually lead to crack-
ing [22–24].
In practice, CO2 penetrates into the concrete mass by
diffusion from the surface layer. Thus, a carbonation front
appears that moves into the structure. A model for Fig. 4. Evolution in time of the rebars diameter and associated state of
computing the carbonation depth xc is proposed by the damage.
CEB Task Groups 5.1 & 5.2 [25]. The simplified version
retained in the present paper reads corrosion current density icorr :
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2C 0 DCO2 Z ¼ fDCO2 ; C 0 ; a; e; f0 ; icorr g. (40)
xc ðtÞ ¼ t, (37)
a The damage criterion is defined at a given time instant by
where DCO2 is the coefficient of diffusion of carbon dioxide the fact that the residual rebars diameter (Eq. (39))
in dry concrete, C 0 is the carbon dioxide concentration in becomes smaller than a prescribed fraction ð1  lÞ of its
the surrounding air and a is the binding capacity, i.e. the initial value:
amount of carbon dioxide necessary for complete carbona-
gðZ; tÞ ¼ fðtÞ  ð1  lÞf0 . (41)
tion of a concrete volume. It is supposed that corrosion
immediately starts when carbonation has attained the Remarking that Eq. (39) rewrites
rebars. Denoting by e the concrete cover, the time
fðtÞ ¼ minðf0 ; f0  2icorr kðt  T init ÞÞ (42)
necessary for corrosion to start, called initiation time, reads
for reasonable values of t that do not lead to ‘‘negative’’
ae2
T init ¼ . (38) values of fðtÞ, Eq. (41) becomes
2C 0 DCO2
gðZ; tÞ ¼ minðlf0 ; lf0  2icorr kðt  T init ÞÞ. (43)
If generalized corrosion is considered, the loss of metal due
to corrosion is approximately uniform over the whole Thus, the probability of failure may be interpreted as that
surface. In this case, Faraday’s law indicates that a unit of a series system:
corrosion current density (or corrosion rate) corresponds Pf ðtÞ ¼ Pðfl f0 p0g [ flf0  2icorr kðt  T init Þp0gÞ. (44)
to a uniform corrosion penetration of k ¼ 11:6 mm=year. If
a constant annual corrosion rate is supposed, the expres- As the rebars diameter is positive in nature, the above
sion of the rebars diameter as a function of time eventually system event reduces to its second component. Thus
reads Pf ðtÞ ¼ Pðlf0  2icorr kðt  T init Þp0Þ. (45)
(
f0 if tpT init Following these remarks and using (38), the limit state
fðtÞ ¼ (39)
max½f0  2icorr kðt  T init Þ; 0 if t4T init : function (41) eventually reads
!
Fig. 4 sketches the evolution in time of the rebars ae2
diameter together with the associated damage on the gðZ; tÞ ¼ lf0  2icorr k t  (46)
2C 0 DCO2
structure. From experimental evidence [26,27], it is possible
to associate a value l (representing the relative loss of
rebar’s diameter) to a given state of damage (i.e. crack 5.2. Probabilistic problem statement
initiation, severe cracking, spalling, etc.). For instance, a
value of l ¼ 0:521% is consistent with the apparition of In order to illustrate the concept of extent of damage
cracks. described above, we consider a concrete beam of length
In a probabilistic context, the random parameters are L ¼ 10 m. This beam is reinforced by a single longitudinal
those appearing in Eqs. (38), (39), namely the coefficient of reinforcing bar whose initial diameter is modelled by a
diffusion of carbon dioxide DCO2 , the surface carbon lognormal random variable f0 . The concrete cover eðxÞ is a
dioxide concentration C 0 , the binding capacity a, the univariate homogeneous lognormal random field, obtained
concrete cover e, the rebars initial diameter f0 and the by exponentiation of a Gaussian random field, whose
 ARTICLE IN PRESS
418 B. Sudret / Reliability Engineering and System Safety 93 (2008) 410–422

properties are given below. This allows to model the the various approaches is observed. The maximal discre-
imperfections in placing the rebar into the falsework. pancy between FORM and the field discretization
The parameters fC 0 ; ag are modelled by lognormal approach is less than 5%, and less than 2% for tX40
random variables. The coefficient of diffusion DCO2 is years. Part of this discrepancy is due to the linearization in
modelled as a homogeneous lognormal random field. The FORM: indeed, if MCS is used for evaluating the point-in-
corrosion current density icorr ðxÞ is supposed to be inversely space probability of failure instead of FORM, the latter
proportional to the concrete cover (see [6]): discrepancy reduces to 3%, and less than 0.5% for tX40
e0 years.
icorr ðxÞ ¼ i0corr . (47) The standard deviation of the extent of damage is
eðxÞ
computed by three approaches as well:
In this equation, e0 ¼ 50 mm is the mean concrete cover
and i0corr is a lognormal random variable. The above  Eq. (16) where the system reliability problem under the
expression has the following interpretation: variable i0corr integral is solved by FORM.
models the overall uncertainty on the corrosion rate,  Eq. (16) where this problem is solved by MCS (100,000
whereas its spatial variability is perfectly correlated to that
samples were used).
of the concrete cover. The mean value of the corrosion rate
 The field discretization approach.
is 2:08 mA=cm2 . The parameters describing these six input
quantities (two random fields and four random variables)
The standard deviation of the extent of damage obtained
are shown in Table 1.
by the three approaches is plotted in Fig. 6 (the values are
also shown in Table 2, columns #5–7). Here again, the
5.3. Mean and variance of the extent of damage agreement between the various approaches is excellent. The
maximal discrepancy between the analytical and the field
The mean value of the damage length is computed by discretization approaches is less than 9%, and about 0.2%
three approaches: for tX40 years. The accuracy of the analytical formula
(Eq. (16)) increases with larger time horizons. The use of
 Eq. (11), where the point-in-space probability of failure,
obtained by freezing the spatial coordinate x (i.e.
replacing the random fields eðxÞ and DCO2 ðxÞ by 2
Random Field (MCS)
lognormal random variables) is computed by FORM 1.8 Analytical (MCS)
analysis. Analytical (FORM)
1.6
 Eq. (11), where the point-in-space probability of failure
Mean extent of damage

is obtained by MCS (100,000 samples were used). 1.4

 The field discretization approach described in Section 1.2
4.3 and Fig. 3. The EOLE discretization of both input
fields was carried out using a regular grid consisting in 1
N ¼ 81 points over ½0; 10 m. A number r ¼ 16 points 0.8
was retained in the spectral decomposition (Eq. (27)).
0.6
This allows to get a maximal relative discretization error
on the field variance of 0.1% (as shown in Fig. 2(b)). A 0.4
total number of 10,000 spatial realizations of the limit 0.2
state function was used.
0
10 20 30 40 50 60
The mean damage length obtained by the three
Time (years)
approaches is plotted in Fig. 5 (the values are also shown
in Table 2, columns #2–4). An excellent agreement between Fig. 5. Evolution in time of the mean extent of damage.

Table 1
Probabilistic input data

Parameter Type of PDF Mean value Coef. var. (%) A.c.fa

Rebars diameter f0 Lognormal 10 mm 10 –

Diffusion coefficient DCO2 Lognormal 5  108 m2 =s 30 rD ðx1 ; x2 Þ
Surface concentration C 0 Lognormal 6:2  104 kg=m3 30 –
Binding capacity a Lognormal 80 kg=m3 30 –
Nominal corrosion rate i0corr ðxÞ Lognormal 1 mA=cm2 25 –
Concrete cover eðsÞ Lognormal 50 mm 20 re ðx1 ; x2 Þ
2 =y2 2 =y2
a
Autocorrelation coefficient function of the underlying Gaussian field rD ðx1 ; x2 Þ ¼ epðx1 x2 Þ D, yD ¼ 2 m; re ðx1 ; x2 Þ ¼ epðx1 x2 Þ e, ye ¼ 2 m.
 ARTICLE IN PRESS
B. Sudret / Reliability Engineering and System Safety 93 (2008) 410–422 419

Table 2 (Pentium M processor at 1.6 GHz, 512 MB RAM) to get

Mean and standard deviation of the extent of damage—comparaison of the curves in Figs. 5, 6, whereas the analytical approach
approaches
(FORM and MCS) requires less than 2 min.
Time Mean value Standard deviation
(years) 5.4. Histogram of the extent of damage
Eq. (11) Eq. (11) Random Eq. (16) Eq. (16) Random
field field
MCS FORM MCS MCS FORM MCS
From the MCSs, it is possible to plot histograms of the
extent of damage, see Fig. 7. It is first observed that there is
20 0.038 0.039 0.037 0.239 0.245 0.224 always a non-zero number of spatial realizations of the
22 0.062 0.061 0.058 0.310 0.321 0.294 limit state function that are strictly positive, meaning that
24 0.088 0.091 0.089 0.376 0.406 0.382
the associated realization of the damage length eð½0; L; tÞ is
26 0.125 0.128 0.129 0.479 0.498 0.482
28 0.171 0.174 0.175 0.567 0.597 0.582 exactly equal to zero. In other words, there is a probability
30 0.228 0.228 0.230 0.677 0.701 0.686 spike on zero, as shown in Fig. 7. This spike is all the
32 0.296 0.291 0.295 0.781 0.810 0.799 greater since the time instant is smaller (the spikes in the
34 0.365 0.363 0.370 0.902 0.921 0.914 figure are not on scale). Similarly, when time increases (e.g.
36 0.450 0.442 0.454 1.016 1.035 1.034
t ¼ 60 years), another spike appears for eð½0; L; tÞ ¼ L.
38 0.537 0.530 0.543 1.134 1.149 1.151
40 0.638 0.626 0.640 1.245 1.264 1.265 This represents cases where the beam is fully damaged.
42 0.740 0.728 0.743 1.350 1.377 1.379 It clearly appears in Fig. 7 that the damage length has a
44 0.849 0.837 0.852 1.456 1.489 1.492 complex PDF. Thus, its approximation by a Gaussian
46 0.961 0.952 0.967 1.578 1.598 1.599 PDF (as e.g. in [6]), due to the application of the central
48 1.084 1.073 1.088 1.685 1.705 1.705
limit theorem on the summation of independent damaged
50 1.211 1.199 1.214 1.781 1.809 1.808
52 1.341 1.329 1.344 1.883 1.909 1.906 areas, is probably very rough. This is due to the fact that:
54 1.478 1.462 1.478 1.972 2.005 2.003
56 1.616 1.600 1.614 2.066 2.097 2.093  it is impossible to discretize the structure into really
58 1.755 1.739 1.753 2.155 2.184 2.181 independent subdomains, as soon some input parameters
60 1.896 1.882 1.898 2.251 2.267 2.264
are modelled by random variables (i.e. spatially perfectly
correlated);
 the extent of damage is by definition bounded by 0 and
2.5 jDj.
Random Field (MCS)
Standard deviation of the extent of damage

Analytical (MCS)
Analytical (FORM) 5.5. Probability of no corrosion
2

The probability of having the beam completely sound at

1.5 a given time instant ðPðEð½0; L; tÞ ¼ 0ÞÞ is an interesting
information to assess the state of damage in time. In the
case of uni-dimensional problems ðd ¼ 1) as the current
1 example, a closed-form expression for this quantity can be
derived. Indeed

0.5
PðEð½0; L; tÞ ¼ 0Þ ¼ 1  PðEð½0; L; tÞ40Þ
¼ 1  Pð9x 2 ½0; L; gðZðxÞ; tÞp0Þ
!
[
0 ¼1P gðZðxÞ; tÞp0 , ð48Þ
10 20 30 40 50 60
x2½0;L
Time (years)
where the right-hand side is nothing but the space-variant
Fig. 6. Evolution in time of the standard deviation of the extent of probability of failure defined in Eq. (7).
damage. This space-variant problem can be considered as a
spatial first passage problem, and thus solved by techniques
developed in time-variant reliability analysis [11]. The
MCS for evaluating the system probability of failure
spatial outcrossing rate from the undamaged state to the
(instead of FORM) does not help much. Indeed, part of the
damage state is defined at each time instant by
residual error is due to the approximation in the
quadrature of the integral [16]. nþ ðx; tÞ ¼ lim PðgðZðxÞ; tÞ40 \ gðZðx þ hÞ; tÞp0Þ=h.
h!0
The various algorithms to compute the mean and
standard deviation of the extent of damage are implemen- (49)
ted in MathCad. In terms of efficiency, the field discretiza- In case of homogeneous problems, this quantity is
tion approach requires about 5 h on a standard PC independent of x. If the occurrence of spatial outcrossing
 ARTICLE IN PRESS
420 B. Sudret / Reliability Engineering and System Safety 93 (2008) 410–422

10 years 20 years
1000 Probability spike 1000 Probability spike

400 400

0 0
0 2 4 6 8 10 0 2 4 6 8 10
m m

30 years 40 years
1000 Probability spike 1000 Probability spike

400 400

0 0
0 2 4 6 8 10 0 2 4 6 8 10
m m

50 years 60 years
1000 Probability spike 1000 Probability spike

400 400 Probability spike

0 0
0 2 4 6 8 10 0 2 4 6 8 10
m m

Fig. 7. Histogram of the extent of damage (m) at various time instants (10,000 samples).

is considered as a Poisson process [14, Chapter 6], the Table 3

space-variant probability of failure may be estimated in the Probability of no corrosion in the beam ðPðEð½0; L; tÞ ¼ 0ÞÞ
homogeneous case by Time (years) MCS Eq. (51)
nþ ðx0 ;tÞ
PðEð½0; L; tÞ40Þ 1e , (50) 20 0.960 0.956
22 0.939 0.935
where the outcrossing rate is computed at any point x0 2 24 0.913 0.910
½0; L due to homogeneity. Hence the probability of no 26 0.886 0.880
corrosion: 28 0.857 0.847
30 0.823 0.812
þ ðx ;tÞ
PðEð½0; L; tÞ ¼ 0Þ en 0 (51) 32 0.787 0.774
34 0.750 0.736
The outcrossing rate may be evaluated by the PHI2 method 36 0.715 0.698
[15,28] which is based on the solution of the parallel system 38 0.681 0.660
reliability problem in Eq. (49), see the references for details. 40 0.642 0.623
42 0.607 0.587
Table 3 shows the probability of no corrosion at various 44 0.572 0.553
time instants, computed either by the field discretization 46 0.535 0.522
approach (fraction of the number of spatial realizations of 48 0.502 0.492
the limit state function for which no corrosion is observed) 50 0.470 0.464
or by Eq. (51). 52 0.437 0.438
54 0.409 0.414
It appears that the analytical approximate formula 56 0.382 0.393
provides an accurate estimation of the probability of 58 0.355 0.373
having no corrosion, the discrepancy compared to the 60 0.330 0.355
MCS results being less than 6%. However, it is expected
that the accuracy will decrease when time increases, due to
due to lack of data. In this section, the influence of the
the fact that the Poissonian assumption on the outcrossing
shape of the autocorrelation coefficient function rD ðx1 ; x2 Þ
becomes incorrect when many (possibly correlated) out-
is studied, as well as that of the scale of fluctuation. The
crossings occur.
description of the random field modelling the concrete
cover is the same as in Table 1. As shown in Section 3.2,
5.6. Influence of the autocorrelation coefficient function changing rD ðx1 ; x2 Þ will not change the expected extent of
damage, but rather influence the variance of the latter.
As mentioned above, the autocorrelation coefficient As the analytical approach described above allows to
function for the input random fields is not well known compute the standard deviation of the extent of damage in
 ARTICLE IN PRESS
B. Sudret / Reliability Engineering and System Safety 93 (2008) 410–422 421

3 The mean and standard deviation of the extent of damage

Type A
obtained by the various approaches compare very well with
Standard deviation of the extent of damage

2.8 Type B
each other. The computational cost of the analytical
2.6 approach is about two orders of magnitude smaller than
2.4 that of the direct MCS.
From the histograms of the extent of damage, it appears
2.2
that probability spikes corresponding to ‘‘fully sound’’ or
2 ‘‘fully damaged’’ structures exist. The approximation of the
1.8
extent of damage by a Gaussian random variable is thus
proven to be irrelevant in this context. These histograms
1.6 may be used in the context of reliability analysis, e.g. to
1.4 compute the probability that the extent of damage is
greater than a given value, say 5% of the original length (or
1.2
surface) of the structure. The existence of probability
1 spikes shall warn the analyst that classical reliability
0 2 4 6 8 10 methods such as FORM/SORM may be inappropriate
Scale of fluctuation (m) for solving this problem.
Fig. 8. Influence of the shape of the autocorrelation coefficient function Of particular interest in reliability analysis is the problem
and the scale of fluctuation of DCO2 . of computing the probability of having no damage along
the structure ðEðD; tÞ ¼ 0Þ. A technique for computing
directly this quantity is finally proposed in the case of one-
a matter of seconds, it is possible to use it for carrying out a dimensional problems (see Application example). This
parametric study. As an illustration, the standard deviation technique, inspired by time-variant reliability methods,
of the extent of damage at t ¼ 60 years is considered here. does not require the discretization of the input random
The evolution of this quantity as a function of the scale of fields either.
fluctuation of the coefficient of diffusion DCO2 is plotted in Although illustrated on a simple degradation model, the
Fig. 8 for two shapes of autocorrelation coefficient proposed approach is quite general and could be applied
functions (Types A, B), see Eqs. (29)–(30). efficiently to chloride contaminated structures such as
From this figure, it appears that the standard deviation is bridge decks.
rather insensitive to the parameters. Indeed, the value
varies between 6% and þ11% with respect to the original
scenario presented in Section 5.3 (Type B, yD ¼ 2 m). This Acknowledgments
is an interesting result since the shape and scale parameters
of the autocorrelation coefficient functions are difficult to The author would like to thank Mr G. Defaux and Dr
infer in practice due to lack of data. M. Pendola (Phimeca Engineering S.A.) for fruitful
discussions on space-variant reliability problems. These
6. Conclusions discussions were made possible through the joint research
agreement between Electricité de France, the French
The characterization of the extent of damage in Institute for Advanced Mechanics (IFMA, Pr. Lemaire)
probabilistic degradation models of structures is of major and Phimeca Engineering S.A.
interest, especially for optimizing maintenance policies.
This requires describing the parameters of the degradation
Appendix A. Proof of Eqs. (16), (17)
models as random fields.
The paper presents analytical derivations for the mean
One-dimensional case: D ¼ ½0; L. Suppose f ðx1 ; x2 Þ 
and standard deviation of the extent of damage. The
f~ðx1  x2 Þ, where f~ðzÞ is an even function of its argument.
former appears independent from the correlation structure RL RL
of the input fields. Both quantities may be computed by Assume the integral I 1 ¼ 0 0 f ðx1 ; x2 Þ dx1 dx2 exists and
analytical formulae without any discretization of the is to be computed. The following mapping is used:
random fields. x1 þ x2
In order to assess the accuracy of the formulae, an u¼  1,
L
efficient Monte Carlo-based framework for computing
spatial realizations of the extent of damage is proposed, x2  x1
v¼ . (52)
based on the EOLE discretization method. On top of mean L
and standard deviation, the direct MCS of the extent of The integral rewrites
damage also yields histograms of the latter. Z Z
L2 1 aðvÞ=2 ~
All the methods are illustrated on the example of a RC I1 ¼ f ðLvÞ du dv. (53)
beam submitted to carbonation-induced rebars corrosion. 2 1 aðvÞ=2
 ARTICLE IN PRESS
422 B. Sudret / Reliability Engineering and System Safety 93 (2008) 410–422

Since the integrand does not depend anymore on u, the second international forum on engineering decision making (IFED),
integration with respect to u provides aðvÞ. Moreover, due Lake Louise, Canada; 2006.
to the fact that f~ is even, the integral with respect to v is [10] Li Y. Effect of spatial variability on maintenance and repair decisions
for concrete structures. PhD thesis, Delft University of Technology;
twice that computed over ½0; 1. Finally, for vX0, it is easy 2004.
to show that aðvÞ ¼ 2  2v. Thus [11] Der Kiureghian A, Zhang Y. Space-variant finite element reliability
Z 1 analysis. Comput Meth Appl Mech Eng 1999;168:173–83.
I 1 ¼ L2 f~ðL vÞð2  2vÞ dv. (54) [12] Sudret B, Defaux G, Pendola M. Introducing spatial variability in the
0 lifetime assessment of a concrete beam submitted to rebar’s
corrosion. In: Proceedings of the second international forum
Two-dimensional case: D ¼ fðx; yÞ 2 ½0; L1   ½0; L2 g. on engineering decision making (IFED), Lake Louise, Canada;
Suppose f ðx1 ; y1 ; x2 ; y2 Þ  f~ðx1  x2 ; y1  y2 Þ, where 2006.
~
f ðx; yÞ is an even function of its arguments. The integral [13] Petryna YS, Krätzig WB. Computational framework for long-term
to be computed is of the form: reliability analysis of RC structures. Comput Meth Appl Mech Eng
Z L1 Z L1 Z L2 Z L2 2005;194:1619–39.
[14] Melchers R-E. Structural reliability analysis and prediction. New
I2 ¼ f~ðx1  x2 ; y1  y2 Þ dx1 dx2 dy1 dy2 . York: Wiley; 1999.
x1 ¼0 x2 ¼0 y1 ¼0 y2 ¼0
[15] Andrieu-Renaud C, Sudret B, Lemaire M. The PHI2 method: a way
(55) to compute time-variant reliability. Reliab Eng Syst Safety
2004;84:75–86.
The mapping in Eq. (52) can be applied on each coordinate [16] Koo H, Der Kiureghian A. FORM, SORM and simulation
ðx; yÞ. Thus techniques for nonlinear random vibrations. Technical Report no.
Z 1Z 1 UCB/SEMM-2003/01, University of California at Berkeley: 2003.
2 2
I 2 ¼ L1 L2 f~ðL1 v1 ; L2 v2 Þð2  2v1 Þð2  2v2 Þ dv1 dv2 . 185pp.
0 0 [17] Abramowitz M, Stegun IA, editors. Handbook of mathematical
(56) functions. New York: Dover Publications, Inc.; 1970.
[18] Li CC, Der Kiureghian A. Optimal discretization of random fields. J
Eng Mech 1993;119(6):1136–54.
References [19] Sudret B, Der Kiureghian A. Stochastic finite elements and reliability:
a state-of-the-art report. Technical Report no UCB/SEMM-2000/08,
[1] Engelund S, Sorensen J. A probabilistic model for chloride-ingress University of California, Berkeley; 2000, 173pp.
and initiation of corrosion in reinforced concrete structures. Struct [20] Vanmarcke E. Random fields: analysis and synthesis. Cambridge,
Saf 1998;20:69–89. MA: The MIT Press; 1983.
[2] Stewart M, Rosowsky D. Time-dependent reliability of deteriorating [21] Der Kiureghian A, Ke J-B. The stochastic finite element method in
reinforced concrete bridge decks. Struct Saf 1998;20:91–109. structural reliability. Probab Eng Mech 1988;3(2):83–91.
[3] Val D, Stewart M, Melchers R. Effect of reinforcement corrosion on [22] Liu Y, Weyers RE. Modelling the time-to-corrosion cracking in
reliability of highway bridges. Eng Struct 1998;20:1010–9. chloride contaminated reinforced concrete structures. ACI Mater J
[4] Vu AT, Stewart M. Structural reliability of concrete bridges including 1998;95:675–81.
improved chloride-induced corrosion models. Struct Saf [23] Thoft-Christensen P. FEM modelling of the evolution of corrosion
2000;22:313–33. cracks in reinforced concrete structures. In: Maes M, Huyse L,
[5] Li Y, Vrouwenvelder T, Wijnants T, Walraven J. Spatial variability editors. Proceedings of the 11th IFIP WG7.5 conference on reliability
of concrete deterioration and repair strategies. Struct Concr and optimization of structural systems, Banff, Canada; 2004. p.
2004;5:121–30. 221–8.
[6] Vu KAT, Stewart MG. Predicting the likelihood and extent of [24] Bhargava K, Ghosh AK, Mori Y, Ramanujam S. Model for cover
reinforced concrete corrosion-induced cracking. J Struct Eng ASCE cracking due to rebar corrosion in RC structures. Eng Struct
2005;131(11):1681–9. 2006;28(8):1093–109.
[7] Darmawan MS, Stewart M. Spatial time-dependent reliability [25] Coll. New approach to durability design—an example for carbona-
analysis of corroding pretensioned prestressed concrete bridge tion induced corrosion. Technical Report no. 238, Comité Euro-
girders. Struct Saf 2007;29(1):16–31. international du béton; 1997.
[8] Stewart MG. Spatial variability of damage and expected maintenance [26] Alonso C, Andrade A, Rodriguez J, Diez JM. Factors controlling
costs for deteriorating RC structures. Struct Infrastruct Eng cracking of concrete affected by reinforcement corrosion. Mater
2006;2(2):79–90. Struct 1998;31:435–41.
[9] Stewart MG, Mullard JA, Drake BJ. Utility of spatially variable [27] Broomfield P. Corrosion of steel in concrete. E&FN Spon; 1997.
damage performance indicators for improved safety and maintenance [28] Sudret B. Analytical derivation of the outcrossing rate in time-variant
decisions of deteriorating infrastructure. In: Proceedings of the reliability problems. Struct Infrastruct Eng 2007, to appear.