Reliability Engineering and System Safety 87 (2005) 377385

www.elsevier.com/locate/ress

Decision analysis for deteriorating structures

Dimitri V. Vala,*, Mark G. Stewartb
a

Department of Structural Engineering and Construction Management, Faculty of Civil and Environmental Engineering, TechnionIsrael Institute
of Technology, Haifa 32000, Israel
b
Centre for Infrastructure Performance and Reliability, School of Engineering, The University of Newcastle, Newcastle, NSW 2308, Australia
Received 7 February 2003; received in revised form 30 July 2003; accepted 28 June 2004
Available online 11 September 2004

Abstract
Measures that improve durability of a structure usually increase its initial cost. Thus, in order to make a decision about a cost-effective
solution the life-cycle cost of a structure including cost of structural failure needs to be considered. Due to uncertainties associated with
structural properties, loads and environmental conditions the cost of structural failure is a random variable. The paper derives probability
distributions of the cost of failure of a single structure and a group of identical structures when single or multiple failures are possible during
the service life of a structure. The probability distributions are based on cumulative probabilities of failure of a single structure over its service
life. It is assumed that failures occur at discrete points in time, the cost of failure set at the time of decision making remains constant for a
particular design solution and the discount rate is a deterministic parameter not changing with time. The probability distributions can be
employed to evaluate the expected life-cycle cost or the expected utility, which is then used in decision making. An example, which considers
the selection of durability specifications for a reinforced concrete structure built on the coast, illustrates the use of the derived probability
distributions.
q 2004 Elsevier Ltd. All rights reserved.
Keywords: Decision making; Durability; Life-cycle cost; Utility; Uncertainty

1. Introduction
Structures deteriorate with time if subject to aggressive
environments. As a result of deterioration owners of aging
structures need to allocate significant ever increasing
financial resources for their maintenance, repair or replacement. Improving the durability of new structures can reduce
costs associated with their future service life. However,
measures that improve durability usually increase the initial
cost of a structure. Thus, in order to make a decision about a
cost-effective solution the life-cycle cost of a structure
needs to be considered, i.e. all costs associated with design,
construction, maintenance and possible failure modes of the
structure.
An assessment of the life-cycle cost of a structure in an
aggressive environment is associated with significant
* Corresponding author. Tel.: C972-4-829-2276; fax: C972-4-8295697.
E-mail address: [email protected] (D.V. Val).
0951-8320/$ - see front matter q 2004 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ress.2004.06.006

uncertainty. Sources of the uncertainty include structural

properties, loads and environmental conditions, and their
effects on structural properties (e.g. fatigue, deterioration
processes such as corrosion, alkali-silica reaction, etc.). As a
result the life-cycle cost is a random variable and its
expected value serves as a criterion for decision making
(e.g. [14]). This expected life-cycle cost decision criterion
is based on the classical von NeumannMorgenstern
decision theory [5] under the assumption that utility is
simply expressed in monetary values, i.e. minimizing the
expected life-cycle cost is equivalent to maximizing the
expected utility. However, the simple replacement of
utilities by monetary values may not be appropriate when
consequences for financial well being of the owner of a
structural asset differ significantly depending on the amount
of money the owner will need to spend on the structure over
its service life. For example, the owner of a structure
allocates a certain amount of money for its design,
construction and maintenance and if life-cycle cost of the
structure will exceed this amount the owner expects serious

378

D.V. Val, M.G. Stewart / Reliability Engineering and System Safety 87 (2005) 377385

financial problems, including possible bankruptcy. In this

situation the owner may assign quite different utilities for
monetary values below and above of the allocated amount
so that there will no longer be linear proportionality between
utilities and monetary values. In this case a decision based
on the minimum expected life-cycle cost is no longer
equivalent to that based on the maximum expected utility
and does not reflect the owners preferences. In order to
estimate the expected utility the whole distribution of the
life-cycle cost needs to be known as well as a utility function
relating utilities with the corresponding life-cycle cost
values.
The present paper considers probability distributions of
the life-cycle cost of a single structure and of a group of
similar structures when either single failure or multiple
failures are possible over the service life of a structure. The
distributions are estimated based on the cumulative
probabilities of failure of a single structure over its service
life. The use of these distributions for decision making is
illustrated by an example, which estimates limits on the cost
of durability improvement measures for reinforced concrete
structures in marine environment by maximizing expected
utility. Results are obtained for different utility functionsa
linear function (i.e. utilities are linearly related to life-cycle
costs and minimizing the expected life-cycle cost is
equivalent to maximizing the expected utility) and two
bilinear functions representing risk averse and risk prone
attitudes (in this case minimizing the expected life-cycle
cost is not equivalent to maximizing the expected utility).
Comparison of the results also illustrates the effect of
different utility functions on decision making.

where cF is the cost of failure set at the time of decision

making, t the time of failure, and r the discount rate. The
discount rate used in this study is the real discount rate,
which excludes the rate of general price inflation. Discount
rates are influenced by a number of economic, social and
political factors and can be quite variable (more information
on the selection of values of the discount rate can be found
elsewhere [6]).
Due to uncertainties associated with structural properties,
loads and environmental conditions structural failures
are random events with time-dependent probabilities of
occurrence. It is typical to consider failures at discrete
points in time so that their probabilities are equal to the
cumulative probability of failure over a corresponding time
interval, e.g. one failure event per year with a corresponding
annual probability. Thus, CF(T) is a discrete random
variable which at failure times ti assumes different values,
ci, given by Eq. (3), i.e.
ci Z

cF
1 C rti

(4)

with probabilities of occurrence, pi. Subsequently, LCC(T)

becomes a discrete random variable as well. Decisions are
then often based on the expected life-cycle cost, E[LCC(T)]
ELCCT Z CC0 C CDI C ECF T

(5)

where E[CF(T)] is the expected cost of failure. For example,

for a single structure, which can fail only once during T
years of service, and when cF is assumed the same for all
possible failure modes E[CF(T)] is estimated as
ECF T Z

M
X

pi c i

(6)

iZ1

2. Life-cycle cost and utility

Generally, to compare different alternatives with the
same benefits the life-cycle cost of a structure up to time T,
LCC(T), can be expressed as
LCCT Z CD C CC C CIN T C CM T C CF T

(1)

where CD is the design cost, CC the construction cost

(materials and labor), CIN(T) the cost of inspections, CM(T)
the cost of maintenance, and CF(T) the cost of failure
(damages, cost of life, injury). If to assume that CD, CIN(T)
and CM(T) are similar for different alternatives and that
CCZCC0CCDI, where CC0 is the lowest construction cost
among considered alternatives and CDI the cost of durability
improvements, Eq. (1) may be re-written as
LCCT Z CC0 C CDI C CF T

(2)

Failures of different alternatives may occur at different

times so in order to obtain consistent results costs of failure
are discounted to a present value (e.g. [6]), i.e.
cF
CF T Z
(3)
1 C rt

where M is the number of points in time at which the

possibility of failure occurrence is considered. An alternative with the minimum expected life-cycle cost is then
selected as the optimal alternative.
When utilities can be replaced by monetary (i.e. cost)
values the decision criterion using the minimum expected
life-cycle is based on the classical von NeumannMorgenstern decision theory [5]. In this case the minimum expected
life-cycle cost criterion is equivalent to the maximum
expected utility criterion of the von NeumannMorgenstern
theory. However, in general the relationship between
utilities and monetary values depends on the attitude of
the owner of a structure towards risk (i.e. losses associated
with the structure maintenance over its service life including
failure). Depending on the owner attitude towards risk the
utility function, which represents the relationship between
utilities and monetary values, can be concave (risk averse
attitude), convex (risk prone attitude), or linear (risk neutral
attitude when utilities are equivalent to monetary values).
The following simple example illustrates the effect of the
shape of the utility function on decisions concerning the
choice of durability improving measures. A decision maker

D.V. Val, M.G. Stewart / Reliability Engineering and System Safety 87 (2005) 377385

Fig. 1. Decision tree of example problem.

has to choose between two alternativesa1 with the

construction cost of CC1ZCC0 and the probability of failure
of Pf1Z0.6, and a2 with improved durability performance
so that CC2ZCC0CCDI and Pf2Z0.1. It is assumed that the
cost of failure, cFi, is the initial construction cost, CCi, plus
the cost of demolition and other direct costs assumed equal
to CC0 such that cFiZCCiCCC0 (iZ1,2) and discount rate is
not taken into account. The decision tree of the problem is
shown in Fig. 1, where u($) denotes utility (all costs are
equivalent to losses and thus taken as negative).
The minimum possible value of the life-cycle cost is CC0
(i.e. the cheapest alternative, no failure) and the corresponding utility u(KCC0)Z1, while the maximum possible lifecycle cost is 3CC0C2CDI (i.e. the highest cost alternative,
failure) and the corresponding u(K3CC0K2CDI)Z0. Three
hypothetical utility functions representing risk averse, risk
prone and risk neutral attitudes are shown in Fig. 2.
According to the maximum expected utility criterion the
alternative a2 should be chosen if E[uja2]OE[uja1]. The
upper limit on CDI, which ensures this condition, has been
evaluated for the three utility functions. In the case of the
linear utility function (i.e. risk neutral attitude which is
equivalent to decision making based on the minimum
expected life-cycle cost) CDI should not exceed 0.91CC0. As
expected for the concave utility function (i.e. risk averse
attitude) CDI may be higher (!1.25CC0), i.e. the owner
prefers to invest more money initially to reduce risk of
higher losses in the future, while for the convex utility
function (i.e. risk prone attitude) CDI!0.83CC0, i.e. the
owner is ready to take higher risk by saving on the initial
investment.

3. Distributions of cost of failure

The first step of a decision analysis based on expected
utilities is to evaluate cumulative failure probabilities of

Fig. 2. Utility functions.

379

a structure over its service life, which is obtained by

a probabilistic time-dependent analysis of the structure
taking into account uncertainties associated with loads,
the structure properties and environmental conditions
(e.g. [4,7,8]). The next step is to determine the probability
distribution of life-cycle cost, which then can be combined
with owner-specified utility functions to evaluate expected
utilities. If the cost of failure is the only component of the
life-cycle cost analysis, which is considered as a random
variable, then only the probability distribution of the
cost of failure is necessary to determine the distribution of
life-cycle costs. This section describes how probability
distributions of the cost of failure of a single structure or of
a group of identical structures with single or multiple
possible failures over the service life of a structure can be
determined based on the cumulative probabilities of
structural failure.
3.1. Single structure
3.1.1. Single failure during service life
For a single structure with only one possible failure
during its service life the probability distribution of the cost
of failure without taking into account the discount rate is
(
CF Z cF
p1 ;
f CF Z
(7)
p2 Z 1 K p1 ; CF Z 0
where p1 and p2 are cumulative probabilities of failure and
no failure of the structure during its service life,
respectively.
If CF represents the present value cost of failure (i.e.
discount rate is taken into account) then its probability
distribution is given as
(
cF
Pf ti K Pf tiK1 ; CF Z
f CF Z
(8)
1 C rti
1 K Pf tM ;
CF Z 0
where Pf(ti) is the cumulative probability of failure at time ti
(iZ1,2,.,M), M the number of points in time at which
failure may occur, t0Z0, and tM denotes the latest possible
time of failure. It is assumed that repair/replacement of a
failed structure (say for serviceability failure) will occur
immediately after the structure is inspected. The time
between inspections, Dt, is defined as DtZ ti K tiK1 :
3.1.2. Multiple failures during service life
Multiple failures are possible when serviceability limit
states of a structure are considered (e.g. cracking, spalling,
rust, delamination). Assume that failures occur at discrete
points in time ti (iZ1,2,.,M), i.e. at the times when a
structure is inspected and violation of a serviceability limit
state can be detected so that i represents a sequential number
of an inspection and M the total number of inspections.
If failure detected during an inspection is denoted as 1
and no failure as 0 then combinations of failures

380

D.V. Val, M.G. Stewart / Reliability Engineering and System Safety 87 (2005) 377385

and no failures of a structure over its service life are

b(x;N,p), where p is the probability of failure of a single
represented by sequences of 1s and 0s of the length M (e.g.
structure from the group during its service life. Thus, the
the sequence containing all 0s represents the situation when
cost of failure (without taking into account the discount rate)
no failure occurs during the service life of the structure).
of N identical structures is a discrete random variable with
The total possible number of such sequences is 2M. It is also
the following probability distribution
assumed that a structure is repaired/replaced as soon as its
8
N!
<
serviceability failure has been detected and that after
px 1 K pNKx ; CF Z xcF
(10)

Z
f
C
N K x!x!
repair/replacement the structure is restored to its original
F
:
N
1
K
p
;
C
Z
0
undamaged condition.
F
If CF is treated as the present value cost of failure (i.e.
If the present value of the cost of failure is of interest (i.e.
discount rate is taken into account) then for each sequence
discount
rate is taken into account) then the probability
CF assumes a different value. Thus, CF is a discrete random
distribution
of the cost of failure of a single structure, which
variable with the following probability distribution
8(
)
Nf
Nf
X
>
1
< Y
Pf ti;j K Pf ti;jK1  1 K Pf ti;M K ti;Nf ; CF Z cF
f CF Z
(9)
1 C rti;j
jZ1
>
: jZ1
1 K Pf ti;M ;
CF Z 0
where Nf%M is the total number of failures (i.e. the
sequence contains Nf of 1s), i1 ; i2 ; .; iNf the numbers of
inspections at which failures are detected, ti,j the time of
the ij inspection, and ti,0Z0. In order to obtain the
complete probability distribution of CF all possible
sequences of 1s and 0s of the length M have to be
generated and for each sequence the corresponding values
of CF and their probabilities of occurrence is calculated
by Eq. (9).
If the discount rate is not taken into account then values
which can assume CF depend only on the number of failure
incidents during a structure service life, i.e. CFZNfcF
(NfZ0,1,.,M). Thus, CF can assume MC1 different
values, including a zero value for the no failure case.
The probability corresponding to each of these CF values
can be calculated as the sum of values given by Eq. (9) for
the same Nf.
Note that Eqs. (8) and (9) do not presume stochastic
independence of failure events. Stochastic dependence
between occurrences of a single failure event at different
points in time may be taken into account in calculations of
cumulative probabilities Pf(ti), while Eq. (9) accounts for
stochastic dependence which exists between the occurrence
of a number of failure events during service life of a
structure.

can assume MC1 different values, is given by Eq. (8).

The probability that x1 structures fail at time t1 (the cost of
failure of one structure from this group is c1 Z cF =1C rt1
and its probability of failure p1 Z Pf t1 K Pf 0; x 2
structures fail at time t2 c2 Z cF =1C rt2 and p2 Z Pf t2 K
Pf t1 ; .; and xMC1 structures do not fail at all (cMC1Z0
and pMC1 Z 1K Pf tM ) is represented by a multinomial
distribution f x1 ; x2 ; .; xMC1 ; p1 ; p2 ; .; pMC1 ; N: The cost
of failure of N identical structures is then a discrete random
variable with the following probability distribution
8
M
X
>
N!
<
MC1
px11 px22 ; .; pxMC1
; CF Z
ci xi
f CF Z x1 !x2 !; .; xMC1 !
iZ1
>
: N
pMC1 ;
CF Z 0
(11)

3.2. Group of similar structures

3.2.2. Multiple failures during service life

If the present value of the cost of failure is of interest (i.e.
discount rate is taken into account) then the probability
distribution of the cost of failure of a single structure is
given by Eq. (9) and the total possible number of values
which can assume this random variable is 2M. In order to
obtain the probability distribution of CF for N identical
structures all possible combinations of distributing N
structures between 2M values need to be generated.
The total number of such combinations can be calculated
using Eq. (12) where MC1 is replaced by 2 M.
The probability distribution of CF will be similar to that

Probability distributions of the cost of failure of N

identical structures built and then being in service in similar
conditions (i.e. it can be assumed that cumulative
probabilities of failure are the same for each of the
structures) are considered.
3.2.1. Single failure during service life
For a group of N identical structures with only one
possible failure during their service life the number of
structures which fail, x, follows a binomial distribution

In order to obtain the complete probability distribution of

CF all possible combinations of failures of N structures at M
different times and no failure need to be generated and for
each combination the corresponding values of CF and their
probabilities of occurrence can be calculated by Eq. (11).
Note that the total number of such combinations, V, is equal to
VZ

M C N!
M!N!

(12)

D.V. Val, M.G. Stewart / Reliability Engineering and System Safety 87 (2005) 377385

381

given by Eq. (11), except that ci and pi are evaluated using

Eq. (9).

4. Example
To illustrate the use of the probability distributions of
the cost of failure described above in decision analysis the
following example is considered. For a reinforced
concrete structure (or a group of identical structures),
which will be built right on the coast, it is necessary to
select a design specification which will ensure the
minimum expected life-cycle cost (or, if different, the
maximum expected utility) for 100-year service life of
the structure. A major problem for reinforced concrete
structures located on the coast is corrosion of reinforcing
steel caused by chloride contamination from seawater or
sea spray. Corrosion develops with time and leads to
reduction of the cross-sectional area of reinforcing steel
and cracking and spalling of the concrete cover that
affects both resistance and serviceability performance of a
reinforced concrete structure. To protect reinforcing steel
against corrosion and improve durability of a structure it
is usually recommended to use higher grade concrete and
thicker concrete cover.
Three design specifications are compared herein: poor
(in terms of durability)25-MPa concrete and 30-mm
cover, fair32-MPa concrete and 50-mm cover, and
good40-MPa concrete and 70-mm cover. In calculation
of life-cycle cost only the cost of serviceability failure
caused by spalling of the concrete cover is taken into
account. It is assumed that as soon as spalling of the
concrete cover has been detected during an inspection the
structure should be repaired or replaced (i.e. the cost of
failure is the cost of repair/replacement). Note that in
practice it is usually necessary to take into account also the
cost of failure associated with ultimate (strength) limit states
and, of course, the probability distributions presented above
can be used to calculate this cost. However, the purpose of
this example is not to present a comprehensive life-cycle
cost analysis but merely to illustrate the use of these
probability distributions in such an analysis. For this
purpose it is sufficient to consider only serviceability
failures, especially since multiple failures during a structure
service life are normally associated only with serviceability
limit states.

Fig. 3. Cumulative probabilities of failure (i.e. spalling).

initiation, propagation and cracking and spalling of the

concrete cover [4,7]. Results are shown in Fig. 3.
Probability distributions and corresponding cumulative
probabilities of the cost of failure are calculated for a single
structure with single failure (i.e. spalling) over its service
life using Eq. (8) and the cumulative probabilities of failure
shown in Fig. 3. It is assumed that repair/replacement
significantly improves durability performance of the
structure so that reoccurrence of spalling can be neglected.
The cumulative probabilities of CF obtained for rZ4% and
three different time intervals, Dt, between inspections
are shown in Fig. 4. Note that M is determined as the
nearest whole number equal to or less than (TK1)/Dt, where
TZ100 years is the service life of the structure. For
simplicity it is assumed that the cost of failure, cF, set at
the time of decision making is the same for all of the
design specifications. The results demonstrate clearly that
for better durability specifications higher failure costs occur
with lower probabilities. Note that as the time between

4.1. Probability distributions of cost of failure

The probability distributions of the cost of failure are
based on cumulative probabilities of failure of a single
structure over its service life. Thus, initially the cumulative
probabilities of failure (i.e. spalling of the concrete cover)
are calculated for the three design specifications of a
reinforced concrete structure located on the coast. This is
carried out using previously developed models for corrosion

Fig. 4. Cumulative probabilities of cost of failure for single structure with

single failure possible over its service life.

382

D.V. Val, M.G. Stewart / Reliability Engineering and System Safety 87 (2005) 377385

inspections (Dt) increases the time of repair/replacement is

increasingly deferred, resulting in lower present value
failure costs.
Probability distributions and corresponding cumulative
probabilities of the cost of failure have also been
calculated for a single structure with multiple failures
possible over the structure service life using Eq. (9). It is
assumed in this case that repair/replacement restores the
structure to its original condition without improving its
future durability performance. The resulting cumulative
probabilities of CF obtained for rZ4% are shown in
Fig. 5. As expected the maximum possible cost of failure
is higher than in the case when only a single failure can
occur (see Fig. 4).
Probability distributions and corresponding cumulative
probabilities of the cost of failure are then calculated for a
group of identical structures with the fair durability
specification when single (DtZ10 years) as well as multiple
(DtZ20 years) failures are possible over the service life of
a structure. Results are shown in Figs. 6 and 7.
For comparison the cumulative probabilities of CF are
presented for one structure from a group (i.e. CF/N). As it
can be seen the variability of CF decreases with an increase
in the number of structures, N, in a group. It can be shown
that the coefficient
of variation of CF is inversely
p
proportional to N :
4.2. Choice of optimal durability specifications
The minimum construction cost will obviously be for the
poor durability specification denoted herein as CC0. The
construction costs for fair and good specifications are
then denoted as CC1ZCC0CCDI,1 and CC2ZCC0CCDI,2,
respectively, where CDI,1 and CDI,2 are cost increases due
to the use of higher grade concrete and thicker concrete
cover. It is assumed that the cost of failure set at the
time of decision making for each design specification is

Fig. 5. Cumulative probabilities of cost of failure for single structure with

multiple failures possible over its service life.

Fig. 6. Cumulative probabilities of cost of failure for group of N structures

with fair design specification and single failure possible.

cFiZCCiCCC0, where iZ0,1 and 2 for poor, fair, and

good design specifications, respectively. To select the
optimal design specification the corresponding expected
utilities need to be estimated and compared. This is carried
out for three hypothetical utility functions (i.e. functions
relating utility with life-cycle cost):
(i) a linear function when results based on expected
utilities are equivalent to those based on expected lifecycle costs (represents risk neutral attitude),
(ii) a concave bilinear function (represents risk averse
attitude), and
(iii) a convex utility function (represents risk prone
attitude).
These utility functions are shown in Fig. 8, where Cmax
denotes the maximum life-cycle cost possible among the
compared alternatives. Accordingly, for a value of LCC less

Fig. 7. Cumulative probabilities of cost of failure for group of N structures

with fair design specification and multiple failures possible.

D.V. Val, M.G. Stewart / Reliability Engineering and System Safety 87 (2005) 377385

383

Fig. 8. Example: utility functions.

than or equal to Cmax the corresponding utility is calculated

as
8
CC0 % LCC% 2CC0
>
< 1;
(13)
uLCC Z Cmax K LCC
>
; 2CC0 ! LCC% Cmax
:
Cmax K 2CC0

Fig. 10. Preferable specifications depending on cost of durability

improvements for a single structurerisk averse utility function.

where L is the total number of discrete random variable CF

values. After expected utilities for the different design
specifications have been calculated a preferable design
specification is selected as a design specification with the
maximum expected utility. Of course, as seen from Eq. (15)

the selection of the preferable design specification will

depend on its initial construction cost, CCi, and subsequently, on the costs of durability improvements for the
fair and good design specifications, i.e. CDI,1 and CDI,2.
In all the following analyses it is assumed that the discount
rate rZ4%.
For a single structure with only a single failure possible,
which will be inspected every 10 years (i.e. DtZ10 years
and MZ9), a decision analysis with the linear utility
function (i.e. decision is based on the minimum expected
life-cycle cost) yields the following results (see Fig. 9). The
poor specification is preferable when CDI,1O0.269CC0 and
CDI,2O0.410CC0, the fair when CDI,1!0.269CC0 and
CDI,2O0.124CC0C1.063 CDI,1, and in the rest of the cases
the good specification should be chosen.
Results for the same structure for the risk averse utility
function are shown in Fig. 10. This utility function reflects
risk averse behavior so that the better durability specifications remain preferable for the costs of durability improvement higher than in the previously considered analysis with
a risk neutral attitude. Also note that the border between the
fair and good specifications becomes nonlinear. Results
for the risk prone utility function are shown in Fig. 11.
As expected, in this case the better durability specifications
will be preferable for lower costs of durability improvement
than in the case of a risk neutral attitude (compare with
Fig. 9). Clearly, including the variability of failure costs
affects the maximum expected utility for nonlinear utility
functions.

Fig. 9. Preferable specifications depending on cost of durability improvementslinear utility function.

Fig. 11. Preferable specifications depending on cost of durability

improvements for a single structurerisk prone utility function.

for the risk averse utility function, and as

(
2CC0 K LCC
; CC0 % LCC% 2CC0
uLCC Z
CC0
0;
2CC0 ! LCC% Cmax

(14)

for the risk prone utility function, where LCCZ CCi C

CF cFi iZ 0; 1; 2: Possible values of CF depend on the
design specification (i.e. cFi) and are calculated according to
an appropriate distribution (see Section 3). Probabilities of
occurrence related to these values of CF, f(CF), are also
probabilities of occurrence of the corresponding values of
LCC and, subsequently, of the corresponding utilities (see
Eqs. (13) and (14)) and are used to calculate the expected
utility for a given design specification i
Euji Z

L
X

uLCCl f CF;l

lZ1

L
X

uCCi C CF;l cFi f CF;l

(15)

lZ1

384

D.V. Val, M.G. Stewart / Reliability Engineering and System Safety 87 (2005) 377385

For a risk neutral attitude (i.e. linear utility function)

there is no difference in a decision analysis if a single
structure or a group of identical structures is considered
since the expected cost of failure of N identical structures
equals N times the expected cost of failure of a single
structure. Thus, the results presented in Fig. 9 for a
single structure are also valid for a group of N identical
structures for a risk neutral attitude. However, when the
utility function is nonlinear (i.e. risk averse or risk prone
attitude) the relation between the expected utilities for
different design specifications will depend not only on the
expected values of CF but on the variabilities of CF as
well and as it has been noted before for a group of N
identical structures the variability of CF decreases with an
increase in N. Thus, results for a group of N identical
structures will be different from those obtained for a
single structure and will depend on N. For example,
Fig. 12 shows results of a decision analysis with the risk
averse utility function for a group of five identical
structures with a single failure possible, which will be
inspected every 10 years. As can be seen, the results are
different from those obtained for a single structure
(compare with Fig. 10).
To examine further the influence of the reduced
variability of CF for a group of identical structures on
decision making the poor and good durability specifications have been compared for different values of N for a
group of structures with single failure possible and DtZ10
years (for simplicity the fair specification has not been
considered). From the previously presented results for a
single structure (i.e. NZ1) and the risk averse utility
function the good specification should be selected if
CDI,2!1.030CC0 (see Fig. 10). However, as the number of
structures increases and, correspondingly, the variability of
CF decreases the upper limit of CDI,2 (i.e. the maximum
value of CDI,2 for which the good design specification still
remains optimal) decreases as wellfor NZ2 the good
specification is preferable when CDI,2 for one structure is
less than 0.914CC0, while for NZ10 CDI,2 should be less
than 0.771CC0 (see Fig. 13).

Fig. 13. Relationship between upper limit of CDI,2 and number of

structures when poor and good design specifications are compared using
risk averse utility function.

5. Conclusions
Probability distributions of the cost of failure of a single
structure and a group of identical structures with single or
multiple possible failures during service life have been
derived. The main assumptions used for it are: (i) failures
occur at discrete points in time; (ii) the cost of failure, cF, set
at the time of decision making is constant for a particular
design solution, i.e. it does not depend on time of failure of a
failure mode (if more than one failure mode are considered);
(iii) the discount rate has a constant value which does not
change with time. When a group of structures is considered
it is assumed that the structures are almost identical, built in
similar environmental conditions approximately at the same
time and their failures are not correlated. The probability
distributions are based on cumulative probabilities of failure
of a single structure, which can be found by a probabilistic
time-dependent analysis of the structure. The distributions
can be used in decision analysis to evaluate the expected
life-cycle cost or, when a utility function is nonlinear, the
expected utility. The use of these probability distributions to
represent variability of failure costs has been illustrated by
an example, which considers the selection of durability
specifications for a reinforced concrete structure (or a group
of identical structures) built on the coast. Results have been
presented for three utility functionsa linear function that
is equivalent to the use of the expected life-cycle cost, and
nonlinear utility functions representing risk averse and risk
prone attitudes. It was clearly demonstrated that the
variability of failure costs affected the expected utility for
nonlinear utility functions.

Acknowledgements
Fig. 12. Preferable specifications depending on cost of durability
improvements for a group of five identical structuresrisk averse utility
function.

The first author acknowledges the Fund for the

Promotion of Research at the Technion and the Technion

D.V. Val, M.G. Stewart / Reliability Engineering and System Safety 87 (2005) 377385

V.P.R. FundM.R. Saulson Research Fund for funding

support of this research.

References
[1] Mori Y, Ellingwood BR. Maintaining reliability of concrete structures.
II: optimum inspection/repair. ASCE J Struct Eng 1994;120(3):84662.
[2] Frangopol DM, Lin KY, Estes AC. Life-cycle cost design of
deteriorating structures. ASCE J Struct Eng 1997;123(10):1390401.
[3] Stewart MG. Spalling risks, durability and life-cycle costs for
RC structures. Safety, risk, and reliabilitytrends in engineering.
Proceedings of the International Conference, Malta; p. 53742.

385

[4] Val DV, Stewart MG. Reliability-based life-cycle cost analysis of

reinforced concrete structures in marine environments. Structural
safety and reliability. Proceedings of the Eighth International
Conference on Structural Safety and Reliability, Newport Beach, CA,
USA [CD-ROM]. 2001.
[5] Von Neumann J, Morgenstern O. Theory of games and economic
behavior. Princeton, NJ: Princeton University Press; 1944.
[6] Ruegg RT, Marshall HE. Building economics: theory and practice.
New York, NY: Van Nostrand Reinhold; 1990.
[7] Vu KAT, Stewart MG. Structural reliability of concrete bridges
including improved chloride-induced corrosion models. Struct Saf
2000;20(4):31333.
[8] Estes AC, Frangopol DM. Bridge lifetime system reliability under
multiple limit states. ASCE J Bridge Eng 2001;6(6):5238.