Structural Safety 72 (2018) 54–64

Contents lists available at ScienceDirect

Structural Safety
journal homepage: www.elsevier.com/locate/strusafe

Optimum design and damage control for load sequences

Jorge L. Alamilla a,⇑, Dante Tolentino b
a
Instituto Mexicano del Petróleo, Eje Central Lázaro Cárdenas Norte 152, Col. San Bartolo Atepehuacan, Gustavo A. Madero, México DF c.p. 07730, Mexico
b
Instituto Politécnico Nacional, ESIA Zacatenco, Gustavo A. Madero, Col. La Escalera, Gustavo A. Madero, México, DF c.p. 07320, Mexico

a r t i c l e i n f o a b s t r a c t

Article history: A probabilistic approach to decision-optimal design and damage control is developed for structural sys-
Received 18 May 2017 tems that can gradually accumulate damage by nonlinear behavior under sequences of dynamic loads,
Received in revised form 19 December 2017 whose occurrence can be idealized by renewal stochastic processes. To minimize consequences and dam-
Accepted 20 December 2017
ages during the life cycle of a structure, a damage threshold is established as a measure of damage con-
Available online 28 December 2017
trol. If structural damages are lesser than such a damage threshold the structure is not repaired,
otherwise the structure is repaired or rebuilt. The proposed approach is generalized and capable of
Keywords:
describing particular cases for the optimization of expected losses. One of them is the well-known case
Cumulative damage
Damage threshold
used as a basis for many current design criteria in which it is assumed that the structure is repaired or
Optimization rebuilt systematically after some damage or failure. The present work extends the ideas and models
Losses reported in several seminal papers. However, the proposed approach has the advantage that it takes into
Costs account cumulative structural damage over time, allows evaluating objectively the cost of damages and
Life cycle sets an optimum repairing damage threshold. Finally, the probabilistic formulation is illustrated through
Renewal process its application to a building subjected to sequences of earthquakes.
Present value Ó 2017 Elsevier Ltd. All rights reserved.

1. Introduction of the expected damage losses conditioned on the structural sur-

It is now recognized that design codes should ethically generate
optimal designs for society in the sense that they have to consider
protecting people and property [1]. A useful tool to achieve this
goal is decision theory [2,3]. On the basis of this theory and accord-
ing to the annals of UNAM (National University of Mexico) in 1968,
Esteva [4] proposed the first model of design decisions for struc-
tures subjected to cyclic loads (earthquakes) that occur sporadi-
cally according to a stochastic Poisson process. Later on, between
1971 and 1974, Rosenblueth and Mendoza [5], Rosenblueth [6]
and Hasofer [7] generalized the ideas of [4], by considering that
loads are generated according to a stochastic renewal process, in
which the Poisson process is only a particular case. In these works,
the optimization (minimizing the present value of expected losses
or maximizing benefits) is carried out by taking into account only
initial and failure costs, because the formulation is related exclu-
sively to a failure model. These kinds of models consider the con-
tributions to total losses of all failure events occurring over time,
since the structure is rebuilt each time it fails. Then, in 1976,
Rosenblueth [8] additionally considered explicitly the inclusion
⇑ Corresponding author.
E-mail addresses: jalamill@imp.mx (J.L. Alamilla),
(D. Tolentino).
vival. Long after, in 2000, Rackwitz [9] took the Rosenblueth’s
and Hasofer’s approach [5–8] and made some improvements to
the model, but without considering the damage and the associated
damage costs conditioned on the structural survival. However, in
other works [10–12] damage and repair costs are taken into
account in the optimization process. This kind of model, based
on renewal process, has been used by some researchers to perform
engineering applications in the optimization of structural systems
[13–16].
A fundamental assumption in all these works indicate that ‘‘the
structural capacity is immediately restored after each load event
causing damage”, which implies that the failure probability for
uncertain load events is constant and even those damages that
are not identified or visible are repaired. This means that there is
no cumulative damage and every time that the structure is
exposed to the next load its capacity remains intact. This assump-
tion is also the backbone of modern structural design codes,
because the design criteria refer to the behavior of structural sys-
tems with intact mechanical properties subjected to dynamic loads
with different probabilities of occurrence. Codes do not explicitly
take into account changes in the vulnerability of the structure dur-
ing its service life, let alone specify acceptable damage thresholds.
The above assumption has the advantage that mathematical anal-
dtolentino@ipn.mx ysis leads to simple analytical models and results. However, in

https://doi.org/10.1016/j.strusafe.2017.12.006
0167-4730/Ó 2017 Elsevier Ltd. All rights reserved.
 J.L. Alamilla, D. Tolentino / Structural Safety 72 (2018) 54–64 55

Z d
engineering practice the repair or rebuild decisions are made from
an assessment of structural integrity, and in many cases this
p½Dn > d ¼ p½DDn > d  xjxf Dn1 ðxÞdx ð3Þ
0
assessment is virtual and subjective. Therefore, the assumption
of systematically repairing or rebuilding the structure at any level Each increment is conditioned on the previous state of struc-
of damage after each load event is difficult to satisfy, whereas in tural integrity. Here, f Dn1 ðÞ is the probability density function of
other cases, it may be appropriate. damage state to the occurrence of the ðn  1Þ-th dynamic load,
On the other hand, under this kind of approach, neither the which is obtained as the derivative with respect to d of
damage probabilities are estimated from an explicit formulation p½Dn1 6 d. Regarding that damage remains constant between load
nor the damage costs are explicitly estimated for given levels of events and that the increment of damage DDn not only depends on
damage. For this reason, the damage costs are estimated roughly, the last state of damage but also on the load of intensity Y ¼ y, Eq.
typically associated with a load intensity indicator. Apparently, (3) can be expressed in extended form as
only the initial costs and expected consequences due to structural Z 1 Z d
failures are described adequately, because the stochastic formula- p½Dn > d ¼ p½DDn ðyÞ > d  xjx; yf Y ðyÞf Dn1 ðxÞdx dy ð4Þ
0 0
tion [4–7,9] only considers the failure condition, meanwhile the
probability of damages and damage costs related to the structural The probability distribution function of cumulative damage
survival associated with each load are not explicitly considered. F Dn ðdÞ ¼ 1  p½Dn > d, associated with the n-th load, for all d P 0,
This paper describes an explicit and generalized probabilistic is expressed as
formulation to make decisions for design and damage control of 8
> 1 n¼0
structural systems whose structural vulnerability can change dur- <R1
ing its planned service life due to the impact of stochastic loads F Dn ðdÞ ¼ 0
F DD1 ðdjyÞf Y ðyÞdy n ¼ 1
>
:R1 Rd
with uncertain intensities. The development follows the ideas of 0
F ðd  xjx; yÞf Y ðyÞf Dn1 ðxÞdx dy n ¼ 2; . . .
0 DDn
the above mentioned seminal papers [4–9]. The first part of this ð5Þ
work outlines the basic stochastic characteristics useful to describe
the evolution of damage in structural systems. After this, a gener- Here F DD1 ðjÞ and F DDn ðjÞ denote the cumulative distribution
alized formulation and a particular case are described to quantify functions of random damage increments DD1 and DDn , respec-
the present value of the expected losses of structures with cumu- tively. In Eq. (5), f Dn ðxÞ ¼ dF Dn ðxÞ=dx is the probability density func-
lative damage. Finally, the proposed formulation is applied to a tion of cumulative damage to the n-th load, dðÞ=dx indicates the
reinforced concrete building subjected to seismic sequences. derivative with respect to x. For n ¼ 0, f D0 ðdÞ ¼ 1, if d ¼ 0; and
R1
f D0 ðdÞ ¼ 0, if d–0. However, 1 f D0 ðxÞdx ¼ 1, in other words
2. Stochastic properties of cumulative damage f D0 ðÞ is a Dirac’s delta function. The probability distributions
F Dn ðdÞ, n ¼ 2; . . ., given by Eq. (5) should be obtained recursively.
This section outlines the basic stochastic properties to develop Fig. 1 shows schematically the form of f Dn ðÞ, for a given integer
the generalized optimization model presented in the next sections. value of n, in which three characteristic states of damage can be
In this work we focus only on the probability density functions of identified: 1) no damage, 2) a certain level of damage and 3)
the time to the n-th, n ¼ 1; 2; . . ., exceedance of a given damage collapse.
threshold. On the other hand, substituting Eq. (5) into Eq. (2), and express-
ing p½DðtÞ > d in terms of its complement, the probability distribu-
2.1. Probability functions of the time to the n-th exceedance of a given tion function F DðtÞ ðdÞ, Eq. (2) is rewritten as
damage threshold
X
1 X
1
F DðtÞ ðdÞ ¼ 1  pn ðtÞ½1  F Dn ðdÞ ¼ pn ðtÞF Dn ðdÞ ð6Þ
Cumulative damage over time t is described by the random n¼1 n¼0
function DðtÞ; it takes values in the interval [0, 1]. DðtÞ ¼ 0 indi-
According to the renewal theory [17], pn ðÞ is expressed in terms
cates nonstructural damage and DðtÞ P 1 indicates total damage.
of the probability distribution functions F Sn ðÞ and F Snþ1 ðÞ, which
DðtÞ is represented as the sum of random increments of damage
describe the time to the n-th and to the ðn þ 1Þ-th load respec-
DDn ð; Þ, as follows
tively, and is written as
X
NðtÞ
DðtÞ ¼ DDn ðt  Sn ; Y n Þ ð1Þ
n¼1

NðÞ describes the cumulative numbers of load events that have fDn(d)
impacted the structure over time through a stochastic renewal
process [17]. Damage increments DDn ð; Þ, n ¼ 1; 2; . . . are random
variables that quantify on a dimensionless uniform scale [0, 1] the
potential structural degradation each time that the structure is
subjected to a dynamic load. Sn denotes the random variable of
the time to the occurrence of the n-th load with intensity Y n ¼ Y,
also treated as a random variable. Here f Y ðÞ is the probability den-
sity function of the intensities of dynamic loads. The probability
p½, that at time t, DðÞ exceeds a given value of d, can be written as
" # No damage Collapse
X
1 X
n
p½DðtÞ > d ¼ pn ðtÞp DDi ðt  Si ; Y i Þ > d ð2Þ
n¼1 i¼1
P
For conciseness, DDi ¼ DDi ð; Þ and Dn ¼ ni¼1 DDi . Furthermore, 0
Damage, d 1
Dn ¼ DDi ¼ 0 if d < 0. Thus, the damage Dn at the n-th occurrence
can be expressed as Dn ¼ Dn1 þ DDn . According to this, Fig. 1. Probability density function of damage at the n-th stochastic load.
56 J.L. Alamilla, D. Tolentino / Structural Safety 72 (2018) 54–64

pn ðtÞ ¼ F Sn ðtÞ  F Snþ1 ðtÞ ð7Þ the first term of the sum in Eq. (13) is associated with the occur-
rence of the k-th dynamic load, since at least k events must occur
Substituting Eq. (7) into Eq. (6), this can be expressed as
before the structure is rehabilitated or rebuilt for the k-th time.
X
1
F DðtÞ ðdÞ ¼ 1  F Sn ðtÞpDn ðdÞ ð8Þ
n¼1
2.2. Probability functions regarding systematic repair or rebuilding
after each damage
where, pDn ¼ pDn ðdÞ is the probability that at the occurrence of the
n-th load the damage is greater than d and is obtained as If we assume that the structural capacity is immediately
restored after each load event that causes damage, then the distri-
pDn ðdÞ ¼ F Dn1 ðdÞ  F Dn ðdÞ ð9Þ
bution function of damage to the n-th load, given by Eq. (5), is
On the basis of Eq. (8), the probability that a structure is transformed into Eq. (15). In this case, since the structure recovers
repaired or rebuilt at time t due to the structural damage that its initial structural capacity after each damage, then DDn ¼ DD1
exceeds a given damage threshold d0 is defined as and the damage increments are independent and identically dis-
tributed random variables. For this particular case the structure
X
1
p½DðtÞ > d0  ¼ 1  F DðtÞ ðd0 Þ ¼ F Sn ðtÞpDn ðd0 Þ ð10Þ does not accumulate damage.
n¼1  n
F Dn ðdÞ ¼ F D1 ðdÞ ð15Þ
According to this definition, the probability density function
g 1 ðtÞ ¼ dðp½DðtÞ > d0 Þ=dt of time to the first repair or replacement Considering the above assumption and in a similar way to
is given by Eq. (11). Where dðÞ=dt indicates the derivative with [4–9], if we are only interested in quantifying the structural failure,
 n
respect to t. then F Dn ð1Þ ¼ F D1 ð1Þ ¼ ð1  pF Þn (see Fig. 1). Here, pF is the fail-
ure probability of a structure subjected to an uncertain dynamic
X
1
g 1 ðtÞ ¼ f Sn ðtÞpD1;n ðd0 Þ ð11Þ load. This probability is obtained according to Eq. (15) regarding
n¼1 Eq. (5), in which pF ðyÞ is the failure probability for given load inten-
sities Y ¼ y.
Here pD1;n ¼ pDn and f Sn ðÞ is the probability density function of
Z 1
time to the occurrence of the n-th dynamic load, which is obtained
pF ¼ pF ðyÞf Y ðyÞdy ð16Þ
according to Eq. (12). In this equation, f S ðÞ is the probability den- 0
sity function of time between load occurrences.
Eqs. (11) and (13) are transformed into the probability density
Z t function of time to the first and k-th failure, given by Eq. (17),
f Sn ðtÞ ¼ f S ðt  sÞf Sn1 ðsÞds; n ¼ 2; 3; . . . ð12Þ where k ¼ 1; 2; . . .
0
X
1  
Once the system is rebuilt or rehabilitated for the first time, the ðn  1Þ!
g F k ðtÞ ¼ f Sn ðtÞ ð1  pF Þnk pF k ; n ¼ k; k þ 1; . . .
process restarts until it is restored for the second time and so on. n¼k
ðk  1Þ!ðn  kÞ!
According to the renewal theory [17] the probability density func-
ð17Þ
tion g k ðÞ that at time t the k-th repair or rebuilt is obtained recur-
sively by convolution integral If k ¼ 1, Eq. (17) becomes the probability density function of
Z t X
1 time to the first failure
g k ðtÞ ¼ gðt  sÞg k1 ðsÞds ¼ f Sn ðtÞpDk;n ðd0 Þ; n ¼ k; k þ 1; . . . X
1
0 n¼k g F 1 ðtÞ ¼ f Sn ðtÞfð1  pF Þn1 pF g ð18Þ
ð13Þ n¼1

where The expression between brackets in Eq. (17) denotes a Negative

Binomial mass probability function, that in the following will be
X
n
denoted as pF k;n ¼ pDk;n , meanwhile the expression between brack-
pDk;n ðd0 Þ ¼ pDk1;j1 pDnjþ1 ; k ¼ 2; 3; . . . ð14Þ
j¼k ets in Eq. (18) denotes a geometric mass probability function.
The convolution of this latter probability function results in the
Here g k1 ðÞ is the probability density function of time at the Negative Binomial pF k;n .
P
ðk  1Þ-rehabilitation, k ¼ 2; 3 . . ., and gðtÞ ¼ 1 n¼1 f Sn ðtÞpDn ðd0 Þ is
the probability density function of time between rehabilitation or
2.3. Probability of no repair to the occurrence of the n-th load event
rebuilt. Note that gðÞ and g 1 ðÞ are equal if the probability density
function of the time to the occurrence of the first load are also
In addition to the above equations, we are interested in estimat-
equal; in this case it is said that the stochastic process is ordinary.
ing the probability pD n ¼ pD n ðd0 Þ that at the occurrence of the n-th
Rackwitz [9] states that a modified renewal process should be used
stochastic load, the structural damages are less than the damage
instead of an ordinary renewal process. In a modified process the
threshold d0 , which is also equal to the event of no repair to the
probability density function of time to the first occurrence is given
occurrence of the n-th load. The assessment of this probability
by f S1 ðtÞ ¼ ð1  F S ðtÞÞ=l where F S ðtÞ is the cumulative distribution
takes into account the accumulation of damage from the last pos-
function of time between events and l is its mean value. The main
sible rehabilitation or rebuilding, which is quantified according to
reason to use a modified process is related to the fact that the
Eq. (19). This probability is obtained from the complement of the
structural life cycle does not start exactly at the occurrence of a
probability of the union of all independent events, which describes
stochastic load.
that at the occurrence of the n-th load the structure is rehabilitated
The probability density functions, given respectively by Eqs.
or rebuilt for the k-th time.
(11) and (13), describe the time in which the structure is rehabili-
tated or rebuilt for the first and k-th time, since a given damage X
n

threshold is exceeded. If the damage threshold value is one, then pD n ðd0 Þ ¼ 1  pDk;n ðd0 Þ ð19Þ
k¼1
Eqs. (11) and (13) are transformed into probability density func-
tions of time to the first and k-th failure, respectively. Note that Substituting Eq. (14) into the latter
 J.L. Alamilla, D. Tolentino / Structural Safety 72 (2018) 54–64
" #
X
n X
n In addition to the policy of rehabilitation or reconstruction after
pD n ðd0 Þ ¼ 1  pD1;n þ pDk1;j1 pDnjþ1 ð20Þ
k¼2 j¼k
If we assume that the structure is immediately restored after
each damage or failure and we are only interested in failure prob-
ability, then pDk;n is a Negative Binomial probability mass function.
Under this last assumption, it can be easily verified that
pD n ¼ 1  pF is constant for all n, and describes the probability that
the structure does not fail (reliability) when it is exposed to a load
of uncertain intensity. This result is consistent, because every time
the structure is subjected to a stochastic load its capacity remains
intact or renewed and therefore its chance of survival is the same
for all n. Eq. (20) and its assessment is a primary concern when
quantifying expected costs related to the non-exceedance of a
given damage threshold.
3. Cost-optimization model
On the basis of [4–6] and the stochastic characteristics pre-
sented above, the decision problem is formulated as the optimiza-
tion of an objective function. This optimization can be treated as
either a maximization or a minimization problem. In the maxi-
mization problem, up to now a convenient objective function to
maximize is the function shown in Eq. (21) and in this work, with-
out loss of generality, it is expressed in monetary terms. However,
the evaluation of intangible losses and the ethical issues involving
their estimation are beyond the scope of this work.
Z ¼ B  C0  L ð21Þ
B ¼ BðbÞ is the expected benefit during the life cycle of the
structure, C 0 ¼ C 0 ðbÞ is the initial cost, comprising the cost of the
structural system’s design and build. These benefits and costs are
a function of vector b, consisting of those parameters and/or design
procedures that are the subject of the decision. L ¼ LðbÞ is the pre-
sent value of the total expected losses due to damages and failures
and is a function of vector b.
In the minimization problem [8], the objective function given
by Eq. (21) is recast as
Z ¼ C0 þ L ð22Þ
Based on, Eq. (21) or Eq. (22), the optimized parameters satisfy
that @ZðbÞ=@b ¼ 0, where @ðÞ=@b indicates a partial derivative. To
quantify L, a damage threshold d0 is considered. Each time the
structure is subjected to a dynamic load, one of the following
two scenarios occurs. In the first, damage threshold d0 is exceeded
requiring rehabilitation actions, which result in structure recovery
to its original condition; in this scenario, the expected losses
LE ¼ LE ðbÞ include costs for nonstructural damages and structural
rehabilitations. In the second scenario, the damage threshold is
not exceeded and, therefore, the structure is not repaired or rebuilt,
then it can accumulate damage. In this scenario, the present value
of expected losses LNE ¼ LNE ðbÞ only includes nonstructural costs. In
the following sections, we distinguish between structural losses
which are related to repairs or rebuilding of the structural system,
and nonstructural losses which are related to equipment, furniture,
partition walls, injuries, etc. Here, these two types of losses are a
function of structural damage. Thus, the total present value of
expected damage losses is quantified as shown in Eq. (23). The
damage threshold d0 is part of vector b; however d0 will be treated
as outside vector b. In the following, all costs and probabilities are
functions of the vector b.
L ¼ LE þ LNE ð23Þ
57
the exceedance of a given damage threshold, it is considered that
the structure is rehabilitated or systematically rebuilt until t > T,
where T is the time during which structural facilities are needed
in order to obtain planned benefits. In the following, each one of
the expected costs involved in Eqs. (21) to (23) are evaluated. How-
ever, the decision problem is treated as a minimization problem.
3.1. Present value of expected benefits
According to [4–6], it may be considered that the benefits
bðtÞ ¼ b obtained during the life cycle of the structure are constant.
According to [9], under the assumption that the structure is imme-
diately repaired or rebuilt, the present value of expected benefits
are obtained as follows
Z T
b
B¼b exp½ctdt ¼ ½1  exp½cT ð24Þ
0 c
In Eq. (24), exp(ct) is the actualization function or capitaliza-
tion function at time t, in years, which transforms the benefits into
the present value and c is the discount rate. If T ! 1, B ¼ b=c.
3.2. Present value of expected losses associated with a damage
threshold exceedance
In general, almost all damaged civil structures are rehabilitated
or rebuilt, or at least it is reasonable to think that this will always
happen because of the demand for civil infrastructure by service
users and society [9]. For this reason, the seminal works [4–9] have
shown that each time the structure fails, the failure cost is dis-
counted and the structure is immediately renewed. In these works,
for the k-th failure, the present value of the failure cost is estimated
R1
as C F 0 ect g F k ðtÞdt. Where, C F is the failure cost, and g F k ðtÞdt is the
k-th structural failure probability at time interval ½t; t þ dt. Thus,
the total losses are the sum of each failure cost as
Z
X
1 1
CF ect g F k ðtÞdt: ð25Þ
k¼1 0
According to the above, LE is quantified as follows
Z
X
1 X
1 T
LE ¼ Lnkþ1 pDk;n ect f Sn ðtÞdt ð26Þ
k¼1 n¼k 0
Each term of the inner sum quantifies the present value of
expected loss associated with each load occurrence, such that the
damage threshold is exceeded for the k-th time. The outer sum
quantifies all k ¼ 1; 2; . . . possible damage exceedances. To quantify
the inner sum the probability density functions given by Eqs. (11)
and (13) are used.
When the structure is exposed to the n-th stochastic load, the
expected loss is Lnkþ1 and the probability that damage d0 is
exceeded for the k-th time at the n-th load occurrence is pDk;n .
The integral in Eqs. (25) and (26) quantifies the capitalization
and is associated with the occurrence of each load n ¼ f1; 2; . . .g
at time t with probability f Sn ðtÞdt. The foregoing paragraphs are
illustrated in Fig. 2.
In Eq. (26), Ln , n ¼ f1; 2; . . .g, is the expected cost that would be
paid as a result of repair or rebuilt and for nonstructural damages
to the occurrence of the n-th load. These costs are estimated
according to Eq. (27).
( R1
q1 ðl ðdÞ þ lD ðd;0ÞÞf DD1 ðdÞdd if n ¼ 1
d0 R
Ln ¼ R 1 R d0
qn d0 0 ðlR ðdÞ þ lD ðd;xÞÞf DDn ðd  xÞf Dn1 ðxÞdxdd if n ¼ 2; 3;. ..
ð27Þ
58 J.L. Alamilla, D. Tolentino / Structural Safety 72 (2018) 54–64

Total nonstructural
6
failure losses, C F C0

5
Nonstructural losses, l D C0
4
Rehabilitation and
rebuild costs, l R C 0

Losses
3

2 Initial cost C0 , C0 C0

0.0 0.2 0.4 0.6 0.8 1.0

Fig. 2. Schematic description of the discounted process. Damage, d

In Eq. (27), for n P 2 the inner integral indicates that at the Fig. 3. Schematic description of the losses as a function of damage index.
occurrence of the ðn  1Þ-th load the threshold d0 was not
exceeded, while the outer integral considers that at the n-th load X
1 Z T
event, the damages exceed the threshold d0 . The factor qn satisfies LNE ¼ Hn pD n ect f Sn ðtÞdt ð28Þ
n¼1 0
that the area associated with the limits of integration of the prob-
ability density function f Dn ðb; dÞ is equal to 1. For n ¼ 1, the initial Hn is estimated from Eq. (29) and describes the expected loss
damage is equal to zero with probability equal to one. The trunca- associated with the occurrence of the n-th stochastic load, in which
tion is necessary to obtain the mean value associated with the lim- the structural damage is less than the threshold d0 . In this scenario,
its of integration. Hn , j ¼ f1; 2; . . .g and consequently the present value of expected
The losses are expressed as a function of structural damage losses LNE only includes the nonstructural losses lD as a function
which may be related to any physical damage measurement. As structural damage. The assessment of Hn is obtained from the con-
shown in Eq. (27), the expected cost Ln , n ¼ f1; 2; . . .g involves tributions of the different ways in which the structural damage is
the costs lR ¼ lR ðdÞ due to the single action of repair or rebuilt less than d0 exactly at the occurrence of the n-th load. The proba-
and the costs of nonstructural losses lD ¼ lD ðd; dI Þ for each possible bilities of these different ways are estimated according to Eq. (20).
damage state. Note that according to the structure of Eq. (27), Ln The factor kn in Eq. (29) is quantified by Eq. (30) and satisfies that
can also be written as Ln ¼ LRn þ LDn , where LRn is related to the the sum of such probabilities is equal to one, for each n ¼ 1; 2; . . .
integral associated with lR , meanwhile LDn is related to the integral
X
n X
n
associated with lD . In this sense, the present value of expected Hn ¼ kn ð1  pD1;n Þhn þ kn pDk1;j1 ð1  pDnjþ1 Þhnjþ1 ð29Þ
losses LE due to the exceedance of d0 , given by Eq. (26), can also k¼2 j¼k
be written as LE ¼ LER þ LED , where LER is the present value of
expected costs due to repair or rebuilt and LED is the present value 1
X
n X
n

of expected losses due to nonstructural damage. kn ¼ ð1  pD1;n Þ þ pDk1;j1 ð1  pDnjþ1 Þ ð30Þ

k¼2 j¼k
Here, the losses lD are a function of the damage increments and
also depend on the initial structural damage dI before the load appli- For each manner in which the structural damage is less than d0
cation. These losses as a function of structural damage are shown exactly at the occurrence of the n-th load, the mean nonstructural
schematically in Fig. 3 and should be similar to those described in cost is obtained as follows
[18]. According to these authors, in the case of reinforced concrete 8 R
< r 1 d0 lD ðz; 0Þf ðzÞdz if n ¼ 1
buildings, the costs of nonstructural damage are less than the costs 0 DD1
hn ¼
of rehabilitation for damage values lower than 0.55; while for higher : r R d0 R d0 l ðd; xÞf ðd  xÞf
n 0 0 D DDn Dn1 ðxÞdx dd if n ¼ 2; 3; . . .
damage values costs of nonstructural damage are significantly
increased, up to 4 times the initial cost C 0 . These nonstructural costs ð31Þ
include the losses by contents, injury, loss of life, and interruptions. The first integral operates over the function f Dn1 ðÞ and satisfies
It is noted that from a damage value of 0.5, the cost of rehabilitation
for N P 2 that at ðn  1Þ-th load event, the damage should be less
is constant, equal to C 0 ð1 þ nÞ, since it is believed that from this level
than d0 ; while the second integral that operates over the function
of damage, the rehabilitation of the structure may not be technically
f DDn ðÞ, satisfies that at the occurrence of the n-th load, the damages
or economically viable, so the structure has to be demolished and
are also lower than the threshold d0 . In Eq. (31), the factor r n satis-
rebuilt. Here, a cost nC 0 is considered for demolition and cleaning,
fies that the area within the limits of integration of the probability
which in accordance with [19] n  0:15. The estimation of the losses
density function f Dn ðb; dÞ is equal to 1.
lD and lR requires further research.

3.4. Present value of expected total losses

3.3. Present value of expected losses associated with non-exceedance
of a damage threshold According to what was discussed above and considering Eqs.
(23), (22) is written in expanded form as follows:
The present value of expected total losses LNE due to the non-
Z ¼ C 0 þ LE þ LNE ð32Þ
exceedance of d0 is obtained as follows
 J.L. Alamilla, D. Tolentino / Structural Safety 72 (2018) 54–64 59

Substituting Eqs. (26) and (28) into Eq. (32), this latter is written 3.5. Present value of expected losses considering immediate
in expanded form as rehabilitation after each load inducing-damage

X
1 X
1 Z T X
1 As a particular case, we adopt the assumptions that the struc-
Z ¼ C0 þ Lnkþ1 pDk;n ect f Sn ðtÞdt þ Hn pD n tural capacity is immediately restored or rebuilt after each load
0
k¼1 n¼k n¼1
event that causes damage or failure and that T ! 1. Such assump-
Z T
tions were established in the seminal work of Rosenblueth [8]. In
 ect f Sn ðtÞdt ð33Þ
0 this case, Eq. (34) is recast as
X
1 X
1
 n1
For most types of renewal processes, the probability functions Z ¼ C0 þ Lnkþ1 pF k;n f S1 ðcÞf S ðcÞ
f Sn ðtÞ are difficult to evaluate in closed analytical form due to the k¼1 n¼k

sequential convolving process. For this reason, the integrals X1

 n1
R T ct þ Hn ð1  pF Þf S1 ðcÞf S ðcÞ ð39Þ
0
e f Sn ðtÞdt in Eq. (33) are also difficult to assess. However, the n¼1
value of these integrals can be obtained adequately by means of
Due to the fact that there is no accumulation of damage, Ln ðÞ
Monte Carlo simulation in which the load occurrence times X j ,
given by Eq. (27) is rewritten as follows
j ¼ 1; . . . n, are simulated. When T ! 1, the integral denotes the Z Z
1 1
Laplace transform of f Sn ðÞ and Eq. (33) adopts the following form Ln ¼ qn lR ðxÞf D1 ðxÞdx þ qn lD ðzÞf D1 ðzÞdz ð40Þ
1 1

X
1 X
1
 n1 Using Fig. 1 as a reference, it is possible to verify that in Eq. (40)
Z ¼ C0 þ Lnkþ1 pDk;n f S1 ðcÞf S ðcÞ
k¼1 n¼k
q1
n ¼ pF , f D1 ð1Þ ¼ pF . According to Fig. 3, lR ð1Þ ¼ C 0 ð1 þ nÞ and
X
1 lD ð1Þ ¼ C F , so that Eq. (40) is constant for all n ¼ f1; 2; . . .g and
 n1
þ Hn pD n f S1 ðcÞf S ðcÞ ð34Þ can be expressed as
n¼1
Ln ¼ C 0 ð1 þ nÞ þ C F ð41Þ
On the other hand, the occurrence of load events is commonly
represented by a homogeneous stochastic Poisson process with On the other hand, Eq. (29), which quantifies Hn losses, is trans-
parameter m0 . This parameter describes the mean number of events formed into Eq. (42), where r 1
n ¼ 1  pF . Due to the assumption
per unit of time. In this stochastic process, f Sn ðÞ is expressed that the structure is immediately restored whenever it experiences
explicitly by a Gamma probability density function (when n is inte- a damage, even if the structural damage is less than d0 , Eq. (42)
ger, this function is also known as Erlang probability function) and takes into account the rehabilitation costs lR ðÞ for this particular
the following integral is obtained case. Hn can also be viewed as the cost due to the fact that the
structure survives the n-the load.
Z  n Z Z
T
m0 Cðn; ðc þ m0 ÞTÞ 1e 1
ect f Sn ðtÞdt ¼ ð35Þ Hn ¼ r n ðlD ðdÞ þ lR ðdÞÞf DD1 ðdjyÞf Y ðyÞdy dd ð42Þ
0 c þ m0 CðnÞ 0 0

Here Cð; Þ and CðÞ denote the incomplete and complete e is a positive value close to zero so that the integral is carried
Gamma function, respectively. In particular, due to lack of memory out within the range [0, 1). The structure of this last equation jus-
of a Poisson process, any modified stochastic process is equal to an tifies the Rosenblueth [8] approach in which the damage losses are
ordinary renewal process, thus expressed in terms of intensities Y of dynamic loads. In compact
form, the previous equation is rewritten as follows

f S1 ðcÞf S
n1 n
ðcÞ ¼ f S ðcÞ ð36Þ Z 1e
1
Hn ¼ ðlD ðdÞ þ lR ðdÞÞf DD1 ðdÞdd ¼ C D þ C R ð43Þ
If the arrival of dynamic loads is described by a homogeneous 1  pF 0

Poisson process, Eq. (33) is transformed into the following The resulting values of Eqs. (41) and (43) are constant for all
n ¼ f1; 2; . . .g and can be expressed in terms of losses due to non-
X
1 X
1  n
m0 Cðn; ðc þ m0 ÞTÞ structural damage C D and rehabilitation costs C R , which integrate
Z ¼ C0 þ Lnkþ1 pDk;n
k¼1 n¼k
c þ m0 CðnÞ into Eq. (43) the functions of losses lD ðb; dÞ and lR ðb; dÞ, respectively.
X
1  n Based on the foregoing, Eq. (39) is expressed as follows
m0 Cðn; ðc þ m0 ÞTÞ
þ Hn pD n ð37Þ
n¼1
c þ m0 CðnÞ Z ¼ C 0 þ ðC 0 ð1 þ nÞ þ C F Þ
( )
If T ! 1, Eq. (37) adopts the form X 1 X1
ðn  1Þ! nk k  n1
 ð1  pF Þ pF f S1 ðcÞf S ðcÞ
ðk  1Þ!ðn  kÞ!
X
1 X
1  n X
1  n k¼1 n¼k
m0 m0 X
1
Z ¼ C0 þ Lnkþ1 pDk;n þ Hn pD n ð38Þ  n1
k¼1 n¼k
c þ m0 n¼1
c þ m0 þ ðC D þ C R Þ ð1  pF Þf S1 ðcÞf S ðcÞ ð44Þ
n¼1
Eq. (33) is generalized and can be applied to various types of
structures exposed to different types of stochastic loads; it also In the above equation, the sum within the bracket is the Laplace
allows quantifying objectively the losses associated with the total transform of the probability density function of time to the k-th
damages and have the advantage of incorporating in a single for- failure and is expressed as
mat the contribution of losses associated with all possible states X
1
 ðn  1Þ!
n1
of damage, including the collapse. The optimal solution is obtained g F k ðcÞ ¼ f S1 ðcÞf S ð1  pF Þnk pF k
ðcÞ
from the probability distributions of damages, which are indepen- n¼k
ðk  1Þ!ðn  kÞ!
dent of the time. In addition, these distributions can be related  k
 k1 pF
explicitly and objectively, jointly with design and maintenance ¼ f S1 ðcÞf S ðcÞ  ð45Þ
1  ½1  pF f S ðcÞ
indicators.
60 J.L. Alamilla, D. Tolentino / Structural Safety 72 (2018) 54–64

Substituting Eq. (45) into Eq. (44), this is expressed as follows representative realizations of the ground motion at the site. The
X
1  k simulation of each realization of the ground motion for a given
 k1 pF intensity described by f Y ðÞ is carried out in two steps. In the first
Z ¼ C 0 þ ðC 0 ð1 þ nÞ þ C F Þ f S1 ðcÞf S ðcÞ 
k¼1
1  ½1  pF f S ðcÞ step, the most likely magnitude and distance to the seismic source
X
1
 n1
are obtained by means of a disaggregation process [25]; in the sec-
þ ðC D þ C R Þ ð1  pF Þf S1 ðcÞf S ðcÞ ð46Þ ond step, a realization of a ground motion is obtained from a sim-
n¼1
ulation of a filtered white noise, whose non-stationary properties
After evaluating both sums in Eq. (46), Z is expressed as are obtained from evolutionary functions of amplitude and fre-
 quency for given magnitude and distance.
f S1 ðcÞ Here, the structural performance is recognized to depend on the
Z ¼ C 0 þ ½ðC 0 ð1 þ nÞ þ C F ÞpF þ ðC D þ C R Þð1  pF Þ  ð47Þ
1  f S ðcÞ interaction between structural dynamic properties and evolution-
ary properties of the stochastic loads (frequency content, duration,
If the occurrence of loads is described by a Poisson process with
time variation of instantaneous intensity). This implies that the
parameter m0 , Z is quantified as
structural capacity and the load demand are not separable, much
m0 pF m0 ð1  pF Þ less can they be treated as independent random variables. In this
Z ¼ C 0 þ ½C 0 ð1 þ nÞ þ C F  þ ðC D þ C R Þ ð48Þ
c c work, the quantification of the total structural cumulative damage
was performed using the damage index. The damage index is a
Under the specific conditions described above, Esteva [4] devel-
function of inelastic displacement during each cycle of loading
oped the first two terms, from left to right, of Eq. (48), while Haso-
and unloading. The maximum amplitudes di ¼ jy2i  y1i j of each i-
fer [7] obtained these two terms considering a renewal process as
th inelastic cycle that contribute to damage are measured at the
is indicated in Eq. (47). In the work of Rosenblueth [8], Eqs. (47)
top of the building. Here, y2i and y1i are the maximum displace-
and (48) were obtained and considered the losses due to failures,
but also those corresponding to damages in which the structure ments of each cycle (maximum to compression and tension). The
survives. These expressions have served as the starting point in relative increment Dd of damage associated with a given dynamic
the establishment of risk-based seismic design criteria. Eq. (48) load is quantified as
has the advantage that the optimization is relatively simple and XnT
is a function of the annual failure rate m0 pF , which does not change di I
Dd ¼ Xni¼1 ¼ ð49Þ
over time. F
d Fi
IF
i¼1

4. Case study The fatigue index value I is the sum of nT amplitudes of inelastic
deformations di ¼ y2i  y1i , i ¼ 1; . . . ; nT , associated with a given
The generalized formulation is exemplified by optimizing the dynamic load. If nT ¼ 0 then I ¼ 0. The failure fatigue index IF value
damage threshold above which a building located in a seismic zone is the result of the sum of nF 6 nT inelastic deformation amplitudes
is repaired or replaced whenever such damage threshold is dF i , i ¼ 1; . . . ; nF , leading the structural system to incipient failure.
exceeded. The studied system is a 10-story reinforced concrete Based on [26–28], incipient failure is obtained by scaling the
building. The nonlinear dynamic response is determined based dynamic load by an appropriate scale factor, so that the structural
on a plane structural frame, which is evaluated by means of Drain collapse corresponds to the start of the unbounded displacement
2D (computer program), developed by [20]. The cyclic behavior of on the top of the structural system. If this scale factor is less than
non-linear beams and columns is described by the constitutive or equal to one, then Dd ¼ 1. Here Dd takes values in the interval
function developed and implemented by [21] in Drain 2D. This [0, 1]. Dd ¼ 0 indicates no damage relative increment, while
constitutive function describes the nonlinear cyclic behavior of Dd ¼ 1 means the collapse of the structure.
bending moment vs rotation at the ends of structural elements, The damage probability distributions F Dn ðÞ are systematically
and considers the stiffness and resistance degradation in each cycle obtained from a sample of realizations of the structural damage
according to a cumulative damage index that follows the rule of process, from an undamaged state until collapse. This process
Miner. When computing the structural dynamic response, the
uncertainty in gravity loads (dead and live), and in the mechanical
properties of structural elements (geometric dimensions,
strengths, and stiffness), were considered. A sample random vector
of these properties was simulated to describe the designed struc-
ture in a likely real performance context in which its properties
are known in an uncertain manner. 100 structures with uncertain
properties were generated in this way, with a mean fundamental
period T ¼ 1:57 s, and a coefficient of variation of 0.04.
The occurrence process of seismic intensity is represented by a
stochastic Poisson process with parameter m0 ¼ 0:627 which is
associated with a minimum seismic intensity y0 ¼ 50 cm=s2 , spec-
ified in terms of the spectral ordinate corresponding to the mean
period of the structures. The m0 parameter describes the annual
mean number of seismic intensities greater than seismic intensity
y0 . The exceedance rate mðyÞ function is shown in Fig. 4, and it was
estimated according to the classic approach [22] developed by
Cornell [23] and Esteva [4]. This function is linked with f Y ðÞ by
the relation m0 f Y ðyÞ ¼ dmðyÞ=dy.
The ground motions are represented by seismic accelerations
simulated by a non-stationary stochastic process. The simulation
process is described in [24], which was adapted to perform Fig. 4. Seismic intensity rate function.
 J.L. Alamilla, D. Tolentino / Structural Safety 72 (2018) 54–64 61

describes how the structure enter sequentially into possible 1.0

damage states while at the same time is subjected to a likely
sequence of dynamic loads. A realization is characterized by a par-
ticular structure that is randomly obtained in this work from the 0.8
sample of 100 structures, and a likely random sequence of ground
motions of given intensities Y ¼ y, which are obtained from a sim-
ulation process. In general the realizations are obtained by Monte 0.6
Carlo simulations and the distributions F Dn ðÞ are obtained from the

p[D(t)>d]
sample of damage values associated with the n-th load of each d=0.1
realization. In this work we do not fit a suitable analytical function d=0.5
0.4 d=1.0
to describe such distributions, instead we work with the empirical
distributions obtained from such samples. To estimate the cumula-
tive damage value to the n-th load for each realization a dynamic
response analysis for the structure with intact mechanical proper- 0.2
ties is performed: the structure is subjected sequentially to two
dynamics loads separated by a period of time of several seconds,
so that the structure is not moving when the second load is 0.0
applied. The involved two loads are the ðn  1Þ-th and the n-th 0 100 200 300 400 500
dynamic loads. The damage state associated with the n-th load is time, t (years)
estimated by means of the following procedure. The ðn  1Þ-th load
is scaled by a factor so that the structural response associated with Fig. 5. Probability functions that a given damage threshold is exceeded at time t.

this load results in a given value of damage obtained by Eq. (49).

When the structure is subjected to the n-th dynamic load, the
structure can endure or increase its damage level, which is also cal-
culated by Eq. (49). This last value of damage is used as a given
value to evaluate the damage related to the (n þ 1Þ-th load. This
procedure is recursive up to reach the structural collapse, which
corresponds to the start of the unbounded displacement on the
top of the structural system. Finally, the cumulative damage at
the end of the k-th dynamic load is the result of the history of
the relative increments of damage Ddi , i ¼ 1; . . . ; m, which are
linked to a sequence of N loads that causes structural collapse.
Because the damage increments are relative to each dynamic load,
the sum of these increments is not equal to one. According to the
above, the cumulative damage at the k-th dynamic load, where
0 6 k 6 m, is estimated as
Pk
Ddi
dk ¼ Pi¼1
m ð50Þ
i¼1 Ddi
Fig. 5 shows the probability functions p½DðtÞ > d for different
levels of damage exceedance. These probabilities functions are
related to Eqs. (6) and (8). The probability that the structure
exceeds a given damage level at time t, is less than or equal to
the probability of exceeding minor levels of damage. In other
words, if dj > di , then p½DðtÞ > dj  6 p½DðtÞ > di . The undamaged
structure has the initial ratio T=T S  0:78; the lateral stiffness
damage costs lD ¼ lc þ le þ lin þ lf , are expressed as the sum of
losses per contents lc , direct economic losses from rentals le , losses
from injuries lin , and fatalities lf . The losses are estimated as fol-
2
lows: lc ¼ 0:5C 0 Dd, if 0 6 Dd 6 1; le ¼ 480ADd , where A is the total
2 4
floor area of the building in m2 ; lin ¼ 221ADd and lf ¼ 1228:5ADd .
These cost functions correspond to the damage index proposed by
[29]. However, for simplicity, here it is assumed that the costs are
associated in the same way with the values of damage given by Eq.
(50) and that they are a function of the relative increases of dam-
age Dd ¼ d  dI , in which dI is the initial damage at the n-th load. In
future works, it is necessary to develop cost functions of damage
that consider cumulative damage.
Fig. 6 shows the present value of the expected losses due to ini-
tial cost C 0 , non-structural damage costs and rehabilitation and
rebuild costs as a function of given possible thresholds of accep-
tance of damages d0 . These losses are normalized with respect to
C 0 . The expected cost LE decreases with d0 , and goes from a maxi-
mum loss of 0:28C 0 to value 0:06C 0 , approximately. The nonstruc-
tural costs LED contribute approximately 96% to LE . On the contrary,
LNE losses increase with d0 . Here, the costs associated with non-
1.5

decreases with the impact of dynamic loads and therefore Z C0 C0 C0

T=T S ! 1. Thus, this structure goes into levels of greater damages
faster than those of the second set. This is in accordance with the
Expected looses / C 0

fact that the slopes of the distribution function for time periods
below 100 years decrease with the damage levels, and the velocity 1.0
at which structures go into levels of greater damages decreases
with the increase in the level of damage.
On the other hand, the costs of damage include costs of repair
and/or rebuilding, loss of content, costs of injuries, costs associated
with fatalities and the corresponding economic impact of the 0.5
losses. Here, we have adopted the functions of damage reported
by Ang and De Leon [18], which were obtained on the basis of costs LE C 0 LNE C0 LE D C 0
of damages reported in Mexico City during the strong earthquake
occurred in 1985. The costs of repair lR are specified in terms of d*
the structural initial cost C 0 and the index structural damage d, 0.0
as follows: if lR ¼ 1:64C 0 d then 0 6 d < 0:5; if lR ¼ C 0 then 0.00 0.05 0.10 0.15 0.20
0:5 6 d 6 1. The function of rehabilitation and rebuild costs pre- Damage threshold
sents a discontinuity in d ¼ 0:5, since for damage values higher
than this value the structure is required to be replaced. However, Fig. 6. Expected losses as a function of damage threshold.
62 J.L. Alamilla, D. Tolentino / Structural Safety 72 (2018) 54–64

Fig. 7. Elements that contribute to the present value of expected losses associated with a damage threshold exceedance.

structural damages determine the form of the objective function Z,
since the repair costs are relatively small. The acceptable damage
threshold that minimizes the total losses is small, equal to

d0 ¼ 0:008. The Z function shows an abrupt fall around the optimal
value, which is basically due to the shape of f Dn ðdÞ. The shape of
f Dn ðdÞ is narrow for the first loads and takes small damage values,
meanwhile with n this function takes greater damage values and
tends to have a flat shape. f Dn ðdÞ determines the value of the prob-
abilities pDk;n , the value of expected costs Ln and Hn , and the contri-
bution to the present value of the expected total losses of each
term associated with each load. This can be corroborated in
Fig. 7 that shows the probabilities pDk;n and the present value of
 n1
the expected losses Ln;k ¼ Lnkþ1 pDk;n f S1 ðcÞf S ðcÞ for the first 100
loads and for the first three repairs, k ¼ 1; 2; 3. Every plot in
Fig. 7 shows the damage threshold values for d0 ¼ 0:001, 0.008
and 0.20. As was expected, the corresponding values of pDk;n for
d0 ¼ 0:008 and 0.20 are similar, but the values of these probabili-
ties are very different for d0 ¼ 0:20. However, the number of Ln;k
 J.L. Alamilla, D. Tolentino / Structural Safety 72 (2018) 54–64
takes relatively greater values for d0 ¼ 0:001 than for damage
thresholds 0.008 and 0.20. The Ln;k values are similar for d0 with
values of 0.008 and 0.20. Furthermore, the expected time to the
 several seminal works. It is shown that under certain hypothesis,
occurrence of d0 is equal to E½T D1  ¼ 20 years, which is obtained
as follows
X
1
 the loads occurrence are described by a Poisson process, these
E½T D1  ¼ pD1;n ðd0 ÞE½Sn  ð51Þ
n¼1
According to the exceedance rate shown in Fig. 4, the seismic

intensity associated with d0 is equal to 350 cm/s2. Actually, this
intensity may control the structural design. Furthermore, accord-
ing to the proposed approach developed by [25] which was imple-
mented to this particular soil site, seismic events that can cause

damage level exceedance d0 range from magnitudes of 6.8 to 8.3
with distances from 150 to 400 km. However, the most likely
events that can cause this exceedance are in the range of magni-
tudes from 7.7 to 8.0; while the range of distances is between
300 and 350 km, which correspond to seismic subduction events
off the southern Pacific coast of Mexico.
Finally, the theoretical result in which the acceptable damage
threshold or optimum threshold that minimizes the total losses
is relatively small is congruent in a context in which the owner will
take repairing actions if his structure has some level of structural
damage, even if it is small.
5. Discussion
The proposed approach leads us to minimize the expected con-
sequences from different possible damage scenarios, in which the
distribution of damages of each scenario is linked to their corre-
sponding detailed consequences. Many of the proposed design cri-
teria are mainly based on a reliability scheme, where the decision
parameters are obtained from the minimization of the correspond-
ing losses as a function of structural reliability. This latter indicates
that nonstructural damages are treated marginally or secondarily,
they are not treated in an explicit manner. Here, we minimize
losses associated with levels of damage, including structural
failures.
The proposed approach allows to consider two key points. The
first one is the dependence between increments of damage as a
function of how the dynamic loads enter sequentially. The second
one is the coupling and interaction between structural dynamic
properties and evolutionary properties of the stochastic loads (fre-
quency content, duration, time variation of instantaneous inten-
sity). This last point is extensively recognized in structural
engineering [26–28] where the damages are related to a structural
nonlinear behavior. Commonly these two points are underesti-
mated in structural life-cycle optimization problems.
Due to the explicit treatment of damages, a relationship
between the damage index and physical damage could be estab-
lished in order to have better quantification and identification of
physical expected damages and better policies of damage control.
The nonstructural damages are important. This specific issue was
not addressed in this work and requires further research. In addi-
tion, in future works, it is necessary to develop utility functions
or cost functions to quantify consequences that can take into
account other issues (risk attitudes, quality of life [30], value of life,
health, environmental damage and others [31]) which were not
considered herein.
6. Conclusions
This work has systematically presented the development of a
generalized formulation useful to quantify the present value of
63
expected losses of structural systems, which can accumulate dam-
age when they are exposed to dynamic load sequences. The pro-
posed approach extends and generalizes the ideas described in
our formulation may result in the models proposed by these works.
When the structural service planned time is relatively large and
models have the advantage that the optimization is relatively sim-
ple and is a function of the annual failure rate. The proposed
approach is, indeed, mathematically more elaborated, however
the optimization does not depend on time, but is only a function
of the load sequence. Furthermore, it allows answering to several
questions that are important for making decisions, for example:
what is the optimal damage threshold? What are the involved
costs? What is the expected time to the occurrence of this damage
threshold? The answers to these questions allow making better
decisions for design and maintenance of structures subjected to
uncertain loads.
Acknowledgements
The authors are grateful for the comments and criticism of the
anonymous reviewers which helped to improve the manuscript.
References
[1] Rosenblueth E. Towards optimum design through building codes. J Struct Div-
ASCE 1976;102(3):591–607.
[2] Benjamin JR, Cornell CA. Probability, statistics, and decision for civil
engineers. New York: MacGraw-Hill, Inc; 1970.
[3] Keeney RL, Raiffa H. Decisions with multiple objectives: preferences and value
tradeoffs. Cambridge: Cambridge University Press; 1993.
[4] Esteva L. Bases para la formulación de decisiones de diseño sísmico [Ph.D.
thesis]. National University of Mexico; 1968.
[5] Rosenblueth E, Mendoza E. Reliability optimization in isostatic structures. J
Eng Mech Div-ASCE 1971;97(6):1625–42.
[6] Rosenblueth E. Safety and structural design. In: Bresler B, editor. Reinforced
Concrete Engineering. New York: John Wiley & Sons; 1974. p. 407–516.
[7] Hasofer AM. Design for infrequent overloads. Earthquake Eng Struct 1974;2
(4):387–8.
[8] Rosenblueth E. Optimum design for infrequent disturbances. J Struct Div-ASCE
1976;102(9):1807–25.
[9] Rackwitz R. Optimization-the basis of code –making and reliability
verification. Struct Saf 2000;22:27–60.
[10] Joanni A, Rackwitz R. Cost-benefit optimization for maintained structures by a
renewal model. Reliab Eng Syst Saf 2006;93:489–99.
[11] Streicher H, Joanni A, Rackwitz R. Cost-benefit optimization and risk
acceptability for existing, aging but maintained structures. Struct Saf
2008;30:375–93.
[12] Rackwitz R, Joanni A. Risk acceptance and maintenance optimization of aging
civil engineering infrastructures. Struct Saf 2009;31:251–9.
[13] Takahashi Y, Der Kiureghian A, Ang AH-S. Life-cycle cost analysis based on a
renewal model of earthquake occurrences. Earthquake Eng Struct 2004;33
(7):859–80.
[14] Goda K, Hong HP. Optimal seismic design for limited planning time horizon
with detailed seismic hazard information. Struct Saf 2006;28:247–60.
[15] Garcia-Perez J, Castellanos F, Diaz O. Occupancy importance factor in
earthquake engineering. Eng Struct 2005;27:1625–32.
[16] Esteva L, Díaz-López OJ, Vásquez A, León JA. Structural damage accumulation
and control for life cycle optimum seismic performance of buildings. Struct
Infrastruct E 2016;12(7):848–60.
[17] Cox DR. Renewal theory. London: Methuen; 1962.
[18] Ang AH-S, De Leon D. Determination of optimal target reliabilities for design
and upgrading of structures. Struct Saf 1997;19(1):91–103.
[19] Montes-Iturrizaga R, Heredia-Zavoni R, Esteva L. Optimal maintenance
strategies for structures in seismic zones. Earthquake Eng Struct
2003;32:245–64.
[20] Powell GH. DRAIN 2D User’s Guide. Earthquake engineering research
center. Berkeley: University of California; 1973.
[21] Campos D, Esteva L. Modelo de comportamiento histerético y de daño para
vigas de concreto reforzado. In: 11th Congreso Nacional de Ingeniería Sísmica.
Veracruz, Mexico 1997;1:567–573.
[22] McGuire RK. Probabilistic seismic hazard analysis: early history Review.
Earthquake Eng Struct Dyn 2008;37:329–38.
[23] Cornell A. Engineering seismic risk analysis. Bull Seism Soc Am 58, 1583–1606.
[24] Alamilla J, Esteva L, García-Perez J, Diaz-Lopez O. Evolutionary properties of
stochastic models of earthquake accelerograms: their dependence on
magnitude and distance. J Seismol 2001;5(1):1–21.
64 J.L. Alamilla, D. Tolentino / Structural Safety 72 (2018) 54–64
[25] Alamilla J, Esteva L, García-Perez J, Diaz-Lopez O. Simulating earthquake
ground motion at a site, for given intensity and uncertain source location. J
Seismol 2001;5:475–85.
[26] Shome N, Cornell CA. Probabilistic seismic demand analysis of non linear
structures. Report No. RMS-35. Stanford University; 1999.
[27] Vamvatsikos D, Cornell CA. Incremental dynamic analysis. Earthquake Eng
Struct 2002;31:491–514.
[28] Alamilla JL, Esteva L. Seismic reliability functions for multistory frame and
wall-frame systems. Earthquake Eng Struct 2006;35:1899–924.
[29] Park YJ, Ang AH-S. Mechanistic seismic damage model for reinforced concrete.
J Struct Eng-ASCE 1985;111(4):722–39.
[30] Pandey MD, Nathwani JS, Lind CC. The derivation and calibration of the life
quality index (LQI) from economical principles. Struct Saf 2006;28:341–60.
[31] Ditlevsen O, Friis-Hansen P. Cosdt and benefit including value of life, health
and environmental damage measured in time units. Struct Saf
2009;31:136–42.