This article was downloaded by: [ETH Zurich]

On: 08 May 2014, At: 05:33

Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House,
37-41 Mortimer Street, London W1T 3JH, UK

Structure and Infrastructure Engineering:

Maintenance, Management, Life-Cycle Design and
Performance
Publication details, including instructions for authors and subscription information:
http://www.tandfonline.com/loi/nsie20

Investigation of the use of a Weibull model for the

determination of optimal road link intervention
strategies
a a
Nam Lethanh & Bryan T. Adey
a
Institute of Construction and Infrastructure Management, Swiss Federal Institute of
Technology (ETHZ), 8093, Zurich, Switzerland
Published online: 22 Jan 2013.

To cite this article: Nam Lethanh & Bryan T. Adey (2014) Investigation of the use of a Weibull model for the determination
of optimal road link intervention strategies, Structure and Infrastructure Engineering: Maintenance, Management, Life-Cycle
Design and Performance, 10:5, 684-696, DOI: 10.1080/15732479.2012.758641

To link to this article: http://dx.doi.org/10.1080/15732479.2012.758641

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained
in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no
representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the
Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and
are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and
should be independently verified with primary sources of information. Taylor and Francis shall not be liable for
any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever
or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of
the Content.

This article may be used for research, teaching, and private study purposes. Any substantial or systematic
reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any
form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://
www.tandfonline.com/page/terms-and-conditions
 Structure and Infrastructure Engineering, 2014
Vol. 10, No. 5, 684–696, http://dx.doi.org/10.1080/15732479.2012.758641

Investigation of the use of a Weibull model for the determination of optimal road link
intervention strategies
Nam Lethanh* and Bryan T. Adey1
Institute of Construction and Infrastructure Management, Swiss Federal Institute of Technology (ETHZ), 8093 Zurich, Switzerland
(Received 20 February 2012; final version received 9 September 2012; accepted 17 October 2012; published online 22 January 2013)

In this paper, a probabilistic model for the determination of optimal intervention strategies (OISs) for a road link composed
of multiple objects that are affected by gradual deterioration processes is investigated. The model is composed of a
deterioration part and a strategy evaluation part. In the deterioration part, a Weibull hazard function is used to represent the
deterioration of the individual objects, where the values of the model parameters are to be estimated using inspection data.
A threshold condition state (CS) for each object is defined, at which an intervention must be executed. The results of the
deterioration part are used as inputs in the strategy evaluation part, in which OISs for individual objects and for the link as a
whole are determined. The determination of the optimal strategies takes into consideration impacts on multiple stakeholders.
The model is demonstrated by determining the OISs for a fictive road link composed of one bridge and two road sections.
Downloaded by [ETH Zurich] at 05:33 08 May 2014

The main strengths of the methodology are that past deterioration is taken into consideration and that it is possible to
consider the execution of interventions simultaneously and, therefore, associated reductions in impacts that normally occur
when interventions are grouped. The main weakness of the methodology is that the condition of the objects is represented
using only two CSs, i.e. fully operational and not fully operational.
Keywords: Weibull analysis; total cost analysis; multi-stakeholder approach; road asset management

1. Introduction (1) estimate the average deterioration of all similarly

Inadequate performance of an object, e.g. a bridge or a classified objects or elements and (2) estimate the
road section, in a road link that results in the inability of deterioration of each object or element. For the former,
the link to provide an adequate level of service results in statistical methods are often used, and for the latter,
negative impacts on the stakeholders of the road. In order methods based on deterministic physical models are often
to manage the road link in a way to minimise negative used.
impacts on stakeholders, i.e. to determine optimal In this paper, a methodology for the determination of
intervention strategies (OISs), it is necessary to under- OISs for a road link composed of multiple objects, where
stand how the objects in the link will change over time the condition of each object is represented using two
and to determine the optimal times to execute an condition states (CSs), is proposed. The methodology is
intervention and the types of interventions to be executed composed of a deterioration part and a strategy evaluation
at these times.2 part. In the deterioration part, a Weibull hazard function is
The process of determining the impacts on stake- used in a probabilistic model to represent the deterioration
holders that are incurred when intervention strategies (ISs) of the individual objects, where the values of the model
are followed is often referred to as life cycle cost (LCC) parameters are to be estimated using inspection data.
analysis (Kobayashi & Kuhn, 2007; Woodward, 1997). A probabilistic deterioration model was selected to take
The results of analyses are then used to determine the into consideration the uncertainty in deterioration predic-
strategy that results in the lowest overall costs over the life tion, which is increasingly seen as both possible and
cycle of the object (Adey et al., 2010; Jido, Otazawa, & necessary in order to make appropriate decisions on when
Kobayashi, 2008; Kobayashi & Kuhn, 2007). One interventions should be executed (Frangopol, Kallen, &
important, but subtle, feature of the process of determining Noortwijk, 2004; Frangopol & Neves, 2008; Jido et al.,
the OIS is that it is dependent on the deterioration models 2008; Madanat, Bulusu, & Mahmoud, 1995; Nakat &
used to predict the deterioration of an object. Deterioration Madanat, 2008). It is seen as increasingly possible due to
models used for this purpose can be classified as those that the increasing ease with which sufficient data can be
collected, stored and processed in order to determine

*Corresponding author. Email: lethanh@ibi.baug.ethz.ch

q 2013 Taylor & Francis
 Structure and Infrastructure Engineering 685

correct probabilistic models (Ker, Lee, & Wu, 2008; between states can be described by a stochastic variable
Wang, Mahboub, & Hancher, 2005). It is seen as t [ ½0; 1
that represents the time to depart from CS1
increasingly necessary because the correct determination (Lancaster, 1990) and that the probability of transition of
of OISs can result in significant reductions in the negative an object from state 1 to state 2 can be represented by a
impacts related to road use and it is not possible to predict Weibull distribution function. The Weibull distribution
the deterioration of an object exactly, i.e. deterministi- function is used because it is not memoryless, over-
cally. Coupled with this increasing necessity is also an coming some of the criticisms of the widely used
increasingly wide spread use of probabilistic models in exponential distribution (Gertsbakh, 1997; Gertsbakh,
infrastructure management decision making (Frangopol 2000; Marquez, 2007), and since it has been found to be
et al., 2004; Frangopol & Neves, 2008), which has the a good representation of certain deterioration processes
additional benefit of increasing their acceptance in in the past (Agrawal, Kawaguchi, & Chen, 2010;
management decision making. For example, it is common Kobayashi & Kaito, 2010; Kobayashi, Kaito, & Lethanh,
that state-of-the-art bridge management systems such as 2010). It is noted that the suitability of this model should
KUBA and PONTIS use probabilistic models (FHWA, be checked using the data related to the specific objects
2005). The models used in these systems are often based in question.
on Markov chain theory, meaning that a range of discrete The deterioration and intervention (or renewal)
CSs are used to represent the physical condition of civil process is illustrated in Figure 1. Time tk is the duration
infrastructure objects; it is assumed that the transition that the object is in CS1. When the object reaches CS2,
probabilities are stationary and the historical performance intervention is required to bring it back to CS1, and the
Downloaded by [ETH Zurich] at 05:33 08 May 2014

of the object is not directly considered. cycle of deterioration and intervention process is repeated.
The results of deterioration part are used as inputs in As t is a stochastic variable, it has a probability
the strategy evaluation part, in which OISs for individual distribution function FðtÞ and a probability density
objects and for the link as a whole are determined. The distribution f ðtÞ. The probability of remaining in CS1
determination of the OIS takes into consideration impacts (hereafter referred to as survival probability) expressed by
on multiple stakeholders. To demonstrate the functioning survival function Fð ~ tÞ can be defined according to the
of the methodology, the OIS for a road link composed of value of failure probability FðtÞ as:
three objects is determined. ~ tÞ ¼ 1 2 FðtÞ:
Fð ð1Þ
The probability of the object being in CS1 until time t
2. Methodology and then entering CS2 for the first time during the interval
t þ Dt can be regarded as the hazard function, which is
2.1. Deterioration part given by:
To model deterioration, it is assumed that an object can
be in only one of the two CSs: (1) fully operational or (2) f ðtÞDt
li ðtÞDt ¼ ; ð2Þ
not fully operational; that the transition of the object ~ tÞ
Fð

Figure 1. Deterioration and intervention process.

 686 N. Lethanh and B.T. Adey

the probability that the object transitions from CS1 to CS2 independent from all other objects, is given by:
depend greatly on the elapsed time that it has spent in CS1,
i.e. the hazard function has a memory. Using the Weibull ln Lða; m : ts Þ
distribution function, the hazard function is given by: " #
X
S ð1 2 ds Þ ð2atm
s Þ
lðtÞ ¼ amt m21 ; ð3Þ ¼ :
s
þds {ln a þ ln m þ ðm 2 1Þln ts 2 atm
s }

where a is the so-called arrival density and m is the ð9Þ

acceleration or shape parameter (Lancaster, 1990).
The parameter a can be expressed in multiplicative where ds is a binary variable which has value of 1 when
form of the unknown parameter b and the characteristic CS2 is observed and 0 otherwise and ts is used to represent
variables (covariates) x, i.e. the factors that affect the rate of the evolution of time for object s.
deterioration, for example the daily traffic volume (DTV), For ease of mathematical manipulation, the logarithm
ambient temperature and pavement thickness. of both sides of Equation (9) is taken and often referred to
as the log-likelihood function.
X
M
a¼ bi x i ; ði ¼ 1; . . . ; MÞ; ð4Þ Y
S  ð12ds Þ  m ds
i¼1
Lða; m : ts Þ ¼ {F
tm
s f ðts Þ
s

where M is the total number of covariates and the value of Y

S  ð12ds Þ  ds
Downloaded by [ETH Zurich] at 05:33 08 May 2014

first covariate is 13. ¼ s Þ

expð2atm amtm21
s s Þ
expð2atm :
The probability density function f ðtÞ and survival s
~ tÞ of the Weibull hazard function are given by
function Fð ð10Þ
Equations (5) and (6), respectively:

f ðtÞ ¼ amt m21 expð2at m Þ; ð5Þ

2.2. Evaluation part
~ tÞ ¼ expð2at Þ:
Fð m
ð6Þ 2.2.1. Steps
Once the deterioration curves for each object are
In order to obtain the values of the parameters a and m, the determined, the OIS for the link can be determined. The
maximum likelihood estimation method can be used, steps required to do this are given in Table 1. More in-
where the parameter values ðu1 ¼ a; u2 ¼ mÞ, which depth explanation of some steps is given in the remainder
maximise the logarithmic likelihood function (9) of Section 2.2. Their exact location is also indicated in
u^ ¼ ðu^1 ; u^2 Þ, i.e. satisfy: Table 1.

› ln LðJ; u^Þ
¼ 0; ði ¼ 1; 2Þ; ð7Þ
›ui 2.2.2. Determination of OIS for each object
An IS is used to ensure that an object does not enter CS2
where L is the maximum likelihood function, J is the set where an adequate level of service would not be provided,
of observed data and the most likely values of u^ ¼ ðu^1 ; u^2 Þ may be to execute a preventive intervention at time z and
are estimated by using numerical iterative procedures such to execute a corrective intervention if the object enters CS2
as Newton method 4 for simultaneous equations (Equation before time z, i.e. in the period from ð0; z
. This type of IS
(8)) (Kelley, 1999). In order to test these values for is often referred to as age replacement (Gertsbakh, 1997;
statistical significance, the probabilistic t-test and the Gertsbakh, 2000). The explanation of the probability of
asymptotic covariance matrix (Equation (8)) 5 can be used transitioning from CS1 to CS2 is given in Section 1.
(Cramer, 1946). When following this IS, one can envision that impacts

 are incurred by stakeholders in two ways:
X
^
› ln LðJ; uÞ 21
ðu^Þ ¼ : ð8Þ
›u›u 0 (1) During the execution of interventions (ICu; p )6: e.g.
the owner has to pay for the manual labour required to
To be clear about the maximum likelihood estimation execute the intervention, and the user has additional
method, the likelihood of the values of the model travel time due to the required detours.
parameters, given the set of observed data J, e.g. CSs, (2) When the object is in CS2 but the execution of the
time to failure of all similar object, sðs ¼ 1; . . . ; SÞ and intervention has not yet begun (SCu ): e.g. the owner
assuming that the deterioration of each object is has to pay for the manual labour required to erect
 Structure and Infrastructure Engineering 687

Table 1. Steps to determine the OIS for the link.

Step Description Section

1 Calculate the impacts when an object is in CS2 and no intervention is being executed, SCu , and the impacts when an 2.2.2
intervention is being executed, ICu .
2 Determine the OIS for each object, e.g. for every t*k an intervention should be executed. 2.2.2
3 Determine the time to intervention for each object taking into consideration its actual condition (e.g. optimal 2.2.2
intervention time is 10 years, but the object has already been in operation 6 years, then the time to intervention is 4
years).
4 Determine the types of ISs to be investigated.
5 Determine the values of the reduction factors to be used for each investigated type of IS. 2.2.3
6 Select a type of IS
7 Determine the time of the next intervention to be executed on the road link, t
8 Estimate Pthe


total impact from time t to the time when the execution of the next intervention. 2.2.4
9 Check Kk¼1 icp;l k ðtk Þ, if # B ðt k Þ, then all impacts for planned interventions in t k are within the set limits and
p;l

therefore all proposed interventions are executed, then move to step 11. Otherwise, move to step 10.
10 Use priority rule to identify the interventions to be executed. 2.2.5
10.1 Order the objects that are candidates for intervention in decreasing order of contribution to impact incurred due
to arriving in CS2 in year tk .
10.2 Select first object, k ¼ 1, ICp;l ¼ 0.
10.3 Check ICp;l k , B ðt k Þ, if yes, object k is selected, go to 10.4. if no, object k is not selected and deferred to next
p;l

intervention time, go to 10.5.

Downloaded by [ETH Zurich] at 05:33 08 May 2014

10.4 Set ICp;l p;l p;l

k ¼ ICk þ IC , B ðt k Þ ¼ B ðt k Þ 2 ICk , and k ¼ k þ 1 and then go to 10.3.
p;l p;l p;l

10.5 Set k ¼ k þ 1 and then go to 10.3.

11 Go to step 7 if the end of the investigated time period has not been reached, otherwise go to step 12.
12 If all types of ISs have been analysed, select the IS with lowest total impacts as the OIS and go to step 13. If not all
types of ISs have been analysed, select another type of IS and go to step 7.
13 Stop.
Note: t*k is referred to as optimal intervention time for object k, corresponding to z * in Equation (19). It is also noted that the notation k for object in this section is different from s
in Section 2.1, e.g. k can be road section or bridge, while s refers to objects with similar structural characteristics.

signs to reduce the number of lanes in use on a bridge, vehicle operation cost during the execution of an
and the user has additional travel time due to the intervention can be estimated as a function of daily traffic
congestion that this restriction would cause. volume, gasoline unit price, type of vehicle, condition of
road, etc. (Kumares & Samuel, 2007). Values of icu;pl and
This is illustrated in Figure. 2, where tA is the time the scul can be either positive or negative.
object enters CS2, tB denotes the start of the intervention The expected total impact incurred by stakeholders
and tC denotes the end of the intervention, where the between when an object enters CS2 and when an
object is restored to CS1. intervention is started is given by Equation (12). The
IC and SC are given by Equations (11a) and (11b), expected total impact incurred when an object is in CS2
respectively.

X
L
ICu;p ¼ icu;
l ;
p
ð11aÞ
l¼1

X
L
SCu ¼ scul ; ð11bÞ
l¼1

where icu;l
p
and scul are the impacts incurred by each
stakeholder group l ¼ ð1; . . . ; LÞ. The superscripts u and p
are referred to as for ‘unplanned’ and ‘planned’. As the
same type of intervention is applied in both two cases, it is
icu; p ¼ icu ¼ icp .
Impacts incurred by each stakeholder group l can be
estimated by the use of empirical models (Adey, Lethanh,
& Lepert, 2012; Kumares & Samuel, 2007). For example, Figure 2. Graphical representation of impact IC and SC.
 688 N. Lethanh and B.T. Adey

Table 2. Object information.

Objects Type Material Length (m) Width (m) DTV vehicle

1 Road section Asphalt concrete 290 5.7 600
2 Bridge Concrete 25 6.5 600
3 Road section Asphalt concrete 160 5.7 600

and an intervention is being executed is given by Equation of total impact TCa ðzÞ is given by:
(13). The first part of Equation (13) is used to estimate the
expected impact due to intervention if the object reaches to
CS2. The second part of Equation (13) is used to estimate X
ð 
w 2rt
the expected impact if the object survived until time z and TC ðzÞ ¼ VðzÞ þ
a
f ðtÞTC ðz Þ e
a
dt
an intervention is executed. Similar formulations can be zw
found in Gertsbakh (1997, 2000) and Lethanh (2009). X
þ e2rz ~ TCa ðz w Þ;
FðzÞ ð14Þ
zw
ðz
Downloaded by [ETH Zurich] at 05:33 08 May 2014

ESCðzÞ ¼ SCu f ðtÞ expð2rtÞ dt; ð12Þ

0
where VðzÞ ¼ ESCðzÞ þ EICðzÞ and a is intervention
type to be followed. The second and third polynomials
of Equation (14) represent the recursive form of
Bellman equation in dynamic programming (Bachmann
ðz & Konik, 1984; Bellman, 1955; Howard, 1960; Howard,
EICðzÞ ¼ ICu f ðtÞ expð2rtÞ dt 1971) and represent the expected total impact from the
0 next investigated time interval. As the same type of
~
þ FðzÞIC p
expð2rzÞ; ð13Þ intervention a will be repeated over and infinite time
horizon and expected impact in each interval is
considered to be equivalent, it is approximated that
TCa ðzÞ ¼ TCa ðz w Þ, and therefore, Equation (14) can be
where f ðtÞ represents the probability of entering CS2 expressed as:
before a planned intervention is executed and FðzÞ ~
represents that probability of surviving until the time
when a planned intervention is executed. r is the discount ðz
rate.7 TCð0 : zÞ ¼ f ðtÞ{SCu þ ICu þ TCð0 : zÞ} expð2rtÞ dt
In order to determine the OIS, it is necessary to 0 ð15Þ
estimate the total impact related to each IS. If an OIS is to
~
þ FðzÞ{IC p
þ TCð0 : zÞ} expð2rzÞ:
be evaluated over a fixed time period, the net present value

Table 3. Deterioration information.

Deterioration
parameters CS definition Other impacts related to
entering CS2 (SCu )
Objects m a CS1 CS2 (mu)
1 2.010 0.020 Average roughness of less Average roughness greater or 20,000
than 100 mm/m equal to 100 mm/m
2 2.130 0.003 A wearing index value A wearing index valuea of greater than or 5000
of less than 0.75 equal to 0.75
3 2.020 0.020 Average roughness of less Average roughness greater or 10,000
than 100 mm/m equal to 100 mm/m
a
Wearing index value represents deterioration of bridge (Brodsky et al., 2006).
 Structure and Infrastructure Engineering 689

In order to obtain an explicit form of TC, it is

Owner impacts related to a

unplanned intervention
necessary to define GðzÞ and LðzÞ as:
ðz

ICu (mu)
GðzÞ ¼ f ðtÞ expð2rtÞ dt

72,217

10,375

39,456
0
ðz
¼ amt m21 expð2at m 2 rtÞ dt; ð16Þ
0

~ expð2rzÞ ¼ expð2az m 2 rzÞ:

LðzÞ ¼ FðzÞ ð17Þ

and substituting these equations into Equation (15) gives:

Owner impacts related to a
planned intervention

ðSCu þ ICu ÞGðzÞ þ ICp LðzÞ

TCð0 : zÞ ¼ : ð18Þ
ICp (mu)
72,217

10,375

39,456
1 2 GðzÞ 2 LðzÞ

The objective function TCð0; zÞ from which the minimum

expected total impact of all ISs, and therefore, the OIS is
then given by:
Downloaded by [ETH Zurich] at 05:33 08 May 2014

TCð0Þ ¼ min{TCð0 : zÞ}: ð19Þ

Restored object to CS1 with

Restored object to CS1 with

Restored object to CS1 with

The OIS, i.e. the optimal interval z * , is then estimated by

taking the first derivative of Equation (18) and by setting it
100% probability

100% probability

100% probability

equal to 0, as shown in the following equation:

Effectiveness

d TCð0 : zÞ cðzÞ
¼ ¼ 0; ð20Þ
dz {1 2 gðzÞ 2 LðzÞ}2

where

cðzÞ ¼ ðSCu þ ICu ÞG0 ðzÞ þ ICp L0 ðzÞ

Resurfacing: resurfacing with 5 cm thickness of asphalt concrete

Resurfacing: resurfacing with 5 cm thickness of asphalt concrete

þ SCu {LðzÞ0 GðzÞ 2 G0 ðzÞLðzÞ}; ð21Þ

Intervention

dGðzÞ dLðzÞ
GðzÞ0 ¼ and LðzÞ0 ¼ :
dz dz
When cðzÞ ¼ 0, the optimal interval z * has been found. A
cover without exposing the reinforcement

numerical algorithm to solve Equation (18) is explained in

Reconstruction: removal of the concrete

new chloride-free concrete cover layer

Appendix.
In the determination of the OIS, it is assumed that an
intervention is executed when an object is in CS2 and than
an intervention restores the object with 100% certainty to
CS1.
Intervention information.

of the same thickness

and the addition of a

2.2.3. Determination of the variations in impacts due to

the timing of multiple interventions
The determination of the OIS for the road link requires
Types

special attention to the variations in impacts due to the

timing of multiple interventions, e.g. executing two
interventions at the same time is less expensive than
executing the same two interventions separately. In this
Table 4.

Objects

methodology, these impacts are expected when multiple

interventions that are simultaneously executed are
1

3
 690 N. Lethanh and B.T. Adey

Table 5. Investigated IS types.

Objects with interventions executed Objects with interventions executed Reduction factors for objects
Strategy type simultaneously alone with simultaneous interventions
1 – 1, 2, 3
2 1, 2 3 0.84
3 1, 3 2 0.90
4 2, 3 1 0.85
5 1, 2, 3 – 0.90

estimated by multiplying the impacts expected if only one where f k ðzÞ is the expected failure probability of object k
intervention is executed with appropriate reduction in an elapsed time of zðz ¼ ½t; tk 2 t
Þ. t is the start time of
factors. For example, if an intervention of type A is the investigation and tk is the time of intervention for
expected to cost 10 monetary units (mu) if executed alone, object k. The terms icu;l u;l
k and sck are impacts incurred by
then two interventions of type A would cost 20 mu if stakeholder group l if object k enters CS2.
executed at entirely different times. If, however, two
interventions of type A were to be executed simul-
taneously and had a reduction factor of 0.9, then they
Downloaded by [ETH Zurich] at 05:33 08 May 2014

would cost 18 mu (20 £ 0.9). 3. Example

3.1. Description
The use of the methodology is demonstrated by
2.2.4. Determination of intervention candidates determining the OIS for a fictive road link consisted of
The time to the next intervention is the minimum of the one bridge and two road sections when there is no budget
times to intervention of each object t*k ðk [ KÞ. The constraint and when there is a budget constraint of 3500 mu
number of objects to be considered in the year of next per year. The information related to the object, to
intervention is deterioration and to the interventions to be executed is
given in Tables 2– 4, respectively. The investigated ISs and
ðK
¼ arg min {tk }Þ the reduction factors to be used in the estimation of the
k[K impacts when multiple interventions are executed
.
simultaneously are given in Table 5. All numbers are
fictive but are considered sufficient to demonstrate the
methodology. A value of 2% was used as the discount rate,
2.2.5. Using priority rule
but to investigate the effect of variations in the discount
The determination of the OIS for the road link requires rate the analyses were carried out with discount rates of 6%
special attention to the constraints on impacts, e.g. limits and 10%. The probabilities of each object staying in CS1
on the maximum financial resources that can be allocated over time (without the execution of an intervention) are
to all objects or the maximum travel time that can be shown in Figure 3.
incurred due to interventions on all objects in a specific
time period. These constraints are dealt with in this
methodology by implementing a priority rule that selects
some interventions to be postponed to future time periods
if their simultaneous execution results in the exceedance of 1.0
Object 1
an impact constraint. To determine if the value of an Object 2
Survival probability

0.8 Object 3
impact is exceeded in a time period, the summation over
the investigated time period is made of the probabilities of 0.6
the object being in each CS at each instant of time
multiplied with the impact if the object were in that CS 0.4
over the specific time period. As shown in the following
0.2
equation for the impacts on the stakeholders:
0.0
ð tk 2t X
L   0 5 10 15 20 25 30
Dk ð
tk Þ ¼
u;*
icu; l u; l
k þ sck f k ðzÞ expð2rzÞ dz;
t Time (years)
l¼1
ð22Þ Figure 3. Deterioration.
 Downloaded by [ETH Zurich] at 05:33 08 May 2014

Table 6. OISs under five strategy types (discount rate 2%).

No budget constraint Budget constraint of 3500 mu/year

Average Average
Intervention Intervention annual Reduction in average annual Intervention Intervention annual Reduction in average annual
Strategy duration interval impacts impacts compared to OIS of duration interval impacts impacts compared to OIS of
type Object (days) (years) (mu) type 1 (mu) (days) (years) (mu) type 1 (mu)
1 1 30 19 3738 30 29 4251
2 8 25 205 8 25 205
3 15 16 2644 15 16 2644
Total 6587 0 7100 0
2 1þ2 30 23 2916 30 26 3441
3 15 16 2644 15 16 2644
Total 5560 1027 6085 1015
3 1þ3 30 17 6140 30 23 6492
2 8 25 205 8 26 728
Total 6345 241 7220 2 120
Structure and Infrastructure Engineering

4 1 30 19 3739 30 19 3739
2þ3 15 23 1813 15 28 2119
Total 5552 1034 5858 1242
5 1þ2þ3 30 24 4548 30 25 5152
Total 4548 2038 5152 1948
691
 Downloaded by [ETH Zurich] at 05:33 08 May 2014

692

Table 7. OISs under five strategy types (discount rate 6%).

No budget constraint Budget constraint of 3500 mu/year

Average Average
Intervention Intervention annual Reduction in average annual Intervention Intervention annual Reduction in average annual
Strategy duration interval impacts impacts compared to OIS of duration interval impacts impacts compared to OIS of
type Object (days) (years) (mu) type 1 (mu) (days) (years) (mu) type 1 (mu)
1 1 30 21 1033 30 21 1033
2 8 29 49 8 29 49
3 15 17 763 15 17 763
Total 1845 0 1845 0
2 1þ2 30 26 785 30 26 785
3 15 17 763 15 17 763
Total 1545 300 1545 300
3 1þ3 30 19 1681 30 19 1681
N. Lethanh and B.T. Adey

2 8 29 49 8 29 49
Total 1730 115 1730 115
4 1 30 21 1033 30 21 1033
2þ3 15 26 487 15 26 487
Total 1520 325 1520 325
5 1þ2þ3 30 27 1212 30 27 1212
Total 1212 633 1212 633
 Downloaded by [ETH Zurich] at 05:33 08 May 2014

Table 8. OISs under five strategy types (discount rate 10%).

No budget constraint Budget constraint of 3500 mu/year

Average Average
Intervention Intervention annual Reduction in average annual Intervention Intervention annual Reduction in average annual
Strategy duration interval impacts impacts compared to OIS of duration interval impacts impacts compared to OIS of
type Object (days) (years) (mu) type 1 (mu) (days) (years) (mu) type 1 (mu)
1 1 30 23 520 30 23 520
2 8 34 21 8 34 21
3 15 18 399 15 18 399
Total 940 0 940 0
2 1þ2 30 30 372 30 30 372
3 15 18 399 15 18 399
Total 771 169 771 169
3 1þ3 30 21 839 30 21 839
2 8 34 21 8 34 21
Total 860 80 860 80
Structure and Infrastructure Engineering

4 1 30 23 520 30 23 520
2þ3 15 30 231 15 30 231
Total 751 189 751 189
5 1þ2þ3 30 32 575 30 32 575
Total 575 365 575 365
693
 694 N. Lethanh and B.T. Adey

3.2. Results reduction factors and to define the types of ISs that should
When there is no budget constraint and a discount rate of be investigated in a way that does not result in a
2% is used, the OIS executes interventions on all three combinatorial explosion of the number of OISs (i.e. pro IS
objects every 24 years for an average impact of 4548 mu/ type) that need to be determined before the OIS for the
year (Table 2). When there is a budget constraint of road link can be determined.
3500 mu/year, the OIS executes interventions on all three A potential weakness of the methodology is that it only
objects every 25 years for an average impact of 5152 mu/ uses binary states to represent the condition of the objects,
year. The OISs in both cases are of type 5 and result in making it not possible to distinguish between different
savings of 2038 and 1948 mu/year, respectively, when types of interventions for each object, something which is
compared to the OISs of type 1. The OISs of each type of ISs done in many existing methodologies, albeit on the single
under each budget scenario, as well as the intervention object level. This simplification makes it necessary to
duration per object, the intervention interval per object, the make a number of broad approximations which result in a
average annual impacts and the reduction in average annual significant loss of information in the determination of
impact compared to the OIS of type 1, are given in Table 6. optimal strategies, e.g. it is necessary to assume that road
When there is no budget constraint and discount rates users have the same level of service over the entire range
of 6% and 10% are used, the OIS executes interventions on of physical condition of a road section between as new and
all three objects every 27 years and 32 years, respectively, failed. It is believed that models that make it possible to
for an average impact of 1212 and 575 mu/year (Tables 7 take into consideration more than two CSs may be more
and 8). When there is a budget constraint of 3500 mu/year, suitable for the determination of OISs for road sections
Downloaded by [ETH Zurich] at 05:33 08 May 2014

the OIS in each case is the same. This is because the affected by gradual deterioration processes. The investi-
budget of 3500 mu/year no longer has an effect on the gated methodology may be more suitable when attempting
interventions that can be executed in each year. It can be to determine OISs for a road link affected by the processes
seen by comparing the values given in Tables 6 – 8 that the that result in sudden deterioration, such as flooding, where
use of different discount rates in the range chosen has no it is more appropriate to view infrastructure objects as
effect on the optimality of the ISs but has an effect on the fully operational or not fully operational.
average time between interventions. The use of increas-
ingly higher discount rates increasingly enlarges the
average time between interventions.
4. Conclusions
In this paper, a probabilistic model for the determination of
OISs for a road link composed of multiple objects is
3.3. Discussion investigated. In the model, the performance of each object
It can be seen from the example that the proposed Weibull that comprises that link is described by two CSs. The
model can be used to determine OISs for road links transition time between states is considered as a stochastic
consisted of multiple objects, allowing for consideration of variable, which can be represented by a model based on
the past deterioration of each object (deterioration part) the Weibull hazard function. The parameters of Weibull
and the changes in the impacts that occur due to the timing hazard function are to be estimated using inspection data.
of multiple interventions (through the reduction factors in The determination of OISs is carried out based on the
the evaluation part). evaluation of total impacts incurred by all stakeholders.
It can, therefore, be also seen that the investigated An example was shown to demonstrate the applicability of
methodology can be used to determine OISs on road links the methodology.
that include grouped interventions, and, therefore, interven- The main strengths of the methodology were that
tions on individual objects even if it is not the right time deterioration history was taken into consideration and that
(either earlier or later) according to their ‘own’ OIS. For it was possible to consider the execution of interventions
example, the OIS for the road link (with a 2% discount rate) simultaneously and therefore associated reductions in
actually shows that it is better to execute an intervention on impacts that normally occur when interventions are
object 1 every 24 years even if its ‘own’ OIS indicates that it grouped. The main weakness of the methodology was
is best to execute an intervention every 19 years. This means that the condition of objects is represented using only two
that without consideration of the grouping of interventions, CSs, i.e. fully operational and not fully operational.
something not done in many existing methodologies, it is in Future work should be focused on the testing of this
many cases not possible to determine the OIS. methodology on real road link with many more objects.
Although the methodology is only demonstrated using This would include the investigation of the sensitivity of
a road link consisted of three objects, it can be potentially the OIS determined using this methodology to changes in
extended to a road link consisted of many more objects. the values of the deterioration model parameters and of the
The main challenges with this are to determine the correct impacts associated with interventions.
 Structure and Infrastructure Engineering 695

Acknowledgements Frangopol, D.M., Kallen, M.J., & Noortwijk, J.M.V. (2004).

The authors thank ERANET-ROAD a research-oriented entity under Probabilistic models for life-cycle performance of deterio-
the directorship of European commission, for the financial support of ration structures: Review and future directions. Progress in
this work, completed as part of the project entitled ‘Stakeholder Structural Engineering and Materials, 6(4), 197– 212.
Benefits and Road Intervention Strategies (SABARIS)’. Frangopol, D.M., & Neves, L.A.C. (2008). Optimization of life-
cycle maintenance strategies under uncertainties: Role of
inspections trends in computational structures technology.
Stirlingshire: Saxe-Coburg Publications.
Notes Gertsbakh, I.B. (1997). Models of preventive maintenance.
1. Email: adey@ibi.baug.ethz.ch. Amsterdam: North-Holland Publishing Company.
2. It is noted at the outset of the paper that the term ‘optimal’ can Gertsbakh, I. (2000). Reliability theory with applications to
be understood as referring to the ‘best’ IS among possible ones. preventive maintenance. New York, USA: Springer.
3. The function assumed for a can take different forms in Gruber, J. (2007). Cost-benefit analysis. Public finance and
regression analysis, e.g. exponential (Lancaster, 1990). public policy. Chapter 8 (pp. 201– 223), New York, USA:
4. Newton method is a method to find the successively better Worth Publishers.
approximations to the roots of a real-valued function. Howard, R. (1960). Dynamic programming and Markove
5. Values of t-test should be greater than 1.96 for 95% processes. Cambridge, MA: The Massachusetts Institute of
confidence and the covariance matrix should be non-singular. Technology.
6. The superscripts u and p refer to ‘unplanned’ and ‘planned’, Howard, R.A. (1971). Dynamic probabilistic systems, volume II:
respectively. As the same type of intervention is selected for Semi-Markov and decision processes. New York: Wiley.
both preventive and corrective intervention, it is Jido, M., Otazawa, T., & Kobayashi, K. (2008). Optimal repair
ICu; p ¼ ICu ¼ ICp . and inspection rules under uncertainty. Journal of Infra-
7. The discount rate is a variable used to help balance impacts structure Systems, 14(2), 150– 158.
Downloaded by [ETH Zurich] at 05:33 08 May 2014

that occur at different periods of time. The value of the Kelley, C.T. (1999). Iterative Methods for Optimization.
discount rate can greatly affect the present value of a future Frontiers in applied mathematics. Philadelphia: SIAM.
impact. The proper discount rate should represent the Ker, H.-W., Lee, Y.-H., & Wu, P.-H. (2008). Development of
opportunity cost of what else could be accomplished with fatigue cracking prediction models using long-term pave-
those same resources (Gruber, 2007). The higher the ment performance database. Journal of Transportation
discount rate, the lower the value of impacts that occur in Engineering, 134(11), 477– 482.
the future when compared with the value of impacts incurred Kobayashi, K., & Kaito, K. (2010). Random proportional
today. Usually, the higher the discount rate the more likely it Weibull hazard model: Application to large-scale infor-
is that interventions will be postponed. As small changes in mation systems. Facilities, 29(13/14), 1– 21.
the value of the discount rate can have a large effect on the Kobayashi, K., Kaito, K., & Lethanh, N. (2010). Deterioration
valuation of ISs, it is normally suggested to include it in forecasting model with multistage Weibull hazard functions.
appropriate sensitivity analyses before decisions are made. Journal of Infrastructure Systems, 16(4), 282– 291.
Kobayashi, K., & Kuhn, K. (2007). Decentralized life-cycle cost
evaluation and aggregated efficiency. The management and
References measurement of infrastructure: Performance, efficiency and
Adey, B.T., Herrmann, T., Tsafatinos, K., Luking, J., Schindele, innovation. Northampton, MA: New Horizons in Regional
N., & Hajdin, R. (2010). Methodology and base cost models Science.
to determine the total benefits of preservation interventions Kumares, C.S., & Samuel, L. (2007). Transportation decision
on road sections in Switzerland. Structure and Infrastructure making: Principles of project evaluation and programming.
Engineering, 8(7), 639– 654. Hoboken, New Jersey: Wiley.
Adey, T.B., Lethanh, N., & Lepert, P. (2012). An impact Lancaster, T. (1990). The econometric analysis of transition data.
hierarchy for the evaluation of intervention strategies for New York: Cambridge University Press.
public roads. 4th European Pavement and Asset Management Lethanh, N. (2009). Stochastic optimization methods for infra-
Conference, 5– 7 September, Malmo, Sweden. structure management with incomplete monitoring data (PhD
Agrawal, A.K., Kawaguchi, A., & Chen, Z. (2010). Deterioration thesis). Graduate School of Engineering, Kyoto University.
rates of typical bridge elements in New York. Journal of Madanat, S., Bulusu, S., & Mahmoud, A. (1995). Estimation of
Bridge Engineering, 15(4), 419– 429. infrastructure distress initiation and progression models.
Bachmann, W., & Konik, D. (1984). On stabilization of Journal of Infrastructure Systems, 1(3), 146–150.
decentralized dynamic output feedback systems. Systems Marquez, A.C. (2007). The maintenance management frame-
and Control Letters, 5(2), 89– 95. work: Models and methods for complex systems mainten-
Bellman, R. (1955). Equipment replacement policy. Journal of the ance. Springer series in reliability engineering, London:
Society for Industrial and Applied Mathematics, 3(3), 133–136. Springer.
Brodsky, G., Muzykin, R., Brodskaia, E.S., Ponomarev, Y.A., Nakat, Z., & Madanat, S. (2008). Stochastic duration modeling of
Yenyutin, Y.A., & Vlasova, M.S. (2006). Analysis of pavement overlay crack initiation. Journal of Infrastructure
parameters of structure deterioration models within the Systems, 14(3), 185– 192.
Moscow bridge management system. Structure and Infra- Wang, Y., Mahboub, K.C., & Hancher, D.E. (2005). Survival
structure Engineering, 2(1), 13 – 21. analysis of fatigue cracking for flexible pavements based on
Cramer, H. (1946). Mathematical methods of statistics. long-term pavement performance data. Journal of Transpor-
Princeton, NJ: Princeton University Press. tation Engineering, 131(8), 608– 616.
FHWA (2005). Bridge management systems: Meeting the Woodward, D.G. (1997). Life cycle costing-theory, information
challenges of managing bridge assets. FOCUS: Accelerating acquisition and application. International Journal of Project
Infrastructure Innovations, FHWA-HRT-06-021. Management, 15(6), 335– 344.
 696 N. Lethanh and B.T. Adey

Appendix: Solution to Gamma function Substituting Equations (A.4) and (A.5) into Equation (A.1), the
Solution to Gamma function in the equation Cð0; zÞ following results are obtained:

ðSC þ ICÞGðzÞ þ ICLðzÞ ðSC þ ICÞ½GðzÞ þ LðzÞ 2 1
þ SC þ IC 2 SCLðzÞ

Cð0 : zÞ ¼ ; ðA:1Þ Cð0; zÞ ¼
1 2 GðzÞ 2 LðzÞ 1 2 LðzÞ 2 GðzÞ
SC þ IC 2 SCLðzÞ
where GðzÞ and LðzÞ functions are defined as follows: ¼ Ðz 2 ðSC þ ICÞ:
r 0 LðtÞ dt
ðz
ðA:6Þ
GðzÞ ¼ f ðtÞ expð2rtÞ dt
0
ðz In order to solve the integration of function LðzÞ, the general form
of expanding the integration into following discrete series is
¼ amt m21
expð2at 2 rtÞ dt;
m
ðA:2Þ used.
0
ð kdt
~ expð2rzÞ ¼ expð2az m 2 rzÞ;
LðzÞ ¼ FðzÞ ðA:3Þ
Ik ¼ f ðxÞ dx; ðA:7Þ
0
Gamma function in Equation (A.2) can be extended in the
following way: in which k is the number of iteration and dt is the very small
ðz amount of time. For example, value of d can become d ¼ 0:01 or
0.001 or even smaller.
GðzÞ ¼ ðamt m21 þ r 2 rÞ expð2at m 2 rtÞ dt
0 ð ðkþ1Þdt ð ðkþ1Þdt
ðz I kþ1 ¼ f ðxÞ dx ¼ I k þ f ðxÞ dx
Downloaded by [ETH Zurich] at 05:33 08 May 2014

, 2 expð2at m 2 rtÞ dð2at m 2 rtÞ 0 k:dt

0
ðz ðA:4Þ ½f ðkdtÞ þ f {ðk þ 1Þdt}
dt
¼ Ik þ : ðA:8Þ
2 r expð2at m 2 rtÞ dt 2
0
ðZ To this point, the value of integration can be easily estimated by
numerical calculation. We substitute Equation (A.6) and use
¼ 1 2 LðzÞ 2 r expð2at m 2 rtÞ dt: Newton method to estimate the minimum value of Cð0; ZÞ with
0
respect to the increasing number of year Z. Apart from this
The denominator in Equation (A.1) becomes method, the Simpson rule for solving integration can also be
applied. However, a comparison with various small values of d
ðz
proves that the above method is sufficient enough.
1 2 LðzÞ 2 GðzÞ ¼ r expð2at m 2 rtÞ dt: ðA:5Þ
0