Structure and Infrastructure Engineering

Maintenance, Management, Life-Cycle Design and Performance

ISSN: 1573-2479 (Print) 1744-8980 (Online) Journal homepage: http://www.tandfonline.com/loi/nsie20

A comparative study of time-based maintenance

and condition-based maintenance for optimal
choice of maintenance policy

Jeongyun Kim, Yongjun Ahn & Hwasoo Yeo

To cite this article: Jeongyun Kim, Yongjun Ahn & Hwasoo Yeo (2016): A comparative study of
time-based maintenance and condition-based maintenance for optimal choice of maintenance
policy, Structure and Infrastructure Engineering, DOI: 10.1080/15732479.2016.1149871

To link to this article: http://dx.doi.org/10.1080/15732479.2016.1149871

Published online: 29 Feb 2016.

Submit your article to this journal

Article views: 53

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at

http://www.tandfonline.com/action/journalInformation?journalCode=nsie20

Download by: [Universite de Lorraine] Date: 18 May 2016, At: 06:36

 Structure and Infrastructure Engineering, 2016
http://dx.doi.org/10.1080/15732479.2016.1149871

A comparative study of time-based maintenance and condition-based maintenance

for optimal choice of maintenance policy
Jeongyun Kim , Yongjun Ahn and Hwasoo Yeo
Civil and Environmental Engineering, Korea Advanced Institute of Science and Technology, Daejeon, Republic of Korea

ABSTRACT ARTICLE HISTORY

Cost-effective maintenance of infrastructure systems within an acceptable level of safety and performance Received 16 March 2015
is the major concern of managing agencies. Recent maintenance approaches have offered two distinct Revised 8 December 2015
maintenance policies: time-based maintenance (TBM) and condition-based maintenance (CBM). This Accepted 24 December 2015
paper compares the two policies under different cost environments for stochastically deteriorating
infrastructures. The performance of TBM and CBM is evaluated from the viewpoint of condition transition
KEYWORDS
and life cycle cost. We found the optimal maintenance solutions for TBM and CBM using dynamic Inspections; time-based
Downloaded by [Universite de Lorraine] at 06:36 18 May 2016

programming and performed a simulation study. The simulation study showed that TBM causes some maintenance; condition-
unexpected deterioration that leads to high cost, while CBM maintains a certain level of condition steadily based maintenance;
under consistent inspection, which enables steady spending at the management level. The life cycle cost optimisation; decision-
under CBM is relatively symmetric and has a more concentrated distribution than TBM, which has a large making; stochastic models;
number of outliers from unexpected deteriorations. Finally, we evaluated the life cycle cost with a change life cycle cost
in the inspection–repair cost ratio to find the most appropriate cost environment for each maintenance
policy. While CBM needs periodic inspections, it still has more advantages than TBM when the inspection
cost is relatively low.

1. Introduction a lot to fix not only the failed part but also the other deteriorating
parts that may be near failure.
An infrastructure or a system of infrastructures is less likely
As the expenditure on maintenance has increased due to the
to fail when it is fairly new after construction, and accordingly ageing impact, researchers become interested in not only mini-
maintenance is not a big concern to the owner of the infrastruc- mum maintenance cost but also in minimum total life cycle cost
tures or to the users. However, with the increasing use of the of the target system. Consequently, a new maintenance strategy,
infrastructures, they begin to deteriorate implying that the per- preventive and predictive maintenance (PPM) has been intro-
formance or state begins to be influenced by the ageing effect. duced to minimise the total life cycle cost of the infrastructure
As the number the infrastructures to manage increases, and system. Compared to the corrective maintenance strategy, PPM
finally in later times the agency may experience budget deficits is an advanced type because it reduces the deterioration rate,
to maintain the systems of the infrastructures as the US cases avoids unnecessary repairs, solves potential problems and also
in recent years. Therefore, under the situation of infrastructure prevents unexpected system failures (Ben-Daya & Alghamdi,
deficit that many agencies are facing today, efficient and effective 2000; Carnahan, 1988). The preventive strategy adopts proactive
maintenance strategies and reducing cost to within the allowable repair actions before the state of the infrastructure drops to an
level of safety have become the main concern of infrastructure unacceptable state. In this way, it can prevent critical failures
management agencies that are in charge of keeping the infra- and reduce the rate of deterioration. The predictive maintenance
structures within the serviceable level. strategy complements the preventive maintenance by predict-
Before the 1950s, reactive maintenance, which is also known ing the future state of an infrastructure system and implement-
as corrective maintenance, was widely accepted as a maintenance ing optimal maintenance for the system. Both maintenance
strategy. Reactive maintenance carries out maintenance work strategies share the same goal: to minimise the total life cycle
when it finds that the system performance is below the intended cost of the target system. Various approaches have been intro-
level. Since this maintenance strategy does not consider the tar- duced following the PPM strategies to predict the future state
get system before failure, the agency is not aware of any minor of a system and apply proactive repair actions in accordance
abnormalities before a critical failure occurs. Thus, when a crit- with the state of the system. Recent approaches embrace two
ical failure of the system occurs, the agency usually has to spend distinct maintenance policies: time-based maintenance (TBM)

CONTACT Hwasoo Yeo [email protected]

© 2016 Taylor & Francis
 2 J. Kim et al.

and condition-based maintenance (CBM). In this paper, these Petcherdchoo (2006), Liu and Frangopol (2006) and Frangopol
two policies are reviewed and compared in order to make best and Liu (2007) considered uncertainties existing in predicting
possible selection under diverse environments. condition and safety of the system and solved the probabilistic
This paper consists of five sections. Section 2 provides a multi-objective optimisation of a bridge.
literature review followed by Section 3, which describes the More recently, Yeo, Yoon, and Madanat (2010, 2013) devel-
approaches for finding optimal maintenance options under the oped a method for maintenance optimisation with a two-stage
TBM and CBM policies. Section 4 provides a simulation study bottom-up approach: facility-level optimisation and system-level
for maintenance policy selection for stochastically deteriorating optimisation. To reflect the reality in the maintenance field, it
infrastructures. Finally, Section 5 summarises the findings and imposed the budget constraint as an economic issue and dealt
provides the conclusion of this work. with the heterogeneity of infrastructure systems. (Barone &
Frangopol, 2014) proposed a maintenance optimisation model,
accounting for the structural system effects in a system in which
2. Previous research
components deteriorate time dependently, covering the series,
TBM is a period-based maintenance policy, in which mainte- parallel and series–parallel configurations of a three-component
nance actions are carried out periodically with predetermined system.
schedules (Peng, Dong, & Zuo, 2010). TBM is known to slow As various approaches to planning maintenance scheduling
down the degradation of the system by applying preventive main- have been proposed, research for comparing their advantages
tenance activities while the system is still in operation. When and drawbacks have been also widely conducted (Barone &
the system follows its hazard rate as it ages, TBM of a repairable Frangopol, 2014). In particular, many researchers have compared
Downloaded by [Universite de Lorraine] at 06:36 18 May 2016

system is of great importance. Herein, the hazard rate of a system the performance of CBM and TBM and so far, most of them
at the age t represents the failure rates of a system when it has concluded that CBM is more cost efficient for analysing the per-
survived until time t. More frequent preventive maintenance to formance on the life cycle cost basis (Garg & Deshmukh, 2006).
the extent that the maintenance cost is kept under the budget For example, Mann, Saxena, Knapp, Saxena, and Knapp (2012)
is recommended to keep the system safe. Therefore, the most and Ahmad and Kamaruddin (2012) showed that the application
important and practical research problem for this situation is to of CBM appeared more realistic than TBM based on both data
design the most cost-effective TBM. A number of TBM policies availability/accuracy and data analysis/modelling in most cases.
have been proposed in the literature. Pham and Wang (1996) and However, it is too early to conclude that CBM always performs
Wang (2002) summarised various types of periodic maintenance better than TBM in terms of life cycle cost. As CBM needs more
policies. In addition, many authors discussed the problem of cost for periodic inspection and the decision-making process,
planning the optimal schedule for the TBM by determining the if the repair cost is minor and the inspection cost (IC) is higher,
fixed length of the time interval between preventive maintenance TBM may have advantages over CBM. The ICs, including labour
activities to minimise the total life cycle cost of the system. Brown charges, are supposed to vary according to the level of economic
and Proschan (1983) suggested an imperfect model for the sys- development of a country. Therefore, to select the best policy, it
tem under stochastic deterioration, and Fontenot and Proschan is important to find the applicable conditions of CBM and TBM
(1984) developed the optimal TBM model based on Brown and according to the maintenance cost situations varying country
Proschan’s (1983) model. Canfield (1986) introduced a periodic by country.
preventive maintenance policy for slowing down the degradation This paper compares these two maintenance methods: TBM
process of the target system. Subsequently, Chiang and Yuan and CBM. Since countries have different industrial environ-
(2001) proposed the control-limit preventive maintenance policy ments, the maintenance policy should reflect those factors
with continuous and periodic inspection of the system. Chen when choosing the suitable policy of maintenance. The advan-
and Trivedi (2001) obtained the optimal maintenance interval tage of TBM is simple application and economic benefits, while
using hypo-exponential distribution, Weibull distribution and a the disadvantage is potential failures, and higher life cycle cost
general increasing failure rate distributed for maintenance inter- caused by not considering the current state of an infrastructure.
val. Tsai, Wang, and Teng (2001) evaluated unit cost life and In comparison, CBM has the advantage of providing the optimal
optimised activities combination using a genetic algorithm. Lim solution, while at the same time it has the disadvantage of high
and Park (2007) optimised maintenance cost by determining the cost and complicated procedures for optimal decision-making.
average cost-rate of system operation with the assumption that For this reason, to determine the maintenance method between
failure times follow a Weibull distribution. CBM and TBM, we investigated the conditions to compare the
Beyond minimising the maintenance cost, some researchers robustness of infrastructure maintenance policy selection and
have focused on minimising the life cycle cost of the system, execute a sensitivity analysis for both methods.
which is the sum of the cost for the construction, the manage-
ment and the destruction. Kong and Frangopol (2003) proposed
3. Optimal solution methodology for TBM and CBM
a method for evaluating the expected life cycle maintenance
cost for deteriorating structures based on a modified event tree Assume an infrastructure system that is composed of a single
analysis. In addition, Liu and Frangopol (2005) formulated and facility. A managing agency is to choose a maintenance policy
solved the maintenance planning of bridges as multi-objective between TBM and CBM based on the evaluated performance.
optimisation, concerning not only life cycle maintenance cost For this purpose, many factors such as total life cycle cost, safety
but also lifetime condition and safety level. Based on this frame- and the performance robustness from each maintenance policy
work, Neves, Frangopol, and Cruz (2006); Neves, Frangopol, and can be considered. For system evaluation, first assume that the
 Structure and Infrastructure Engineering 3

state or condition of a facility can be represented on a numeric

scale. The condition or the service performance level of the infra-
structure system is represented by discrete numbers ranging from
1 (worst condition) to 10 (best condition). As the infrastructure
system deteriorates stochastically, the condition changing over
time is assumed to represent the state of the infrastructure.
In this paper, Markov process is applied to represent the
deterioration of the system over a time period. Then, it can
be assumed that the next state of the system depended only
on the current state. The probability that the infrastructure of
which state is I changes to state j after applying maintenance
activity
( ) (a) is specified as the transition probability matrix,
Pa i, j , ∀i = 1, … , 10, ∀j = 1, … , 10. Let a finite planning
horizon of length T. Note that the Markov process is thought
to be efficiently applied to many infrastructure systems includ-
ing pavements and bridge decks with simplicity. However, some
other systems such as mechanical parts cannot be accurately
modelled by the Markov process based on the condition that Figure 1. Optimisation of TBM action intervals.
their deterioration process is governed more by an age impact
Downloaded by [Universite de Lorraine] at 06:36 18 May 2016

than the condition. Therefore, this study focuses on the infra-

structure systems that can be modelled with a Markov process. Figure 2 presents an example of TBM optimisation results.
Assume that there exist three activities: a1 (Do Nothing) a2
(Routine Maintenance), a3 (Reconstruction). The agency is to
3.1. Optimal solution method for TBM
choose activity intervals for a2 and a3. By iterating the combi-
Under TBM maintenance policy, maintenance actions are nations of the intervals of the two activities and evaluating the
applied with constant time interval which is predetermined. average life cycle cost from multiple instances, TBM finds the
Hence, optimal TBM is to find the maintenance time intervals optimal maintenance intervals, which is found as (2, 5).
that minimises the total life cycle cost. If there exist three options
for maintenance actions: do nothing, repair and reconstruction,
3.2. Optimal solution method for condition-based
our objective is to find the optimal combination of maintenance
maintenance
time intervals of maintenance activities, ta1 , ta2 , ta3. To find the
optimal intervals, the total life cycle cost with different combi- The CBM uses condition information obtained by inspection
nations of maintenance actions is first calculated. Because the or monitoring. Among the CBM strategies, the most advanced
number of the combinations is limited, an iterative search is used. methodology is ‘predictive maintenance’, in which the condi-
The repair action interval is iterated between one and five years, tion information is used for predicting the future condition of
and the reconstruction interval is iterated between two and eight an infrastructure. By using the prediction information of the
times of the interval of repair action. When iterating intervals for future condition, the highest level of efficiency of maintenance
repair and reconstruction action, life cycle costs, which change sto- actions can be attained and it can save a lot in terms of life cycle
chastically according to the transition matrix, are calculated. Thus, cost. Here, in this research, the objective of the CBM is to find
by generating multiple runs for same combination can determine the the optimal maintenance activity for each year based on the
optimal combination of intervals giving the minimum average life inspection result. It is assumed that inspections are performed
cycle cost. The procedure of this optimisation is illustrated in Figure 1. every year at the beginning of the year, and the current state of

Figure 2. TBM solution using the proposed approach.

 4 J. Kim et al.

Figure 3. CBM solution using the proposed approach.

Downloaded by [Universite de Lorraine] at 06:36 18 May 2016

Figure 4. Dynamic programming solution for life cycle cost optimisation under stochastic deterioration.

the infrastructure is identified. Figure 3 illustrates an example expected costs for the possible conditions in the next time step.
of the optimal maintenance activities set for T years using the Then, the action with minimum expected cost is chosen as a
CBM approach. solution for the condition.
Solving the CBM optimisation problem, many different meth- Similar to TBM optimisation, other costs such as user costs
ods can be used. Among them, dynamic programming is one can be included in the Bellman equations. A certain threshold
of the most widely used solution algorithms for maintenance value can be set not to allow condition progress to the unusable
optimisation of stochastically deteriorating problems. Dynamic conditions. This can be implemented using constraints in optimi-
programming for optimal activity a* and its expected cost-to-go sation formulation. Figure 4 illustrates a dynamic programming
V* can be formulated as the following equations (Yeo et al., 2013): solution method with three maintenance activities and 10 states
∑ ( ) ( ) of a given infrastructure. In a dynamic programming method, the
a∗ (i, t) = argmina∈A {Ca (a, i) + Cur (a) + Cus (i) + 𝛼 V j, t + 1 Pa i, j } CBM tries to find the optimal solution for each year according to
j∈S
(1) the expected cost-to-go from the current year t to the final year
T, which is calculated from the condition of the facility. Then,
∑ ( ) ( )
V ∗ (i, t) = mina∈A {Ca (a, i) + Cur (a) + Cus (i) + 𝛼 V j, t + 1 Pa i, j } the final objective is to find the optimal maintenance activity at
j∈S year 1 that minimises the expected life cycle cost, which is the
(2) expected cost-to-go from year 1 to T. In other words, iterating
where Ca(a, i) : repair cost for activity a in state i; Cur (a): user cost backward in time, the optimal policy can be found for the current
during the maintenance activity a; Cus (i): user cost in state i; A: year t (t = 1, 2, 3, …, T).
set of maintenance activities, A(= a)1 , a2 , a3; S: set of system con-
dition, S = {1, … , 10}; and Pa i, j : transition probability from
4. Simulation comparison of TBM and CBM
state i to j under maintenance activity a.
In addition, Ca(a, i) is the direct repair cost spent by the To evaluate the proposed optimisation algorithm and demon-
agency, and Cur (a) is the user cost from the execution of activity strate the applicability to realistic problems, a highway pavement
a in state i. It includes the delay cost from the additional traf- section with transition probability matrix and action costs were
fic congestion. Also, Cus (i) is the user cost incurred from the generated and simulated.
infrastructure conditions. For example, in pavement case, the
roadway roughness causes additional user cost such as vehicle
4.1. Test system creation
maintenance cost. As shown in Equations (1) and (2), for each
condition at a certain time, it calculates the expected cost when For comparison purpose of maintenance policies, a virtual high-
it chooses an action based on the transition probability and way pavement system with a single section was created. The state
 Structure and Infrastructure Engineering 5

Table 1. Activity costs ($1000).

Maintenance Activity a1 a2 (CBM) a3

Activity cost State 10 Acceptable 0 30 500
9 0 70 500
8 0 110 500
7 0 150 500
6 0 190 500
5 0 230 500
4 0 270 500
3 Unacceptable 0 310 500
2 0 350 500
1 0 390 500

of the section is represented by discrete values ranging from 1 to salvage value and penalty cost. Salvage is used as the final cost
10, and simulations are run with a stochastic deterioration pro- denoted as V(T + 1), which is used for the first backward iter-
cess represented by the transition probability matrix based on the ation for finding V(T) in a dynamic programming solution. It
Markov process. The planning horizon T was set to 40 years, and provides the remaining value of the infrastructure system at the
the interest rate was set to 5%. There are three maintenance activ- end of the analysis period, so it is treated as a negative cost. The
ities: do nothing, repair and reconstruction. For the transition penalty cost is user cost, Cus (i) incurred from the lower condition
probability matrix and cost for maintenance activities, Smith & of the pavement.
Walls (1998) is used. Table 1 shows the agency activity costs used There are various evaluation criteria. One of them was intro-
Downloaded by [Universite de Lorraine] at 06:36 18 May 2016

for TBM and CBM. Note that the pavement states lower than four duced by Liu and Frangopol (2005) and Barone and Frangopol
are assumed to be unacceptable by the low serviceability, which (2014), who suggested various performance indicators such as
is incorporated as a constraint in optimisation procedures. For risk, reliability, hazard and cost in their multi-objective optimisa-
the TBM cases, it is assumed that the same maintenance activity tion framework. In this study, focus is given on the optimisation
would be conducted on the pavement section and the mainte- of the life cycle cost of the test infrastructure system. To satisfy
nance cost for repair action was equally set to 200. the level of safety and serviceability, evaluation of the life cycle
The transition probability matrix provides the probabilities cost involves penalty costs depending on the condition of the
of state transitions in a pavement segment after a maintenance pavement system with a constraint on the state of the system.
activity over one-year period. The matrices shown below are the For example, if the state of the pavement is low enough to cause
transition probability matrices for do nothing (P1) and repair severe inconvenience to users, it causes vehicle maintenance cost
(P2). For reconstruction activity, P3 is defined as a column vector and travel time delay. The user costs derived from all these incon-
with P3 (i, j) = 1, and P3 (i, j ≠ 1) = 0. veniences can be included in the penalty cost as shown in the
Table 2. In order to avoid the state of potential danger of failure,
⎛ .68 .32 ⎞ high value of penalty cost for TBM is set, and the solutions with
⎜ .78 .22 ⎟ the states lower than four in CBM are not allowed.
⎜ ⎟
⎜ .75 .25 ⎟
⎜ ⎟
⎜ .67 .33 ⎟ 4.2. Condition transition under TBM and CBM Policy
⎜ .66 .34 ⎟
P1 = ⎜ ⎟ To evaluate the maintenance policies, this study applied the opti-
⎜ .68 .32 ⎟ mal solution algorithms for the test system under the TBM and
⎜ .5 .5 ⎟ CBM policy. The initial condition (state) of the test infrastructure
⎜ ⎟
⎜ .55 .45 ⎟ system is set to state 8, which is the fair condition with high
⎜ .49 .51 ⎟ serviceability. The conditions of the pavement are compared for
⎜ ⎟ 40-year planning horizon. In each simulation run, the infrastruc-
⎝ 1.00 ⎠
ture conditions change randomly for each time period according
⎛ .80 .20 ⎞ to distribution defined by the transition probability matrices.
⎜ .68 .32 ⎟
⎜ ⎟ Therefore, different condition progress results are obtained for
⎜ .78 .22 ⎟ each simulation run. Figure 5 depicts two examples showing the
⎜ .75 .25 ⎟ condition progress under the TBM policy. The TBM optimisation
⎜ ⎟
⎜ .67 .33 ⎟ found the optimal repair interval as 5 years, and reconstruction is
P2 = ⎜ ⎟ not needed for both examples. As shown in the figure, the infra-
⎜ .66 .34 ⎟ structure conditions remain between state 7 and 10 for 40 years.
⎜ .68 .32 ⎟
⎜ ⎟ In the same manner as for the TBM cases, Figure 6 presents
⎜ .5 .5 ⎟ two examples of condition transition under the CBM policy.
⎜ .55 .45 ⎟ Reflecting the characteristics of CBM, the agency is supposed
⎜ ⎟ to execute inspection every year for 40 years to implement repair
⎝ .49 .51 ⎠
in a timely manner. Consequently, the CBM maintained infra-
Annual inspections are performed at the beginning of the year structure condition as state 8 or state 9, which implies that CBM
under CBM policy, and the cost is set to 10. Table 2 shows the keeps more stable and constant serviceability than TBM does.
 6 J. Kim et al.

Table 2. Salvage value and penalty cost for the test infrastructure system.

State 10 9 8 7 6 5 4 3 2 1
Salvage value −250 −233 −219 −200 −185 −170 −150 −125 −114 −100
Penalty cost 0 0 0 50 100 150 300 500 1000 2000
Not allowed in CBM
Downloaded by [Universite de Lorraine] at 06:36 18 May 2016

Figure 5. TBM condition transition examples over planning horizon.

Figure 6. CBM condition progress examples over planning horizon.

4.3. Simulation comparison proceed to the unallowable conditions below 4, which is a serious
problem potentially leading to infrastructure failure.
The difference between the two maintenance policies in condi-
Figure 7(b) represents the expected conditions (the average
tion transition can be more clearly suggested using the results
conditions), which shows that the condition is deteriorated in
from more simulation runs. To evaluate the overall trend of the
the years with no maintenance action, and it is recovered to state
condition transition, 10,000 simulation runs generated for the
9 with the regular repair action done every 5 years.
analysis. Each simulation run executed the optimal maintenance
activity under the TBM and CBM policy for 40 years with sto-
chastic transition of the infrastructure condition according to 4.3.2. CBM simulation result
the transition matrices defined. Figure 8 illustrates the density progress distribution of conditions
and expected conditions from 10,000 CBM simulation runs. As
4.3.1. TBM simulation result shown in Figure 8(a), CBM starts from state 8 at the initial year,
Figure 7(a) shows the condition distribution progress from simu- but it keeps its condition between state 8 or 9. The average condi-
lations runs for 40 years. The dark region denotes a high-density tion value from 10,000 simulation runs shows an almost constant
area in which many cases fall among 10,000 runs. Note that the pattern, which is a significant difference from the TBM result.
column total is 10,000. Because the initial condition is set to 8, This steady result shows the effect of annual inspection and the
state 8 in the initial year is the darkest with a 10,000 popula- optimised maintenance decisions of CBM.
tion. As the simulation year progresses, it can be noticed that
the condition of the infrastructure begins to distribute to other
4.4. Cost comparison
conditions. Although the optimal interval for partial repair is
set to 5 years, it is hard to keep all the cases higher than state In spite of the benefits of the stable management of conditions,
5, because of the stochastic nature of the deterioration process. the CBM approach has a weakness in the relatively high IC com-
Even in some cases, it can be found the infrastructure conditions pared to the TBM. To compare the two policies in terms of cost,
 Structure and Infrastructure Engineering 7
Downloaded by [Universite de Lorraine] at 06:36 18 May 2016

Figure 7. TBM condition transition (a) and expected condition (b) for 10,000 samples.

Figure 8. CBM condition transition (a) and expected conditions (b) for 10,000 simulations runs.

the annual costs and life cycle cost are investigated. The annual 4.4.1. TBM costs
cost consists of the agency cost and the user cost. The agency cost Figure 9 shows two example cases under the TBM policy. The
is the sum of action cost and IC, while the user cost includes a user cost, the agency cost and the total cost are shown together.
penalty for low serviceability caused by the low condition of the As mentioned in the condition analysis, the optimal maintenance
infrastructure. And the total cost is the sum of the agency cost time interval in the TBM was found to be 5 years. The agency
and the user cost. cost and the total cost have high values once in five years. The
 8 J. Kim et al.
Downloaded by [Universite de Lorraine] at 06:36 18 May 2016

Figure 9. Two example cases for TBM cost data.

maximum value of the user cost from the results is 50, which be analysed in relation to the condition progress results provided
implies that the condition of the pavement does not fall below in the previous section. First, in the condition result, note that
state 7 for either of the cases. TBM has several cases in which the condition drops below state
8. These cases incur high user costs leading to increased life cycle
4.4.2. CBM costs cost. Second, while TBM has a constant maintenance cost of 200,
Figure 10 shows two cases of annual cost progression under CBM finds a variable cost depending on the state of the system.
CBM. CBM includes the IC every year and shows more frequent
maintenance cost than TBM. These periodical inspections and 4.4.4. Life cycle cost vs. average annual cost
maintenance activity decision with life cycle cost optimisation To show the relationship between the average annual cost and
makes lower repair cost and total life cycle cost than TBM. Note life cycle costs for TBM and CBM, 300 samples were evaluated.
that in both cases, there are no penalty costs imposed to a con- The average annual cost is defined as the average cost that the
dition of less than 8, which implies that CBM optimisation does agency and users pay annually for 40 years. Figure 12 compares
not allow the state to drop below 8. However, the first case has 7 the results under TBM and CBM: black marks are for TBM and
times the repair actions, while the second one has 17 times the grey ones are for CBM. In TBM cases, all the average annual
repair actions. agency costs are 42.5 or 47.5. Although the cost for partial repair
is 200, the average annual cost is 40 (200/5 years), the penalty for
4.4.3. Life cycle cost low conditions explains the portion which is higher than 40. The
The life cycle costs are evaluated by performing 10,000 simula- range of the life cycle cost of the TBM was [600, 3000], which is
tion runs to compare the cost efficiency of the two maintenance wider than that of the CBM. In the result of the CBM, life cycle
strategies. Figure 11 compares the life cycle cost distribution of cost also increases as the average annual cost increases. Because
the IC = 30 case for TBM and CBM. TBM has higher life cycle under the CBM the maintenance action decision is based on
cost and greater cost deviation than CBM. High life cycle cost the stochastically changing conditions of the infrastructure, the
means less cost efficiency, and a high variation in cost indicates agency cost increases whenever the condition of the infrastruc-
less reliability of the policy. Therefore, life cycle cost results can ture is bad, which increases the life cycle cost. Although life cycle
 Structure and Infrastructure Engineering 9
Downloaded by [Universite de Lorraine] at 06:36 18 May 2016

Figure 10. Two example cases of CBM cost data.

Figure 11. Life cycle cost result of IC = 30.

 10 J. Kim et al.

cost shows an increasing tendency as the average annual cost IC, while the user cost includes a penalty for low serviceability
increases, the maximum value still remains under 1500, with a caused by the low condition of the infrastructure. The action
minimum value of 600. It has a relatively narrow range compared cost for repair and reconstruction includes material costs and
to the TBM cases. labour costs, which are commonly applied to the TBM and CBM.
However, the IC is not fairly applicable to both maintenance
policies. It rather has more influence in CBM than TBM, since
4.5. Impact of IC
CBM needs periodic inspections done for the life cycle of the
In the cost analysis, the agency cost and the user cost are eval- infrastructure. Therefore, here, the impact of the IC change is
uated. The agency cost is the sum of the action cost and the further investigated.

4.5.1. Life cycle cost distribution and IC

CBM needs an annual inspection for accurate determination
of the state, while TBM skips this procedure. IC is one of the
main factors for economic feasibility of TBM and CBM. To
investigate the impact of inspection on life cycle cost, the life
cycle cost is evaluated based on the variable IC: 10, 50 and 80
by running the simulations as in the previous examples. The
results are illustrated in Figures 13–15. As the IC increases, the
life cycle cost also increases in both TBM and CBM. However,
Downloaded by [Universite de Lorraine] at 06:36 18 May 2016

note that TBM has a right skewed distribution of the life cycle
cost compared to CBM with a higher variance. When IC
changes from 10, 50 to 80, the mean value of the life cycle cost
of TBM changes from 1121.0, 1243.0 to 1349.0 while that of
CBM changes from 626.9, 1310.0 to 1818.0. It is clear that the
life cycle cost change in CBM is much greater than that in the
Figure 12. Life cycle cost and average annual cost from simulation result.

Figure 13. Life cycle cost result of IC = 10.

Figure 14. Life cycle cost result of IC = 50.

 Structure and Infrastructure Engineering 11

Figure 15. Life cycle cost result of IC = 80.

Downloaded by [Universite de Lorraine] at 06:36 18 May 2016

Figure 16. Life cycle cost and inspection–repair ratio 𝛿.

TBM result. If the IC is greater than 50, the mean life cycle cost implies higher sensitivity of life cycle cost of CBM than the one of
of CBM exceeds that of TBM. TBM on 𝛿. Using linear regression analysis, 𝛿 range for selecting
either TBM or CBM is found. The result is represented in the
4.5.2. Inspection-repair cost ratio following manner:
This subsection provides a simple standard for maintenance pol- Inspection cost
icy decision between TBM and CBM based on cost-efficiency If 0 < < .1617 CBM is selected.
Repair cost
according to the impact of IC for maintenance policies. The
inspection–repair cost ratio is defined as follows: Inspection cost
Inspection cost If .1617 < , TBM is selected.
Repair cost
𝛿= (3)
Repair cost (Action cost)
5. Conclusions
If the inspection–repair cost ratio 𝛿 is big enough, it is more
advantageous to select TBM, and a low CBM will be better. Then, This paper provides a comparative study of TBM and CBM under
how large should 𝛿 be to select TBM? And how small should 𝛿 different cost environments for stochastically deteriorating infra-
be to select CBM? structures. For evaluation purposes, the optimal maintenance
To find the range of inspection–repair cost ratio, the life cycle solutions for TBM and CBM is found using dynamic program-
cost is analysed by setting the IC as a variable. The result is pro- ming and performed a simulations study by generating 10,000
vided in Figure 16, in which the CBM life cycle cost increases simulation runs for each policy. The findings from the simulation
more rapidly than the TBM life cycle cost as 𝛿 increases, which study can be summarised as follows:
 12 J. Kim et al.

Firstly, TBM causes some unexpected deterioration which Canfield, R. V. (1986). Cost optimization of periodic preventive
increases cost, while CBM steadily maintains a certain level of con- maintenance. IEEE Transactions on Reliability, 35, 78–81. doi:10.1109/
TR.1986.4335355
dition under consistent inspection, which enables steady spending Carnahan, J. V. (1988). Analytical framework for optimizing pavement
on the management level. Even though it is easier to execute TBM maintenance. Journal of Transportation Engineering, 114, 307–322.
on site, it shows less reliability with a wide distribution of life cycle Chen, D., & Trivedi, K.S. (2001). Analysis of periodic preventive maintenance with
costs resulting from stochastic state transition leading to failures. In general system failure distribution. Dependable Computing, Proceeding of
comparison, CBM provides a more reliable state with less probability Eighth Pacific Rim International Symposium, Seoul, (103–107).
Chiang, J. H., & Yuan, J. (2001). Optimal maintenance policy for a
of sudden failures and critical errors, although it has higher system Markovian system under periodic inspecction. Reliability Engineering
requirements for inspection and analysis for optimal decision-mak- & System Safety, 71, 165–172.
ing. The simulation study shows that TBM causes unexpected dete- Fontenot, R. A., & Proschan, F. (1984). Some imperfect maintenance
rioration once in every 2000 tests, while CBM maintains a certain models. In: M. Abdel-Hameed, E. Çinlar and J. Quinn. (Eds.), Reliability
level of condition steadily under periodic inspection. The life cycle theory and models, 1–20. Orlando, FL: Academic Press.
Frangopol, D. M., & Liu, M. (2007). Bridge network maintenance
cost under CBM has a relatively symmetric and more concentrated optimization using stochastic dynamic programming. Journal of
distribution than TBM which has a large number of outliers from Structural Engineering, 133, 1772–1782.
unexpected deteriorations. Garg, A., & Deshmukh, S.G. (2006). Maintenance management: Literature
Secondly, this study introduced the inspection–repair cost ratio review and directions. Journal of Quality in Maintenance Engineering,
and evaluated the life cycle cost with the change of the inspection– 12, 205–238. doi:10.1108/13552510610685075
Kong, J. S., & Frangopol, D. M. (2003). Evaluation of expected life-cycle
repair cost ratio to find the proper cost environment for each main- maintenance cost of deteriorating structures. Journal of Structural
tenance policy. Though CBM needs periodic inspections, CBM has Engineering, 129, 682–691.
Downloaded by [Universite de Lorraine] at 06:36 18 May 2016

a bigger advantage than TBM has when the IC is relatively low. Lim, J., & Park, D. (2007). Optimal periodic preventive maintenance
The method adopted in this study can be used in selecting an schedules with improvement factors depending on number of preventive
appropriate maintenance policy in various industrial environments maintenances. Asia-Pacific Journal of Operational Research, 24, 111–124.
Liu, M., & Frangopol, D. M. (2005). Multiobjective maintenance planning
according to the cost structures and infrastructure types. With the optimization for deteriorating bridges considering condition, safety,
statistical transition probability matrices and the related costs, the and life-cycle cost. Journal of Structural Engineering, 131, 833–842.
administrating agency can select a better maintenance policy from Liu, M., & Frangopol, D. M. (2006). Optimizing bridge network
the comparisons on the life cycle cost. Furthermore, the proposed maintenance management under uncertainty with conflicting criteria:
method for evaluating TBM and CBM in this study can be applied life-cycle maintenance, failure, and user costs. Journal of Structural
Engineering, 132, 1835–1845.
and extended to other systems such as bridges and mechanical Mann, L., Saxena, A., Knapp, G. M. Jr., Saxena, A., & Knapp, G. M. (2012).
systems with stochastic deterioration, or large systems with multi- Statistical-based or condition-based preventive maintenance ? Journal
ple facilities. Since this study is limited to an infrastructure system of Quality in Maintenance Engineering, 1, 46–59.
composed of a single facility, an extended study to include multiple Neves, L. A. C., Frangopol, D. M., & Cruz, P. J. S. (2006). Probabilistic
facilities, which is more realistic, is needed. lifetime-oriented multiobjective optimization of bridge maintenance:
Single maintenance type. Journal of Structural Engineering, 132, 991–
1005.
Disclosure statement Neves, L. A. C., Frangopol, D. M., & Petcherdchoo, A. (2006). Probabilistic
lifetime-oriented multiobjective optimization of bridge maintenance:
No potential conflict of interest was reported by the authors. Combination of maintenance types. Journal of Structural Engineering,
132, 1821–1834.
Peng, Y., Dong, M., & Zuo, M.J. (2010). Current status of machine
Funding prognostics in condition-based maintenance: A review. The
International Journal of Advanced Manufacturing Technology, 50, 297–
This work was supported by the Research Program-Condition Based Smart
313. doi:10.1007/s00170-009-2482-0
Maintenance Critical Technology [program number 13RTRP-C068243-01]
Pham, H., & Wang, H. (1996). Imperfect maintenance. European Journal of
and funded by the Korea Agency for Infrastructure Technology
Operation Research, 94, 425–438.
Advancement (KAIA).
Smith, M.R., & Walls, J. (1998). Life-Cycle Cost Analysis In Pavement
Design: Interim Technical Bulletin FHWA-SA-98-079. Federal Highway
Administration: Research Report.
ORCID Tsai, Y.-T., Wang, K.-S., & Teng, H.-Y. (2001). Optimizing preventive maintenance
Jeongyun Kim http://orcid.org/0000-0001-5004-7630 for mechanical components using genetic algorithms. Reliability Engineering &
Yongjun Ahn http://orcid.org/0000-0003-4153-3420 System Safety, 74, 89–97. doi:10.1016/S0951-8320(01)00065-5
Hwasoo Yeo http://orcid.org/0000-0002-2684-0978 US Department of Transportation, Federal Highway Administration.
(1998). Life-cycle cost analysis in pavement: search of better investment
decisions Pavement Division Interim Technical Bulletin, Publication
References No. FHWA-SA-98-079.
Wang, H. (2002). A survey of maintenance policies of deteriorating
Ahmad, R., & Kamaruddin, S. (2012). An overview of time-based and systems. European Journal of Operational Research, 139, 469–489.
condition-based maintenance in industrial application. Computers and Yeo, H., Yoon, Y., & Madanat, S. (2010). Maintenance optimization for
Industrial Engineering, 63, 135–149. doi:10.1016/j.cie.2012.02.002 heterogeneous infrastructure systems: Evolutionary algorithms for
Barone, G., & Frangopol, D.M. (2014). Life-cycle maintenance of bottom-up methods. In: K. Gopalakrishnan & S. Peeta (Eds.), Sustainable
deteriorating structures by multi-objective optimization involving and resilient critical infrastructure systems: Simulation, modeling, and
reliability, risk, availability, hazard and cost. Structural Safety, 48, 40–50. intelligent engineering. (pp. 185–200). Berlin: Springer-Verlag.
Ben-Daya, M., & Alghamdi, A.S. (2000). On an imperfect preventive Yeo, H., Yoon, Y., & Madanat, S. (2013). Algorithms for bottom-up
maintenance model. International Journal of Quality & Reliability maintenance optimisation for heterogeneous infrastructure systems.
Management, 17, 661–670. doi:10.1108/02656710010317065 Structure and Infrastructure Engineering, 9, 317–328.
Brown, M., & Proschan, F. (1983). Imperfect repair. Journal of Applied
Probability, 20, 85l–859.