The influence of practical factors on the benefits of condition-based

maintenance over time-based maintenance

Bram de Jongea,∗, Ruud Teuntera , Tiedo Tingac,b

a Department of Operations, Faculty of Economics and Business, University of Groningen, The Netherlands
b Faculty of Military Sciences, Netherlands Defence Academy, Den Helder, The Netherlands
c Faculty of Engineering Technology, University of Twente, The Netherlands

Abstract
Recent developments in condition monitoring technology have led to an ongoing shift from time-based
maintenance (TBM) to condition-based maintenance (CBM). Although CBM allows for more effectively
planned maintenance actions, its relative performance strongly depends on the behavior of the deterioration
process, the severity of failures, the required setup time, the accuracy of the condition measurements, and
the amount of randomness in the deterioration level at which failure occurs. The contribution of this paper is
twofold. First, we review studies that compare CBM with TBM, and studies that consider the above factors
in combination with a CBM model. Second, whereas existing studies confine themselves to a few examples,
we perform a numerical investigation to derive insights on the effects of the various characteristics on the
relative benefit of CBM. The results can be used by companies to decide what factors are most important
when considering to implement CBM, and to assess whether the benefit of CBM during the operational
phase outweighs the additional costs during the life cycle of equipment. This study allows for follow-up
research to quantify and generalize the insights obtained, and to analyze interaction effects.
Keywords: Condition-based maintenance, time-based maintenance

1. Introduction

Due to ongoing automation of production processes and increasing reliance on expensive production
equipment, the importance of effectively planned and performed maintenance activities is growing, and
both the portion of employees working in maintenance and the maintenance costs are increasing (Zio and
Compare, 2013). As an illustration, over a quarter of the total workforce in the process industry, and up
to 30% in the chemical industry, deal with maintenance operations (Waeyenberg and Pintelon, 2002). In
refineries, the maintenance and operations departments are usually the largest (Dekker, 1996). Further-
more, maintenance costs typically account for 15-70% of the total value of the end product (Bevilacqua
and Braglia, 2000; Mobley, 2002), the amount of money spent on maintenance of engineering structures
and infrastructures is increasing continuously (Van Noortwijk, 2009), and medical equipment maintenance
nowadays demands large sums from hospital budgets (Cruz et al., 2014).
Many firms still apply ‘traditional’ time-based maintenance (TBM) strategies, which are easy to imple-
ment as only the time that a unit is in service has to be recorded. However, substantial remaining useful
life is wasted if the machine is still in reasonable condition when preventive maintenance is performed, and
a breakdown might occur if it happens to deteriorate faster than expected. Due to the increasing technical
possibilities to monitor, store, and analyze conditions, condition-based maintenance (CBM) strategies are
gaining popularity (Chen et al., 2015; Do et al., 2015; Jardine et al., 2006; Scarf, 2007; Sloan and Shan-
thikumar, 2000). Condition-based maintenance generally results in more effectively scheduled preventive
maintenance, and, in the ideal case, preventive maintenance that is performed just before failure.
The relative benefit of CBM, however, strongly depends on the behavior of the deterioration process
and the severity of failures. Furthermore, it is affected by various practical factors that are often present
in practice, viz., required planning time, imperfect condition monitoring, and variation in the deterioration
level at which failure occurs. CBM should only be applied if this relative benefit outweighs the efforts and
costs during the entire life cycle that are required to apply CBM (Frangopol and Liu, 2007; Nilsson and
Bertling, 2007; Schuman and Brent, 2005; Van Noortwijk and Frangopol, 2004). The requisites to switch
from time-based to condition-based maintenance include condition monitoring equipment and software to

∗ Corresponding author. E-mail address: b.de.jonge@rug.nl.

Preprint submitted to Elsevier October 17, 2017

store, analyze, and initiate maintenance actions (Ahmad and Kamaruddin, 2012; Scarf, 2007). Companies
that are interested in implementing condition-based maintenance must also consider the risks related to the
lack of experience (Zio and Compare, 2013). Furthermore, they should realize that CBM requires a dynamic
scheduling of maintenance activities, whereas they might not have the capability for such flexible planning.
The first contribution of this paper is to review studies that compare condition-based and time-based
maintenance, as well as studies that consider the above practical factors in a CBM model. Although
both CBM and TBM have received ample attention in the scientific literature, few studies compare them.
Moreover, existing comparative studies confine themselves to a few examples. Insights on how the vari-
ous characteristics influence the performance of condition-based and time-based maintenance are lacking.
Therefore, our second contribution is to derive insights on the effects of the various characteristics on the
relative benefit of CBM from a numerical investigation. We start with the effects of the behavior of the
deterioration process and the severity of failures. Thereafter, we extend our model and analyze the effects
of the practical factors on the relative performance of CBM. The obtained insights are useful in practice to
decide what factors are most important when considering to switch from TBM to CBM, and for avoiding
the risk of switching from TBM to CBM in situations where benefits do not outweigh costs.
The remainder of this paper is organized as follows. In Section 2 we review existing studies that
compare condition-based maintenance with time-based maintenance, and studies that consider planning
time, accuracy of condition measurements, and predictability of the failure deterioration level. The approach
that we use to compare the two maintenance strategies is discussed in Section 3. This section also contains
formal definitions of the condition-based maintenance and the time-based maintenance strategy that we
adopt. In Section 4 we consider the effect of the behavior of the deterioration process and of the severity of
failures on the relative performance of CBM. In Section 5 we point out how this performance is influenced
by required planning time, imperfect condition information, and predictability of the deterioration level at
which failure occurs. We end with conclusions and suggestions for future research in Section 6.

2. Literature Review

We start this section with a review of studies that compare condition-based maintenance with time-
based maintenance. Thereafter, in Section 2.2, we review studies that consider various practical factors
that influence the relative performance of condition-based maintenance.

2.1. Comparative Studies

In this section we review studies that compare time-based maintenance with condition-based mainte-
nance.
The most simple models are those that consider a small number of deterioration states. McKone and
Weiss (2002) consider the integration of condition-based maintenance with traditional periodic preventive
maintenance. The available condition information is limited to a signal of a potential failure that might
be received before the actual failure. The probability that this signal is received depends on the prediction
accuracy, and the time between the signal and the failure depends on the prediction precision. The perfor-
mance of the condition-based maintenance strategy depends on the prediction accuracy and precision. In
some situations, periodic preventive maintenance or a combination of condition-based and periodic preven-
tive maintenance is preferred. Paté-Cornell et al. (1987) use a Markov chain with four states to model the
deterioration process of a production system. Time-based maintenance and three types of condition-based
maintenance are considered. The latter are based on inspections of the product, signals of the machine, and
signals provided by the use of the product. Zhang et al. (2014) develop an adaptive discrete-state model
based on Bayesian Belief Network theory. Both Paté-Cornell et al. (1987) and Zhang et al. (2014) consider
a single illustrative example, and no attempt is made to generalize the results.
Other studies consider deterioration processes with a continuous state space. Pandey et al. (2009) use
gamma deterioration processes and linear deterioration processes with a random (but fixed) rate. Xiang
et al. (2012) also adopt linear processes, but consider a rate that depends on the environment in which the
system operates. This environment is represented by a continuous-time Markov chain with three states.
Crowder and Lawless (2007) consider gamma and Wiener processes, and Zio and Compare (2013) adopt
the randomized Paris-Erdogan fatigue crack growth model. For the main part of their analysis, Pandey
et al. (2009) consider the threshold deterioration level that triggers preventive maintenance as fixed. Only a
limited investigation also includes the threshold deterioration level as a decision variable. Condition-based
maintenance turns out to be preferred over time-based maintenance only if the coefficient of variation of
the lifetime exceeds a certain level. Crowder and Lawless (2007) and Zio and Compare (2013) compare
the optimal condition-based maintenance policy with the optimal time-based maintenance policy, but they

2
only do so for a single specification of the parameters. In both studies, the performance of condition-based
maintenance turns out to be much better than that of time-based maintenance for the considered parameter
settings, but general insights are lacking. Xiang et al. (2012) include randomness in the deterioration level
at which failure occurs, and show that there is potential cost saving through implementing a condition-
based maintenance policy as opposed to time-based maintenance. No insights are presented on the effect of
changes in the other model parameters on these cost savings.
More sophisticated models are considered by Huynh et al. (2011) and by Bouvard et al. (2011). The
former combine failures due to deterioration with failures due to shock events. Because failures are not self-
announcing but should be identified by inspections, a cost is introduced for system inactivity. A condition-
based strategy with periodic inspections is compared with a purely time-based block replacement strategy. A
clear effect of the type of condition deterioration is lacking, as only two deterioration processes are considered
(high and low variance). The influence of the values of the cost parameters is studied in more detail and
the relative benefit of the condition-based strategy turns out to increase in the preventive replacement cost.
Bouvard et al. (2011) develop a maintenance model that dynamically optimizes the maintenance decisions
for a multi-component system at each periodic inspection time. Maintenance actions are grouped to reduce
maintenance costs. A system with three components is considered as an example and it is shown that
the use of condition information leads to lower costs compared with the case that this information is not
used. Cost savings are most significant for short times between inspections and moderate variances of the
underlying gamma processes that are used to model deterioration of the components.
Summarizing, we conclude that only a few general insights on the benefits of condition-based maintenance
compared with time-based maintenance are provided by the current literature.

2.2. Practical Factors Influencing the Benefits of CBM

We continue with a review of studies that include various practical factors that influence the relative
benefit of CBM. These factors are required planning time, imperfect condition monitoring, and uncertainty
in the deterioration level at which failure occurs.

2.2.1. Planning Time.

In many practical situations, repairmen are not continuously available (Iravani and Krishnamurthy, 2007;
Koole, 1995), and spare parts may not be on stock and have to be ordered (Elwany and Gebraeel, 2008; Li
and Ryan, 2011; Panagiotidou, 2014). If so, a certain planning time (in the literature also called lead time
or delay time) is required between initiating and performing a maintenance action. We remark that joint
optimization of maintenance and spare parts inventories has been considered by a number of authors. This
is beyond the scope of our study; we refer to Van Horenbeek et al. (2013) for a recent review.
A required planning time to perform preventive maintenance in combination with a continuously moni-
tored unit that deteriorates according to a gamma process is considered by various authors. They all assume
that maintenance is scheduled if the level of deterioration exceeds an alarm level, and that failure occurs if
the level of deterioration exceeds a fixed failure level. The alarm level can be specified and is the decision
variable of the considered models. Bérenguer et al. (2003) introduce a maintenance duration that depends
on the current deterioration state. Approximations for the asymptotic unavailability are obtained and
preventive maintenance policies that minimize these estimated asymptotic unavailabilities are determined.
Two values for the delay time are compared and the longer planning time results in a greater unavailability.
Grall et al. (2006) focus on the asymptotic behavior of the reliability function and show how to compute
the asymptotic failure rate of the system. Saassouh et al. (2007) extend the model with a possible sudden
change in the deterioration process. The latter two studies do not assess the effect of changing the required
planning time.
Bouvard et al. (2011) study a multi-component system with grouped maintenance actions to reduce
maintenance costs. The proposed model includes a certain minimal time required to prepare and organize
maintenance. However, the effect of the required planning time on the performance of the maintenance
policies is not studied. Van Oosterom et al. (2014) allow for both immediate and postponed preventive
maintenance actions upon the identification of a defect at an inspection. This results both in a better
utilization of the useful life and in a reduced maintenance cost. They find that if the cost difference between
a planned maintenance action and an immediate maintenance action is sufficiently large, maintenance actions
should always be planned in advance.

2.2.2. Imperfect Condition Information.

According to Ghasemi et al. (2010), condition information may contain noise due to errors of mea-
surement and interpretation, and due to the limited accuracy of the measurement’s instruments. Typical

3
condition monitoring techniques like vibration and oil debris monitoring, which are widely applied in indus-
try, generally result in such inaccuracies. These techniques can therefore be considered as imperfect (Wang
and Christer, 2000). Also when considering the crack growth of a mechanical component subject to fatigue
degradation, observations of the crack depth at inspections are just estimations of the true values (Zio and
Compare, 2013).
The most simple models that include imperfect condition monitoring contain two (Berrade et al., 2013a;
Panagiotidou and Tagaras, 2010) or three (Berrade et al., 2013b) deterioration states, and inspections reveal
the true system state with specific given probabilities. These basic models are extended with inspections
that might induce failures (Flage, 2014), a distinction between minor and major inspections (Wang et al.,
2014), and aperiodic inspections (Leung, 2001). Ghasemi et al. (2010) model the deterioration state using
a discrete time Markov process, Makis and Jiang (2003) and Moghaddass and Zuo (2012) use a continuous
time Markov process. They all relate the unobservable system state to the observed system state through an
observation probability matrix. Le and Tan (2013) also use a continuous time Markov process and combine
imperfect continuous monitoring with perfect inspections. MacPherson and Glazebrook (2014) consider
continuously monitored two-phase systems. Transitions from the fault free to the worn state might not be
observed, and indications of transitions might also be false.
Imperfect condition information has also been combined with deterioration processes modeled by con-
tinuous stochastic processes. Kallen and Van Noortwijk (2005) use a gamma deterioration process, Ye et al.
(2013) a Wiener process with positive drift, Peng and Tseng (2009) a linear trend with random coefficient
plus a Brownian motion as a second random effect, and Zio and Compare (2013) a Randomized Paris-
Erdogan fatigue crack growth model. In all cases, inspections have to be performed to obtain condition
information, and the measurement error is modeled by independent normal distributions in all cases.
Most of the aforementioned studies lack a clear translation from the level of uncertainty in the observed
condition information to the performance of condition-based maintenance. Xiang et al. (2012) consider
a model in which the rate of deterioration depends on the environment, and the measurement error is
modeled by a normal random variable. However, the assumption is made that this error is constant over
time, which may often be unrealistic. Their analysis indicates that even small levels of measurement errors
can render condition-based maintenance no better or even worse than time-based maintenance. The study
of Zio and Compare (2013) indicates that for large measurement errors the performance of condition-based
maintenance gets worse if the number of inspections increases. However, they model all measurement
errors by independent normal distributions and perform preventive maintenance when a single observed
deterioration level exceeds a safety threshold. However, if the number of inspections is large, it becomes
likely that one of the observed deterioration levels exceeds the safety threshold while the real deterioration
level is still low. This can be avoided by using a decision rule for initiating preventive maintenance that does
not only take the most recent condition measurement into account. An alternative is to model the noise
by a Brownian motion, as for example done by Elwany et al. (2011) and Li and Ryan (2011). According
to Elwany and Gebraeel (2008), this is suitable for applications where correlation exists between successive
error fluctuations in sensor readings. Furthermore, such a continuous-time process may also be appropriate
for modeling measurement errors when conditions are monitored continuously.

2.2.3. Uncertain Failure Level.

According to Jiang (2013), uncertainty in the failure level is especially relevant if the deterioration
process is represented by several condition variables that are combined into a composite condition variable.
Abdel-Hameed (2004), Van Horenbeek and Pintelon (2013), Wang et al. (2011), and Xiang et al. (2012)
model the deterioration level at which failure occurs as a random variable. Some of them assume a specific
distribution (the normal, exponential, Weibull, gamma, and lognormal distribution are used). Jiang (2013)
makes this random variable time-dependent.
Other models where failure does not occur at a fixed level of deterioration are those in which the
failure rate depends on condition variables. Kong and Park (1997) and Park (1988) consider a failure rate
that depends on a single condition variable. Examples of models that allow the failure rate to depend on
multiple condition variables are the proportional hazards model (PHM) and the accelerating failure time
(AFT) model. The PHM is used most often; see Kumar and Klefsjö (1994) for a review. Authors that
use this model include Ghasemi et al. (2010), Jiang and Jardine (2008), Louit et al. (2011), Tian and Liao
(2011), and Zuashkiani et al. (2009). The AFT model is used by Newby (1988).

4
3. Approach

We consider a single-unit system, i.e., a unit that can only be maintained in its entirety. The unit is
monitored continuously and its deterioration is modeled using a gamma process. If the unit fails, a corrective
maintenance action has to be performed. Preventive maintenance can be performed before failure of the
unit. Both maintenance types make the unit as-good-as-new, maintaining and replacing the unit are thus
interchangeable notions.
Corrective maintenance is assumed to be more expensive than preventive maintenance because failures
occur unexpectedly and are likely to have severe consequences. This is a common assumption in studies on
preventive maintenance planning (e.g. De Jonge et al., 2015a,b; Sheu and Zhang, 2013; Shore, 1996, 1998;
Zitrou et al., 2013). The cost of corrective maintenance is normalized to 1, and the cost of performing
preventive maintenance (based on condition or time) is denoted by c < 1. We use the cost rate, i.e. the
mean cost per unit of time, as the optimality criterion. Standard renewal theory can be used because of the
as-good-as-new property of maintenance, see for example (Rackwitz, 2001).
Section 3.1 discusses our selection of the gamma deterioration process and summarizes its most impor-
tant properties. The condition-based and time-based maintenance strategy that we compare are formally
described in Section 3.2. Because, different from most of the existing literature, the model that we consider
takes all practical factors discussed in Section 2.2 into account, it is rather complex. Therefore, simulation
is used to assess the performance of the maintenance strategies. Section 3.3 motivates the choice for this
research method and discusses the setup of the simulations.

3.1. Deterioration Process

Various stochastic processes can be used to model stochastic deterioration. Examples are discrete-
time or continuous-time Markov processes, Brownian motions with drift, compound Poisson processes, and
gamma processes. In this paper we will use the latter. The gamma process was introduced in the area
of reliability by Abdel-Hameed (1975). It is a rather flexible process that is applicable to model a wide
variety of deterioration processes. Therefore, by considering various gamma processes, we obtain a broad
impression of the relative performance of CBM. According to Van Noortwijk (2009), the gamma process is
most appropriate to model monotonic and gradual deterioration. He also lists some studies in which the
gamma process fits well to deterioration data. Cinlar et al. (1977) give a comprehensive justification of the
gamma process from a physical and practical point of view.
We model the level of deterioration of the unit using a stationary (also called homogeneous) gamma
process. We use the following definition for the density funtion f of the gamma distribution with shape
parameter α > 0 and scale parameter β > 0:
1 t
fα,β (t) = α
tα−1 e− β , t > 0,
Γ(α)β
R∞
in which Γ(α) = 0 z α−1 e−z dz denotes the gamma function. The stationary gamma process has a shape
function at with shape parameter a > 0 and a scale parameter b > 0. It is a continuous-time process
{X(t) : t ≥ 0} with X(τ ) − X(t) ∼ fa(τ −t),b for τ > t ≥ 0. The expectation and the variance of the
deterioration level X(t) √ at time t are respectively given by E(X(t)) = abt and Var(X(t)) = ab2 t. For
convenience, we let σ = ab denote the standard deviation of the level of deterioration at time 1. Figure
1 shows some illustrative sample paths of the stationary gamma process. The gamma process is a jump
process, and the occurrence of larger jumps becomes more likely if σ increases. Gamma processes with
larger variations are suitable to model deterioration processes with possible sudden large increases in the
deterioration level on which a CBM strategy cannot respond quickly enough.
We assume that the unit fails if the level of deterioration exceeds a certain threshold level L. The
distribution function FL of the time until failure equals
Z ∞
Γ(at, Lb−1 )
FL (t; a, b) = P (X(t) > L) = fat,b (x) dx = , t > 0,
L Γ(at)
in which
Z ∞
Γ(α, x) = z α−1 e−z dz, x ≥ 0, α > 0,
x

denotes the upper incomplete gamma function.

For the ease of representation, we change the scales that we use to express deterioration levels and time.
Firstly, we rescale the values of the deterioration levels such that failure occurs if deterioration level L = 1

5
is exceeded. Secondly, in order to obtain a cost rate of 1 under a purely corrective maintenance strategy, we
rescale the time such that the mean time to failure equals 1. We note that the latter does not imply that
the mean level of deterioration at time 1 equals 1, and thus that ab = 1. Realizing that the probability of
failure between time t1 and time t2 equals FL (t2 ; a, b) − FL (t1 ; a, b), and using a sufficiently small number
∆, the mean time to failure can be accurately approximated by
∞
X
M T T F (a, b) = i∆[FL ((i + 1)∆; a, b) − FL (i∆; a, b)].
i=0

The mean time to failure equals 1 for combinations of values for a and b that satisfy M T T F (a, b) = 1.
These combinations are determined numerically.

X(t) 1

Medium σ

Low σ

High σ

0
0 1
t

Figure 1: Sample paths of a stationary gamma process with low (σ = 0.05), medium (σ = 0.5), and high (σ = 5) standard
deviation, respectively.

3.2. Maintenance Strategies

Under the condition-based maintenance (CBM) strategy, we perform preventive maintenance if the level
of deterioration exceeds a certain threshold level M , which is the decision variable of this strategy. This
commonly used strategy is often called the control-limit strategy, see for example (Jiang, 2010). Obviously,
M should not be chosen too low, as this results in performing preventive maintenance too often. On the
other hand, if M is chosen too high, and if, for example, the deterioration process is a jump process or if
a planning time is required before preventive maintenance can be performed, the failure level L might be
exceeded before preventive maintenance is performed.
The time-based maintenance (TBM) strategy that we adopt is the age-based maintenance strategy, see
(Barlow and Hunter, 1960). Under this strategy, preventive maintenance is performed if the unit reaches
a certain age T , which is the decision variable of this strategy. A too low value for T leads to preventive
maintenance that is performed too frequently, while a too high value results in too many failures. The level
of deterioration is obviously not observed if this strategy is applied, only failures are observed.

3.3. Simulation
The evaluation of maintenance strategies for units with a deterioration level that evolves according to a
gamma process involves some integrals that are very complicated and burdensome to compute numerically
(Mercier and Castro, 2013). The main reason for this is the ‘overshoot behavior’ (Nicolai, 2008) of the
gamma process caused by the fact that the gamma process is a jump process. This behavior complicates
the exact determination of the distribution of the level of deterioration at the moment that it exceeds a
certain threshold value. Nicolai et al. (2009) mention that in most studies the overshoot of the gamma
process is not mentioned at all. To cope with these complexities of evaluating maintenance strategies
with an underlying gamma process, approximations have been studied (e.g. Bérenguer et al., 2003; Huynh
et al., 2011) and simulations have been used (e.g. Bouvard et al., 2011; Grall et al., 2006; Kallen and
Van Noortwijk, 2005; Mercier and Castro, 2013). Furthermore, the inclusion of various practical factors in
Section 5 makes a numerical analysis even more complex. In the current study we use simulation. Although
this obviously leads to approximations of the cost rate, by using a sufficiently large number of iterations
these approximations become very accurate.
We have simulated gamma processes using gamma increment sampling, see (Van Noortwijk, 2009) for
details. This method starts with choosing a small time interval and iteratively simulates the level of

6
deterioration at the end of subsequent time intervals by taking samples of the additional deterioration
during a time interval. In our simulations we use time steps with length 0.01. Furthermore, for specific
values of the parameters and the relevant decision variable (M or T ), we simulate 100,000 paths of the
gamma process until failure occurs. This number of iterations is sufficiently large to obtain smooth curves
for the cost rates as functions of the decision variables, implying that the selected optimal values of the
decision variables are not caused by coincidence. We note that, instead of using simulation, the performance
of the time-based maintenance strategy could also have been determined based on the lifetime distribution
function FL as stated in Section 3.1.

4. A Comparison of CBM and TBM

We start our comparison of the performances of condition-based maintenance and time-based mainte-
nance with an initial setting of the parameters, the so-called base case. Thereafter, we will assess the effect
of changing the (values of the parameters that drive the) behavior of the deterioration process and the cost
structure.
In our base case, we select parameter values rather arbitrarily. Other values reveal similar patterns, and
those are of interest to us rather than specific outcomes. We set the shape parameter a of the gamma process
equal to 5. In order to obtain a mean time until failure of 1, the value of the scale parameter b follows to
be approximately 0.2246 (see Section 3.1). Furthermore, we assume that the relative cost of performing
preventive maintenance equals c = 0.2. Figure 2 shows the cost rate for both strategies as functions of the
respective decision variables. The upper dotted lines show the cost rate if only corrective maintenance is
performed. This cost equals 1 because both the time until failure and the failure cost are normalized to 1.
The lower dotted lines show the cost rate in the ideal case in which we would be able to perform preventive
maintenance just before each failure. This cost obviously equals c = 0.2.
A comparison of Figures 2 (a) and (b) indicates that the cost rate under the optimal condition-based
maintenance strategy is lower than the cost rate under the optimal time-based maintenance strategy, and
thus that the use of condition information allows for cost savings. However, there is still a gap between the
cost in the ideal case and the cost under the optimal condition-based maintenance strategy, implying that
we are not able to perform condition-based maintenance just before failure. This is caused by the fact that
the deterioration process makes jumps; if we choose the maintenance threshold M very high, it is very likely
that failure occurs at the same moment at which the threshold is exceeded. Thus, the potential cost saving
of applying condition-based maintenance decreases if sudden large increases in the deterioration level occur.
A similar effect occurs if the deterioration process would be continuous, but if condition information is only
observed by inspections. In that case, the observed levels of deterioration are also non-continuous. In this
paper we consider continuously monitored equipment; studies on inspection-based CBM include Chen et al.
(2015), Golmakani and Fattahipour (2011), and Jiang (2010).

1 1
Cost
Cost

TBM

CBM

0 0
0 1 0 1
T M
(a) (b)

Figure 2: The cost rate for a = 5, b ≈ 0.2246, and c = 0.2 as a function of the maintenance age T for the TBM strategy (a)
and of the maintenance threshold M for the CBM strategy (b).

We continue to explore the effect of the behavior of the deterioration process on the performance of
CBM and TBM. As outlined in Section 3.1, the variation
√ in the deterioration process depends on the values
of a and b. We use the standard deviation σ = ab of the level of deterioration at time 1 as a measure

7
of this variation. Figure 3 (a) shows the cost rate under the optimal TBM and under the optimal CBM
strategy, for various values of σ. Figure 3 (b) shows the difference between these costs.
In the extreme case with σ = 0, the level of deterioration accrues linearly without any variation and the
time until failure is deterministic. Under both maintenance strategies this enables us to perform preventive
maintenance just before failure, resulting in a cost rate of c = 0.2. Although the costs are low in this case,
there is no cost benefit of condition-based maintenance compared with time-based maintenance. In the
other extreme case with σ very high, failure is caused by a single shock occurring at a random moment in
time, resulting in an exponentially distributed lifetime. Both time-based maintenance and condition-based
maintenance cannot prevent such a failure, implying that only corrective maintenance will be performed
and that the cost rate is 1. Again, the use of condition information does not allow for any cost savings. In
between these extremes, CBM does offer a benefit and we next explore when this benefit is largest.
Figure 3 (a) indicates that a small level of variation in the deterioration process has a relatively large
impact on the performance of TBM. This results from the fact that the increased variance in the time until
failure makes it impossible to determine a TBM strategy that always performs maintenance just before
failure. The impact on CBM is smaller; for minor variance in the deterioration process maintenance can
still be scheduled relatively effectively if condition information is available. If the variance increases further,
failure caused by sudden large increases in the deterioration process on which a CBM strategy cannot
respond quickly enough become more likely. This results in a declining performance of this strategy, both
absolute and relative compared to TBM. The benefit of CBM over TBM is thus largest for small but positive
levels of variation in the deterioration process, as depicted in Figure 3 (b).

1 0.25
Cost

Cost difference

TBM
CBM

0 0
0 5 0 5
σ σ
(a) (b)

Figure 3: The cost rate under the optimal CBM and TBM strategy (a) and the difference between these cost rates (b) for a
varying standard deviation σ of the gamma deterioration process.

We now return to the initial gamma deterioration process and consider the effect of the cost structure
on the performance of CBM and TBM. To that end, we vary the relative cost c of performing preventive
maintenance. Figure 4 (a) shows the cost rate under the optimal TBM and under the optimal CBM strategy,
for various values of c. Because the cost rate in the ideal case with preventive maintenance just before failure
depends on the cost c of preventive maintenance, the corresponding dotted line is now a diagonal line. Figure
4 (b) shows the difference between the cost under the optimal TBM and the optimal CBM strategy.
Again, two extreme cases can be distinguished. If the cost of preventive maintenance is extremely small,
preventive maintenance can be performed very frequently. Both under CBM and under TBM no failures
occur, resulting in a cost rate close to 0. If, on the other hand, the cost of preventive maintenance is almost
equal to the cost of a failure, both maintenance strategies do not allow for any cost savings. Thus, in both
extreme cases, applying condition-based maintenance does not result in any cost savings compared with
time-based maintenance.
Figure 4 (a) shows that the cost of TBM increases more steeply in the preventive maintenance cost for
moderate values of that cost, due to the increasing costs of maintaining the system when it is not close to
failure and of preventive maintenance that is scheduled too late. Figure 4 (b) shows that this results in a
substantial relative cost benefit of CBM already for low preventive maintenance costs c. This cost benefit is
retained for all realistic preventive maintenance costs and only vanishes if the preventive maintenance cost
is very high.

8
 1 0.25

Cost

Cost difference
TBM
CBM

0 0
0 1 0 1
c c
(a) (b)

Figure 4: The cost rate under the optimal CBM and TBM strategy (a) and the difference between these cost rates (b) for a
varying relative preventive maintenance cost c.

5. Practical Factors Affecting the Benefits of CBM

So far, we assumed that maintenance can be performed immediately when a certain level of deterioration
is reached, that the exact level of deterioration can be observed without any errors, and that failure always
occurs at the exact same level of deterioration. In this section, we relax these assumptions and assess the
influence on the relative performance of CBM.

5.1. Planning Time

Let us assume that performing preventive maintenance requires a fixed planning time, denoted by s. The
addition of this planning time slightly changes the CBM strategy. Preventive maintenance is scheduled at the
first time that the level of deterioration exceeds level M . Denoting this time by t(M ) = min{t : X(t) > M },
preventive maintenance is performed at time t(M )+s. Because the unit still deteriorates during the planning
time, a breakdown might occur between planning and performing preventive maintenance. In this case, we
assume that the cost c of the preventive maintenance action that is already scheduled does not have to be
paid, but the higher corrective maintenance cost of 1 is incurred instead. Note that as long as the planning
time s is smaller than the optimal maintenance age, the planning time does not affect the TBM strategy as
preventive maintenance will just be planned s time units before the optimal maintenance age is reached.
Figure 5 (a) shows the cost rate as a function of the decision variable M for the base case considered
in Section 4, and for the same case with a planning time of s = 0.1 time units. Under the presence of
a planning time, condition-based maintenance should already be initiated at a lower deterioration level.
Furthermore, the cost rate under the optimal CBM strategy increases if a planning time is required. This
is easily explained by the fact that the condition information cannot be utilized anymore once preventive
maintenance has been planned. Figure 5 (b) gives a more complete view of the effect of a planning time on
the cost rate under the optimal CBM strategy. This cost turns out to increase more or less linearly in the
planning time, resulting in a cost benefit of CBM over TBM that decreases linearly in the planning time. If
the planning time equals the maintenance age under the optimal TBM strategy, the condition information
does not allow for any cost savings anymore. Preventive maintenance is scheduled at time 0 under both
strategies in that case.

5.2. Imperfect Condition Information

We use a Brownian motion to model the difference between the real level of deterioration X(t) and the
observed level of deterioration. The Brownian motion is a continuous-time process {W (t) : t ≥ 0} with the
following properties:
1. W (0) = 0 with probability 1;
2. W (τ ) − W (t) ∼ N (0, τ − t) for τ > t ≥ 0, where N (µ, σ 2 ) denotes the normal distribution with mean
µ and variance σ 2 ; and
3. W (t) has independent increments.

9
 1

Cost
Cost
0.6 TBM

CBM

CBM (s = 0.1)
CBM (s = 0)

0 0
0 1 0 0.5
M s
(a) (b)

Figure 5: The cost rate for the CBM strategy for the base case with s = 0 and with s = 0.1 as a function of the preventive
maintenance threshold M (a) and the cost rate under the optimal TBM strategy and CBM strategy as a function of the
planning time s (b).

The observed deterioration level is X(t) + σp W (t), where σp ≥ 0 can be seen as a measure of the level
of uncertainty in the obtained condition information. Obviously, if σp = 0 then condition monitoring is
perfect.
Under imperfect condition information, condition-based maintenance is performed if the observed level
of deterioration exceeds a certain threshold level M . The threshold level that minimizes the cost rate con-
stitutes the optimal CBM strategy. Because measurement errors make the obtained condition information
less valuable, they result in a declining performance of condition-based maintenance. Measurement errors
only influence the observed information; the underlying process and the lifetime distribution do not change.
Therefore, they do not affect time-based maintenance.
Figure 6 (a) shows the cost rate under the optimal CBM and the optimal TBM strategy, Figure 6 (b)
shows the difference between these costs. The base case introduced in Section 4 is again considered, and
the level of uncertainty σp in the obtained condition information is varied. It turns out that a small level of
uncertainty in the obtained condition information only has a minor effect on the cost rate under the optimal
CBM strategy. If the uncertainty increases further, the cost rate tends to increase linearly in the level of
uncertainty. When the level of uncertainty exceeds a certain threshold, the use of condition information
does not result in any cost benefits anymore compared to time-based maintenance and may even worsen
performance.
We note that the control-limit strategy initiates maintenance actions based only on the actual observed
deterioration level, which serves as an estimate for the true deterioration level when condition measurements
are imperfect. In such cases with measurement errors, however, the estimate of the true deterioration level
can be improved by using statistical inference methods that also take historical condition measurements
into account. Such procedures improve condition-based maintenance strategies, see e.g. (Christer et al.,
1997; Makis and Jiang, 2003), and provide opportunities for future research.

5.3. Uncertain Failure Level

We model the uncertainty in the deterioration level at which failure occurs using a normal distribution
with mean 1 and standard deviation σf . The value of σf can be seen as a measure of the level of uncertainty
in the critical deterioration level. Because a negative deterioration level does not make sense we left truncate
the normal distribution at value 0, and to maintain an average deterioration level of failure of 1, we also
right truncate the normal distribution at value 2. These truncations have only a minor effect for the values
of σf that we consider.
Contrary to imperfect condition information, uncertainty in the failure threshold does affect the life-
time distribution and thereby the performance of time-based maintenance. Uncertainty in the failure level
increases the variance of the time until failure, thereby worsening the performance of the time-based main-
tenance strategy. It also makes the observed condition information less valuable, leading to a declining
performance of the condition-based maintenance strategy as well. Figure 7 (a) shows the cost rate un-
der the optimal time-based and the optimal condition-based maintenance strategy for a varying level of
uncertainty in the failure threshold. Figure 7 (b) shows the difference between these costs.

10
 Cost
TBM 0.25

Cost difference
0.6

CBM

0 0
0 0.5 0 0.5
σp σp
(a) (b)

Figure 6: The cost rate under the optimal CBM and TBM strategy (a) and the difference between these cost rates (b) for a
varying level of uncertainty σp in the obtained deterioration information.

It turns out that uncertainty in the failure threshold has a greater impact on condition-based maintenance
than on time-based maintenance, implying that the relative benefit of condition-based maintenance decreases
if this uncertainty increases. Similar to uncertainty in the obtained condition information, a small level of
uncertainty in the failure threshold only has a minor effect on the cost difference between CBM and TBM.

0.8 0.25
Cost

Cost difference

TBM

CBM

0 0
0 0.5 0 0.5
σf σf
(a) (b)

Figure 7: The cost rate under the optimal CBM and TBM strategy (a) and the difference between these cost rates (b) for a
varying level of uncertainty σf in the failure level.

6. Conclusions and Future Extensions

We have considered the benefit of condition-based maintenance compared with time-based maintenance.
We started with a literature review of studies that compare CBM with TBM, and of studies that consider
required planning time, imperfect condition monitoring, and variation in the deterioration level at which
failure occurs in a CBM model. These practical factors affect the relative benefit of CBM. It turned out
that existing studies confine themselves to a few examples and that general insights on how the various
characteristics influence the benefits of CBM are lacking. Subsequently, we considered a model for a single,
continuously monitored unit that deteriorates gradually over time. We performed simulations to first analyze
the effect of the behavior of the deterioration process and of the cost structure on the relative performance
of CBM. Thereafter, we investigated the influence of the practical factors mentioned above.
The behavior of the deterioration process turns out to be more important for the relative cost benefit
of condition-based maintenance than the cost of performing preventive maintenance. The cost difference
between CBM and TBM is substantial for a small level of variation in the deterioration process, but
diminishes quite rapidly if this variation increases. The actual preventive maintenance cost is less important
for the magnitude of the relative cost saving of CBM; the cost benefit of CBM compared with TBM is

11
substantial for a wide range of preventive maintenance costs. Only for extremely small or extremely large
preventive maintenance costs, the potential cost benefit is limited.
From the practical factors that influence the relative benefit of CBM, required planning time and im-
perfect condition information only influence the performance of CBM. Uncertainty in the failure level has a
negative effect on both CBM and TBM. However, because the effect on CBM is stronger, uncertainty in the
failure level also worsens the relative performance of CBM. The cost benefit of CBM turns out to decrease
linearly in the planning time, and is negated completely when the planning time equals the maintenance
age of the optimal TBM strategy. The effect of a planning time is thus substantial if it is large compared
to the optimal maintenance age.
Both imperfect condition information and uncertainty in the failure level only have a minor effect on the
relative cost benefit of CBM if the respective uncertainties are small. The marginal effects become stronger
and significantly impact the cost benefit of CBM if the uncertainties increase further. A notable difference
between these two effects is that the cost benefit gradually decreases but continues to be positive if the
uncertainty in the failure level increases; whereas a large level of uncertainty in the obtained deterioration
information might make CBM perform worse than TBM.
Our results show that all factors can significantly affect the benefit of condition-based maintenance over
time-based maintenance. The obtained insights are useful for companies to assess the relative importance
of the different factors in specific practical situations, and to judge whether the relative benefit of CBM
outweighs the additional costs for e.g. monitoring equipment and collecting, storing and analyzing data.
Further research could be devoted to a more detailed quantification of the insights obtained, and the
extent to which these are generalizable. We have modeled the deterioration using a stationary gamma
process, which is a rather flexible and appropriate to model various deterioration processes in practice.
Future studies could consider non-stationary gamma processes (Nicolai et al., 2007), or other stochastic
processes as Wiener processes or inverse Gaussian processes (Chen et al., 2015). The potential required
planning time is assumed to be fixed, and studying the effect of uncertainty in the planning time is also of
interest. Furthermore, we have made some modeling assumptions regarding the measurement errors and
the randomness in the failure level. The obtained effects could be validated by making alternative modeling
choices. Another reasonable stochastic process to model measurement errors is a mean-reverting Brownian
process, which has the property that the probability distribution of measurement errors is stationary over
time. The random distributions that model the uncertainty in the failure level and the random planning
time, as well as the stochastic process that models the imperfect condition monitoring, could also be based
on real life data. Finally, we have repeatedly considered the effect of a single factor on the benefit of CBM.
The manner in which these factors interact with each other also constitutes interesting future research
opportunities.

References

Abdel-Hameed, M., 1975. A gamma wear process. IEEE Transactions on Reliability R-24 (2), 152–153.
Abdel-Hameed, M., 2004. Optimal predictive maintenance policies for a deteriorating system: The total
discounted cost and the long-run average cost cases. Communications in Statistics 33 (3), 735–745.

Ahmad, R., Kamaruddin, S., 2012. An overview of time-based and condition-based maintenance in industrial
applications. Computers & Industrial Engineering 63 (1), 135–149.
Barlow, R., Hunter, L., 1960. Optimum preventive maintenance policies. Operations Research 8 (1), 90–100.
Bérenguer, C., Grall, A., Dieulle, L., Roussignol, M., 2003. Maintenance policy for a continuously monitored
deteriorating system. Probability in the Engineering and Informational Sciences 17 (2), 235–250.
Berrade, M. D., Cavalcante, C. A. V., Scarf, P. A., 2013a. Modelling imperfect inspection over a finite
horizon. Reliability Engineering and System Safety 111, 18–29.
Berrade, M. D., Scarf, P. A., Cavalcante, C. A. V., Dwight, R. A., 2013b. Imperfect inspection and re-
placement of a system with a defective state: A cost and reliability analysis. Reliability Engineering and
System Safety 120, 80–87.
Bevilacqua, M., Braglia, M., 2000. The analytic hierarchy process applied to maintenance strategy selection.
Reliability Engineering and System Safety 70 (1), 71–83.

12
Bouvard, K., Artus, S., Bérenguer, C., Cocquempot, V., 2011. Condition-based dynamic maintenance
operations planning & grouping. application to commercial heavy vehicles. Reliability Engineering and
System Safety 96 (6), 601–610.
Chen, N., Ye, Z.-S., Xiang, Y., Zhang, L., 2015. Condition-based maintenance using the inverse Gaussian
degradation model. European Journal of Operational Research 243, 190–199.

Christer, A. H., Wang, W., Sharp, J. M., 1997. A state space condition monitoring model for furnace erosion
prediction and replacement. European Journal of Operational Research 101 (1), 1–14.
Cinlar, E., Bazant, Z. P., Osman, E., 1977. Stochastic process for extrapolating concrete creep. Journal of
the Engineering Mechanics Division 103 (EM6), 1069–1088.

Crowder, M., Lawless, J., 2007. On a scheme for predictive maintenance. European Journal of Operational
Research 176 (3), 1713–1722.
Cruz, A. M., Haugan, G. L., Rincon, A. M. R., 2014. The effects of asset specificity on maintenance financial
performance: An empirical application of transaction cost theory to the medical device maintenance field.
European Journal of Operational Research 237 (3), 1037–1053.

De Jonge, B., Dijkstra, A. S., Romeijnders, W., 2015a. Cost benefits of postponing time-based maintenance
under lifetime distribution uncertainty. Reliability Engineering and System Safety 140, 15–21.
De Jonge, B., Klingenberg, W., Teunter, R. H., Tinga, T., 2015b. Optimum maintenance strategy under
uncertainty in the lifetime distribution. Reliability Engineering and System Safety 133, 59–67.

Dekker, R., 1996. Applications of maintenance optimization models: a review and analysis. Reliability
Engineering and System Safety 51 (3), 229–240.
Do, P., Voisin, A., Levrat, E., Iung, B., 2015. A proactive condition-based maintenance strategy with both
perfect and imperfect maintenance actions. Reliability Engineering and System Safety 133, 22–32.

Elwany, A. H., Gebraeel, N. Z., 2008. Sensor-driven prognostic models for equipment replacement and spare
parts inventory. IIE Transactions 40 (7), 629–639.
Elwany, A. H., Gebraeel, N. Z., Maillart, L. M., 2011. Structured replacement policies for components with
complex degradation processes and dedicated sensors. Operations Research 59 (3), 684–695.
Flage, R., 2014. A delay time model with imperfect and failure-inducing inspections. Reliability Engineering
and System Safety 124, 1–12.
Frangopol, D. M., Liu, M., 2007. Maintenance and management of civil infrastructure based on condition,
safety, optimization, and life-cycle cost. Structure and Infrastructure Engineering 3 (1), 29–41.
Ghasemi, A., Yacout, S., Ouali, M.-S., 2010. Parameter estimation methods for condition-based maintenance
with indirect observations. IEEE Transactions on Reliability 59 (2), 426–429.
Golmakani, H. R., Fattahipour, F., 2011. Optimal replacement policy and inspection interval for condition-
based maintenance. International Journal of Production Research 49 (17), 5153–5167.
Grall, A., Dieulle, L., Bérenguer, C., Roussignol, M., 2006. Asymptotic failure rate of a continuously
monitored system. Reliability Engineering and System Safety 91 (2), 126–130.

Huynh, K. T., Barros, A., Bérenguer, C., Castro, I. T., 2011. A periodic inspection and replacement policy
for systems subject to competing failure modes due to degradation and traumatic events. Reliability
Engineering and System Safety 96 (4), 497–508.
Iravani, S. M., Krishnamurthy, V., 2007. Workforce agility in repair and maintenance environments. Man-
ufacturing & Service Operations Management 9 (2), 168–184.
Jardine, A. K. S., Lin, D., Banjevic, D., 2006. A review on machinery diagnostics and prognostics imple-
menting condition-based maintenance. Mechanical Systems and Signal Processing 20, 1483–1510.
Jiang, R., 2010. Optimization of alarm threshold and sequential inspection scheme. Reliability Engineering
and System Safety 95 (3), 208–215.

13
Jiang, R., 2013. A multivariate CBM model with a random and time-dependent failure threshold. Reliability
Engineering and System Safety 119, 178–185.
Jiang, R., Jardine, A. K. S., 2008. Health state evaluation of an item: A general framework and graphical
representation. Reliability Engineering and System Safety 93 (1), 89–99.

Kallen, M. J., Van Noortwijk, J. M., 2005. Optimal maintenance decisions under imperfect inspection.
Reliability Engineering and System Safety 90 (2-3), 177–185.
Kong, M. B., Park, K. S., 1997. Optimal replacement of an item subject to cumulative damage under
periodic inspections. Microelectronics Reliability 37 (3), 467–472.

Koole, G., 1995. Optimal repairman assignment in two symmetric maintenance models. European Journal
of Operational Research 82 (2), 295–301.
Kumar, D., Klefsjö, B., 1994. Proportional hazards model: a review. Reliability Engineering and System
Safety 44, 177–188.
Le, M. D., Tan, C. M., 2013. Optimal maintenance strategy of deteriorating system under imperfect mainte-
nance and inspection using mixed inspection scheduling. Reliability Engineering and System Safety 113,
21–29.
Leung, F. K., 2001. Inspection schedules when the lifetime distribution of a singe-unit system is competely
unknown. European Journal of Operational Research 1 (132), 106–115.

Li, R., Ryan, J. K., 2011. A bayesian inventory model using real-time condition monitoring information.
Production and Operations Management 20 (5), 754–771.
Louit, D. M., Pascual, R., Banjevic, D., Jardine, A. K. S., 2011. Condition-based spares ordering for critical
components. Mechanical Systems and Signal Processing 25 (5), 1837–1848.
MacPherson, A. J., Glazebrook, K. D., 2014. How and when to deploy error prone sensors in support of the
maintenance of two-phase systems with ageing. IEEE Transactions on Reliability 63 (1), 118–133.
Makis, V., Jiang, R., 2003. Optimal replacement under partial observations. Mathematics of Operations
Research 28 (2), 382–394.
McKone, K. E., Weiss, E. N., 2002. Guidelines for implementing predictive maintenance. Production and
Operations Management 11 (2), 109–124.
Mercier, S., Castro, I. T., 2013. On the modelling of imperfect repairs for a continuously monitored gamma
wear process through age reduction. Journal of Applied Probability 50, 1057–1076.
Mobley, R. K., 2002. An introduction to predictive maintenance, 2nd Edition. Butterworth-Heinemann,
New York, NY, USA.

Moghaddass, R., Zuo, M. J., 2012. A parameter estimation method for a condition-monitored device under
multi-state deterioration. Reliability Engineering and System Safety 106, 94–103.
Newby, M., 1988. Accelerated failure time models for reliability data analysis. Reliability Engineering and
System Safety 20 (3), 187–197.

Nicolai, R. P., 2008. Maintenance models for systems subject to measurable deterioration. Ph.D. thesis,
Erasmus University of Rotterdam-Tinbergen Institute, The Netherlands.
Nicolai, R. P., Dekker, R., Van Noortwijk, J. M., 2007. A comparison of models for measurable deterioration:
An application to coatings on steel structures. Reliability Engineering and System Safety 92 (12), 1635–
1650.

Nicolai, R. P., Frenk, J. B. G., Dekker, R., 2009. Modelling and optimization imperfect maintenance of
coatings on steel structures. Structural Safety 31 (3), 234–244.
Nilsson, J., Bertling, L., 2007. Maintenance management of wind power systems using condition monitoring
systems - life cycle cost analysis for two case studies. IEEE Transactions on Energy Conversion 22 (1),
223–229.

14
Panagiotidou, S., 2014. Joint optimization of spare parts ordering and maintenance policies for multiple
identical items subject to silent failures. European Journal of Operational Research 235 (1), 300–314.
Panagiotidou, S., Tagaras, G., 2010. Statistical process control and condition-based maintenance: A mean-
ingful relationship through data sharing. Production and Operations Management 19 (2), 156–171.

Pandey, M. D., Yuan, X.-X., Van Noortwijk, J. M., 2009. The influence of temporal uncertainty of deterio-
ration on life-cycle management of structures. Structure and Infrastructure Engineering 5 (2), 145–156.
Park, K. S., 1988. Optimal wear-limit replacement with wear-dependent failures. IEEE Transactions on
Reliability 37 (3), 293–294.

Paté-Cornell, M. E., Lee, H. L., Tagaras, G., 1987. Warning of malfunction: the decision to inspect and
maintain production processes on schedule or on demand. Management Science 33 (10), 1277–1290.
Peng, C.-Y., Tseng, S.-T., 2009. Mis-specification analysis in linear degradation models. IEEE Transactions
on Reliability 58 (3), 444–455.
Rackwitz, R., 2001. Optimizing systematically renewed structures. Reliability Engineering and System
Safety 73 (3), 269–279.
Saassouh, B., Dieulle, L., Grall, A., 2007. Online maintenance policy for deteriorating system with random
change of mode. Reliability Engineering and System Safety 92 (12), 1677–1685.
Scarf, P. A., 2007. A framework for condition monitoring and condition based maintenance. Quality Tech-
nology & Quantitative Management 4 (2), 301–312.
Schuman, C. A., Brent, A. C., 2005. Asset life cycle management: towards improving physical asset per-
formance in the process industry. International Journal of Operations & Production Management 25 (6),
566–579.
Sheu, S.-H., Zhang, Z. G., 2013. An optimal age replacement policy for multi-state systems. IEEE Trans-
actions on Reliability 62 (3), 722–734.
Shore, H., 1996. Optimum schedule for preventive maintenance: A general solution for a partially specified
time-to-failure distribution. Production and Operations Management 5 (2), 148–162.
Shore, H., 1998. Solution procedures with limited sample data for the optimal replacement problem. Pro-
duction and Operations Management 7 (4), 417–422.
Sloan, T. W., Shanthikumar, J. G., 2000. Combined production and maintenance scheduling for a multiple-
product, single-machine production system. Production and Operations Management 9 (4), 379–399.
Tian, Z., Liao, H., 2011. Condition based maintenance optimization for multi-component systems using
proportional hazards model. Reliability Engineering and System Safety 96 (5), 581–589.

Van Horenbeek, A., Buré, J., Cattrysse, D., Pintelon, L., Vansteenwegen, P., 2013. Joint maintenance
and inventory optimization systems: A review. International Journal of Production Economics 143 (2),
499–508.
Van Horenbeek, A., Pintelon, L., 2013. A dynamic predictive maintenance policy for complex multi-
component systems. Reliability Engineering and System Safety 120, 39–50.
Van Noortwijk, J. M., 2009. A survey of the application of gamma processes in maintenance. Reliability
Engineering and System Safety 94 (1), 2–21.
Van Noortwijk, J. M., Frangopol, D. M., 2004. Two probabilistic life-cycle maintenance models for deterio-
rating civil infrastructures. Probabilistic Engineering Mechanics 19 (4), 345–359.

Van Oosterom, C. D., Elwany, A. H., Çelebi, D., Van Houtum, G.-J., 2014. Optimal policies for a delay
time model with postponed replacement. European Journal of Operational Research 232 (1), 186–197.
Waeyenberg, G., Pintelon, L., 2002. A framework for maintenance concept development. International
Journal of Production Economics 77 (3), 299–313.

15
Wang, W., Carr, M., Kobbacy, K. A. H., Xu, W., 2011. A model for residual life prediction based on
brownian motion with an adaptive drift. Microelectronics Reliability 51 (2), 285–293.
Wang, W., Christer, A. H., 2000. Towards a general condition based maintenance model for a stochastic
dynamic system. Journal of the Operational Research Society 51 (2), 145–155.

Wang, W., Zhao, F., Peng, R., 2014. A preventive maintenance model with a two-level inspection policy
based on a three-stage failure process. Reliability Engineering and System Safety 121, 207–220.
Xiang, J., Cassady, C. R., Pohl, E. A., 2012. Optimal maintenance policies for systems subject to a marko-
vian operating environment. Computers & Industrial Engineering 62 (1), 190–197.

Ye, Z.-S., Wang, Y., Tsui, K.-L., Pecht, M., 2013. Degradation data analysis using wiener processes with
measurement errors. IEEE Transactions on Reliability 62 (4), 772–780.
Zhang, X., Kang, J., Jin, T., 2014. Degradation modeling and maintenance decisions based on bayesian
belief networks. IEEE Transactions on Reliability 63 (2), 620–633.
Zio, E., Compare, M., 2013. Evaluating maintenance policies by quantitive modeling and analysis. Reliability
Engineering and System Safety 109, 53–65.
Zitrou, A., Bedford, T., Daneshkhah, A., 2013. Robustness of maintenance decisions: Uncertainty modelling
and value of information. Reliability Engineering and System Safety 120, 60–71.
Zuashkiani, A., Banjevic, D., Jardine, A. K. S., 2009. Estimating parameters of proportional hazards model
based on expert knowledge and statistical data. Journal of the Operational Research Society 60 (12),
1621–1636.

16