See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/235310886

Optimal preventive maintenance and replacement schedules with

variable improvement factor

Article  in  Journal of Quality in Maintenance Engineering · August 2010

DOI: 10.1108/13552511011072916

CITATIONS READS
23 1,002

2 authors:

Kamran S. Moghaddam John S. Usher

Clayton State University University of Louisville
23 PUBLICATIONS   352 CITATIONS    57 PUBLICATIONS   1,080 CITATIONS   

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Kamran S. Moghaddam on 28 October 2015.

The user has requested enhancement of the downloaded file.

 The current issue and full text archive of this journal is available at
www.emeraldinsight.com/1355-2511.htm

METHODOLOGY AND THEORY Optimal

preventive
Optimal preventive maintenance maintenance
and replacement schedules with
271
variable improvement factor
Kamran S. Moghaddam and John S. Usher
Department of Industrial Engineering, University of Louisville, Louisville,
Kentucky, USA

Abstract
Purpose – This paper seeks to develop and present a new mathematical formulation to determine the
optimal preventive maintenance and replacement schedule of a system.
Design/methodology/approach – The paper divides the maintenance-planning horizon into
discrete and equally-sized intervals and in each period decide on one of three possible actions: maintain
the system, replace the system, or do nothing. Each decision carries a specific cost and affects the
failure pattern of the system. The paper models the cases of minimizing total cost subject to a
constraint on system reliability, and maximizing the system reliability subject to a budgetary
constraint on total cost. The paper presents a new mathematical function to model an improvement
factor based on the ratio of maintenance and repair costs, and show how it outperforms fixed
improvement factor models by analyzing the effectiveness in terms of cost and reliability of the
system.
Findings – Optimal decisions in each period over a planning horizon are sought such that the
objectives and the requirements of the system can be achieved.
Practical implications – The developed mathematical models for this improvement factor can be
used in theoretical and practical situations.
Originality/value – The presented models are effective decision tools that find the optimal solution
of the preventive maintenance and replacement scheduling problem.
Keywords Preventive maintenance, Optimization techniques, Manufacturing systems
Paper type Research paper

1. Introduction
Preventive maintenance and replacement describes a wide range of activities designed
and performed in order to improve the overall reliability and availability of a system.
All types of systems found in manufacturing, health care, communications, and
transportation industries have prescribed maintenance plans designed by reliability
engineers to reduce the risk of system failure. Preventive maintenance and replacement
activities generally consist of inspection, cleaning, lubrication, adjustment, alignment,
and/or replacement of systems that wear-out. Regardless of the specific system,
preventive maintenance and replacement activities can be categorized in one of two
ways, maintenance or replacement. An example of system maintenance would be Journal of Quality in Maintenance
checking and, if needed, correcting the air pressure in the tires of an automobile. Note Engineering
Vol. 16 No. 3, 2010
that this activity changes the aging characteristics of the tires and, if done correctly, pp. 271-287
ultimately decreases their rate of occurrence of failure. An example of replacement q Emerald Group Publishing Limited
1355-2511
would be replacing the tires with new ones. DOI 10.1108/13552511011072916
JQME It is clear that preventive maintenance and replacement strategies involve a basic
trade-off between the conflicting desires to minimize of the total costs of maintenance
16,3 and replacement activities and maximization of the overall reliability of the system.
System that are maintained and/or replaced often will experience high reliability (low
cost of failures), but with high costs of maintenance and replacement. Designers of
preventive maintenance and replacement policies must weigh these individual costs
272 along with cost of system failure in order to minimize the overall cost of system
operation. They may also be interested in maximizing the system reliability, subject to
some sort of budget limitations.
In this research, mathematical models for planning the preventive maintenance and
replacement schedules for a repairable system are presented. In each period, (which
could be defined as an hour, a day, a week, a month, etc.) we assume that one of three
distinct actions can be performed:
(1) Do nothing. In this case, no action is to be performed. This is often addressed to
as leaving the system in a state of “bad-as-old”, where it continues to age
normally.
(2) Maintenance. In this case, the system is maintained or repaired, which places it
into a state somewhere between “good-as-new” and “bad-as-old”. In this
research, we assume that the maintenance action reduces the effective age of the
system by a stated percentage of its actual age. This is only applicable to
systems subject to wear out, that experience an increasing ROCOF. As such, the
reduction in effective age places the system at a lower point on the increasing
ROCOF function.
(3) Replacement. In this case, the system is replaced by a new one, immediately
placing it in a state of “good-as-new”, i.e. its age is effectively returned to time zero.
The preventive maintenance and replacement scheduling problem can be defined as
designing a sequence of actions (1), (2) or (3) for the system for each period in the
planning horizon such that overall costs are minimized and/or reliability is maximized.
The organization of the paper is as follows. In section 2, we briefly review the
current literature. We describe the failure rate, and characteristics of maintenance and
replacement activities used in our models in section 3. In section 4, we develop and
present optimization models that include a new mathematical model for defining the
improvement factor. We discuss how to find optimal preventive maintenance and
replacement policies. In section 5, we present a numerical example and demonstrate
computational results. Conclusions and recommendations for further study are
presented in section 6.

2. Literature review
Nakagawa (1988) presents models that utilize an improvement factor approach to
maintenance planning. His work has been referenced by many researchers. He
develops two analytical models in order to find optimal preventive maintenance
schedules based on an assumption of increasing failure rate over time. The first model,
called a preventive maintenance hazard rate model, calculates the average failure cost
of minimal repairs along with costs of preventive maintenance and replacement actions
under the assumption that preventive maintenance actions reduce the next effective
age to zero. He also assumes the failure rate is increased by increasing the frequency of
preventive maintenance actions. Furthermore, this model assumes that maintenance Optimal
activities take place at fixed intervals between each predetermined replacement. The preventive
second model, called an age reduction preventive maintenance model, considers the
average failure cost of minimal repairs as well as costs of preventive maintenance and maintenance
replacement actions by assuming that the effective age of a system is reduced by an
improvement factor after performing minimal repairs. In order to find the optimal
schedule, both models are optimized by calculus methods. He applies the models in a 273
numerical example and describes that based on obtained computational results the
second model is more practical than the first model.
Jayabalan and Chaudhuri (1992) propose another often-cited work on age reduction
and improvement factors models. They develop an optimization model and a
branching algorithm that minimizes the total cost of preventive maintenance and
replacement activities. They assume a constant improvement factor and define a
required failure rate. In addition, they assume a zero failure cost and do not consider
time value of money for future costs. Their algorithm determines an optimal schedule
of maintenance actions before each replacement action in order to minimize the total
cost in a finite planning horizon. They utilize FORTRAN programming environment to
implement the algorithm and prove the effectiveness of the algorithm via several
numerical examples.
Dedopoulos and Smeers (1998) develop a nonlinear optimization model to find the
best preventive maintenance schedule by considering the degree of age reduction as
the variable in the model. The researchers assume a constant improvement factor but a
variable amount of age reduction, which depends on the schedule of preventive
maintenance actions. They define amount of age reduction, time and duration of
preventive maintenance activities as the decision variables and consider fixed cost and
variable cost for maintenance actions. They present the variable cost as a function of
the amount of age reduction and duration of action and the effective age of the system.
Moreover, they present the failure rate in each period as a recursive function of age
reduction from a previous period and consider the net profit as the objective function in
the model. They implement the model in GAMS programming environment and use
GAMS/MINOS optimization software. Finally, the effectiveness of the model is shown
via three numerical examples. Martorell et al. (1999) present an age-dependent
preventive maintenance model based on the surveillance parameters, improvement
factor, and environmental and operational conditions of the equipment in a nuclear
power plant. They consider risk and cost as the main criteria of the model based on the
age of the system and perform a sensitivity analysis to show the effect of the
parameters on the preventive maintenance policies. They express that the results
obtained from their model are different from those resulted from the models that do not
consider the improvement factor parameter and working conditions.
Lin et al. (2001) combine the models that were developed by Nakagawa (1988)
and present hybrid models in which the effects of each preventive maintenance
action are considered in two ways; one for its immediate effects and the other one
for the lasting effects when the equipment is put to use again. The authors
construct two models that reflect the concept of maintainable and non-maintainable
failure modes. In the first model, they assume that preventive maintenance and
replacement time are independent decision variables and consider the mean cost
rate as the objective function that should be minimized. In the second model, they
JQME assume that preventive maintenance activities are performed whenever the failure
16,3 rate of the system exceeds the certain level and same as the first model, the mean
cost rate is considered as the objective function. Finally, they present numerical
examples to show the application of the developed models and mention that for a
system with failure rate corresponds to Weibull distribution function, optimal
schedules can be achieved analytically, but for the general case, it cannot be solved
274 by analytic methods. Cheng and Chen (2003) consider the improvement factor as a
variable of total number of preventive maintenance actions performed over the
planning horizon, and the cost ratio of preventive maintenance to replacement
actions. They assume different types of restoration effect based on the cost ratio of
maintenance and replacement actions and propose three different models. They
consider the total number of preventive maintenance actions as the decision variable
and develop an objective function to minimize the total cost of the system. By using
a numerical analysis method, they mention that the proposed improvement factor
model provides a variety of options to evaluate the restoration effect of a
deteriorating system.
Xi et al. (2005) develop a sequential preventive maintenance optimization model
over a finite planning horizon. They define a recursive hybrid failure rate based on the
improvement factor concept and increasing failure rate in order to estimate the systems
reliability in each period of planning horizon. In addition, they consider the total cost of
preventive maintenance activities and assume that the mean cost in each period is a
function of required reliability and the improvement factor parameter. Finally, they
utilize a simulation approach to optimize the model and mention that the
computational results can be used in a maintenance decision support system for job
shop scheduling problems. Jaturonnatee et al. (2006) develop an analytical model in
order to find the optimal preventive maintenance schedule of leased equipment by
minimizing the total cost function. They define maintenance actions as preventive and
corrective, each with associated costs, and then consider the concept of reduction in
failure intensity function along with penalty costs due to violation of leased contact
issues. They present a numerical example for a system with Weibull failure rate, solve
the model analytically, and examine the effect of penalty terms on the optimal
preventive maintenance policies.
Bartholomew-Biggs et al. (2006) present several preventive maintenance scheduling
models that consider the effect of imperfect maintenance on effective age of the system.
The researchers develop optimization models that minimize the total cost of preventive
maintenance and replacement activities. In this study, they assume a known failure
rate to express the expected failures as a function of age and consider age reduction in
the effective age, based on the concept of an improvement factor. They develop a new
mathematical programming formulation to achieve optimal maintenance schedule and
utilize automatic differentiation as numerical approach, instead of an analytical
approach, to compute the gradients and Hessians in the optimization procedure, which
is a global minimization of a non-smooth performance function. Finally, the
effectiveness of the proposed model and algorithm is shown in several numerical
examples.
El-Ferik and Ben-Daya (2006) present an age-based hybrid model for imperfect
preventive maintenance scheduling. The authors review different policies and propose
a new sequential age-based analytical model that assumes imperfect preventive
maintenance activities reduce the effective age of the system but increase the failure Optimal
rate. They present mathematical formulations to determine the adjustment factors for preventive
both failure rate and the age reduction coefficient. They construct an optimization
model based on their analytical models, consider the total cost as the objective function, maintenance
and solve the optimization model via a new heuristic algorithm in a numerical example.
One of the recent works on methods for estimating age reduction factor is presented by
Che-Hua (2007). In this research, he considers an optimal preventive maintenance for a 275
deteriorating one-component system via minimizing the expected cost over a finite
planning horizon. He develops a mathematical model for estimating improvement
factor to measure the restoration of system under the minimal repair. The proposed
improvement factor is a function of effective age of the system, the number of
preventive maintenance actions, and the cost ratio of maintenance action to the
replacement action. Finally, he obtains an optimal preventive maintenance schedule for
a case study with the Weibull hazard function by applying a particle swarm
optimization method.
Cheng et al. (2007) present a paper about models to estimate the degradation rate
of the age reduction factor. They present two optimization models, which minimize
the cost subject to required reliability. The first model has a periodic preventive
maintenance time interval for every replacement and the second one contains the
maintenance schedule where the time interval between the final maintenance and
replacement is not constant. Lim and Park (2007) present three analytical preventive
maintenance models that consider the expected cost rate per unit time as the
objective function. In this research, they assume that each preventive maintenance
activity reduces the starting effective age but does not change the failure rate and
consider the improvement factor as the function of number of preventive
maintenance activities. They also assume that the failure function corresponds to a
Weibull distribution function and develop a mathematical formulation for three
different situations; preventive maintenance period is known, number of preventive
maintenance is known, and number and period of preventive maintenance is
unknown. They obtain optimal preventive maintenance and replacement schedules
by taking an analytical approach and apply them to a numerical example to show
an application of their models.
By reviewing the literature, we find that most researchers assume a constant
improvement factor and develop optimization models to determine the optimal
schedule of preventive maintenance actions; see Jayabalan and Chaudhuri (1992),
Martorell et al. (1999) and Martorell et al. (1999). Some assume a constant improvement
factor with a variable amount of age reduction, which depends on when maintenance
actions are performed; see Dedopoulos and Smeers (1998). Nakagawa (1988) assumes a
variable improvement factor as a function of time intervals before any replacement and
presents equations (1) and (2) for hazard rate improvement factor and effective age
improvement factor which are also used by Lim and Park (2007):
2k þ 1
ak ¼ fork ¼ 1; :::; n ð1Þ
kþ1

k
bk ¼ fork ¼ 1; :::; n ð2Þ
kþ1
JQME Lin et al. (2001) consider equations (3) and (4) for the same purpose, which are also used
by El-Ferik and Ben-Daya (2006) and Bartholomew-Biggs et al. (2006):
16,3
6k þ 1
ak ¼ fork ¼ 1; :::; n ð3Þ
5k þ 1

276 bk ¼
k
fork ¼ 1; :::; n ð4Þ
2k þ 1
In this paper, we introduce a more practical formulation to model the improvement
factor in preventive maintenance and replacement scheduling problems. The proposed
function assumes the variable improvement factor and it is based on ratio of
maintenance and replacement costs and the effective age of the system.

3. Effects of maintenance, replacement, and do nothing actions

Notation:
. T : Length of the planning horizon.
. J : Number of intervals.
. l : Characteristic life (scale) parameter of system.
.
b : Shape parameter of system.
.
a : Improvement factor of system.
.
F : Unexpected failure cost of system.
.
M : Maintenance cost of system.
.
R : Replacement cost of system.
.
RR : Required reliability of system.
.
GB : Given budget for system.
.
X j : Effective age of system at the start of period j.
.
X 0j : Effective age of system at the end of period j.
(
1 if component at period j is maintained
mj ¼
0 otherwise

(
1 if component at period j is replaced
rj ¼
0 otherwise

Suppose we have a new repairable and maintainable system, which is subject to

deterioration. The system is assumed to have an increasing rate of occurrence of failure
(ROCOF), vðtÞ, where t denotes actual time, ðt . 0Þ. We assume that the ROCOF
corresponds to the well-known Non-Homogeneous Poisson Process (NHPP) and takes
on the following form:
vðtÞ ¼ l · b · t b21 ð5Þ
l and b are the characteristic life (scale) and the shape parameters of the system Optimal
respectively. For more on this well-known stochastic process see Ascher and Feingold preventive
(1984).
We seek to find a schedule of future maintenance and replacement actions for the maintenance
system over the planning period [0, T ], which is segmented into J discrete intervals,
each of length T/J. At the end of period j, the system is either, maintained, replaced,
or no action is performed. We assume that maintenance or replacement activities in 277
period j reduce the “effective age” and thus the failure rate of the system. For
simplicity we also assume that these activities are instantaneous, i.e. the time
required to replace or maintain is negligible, relative to the size of the interval, and
thus is zero. However, we do consider a cost associated with each repair or
maintenance action.

3.1 Maintenance
We define effective ages of a system at the start and end of each period denoted by X j ,
and X 0j respectively and presented an equation to relate them to each other by the
length of each period T/J as follows:

T
X 0j ¼ X j þ forj ¼ 1; :::; T ð6Þ
J

In addition, we assume the initial age of the system is equal to zero. We also assume
that the maintenance activity occurs at the end of the each period and effectively
reduces the age of the system at the start of the next period based on an “improvement
factor” (aka “age reduction factor”). This kind of maintenance does not change the
failure characteristics of the system but reduces its effective age, and is sometimes
referred to as “minimal repair”:

X jþ1 ¼ a · X 0j for i ¼ 1; :::; N ; j ¼ 1; :::; Tandð0 # a # 1Þ ð7Þ

Note that when a ¼ 0, the effect of maintenance is to return the system to a state of
“good-as-new” and it corresponds to replacement of the system. When a ¼ 1,
maintenance has no effect, and the system remains in a state of “bad-as-old” which
corresponds to “do nothing”. Without lose of generality, we can always assume that
0 # a # 1. The maintenance action at the end of period j results in an instantaneous
drop in the ROCOF of the system, as shown in Figure 1. Thus at the end of period j, the
system ROCOF is vðX 0j Þ. At the start of period j þ 1 we find that the ROCOF drops to
vðX 0j Þ.
Suppose a system is maintained during its service life without any replacement. We
can calculate its effective age at the start and end of each period as a function of length
of each period, number of maintenance actions, and amount of improvement factor
based on the following equations:
 8
JQME >
>
>
>
X 1 ¼ 0; X 01 ¼ X 1 þ TJ ¼ TJ
>
16,3 >
>
>
>
>
X 2 ¼ a · X 01 ¼ a £ TJ ; X 02 ¼ X 2 þ TJ ¼ TJ ða þ 1Þ
>
>


>
> X 3 ¼ a · X 02 ¼ TJ a 2 þ a ; X 03 ¼ X 3 þ TJ ¼ TJ a 2 þ a þ 1
<


 ð8Þ
>
> X 4 ¼ a · X 03 ¼ TJ a 3 þ a 2 þ a ; X 04 ¼ X 4 þ TJ ¼ TJ a 3 þ a 2 þ a þ 1
>
>
>
> ..
>
>
>
278 >
>
>
>
>


.


: X j ¼ a · X 0j ¼ TJ a j21 þ a j22 þ a j23 þ · · · þ a ; X 0j ¼ X j þ TJ ¼ TJ a j21 þ a j22 þ a j23 þ · · · þ a þ 1

!
T X
j21
r
Xj ¼ a ð9Þ
J r¼1

!
T X
j21


X 0j ¼ a þ1r
ð10Þ
J r¼1

Now if we assume an unlimited service life for a system with a large number of
maintenance actions, we can find the lower-bound for its effective age by taking a limit
of the effective ages when number of maintenance actions goes to infinity (note that
0 , a , 1):
!!
T X r T  a 
j21


lim X j ¼ lim a ¼ ð11Þ
j!1 j!1 J r¼1
J 12a

!!  
  T X
j21

r  T 1
lim X 0j ¼ lim a þ1 ¼ ð12Þ
j!1 j!1 J r¼1
J 12a

The equations (11) and (12) provide a useful perspective to figure out how maintenance
actions affect the effective ages of a system over a long-term planning horizon. For
example, suppose the length of each period is equal to one month, the planning horizon
is long enough and the system is maintained every month with an improvement factor
equal to 0.8, which means that each maintenance action reduces the effective age by 20
percent. Under these assumptions, a lower-bound for the starting and ending effective

Figure 1.
Effect of period-j
maintenance on system
ROCOF
ages in any given month would be 4 and 5 months respectively. That is, the best one Optimal
could ever do over the long run, by maintaining a system every period, would be to
have a minimum effective age of 4 months at the start of any future period. These
preventive
values can be therefore be considered as the minimum for starting and ending effective maintenance
ages of the system over the long run. In addition, if a maintenance is performed on the
system in period j, a maintenance cost constant M is incurred at the end of the period.
279
3.2 Replacement
If the system is replaced at the end of period j with a new identical one then, it is
returned to a state of “good-as-new” and the ROCOF of system instantaneously drops
from vðX 0j Þ to vð0Þ as shown in Figure 2. We also assume a replacement cost for the
system, denoted R:
X jþ1 ¼ 0 for j ¼ 1; . . .; T ð13Þ

3.3 Do nothing
If no action is performed in period j, then no effect will be on the ROCOF of the system,
and we find that:
T
X 0j ¼ X j þ for j ¼ 1; :::; T ð14Þ
J

X jþ1 ¼ X 0j for j ¼ 1; :::; T ð15Þ

vðX jþ1 Þ ¼ vðX 0j Þfor j ¼ 1; :::; T ð16Þ

When we plan the future schedule of operation for the system, we must consider the
inevitable costs caused by unexpected system failures. Based on the assumptions, at
the start of period j ¼ 1 however, we cannot know when such failures will be observed.
However, we know that if the system carries a high ROCOF through a period, then the
system is at risk of experiencing high number, and hence the cost, of failures. On the
other hand, a low ROCOF in period j should yield a low cost of failure. In order to take
into account this unplanned cost, we propose the computation of the expected number

Figure 2.
Effect of period-j
replacement on system
ROCOF
JQME of failures in each period. Here we compute the expected number of failures of system
in period j, as:
16,3
Z X0
  j
E Nj ¼ vðtÞdtfor j ¼ 1; :::; T ð17Þ
Xj

280 Under the NHPP failure rate, we can find the expected number of system failures in
period j to be:
Z X0   
  j b
b
E Nj ¼ l · b · t b21 dt ¼ l X 0j 2 X j for j ¼ 1; :::; T ð18Þ
Xj

We assume that the cost of each failure is F (in units of $/failure event). Hence
regardless of any maintenance or replacement actions (which are assumed to occur at
the end of the period) in period j, there is still a cost associated with the possible and
unexpected failures that can occur during the period.

4. Optimization models
Using the results of section 3, we now present models to minimize total cost subject to
the constraint that some minimum level of system reliability over the planning horizon
is achieved.
From our definitions of each type of cost, we can write the total cost function as follows:
XT   
b
b
 
TotalCost ¼ F ·l X 0j 2 X j þ M · mj þ R · r j ð19Þ
j¼1

This function computes total cost as a simple sum of system costs in each period based
on any maintenance or repair cost, and the cost of the expected number of failures. It is
certainly possible to compute a more accurate economic measure of present value of
these future costs by using a suitable interest rate, or even including the effects of
inflation, by adding an inflation rate into the calculation of future costs. While these
may make the model more accurate, we have avoided those minor refinements for the
sake of notational simplicity.
One may also be interested in determining the system reliability (probability of
operating without failure survival over the planning horizon). Based on the assumption
on a NHPP, we can define the system reliability in period j (the probability of surviving
to the end of period j given survival to the start of period j) as follows:
hR X 0 i h 
b
b  i
2 X j vðtÞdt 2 l X 0j 2 X j
Rj ¼ e j
¼e forj ¼ 1; :::; T ð20Þ
Therefore, the probability of the system surviving the entire planning horizon is:
h 
b
b  i
YT
2 l X 0j 2 X j
Reliability ¼ e ð21Þ
j¼1

Now we must define the improvement factor because it affects the effective ages of the
system, which in turn affects the cost and reliability defined in the two previous
objective functions. Here we will propose a new improvement factor model as a Optimal
function of maintenance and replacement costs, and effective age of the system at the
end of previous period:
preventive
! maintenance
 
0 R2M X 0j21
aj ¼ fðR; M ; X j21 Þ ¼ · forj ¼ 1; :::; T ð22Þ
R X 0j21 þ 1
281
The first term is the constant coefficient based on the ratio of difference of replacement
and maintenance costs, which is always between zero and one. It is designed so that if a
costly maintenance action is performed on a system, the effective age improves more
than when inexpensive maintenance is performed. That is, more expensive
maintenance results in a greater amount of age reduction. For example, overhauling
an engine results in more age reduction that does changing the oil. Note that if
maintenance cost is equal to the replacement cost, the numerator of the fraction
becomes zero, and maintenance action will coincide with replacement action. On the
other hand, if the maintenance cost is equal to zero, the ratio becomes one and it means
that maintenance does not affect the effective age and it can be considered as do
nothing.
The second term is a ratio is the effective age at the end of previous period, which is
always less than one. The minimum value is obtained whenever the system is replaced
at the previous period. It can be seen that the ratio increases by increasing the effective
age and the amount of age reduction decreases as the system ages over the planning
horizon.
We present the following nonlinear mixed-integer programming model that
minimizes the total cost subject a required reliability of the system:
T 
X  
b
b
 
MinTotalCost ¼ F ·l X 0j 2 X j þ M · mj þ R · r j
j¼1

s:t: :
X1 ¼ 0
X j ¼ ð1 2 mj21 Þð1 2 r j21 ÞX 0j21 þ mj21 ðaj · X 0j21 Þj ¼ 2; :::; T
X 0j ¼ X j þ TJ j ¼ 1; :::; T ð23Þ
mj þ r j # 1j ¼ 1; :::; T
h 
b i
Y
T
2 l X 0j 2ð X j Þ
b

e $ RR
j¼1

mj ; r j ¼ 0or1j ¼ 1; :::; T
X j ; X 0j $ 0j ¼ 1; :::; T

The first constraint addresses the initial age of the system at the beginning of the
planning horizon. The second and third sets calculate the effective age of the system
based on preventive maintenance activities recursively. The fourth set prevents the
JQME occurrence of simultaneous maintenance and replacement actions on the system. The
main constraint of the model ensures that reliability of the system is greater or equal
16,3 than the required reliability. Finally, the last two sets of constraints restrict the
decision variables to be binary and positive.
Alternatively, we can modify the formulation and introduce a budgetary constraint,
GB. The objective of this model is to maximize the reliability, through our choice of
282 maintenance and replace decisions, such that we do not exceed the budgeted total cost.
This model can be formulated as:
h 
b
b  i
Y
T
2 l X 0j 2 X j
Max Re liability ¼ e
j¼1

s:t: :
X1 ¼ 0
X j ¼ ð1 2 mj21 Þð1 2 r j21 ÞX 0j21 þ mi;j21 ðaj · X 0j21 Þj ¼ 2; :::; T
X 0j ¼ X j þ TJ j ¼ 1; :::; T ð24Þ
mj þ r j # 1j ¼ 1; :::; T
T 
X  
b
b
 
F ·l X 0j 2 X j þ M · mj þ R · r j # GB
j¼1

mj ; r j ¼ 0or1j ¼ 1; :::; T
X j ; X 0j $ 0j ¼ 1; :::; T

5. Computational results
In order to show the effectiveness of the proposed mathematical models, we consider a
system with l ¼ 0:00025 and b ¼ 2:20 as the characteristic life (scale) and the shape
parameters and consider failure, maintenance, and replacement costs equal to $2,500,
$300, $1,500 respectively. In addition, we assume RR ¼ 92 percent as the required
reliability for model 1, GB ¼ $6000 as the given budget for model 2, and 36 months as
the planning horizon. In addition, we define three types of improvement factor function
as follows:
 
R2M
a1;j ¼ f1 ðR; M Þ ¼ ð25Þ
R
!
X 0j21
a2;j ¼ f2 ðX 0j21 Þ ¼ ; forj ¼ 1; :::; T ð26Þ
X 0j21 þ 1

  !
R2M X 0j21
a3;j ¼ f3 ðR; M ; X 0j21 Þ ¼ · ; forj ¼ 1; :::; T ð27Þ
R X 0j21 þ 1
The first function (25) calculates the improvement factor of the system based on the Optimal
ratio of difference of replacement and maintenance costs, which is constant over the
planning horizon. The second function (26) is a simple version of the original
preventive
formulation that uses only the ratio of effective age at the end of previous period. The maintenance
last function (27) is the original proposed model.
We employ the functions in our optimization models (23) and (24) and solve them
using LINGO (www.lindo.com) software. The optimal objective function value for both 283
models with different improvement factor functions is presented in Table I. As we can
see, by applying a variable improvement factor, equations (26) and (27), we can obtain
lower optimal value in model 1, minimizing total cost subject to reliability constraint,
and higher optimal value in model 2, maximizing overall reliability subject to
budgetary constraint, than considering constant improvement factor function;
equation (25). It is clear that variable improvement factor functions have an
advantage over constant improvement factor in terms of objective function value in
optimal solution.
Tables II and III illustrate the optimal schedules based on different improvement
factor functions in both models. As can be seen, by using a variable improvement
factor, the optimal schedules contain more maintenance activities than replacements
activities. Especially, by applying the third function the optimal schedule consists of
only maintenance actions. We also plot the variation of improvement factor functions
over the planning horizon in Figures 3 and 4. It can be seen that the constant coefficient
smoothes the second function and reduces its variability. We can state that equation
(27), which combines maintenance and replacement costs as a constant coefficient
along with effective age of the system as an independent variable can model the
improvement factor variations very well and result the better optimal solution than the
second function.
Finally, we conclude that the proposed improvement factor model has advantage
over the constant improvement factor and the variable improvement factor function
that uses just concept of effective age without considering maintenance and
replacement costs.

6. Conclusion
In this paper, we reviewed current improvement factor models applied in maintenance
scheduling optimization models. We developed mathematical equations to determine a
lower-bound for effective age of a maintainable and repairable system over a long-term
planning horizon. We presented two nonlinear mixed-integer optimization models to
find the optimal preventive maintenance and replacement scheduling for the system.
These models seek to minimize the total cost subject to achieving some minimal
reliability and maximize the reliability subject to a budgetary constraint. We also
introduced a variable improvement factor model and analyzed it using computational

Model 1 Model 2
Improvement factor Total cost Reliability (%) Reliability (%) Budget

Function 1 $8,002.54 92 89.45 $6,000 Table I.

Function 2 $7,707.74 92 89.66 $6,000 Optimal objective
Function 3 $6,506.86 92 91.17 $6,000 function values
JQME
Month/function Function 1 Function 2 Function 3
16,3
1 – – –
2 – – –
3 – – M
4 – – –
284 5 – – M
6 R R M
7 – – M
8 – M M
9 – – –
10 R – –
11 – – –
12 – – –
13 – – –
14 – R –
15 M – M
16 – – M
17 – – M
18 R – –
19 – R M
20 – – M
21 – M M
22 – M –
23 – M –
24 – – –
25 R – –
26 – – –
27 – – R
28 – R –
29 R – –
30 – M –
31 – – M
32 – – –
Table II. 33 – – –
Optimal maintenance and 34 – – –
replacement schedules in 35 – – –
model 1 36 – – –

results of the optimization models. We showed the advantages of it over the constant
models.
The developed models in this paper can be applied in a wide variety of industries
such as semiconductor manufacturing, transportation, material handling, and power
generation. Maintenance engineers and managers can use these models to design
maintenance plans in order to meet systems requirements and objectives. These
models can also be used to quickly generate new preventive maintenance and
replacement plans even after unexpected failures occur in the system. In such a
situation, the original schedule should be updated and the new optimal schedule will be
used over the remaining of the planning horizon. They can also be applied into
condition-based simulation models as a real-time optimization procedure to refine and
update maintenance plans during the simulation run. Future work in this area is
 Optimal
Month/function Function 1 Function 2 Function 3
preventive
1 – – – maintenance
2 – M –
3 – – –
4 – – M
5 M – – 285
6 – – M
7 – – –
8 – – M
9 R R M
10 – – M
11 – M –
12 – M M
13 – M M
14 – – M
15 – – –
16 R – M
17 – – –
18 – – M
19 – – M
20 – R –
21 M – M
22 M – M
23 – – –
24 – – M
25 – – M
26 R – –
27 – – M
28 – R M
29 – – M
30 – – M
31 M – –
32 – – –
33 – – – Table III.
34 – – – Optimal maintenance and
35 – – – replacement schedules in
36 – – – model 2

Figure 3.
Variation of improvement
factor functions in model 1
JQME
16,3

286
Figure 4.
Variation of improvement
factor functions in model 2

needed to investigate the effects of other constraints on system availability and

demand for limited maintenance resources.
It is also recommended to use of more complex stochastic programming techniques
for developing optimal preventive maintenance plans that better account for the
stochastic nature of the problem.

References
Ascher, H.E. and Feingold, H. (1984), Repairable Systems Reliability: Modeling, Inference,
Misconceptions, and Their Causes, Marcel Dekker, New York, NY.
Bartholomew-Biggs, M., Christianson, B. and Zuo, M. (2006), “Optimizing preventive
maintenance models”, Computational Optimization and Applications, Vol. 35 No. 2,
pp. 261-79.
Che-Hua, L. (2007), “Preventive maintenance scheduling via particle swarm optimization
method”, International Multi Conference of Engineers and Computer Scientists, 21-23
March, Kowloon, China, Vol. 1, pp. 121-6.
Cheng, C.Y. and Chen, M.C. (2003), “The periodic preventive maintenance policy for deteriorating
systems by using improvement factor model”, International Journal Applied Science and
Engineering, Vol. 1 No. 2, pp. 114-22.
Cheng, C.Y., Chen, M. and Guo, R. (2007), “The optimal periodic preventive maintenance policy
with degradation rate reduction under reliability limit”, 2007 IEEE International
Conference on Industrial Engineering and Engineering Management, 2-4 December,
Singapore, pp. 649-53.
Dedopoulos, I.T. and Smeers, Y. (1998), “Age reduction approach for finite horizon optimization
of preventive maintenance for single units subject to random failures”, Computers and
Industrial Engineering, Vol. 34 No. 3, pp. 643-54.
El-Ferik, S. and Ben-Daya, M. (2006), “Age-based hybrid model for imperfect preventive
maintenance”, IIE Transactions, Vol. 38 No. 4, pp. 365-75.
Jaturonnatee, J., Murthy, D.N.P. and Boondiskulchok, R. (2006), “Optimal preventive
maintenance of leased equipment with corrective minimal repairs”, European Journal of
Operational Research, Vol. 174 No. 1, pp. 201-15.
Jayabalan, V. and Chaudhuri, D. (1992), “Cost optimization of maintenance scheduling for a
system with assured reliability”, IEEE Transactions on Reliability, Vol. 41 No. 1, pp. 21-5.
Lim, J.H. and Park, D.H. (2007), “Optimal periodic preventive maintenance schedules with
improvement factors depending on number of preventive maintenances”, Asia-Pacific
Journal of Operational Research, Vol. 24 No. 1, pp. 111-24.
Lin, D., Zuo, M.J. and Yam, R.C.M. (2001), “Sequential imperfect preventive maintenance models Optimal
with two categories of failure modes”, Naval Research Logistics, Vol. 48 No. 2, pp. 172-83.
Malik, M.A.K. (1979), “Reliable preventive maintenance scheduling”, AIIE Transactions, Vol. 11
preventive
No. 3, pp. 221-8. maintenance
Martorell, S., Sanchez, A. and Serradell, V. (1999), “Age-dependent reliability model considering
effects of maintenance and working conditions”, Reliability Engineering and System Safety,
Vol. 64 No. 1, pp. 19-31. 287
Nakagawa, T. (1988), “Sequential imperfect preventive maintenance policies”, IEEE
Transactions on Reliability, Vol. 37 No. 3, pp. 295-8.
Usher, J.S., Kamal, A.H. and Hashmi, S.W. (1998), “Cost-optimal preventive maintenance and
replacement scheduling”, IIE Transactions, Vol. 30 No. 12, pp. 1121-8.
Xi, L.F., Zhou, X.J. and Lee, J. (2005), “Research on sequential preventive maintenance policy in
finite time horizon”, Computer Integrated Manufacturing Systems, Vol. 11 No. 10,
pp. 1465-8.

About the authors

Kamran S. Moghaddam is a Post-Doctoral Associate in the Department of Industrial Engineering
at the University of Louisville. He received his BS degree in Applied Mathematics from the
University of Tehran, his MS degree in Industrial Engineering from Tehran Polytechnic and his
PhD degree in Industrial Engineering from the University of Louisville in 2001, 2003, and 2010
respectively. He was an Industrial Engineer, Transportation Systems Analyst, and later a Project
Control Manager at the Tehran Comprehensive Transportation and Traffic Studies Company
(TCTTS) during 2002-2006. His research interests are systems reliability optimization,
healthcare operations management, supply chain management, and applied operations research.
Kamran S. Moghaddam is the corresponding author and can be contacted at:
[email protected]
John S. Usher is a Professor and Chair of the Department of Industrial Engineering at the
University of Louisville. He received his BS and MEng degrees in Industrial Engineering from
the University of Louisville in 1980 and 1981, respectively. He received his PhD in Industrial
Engineering from North Carolina State University in 1987. While completing this degree he was
employed by IBM in Research Triangle Park, North Carolina for two years as a Reliability
Engineer. He has conducted funded research in the areas of systems reliability, quality control,
and facilities design and analysis for organizations such as the National Science Foundation, US
Navy, IBM, AT&T Bell Laboratories, General Electric, Quaker and KFC. He is Past Director of
the Facilities Planning and Design Division of IIE and a Past President of the College Industry
Council for Material Handling Education (CIC-MHE). He is also member of IIE, ASQ, and IEEE,
and is a Registered Professional Engineer in the State of Kentucky.

To purchase reprints of this article please e-mail: [email protected]

Or visit our web site for further details: www.emeraldinsight.com/reprints

View publication stats