Professional Documents
Culture Documents
a r t i c l e i n f o a b s t r a c t
Article history: In case of leasing, the user rents equipment for a predetermined time. During this period, all maintenance
Available online 16 May 2013 actions are performed by the lessor. The aim of this research paper consists in determining an optimal
maintenance policy for ensuring a minimum reliability, required by the customer. Two strategies are pro-
Keywords: posed: the first consists in performing preventive actions whenever the system reliability reaches a pre-
Leasing defined reliability threshold. These actions are characterized by a reduction of the system age. The
Preventive maintenance objective is therefore to determine the effectiveness factor of the optimal maintenance minimizing main-
Optimization
tenance costs. For the second strategy, ‘‘improving’’ actions replace corrective actions during an interval
Simulation
to be determined to minimize maintenance costs. The first strategy will be solved using a numerical pro-
cedure and the second strategy uses an algorithm of discrete event simulation.
Ó 2013 Elsevier Ltd. All rights reserved.
0360-8352/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.cie.2013.05.004
594 J. Schutz, N. Rezg / Computers & Industrial Engineering 66 (2013) 593–600
intervals between two preventive maintenance activities. Periodic C VI variable cost of improving action
policy is characterized exclusively by the search of the number of
bzczP0 floor of z
preventive actions. In this case, the time intervals are obtained by
dzezP0 ceiling of z
dividing the time horizon by the number of preventive maintenance
actions (Barlow & Proschan, 1967). In their paper, Zhou, Xi, and Lee
(2007) propose a new policy of preventive maintenance. The latter
is based on system reliability. The authors seek the optimal reliabil-
2.2. Problem description
ity threshold to reach in order to perform preventive actions.
Our research work is based on the maintenance policy devel-
In this paper, we consider new equipment available to a cus-
oped by Zhou et al. Unlike the author, the reliability threshold is re-
tomer. During the lease period, the lessor must perform all correc-
garded as a data given by the customer and we seek to determine
tive and preventive actions. Compared to the duration of the lease
the effectiveness degree of preventive maintenance. Indeed, this
period, the times related to maintenance actions, corrective or pre-
degree is a variable to determine the cost of preventive mainte-
ventive, are considered negligible.
nance actions. Moreover, we propose a new threshold, named
Corrective actions are intended to return the equipment in
improving reliability threshold. After this threshold, minimal cor-
working condition and the failure rate is unchanged. Therefore, it
rective actions are replaced by imperfect maintenances.
corresponds to the minimal repair model ‘‘As Bad As Old’’ (Barlow
The rest of this paper will be organized as follows. Section 2
& Proschan, 1967). Preventive maintenances are characterized by
gives a description of the problem/maintenance strategy. Section 3
an age reduction of the system according to model II, developed
presents the mathematical formulation and the methodology used
by Kijima (1989) also known as ARA model (Arithmetic Reduction
to solve this problem. Section 4 focuses on a numerical approach to
of Age Memory 1) (Doyen & Gaudoin, 2004). Unlike model I,
illustrate our methodology. Finally, Section 5 provides the conclu-
where the effect of maintenance is to reduce the functional age
sion and the future works.
by an amount proportional to the time elapsed since the last main-
tenance, model II is characterized by a reduction of the functional
2. Problem statement age itself. After the ith preventive maintenance, the functional age
of the system can be expressed as:
Before we describe precisely the problem, it seems appropriate
to specify the different notations employed in the remainder of this Ai ¼ ð1 qÞ ðAi1 þ xi Þ ð1Þ
paper. When the effectiveness factor is equal to 0 and 1, preventive
maintenances are considered minimal and perfect, respectively.
2.1. Notations For all values in the interval ]0, 1[, preventive activities are re-
garded as imperfect actions. Consequently, the equipment is re-
The various concepts and notations used in this article are de- stored to an intermediate state, between AGAN (As Good As
scribed below: New) and ABAO (As Bad As Old). As the effectiveness factor is in-
cluded in the interval ]0, 1[, it is logical that the cost of preventive
L lease period actions depends on the effectiveness of these preventive actions.
{n}n>0 number of preventive maintenance actions However, it would be illogical to consider a cost proportional to
{q}0<q61 effectiveness factor of preventive maintenance the effectiveness factor. For example, in a lease strategy, mainte-
actions nance crews associated with the lessor are not located where the
qn optimal effectiveness factor of preventive equipment is. It is therefore coherent that the cost of preventive
maintenance for a predetermined n action should be divided into a fixed cost (displacement, etc.)
{Ai}iP0 functional age of the equipment after the ith and a variable cost depending on the desired effectiveness (quali-
preventive maintenance action fied maintenance personnel, duration, etc.). The cost model can
{xi}iP1 duration of preventive maintenance interval, i.e. be expressed as follows:
time between the (i 1)th and ith preventive ðqþ0:5Þ
maintenances C P ðqÞ ¼ C FP þ C VP ð2Þ
{yi}iP1 duration for each ith preventive maintenance
interval, during which failures are fixed by minimal The use of the power (q + 0.5) allows to obtain a cost that is not
actions proportional to the effectiveness factor. To reach a ‘‘perfect’’ main-
F(t) cumulative density function tenance, like a replacement, it seems logical that the cost is high; it
k(t) Hazard rate function is based on the Pareto principle.
RTH_P reliability threshold required to perform preventive This proposed maintenance strategy aims to ensure the cus-
activities tomer a minimum reliability of its equipment. The aim of this pa-
RTH_I reliability threshold at which corrective actions are per is to find the total cost of maintenance actions for a minimum
replaced by improving actions desired reliability and consequently, to determine the optimum
C() total cost of maintenance dates of preventive maintenance. It is also possible to determine
CC cost of correction action the minimum reliability required to minimize the total mainte-
CP(q) cost of preventive activity based on effectiveness nance costs. In the models developed in this paper, the preventive
factor maintenance effects are better than corrective maintenance effects
C FP fixed cost of preventive action (restored to a state ‘‘As Bad As Old’’) and the cost is based on the
effectiveness factor. Meanwhile, corrective maintenance is unex-
C VP variable cost of preventive action
pected and the cost is usually more important than a preventive
CI(q) cost of improving activity based on effectiveness
action. Also, the chosen modeling cost assumes the corrective
factor
maintenance cost is equivalent to an average preventive action
C FI fixed cost of improving action
cost (with q = 0.5)
J. Schutz, N. Rezg / Computers & Industrial Engineering 66 (2013) 593–600 595
total maintenance cost is expressed, with the factor q, by the fol-
C P ðqÞ > C C ¼ C P ðqÞ > C P ðqÞ ð3Þ
0:5<q61 q¼0:5 06q<0:5 lowing equation:
CðqÞ ¼ C C /ðqÞ þ C P ðqÞ n ð6Þ
2.2.1. Policy I In Eq. (6), /(q) represents the average number of failure for the
For the first policy, preventive actions are performed when the lease period L. The number of preventive maintenance activities is
system reliability reaches the reliability threshold fixed in the leas- computed as follows:
ing contract. Fig. 1 gives a representation of the system reliability
L x1
evolution over time. n¼ ð7Þ
q x1
The aim consists in searching the optimal value of q which min-
imizes the total maintenance cost. The intervals {xi}i>2 are constant and equal to q x1 because the
effectiveness factor is constant and the same for each preventive
maintenance interval. In fact, intervals xi vary between 0 (if
2.2.2. Policy II
Policy II is based on the policy I for which improvements are
q = 0) and x1 (if q = 1). When q is in ]0, 1[, the interval duration
xi is determined by the time to reach the reliability threshold
added to the maintenance strategy described previously. We seek
RTH_P from the virtual age Ai1. As the virtual age is given by
an interval ½RTH I ; RTH P Þ during which the corrective maintenance
Eq. (1), the preventive maintenance interval duration is
actions will be replaced by maintenance actions called ‘‘improv-
(x1 (1 q) x1) = q x1.
ing’’. The effects of these improving actions are the same as the ef-
For the maintenance strategy, based on an age reduction of the
fect of preventive actions, i.e. based on Kijima Model II (age
system, the average number of failures is expressed by:
reduction based on the effectiveness factor q).
Z Pn
As these ‘‘improving’’ actions are not planned, and they improve x1 n1 Z
X Ai þxiþ1 Z L A
i¼1 i
the state of the system, the cost will exceed any corrective and pre- /ðqÞ ¼ kðtÞdt þ kðtÞdt þ kðtÞdt ð8Þ
0 i¼1 Ai An
ventive actions. The various costs are defined as follows: Z Z Z
x1 x1 Lnqx1
The costs of improving actions are based on preventive costs, i.e. As preventive actions are performed when the reliability
with fixed and variables costs. They are defined by: reaches the threshold RTH_P, we obtain the following equation:
ðqþ0:5Þ Z x1
C I ðqÞ ¼ C FI þ C VI ð5Þ kðtÞdt ¼ lnðRTH P Þ ð10Þ
0
So, for this new maintenance policy, the decision variables are The optimal value of the effectiveness factor q⁄ is obtained by
RTH_I and q. canceling the partial derivative of cost with respect to the effec-
tiveness factor. However, the value of q appears in the determina-
3. Mathematical formulation tion of the number n (Eq. (7)) and consequently in the floor.
Drawing on the work of Yeh and Chang (2007), one possibility
3.1. Policy I would be to consider n and q as two decision variables and to can-
cel the following two partial derivatives:
For the first maintenance policy, noted Policy I, the aim consists @ Cðq;nÞ @C P ðqÞ
in determining, for the lease period, the relationship between the ¼ C C n x1 ½kðð1 qÞ x1 Þ kðL n q x1 Þ þ n ¼ 0 ð11Þ
@q @q
effectiveness of maintenance actions and the total maintenance "Z #
x1
@ Cðq;nÞ
cost. Subsequently, we seek the optimal value of the effectiveness ¼ CC kðtÞdt q x1 kðL n q x1 Þ þ C P ðqÞ ¼ 0 ð12Þ
@n ð1qÞx1
factor to minimize maintenance costs. For the lease period L, the
However, we note that the two decision variables are incorpo- By applying the same methodology to determine the maximum
rated into the function of failure rate. Solving these two equations value for n, we obtain nmax = 1 when q ? 0. To avoid this problem,
requires the use of a given distribution. Therefore, without a given we will ensure the existence of a local minimum, which validates
distribution, it is not possible to obtain analytical expressions giv- the existence of an optimal number of preventive maintenance.
ing directly the results of n and q. To avoid this mathematical prob-
lem, we opted for a numerical solution. In this case, the 9 n if :
distribution must be specified in the numerical procedure. From C C / qnþ1 ; n þ 1 / qn ; n
Eq. (7), it appears that the value of n is based on the value of q. q þ0:5
þ ðn þ 1Þ C VP
nþ1
L x1
L ½x1 þ ðn 1Þ q x1 > 0q < ð14Þ
ðn 1Þ x1
The proof of this lemma is given by Appendix A. Knowing that
Consequently, for a fixed n, the values of q are in the interval there exists an optimal value of n⁄ which satisfies the relationships
below: expressed by Eq. (17), it is possible to use a numerical procedure
" # (Fig. 3).
L x1 L x1 8
q2 ; ð15Þ C qnþ1 ; n þ 1 C qn ; n P 0
n x1 ðn 1Þ x1 >
>
>
>
>
< C qn ; n C qn1 ; n 1 6 0
We just determine the interval of the effectiveness factor q
n ¼ lim C qn ; n ¼ constant ð18Þ
based on the variable n. Currently, we are seeking to minimize >
> n!nmin
>
>
the number of values in testing to perform as far as the number >
: lim C qn ; n ¼ þ1
n!1
of preventive actions, namely n, is concerned. The number of pre-
ventive actions is minimal when the effectiveness of maintenance
actions tends to 1. Also, we obtain:
3.2. Policy II
ðn þ 1Þ x1 > L
L x1 The second maintenance policy, named Policy II, differs from
n> the policy I by adding improving actions. These actions replace cor-
x1
rective actions when the system reliability is included in the inter-
L x1
nmin ¼ ð16Þ val ½RTH I ; RTH P Þ, with RTH_I > RTH_P. If the reliability of the
x1
equipment reaches directly the value RTH_P, imperfect preventive
J. Schutz, N. Rezg / Computers & Industrial Engineering 66 (2013) 593–600 597
10000 Table 2
Cost variation based on the parameter RTH_P.
Total maintenance cost
5500 actions replace corrective actions when the system reliability is in-
cluded in the interval ½RTH I ; RTH P Þ. The effectiveness of these actions
Total maintenance cost
5400
5300 is similar to preventive maintenance. Indeed, after an ‘‘improving’’
5200 or preventive activity, the system is restored to a previous age, fol-
lowing the value of the effectiveness factor q, based on the Kijima’s
5100
model II. Both strategies have been solved using numerical proce-
5000
dures and discrete event simulation algorithms. These resolution
4900
methods have been chosen to allow a generic resolution, applicable
4800 to any distribution (unlike analytical models developed by Yeh &
4700 Chang (2007), Yeh et al. (2009), Yeh, Lo, & Yu (2011), and Yeh, Kao
0,71 0,74 0,77 0,8 0,83 0,86 0,89 0,92 0,95 0,98 et al. (2011)). In future, this work may be extended by determining,
Reliability threshold at which corrective actions are for each strategy, optimal effectiveness factors for each preventive
replaced by “improving” actions or ‘‘improving’’ action. In these cases, the number of decision vari-
Fig. 6. Cost variation based on the parameter RTH_I for each optimal effectiveness ables increases. Also, the resolution of these issues will be based
factor. on the development of heuristics or the use of meta-heuristics.
These meta-heuristics have already solved works whose theme is
ð21Þ ! fairly near (Schutz, Rezg, & Léger, 2011).
Z x1
2 t
k dt ¼ lnðRTH P Þ
0 100 100 Appendix A
x 2
1
¼ lnðRTH P Þ ð22Þ
100 The proof of the existence of n⁄ is given by:
x ¼ 100 ð lnð0:7ÞÞð2Þ
1
1 C qnþ1 ;n þ 1 C qn ;n P 0
ðq
x1 ¼ 59:72
þ0:5Þ
CC / qnþ1 ;n þ 1 þ C FP þ C VP nþ1
ðn þ 1Þ
The next actions are realized every q x1 time unit. With this ðqn þ0:5Þ
numerical example, we obtain q x1 = 0.7 59.72 = 29.86. CC / qn ;n C FP þ C VP nP0
For the second policy, Table 3 shows changes in the total cost
ðq þ0:5Þ ðqn þ0:5Þ
C C / qnþ1 ;n þ 1 / qn ;n þ ðn þ 1Þ C VP n C VP P C FP
nþ1
depending on the threshold of reliability RTH_I.
For each reliability threshold at which corrective actions are re-
C qn ;n C qn1 ;n 1 6 0
placed by improving actions, the total maintenance cost (average ðqn þ0:5Þ
of three replications for each simulation) is obtained for the opti- C C / qn ;n þ C FP þ C VP n
mal value of q (q⁄) (see Fig. 6). Consequently, the effectiveness fac-
ð n1 þ0:5Þ
q
tor varies following the value of RTH_I. Although improving actions C C / qn1 ;n 1 C FP þ C VP ðn 1Þ 6 0
are more expensive than corrective and preventive ones, we find
C C / qn ;n / qn1 ;n 1 þ n
that for a value of RTH_I equal to 0.74, the total cost of maintenance ðqn þ0:5Þ ðq þ0:5Þ
C VP ðn 1Þ C VP 6 C FP
n1
actions is minimal (4738.36 mu). As a reminder, the minimum cost
obtained for Policy 1 was 4860.69 mu. Indeed, when the system is
near to a preventive action and it fails, it is interesting that main- Moreover, the limits to C qn give:
tenance crews will move only once to restore the system to a pre-
qn þ0:5
vious state rather than twice (corrective action and action lim C qn ; n ¼ C C / qn ; n þ C FP þ C VP n ¼ constant
preventive) within a relatively short time. We note that ‘‘improv- n!nmin |fflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflffl}
constant |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
ing’’ actions are economically attractive when RTH_I is in constant
qn þ0:5
]0.70, 0.85[. For these values, the total maintenance cost is gener-
ally lower than the optimal minimum cost obtained without RTH_I lim C qn ; n ¼ C C / qn ; n þ C P þ C VP F
n ¼ þ1
n!1 |fflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflffl}
as given in Table 1. Instead, we observe that it is not useful to make 0 |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
constant
these improving actions too early because the age reduction does
not prove sufficient to cover costs incurred by these actions in rela- So, there is a number n⁄ which satisfies the following relations:
tion to corrective ones. For policy II, the determination of preven- 8
tive dates is more difficult because they depend of ‘‘improving’’ >
> C qnþ1 ; n þ 1 C qn ; n P 0
>
> C q ; n C q ; n 1 6 0
actions. If an ‘‘improving’’ action is performed, there is no preven- >
< n
n1
tive maintenance action for the considered interval. Otherwise, the n ¼
> n!nmin qn ; n ¼ constant
>
lim C
preventive activity is performed q x1 time unit after the interval >
>
>
: lim C q ; n ¼ þ1
begins. n n!1
Pongpech, J., & Murthy, D. (2006). Optimal periodic preventive maintenance policy Yeh, R. H., Kao, K.-C., & Chang, W. L. (2009). Optimal preventive maintenance policy
for leased equipment. Reliability Engineering & System Safety, 91(7), 772–777. for leased equipment using failure rate reduction. Computers & Industrial
Schutz, J., Rezg, N., & Léger, J.-B. (2011). Periodic and sequential preventive Engineering, 57(1), 304–309.
maintenance policies over a finite planning horizon with a dynamic failure law. Yeh, R. H., Kao, K.-C., & Chang, W. L. (2011). Preventive-maintenance policy for
Journal of Intelligent Manufacturing, 22, 523–532. http://dx.doi.org/10.1007/ leased products under various maintenance costs. Expert Systems with
s10845-009-0313-7. Applications, 38(4), 3558–3562.
Smith, C., & Wakeman, L. (1985). Determinants of corporate leasing policy. Journal Yeh, R. H., Lo, H.-C., & Yu, R.-Y. (2011). A study of maintenance policies for second-
of Finance, 11(3), 895–908. hand products. Computers & Industrial Engineering, 60(3), 438–444.
Yeh, R. H., & Chang, W. L. (2007). Optimal threshold value of failure-rate for leased Zhou, X., Xi, L., & Lee, J. (2007). Reliability-centered predictive maintenance
products with preventive maintenance actions. Mathematical and Computer scheduling for a continuously monitored system subject to degradation.
Modelling, 46(5–6), 730–737. Reliability Engineering & System Safety, 92(4), 530–534.