Available online at www.sciencedirect.

com

MATHEMATICAL
AND
COMPUTER

8CIENCE~DIRECT"

EI.SEVIER

MODELLING

Mathematical and Computer Modelling 39 (2004) 373-380

www.elsevier.com/locate/mcm

An Optimal Maintenance Policy

for a S y s t e m under Periodic Overhaul
J. H. SEO AND D. S. BAI
Korea Advanced Institute of Science and Technology
Department of Industrial Engineering
Taejon 305-701, Korea

(Received December P002; accepted August 2003)

A b s t r a c t - - A preventive maintenance policy with minimal repair at failure, periodic overhaul, and
replacement is considered. A model describing the effect of overhaul is proposed and the expected cost
rate is obtained under negligible or nonnegligible maintenance time. Based on this model, optimal
number of overhauls and optimal interval between overhauls for minimizing the expected cost rate
over infinite time horizon are determined. Under some mild conditions, a unique optimal maintenance
policy exists and a closed-form expression for the optimal interval between overhauls is derived when
the lifetime of the system follows Weibull distribution. Numerical studies are made. (~) 2004 Elsevier
Ltd. All rights reserved.
K e y w o r d s - - P r e v e n t i v e maintenance, Periodic overhaul, Minimal repair, Virtual age.

NOMENCLATURE
T

scheduled interval between

overhauls

scheduled number of overhauls

until the system is replaced; it is
replaced at time NT

v~(T)

virtual age of the system at

overhaul

hn(t)

hazard rate in the n t h overhaul

period; i.e., during ((n - 1)T, nT]

Cl

c2

cost of scheduled overhaul

e3

cost of replacement

C(N, T)
X(n,k)

expected cost rate

n th

cost of minimal repair at failure

system lifetime after (k - 1)th

minimal repairs in the n t h overhaul
period

Y(n,k)

time at kth system failure

F(,~,k)

distribution function of Y(k,~)

maintenance time

1. I N T R O D U C T I O N
A n effective m a i n t e n a n c e policy can reduce frequency of failures a n d u n d e s i r a b l e consequences
of such failures. M a i n t e n a n c e clearly i m p a c t s on c o m p o n e n t a n d s y s t e m reliability; too little
m a i n t e n a n c e m a y result in a n excessive n u m b e r of costly failures a n d poor s y s t e m performance,
a n d therefore, reliability decreases; excessive m a i n t e n a n c e m a y i m p r o v e reliability, b u t the m a i n t e n a n c e cost will be s h a r p l y increased. Therefore, a m a i n t e n a n c e policy which holds t h e b a l a n c e
of t h e two e x p e n d i t u r e s is necessary.

0895-7177/04/$ - see front matter (~) 2004 Elsevier Ltd. All rights reserved.
doi: I0.1016/S0895-7177(04)00028-7

Typeset by .AMS-TEX

374

J.H. SEO AND D. S. BAI

Maintenance can be categorized into corrective and preventive ones. The corrective maintenance (CM), called repair, is an action performed to restore the system to functioning condition
when it fails. Preventive maintenance (PM) is an action performed to prevent failures when the
system is operating. In practice, a system that deteriorates with age usually receives repair and
two P M operations: replacement and overhaul. A replacement restores the system to a new one
and an overhaul is performed to eliminate any probable failure of the system.
Most previous studies on P M policies assume that the system is restored to an as good as
new state after PM. Pierskalla and Voelker [1] and Sherif and Smith [2] provide detailed surveys
on PM policies. Since an overhaul may affect only a limited number of components, it makes a
system 'better than old' but not as good as new. Nakagawa [3], Liu et al. [4], and P h a m and
Wang [5] emphasized the importance of overhaul.
The model describing the effect of overhaul is fundamental for establishing an appropriate P M
policy. Nakagawa [3] and Liu et al. [4] proposed a 'reduction model' and a 'virtual age model',
respectively. The reduction model assumes that the hazard rate after overhaul increases more
quickly than that before overhaul, and each overhaul resets hazard rate function and makes the
system as good as new. The virtual age model assumes that each overhaul decreases the hazard
rate of a system by a fixed amount. The virtual age model is relatively easy to analyze but it
does not changes hazard rate function. In reality, however, each overhaul may not be able to
eliminate all the probable failures. As a result, unlike the replacementl an overhaul cannot make
the system as good as new. In other words, an overhaul can only rejuvenate the system and
bring the condition of the system to a level somewhere between as good as new and just prior
to the overhaul. Since the probable failures not eliminated affect future reliability of the system,
the hazard rate function may become higher after each overhaul is performed. T h a t is, a model
which not only decreases the hazard rate of a system at each overhaul time but also changes
the hazard rate function after overhaul would be more appropriate. Therefore, the following P M
policy is considered; an overhaul is made at periodic times and the system is replaced by a new
one at the N TM overhaul.
This paper is organized as follows: Section 2 proposes a model describing the effect of overhaul.
Expected cost rates under the proposed model are obtained and numerical studies are performed
in the cases of negligible and nonnegligible maintenance times in Sections 3 and 4, respectively.

2. T H E

MODEL

A model is constructed using the virtual age function and increasing the slope of the hazard
rate function after overhaul. Figure 1 depicts hazard rate functions before and after the first
overhaul. The figure shows that an overhaul decreases the hazard rate but not to zero, and
the slope of the hazard rate function becomes larger; i.e., the hazard rate right after overhaul
is hl(t~T) and the slope is the same as h2(t), where t~ (0 < ~ < 1) is a measure describing the
effect of the overhaul on the virtual age. Therefore, the hazard rate right after overhaul can be
described as h2(vl(T)), where vl(T), called the virtual age, satisfies h2(vl(T)) = hl(0T).
We now derive the virtual age of the system after the n t h overhaul. Let t~ be the time of
t h e n t h overhaul, where to = 0; that is, t~ = nT, n = 1 , . . . , N .
Let vn(T) be the virtual
age right after the rt t h overhaul. The virtual age function of the system is a function of two
variables, V(v, T), that specifies the functional relationship between v and T. If the system has
a hazard rate function hn(t) in the n t h overhaul period, then the hazard rate of the system is
h~{V(v,~_I(T),T)} right after t~. Since the hazard rate function changes to h~+l(t) after the
n th overhaul, v~(T) is obtained by formula
h,~+l (v~ (T)) -- h~ {V (vn-1 ( T ) , T ) } .

(1)

Kijima et al. [6] and Kijima [7] measured the effect of the overhaul on the virtual age by a
multiplier ~ (0 < ~ < 1), and used the virtual age function V(v, X) = v+t~X. From this function,

An O p t i m a l Maintenance Policy

375

hazard rate

time

Figure 1. Hazard rate of the p r o p o s e d model.

the virtual age at the

n th

overhaul becomes

(2)

v~ (T) = h ~ l l [h~ (vn-1 (T) + OT)],

where v0(T) = 0.
3. P R E V E N T I V E
NEGLIGIBLE

MAINTENANCE
MAINTENANCE

POLICY:
TIME

Suppose that the system undergoes minimal repair at failure and maintenance time is negligible.
Then, failures over time occur according to a nonstationary Poisson process with failure rate
hn(t) in the n th overhaul period and the expected number of repairs in (Vn-l(T), Vn-l(T) + T]

[v,~-I(T)+Thn~,(t~; dr; see, for example,

is Jv,~-1(T)

[8]. Therefore, the expected cost in a renewal cycle is

cl ~-~=INf v~_I(T)+T,._I(T)h~(t) dt + (N - 1)c2 + c3, where cl is the cost of minimal repair, c2 is the
cost of overhaul, and ca is the cost of replacement with c3 ~ c2. The expected length of a renewal
cycle is NT. Thus, the expected cost rate over infinite time horizon is given by
C1

C (Y, T) =

EN Jv~-x[v~-~(T)+T(T) hn(t) dt + (N n=l

1)c2 + c3

(3)

NT

The optimal (N, T) minimizing the expected cost rate can be obtained by using a procedure
similar to Liu et al. [3] and Nakagawa [4].
Our purpose is to seek both the optimal number of overhauls N* and the optimal time T* which
minimize C(N, T) in (3). The N* can be obtained by the inequalities C(N + 1, T) > C(N, T)
and C(N, T) < C(N - 1, T), which imply

L(N, T) > ca - c2

L(N - 1, T) < -c 3 --- , c 2

and

Cl

Cl

where

/vN(T)+T
L(N,T) =

Nj.~(T )

N Lv~,I(T)+T
n=l

0,

h~(t)dt, if N = 1 , 2 , . . . ,

hN+l(t)dt- E
v,~-i (T)

if N = 0 .

(4)

376

J . H . SEO AND D. S, BAI

From the assumption that h~+l(t)

>

h~(t) for any t > 0, we obtain

L(N,T) - L(N - 1,T) = N kJvN(T)

hN+l(t) dt - J,N-I(T)

hN(t) dt ,

(5)
L ( N , T ) > [ vN(T)+T hg+l(t) dt
_

Jr jr(T)

j~oT hi(t) dt.

Thus, L(N, T) is increasing in N and, if limN-~oo hN(t) --~ oo then it tends to c~ as N --* oo.
Therefore, there exists a finite and unique N* which satisfies (4) for any T > 0.
Next, differentiating C(N, T) with respect to T and setting it equal to 0, we obtain

N[

{h~ (v~_I(T)+T). (v'_l(T) + 1) - h~ (v~-l(T)).V[_z(T)}.T-f~'-KT)+Th,(t)dt

J

k==l

v. _ I (T)

[(g- 1)C2-}-C3]

(6)

c1

If hk (t) is differentiable and strictly increasing to c~, then the left-hand side of (6) strictly increases
to c~. Thus, there exists a finite and unique T* which satisfies (4) for any integer N.
(N*, T*) can be obtained using the following procedure.
(i) Let N1 = 1 and compute T --- T1 satisfying OC(N1,T)
OT
= O.
(ii) Find N = N2 satisfying the inequalities C ( N + I , T J
> C(N, T J and C(N, T J <

c(g

1, T1).

(iii) Compute T = T2 satisfying OC(/V2,T)

_-- 0.
aT
(iv) If Nj -- Nj+I (j = 1, 2,...), set (N*, T*) = (Nj, Tj) and stop; otherwise, go to (ii).
We now obtain C(N, T) and compute the optimal T* for the Weibull hazard rate. If the
hazard rate in the n th overhaul period is hn(t) = a,~/3t~-1 and the virtual age function satisfies
formula (2), then

v~ (T)

~-~. ( ak ~ l/(#-n
k:l

\an+l

OT.

(7)

This can be shown by induction on n. For n : 1, (7) holds since a2jg(vl(T)) #-1 -- hfl(0T) #-1,
and vl(T) = (al/a2) 1/(~-1) OT. Suppose that (7) holds for ( n - 1), n > 1. With hn(t) =
an/3t ~-1, formula (2) becomes

a~+1fl (v~(T) )#-1 = a~13(v~-l(T) + OT)~-1 ,

and
=

{v~-i (T) + OT}

k----1

Therefore, (7) holds for every n >_ 1.

A n O p t i m a l M a i n t e n a n c e Policy

377

From (3) and (7),

C(N,T)

c l R (N, T) + ( N - 1) c2 c3

(s)

NT

where

R(N,T)

= r ~~

o~n

n=l

-\O~n']

kk=l

0+ 1

kk=l

T a b l e 1. O p t i m a l m a i n t e n a n c e policy for negligible m a i n t e n a n c e t i m e .

~=2

c..~3

~=3

N*

T*

C(N*,T*)

17.32

0.3464

10

19,06

0.5970

N*

~-4

T*

C(N*,T*)

N*

T*

5.31

5.21

C(N*,T*)

0.8469

3.16

1.2649

1.5367

2.91

2.4460

el

0.1

20

24.29

0.7135

5.20

2.0929

2.80

3.4525

50

28.95

1.0190

5.61

3.3168

2.86

5.7833

100

32.68

1.3711

5.86

4.9057

2.85

8.9668

17.32

0.3464

5.31

0.8469

3.16

1.2649

10

22.21

0.5852

5.88

1.6577

3.24

2.6723

20

22.68

0~641

5.59

2.3242

2.97

3.8930

50

26.44

1.1156

5.82

3.7988

2.91

6.7509

100

29.33

1.5273

5.86

5.7089

2.80

10.6584

17.32

0.3464

5.31

0.8469

3.16

1.2649

10

21.42

0.6070

5.60

1.7402

3.07

2.8222

20

28.49

0.8074

5.15

2.5255

2.70

4.2776

50

31.35

1.1910

6.65

4.2119

3.27

7.6052

100

33.29

1.6372

6.42

6.3697

3.00

12.0924

0.2

0.3

T a b l e 2. O p t i m a l m a i n t e n a n c e policy a n d cost s a v i n g for selected values of o v e r h a u l

cost.
t3----2
e2
- - x 100(%)
c3

N*

T*

fl----3

C(N*,T*)

A(%)

N*

T*

C(N*,T*)

A(%)

11.67

0.4925

22

3.38

1.2771

32

10

15.41

0.5191

18

3.52

1.3858

27

20

16.65

0.5607

11

4.55

1.5383

19

30

22.21

0.5852

5.88

1.6577

13

40

23.05

0.6073

6.03

1.7416

50

23.86

0.6286

0.6

6.17

1.8236

= 4
c2

- - x 100(%)
c3

N*

T*

C(N*,T*)

~C%)

1.94

1.9804

37

10

2.00

2.1711

31

20

2.54

2.4471

22

30

3.24

2.6723

14

40

3.30

2.8250

50

3.36

3.1205

378

J . H . SEO AND D. S. BAI

--x--fl=4
fl=3
--=--fl=2

- - = - -

40'

30'

20'

10'

10

20

30

40

50

Figure 2. Percentage of cost saving by overhaul.

PE
7"
6"
54"

3"

J"
j"
j"
1\1.._..__1

.-----

~ / I

-60

-40

-20

20

40

, ~, ~, l o o ( ~ )
60 cl

Figure 3. Percentage errors due to incorrect repair cost.

Numerical Study
Suppose that the time to failure of the system follows Weibull distribution with/3 = 2, 3, 4 and

1~an = 100 (0.gZ) n-1 (n = 1, 2 , . . . ) . T h a t is, the mean time to failure in the n th overhaul
period becomes ten percent shorter for every overhaul.
Table 1 gives the values of N*, T*, and the corresponding expected cost rate C(N*, T*) when
t~ = 0.1, 0.2, 0.3, c2/cl = 3, and c3/cl = 3, 10, 20, 50, 100, and shows the following.
(i) The replacement time and the number of overhauls become larger as replacement cost
gets larger.
(ii) For a fixed 8, as/3 gets larger, and hence, the system's hazard rate becomes larger, T*
becomes smaller.
(iii) For a fixed fl, as 8 gets larger, and hence, overhaul is less effective, the replacement time
and the number of overhauls become smaller.
To investigate further the effect of overhaul, additional numerical analyses are performed. Let A
be the percentage of cost savings of the optimal maintenance policy with overhauls over cost of
optimal maintenance policy with only minimal repairs and replacement which means that there
will be no overhauls for the system; that is, A - 100x (C(1, T{) - C(N*, T*))/(C(1, T{)). Table 2

379

An Optimal Maintenance Policy

and Figure 2 give the optimal N*, T*, and the effect of overhaul for 0 = 0.2 and c3/cl = 10, and
show that as the system's hazard rate gets larger, overhaul becomes more effective.
Estimation of the repair cost for a complex system is laborious, so that the effects of incorrect
estimate of repair cost should be investigated. The percentage error of C(N*, T*) due to incorrect
estimate of repair cost is defined as
PE = C ( N ' , T ' ) - C ( N * , T * )
x 100 (%),
(9)

C(N*,T*)

where C(N*, T*) is the minimal cost obtained with the correct repair cost and C(N', T') is the
cost obtained with incorrect estimate of repair cost. Figure 3 gives PE versus (8~ - c l ) / c ~ . 100(%)
for fl -- 2, 0 =- 0.2, c2/cl = 3, and ca/c1 = 10, and shows that underestimating repair cost causes
a larger PE increase than overestimating repair cost.

4. P R E V E N T I V E M A I N T E N A N C E
POLICY--NONNEGLIGIBLE MAINTENANCE

TIME

It is assumed that the system undergoes only minimal repair at failure and each minimal repair
takes constant time R. The maximum possible number of minimal repairs in (0, t) is V(t/R) + 1].
It follows that, in the n th overhaul period,

X(,,j) < t - ( k - 1 ) R

F(n,k)(t)=Pr

=Pr{N(t-(k-1)

R)>_k}.

(10)

{N(x); x >__0} forms a nonhomogeneous Poisson process with intensity h~(t) and
0,

F(n,k)(t) =

t < ( k - 1)R,
k--1 [ _ g n ( t _ (k - 1)R)] j

1 -- exp[-Hn(t - (k - 1)R)] E

J!

t _> (k - 1)R.

(11)

j=0

See, for example, [9].

The probability of k failures occurring in (0, t) is

Pn (t, k) =

(t) -- Ft.,k+1) (t),

(12)

where F(n,0)(t) = 1.
Then the hazard rate h,~(t, R) with minimal repair time R is expressed in terms of hn(t) and
p.(t, k) as follows.
Conditional on the system having survived beyond t and subject to k failures in (0, t], the
probability that it will fail in (t, t + (it] is given by the conditional hazard rate h.(t, R I k). On
removing the conditioning over k, we have
[(T/R)+I]

h,(t, R) =

h~(t, R I k). pn(t, k)

k=0

= h,(t), p,(t, 0)

+ ~

(13)

hn(t-kR).G(n,k)(t)+ hn(t-kR+x).g(.,k)(t+x)dx

k=l

where G(,~,k)(x) = p~(x - R, k)/p~(t, k).

The expected cost rate of formula (3) now becomes
Cl

C(N, T) =

EN f,v,~_,(T)TTv~_l(T) h~(t, R) dt+ (N - 1) c2 + ca

n=l

N (T + R)

(la)

The optimal solution (N*,T*) cannot be obtained analytically, so t h a t numerical methods such
as Powell algorithm and golden section search method may be used.

380

J . H . SEO AND D. S. BAI

Numerical

Study

S u p p o s e t h a t t h e t i m e to failure of t h e s y s t e m follows W e i b u l l d i s t r i b u t i o n w i t h / 3 = 2, 1 / a n =
100 x (0.81) n - l , n = 1, 2 , . . . . T a b l e 3 gives t h e values of N * , T*, a n d t h e c o r r e s p o n d i n g e x p e c t e d
cost r a t e C ( N * , T*) w h e n ~ = 0.2, cl/c2 -~ 3, c3/c2 = 3, 10, 20, 50, a n d R = 0.5, 1, 2, a n d shows
t h a t as t h e m a i n t e n a n c e t i m e g e t s larger, r e p l a c e m e n t t i m e b e c o m e s l a r g e r a n d t h e e x p e c t e d cost
r a t e b e c o m e s smaller.
Table 3. Optimal maintenance policy for nonnegligible maintenance time.

R=0.5
#

c3

R=2.0

R=I.0

N*

T*

C(N*,T*)

N*

T*

C(N*,T*)

N*

T*

C(N*,T*)

3
10
20
50
100

1
2
3
4
5

17.42
24.02
24.58
27.77
30.84

0.3454
0.5625
0.7297
1.0571
1.4336

1
2
3
4
5

18.86
25.38
26.20
28.64
31.56

0.3318
0.5470
0.7112
1.0381
1.3965

1
2
3
4
5

19.66
27.20
28.65
30.69
32.94

0.3250
0.5298
0.6845
1.0211
1.3886

3
10
20
50
100

1
2
3
4
5

5.66
5.89
5.74
5.87
5.91

0.8178
1.6217
2.1349
3.5136
5.3384

1
2
3
4
5

5.78
6.17
5.98
6.05
6.12

0.8140
1.5745
2.0917
3.4546
5.2678

1
2
3
4
5

5.92
6.33
6.14
6.29
6.47

0.8087
1.5334
2.0567
3.3868
5.1826

3
10
20
50
100

1
2
3
4
5

3.28
3.35
3.06
3.00
2.94

1.2306
2.4751
3.5526
6.2418
9.8754

1
2
3
4
5

3.41
3.43
3.17
3.12
3.07

1.2213
2.4124
3.5269
6.1851
9.7412

1
2
3
4
5

3.53
3.57
3.24
3.16
3.11

1.2164
2.3362
3.4875
6.1083
9.6854

Cl

5. C O N C L U S I O N S
W e have p r o p o s e d a n i m p r o v e d m o d e l for d e s c r i b i n g a s y s t e m s u b j e c t t o m i n i m a l r e p a i r a n d
o v e r h a u l l a n d e s t a b l i s h e d o p t i m a l m a i n t e n a n c e policies in t h e cases of negligible a n d nonnegligible m a i n t e n a n c e t i m e s . C o s t m o d e l s a r e c o n s t r u c t e d c o n s i d e r i n g m i n i m a l r e p a i r , overhaul,
a n d r e p l a c e m e n t . N u m e r i c a l s t u d i e s show t h a t o v e r h a u l b e c o m e s m o r e effective as t h e s y s t e m ' s
h a z a r d r a t e g e t s larger, a n d u n d e r e s t i m a t i n g r e p a i r cost is m o r e serious t h a n o v e r e s t i m a t i n g it.
A p o s s i b l e a r e a of f u r t h e r i n v e s t i g a t i o n w o u l d b e t o consider a v a i l a b i l i t y in t h e case of nonnegligible m a i n t e n a n c e t i m e , a n d t o r e l a x t h e a s s u m p t i o n of c o n s t a n t m a i n t e n a n c e t i m e .

REFERENCES
1. W.P. Pierskalla and J.A. Voelker, A survey of maintenance models: The control and surveillance of deteriorating systems, Naval Research Logistics 23, 353-388, (1976).
2. Y.S. Sherif and M.L. Smith, Optimal maintenance models for systems subject to failure--A review, Naval
Research Logistics 28, 47-74, (1981).
3. T. Nakagawa, Periodic and sequential preventive maintenance policies, Journal of Applied Probability 23,
536-542, (1986).
4. X.G. Liu, V. Makis and A.K.S. Jardine, A replacement model with overhauls and repairs, Naval Research
Logistics 42, 1063-1079, (1995).
5. H. Pham and H. Wang, IiJ perfect maintenance, European Journal of Operational Research 94, 425-438,
(1996).
6. M. Kijima, H. Moremura and Y. Suzuki, Periodical replacement problem without assumpting minimal repair,
European Journal of Operational Research 69, 75-82, (1988).
7. M. Kijima, Some results for repairable systems with general repair, Journal of Applied Probability 26, 89-102,
(1989).
8. W.R. Blischke and D.N.P. Murthy, Warranty Cost Analysis, Marcel Dekker, New York, (1994).
9. J.S. Dagpunar, Hybrid minimal repair and age replacement maintenance policies with nonnegligible repair
times, Naval Research Logistics 41~ 1029-1037, (1994).