Adv. Appl.

Probe 20, 479-482 (1988)

Printed in N. Ireland
© Applied Probability Trust 1988

A NOTE ON THE OPTIMAL REPLACEMENT PROBLEM

LAM YEH, * The Chinese University of Hong Kong

Abstract
In this note, we study a new repair replacement model for a deteriorating
system, in which the successive survival times of the system form a
geometric process and are stochastically non-increasing, whereas the
consecutive repair times after failure also constitute a geometric process but
are stochastically non-decreasing. Two kinds of replacement policy are
considered, one based on the working age of the system and the other one
determined by the number of failures. The explicit expressions of the
long-run average costs per unit time under these two kinds of policy are
calculated.
REPLACEMENT PROBLEM; STOCHASTIC MONOTONICITY; RENEWAL
PROCESSES

1. Introduction
Most repair replacement models assume that a failure system after repair will yield a
functioning system which is 'as good as new' i.e., the system is renewed after each failure. An
alternative assumption is to assume that a failure system after repair will function again, but
has the same failure rate and the same effective age at the time of failure: this leads to the
minimal repair model. In both cases, the repair times are often negligible (see e.g. Barlow
and Proschan (1965) or Ascher and Feingold (1984)), although there exist a large number of
repair replacement models where the repair times are taken into account (see e.g. Morimura
(1970)).
However, for deteriorating systems, the problem is different from that described above.
For example, in machine maintenance problems, after repair, because of the deterioration,
the functioning time of a machine will become shorter and shorter so the total functioning
time or the total life of the machine must be finite. On the other hand, in view of the ageing
and accumulative wear, the repair time will become longer and longer and tend to infinity:
i.e., finally the machine is non-repairable. Therefore, we have to consider a repair
replacement model for deteriorating systems as follows: the successive survival times are
non-increasing and will eventually die out, so that the total life of the system is finite;
moreover, the consecutive repair times are non-decreasing and will finally tend to infinity.
One possible way of modelling these deteriorating systems is to use the non-homogeneous
Poisson process. Another possible way is to introduce the following geometric process.
Definition 1. Given a sequence of random variables Xl' X 2 , ••• if for some a> 0,
{an-lXn, n = 1, 2, ...} forms a renewal process, then {Xn, n = 1,2, ... } is a geometric
process. a is called the parameter of the geometric process.
Definition 2. A geometric process is called a non-increasing geometric process if a ~ 1 and
a non-decreasing geometric process if a ~ 1.
Note, if a = 1, then the geometric process reduces to a renewal process.
It is easy to show that for a geometric process, if a > 1, then it is stochastically decreasing
and converges to 0 in probability; furthermore, if a > 1, then U = E:=l X; converges in the

Received 18 June 1986; revision received 25 January 1988.

* Postal Address: Department of Statistics, The Chinese University of Hong Kong, Shatin, N.T.,
Hong Kong.

479

https://doi.org/10.2307/1427402 Published online by Cambridge University Press

 480 Letters to the editor

sense of mean square with P (U < 00) = 1. Whereas, if a < 1, then it is stochastically increasing
and tends to infinity in probability (see Lam Yeh (1987) for more details). Therefore, for the
deteriorating systems considered above, if its successive survival times are increasing
geometrically with a constant rate, then we can model them by using a non-increasing
geometric process. Otherwise, if its successive survival times are decreasing in a general way,
we can still use a non-increasing geometric process as an approximation for modelling the
successive survival times; this is exactly like using a linear function as an approximation of a
general function. Similarly, we can use a non-decreasing geometric process for formulating its
consecutive repair times either precisely or approximately.

2. Model
Now, we are able to state the new repair replacement model as below.
Assumption 1. At the beginning, a new system is used.
Assumption 2. Whenever the system fails, we can repair it. Let X; be the survival time
after the (n -1)th repair, then {Xn , n = 1, 2, ... } forms a non-increasing geometric process
with parameter a ~ 1 and E(X 1 ) = A > O.
Assumption 3. Let J: be the repair time after the nth failure, then {J:, n = 1, 2, ...}
constitutes a non-decreasing geometric process with parameter 0 < b ~ 1 and E(Y1) = J.l ~ 0,
J.l = 0 means that the repair time is negligible.
Assumption 4. {Xn , n = 1, 2, ...} and {Yn , n = 1, 2, ... } are independent.
Assumption 5. The repair cost rate is C, the reward rate whenever the system is working is
r, without loss of generality, we assume that r = 1.
Two kinds of replacement policy are considered in this model.
(i) A replacement policy T is a policy of which we replace the system whenever the
working age of the system reaches T.
The working age T of a system at time t is the cumulative survival time by time t, i.e.,

(1)
u. + V n ~t<Un + 1 + Y:t
Un + 1 + Vn ~ t < Un + 1 + Vn + 1
where
n n

o, = LXi' Vn =L ~ and Uo=O, Vo=O.

i=1 i=1

(ii) A replacement policy N is a policy of which we replace the system at the time of Nth
failure since the last replacement.
Assumption 6. The system may be replaced sometime by a new and identical one. The
replacement cost under policy T is C1; the replacement cost under policy N is C 2 •
Under replacement policy T or N, the problem is to determine an optimal replacement
policy T* or N* respectively, such that the long-run average cost per unit time is minimized.

3. Long-run average cost

Let 1;. be the first replacement time, in general, for n ~ 2, let 1'" be the time between
(n - l)th replacement and nth replacement, then obviously, {1'", n = 1, 2, ... } forms a
renewal process. It is well known that the long-run average cost per unit time is given by
the expected cost incurred in a cycle
(2)
the expected length of a cycle
where a cycle is the time between two consecutive replacements (see, e.g. Ross (1970)).

https://doi.org/10.2307/1427402 Published online by Cambridge University Press

 Letters to the editor 481

Under the replacement policy T, denote the length of a cycle by W, then

(3) W = T + VK, UK < T ~ UK+ 1 , K = 0, 1, . · ..
From Definition 1 and Assumption 3, it follows that E(Yk) = J-l/b k-1. Then

E(W)=E(T+ ~l Yk)
=E[T+ ~l YkI(Uk<T)]
00 1
= T + P, ~l b k - 1 FiT - 0)
where I is the indicator and ~ is the distribution of Ui.
Thus, from (2) the long-run average cost per unit time C1(T) under policy T is given by
1 00

CJ-l k~l b k-1Fk(T - 0) + C 1 - T

(4) C1(T) = 1 00

T + J-l k~l bk-1~(T - 0)

= Ci(T) -1,
where
00 1
(C + 1)J-l ~ b k-1Fk(T - 0) + C1
(5) Ci(T) = k-l
00 1
T + J-l k~l bk-1Fk(T - 0)
Similarly, under replacement policy N, the long-run average cost per unit time C2(N ) is
given by
N-l 1 N 1
CJ-l E b k-1+ C2 - A E k-l
C (N) k_=_l k_=_l_a_
(6) 2 - N 1 N-l 1
E ak-l + J-l
A k=l E bk-1
k=l

= Ci(N)-1
where
N-l 1
(C+1)J-l ~ b k-1+ C2
(7)- C*(N) _ k-l
N=1,2,·· .'.
2 - N 1 N-l 1 '
E ak-l + J-l k=l
Ak=l E bk-1

Finally, we can determine the optimal replacement policy T* and N* by minimizing C1(T)
(or Ci(T)) and C2(N) (or Ci(N)) respectively. Furthermore, the minimization procedure can
be done by analytical or numerical methods.
In practice, we prefer to adopt the optimal policy N* rather than use the optimal policy T*,
because of the much simpler form of (6) or (7).

Acknowledgement
The author is grateful to the referee for his valuable comments which made much
improvement in the presentation.

https://doi.org/10.2307/1427402 Published online by Cambridge University Press

 482 Letters to the editor

References
ASCHER, H. AND FEINGOLD, H. (1984) Repairable Systems Reliability. Marcel Dekker, New York.
BARLOW, R. E. AND PROSCHAN, F. (1965) Mathematical Theory of Reliability. Wiley, New York.
LAM YEH (1987) Geometric Processes and Replacement Problem. Technical Report 42, Depart-
ment of Statistics, The Chinese University of Hong Kong.
MORIMURA, H. (1970) On some preventive maintenance policies for IFR. J. Operat. Res. Soc. Japan.
12, 94-124.
Ross, S. M. (1970). Applied Probability Models with Optimization Applications. Holden-Day, San
Francisco.
Ross, S. M. (1983) Stochastic Processes. Wiley, New York.

https://doi.org/10.2307/1427402 Published online by Cambridge University Press