Вы находитесь на странице: 1из 22

Maintenance Policies With Two-Dimensional

Warranty
Tom Chen
a
Elmira Popova
b,
a
i2 Technologies, Irving, TX 75039, USA
b
Graduate Program in Operations Research and Industrial Engineering,
Department of Mechanical Engineering, The University of Texas at Austin,
Austin, TX 78712, USA
Abstract
A new maintenance policy which minimizes the total expected servicing cost for an
item with twodimensional warranty is proposed. An iterative procedure to estimate
the items failure rate function from historical observations and an optimization al-
gorithm based on Monte Carlo simulation are applied to obtain the best maintenance
policy. Numerical examples and sensitivity analysis are presented.
Key words: TwoDimensional Warranty, Maintenance, Simulation-Optimization

Corresponding Author. Address: Graduate Program in Operations Research and


Industrial Engineering, Department of Mechanical Engineering , The University
of Texas at Austin, Austin, TX 78712, USA, e-mail:elmira@mail.utexas.edu, fax:
(512)-232-1489
Preprint submitted to Elsevier Science 3 March 2002
1 Introduction
Warranties have become a critical segment of the industrial environment. Of-
ten, manufacturers use the warranty as a marketing tool to advertise the
quality of the products. Oering warranty usually results in additional costs
to the manufacturers. In general, the warranty servicing cost varies from 1%
to 10% of the total sales (see McGuire [11]). This cost depends on the war-
ranty length, item reliability and other costs. If warranty is used properly as
a marketing tool, it could result in increase in sales and market share. If the
revenue exceeds the warranty costs, it is worthwhile to sell the products with
warranty (see, for instance, Blischke and Murthy [1], [14], [15]).
There are two types of warranty policies: one and twodimensional. The one
dimensional warranty is characterized by an interval, called the warranty pe-
riod. The twodimensional warranty is characterized by a region in a two-
dimensional plane with one axis representing age and the other one usage.
For instance, when you buy a new car it usually comes with 3 years or 36,000
miles warranty, whichever comes rst. Under this warranty, the buyer is pro-
vided with a warranty coverage for a maximum time period of 3 years or a
maximum usage of 36,000 miles. If the usage is heavy, the warranty can expire
well before 3 years and if the usage is very light, then the warranty can expire
well before the limit of 36,000 miles is reached.
The case of onedimensional warranty is well studied in the literature(see
Nguyen and Murthy [19],[20], and Jack and Dagpunar [9], among all). We will
concentrate on the articles most relevant to our analysis.
The case of twodimensional warranty is discussed in several papers. Murthy
et al.,[16], use twodimensional renewal processes to model the item failure
behavior under the free replacement and no repair assumption. They represent
the expected warranty cost in terms of renewal functions for four dierent two
dimensional warranty policies. The main assumptions are that the failure of
an item is dependent on item usage and that the item can be in one of the two
statesin use or idle. The failure rate is a function of the age of the unit, the
number of times the current unit has been used, and the duration for which
the buyer has used it until time t. Yang and Nachlas [27] dene a structure for
classifying bivariate warranty problems, and methods for their building and
analysis.
Moskowitz and Chun [13] assume that the failure rate is a function of age and
usage and use a Poisson regression to model the number of failures. There
are several papers using Poisson regression to model failure occurences(see
Jorgeson [10], Gart [8] and Weber [23]). The main assumption is that the item
usage U and age T have a linear relationship: i.e., U = rT, where r is the ratio
2
between U and T. We will make the same assumption in our development.
Chun and Tang [2] assume that when a consumer buys the product, a warranty
does not come with the product but instead the customer has an option to
buy the warranty. The warranty price is analyzed with respect to such factors
as the producers risk preference, the consumers risk preference, consumers
perceptions about the product failure rate and the consumers loyalty to the
original producer in repairing failed products.
Two-attribute warranty policies for non-repairable items are described by Kim
and Rao [26]. They derive analytical expressions for the two-dimensional re-
newal function and cost of warranty under the assumption of bivariate ex-
ponential failure distribution. Chun and Tang [24] consider a two-attribute
warranty policy problem as well from a dierent point of view. They use the
warranty criteria to determine the eligibility of a warranty claim. They ana-
lyze several decision models that estimate the total expected cost under these
policies.
Maintenance policies during warranty are analyzed by several authors. Nguyen
and Murthy [19] consider the following policy: For a xed warranty length W,
the item is replaced by a new one if it fails during (0, W ) and replaced by
a repaired item (which doesnt have to be of the same age) if it fails during
(W, W). A failed item is repaired and added to the collection of the repaired
items if the age of the item is less than and has never been repaired before.
Jack and Dagpunar [9] consider an imperfect maintenance policy over a war-
ranty period. They further assume that each preventive maintenance can re-
duce the age of the machine by x units. Under a constant maintenance cost,
they derive an optimal policy. This assumption is later relaxed in Dagpunar
and Jack [5].
Several authors consider the repair/replacement policy following the expira-
tion of warranty from the consumers point of view. Dagpunar and Jack [4]
consider a consumer who purchases a product with a warranty of length W. If
the product fails before W, the manufacturer is assumed to perform minimal
repair. After the expiration of the warranty, the consumer has to bear all the
expenses of failures. They nd a cost limit a

, such that if the cost of repair


C
r
is smaller than a

, the customer should repair the product, otherwise, buy


a new product. A dierent method for estimation of the lifetime distribution
when additional eld data can be gather after the expiration of the warranty
is presented by Oh and Bai [25].
Chung [3] discusses a model with a maximum repair cost and a model with-
out a maximum repair cost. Sahin and Polatoglu [21] study two types of re-
placement policies following the expiration of warranty: the consumer applies
minimal repair for a xed period of time and replace the unit with a new one
3
at the end of this period, or the unit is replaced at the time of the rst fail-
ure following the minimal-repair period. An excellent summary of warranty
economic decision models is given by Thomas and Rao [28].
This paper proposes a new maintenance policy for twodimensional warranty
period. The policy denition is given in 2, simulationoptimization approach
to obtaining the optimal maintenance strategy and sensitivity analysis are
presented in 3.
Notation
W
u
: Warranty usage limit
W
t
: Warranty time limit
V
t
: Optimal time control limit
V
u
: Optimal usage control limit
C
r
: Cost of minimal repair
C
m
: Cost of replacement
(t, u) : Failure rate at age t and usage u
r : Ratio between the usage and time after the purchase, (r = u/t)
s : Ratio between W
u
and W
t
, (s = W
u
/W
t
)
h : Ratio between V
u
and V
t
, (h = V
u
/V
t
)
2 Maintenance policy with optimal control limits V
t
and V
u
Assume that the twodimensional warranty region is dened by the set
{(0, W
U
) (0, W
T
)}
The maintenance policy covers the maximum warranty period W
T
or the max-
imum warranty usage limit W
U
, whichever comes rst.
Policy denition:
If the item fails in the region dened by the set A = {(0, V
t
) (0, V
u
)}, then
the item is replaced by a new item for a cost of C
m
. If the item fails during the
rest of the warranty region, dened by the set B = {(0, W
u
) (0, W
t
)} \ {A},
then the item is minimally repaired for a cost of C
r
(see Figure 1), where V
u
and V
t
are selected to minimize the total expected warranty (or equivalently
maintenance) cost for the warranty region.
Insert Figure 1 here
This policy is a combination of replacement and minimal repair. As a special
case, one might consider a policy which consists of replacements or minimal
4
repairs only. The rst one is obtained by setting V
t
= W
T
, V
u
= W
U
and
V
t
= 0, V
u
= 0 for the second one.
Nguyen and Murthy [19] consider a similar strategy for onedimensional war-
ranty: the failed item is replaced by a repaired one if it fails during the second
part of the warranty period. However, the age of the repaired item could be
greater than the failed item and only an item which has never failed before
can be repaired. This is one of the main reasons we consider the strategy men-
tioned above: the failed item is minimally repaired to bring the item back to
as good as old state and the item can be repaired more than once. It is also
in our interests to search for a single policy that minimizes the total cost over
all customers. However, this policy might not be the best one for a particular
individual.
Suppose that
The usage of the item U
j
and the time T
j
after the purchase have linear
relation with ratio r
i
for customer i (i.e., U
j
= r
i
T
j
). For each customer,
the ratio r
i
could be dierent. For example, grandmothers car has smaller
r
i
than salespersons r
i
.
The probability distribution of r is known or it can be estimated from
historical information and maintenance records.
The item has an increasing failure rate (IFR) in both time and usage, (t, u),
i.e. its probability of failure increases with time or usage. We will adopt the
iterative maximum likelihood estimation procedure described in 2.1.
2.1 Failure rate estimation
Let N be a random variable for the number of failures. Assume that its
distribution is Poisson with parameter (t, u), where t = time, u = usage.
The corresponding probability mass function is:
P(N = k
j
) =
e
(t,u)
(t, u)
k
j
k
j
!
, where k
j
= 0, 1, 2, 3, . . .
The following functional forms for the failure rate are commonly used in the
twodimensional warranty literature:
(t, u) =
1
t
w
+
2
u
w
and
(t, u) = e
(
1
t
w
+
2
u
w
)
(see Moskowitz and Chun [13]).
5
Many other alternative functional forms are also available for the analysis. To
obtain an estimate of (t, u) we will follow the procedure described in Jorgeson
[10] and Weber [23].
As an illustration, let us assume that (t, u) =
1
t
3/2
+
2
u
3/2
. If the sample
size is s, then the likelihood function can be written as:
L(
1
,
2
) =
s

j=1
e

1
t
3/2
j
+
2
u
3/2
j
_
_

1
t
3/2
j
+
2
u
3/2
j
_
k
j
(k
j
)!
The maximum likelihood estimators of
1
and
2
,

1
and

2
, can be obtained
by solving the following equations:
k
s

j=1
t
3/2
j
_

1
t
3/2
j
+
2
u
3/2
j
_
k
j
1
=
s

j=1
t
3/2
j
k
s

j=1
u
3/2
j
_

1
t
3/2
j
+
2
u
3/2
j
_
k
j
1
=
s

j=1
u
3/2
j
This system of equations does not have a close form solution. Bellow is the
recursive procedure recommended by Jorgeson [10].
Let V be a diagonal matrix with elements
v
j
=
1
t
3/2
j
+
2
u
3/2
j
,
X be a matrix of dimension s 2 with elements
[t
3/2
i
u
3/2
i
], i = 1, . . . , s,
and K be an s 1 vector with elements k
i
, i = 1, . . . , s.
Jorgeson [10] shows that the minimum variance linear unbiased estimator for
= (
1
,
2
) is

=
_

1
,

2
_
= (X
T
V
1
X)
1
X
T
V
1
K
Since V is unknown, the problem is to obtain an estimate of V which in turn
gives an estimate of . Let

V
m
denotes the estimate of V obtained on the m
th
iteration and let the corresponding estimate of be
b
m
= (X
T

V
1
m
X)
1
X
T

V
1
m
K
Let

V
0
be the s s identity matrix and dene

V
m+1
= diag[x

1
b
m
, x

2
b
m
, , x

s
b
m
]
6
We repeat the iterations until a convergence is achieved. At that time
b = (X
T

V
1
X)
1
X
T

V
1
K,
where

V is the matrix such that

V
m+1
=

V
m
within a certain tolerance level .
2.2 Policy analysis
The objective function of the policy dened in 2 will consist of the sum of
the replacement and maintenance cost. We would like to obtain the optimal
control limits, V
t
and V
u
, such that its minimum is obtained. Assume that
r is a random variable with probability distribution function G(.). Then the
objective function will equal to
Z (V
t
, V
u
) = min
{VtR
+
,VuR
+
}
E
r
[R
c
+R
m
] (1)
where E
r
[] is the expectation with respect to the random variable r, R
c
=
E [Replacement Cost], and R
m
= E [Minimal Repair Cost].
To reduce the dimensionality of the problem we consider the following cases:
r > max(h, s), min(h, s) < r < max(h, s), r < min(h, s). The probability of
being in each of the above cases can be easily found:
P [r > max(h, s)] = 1 G[max(h, s)] (2)
P [min(h, s) < r < max(h, s)] = G[max(h, s)] G[min(h, s)]
P [r < min(h, s)] = G[min(h, s)]
Conditional on the value of r, the objective function can be rewritten as:
Z (V
t
, V
u
) = min
{VtR
+
,VuR
+
}
{E
r
[R
c
+R
m
|r > max(h, s)] {1 G[max(h, s)]}
+E
r
[R
c
+R
m
|min(h, s) < r < max(h, s)] {G[max(h, s)] G[min(h, s)]}
+E
r
[R
c
+R
m
|r < min(h, s)] G[min(h, s)]}
For each of these cases the policy will reduce to an onedimensional problem.
Let r
i
be the ratio for customer i.
r
i
h and r
i
s: The item reaches V
u
before it reaches V
t
and reaches
W
u
before W
t
. The policy simplies to: If the item fails during [0, V
u
), then
the item is replaced. If the item fails during [V
u
, W
u
), then it is minimally
repaired.
Insert Figure 2(a) here
7
r
i
h and r
i
< s: The item reaches V
u
before it reaches V
t
and reaches W
t
before W
u
.In this case the onedimensional policy is : If the item fails during
(0, V
u
/r
i
), then the item is replaced. If the item fails during (V
u
/r
i
, W
t
), then
it is minimally repaired.
Insert Figure 2(b) here
r
i
< h and r
i
s: The item reaches V
t
before it reaches V
u
and reaches W
u
before W
t
. The policy simplies to: If the item fails during (0, r
i
V
t
), then
the item is replaced. If the item fails during (r
i
V
t
, W
u
), then it is minimally
repaired.
Insert Figure 2(c) here
r
i
< h and r
i
< s: The item reaches V
t
before it reaches V
u
and reaches
W
t
before W
u
.The policy simplies to: If the item fails during (0, V
t
), then
the item is replaced. If the item fails during (V
t
, W
t
), then it is minimally
repaired.
Insert Figure 2(d) here
To obtain the formulae for R
c
and R
m
let us dene N
1
(t, u) to be the process
for the number of replacements in the box (0, t) (0, u) and N
2
(t, u) to be the
process for the number of minimal repairs. The objective function becomes:
min
{VtR
+
,VuR
+
}
E
r
{C
m
E [N
1
(t, u)|(t, u) A] (3)
+C
r
E
_
N
2
(t, u)|(t, u) B, S
N
1
(t,u)
B
_
P
_
S
N
1
(t,u)
B
_
+C
r
E
_
N
2
(t, u)|(t, u) B, S
N
1
(t,u)
A
_
P
_
S
N
1
(t,u)
A
__
where S
N
1
(t,u)
is the time of the last replacement prior to (t, u).
The times of replacement will follow a renewal process since we are replacing
the failed item with a brand new one, i.e. each time we reset the lifetime
clock of the process to 0. The expected number of replacements will equal
to the expected number of renewals. Let m(t, u) be the expected number of
replacements(renewals) from time 0 to t and usage from 0 to u.
If F denotes the distribution function of the time between replacements and
F
n
its n-fold convolution, then m(t) =

n=0
F
n
(t), see Ross ([22], page 56).
Very few distributions have closed form of the renewal function m(t). There are
some authors who use Phase-type distributions to approximate it(see Neuts
[18] and Dekker [6]).
The item is minimally repaired if it fails during the remaining warranty region
B. Then, keep using the repaired item if no other failures occurred till the end
of the warranty region. The expected number of minimal repairs from 0 to
t equals to
_
t
0
h(x)dx, where h(x) is the failure rate function (see Florez and
Feldman [7], page 216.)
8
To obtain a formula for the associated minimal repair cost we need to condition
of the time of the last renewal/replacement, S
N
1
(t,u)
:
R
m
=C
r
_
E
_
N
2
(t, u)|V
t
< t < W
t
, V
u
< u < W
u
, S
N
1
(t,u)
B
_
(4)
P
_
S
N
1
(t,u)
B
_
+E
_
N
2
(t, u)|V
t
< t < W
t
, V
u
< u < W
u
, S
N
1
(t,u)
A
_

P
_
S
N
1
(t,u)
A
__
where the rst term can be simplied as follows:
E
_
N
2
(t, u)|V
t
< t < W
t
, V
u
< u < W
u
, S
N
1
(t,u)
B
_
= (5)
E [N
2
(t, u)|V
t
< t < W
t
, V
u
< u < W
u
, N
1
(V
t
, V
u
) = 0] =
_
A
(t, u)d(t, u)
and
P
_
S
N
1
(t,u)
B
_
= P [N
1
(V
t
, V
u
) = 0] = 1 F(V
t
, V
u
) (6)
The second term in the expected minimal repair cost involves the time of the
last replacement, S
N
1
(t,u)
being inside the rectangle A. Dene (t
0
, u
0
) to be the
coordinates of S
N
1
(t,u)
. Then,
E
_
N
2
(t, u)|V
t
< t < W
t
, V
u
< u < W
u
, S
N
1
(t,u)
A
_
= (7)
_
C
(t, u)d(t, u)
The set C is dened as {(0, W
t
) (0, W
u
)} \ {(0, t
0
) (0, u
0
)}. Following the
key renewal theorem we get the distribution of S
N
1
(t,u)
:
P
_
S
N
1
(t,u)
A
_
= [1 F(V
t
, V
u
)] +
_
A
[1 F(t x, u y)] dm(x, y) (8)
To obtain the complete formula for the expected minimal repair cost we need to
substitute (5), (6), (7), and (8) into (4). The above analysis provides complete
formula for the objective function. Obviously the minimum solution cannot
be obtained with conventional means. For that reason we will nd the optimal
control limits using simulationoptimization algorithm.
We should add that due to the four cases described above and the linear
relationship between time and usage we have to solve only one dimensional
9
problem. For instance, if the original failure rate function equals to (t, u) =
t +u, then for the rst case it reduces to (t, u) = (u/r
i
, u) = (/r
i
) u +
u =
1
(u). If r
i
h, then the item would reach V
u
before it reached V
t
and the
expected number of replacements will equal to: m(V
t
, V
u
) = m[(1/r
i
)V
u
, V
u
] =
m(V
u
). If r
i
< h, then the item would reach V
t
before it reaches V
u
and
the expected number of replacements will equal to m(V
t
, V
u
) = m[V
t
, r
i
V
t
] =
m(V
u
).
3 Simulation-optimization procedure for obtaining the optimal con-
trol limits
In this section we propose a simulation solution to the optimization prob-
lem dened in (3) given the initial twodimensional warranty region [0, W
t
]
[0, W
u
].
First we x the values of the decision variables (V
t
, V
u
) to (V
t1
, V
u1
). Then, we
simulate values of r
i
by sampling from its distribution. (Note that r
i
equals
to the ratio of the usage and time.) Given a value of r
i
we fall into one of
the four cases described above. Given a case we simulate N failure times and
obtain the average cost. Next we take a dierent set of values for (V
t
, V
u
) and
follow the same procedure to get the average warranty cost. The (V
t
, V
u
) which
gives the minimum average cost is the optimal control limit. We use a grid
algorithm with step sizes
t
and
u
to obtain the optimal solution.
The computational examples attempt to show the sensitivity of the objective
function with respect to one of the cost parameters, C
m
, and the failure rate
function, (t, u). The last set of computations show the robustness of the
optimal control limits when the random number seeds change.
3.1 Maintenance cost sensitivity analysis
Assume that the warranty region is dened by W
t
= 3 and W
u
= 4 and
the failure rate function is (t, u)=1.5t
3
2
+ 0.75u
3
2
. In the next examples, we
increase the replacement cost C
m
and keep everything else unchanged.
Example 3.1 C
m
= 2.5, C
r
=1, and r (3, 2), where (3, 2) is the Gamma
distribution with parameters 3 and 2.
From Table 1, the optimal control limits are in the neighborhood of (V
t
=
2.5, V
u
= 3.5). The zoom in results are listed in Tables 2 and 3. The sim-
ulation results show that when V
t
= 2.4 and V
u
= 3.3, the average cost is
10
minimized at 6.7. The average number of replacements is 2.3. As seen in Ta-
ble 3, the values of the neighboring points of (V
t
= 2.4, V
u
= 3.3) are consistent
with the optimal value.
Example 3.2 C
m
= 3, C
r
=1, and r (3, 2).
Similar to Example 3.1 we obtain that the optimal control limits equal to (see
Table 4)
(V
t
, V
u
) = (2.2, 3.1)
The zoom in results are given in Table 5 and the neighboring points analysis
in Table 6.
Example 3.3 C
m
= 3.5, C
r
=1, and r (3, 2).
From Table 7 the control limits are (V
t
, V
u
) = (2.1, 2.9), the zoom in results
are in Table 8 and Table 9.
A summary of the above three examples is given in Table 11. We can make the
following observation: Let V

t
and V

u
be the original optimal control limits.
If the new optimal control limits are V
new
t
and V
new
u
, respectively when C
m
decreases, then they lie outside of the region dened by (0, V

t
) (0, V

u
).
3.2 Comparison of dierent failure rates
Assume that the warranty is dened by W
t
= 3 and W
u
= 4. The replacement
costs are C
m
= 4,C
r
=1 and r (3, 2). The optimal control limits for the
following three dierent failure rates are compared.

1
(t, u) =t
3
2
+ 0.5u
3
2

2
(t, u) =1.5t
3
2
+ 0.75u
3
2

3
(t, u) =2t
3
2
+u
3
2
The failure rates are such that
2
(t, u) = 1.5
1
(t, u) and
3
(t, u) = 2
1
(t, u).
The optimal control limits for the above failure rates are summarized in Table
10. It appears that the parallel shifts of the failure rate function result into
bigger total warranty cost and number of replacements.
3.3 Stability of the control limits
Next we investigate the stability of the control limits for dierent random
seeds. Dierent random number seeds may lead to dierent control limits. If
11
the dierence between control limits is relatively small, we will condence in
the optimal control limits obtained by the simulation. To test their stability
dierent random number seeds are used for the following example:
Example 3.4 C
m
= 3, C
r
= 1, r (3, 2), and (t, u) = 1.5t
3
2
+ 0.75u
3
2
From the simulation results in Table 6, the optimal control limits in that run
are V
t
= 2.2 and V
u
= 3.1 with average cost 7.8. Fifteen more simulation runs
are performed to obtain the new control limits. They are listed in Table 12.
Table 12 shows that the points are close to each other. Figure 3 gives the
scatter plot of the optimal control limits for dierent random number seeds.
4 Conclusion
In this paper we present a maintenance/replacement policy under two dimen-
sional warranty. If the item fails during the region dened by [0, V
t
) [0, V
u
),
the item is replaced. If the item fails during the rest of the warranty region,
the item is minimally repaired.
This policy is very easy to implement. Once the optimal control limits are
obtained, the manufacturers only need to check the actual item usage and time
after purchase and compare that with (V
t
, V
u
) to determine which actions to
take. For those manufacturers that have many branches around the world (like
car dealers), it is particularly attractive. Getting the customers cooperation
is also very important for maintenance/replacement policies under warranty.
This policy can get the customers support very easily. The policy requires
repairs or replacements only if when item fails. Customers regular operations
would not be interrupted.
We study several cases with dierent cost structures and failure rates. Other
than just comparing the simulation results, we empirically show that if the
maintenance cost C
m
increase (minimal repair cost C
r
unchanged), the new
optimal control limits must lie outside of the region dened by the origin
and the original optimal control limits. Similarly, if the maintenance cost C
m
decreases with minimal repair cost C
r
unchanged, the new optimal control
limits must lie outside of the region dened by original optimal control limits
and the warranty limit. We also investigate the stability of the control limits.
Overall, this policy is easy to implement. The control limits are easy to nd. It
provides a good way to serve maintenance interventions under two-dimensional
warranty.
12
References
[1] W. R. Blischke and D. N. P. Murthy. Product warranty management - i: A
taxonomy for warranty policies. European Journal of Operational Research,
62:127148, 1992.
[2] Y. H. Chun and K. Tang. Determining the optimal warranty price based on the
producers and customers risk preferences. European Journal of Operational
Research, 85:97110, 1995.
[3] K-J. Chung. Optimal repair-cost limit for a customer following expiry of a
warranty. Microelectronics and Reliability, 34, 1994.
[4] J. S. Dagpunar and N. Jack. Optimal repair-cost limit for a consumer following
expiry of a warranty. IMA Journal of Mathematics Applied in Business and
Industry, 4:155161, 1992.
[5] J. S. Dagpunar and N. Jack. Preventive maintenance strategy for equipment
under warranty. Microelectronics and Reliability, 34(6):10891093, 1994.
[6] R. Dekker A simple approximation to the renewal function. IEEE Transactions
on Reliability, pages 7175, 1990.
[7] C. Valdez Florez and R. Feldman. A survey of preventive maintenance models
for stochastically deteriorating single - unit systems. Naval Research Logistics,
36:419 446, 1989.
[8] Gart. Regression analysis of Poisson-distributed data. Journal of the American
Statistical Association, 68:935940, 1973.
[9] N. Jack and J. S. Dagpunar. An optimal imperfect maintenance policy over a
warranty period. Microelectronics Reliability, 3(34):529534, 1994.
[10] D. W. Jorgenson. Multiple regression analysis of a Poisson process. Journal of
the American Statistical Association, 56:235245, 1961.
[11] E. P. McGuire. Industrial product warranties: Policies and practices. In The
Conference Board, New York. 1980.
[12] M. A. J. Menezes and I. S. Currim. An approach for determination of warranty
length. International Journal of Research in Marketing, 9:177195, 1992.
[13] H. Moskowitz and Young Hak Chun. A Poisson regression model for two-
attribute warranty policies. Naval Research Logistics, 41:355376, 1994.
[14] D. N. P. Murthy and W. R. Blischke. Product warranty management - ii: An
integrated frame work for study. European Journal of Operational Research,
62:261281, 1992.
[15] D. N. P. Murthy and W. R. Blischke. Product warranty management - iii:
A review of mathematical models. European Journal of Operational Research,
62:134, 1992.
13
[16] D. N. P. Murthy, B. P. Iskandar, and R. J. Wilson. Two dimensional failure
free warranties: Two-dimensional point process models. Operations Research,
43:356366, 1990.
[17] D. N. P. Murthy and D. G. Nguyen. An optimal repair cost limit policy for
servicing warranty. Mathematical Comput. Modeling, 11:595599, 1988.
[18] M. Neuts. Matrix-Geometric Solutions in Stochastic Models: An Algorithm
Approach. Johns Hopkins University Press, Baltimore, 1981.
[19] D. G. Nguyen and D. N. P. Murthy. An optimal policy for servicing warranty.
Journal of the Operational Research Society, 37:10811098, 1986.
[20] D. G. Nguyen and D. N. P. Murthy. Optimal replace-repair strategy for servicing
products sold with warranty. European Journal of Operational Research,
39:206212, 1989.
[21] I. Sahin and H. Polatoglu. Maintenance strategies following the expiration of
warranty. IEEE Transactions on Reliability, 45(2):220228, 1996.
[22] S. Ross. Stochastic Processes. John Wiley and Sons, New York, 1983.
[23] D. C. Weber. Accident rate potential: An application of multiple regression
analysis of a Poisson process. Journal of American Statistical Association,
66:285288, 1971.
[24] Y. H. Chun and K. Tang. Cost analysis of two-attribute warranty policies based
on the product usage rate. IEEE Transactions of Engineering Management,
46:201209, 1999.
[25] Y. S. Oh and D. S. Bai. Field data analysis with additional after-warranty
failure data. Reliability Engineering & System Safety, 72:18, 2001.
[26] H. G. Kim and B. M. Rao. Expected warranty cost of two-attribute
free-replacement warranties based on a bivariate exponential distribution.
Computers & Industrial Engineering, 38:425434, 2000.
[27] S. -C. Yang and J. Nachlas. Bivariate reliability and availability modeling.
IEEE Transactions on Reliability, 50:2635, 2001.
[28] M. U. Thomas and S. S. Rao. Warranty economic decision models: A summary
and some suggested directions for future research. Operations Research, 47:807
820, 1999.
14
V
u
|V
t
0.0 0.5 1.0 1.5 2.0 2.5 3.0
0.0 14.931 14.958 14.973 15.038 14.914 15.030 14.964
0.5 14.870 14.735 14.540 14.412 14.370 14.461 14.470
1.0 14.855 14.160 12.899 12.399 12.286 12.309 12.399
1.5 14.946 13.924 11.303 10.156 9.920 9.918 9.814
2.0 15.004 13.807 10.532 8.694 8.204 8.066 8.190
2.5 14.969 13.829 10.338 8.165 7.320 7.216 7.362
3.0 14.907 13.772 10.384 7.969 6.979 6.879 7.006
3.5 14.934 13.758 10.356 7.998 6.975 6.810 6.983
4.0 14.878 13.848 10.301 8.017 7.133 7.004 7.301
Table 1
The costs for servicing the warranty for C
m
= 2.5, C
r
=1, and r (3, 2)
V
u
|V
t
2.2 2.3 2.4 2.5 2.6 2.7
3.1 6.853 6.852 6.821 6.837 6.841 6.858
3.2 6.851 6.780 6.776 6.799 6.794 6.826
3.3 6.849 6.780 6.776 6.817 6.817 6.824
3.4 6.820 6.829 6.774 6.776 6.799 6.826
3.5 6.820 6.830 6.807 6.800 6.816 6.829
3.6 6.851 6.827 6.784 6.805 6.816 6.889
3.7 6.925 6.836 6.853 6.837 6.905 6.921
Table 2
Zoom in results for C
m
= 2.5, C
r
=1, and r (3, 2)
V
u
|V
t
2.4000 2.4125 2.4250 2.4375 2.4500 2.4625
3.2250 6.844 6.799 6.752 6.810 6.786 6.781
3.2375 6.765 6.798 6.818 6.778 6.779 6.781
3.2500 6.794 6.798 6.768 6.792 6.783 6.835
3.2625 6.834 6.788 6.713 6.758 6.778 6.768
3.2750 6.775 6.789 6.762 6.803 6.791 6.772
3.2875 6.761 6.816 6.811 6.767 6.807 6.800
3.3000 6.760 6.792 6.795 6.741 6.773 6.770
Table 3
Neighboring points for the optimal control limit for C
m
= 2.5, C
r
=1, and r (3, 2)
15
V
u
|V
t
0.0 0.5 1.0 1.5 2.0 2.5 3.0
0.0 14.931 14.958 14.973 15.038 14.914 15.030 14.964
0.5 14.870 14.788 14.629 14.513 14.472 14.570 14.577
1.0 14.855 14.281 13.147 12.707 12.625 12.661 12.749
1.5 14.946 14.070 11.674 10.647 10.477 10.509 10.431
2.0 15.004 13.958 10.982 9.319 8.935 8.862 9.028
2.5 14.969 13.985 10.817 8.879 8.185 8.175 8.389
3.0 14.907 13.932 10.874 8.744 7.943 7.976 8.202
3.5 14.934 13.913 10.861 8.820 8.018 8.023 8.308
4.0 14.878 14.004 10.810 8.866 8.241 8.309 8.762
Table 4
The Costs for servicing the warranty for C
m
= 3, C
r
=1 and dierent values of V
t
and V
u
V
u
|V
t
2.0 2.1 2.2 2.3 2.4 2.5
2.7 8.098 8.004 7.997 8.044 8.011 8.047
2.8 8.072 8.002 7.966 7.935 8.003 8.033
2.9 8.004 7.979 7.891 7.899 7.892 7.993
3.0 7.952 7.931 7.933 7.921 7.931 7.942
3.1 7.989 7.910 7.943 7.880 7.918 7.965
3.2 7.959 7.892 7.866 7.929 7.933 7.967
3.3 7.979 7.977 7.916 7.938 7.902 7.988
3.4 8.011 7.939 7.915 7.957 7.935 7.961
Table 5
Zoom in table for C
m
= 3 and C
r
=1
16
V
u
|V
t
2.1750 2.1875 2.2000 2.2125 2.2250 2.2375
3.0750 7.908 7.912 7.894 7.859 7.950 7.871
3.0875 7.883 7.906 7.925 7.916 7.875 7.881
3.1000 7.880 7.915 7.943 7.882 7.887 7.917
3.1125 7.865 7.905 7.839 7.889 7.901 7.894
3.1250 7.918 7.891 7.860 7.922 7.875 7.892
3.1375 7.918 7.912 7.914 7.892 7.893 7.890
3.1500 7.886 7.896 7.877 7.929 7.879 7.893
Table 6
Costs for the neighboring points around the optimal value for C
m
= 3 and C
r
=1
V
u
|V
t
0.0 0.5 1.0 1.5 2.0 2.5 3.0
0.0 14.931 14.958 14.973 15.038 14.914 15.303 14.964
0.5 14.870 14.842 14.717 14.614 14.573 14.679 14.684
1.0 14.855 14.402 13.394 13.015 12.963 13.012 13.099
1.5 14.946 14.216 12.045 11.139 11.034 11.101 11.049
2.0 15.004 14.110 11.432 9.944 9.667 9.658 9.865
2.5 14.969 14.140 11.296 9.588 9.049 9.134 9.415
3.0 13.932 14.091 11.365 9.520 8.907 9.074 9.397
3.5 14.934 14.072 11.367 9.642 9.061 9.236 9.633
4.0 14.878 14.160 11.318 9.714 9.349 9.614 10.222
Table 7
The maintenance costs for C
m
= 3.5, C
r
=1 and dierent values of V
t
and V
u
V
u
|V
t
1.8 1.9 2.0 2.1 2.2 2.3
2.7 9.060 9.034 8.964 9.004 8.948 9.040
2.8 9.044 9.012 8.945 8.908 8.956 8.953
2.9 9.064 9.009 8.904 8.948 8.926 8.934
3.0 9.058 8.941 8.981 9.002 8.909 8.937
3.1 9.107 8.999 9.006 8.977 8.966 8.975
3.2 9.107 8.999 9.006 8.977 8.966 8.975
3.3 9.078 9.071 9.000 8.973 8.973 9.007
Table 8
Zoom in table for C
m
= 3.5 and C
r
=1
17
V
u
|V
t
2.0500 2.0625 2.0750 2.0875 2.1000 2.1125
2.8250 8.906 8.951 8.951 8.940 8.922 8.962
2.8375 8.964 8.941 8.925 8.910 8.934 8.980
2.8500 8.936 8.955 8.954 8.942 8.908 8.908
2.8625 8.968 8.892 8.872 8.940 8.899 8.965
2.8750 8.929 8.964 8.912 8.900 8.966 8.939
2.8875 8.960 8.951 8.923 8.942 8.904 8.926
2.9000 9.007 8.903 8.966 8.959 8.948 8.915
Table 9
Maintenance costs for the neighboring points around the optimal control limits for
C
m
= 3.5 and C
r
=1
= t
3
2
+ 0.5u
3
2
= 1.5t
3
2
+ u
3
2
= 2t
3
2
+ u
3
2
Control Limit V
t
1.8875 1.9875 2.1500
Control Limit V
u
2.4875 2.8750 2.9250
Average Cost 7.823 9.811 11.418
Avg. of Replacements 1.342 1.860 2.272
Table 10
Summary of the control limits for dierent failure rates
C
m
= 2.5 C
m
= 3.0 C
m
= 3.5
Control Limit V
t
2.4 2.2 2.1
Control Limit V
u
3.3 3.1 2.9
Average Cost 6.7 7.8 8.9
Average of Replacements 2.2 2.1 2.0
Table 11
Summary of the simulation output for dierent cost structures
18
V
t
V
u
Cost of replacements of repairs
1 2.2378 3.0875 7.8306 2.0937 0.7647
2 2.1500 3.0500 7.8330 2.0207 0.9409
3 2.2750 3.1500 7.8346 2.1303 0.7478
4 2.2625 3.1000 7.8354 2.0994 0.7532
5 2.2750 3.0750 7.8326 2.1107 0.7505
6 2.2125 3.1750 7.8282 2.1011 0.8244
7 2.1125 3.0625 7.8370 1.9915 1.0364
8 2.2750 3.1625 7.8472 2.1435 0.7466
9 2.2750 3.0750 7.8326 2.1107 0.7163
10 2.2500 3.0875 7.8310 2.0922 0.7730
11 2.2500 3.1750 7.8280 2.1327 0.7521
12 2.2375 3.1500 7.8296 2.1174 0.7763
13 2.2500 3.1750 7.8220 2.1327 0.7521
14 2.2750 3.1375 7.8301 2.1228 0.7509
15 2.2500 3.1125 7.8399 2.1028 0.7796
Table 12
Control limits for dierent seeds
19


Usage Salespersons car

Wu



Repair


Vu


Replacement Grandmothers car




Vt Wt

Figure 1 : Maintenance policy and the effect of different users



Usage U=r
i
T

Wu
(Wu/r
i
, Wu)

(Vu/r
i
, Vu)
Vu





Vt Wt Time


Figure 2(a) r
i
? h and r
i
? s








Usage

Wu

U=r
i
T



(Wt, r
i
Wt)
(Vu/r
i
, Vu)
Vu




Vt Wt Time

Figure 2(b) r
i
? h and r
i
< s






Usage U=r
i
T

Wu
(Wu/r
i
,Wu)

Vu


(Vt, r
i
Vt)





Vt Wt Time


Figure 2(c) r
i
< h and r
i
? s

Usage

Wu


Vu
U=r
i
T

(Wt, r
i
Wt)
(Vt, r
i
Vt)





Vt Wt Time


Figure 2(d) r
i
<s and r
i
<h









0
0.5
1
1.5
2
2.5
3
3.5
4
0 0.5 1 1.5 2 2.5 3
Vt


Figure 3. Scatter plot of the control limits

Вам также может понравиться