You are on page 1of 7

An Optimal Periodic Replacement Policy for a Product Sold with a Two-Dimensional Warranty

Hennie Husniah 1and Bermawi P. Iskandar 2* Department of Industrial Engineering, Bandung Institute of Technology, Bandung 40132, INDONESIA Email: h_husniah@student.itb.ac.id1 bermawi@mail.ti.itb.ac.id2

Abstract. This paper deals with a periodic replacement policy after the expiry of a warranty for a product sold with a two-dimensional non-renewing failure replacement warranty. We model product failures using the one-dimensional approach and use a more appropriate formulation to model the effect of age and usage to the products degradation. Under the periodic replacement policy, for a given usage rate y, the product is repaired with minimal repair when it fails at age x, x < y and it is replaced with a new one if its age reaches y. We obtain the optimal value of y which minimizes the long-run average cost to the buyer and give a numerical example to illustrate the optimal solution. Keywords: Periodic replacement policy, Two-dimensional warranty, Long-run average cost, Accelerated failure time (AFT) model.

1. INTRODUCTION
This paper deals with a periodic replacement policy after the expiry of warranty for a repairable product sold with a two-dimensional non-renewing failure replacement warranty (NFRW). With NFRW, all failures under warranty are rectified by the manufacturer at no cost to the buyer. From the buyers viewpoint, when the product fails under warranty, the buyer only incurs the cost due to unable to use the product while the failed product is restored. After the warranty expires, all maintenance costs including the costs of each repair and replacement and the cost due to downtime are borne by the buyer. As a result, determining the optimal maintenance policy after the expiry of a warranty, which minimizes the long-run average maintenance cost is of a great interest to the buyer. A review of various maintenance policies developed can be found in Pierskalla and Voelker (1976) and ValdezFlores and Feldman (1989). However, most maintenance policies studied are characterized by one time scale e.g. age. A maintenance policy characterized by two time scales i.e. age and usage has received less attention in the literature (see e.g. Frickenstein and Whitaker (2003)). Nowadays, all durable products are sold with warranty, and hence the study of maintenance policies for such products needs to consider the warranty aspect.

Maintenance policies following the expiry of a warranty have been studied by Sahin and Polatoglu (1996), Jung and Park (2003), Yeh et al. (2007) and Jung, et al. (2008). But these researches deal with the case where a product is sold with a one-dimensional warranty. In automotive industry, products are sold with a two-dimensional warranty. For example, a dump truck is warranted for 36 months or 30000 miles, whichever occurs first. In this paper, we extend a periodic replacement policy to the case of a twodimensional warranty. This replacement policy is suitable for automobiles, trucks, etc., which is normally sold with a two-dimensional warranty in the market place. The outline of the paper is organized as follow. In Section 2 we give the model formulation. The periodic replacement policy considered is dependent on a usage rate and is characterized by one parameter. Section 3 deals with the analysis of the optimal replacement policy. Section 4 presents numerical examples for the case where the product has a Weibull failure distribution. Finally, in Section 5, we conclude with a brief discussion for future research.

2. MODEL FORMULATION 2.1 Warranty Policy and Coverage


The product is sold with a two-dimensional nonrenewing failure replacement warranty (NFRW) with

________________________________________ * : Corresponding Author


232

APIEMS 2008 Proceedings of The 9th Asia Pasific Industrial Engineering & Management Systems Conference
warranty region , the rectangle [0, W ) [0, U ) where W is the time limit and U is the usage limit. With NFRW, all failures under warranty are rectified at no cost to the buyer. It is assumed that the rectification is done through a minimal repair and the repaired product comes with the original warranty. The warranty ceases at the first instance when the age of the product reaches W or its usage reaches U , whichever occurs first. If Wy denotes the warranty expiry time when the usage rate is y, then (see Fig.1)

intensity function ( x y ) which is a non-decreasing function of x and y . Moreover if all repairs are minimal (Barlow and Hunter, 1960) and repair times are negligible, then ( x y ). Murthy and Wilson (1991), Iskandar, et al. (1994), Moskowitz and Chun (1994) and Chun and Tang (1999) assume a linear relationship of the form

( x y ) = ax + byx = ax + bu ,

(3)

if y U / W , W Wy = U / y if y > U / W .

(1)

y >U W

U
y U W

with a, b 0. Lawless et al. (1995) have developed a different method, applying concepts from the accelerated failure time and proportional hazards models (see Lawless (1982) and Blischke and Murthy (2000)) to represent the effect of usage rate on reliability degradation. Conditional on the usage rate, the time to first failure has distribution function F ( x y ) = F0 ( xy ), (4) where F0 (x) is the base failure distribution function. Failures over time follow a non-homogeneous Poisson process with intensity function

( x y ) = y 0 ( xy 1 ),
W Figure 1: Two-dimensional warranty region .
Wy
where 0 ( x) is the base intensity function.

(5)

2.3.2 Modeling first failure 2.2 Usage Rate Modeling


We assume that the usage rate Y (e.g., the annual distance traveled for an automobile) varies across the customer population but it does not change for a given consumer . Y is a random variable with a density function g ( y ), 0 y < . Conditional on Y = y, the total usage at age x is a linear function of x and it is given by u = yx. (2) For a product sold with a two-dimensional warranty, one needs to model the products degradation which takes into account both age and usage. Nat, Iskandar and Murthy (2009) have introduced a more appropriate model which uses the accelerated failure time (AFT) model, to represent the effect of usage rate on degradation. Let y0 denotes the nominal usage rate value associated with component reliability. When the actual usage rate is different from this nominal value, the component reliability can be affected and this in turn affects the product reliability. As the usage rate increases above the nominal value, the rate of degradation increases and this, in turn, accelerates the time to failure. Consequently, the product reliability decreases [increases] as the usage rate increases [decreases]. Using the AFT formulation, if T0 [ Ty ] denotes the time to first failure under usage rate y0 [ y ] then we have
y = 0 . (6) T0 y Furthermore if the distribution function for T0 is given by F0 ( x; 0 ), where 0 is the scale parameter, then the distribution function for Ty is the same as that for T0 but

2.3 Failure Modeling 2.3.1 Approaches to modeling failures


Three approaches can be used to modeling failures for products sold with two-dimensional warranties (See Nat, Iskandar and Murthy (2009)). We use Approach 3 introduced by Murthy and Wilson (1991) to modeling failures. This approach assumes that the usage rate Y varies from customer to customer but is constant for a given customer. For a given usage rate y , failures over time are modeled by a one-dimensional counting process. If failed products are replaced by new ones, then this counting process is a renewal process associated with the conditional distribution F ( x y ) . If failed products are repaired then the counting process is characterized by a conditional

Ty

with scale parameter given by

Nusa Dua, Bali INDONESIA December 3rd 5th, 2008


233

APIEMS 2008 Proceedings of The 9th Asia Pasific Industrial Engineering & Management Systems Conference
y0 0 , y with 1. Hence, we have F ( x; ( y )) = F0 ({ y / y0 } x; 0 ).

( y) =

(7)

We first consider a periodic replacement policy for a given usage rate y and then the policy for various values of usage rate. Let y denote the parameter of a periodic replacement policy for a given usage rate y . The periodic replacement policy is defined as follows. For a given usage rate y , the product is repaired with minimal repair when it fails at age x, x < y , and it is replaced with a new one if its age reaches y .

(8)

The hazard and the cumulative hazard functions associated with F ( x, ( y )) are given by h( x; ( y )) = f ( x; ( y )) /(1 F ( x; ( y )) (9) and
H ( x; ( y )) = h( x; ( y ))dx.
0 x

(10)

We seek the optimal value of y

which minimizes the

where f ( x; ( y )) is the associated density function. For further discussion we define the following notations. Notation: W ,U : Parameters of the two-dimensional warranty Warranty expiry time when the usage rate is y Wy :
X: Y: y0 : y:

long-run average cost to the buyer. If * denote the y optimal value then, as y varies, we have a set of points
( * , y * ) defining a curve as indicated in Fig. 2. Let y y

Product age Usage rate (random variable) Nominal usage rate Parameter of the periodic replacement policy Time to the first failure under usage rate y0 Time to the first failure under usage rate y

denotes a closed region bounded by this curve. Then the replacement policy for a various values of y is defined as follows. The product is repaired with minimal repair when failure occurs in region and it is replaced with a new one if its age reaches y .

T0 : Ty :

0 : Scale parameter ( y) : Scale parameter under usage rate y for AFT : Shape parameter : AFT parameter. F0 (.; 0 ) : Distribution function for T0 0 (.) : Intensity function associated with F0 (.; 0 ) f (.) : Density function of Ty
F (.) :

Distribution function of Ty Intensity function associated with F (.) Conditional intensity function Hazard function Cumulative hazard function Warranty region Maintenance region Cost of minimal repair to rectify a failed product Cost of replacing a failed product by a new one. Cost of downtime for each failure Expected cost per unit time for a given usage rate y Figure 2: Maintenance region

y (.) : (. .) :
h(.) : H (.) : : : Cm : Cr : Cd : J ( y ) :

2.5 Long Run Average Cost


For a given usage rate y , the expected cost per unit time (or the long run average cost) is obtained as follows. Let J ( y ) denotes the expected cost per unit time, which is given by
J ( y ) = E [ Cost per cycle ] E [ Cycle length ] .

(11)

2.4 Periodic Replacement Policy

For this replacement policy, the expected cycle length is y .

Nusa Dua, Bali INDONESIA December 3rd 5th, 2008


234

APIEMS 2008 Proceedings of The 9th Asia Pasific Industrial Engineering & Management Systems Conference
Let Cr and Cm , (Cm > Cr ) denote the cost of each repair and the cost of a replacement, respectively. Since all failures (both in (0, Wy ) and (W y , y ) ) are restored by a minimal repair, then failures occurs according to a nonhomogeneous Poisson process in (0, y ) with intensity function y ( x) = h( x; ( y )). Let Cd denote the cost incurred to the buyer at each minimal repair. Then, the expected cost per cycle is given by
Wy

with lim L y = 0 and lim L y > 0.


y 0

( )

( )

Proof:

' Since y y is IFR, then y y > 0 . Hence, ' y ( y ) + y y ( y ) y ( y ) ' y y ( y ) > 0.

( ) L ( y ) =
y =

( )

Cd

( x )dx + Cr + ( Cd + Cm ) y ( x )dx.
Wy
Wy

(12)

Furthermore, by the definition in (16), it is clear that lim L y = lim L ( 0 ) = 0


y 0

( )

y 0

As a result, the expected cost per unit time is given by


Cd J y =

and
y

( )

y ( x )dx + Cr + ( Cd + Cm ) y

Wy

( x )dx
. (13)

lim L y = lim L ( ) > 0.


y

( )

Theorem 1. Suppose that the product is sold with a twodimensional NFRW and is IFR, y ( y )

3. OPTIMAL POLICY
In this section we obtain an optimal y (i.e. * ) y that

for y [Wy , ) : 1. If L ( )
Cr Cm R Wy

( Cm + Cd )

( ) then

* = . y

minimizes J ( y ) . By differentiating J ( y ) with respect to

y and set it to zero, we have


Wy

2.
Cr Cm

If L Wy <

( )

Cr Cm R Wy

( Cm + Cd )

( ) < L ()
( )

then there

y y ( y )

0 Wy 0

( x ) dx =

( x )dx
. (14)

exist a finite and unique solution * > Wy such y


that
L * = y

( Cm + Cd )

( )

Cr Cm R Wy

( Cm + Cd )

with

the

Define, R W y =

( ) y ( x ) dx, then (14) can be written as


y

corresponding expected cost per unit time Cr + Cm R y R Wy + Cd R y . J * = y

( )

( )

( )

( )

y y ( y ) y ( x ) dx =

Cr Cm R W y

( Cm + Cd )

( ).

Proof:

(15)

If L ( )

Cr Cm R (W y )

( Cm + C d )

then

J ( y ) y

< 0 . Hence,

Next, let L y denotes the left-hand side of (15) i.e.


L y = y y y y ( x ) dx,
0

( )

J ( y ) is a decreasing function of y in the interval Wy , , which means that * = . y

( )

( )

(16)

On the other hand, if L Wy < then from Lemma 1 and unique optimal

( )

Cr Cm R W y

then we have the following lemma and theorem as in Yeh et al. (2007) for the case of one-dimensional warranty.
Lemma 1. If y y

( Cm + C d )

( )<L

() ,

L( y ) is a non-negative and

( )

is IFR, then L y function

( )

is a nonin

increasing function of y . Consequently, there is a finite solution * > Wy , y satisfying

negative

increasing

Nusa Dua, Bali INDONESIA December 3rd 5th, 2008


235

APIEMS 2008 Proceedings of The 9th Asia Pasific Industrial Engineering & Management Systems Conference
J y y

( ) = 0,

and hence L * = y

( )

C r Cm R W y

( Cm + Cd )

( ).

The

corresponding expected cost per unit time is given by C r + C m R y R W y + Cd R y . J * = y

( )

( )

( )

( )

Cr y0 0 . (18) y< ( Cd ( 1) + Cm )Wy To see the proof, we proceed as follows. From Theorem 1, there exists a finite and unique optimal solution satisfying

4. NUMERICAL EXAMPLES
We consider a special case where the distribution function of T0 is given by Weibull distribution function with
F0 ( x; 0 ) = 1 exp( x / 0 ) .

* > Wy whenever L (Wy ) < y


(17) into L Wy < 1 ( y y ) 0 0

Cr Cm R Wy

( )

Cr Cm R Wy

( Cm + Cd )

( ) yields,
,

( Cm + Cd )

( ) . Substituting

<

The conditional failure distribution function, given the usage rate y , is


F ( x; ( y )) = 1 exp( x / ( y )) , with ( y ) given by (7). The hazard function associated with F ( x; ( y )) is given by
y x 1 h( x; ( y )) = . y0 0 As all repairs are minimal, then y ( x ) = h( x; ( y )) and

Wy ( 1) Cd + Cm

Cr

( 1) Cd + Cm y < Cr y0 0 ,
Cr y0 0 . < ( 1) Cd + Cm

(19)

As (19) is equivalent to (18), then the condition for the finite and unique optimal solution * where * > Wy is y y satisfied. In the following numerical example we consider the parameter values satisfying inequality (18) to ensure that the resulting optimal replacement periods are greater then Wy . The parameter values used are: (1) Warranty Policy:
W = 2 (years) and U = 2 ( 104 Km), so U W = 1 ; (2)

its cumulative hazard function is given by Wy . (17) R Wy = y 0 0 y Now we obtain the optimal solution, * . The finite and y

( )

unique optimal solution * y

satisfying * > Wy exists if y

Design Reliability: ( 104 Km per year), 0 = 1 (year) and = 2 ; (3) AFT Model: = 1.5 ; Costs: Cm =1, Cr =10, Cd= 0.25.

Table 1: * for y0 = 2 and 0.4 y 8.5 y


y
0.40 0.60 0.80 1.00 1.02 1.04 1.06 1.08 1.10 1.12

* y
31.57 17.12 11.04 7.80 7.57 7.34 7.13 6.93 6.74 6.56

J * y

( )

y
1.14 1.16 1.18 3.00 4.00 5.00 5.55 7.00 8.50

* y
6.38 6.21 6.05 1.42 0.89 0.62 0.53 0.35 0.24

J * y

( )

0.63 1.16 1.77 2.44 2.51 2.58 2.66 2.73 2.80 2.88

2.95 3.03 3.11 11.98 17.89 24.21 27.46 37.33 46.98

Nusa Dua, Bali INDONESIA December 3rd 5th, 2008


236

APIEMS 2008 Proceedings of The 9th Asia Pasific Industrial Engineering & Management Systems Conference

Figure 3: Maintenance Region for y0 = 2 and 0.4 y 8.5 Table 1 shows the optimal solution * for 0.4 y 8.5 y and its corresponding region is shown in Figure 3. The result in Table 1 indicates that the increase of usage rate decreases the periodic replacement towards the twodimensional warranty limit, Wy . Two-dimensional combination warranty policies. RAIRO Operational Research, 28: 57-75. Iskandar, B.P., and Sandoh, H. (1999) An opportunitybased age replacement policy considering warranty. International Journal of Reliability, Quality and Safety Engineering, 1(3):229-236. Jung, G..M., and Park, D.H. (2003), Optimal maintenance policies during the post-warranty period. Reliability Engineering and System Safety, 82:173-185. Jung, K.M., Han, S.S., and Park, D.H. (2008), Optimization of cost and downtime for replacement model following the expiration of warranty, Reliability Engineering and System Safety, 93:995-1003. Lawless, J. (1982) Statistical Models and Methods for Lifetime Data, Wiley, New York. Lawless J., Hu, J., and Cao, J. (1995), Methods for estimation of failure distributions and rates from automobile warranty data, Lifetime Data Analysis , 1, 227240. Murthy D.N.P., and Wilson, R.J. (1991), Modelling two-dimensional warranties, In: Proceedings of the Fifth International Symposium on Applied Stochastic Models and Data Analysis, Granada, Spain, 481-492. Moskowitz, H., and Chun, Y.H. (1994), A Poisson regression model for two-attribute warranty policies. Naval Research Logistics, 41, 355-375. Nat, J., Iskandar, B.P., Murthy, D.N.P. (2009), A repair-replace strategy based on usage rate for items sold with a two-dimensional warranty. Reliability Engineering and System Safety, 94, 611-617. Pierskalla, W.P., and Voelker, J.A. (1976), A survey of

5. CONCLUDING REMARKS
In this paper, we have studied a periodic replacement policy for a product with a two-dimensional warranty. For the case of two-dimensional warranties, one can study other replacement policies such an age replacement policy or an opportunity-based age replacement (Iskandar and Sandoh, 1999). These topics are currently under investigation.

REFERENCES
Barlow, R.E., and Hunter, L. (1960), Optimal preventive maintenance policies, Operations Research, 8, 90-100. Barlow, R.E., Proschan, F., Hunter, C.H. (1965) Mathematical Theory of Reliability, John Wiley, New York. Blischke, W.R., and Murthy, D.N.P. (2000), Reliability: Modeling,Prediction, and Optimisation, Wiley, New York. Chun, Y.H., and Tang, K. (1999), Cost analysis of twoattribute warranty policies based on the product usage rate. IEEE Transactions in Engineering Management, 46, 201209. Frickenstein, S.G., and Whitaker, L.R. (2003), Age replacement policies in two time scales. Naval Research Logistics, 50, 592-613. Iskandar, B.P., Wilson R.J., and Murthy D.N.P. (1994)

Nusa Dua, Bali INDONESIA December 3rd 5th, 2008


237

APIEMS 2008 Proceedings of The 9th Asia Pasific Industrial Engineering & Management Systems Conference
maintenance models: the control and surveillance of deteriorating systems, Naval Research Logistics, 23, 353388. Sahin, I., and Polatoglu, H. (1996), Maintenance strategies following the expiration of warranty. IEEE Transactions on Reliability, 45(2):220-228. Valdez-Flores, C., and Feldman, R.M. (1989), A survey of preventive maintenance models for stochastically deteriorating single-unit systems. Naval Research Logistics, 36, 419-446. Yeh, R.H., Chen, M.,Y., and Lin, C., Y. (2007), Optimal periodic replacement policy for repairable products under free-repair warranty. European Journal of Operational Research, 176, 1678-1686.

AUTHOR BIOGRAPHIES
Hennie Husniah is a Lecturer in Department of Industrial Engineering, Faculty of Technology, Langlangbuana University, Bandung Indonesia. She is currently a PhD student in the Department of Industrial Engineering, Faculty of Industrial Technology, Bandung Institute of Technology, Indonesia. Her research interests include replacement models and statistical applications in general industrial problems. Her email address is h_husniah@student.itb.ac.id. Bermawi P. Iskandar He obtained BSc (1981) and MSc (1985) degrees in Industrial Engineering from Bandung Institute of Technology, Indonesia, and MEngSc (1989) and PhD (1993) degree from Department of Mechanical Engineering, the University of Queensland. He is currently a Professor at Department of Industrial Engineering, Bandung Institute of Technology. His research interests include Product warranties, Analysis of warranty claim data, Maintenance and reliability models, Facilities layout design, Lot sizing models for a deteriorated production systems. His works have been published in the following journals: Operation Research (USA), Reliability Engineering and Systems Safety, RAIRO-Operations Research (France), International Journal of Reliability, Quality and Safety Engineering, International Journal of Reliability and Application, International Journal of Mathematical and Computer Modelling, Journal of Computers and Operation Research and Journal of TMIITB. He can be reached through his email address bermawi@mail.ti.itb.ac.id.

Nusa Dua, Bali INDONESIA December 3rd 5th, 2008


238

You might also like