Вы находитесь на странице: 1из 7

international journal of

production
economics
ELSEVIER Int. J. Production Economics 51 (1997) 215-221

An algorithm to determine the EOQ for deteriorating items


with shortage and a linear trend in demand
Kun-Jen Chung*, Sui-Fu Tsai
Department of Industrial Management, National Taiwan University of Science and Technology. 43 See 4, Keelung Road, Taipei, Taiwan, ROC

Received 27 March 1996; accepted 25 February 1997

Abstract
In 1991, professors Goswami and Chaudhuri considered the inventory replenishment policy over a fixed planning
period for a deteriorating item having a deterministic demand pattern with a linear trend and shortages. Basically, they
concentrated on the establishment of the inventory model but did not present a concrete algorithm to solve it. This paper
complements the shortcoming of Goswami and Chaudhuris paper. We first show that the total cost function of Goswami
and Chaudhuri is convex. Subsequently, with the convexity, an algorithm to determine the number of reorders, the
interval between two successive reorders and the shortage intervals over a finite time-horizon in an optimal manner is
developed. Many numerical examples illustrate how the algorithm works.

Keywords: EOQ models; Deterioration; Shortages; Linear trend in demand

1. Introduction consider the inventory policy over a fixed planning


period for a deteriorating item having a determinis-
The classical EOQ model is probably the oldest tic demand pattern with a linear trend and shor-
inventory control model. Osteryoung et al. (1986) tags. Hariga and Benkherouf (1994) present
report that these models are still widely used in optimal and heuristic procedures for the inventory
industry, although the assumptions (constant de- replenishment problem in which items are deterio-
mand rate, known inventory holding and setup rating at a constant rate and the demand rates are
cost, no shortages and deterioration allowed, unit changing exponentially with time over a known
production cost independent of the lot size, con- and finite planning horizon. Chung and Ting (1994)
stant production rate, infinite horizon) necessary to propose a heuristic model for determining the in-
justify their use are rarely met. Numerous research ventory replenishment policy for deteriorating
efforts have been undertaken to extend these items with time-proportional demand. Wee (1995)
models to conform more closely to real situations. discusses a replenishment policy of deteriorating
For example, Goswami and Chaudhuri (1991) product where demand declines exponentially over
a fixed time horizon. Benkherouf (1995) presents an
optimal procedure for finding the replenishment
*Tel.: 02-7376324, Fax: 02-7376344 schedule for an inventory system with shortages, in

0925-5273/97/$17.00 Copyright 0 1997 Elsevier Science B.V. All rights reserved


PlI SO925-5273(97)00082-O
216 K.-J. Chung, S.-F. Tsaillnt. J. Production Economics 51 (1997) 215-221

which items deteriorate at a constant rate and de- which there is no-shortage in each interval
mand rates are decreasing over a known and finite [jT, (j + 1) T ] is a fraction of the scheduling
planning horizon. Kim (1995) derives an inventory period T and is equal to KT (0 < K < 1). Short-
model for determining the ordering schedule in ages occur at times (K + j) T (j = 0, 1,2, . . . , n - 2)
which the demand rates are changing linearly with where jT<(K+j)T<(j+l)T, j=O,1,2 ,...,
time and the decay is assumed to be a constant n - 2. Last replenishment occurs at time (n - l)T
rate of the on-hand inventory. Benkherouf and and shortages are not allowed in the last period
Mahmoud (1996) consider an in inventory system [(n - l)T, H]. Let C(n, K) denote the total cost
with increasing demand rates and suggest a proced- over the time horizon [0, H] when there are n re-
ure for determining the optimal replenishment order times over the time horizon H. Based on the
schedule. above notations and assumptions, Goswami and
It is well known that the total cost function based Chaudhuri (1991) show that
on average cost analysis is convex. Furthermore, n-1 n-l
Rachamadugu (1988) shows that the discounted C(n, K) = nA + r C Rj + p 1 Dj + rR,
total cost function is also convex. However, the j=l j=l

property of the total cost function of Goswami and n-l

Chaudhuri (1991) has remained unexplored. In the +~D,+s C Sj (1)


j=l
present paper, we shall prove that under some
circumstances, the total cost function of Goswami where
and Chaudhuris (1991) inventory model is convex.
Using that property, an algorithm to determine the Rj = s [(KeT _ l)e(K+j- i)oT + e(j- l)@T]
optimal interval between two successive reorders
and the optimal shortage intervals can be de-
veloped so as to keep the average system cost to +i[{(K +Kj-K)B2T2
a minimum.
- (2K + j - 1)BT + 2}eK+jPeT

2. The model + {(j - 1)8T - 2}ej- jeTI,

j = 1,2, 3, . . . , n - 1, (2)
A deterministic inventory model has been de-
veloped with the following notations and assump- R, = $[{d(H - nT + T) - l}esH + e(n-l)eT]
tions:
(i) f(t) = a + bt is the demand rate at any time
t where a > 0 and b > 0. + i [{(Hz - nHT + HT)B2
(ii) A is the fixed ordering cost per order.
(iii) r is the inventory holding cost per unit per unit - (2H - nT + T)B + 2)e
time.
+ {(n - 1)BT - 2}e(-)eT], (3)
(iv) p is the unit cost of the inventory item.
(v) s is the shortage cost per unit per unit time.
Dj = i [(eKeT _ l)e(j- i)eT _ KeT]
(vi) H is the fixed time horizon.
(vii) A constant fraction 8 of the on-hand inventory
deteriorates per unit of time.
+ $ [2{(K + j - l)BTeKeT
The total time horizon H has been divided into
n equal parts, each of length T, so that T = H/n. _ (j _ l)eT _ eKeT + l}e(j-l)JT
The reorder times over the time horizon H will be
- t12(K2 + 2Kj - 2K) T 2],
jT (j = 0, 1,2, . . . , n - 1). Initial and final invento-
ries are both zero. We assume that the period for j = 1, 2, 3, . . . , n - 1, (4)
K.-J. Chung, S.-F. Tsaillnt. J. Production Economics 51 (1997) 215-221 217

+ [a + bT(K + j - l)] t3Te(K+j-1)eT}


D, = i [eeH- etneljeT - fl(H - nT + T)]

+sT(n-1) a+2bKT
+ $[2(HH
- l)eeH i

- 2{(n - 1)8T - l}e(-)eT +y(n-4) (8)


1
- 8 {Hz - (n - 1) TZ }]. (5) When n = 1 or n 2 4, d2C(n,K)/dK2 2 0. There-
and fore C(n, K) is convex with respect to K E [0, 11.
This completes the proof. El
Sj=~(K-l)2T2+g(K-l)3T3
In general, C(2, K) and C(3, K) are not necessar-
ily convex with respect to K E [0, l] as the following
+ ;j(K - 1)2 T 3, j = 1,2, 3, . . . , n - 1. (6)
examples show.

Our problem is to determine the values of n and


Example 1. Let n = 2, a = 10, b = 100, T = 1,
K which minimize C(n, K). We first explore the
0 = 0.01, r = 2, p = 1, s = 9 and H = 2. Then we
property of C(n, K).
have d2C(2,0.1)/dK2 = - 567.79 < 0. Hence
C(2,K) is not convex on [0,11.

3. Analysis
Example 2. Let n = 3, a = 10, b = 100, T = 1,
8 = 0.01, r = 2, p = 1, s = 9 and H = 3. Then we
Theorem 1. When n = 1 or n 3 4, C(n, K) is convex
have d2C(3,0.1)/dK2 = - 19.62 < 0. Hence
with respect to K E [0, 11.
C(3,K) is not convex on [IO, 11.
Proof. From
However, if the following conditions hold,
n-1
Wn, W ,.T2 1 {aKeK+j-eT C(2, K) and C(3, K) are convex on [0, 11.
dK = j=l
Theorem 2. If 2a 2 bH, then C(2, K) and C(3,K)
+ bT (K2 + Kj - K)ecK+j- ljeT}
are convex on [0, 11.
n-1
+ pT c ([a + bT(K + j - l)]
j=l Proof.
X(eW+j-l)~T _ I)}. Case 1: If n = 2, then T = H/2. So a + 2bKT +
(bT/2)(n - 4) = bKH + (a - (bH/2)) > 0. Hence
+sT (n - l)[a(K - 1) + bT (K - 1)2 d2C(2, K)/dK2 > 0.
Case2. Ifn=3,thenT=H/3.Soa+2bKT+
+;T(K - l)]. (7) (bT/2) (n - 4) = (2bKH/3) + (a - (bH/6)) > 0. Hence
d2C(3, K)/dK2 > 0.
We have Combining cases 1 and 2, we have completed the
d2C(n,K) n-l proof. 0
dK2 {ue(K+j-l)eT(l + K0T)

+ bTecK+j-jeT[(2K + j - 1) 4. The algorithm


-t- (K2 + Kj - K)t3T]}
Suppose that C(n, K) is convex. In general, the
n-1

+ pT c {bT(e(K+j-l)eT - 1) Newton-Raphason method is frequently used to


j=l locate the optimal point of a convex function whose
218 K.-J. Chung, S.-F. Tsai,iInt. J. Production Economics 51 (1997) 215-221

domain is the whole real line. Although C(n, K) is with respect to K E [0,11. Let K,* be the optimal
convex, the Newton-Raphason method may not point of C(n,K) when n is fixed. The algorithm is
be appropriate for C(n,K). The main reason is described as follows.
that the domain of C(n,K) is the bounded inter-
val [0, l] but not the whole real line. Therefore, The algorithm
some point of the iterative sequence generated
Step 1. Let n be fixed and E > 0.
by the Newton- Raphason method may go out
Step 2. If g(n,O) > 0, set K,,, = 0 and go to
of the interval [0, l] so that the Newton- Raphason
Step 6.
method cannot continue as Example 4 shows. Con-
If g(n,O) < 0 set K, = 0 and KU = 1.
sequently, it is questionable that for a given positive
Step 3. Set Kept = (KL + K,)/2.
integer n, Goswami and Chaudhuri (1991) declare
Step 4. If jg(n,K,,,)I < E, go to Step 6. Other-
that Eq. (11) in Goswami and Chaudhuri (1991) can
wise, go to Step 5.
be solved by any iterative method when the values
Step 5. If g(n,K,,,) > 0, set KU = Kept.
of other parameters are prescribed. Goswami and
If g(n, Kept) < 0, set KL = Kept.
Chaudhuri (1991) mainly concentrate on the
Then, go to Step 3.
modeling of the problem; but they do not present
Step 6. KX = Kept and exit the optimal solution.
a solid algorithm to solve it. This paper
complements the shortcoming of Goswami and The above algorithm is to locate the optimal
Chaudhuri (1991) and develops a simple algorithm solution K,* of C(n,K) when n is fixed. The cri-
to solve it. terion to select the optimal integer n* such that
For a given positive integer n, let dC(n, K)/dK = (n*,Kz*) is the optimal solution of the total cost
g(n,K). Suppose that C(n, K) is convex. Then function C(n, K) of Goswami and Chaudhuri (1991)
s(n,K) is nondecreasing with respect to is described as follows:
K (0 d K < 1). There are two potential cases to Let KT be the optimal solution of C(i,K) when
occur. i is fixed and let n* be the smallest integer such that
Case 1. If g(n,O) 3 0, then g(n,K) > 0 for all
C(n*, K,*.) d C(n* + 1, K,*. + 1)
K (0 < K d 1). Hence C(n, K) will have the min-
imum value at K = 0. Consequently, C(n, 0) is the Then we take (n*,K$) as the optimal solution of
minimum value of C(n, K) when n is fixed. the total cost function C(n,K).
Case 2. Suppose that g(n,O) < 0. Since g(n, 1) 3
0, then there is a point K E (0, l] such that g(n, K)
= 0. Hence C(n,K) has the minimum value at 5. Numerical examples
K = K when n is fixed.
Based on the above arguments, we are in a Example 3. Let A = 90, r = 5, a = 20, b = 2,
position to outline the algorithm to determine p = 0.5, 8 = 0.01, s = 1.5 and H = 11 in appropri-
the optimal point of C(n,K). Before describing ate units. From 2a > bH, we see that C(n, K) is
the algorithm, we need the following theorem. The convex with respect to K (0 d K d 1) for n > 1.
proof of the following theorem can be found in The algorithm described in Section 4 is used to
any calculus book such as Thomas and Finney determine the optimal solution of C(n, K) for n 2 1.
(1992). Table 1 compares our solutions with Goswami and
Chaudhuris solutions. We see that our solutions
Intermediate Value Theorem. Let m be a continuous are all better than those of Goswami and
function on the closed interval [u,v] and let Chaudhuri (199 1).
m(u)m(v) < 0. Then there exists a number d E (u, v) Table 1 shows that total cost C becomes min-
such that m(d) = 0. imum for n = 8 and K = 0.219283. Hence the opti-
mal values of n, K and T become respectively
The following algorithm is based on the above n* = 8, K* = 0.219283 and T* = 1.375. The min-
theorem and the assumption that C(n, K) is convex imum cost becomes 1167.3 18409.
K.-J. Chung, S-F. Tsaillnt. J. Production Economics 51 (1997) 215-221 219

Table 1
Comparisons between two approaches

n T s(n, 0) s(n, 1) Goswami and Chaudhuris Our solutions Comparisons


solutions

K C(n, K) K C(n,K) (W3)

(1) (2) (3) (4) (5) (6) (7) (8) (9)

1 11.0OOOOO 1.OOOOOO 11432.721230 1.000000 11432.721230 1.000000


2 5.500000 - 907.500000 4958.674894 0.240109 3798.028263 0.228372 3797.715900 1.000082
3 3.666667 - 952.684047 4420.423 118 0.240332 2160.697992 0.224979 2160.159069 1.000249
4 2.750000 - 864.409232 3736.117986 0.242140 1573.340631 0.223 111 1572.599866 1.000471
5 2.200000 - 767.950296 3192.12165 0.244322 1320.244887 0.221830 1319.331332 1.000692
6 1.833333 - 683.665983 2773.739865 0.246581 1208.777721 0.220838 1207.715710 1.000879
7 1.571429 - 613.093217 2447.564971 0.248826 1168.509978 0.220010 1167.318409 1.001021
Sb 1.375000 - 554.269012 2187.971775 0.25 1002 1168.376657 0.219283 1167.072152 1.001118
9 1.222222 - 504.925536 1977.192758 0.253001 1192.798720 0.218624 1191.402706 1.001172
10 l.lOOOOO - 463.140501 1802.976121 0.254759 1233.202797 0.218012 1231.738943 1.001188
11 1.OOOOOO - 427.400203 1656.735364 0.217433 1283.015692
12 0.9 16667 - 396.535188 1532.323429 0.216881 1342.066672
13 0.846154 - 369.642461 1425.244373 0.216348 1406.820521
14 0.785714 - 346.020151 1332.143293 0.215830 1475.870277
15 0.733333 - 325.117573 1250.470587 0.215325 1548.229616
16 0.687500 - 306.498197 1178.256448 0.214829 1623.188246
17 0.647059 - 289.812393 1113.956270 0.214342 1700.222659
18 0.611111 - 274.777334 1056.342573 0.213862 1778.939144

The errors of Eqs. (8) and (11) in Goswami and Chaudhuri (1991) are corrected as Eqs. (3) and (7) in this paper. The total costs of C(n, K)
in Table 1 of Goswami and Chaudhuri (1991) are recomputed as Table 1 of this paper shows.

Example 4. Let A = 90, r = 2, a = 7.5, b = 2, Example 5. Let A = 90, r = 5, a = 15, b = 2, p = 1,


p = 0.2,0 = 0.005, s = 10 and H = 6 in appropriate 8 = 0.09, s = 0.5 and H = 8 in appropriate units.
units. From 2a > bH, we see that C(n, K) is convex From 2a > bH, we see that C(n,K) is convex
with respect to K (0 < K < 1) for n B 1. For n = 2 with repect to K (0 d K 6 1) for n > 1. The algo-
if the Newton-Raphason method is used to deter- rithm described in Section 4 is used to obtain the
mine the optimal solution of C(2,K) and if the optimal solutions of C(n,K) for n > 1. We obtain
strarting point K1 is 0.2, we find that Kz is equal to Table 3.
1.32958781 which goes out of the interval [O,l]. Table 3 shows that the total cost C be-
Hence the Newton-Raphason method cannot con- comes minimum for n = 6 and K = 0.034627.
tinue. So the Newton-Raphason method is not Hence the optimal values of n, K and T be-
appropriate for solving C(2, K) although C(2, K) is come respectively n* = 6, K* = 0.034627 and
convex. Consequently the algorithm described in T * = 1.333333. The minimum cost becomes
Section 4 is used to obtain the optimal solutions of 887.877415.
C(n,K) for n 2 1. We obtain Table 2.
Table 2 shows that the total cost C becomes
minimum for n = 3 and K = 0.831247. Hence the 6. Sensitivity analysis
optimal values of n, K and T become respectively
PI* = 3, K* = 0.831247 and T * = 2. The minimum To study the effects of change in the sys-
cost becomes 425.279 150. tem parameter 8 on the optimal cost, the
220 K.-J. Chung, S.-F. Tsai/Int. J. Production Economics 51 (1997) 215-221

Table 2
Optimal solution to Example 4

n T 9 (n, 0) 9 (n, 1) K C(n, K)

6.000000 1.oOOoOO 660.441278


2 3.000000 - 675.00000 246.794891 0.831524 426.134743
3 2.000000 - 759.95317 219.601061 0.831247 425.279150
4 1.500000 - 708.66508 185.410946 0.831086 473.157408
5 1.200000 - 639.24599 158.298135 0.830971 538.952740
6 1.000000 - 574.86393 137.471387 0.830879 613.263874
1 0.857143 - 519.38034 121.246794 0.830799 692.217526
8 0.750000 - 472.33320 108.340726 0.830727 174.158363
9 0.666667 - 432.41473 97.865453 0.83066 1 857.915051
10 0.600000 - 398.33300 89.209733 0.830598 942.965567
I1 0.545455 - 369.00140 8 1.945602 0.830537 1028.945883
12 0.500000 - 343.54877 75.766897 0.830479 1115.616718
13 0.461538 - 321.28607 70.449809 0.830423 1202.814348

Table 3
Optimal solution to Example 5

n T 9 (n, 0) 9 (n, 1) K Cb,K)

1 8.000000 1.000000 7075.683370


2 4.000000 - 120.000000 2677.192429 0.086950 2409.110920
3 2.666667 - 110.921898 2450.758183 0.068166 1387.432335
4 2.000000 - 86.572587 2107.009618 0.055252 1045.132151
5 1.600000 - 64.170335 1824.263658 0.044443 920.730699
6 1.333333 - 45.641042 1603.257040 0.034627 887.877415
7 1.142857 - 30.533700 1429.271491 0.025358 900.427416
8 1.000000 - 18.136087 1289.887717 0.016422 938.011481
9 0.888889 - 7.841958 1176.173412 0.007705 990.746456
10 0.800000 0.813208 1081.843557 0 1053.135927
11 0.721273 8.177270 994.835653 0 1122.041062
12 0.666661 14.511046 934.724160 0 1195.519609
13 0.615385 20.011973 876.336590 0 1272.298261
14 0.571429 24.831370 825.489309 0 1351.523538
15 0.533333 29.086725 780.822997 0 1432.604929
16 0.500000 32.870424 741.283216 0 1515.122627

replenishment times, the optimal cycle inter- can be obtained from the sensitivity analysis based
vals and K* derived by the proposed method, on Tables 4-6.
a sensitivity analysis is performed considering (i) The optimal cost is increasing as 8 increases.
numerical examples. The sensitivity analysis is (ii) The replenishment times are increasing as 0
performed by changing (increasing or decreas- increases.
ing) the parameter by lo%, 25S, 50% taking (iii) The optimal cycle intervals are decreasing as
at a time, keeping the remaining parameters at 8 increases.
their original values. The following inferences (iv) K* is decreasing as 8 increases.
K.-J. Chug, S.-F. TsaijInt. J. Production Economics 51 (1997) 215-221 221

Table 4
Sensitivity analysis of Example 3

Change in parameter 6

- 50% - 25% - 10% 0% 10% 25% 50%

C 1149.238312 1158.175558 1163.636245 1167.072151 1170.071239 1174.620132 1182.338074


n 7 7 7 8 8 8 8
T* 1.571429 1.571429 1.571429 1.375000 1.375000 1.375000 1.375000
K* 0.225364 0.222681 0.221077 0.219283 0.218146 0.216446 0.1213622

Table 5
Sensitivity analysis of Example 4

Change in parameter 6

- 50% - 25% - 10% 0% 10% 25% 50%

C 423.678513 424.527427 424.977976 425.279150 425.580971 426.034921 426.794763


n 2 3 3 3 3 3 3
T* 3 2 2 2 2 2 2
K* 0.832430 0.831770 0.831456 0.831247 0.831038 0.830723 0.830198

Table 6
Sensitivity analysis of Example 5

Change in parameter 6

- 50% - 25% - 10% 0% 10% 25% 50%

C 786.876092 834.171690 865.328366 887.877424 911.967318 949.590859 1007.688698


n 5 6 6 6 6 7 7
T* 1.600000 1.333333 1.333333 1.333333 1.333333 1.142857 1.142857
K* 0.066660 0.047692 0.039775 0.034627 0.029583 0.011015 0.000000

References with exponential time-varying demand. Eur. J. Oper. Res.


79(l), 123-137.
Benkherouf, L., 1995. On an inventory model with deteriorating Kim, D.H., 1995. A heuristic for replenishment of deteriorating
items and decreasing time-varying demand and shortages. items with a linear trend in demand. Int. J. Prod. Econom.
Eur. J. Oper. Res. 86(2) 293-299. 39(3), 265-270.
Benkherouf, L., Mahmoud, M.G., 1996. On an inventory mode1 Osteryoung, J.S., McCarty, D.E., Reinhart, W.J., 1986. Use of
for deteriorating items with increasing time varying demand EOQ mode1 for inventory analysis. Prod. Inv. Mgmt. 27(3),
and shortages. J. Oper. Res. Sot. 47(l), 188-200. 3946.
Chung, K.J., Ting, P.S., 1994. On replenishment schedule for Rachamadugu, R., 1988. Error bounds for EOQ. Naval Res.
deteriorating items with time-proportional demand. Prod. Logist. Quart. 35, 419425.
Planning Control 5(4), 392-396. Thomas, G.B., Finney, R.L., 1992, Calculus and Analytic
Goswami, A., Chaudhuri, K.S., 1991. An EOQ mode1 for de- Geometry. Addison-Wesley, Reading, MA.
teriorating items with shortages and a linear trend in Wee, H.M., 1995. A deterministic lot-size inventory model for
demand. J. Oper. Res. Sot. 42(12), 1105-l 110. deteriorating items with shortages and a declining market.
Hariga, M.A., Benkherouf, L., 1994. Optimal and heuristic in- Comput. Oper. Res. 22(3), 3455356.
ventory replenishment models for deteriorating items

Вам также может понравиться