European Journal of Operational Research 167 (2005) 381–397

www.elsevier.com/locate/dsw

Production, Manufacturing and Logistics

An interactive method for inventory control with fuzzy

lead-time and dynamic demand
a,*
Jayanta Kumar Dey , Samarjit Kar b, Manoranjan Maiti c

a
Department of Mathematics, Mahishadal Raj College, Mahishadal, East-Midnapore 721628, India
b
Department of Mathematics, Haldia Institute of Technology, Haldia, East-Midnapore 721657, India
c
Department of Applied Mathematics with Oceanology and Computer Programming, Vidyasagar University,
Midnapore 721102, India

Received 30 April 2003; accepted 22 March 2004

Available online 3 August 2004

Abstract

Normally, the real-world inventory control problems are imprecisely defined and human interventions are often
required to solve these decision-making problems. In this paper, a realistic inventory model with imprecise demand,
lead-time and inventory costs have been formulated and an inventory policy is proposed to minimize the cost using
man–machine interaction. Here, demand increases with time at a decreasing rate. The imprecise parameters of
lead-time, inventory costs and demand are expressed through linear/non-linear membership functions. These are
represented by different types of membership functions, linear or quadratic, depending upon the prevailing supply
condition and marketing environment. The imprecise parameters are first transformed into corresponding interval
numbers and then following the interval mathematics, the objective function for average cost is changed into respec-
tive multi-objective functions. These functions are minimized and solved for a Pareto-optimum solution by interactive
fuzzy decision-making procedure. This process leads to man–machine interaction for optimum and appropriate
decision acceptable to the decision makerÕs firm. The model is illustrated numerically and the results are presented
in tabular forms.
Ó 2004 Published by Elsevier B.V.

Keywords: Fuzzy lead-time; Fuzzy inventory cost parameters; Interval number; Interactive fuzzy decision making method

*
Corresponding author. Tel.: +91 9434192 752.
E-mail address: [email protected] (J.K. Dey).

0377-2217/$ - see front matter Ó 2004 Published by Elsevier B.V.

doi:10.1016/j.ejor.2003.07.025
382 J.K. Dey et al. / European Journal of Operational Research 167 (2005) 381–397

1. Introduction

Since the development of EOQ model [1], a lot of research works have been made in inventory control
system. In the existing literature, inventory models are generally developed under the assumption of con-
stant or stochastic lead-time. A number of research papers have already been published in this direction by
Das [2], Magson [3] and Foote et al. [4], etc. Recently, Kalpakam and Sapan [5,6] studied a perishable
inventory model with stochastic lead-time. But in real life situations, the lead-time is normally vague,
imprecise and flexible in nature. So it will be more realistic to consider the lead-time as fuzzy in nature.
Similarly, inventory costs are normally assumed to be constant but in reality it is not so. Normally set up
cost, unit holding and shortage costs are described imprecisely. For example, ‘‘unit holding cost is about
C1’’ or ‘‘unit shortage cost is approximately C2 or more’’, etc.
In fuzzy programming problem [7,8], the constraints and the goals are taken as fuzzy sets. It is also as-
sumed that their membership functions (MFs) are known. But it is not always easy for a decision maker
(DM) to specify these functions precisely. Following Zadeh [9], the fuzzy numbers describe the imprecise
coefficients. These imprecise coefficients may then approximate to crisp set of interval numbers.
Grzegorzewski [10] suggested a method to substitute a fuzzy set by a crisp one. Chanas and Kutchta
[11] defined a transportation problem with fuzzy cost coefficients and developed an algorithm to solve
the problem. Ishibuchi and Tanaka [12] developed a concept for optimization of multi-objective program-
ming problem with interval objective function.
In a fuzzy interactive linear/non-linear multi-objective decision making problem, DM plays an impor-
tant role. He has every right to choose the suitable membership functions to achieve his optimum goal.
In this way, an interaction is established with the Sakawa [13,14] proposed a new technique to solve such
type of problems. But till now, none has applied the interactive technique to a practical inventory problem.
In industry, inventory control is a continuous process and the DM interacts and changes his/her decision
as per the situation/environment. So far, the O.R. scientists have ignored this approach. Again, real world
is full of uncertainties in non-stochastic sense, though a very few inventory models have been developed
under such uncertainty. Till now, none has considered fuzzy lead-time, which is, now a day, a reality in
business and production sectors.
In this paper, for the first time, we have introduced interactive fuzzy approach in general form following
Sakawa [13,14] for inventory control system. An inventory model with fuzzy lead-time, fuzzy inventory
costs and time dependent fuzzy demand has been formulated. The interactive methodology developed here
has been successfully applied to find the optimum solutions of the said fuzzy inventory model.
Here, an inventory model is formulated considering time dependent fuzzy demand that increases at a
decreasing rate and partially backlogged shortages. Inventory cost parameters and lead-time are imprecise,
i.e., fuzzy in nature. The said parameters are expressed by fuzzy numbers, which are then converted into
appropriate interval numbers following Grzegorzewski [10].
Here three different scenarios may merge depending upon the placement of order.

Scenario-I. Next order is placed at the time before beginning the shortage.
Scenario-II. Next order is placed at the time of shortage.
Scenario-III. Next order is placed after the occurrence of shortage.
For each scenario, using the concept of interval arithmetic, we have constructed an equivalent multi-
objective deterministic model corresponding to the original problem with interval co-efficient. This
equivalent problem has been solved using interactive fuzzy decision making procedure. Optimum and
Pareto-optimum solutions are derived for transformed problem using a gradient non-linear optimization
technique––Generalized Reduced Gradient (GRG) method. Finally, some numerical examples are used
to illustrate the models.
 J.K. Dey et al. / European Journal of Operational Research 167 (2005) 381–397 383

2. Formulation of the multi-objective problem

We define a general non-linear objective function with coefficients of the decision variables as interval
valued numbers:
P
k Q
n
r
½aLi ; aRi
xjj
i¼1 j¼1
Minimize ZðxÞ ¼
P
l Q
n
q ð1Þ
½bLi ; bRi
xj j
i¼1 j¼1

subject to xj > 0; j ¼ 1; 2; . . . ; n and x 2 S  Rn ;

where S is the feasible region of x, 0 < aLi < aRi , 0 < bLi < bRi and rj, qj are positive numbers.
Now, we exhibit the formulation of the original problem (1) as a multi-objective non-linear problem.
Since the objective function Z(x) is an interval, it is natural that the solution set of (1) should be defined
by preference relations between intervals.
Now from Eq. (1), following the interval arithmeticÕs (cf. [15,16]) we have
P
k Q
n
r
½aCi  aW i ; aCi þ aW i
xj j
i¼1 j¼1
ZðxÞ ¼ ;
P
l Q
n
q
½bCi  bW i ; bCi þ bW i
xj j
i¼1 j¼1
2 ! !3
P
k Q
n
r Q
n
rj P
k Q
n
r Q
n
rj
6 aCi xj j
 aW i jxj j aC i þ aW i jxj j 7 xj j
6 i¼1 j¼1 j¼1 i¼1 j¼1 j¼1 7
6
¼6 !; !7 ð2Þ
7;
4P l Q qj
n Qn
qj Pl Q qj
n Qn
qj 5
bCi x j þ bW i jxj j bCi x j  bW i jxj j
i¼1 j¼1 j¼1 i¼1 j¼1 j¼1

a L i þ a Ri aR  aLi
where aCi ¼ , aW i ¼ i .
2 2
The right limit ZR(x) of the interval objective function Z(x) may be elicited as
!
P k Qn
rj Qn
rj Pk Qn
r
aCi x j þ aW i jxj j a Ri xj j
i¼1 j¼1 j¼1 i¼1 j¼1
Z R ðxÞ ¼ !¼ : ð3Þ
P l Q qj
n Q qj
n Pl Qn
qj
bC i x j  bW i jxj j bL i xj
i¼1 j¼1 j¼1 i¼1 j¼1

Similarly the left limit ZL(x) may be written as

!
Pk Qn
rj Qn
rj Pk Qn
r
aC i x j  aW i jxj j aLi xj j
i¼1 j¼1 j¼1 i¼1 j¼1
Z L ðxÞ ¼ !¼ : ð4Þ
Pl Qn
qj Qn Pl Qn
qj
qj b x
bC i x j þ bW i jxj j R i j
i¼1 j¼1 j¼1 i¼1 j¼1

The center of the objective function ZC(x) can be written as

1
Z C ðxÞ ¼ ½Z R ðxÞ þ Z L ðxÞ
: ð5Þ
2
384 J.K. Dey et al. / European Journal of Operational Research 167 (2005) 381–397

Thus the problem (1) is transformed into

Minimize fZ L ; Z R ; Z C g ð6Þ
subject to the non-negativity constraints of (1), where ZR, ZL and ZC are defined by Eqs. (3), (4) and (5),
respectively.

3. The nearest interval approximation

Here we want to approximate a fuzzy number by a crisp number. Suppose A e and B

e are two fuzzy num-
bers with a-cuts [AL(a), AR(a)] and [BL(a), BR(a)], respectively. Then the distance between A e and Be is
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Z 1 Z 1
e ¼
~ BÞ
disðA; ðAL ðaÞ  BL ðaÞÞ da þ
2
ðAR ðaÞ  BR ðaÞÞ da:
2

0 0

We have to find a closed interval Cd ( A) e that is nearest to A e with respect to metric d. We can do it since each
interval is also a fuzzy number with constant a-cut for all a 2 [0, 1]. Hence ðC d ð AÞÞ e
a ¼ ½C L ; C R
.
Now we have to minimize
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Z 1 Z 1
2
~ ~
disðA; C d ðAÞÞ ¼ ðAL ðaÞ  C L Þ da þ ðAR ðaÞ  C R Þ2 da ð7Þ
0 0

with respect to CL and CR.

In order to minimize disð A; e C d ð AÞÞ,
e it is sufficient to minimize the function DðC L ; C R Þ
2 e e
ð¼ disc ð A; C d ð AÞÞÞ. The first partial derivatives are
Z 1 Z 1
o o
DðC L ; C R Þ ¼ 2 AL ðaÞ da þ 2C L and DðC L ; C R Þ ¼ 2 AR ðaÞ da þ 2C R :
oC L 0 oC R 0
o
Solving, oC L
DðC L ; C R Þ ¼ 0 and oCoR DðC L ; C R Þ ¼ 0 we get
Z 1 Z 1
C L ¼ AL ðaÞ da and C R ¼ AR ðaÞ da: ð8Þ
0 0
 2
2 2 2
e Cd ( A))
H ðC L ; C R Þ ¼ oCo 2 DðC L ; C R Þ. oCo 2 DðC L ; C R Þ  oCLooCR DðC L ; C R Þ ¼ 4 > 0 so D(CL,CR) i.e. dis( A, e is
L R
global minimum.
Therefore the interval
Z 1 Z 1 
~ ¼
C d ðAÞ AL ðaÞ da; AR ðaÞ da ð9Þ
0 0

is the nearest interval approximation of fuzzy number A e with respect to the metric d. Let A
e ¼ ða1 ; a2 ; a3 Þ be
a fuzzy number. The a-level interval of A e is defined as ð AÞ
e ¼ ½AL ðaÞ; AR ðaÞ
. When A e is a TFN then
a
AL(a) = a1 + a(a2  a1) and AR(a) = a3  a(a3  a2).
By the nearest interval approximation method, the lower limit of the interval is
Z 1 Z 1
1
CL ¼ AL ðaÞ da ¼ ½a1 þ aða2  a1 Þ
da ¼ ða2 þ a1 Þ ð10Þ
0 0 2
and the upper limit of the interval is
Z 1 Z 1
1
CR ¼ AR ðaÞ da ¼ ½a3  aða3  a2 Þ
da ¼ ða2 þ a3 Þ: ð11Þ
0 0 2
 J.K. Dey et al. / European Journal of Operational Research 167 (2005) 381–397 385

Therefore the interval number considering Ae as a TFN is [(a1 + a2)/2, (a2 + a3)/2].
Similarly, when A e is a PFN then AL ðaÞ ¼ a2  ða2  a1 Þp ð1  aÞ and AR ðaÞ ¼ a2 þ ða3  a2 Þp ð1  aÞ.
Following the procedure stated above, the interval number is [(2a1 + a2)/3, (a2 + 2a3)/3].

4. Model formulation

4.1. Assumptions and notations

A multi-objective inventory model with imprecise lead-time, inventory costs and time dependent demand
rate is developed under the following assumptions and notations:

(i) Rate of replenishment is infinite.

(ii) Shortages are allowed but backlogged partially.
(iii) The entire lot is delivered in one batch.
(iv) The inventory-planning horizon is infinite and the inventory system involves only one item and one
stocking point.
(v) There is no quantity discount.
(vi) Q be the order quantity which enters into the inventory at time t = 0.
(vii) Q1 be the on hand inventory at time t = t1, when the new order is placed.
(viii) t1 be the reordering time for next cycle.
(ix) t2 be the time of the inventory cycle when on hand inventory reaches zero.
(x) t3  t2 be the duration of the inventory cycle when the units are out of stock.
(xi) f ðtÞ ¼ D0  aebt ðb2; 0Þ be the demand rate per cycle where D0 and a are imprecise in nature i.e.,
e 0 ¼ ðD01 ; D02 ; D03 Þ and ~
D a ¼ ða1 ; a2 ; a3 Þ.
(xii) The inventory carrying cost, C1 per unit per unit time, the shortage cost, C2 per unit per unit time, the
replenishment (ordering) cost C3 per order are imprecise in nature i.e., C e 1 ¼ ðC 11 ; C 12 ; C 13 Þ,
e e
C 2 ¼ ðC 21 ; C 22 ; C 23 Þ and C 3 ¼ ðC 31 ; C 32 ; C 33 Þ.
(xiii) L denotes the lead-time, a fuzzy number. Here e L ¼ ðL1 ; L2 ; L3 Þ. (The wavy bar ÔÕ indicates fuzzifica-
tion of parameter.)

4.2. Diagrammatic representation

Depending upon the time t = t1, the recorder point when the order for the next cycle is placed, following
three different scenarios may arise.

Scenario-I. The order will be placed at t1(<t2) when on hand inventory becomes Q1(<Q) (c.f. Fig. 1).
Scenario-II. The order will be placed at t1( = t2) when inventory level reaches zero (c.f. Fig. 2).
Scenario-III. The order will be placed at t1(>t2) after the occurrence of shortages. Let, at that time, short-
age level is Q2 (c.f. Fig. 3).

The mathematical formulation for Scenario-II and Scenario-III are same as Scenario-I. So, here we for-
mulate only Scenario I and any one can easily obtain the other two.

4.3. Formulation of Scenario-I

4.3.1. Crisp environment

In this scenario, the inventory level q(t) at time t (0 6 t 6 t3) satisfies the following differential equations:
386 J.K. Dey et al. / European Journal of Operational Research 167 (2005) 381–397

Fig. 1. Pictorial representation of Scenario-I.

Fig. 2. Pictorial representation of Scenario-II.

Fig. 3. Pictorial representation of Scenario-III.


dqðtÞ f ðtÞ; 0 6 t 6 t2 ;
¼ ð12Þ
dt df ðt2 Þ; t2 6 t 6 t3 ;
with the boundary conditions
qðtÞ ¼ Q at t ¼ 0;
ð13Þ
¼0 at t ¼ t2 ;
 J.K. Dey et al. / European Journal of Operational Research 167 (2005) 381–397 387

and
qðtÞ ¼ Q1 at t ¼ t1 : ð14Þ

The solutions of the differential equations (12) with the help of (13) are

Q  D0 t þ ab ð1  ebt Þ; 0 6 t 6 t2 ;
qðtÞ ¼
dðt2  tÞf ðt2 Þ; t2 6 t 6 t3 :

Using the condition (13) and (14), we get from above expression
a
Q ¼ D0 t2  ð1  ebt2 Þ ð15Þ
b
and
a
Q1 ¼ Q  D0 t1 þ ð1  ebt1 Þ: ð16Þ
b
The total inventory carrying cost of the system is given by
Z t2  
D0 t22 a a bt2
CH ¼ C1 qðtÞ dt ¼ C 1 Qt2  þ t2  2 ð1  e Þ : ð17Þ
0 2 b b
The total shortage cost of the system is given by
Z t3 n o
1
C S ¼ C 2 qðtÞ dt ¼ dC 2 f ðt2 Þ ðt1  t2 Þ2  2ðt2  t1 ÞL þ L2 ; ð18Þ
t2 2
where t3 = L + t1.
The total average cost for the present model is given by
C3 þ CH þ CS
F ðt1 ; t2 Þ ¼ : ð19Þ
t3

4.3.2. Fuzzy environment

Considering holding cost, shortage cost, set-up costs, lead-time and demand as imprecisely defined, Eq.
(19) is reduced to
C eH þ C
e3 þ C eS
Fe ðt1 ; t2 Þ ¼ ; ð20Þ
t3
where
( )
e 0 t2 ~
D a ~
a
e e
C H ¼ C 1 Qt2  2 bt2
þ t2  2 ð1  e Þ ;
2 b b
n o
e S ¼ 1 dC
C e 2 f ðt2 Þ ðt1  t2 Þ2  2ðt2  t1 Þ e
Lþe
L
2
and t3 ¼ e
L þ t1 :
2

4.4. Deterministic representation of the proposed model

Following Grzegorzewski [10], the fuzzy numbers are now transformed to interval numbers and the
expression (20) is expressed as
388 J.K. Dey et al. / European Journal of Operational Research 167 (2005) 381–397

Fe ðt1 ; t2 Þ  ½F L ; F R
:

(For calculation of FL and FR, see Appendix A.)

Hence, the proposed model can be stated as

Minimize fF L ðt1 ; t2 Þ; F R ðt1 ; t2 Þg: ð21Þ

Normally, the multi-optimization problem (21), in the case of minimization problem, is formulated in a
conservative sense as

Minimize fF C ðt1 ; t2 Þ; F R ðt1 ; t2 Þg; ð22Þ

where FC = (FL + FR)/2.

Here the interval problem (20) is represented as

Minimize fF L ðt1 ; t2 Þ; F C ðt1 ; t2 Þ; F R ðt1 ; t2 Þg; ð23Þ

where t1L 6 t1 6 t1R and t2L 6 t2 6 t2R, t1L, t1R and t2L, t2R are the lower and upper bounds of t1 and t2,
respectively.
The formulation (23) gives better approximate solutions than those obtained from (22). Moreover, by
the formulation (23), the decision maker does have the freedom to choose any one of the three func-
tions––FL, FC, FR for minimization.

5. Interactive approach

Now considering the imprecise nature of DMÕs judgement, DM may have different fuzzy or imprecise
goals for each of the objective functions and hence interactive approach is used for the man–machine
interaction.
To derive the membership functions lF L , lF R , lF C for the objective functions FL, FR, FC respectively from
DMÕs viewpoint, we first calculate individual minimum (i.e. F min min min
L ; F R ; F C ) and individual maximum (i.e.
max max max
F L , F R , F C ) by a non-linear optimization technique.
With the help of individual minimum and maximum, the DM can select any one from among the fol-
lowing three types of membership functions

(i) Linear membership functions.

(ii) Quadratic membership functions.
(iii) Exponential membership functions.

The membership functions lF L , lF R and lF C for the corresponding objective functions FL, FR and FC
may be written as
8
> 1; if F K 6 F 1K ;
<
lF K ¼ d K ; if F 1K 6 F K 6 F 0K ; ð24Þ
>
:
0; if F K P F 0K ;

where F 1K and F 0K are to be chosen such that F min 1 0 max

K 6FK 6FK 6FK and dK is a strictly monotonic decreas-
ing continuous function of FK which may be linear or non-linear.
 J.K. Dey et al. / European Journal of Operational Research 167 (2005) 381–397 389

5.1. Linear membership function (Type-I)

For each objective function, the corresponding linear membership functions are as follows:
8
>
> 1; if F K 6 F 1K ;
< 1
lF K ¼ 1  F KPF K ; if F 1K 6 F K 6 F 0K ; ð25Þ
>
> K
:
0; if F K P F 0K ;

where F 1K and F 0K are to be chosen such that F min 1 0 max

K 6FK 6FK 6FK and P K ¼ F 0K  F 1K is the tolerance of
kth objective function FK (Fig. 4).

5.2. Quadratic membership function (Type-II)

For each of the objective functions, the corresponding quadratic membership functions are
8
>
> 1; if F K 6 F 1K ;
>
<  2
1
lF K ¼ 1  F KPF K ; if F 1K 6 F K 6 F 0K ; ð26Þ
>
> K
>
:
0; if F K P F 0K ;

where F 1K and F 0K are to be chosen such that F min 1 0 max

K 6FK 6FK 6FK and P K ¼ F 0K  F 1K is the tolerance of k-
th objective function FK (Fig. 5).

5.3. Exponential membership function (Type-III)

For each of the objective functions, the corresponding exponential membership function is
8
>
> 1; if F K 6 F 1K ;
>
> "   #
>
< F K F 1
bK K
; if F 1K 6 F K 6 F 0K ;
PK
lF K ¼ a K 1  e ð27Þ
>
>
>
>
>
:
0; if F K P F 0K :

Fig. 4. Pictorial representation of linear lF K .

390 J.K. Dey et al. / European Journal of Operational Research 167 (2005) 381–397

Fig. 5. Pictorial representation of quadratic lF K .

The constants aK > 1, bK > 0 can be determined by asking the DM to specify the three points F 1K , F 0:5 0
K and F K
min 1 0:5 0 max 0 1
such that F K 6 F K 6 F K 6 F K 6 F K and P K ¼ F K  F K is the tolerance of k-th objective function FK
(Fig. 6).

5.4. Fuzzy decision making method

After determining the different linear/non-linear membership functions (MF) for each of the objective
functions, following Bellman and Zadeh [7] and Zimmermann [8], the given problem (23) can be formulated
as
Max k
subject to k 6 lF L ; k 6 lF C ; k 6 lF R ;
ð28Þ
t1L 6 t1 6 t1R ; t2L 6 t2 6 t2R ;
0 6 k 6 1:
With the help of two different types of membership functions given by (25) and (26), the above problem can
be restated for a particular choice of DM as

Fig. 6. Pictorial representation of exponential lF K .

 J.K. Dey et al. / European Journal of Operational Research 167 (2005) 381–397 391

Max k
F L  F 1L
subject to k 6 1  ; if the MF of first objective 2 Type-I;
PL
F C  F 1C
k61  ; if the MF of second objective 2 Type-I;
PC ð29Þ
 2
F R  F 1R
k61  ; if the MF of third objective 2 Type-II;
PR
t1L 6 t1 6 t1R ; t2L 6 t2 6 t2R ;
0 6 k 6 1:

Here DM selects the above membership functions for the corresponding objective functions. Then the
above problem can be solved by a non-linear optimization technique and optimal solution of k, say k is
obtained.
Now after obtaining k , the DM selects the most important objective function from among the objective
functions FL, FR and FC. Here FR is selected as DM would like to minimize his/her worst case. Then the
problem becomes (for k = k )

Min FR
subject to F L 6 mL ; F C 6 mC ; F R 6 mR ;
ð30Þ
t1L 6 t1 6 t1R ; t2L 6 t2 6 t2R ;
0 6 k 6 1;

where

mL ¼ F 1L þ P L ð1  k Þ; if the MF of first objective 2 Type-I;

mC ¼ F 1C þ P C ð1  k Þ; if the MF of second objective 2 Type-I;

pffiffiffiffiffiffiffiffiffiffiffiffiffi
mR ¼ F 1R þ P R 1  k ; if the MF of third objective 2 Type-II:

5.5. Pareto-optimal solution

Now, after deriving the optimum decision variables, Pareto-optimality test is performed according to
Sakawa [13]. Let the decision vector t1 , t2 and the optimum values, F L ¼ F L ðt1 ; t2 Þ, F R ¼ F R ðt1 ; t2 Þ and
F C ¼ F C ðt1 ; t2 Þ are obtained from (30). With these values, the following problem is solving using a non-lin-
ear optimization technique:

Minimize V ¼ ðeL þ eC þ eR Þ
subject to F L þ eL ¼ F L ; F C þ eC ¼ F C ; F R þ eR ¼ F R ;
eL ; eC ; eR P 0; ð31Þ
t1L 6 t1 6 t1R ; t2L 6 t2 6 t2R ;
0 6 k 6 1:

The optimal solutions of (31), say, t1 , t2 , Q, Q1 , F L , F R and F C are called the strong Pareto-optimal solutions
of the problem (23) provided V is very small; otherwise it is weak Pareto-optimum.
392 J.K. Dey et al. / European Journal of Operational Research 167 (2005) 381–397

6. Numerical example

To illustrate the proposed inventory models, following input data are considered.
Input data: d = 0.79, b = 0.75, t1L = 1.4, t1R = 1.6, t2L = 1.62, t2R = 1.95 in proper units.
e 1 ¼ ð1:4; 2; 2:6Þ;
C e 2 ¼ ð3:4; 4; 4:6Þ; C
C e 3 ¼ ð190; 200; 210Þ; e
L ¼ ð0:75; 0:88; 0:95Þ;
a ¼ ð0:45; 0:51; 0:57Þ; D
~ e 0 ¼ ð180; 190; 200Þ: ð32Þ

6.1. Case-I (When fuzzy parameters are TFN)

Considering the above fuzzy parameters as Triangular fuzzy numbers (TFN), the nearest interval
approximations according to Grzegorzewski [10] are
e 1  ½1:7; 2:3
; C
C e 2  ½3:7; 4:3
; e 3  ½195; 205
;
C e
L  ½0:815; 0:915
; ~a  ½0:48; 0:54
;
e 0  ½185; 195
:
D
Following (28) and (30), the problem (23) is solved and the results are presented in Tables 1 and 2.
Let, with the above values, the membership functions of the objective functions be formed of the types as
per Table 3.
Let, at the beginning, analysis is performed to find optimum k with the membership functions FL, FC as
linear (Type-I) and FR as Quadratic (Type-II). The optimum value of k is presented in Table 4.
With this value of k*, the objective function FR is optimized and the optimum results are:
Now, the results obtained from Table 5 are tested for Pareto-optimality and the Pareto-optimal results
are given in Table 6.

Table 1
Individual minimum and maximum of objective functions
Objective Minimum Maximum
functions Scenario-I Scenario-II Scenario-III Scenario-I Scenario-II Scenario-III
FL 219.8972 250.8405 252.7901 248.0994 261.6306 341.5089
FC 279.5084 306.4799 308.7414 303.7400 320.8668 413.0839
FR 337.5597 362.1193 364.6927 359.3806 380.1031 484.6589

Table 2
Input data for F 1K , F 0K
Scenario F 1L F 0L F 1C F 0C F 1R F 0R
I 219.8972 247.8972 279.5084 303.5084 337.5597 358.5597
II 250.8405 260.8405 306.4799 320.4799 362.1193 379.1193
III 252.7901 340.7901 308.7414 412.7414 364.6927 483.6927

Table 3
Possible types of MF for objective functions
Objective functions Type of membership functions
FL Type-I or Type-II or Type-III
FC Type-I or Type-II or Type-III
FR Type-I or Type-II or Type-III
 J.K. Dey et al. / European Journal of Operational Research 167 (2005) 381–397 393

Table 4
Optimal value of k
Maximum k Scenario-I Scenario-II Scenario-III
k* 0.990 0.995 0.999

Table 5
Optimal results when FR is chosen as the most important objective function
Scenario t1 t2 ½QL ; QR
½F L ; F R
F C
I 1.4000 1.7397 [321.3155, 338.7706] [220.1608, 339.3079] 279.7344
II 1.6200 1.6200 [299.1936, 315.4499] [250.8406, 362.1194] 306.4800
III 1.6200 1.6000 [295.4969, 311.5528] [252.7902, 364.6928] 308.7415

Table 6
Pareto-optimal results
Scenario t1 t2 ½QL ; QR
½F L ; F R
FC V
I 1.40 1.7397 [321.3154,338.7704] [220.1608,339.3079] 279.7344 0.00004
II 1.62 1.6200 [299.1940,315.4503] [250.8406,362.1194] 306.4800 0.00005
III 1.62 1.5999 [295.4967,311.5526] [252.7902,364.6928] 308.7415 0.00005

In Table 6, the values of V are quite small and hence, the optimum results in Table 5 are strong Pareto-
optimum and can be accepted. Still, if the decision-maker/practitioner is not satisfied with the outputs, he/
she may perform the above analysis again re-choosing the membership functions for FL, FC and FR as lin-
ear, quadratic and exponential (say). If the second time analysis does not also give the desired result, the
DM may perform the analysis with the other possible different combinations (in this case, 33 times) of
the membership functions and can select the most suitable optimum solution for his/her firm/factory for
implementation.

6.2. Case-II (When fuzzy parameters are PFN)

Considering the fuzzy parameters in (32) as Parabolic fuzzy numbers (PFN), the nearest interval approx-
imations are
e 1  ½1:6; 2:4
;
C e 2  ½3:6; 4:4
; C
C e 3  ½193:33; 206:67
; e
L  ½0:793; 0:927
;
a  ½0:47; 0:55
; D
~ e 0  ½183:33; 196:67
:

In this case, following the same procedures outlined in case-I, the following results are obtained (Tables
7–9).

Table 7
Individual minimum and maximum of objective functions
Objective Minimum Maximum
functions Scenario-I Scenario-II Scenario-III Scenario-I Scenario-II Scenario-III
FL 200.9961 232.1089 233.9321 229.4524 241.5549 316.5873
FC 280.2110 305.8218 308.0554 303.1702 320.0562 411.4067
FR 356.6925 379.5347 382.1787 376.8881 398.5574 506.2261
394 J.K. Dey et al. / European Journal of Operational Research 167 (2005) 381–397

Table 8
Input data for F 1K , F 0K
Scenario F 1L F 0L F 1C F 0C F 1R F 0R
I 200.9961 228.9961 280.2110 302.2110 356.6925 375.6925
II 232.1089 241.1089 305.8218 319.8218 379.5347 398.5347
III 233.9321 315.9321 308.0554 410.0554 382.1787 505.1787

Table 9
Optimal value of k
Maximum k Scenario-I Scenario-II Scenario-III
k* 0.978 0.998 0.999

Table 10
Optimal results when FR is chosen as the most important objective function
Scenario t1 t2 [QL ; QR ] [ F L ; F R ] F C
I 1.40 1.7374 [ 317.9916, 341.2348] [201.6009, 359.4748] 280.5379
II 1.62 1.6200 [ 296.4843, 318.1593] [232.1089, 379.5348] 305.8218
III 1.62 1.6000 [ 292.8209, 314.2287] [233.9321, 382.1787] 308.0554

Table 11
Pareto-optimal results
Scenario t1 t2 ½QL ; QR
½F L ; F R
FC V
I 1.40 1.7374 [317.9915, 341.2347] [201.6009, 359.4748] 280.5379 0.00004
II 1.62 1.6200 [296.4843, 318.1593] [232.1089, 379.5348] 305.8218 0.00002
III 1.62 1.6000 [292.8209, 314.2287] [233.9321, 382.1787] 308.0554 0.00003

Here again, the values of V are very small and hence the optimum values in Table 10 are strong Pareto-
optimum and can be accepted if DM is satisfied. If DM is not satisfied, he/she may perform the analysis
with different combinations of membership functions (Table 11).

7. Discussion

Here, for comparison, we consider the average value of FL and FR. From Tables 5 and 10, it is clear that
in scenario-I, optimum value of the center of interval cost function (i.e., FC) is lower than those in the other
two scenarios. Therefore from this point of view, DM should prefer scenario-I to the others. Here it is also
observed that by considering linear membership function for the fuzzy numbers (i.e., inventory cost param-
eters, Lead-time, etc.) rather than the non-linear ones, the cost incurred by DM is marginally less.

8. Conclusion

The present paper proposes a solution procedure for inventory model with time dependent fuzzy demand
rate where demand increases at a decreasing rate. Here, shortages are allowed and backlogged partially.
 J.K. Dey et al. / European Journal of Operational Research 167 (2005) 381–397 395

Parameters of the demand function, inventory cost parameters and lead-time are taken as fuzzy numbers.
The fuzzy numbers are described by linear/non-linear type membership functions. Fuzzy numbers are then
approximated to an interval number. After that, the problem has been converted into multi-objective inven-
tory problem where the objective functions represented by left limit; right limit and centers of interval func-
tion are minimized. To obtain the solution of the deterministic multi-objective inventory problem, the
interactive fuzzy solution procedure has been used. The advantage of this procedure is that the decision-
maker can easily minimize his worst case. In the model formulation, different scenarios have been consid-
ered depending upon the time of placing the order for the next lot. The proposed demand has a broad area
of applicability. In real life, the period of any product in the market contains three phases: the growth
phase, mature phase and decline phase. During the growth phase the demand for the product should in-
crease over the time because customers are attracted towards the product more and more as the time passes.
However, this expanding market is not unlimited and approaches maturity gradually. This implies
that the demand for the product in the growth phase should increase at a decreasing rate. This type of de-
mand is applicable to all new types of products like Television, Refrigerators and Washing machines etc.
The formulation of the model and the solution procedure presented here are quite general. Though the
problem has been presented in crisp and fuzzy environment, it can be also formulated in fuzzy-stochastic
environment. In the present model, inventory characteristics like deterioration, fixed time horizon, etc. can
be also incorporated easily. The solution procedure outlined here can also be applied to solve the problems
in other areas like pollution, structural analysis, etc. which involve the fuzzy parameter(s) in model
formulation.

Appendix A

e 0 t2  ~
a ½aL ; aR

Q¼D 1  ebt2  ½D0L ; D0R
t2  1  ebt2
b b
h aR aL i
¼ D0L t2  1  ebt2 ; D0R t2  1  ebt2 ¼ ½QL ; QR
;
b b
e
where D 0 ¼ ðD01 ; D02 ; D03 Þ  ½D0L ; D0R
and ~a ¼ ða1 ; a2 ; a3 Þ  ½aL ; aR

where
aR
QL ¼ D0L t2  1  ebt2
b
and
aL
QR ¼ D0R t2  1  ebt2 ;
b
( )
e 0 t2 ~
D a ~
a
e 1 Qt2 
CH ¼ C 2
þ t2  2 1  e 2 bt
2 b b
 
½D0L ; D0R
2 ½aL ; aR
½aL ; aR
bt2
 ½C 1L ; C 1R
½QL ; QR
t2  t2 þ t2  1  e ¼ ½C HL ; C HR
;
2 b b2
where
 
D0R 2 aL aR
C HL ¼ C 1L QL t2  t2 þ t2  2 1  ebt2
2 b b
and
 
D0L 2 aR aL
C HR ¼ C 1R QR t2  t2 þ t2  2 1  ebt2 ;
2 b b
396 J.K. Dey et al. / European Journal of Operational Research 167 (2005) 381–397

e0  ~
f ðt2 Þ ¼ D aebt2  ½D0L ; D0R
 ½aL ; aR
ebt2 ¼ ½D0L  aR ebt2 ; D0R  aL ebt2
½DL ; DR
;

where
DL ¼ D0L  aR ebt
2

and
DR ¼ D0R  aL ebt
2 ;

1 e n 2
o
CS ¼ dC 2 f ðt 2 Þ ð t 1  t 2 Þ 2
 2 ð t 2  t 1 Þ e
L þ e
L
2
1 n o
2
¼ d½C 2L ; C 2R
 ½DL ; DR
 ðt1  t2 Þ  2ðt2  t1 Þ½L1 ; L2
þ ½L1 ; L2
½L1 ; L2
¼ ½C SL ; C SR
;
2
where
1 n o
C SL ¼ dC 2L  DL  ðt1  t2 Þ2  2ðt2  t1 ÞL2 þ L21
2
and
1 n o
2
C SR ¼ dC 2R  DR  ðt1  t2 Þ  2ðt2  t1 ÞL1 þ L22 :
2
Therefore
e 3 þ C H þ C S ½C 3L ; C 3R
þ ½C HL ; C HR
þ ½C SL ; C SR

C
F ðt1 ; t2 Þ ¼  ¼ ½F L ; F R
;
t3 t1 þ ½L1 ; L2

where
C 3L þ C HL þ C SL
FL ¼
t 1 þ L2
and
C 3R þ C HR þ C SR
FR ¼ :
t 1 þ L1

References

[1] F. Harris, Operations and Cost, Factory Management Service, Chicago, A.W. Shaw Co, 1915.
[2] C. Das, Effect of lead-time on inventory; a static analysis, Operations Research 26 (1975) 273–282.
[3] D. Magson, Stock control when the lead-time cannot be considers constant, Journal of the Operational Research Society 30 (1979)
317–322.
[4] B. Foote, N. Kebriaic, H. Kumin, Heuristic policies for inventory ordering problems with long and random varying lead times,
Journal of Operations and Management 7 (1988) 115–124.
[5] S. Kalpakam, K.P Sapan, (S-1, S) perishable system with stochastic lead-time, Mathematical Computational Modeling 21 (1995)
95–104.
[6] S. Kalpakam, K.P Sapan, Stochastic inventory systems, A perishable inventory model, in: International Conference on Stochastic
Process and Their Applications, Anna University, India, 1996, , pp. 232–245.
[7] R.E. Bellman, L.A. Zadeh, Decision-making in a fuzzy environment, Management Science 17 (1970) B141–B164.
[8] H.Z. Zimmermann, Fuzzy linear programming with several objective functions, Fuzzy Sets and Systems 1 (1978) 46–55.
[9] L.A. Zadeh, Fuzzy sets, Information and Control 8 (1965) 338–353.
[10] P. Grzegorzewski, Nearest interval approximation of a fuzzy number, Fuzzy Sets and Systems 130 (2002) 321–330.
 J.K. Dey et al. / European Journal of Operational Research 167 (2005) 381–397 397

[11] S. Chanas, D. Kutchta, Multi-objective programming in optimization of the interval objective functions––a generalized approach,
European Journal of Operational Research 94 (1996) 594–598.
[12] H. Tanaka, H. Ishibuchi, Multi-objective programming in optimization of the interval objective function, European Journal of
Operational Research 48 (1990) 219–225.
[13] M. Sakawa, Interactive fuzzy decision making for multi-objective linear programming problems and application, in: E. Sanche
(Ed.), Proceedings of IFAC Symposium of Fuzzy Information, Knowledge Representation and Decision Analysis, July 1983, pp.
293–298.
[14] M. Sakawa, H. Yano, An interactive fuzzy decision making for multi-objective non-linear programming using augmented
minimax problems, Fuzzy Sets and Systems 20 (1986) 31–41.
[15] R.E. Moore, Method and Application of Interval Analysis, SIAM, Philadelphia.
[16] M. Inuiguchi, Y. Kume, Goal programming problems with interval coefficients and target intervals, European Journal of
Operational Research 52 (1991) 345–360.