Computers & Industrial Engineering 50 (2006) 286295

www.elsevier.com/locate/dsw

Three-machine owshop with two operations per job

to minimize makespan
Ling-Huey Su *, Cheng-Te Lin
Department of Industrial Engineering, Chung-Yuan Christian University, Chung-Li, Taiwan, ROC

Received 14 February 2006; received in revised form 8 June 2006; accepted 8 June 2006
Available online 18 July 2006

Abstract

This work studies three variants of a three-machine owshop problem with two operations per job to minimize make-
span (F3/o = 2/Cmax). A set of n jobs are classied into three mutually exclusive families A, B and C. The families A, B and
C are dened as the set of jobs that is scheduled in machine sequence (M1, M2), (M1, M3) and (M1, M3), respectively, where
(Mx, My) species the machine sequence for the job that is processed rst on Mx, and then on My. Specically, jobs with the
same route (machine sequence) are classied into the same family. Three variants of F3/o = 2/Cmax are studied. First, F3/
GT, no-idle, o = 2/Cmax, in which both machine no-idle and GT restrictions are considered. The GT assumption requires
that all jobs in the same family are processed contiguously on the machine and the machine no-idle assumption requires
that all machines work continuously without idle time. Second, the problem F3/GT, o = 2/Cmax, in which the machine
no-idle restriction in the rst variant is relaxed, is considered. Third, the problem F3/no-idle, o = 2/Cmax with the GT
assumption in the rst variant relaxed is considered. Based on the dominance conditions developed, the optimal solution
is polynomially derived for each variant. These results may narrow down the gap between easy and hard cases of the
general problem.
2006 Elsevier Ltd. All rights reserved.

Keywords: Scheduling; No-idle; Group assumption; Three-machine ow shop; Two operations per job; Makespan

1. Introduction

This work discusses three variants of a three-machine owshop problem with two operations per job to
minimize makespan (F3/o = 2/Cmax). Let Mj, j = 1, 2 and 3, be machine 1, 2 and 3, respectively. The set of
n jobs is classied into three mutually exclusive families A, B and C. Jobs that are scheduled in the same route
(machine sequence) are classied as belonging to the same family. The families A, B and C are dened as the
set of jobs that is scheduled with machine sequence (M1, M2), (M1, M3) and (M2, M3), respectively, where
(Mx, My) species the machine sequence of a job that is processed rst on Mx and then on My. Three variants,
according to the imposition of GT or the no-idle restriction, are proposed. The GT restriction requires that all

*
Corresponding author.
E-mail address: linghuey@cycu.edu.tw (L.-H. Su).

0360-8352/$ - see front matter 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.cie.2006.06.001
 L.-H. Su, C.-T. Lin / Computers & Industrial Engineering 50 (2006) 286295 287

jobs in the same family are contiguously processed. Such a requirement may reect the fact that the setup
times are much longer than the job processing times, or the minimization of the time spent on setup in situa-
tions where capacity is scarce, or the need simply to make the problem more tractable. The no-idle restriction
means that each machine must process jobs without any interruption from the start of processing the rst job
to the completion of the last job. The three cases are as follows. First, the F3/GT, no-idle, o = 2/Cmax problem
in which both machine no-idle and the GT restriction are imposed. Then, the F3/GT, o = 2/Cmax problem in
which machine no-idle restriction in the rst case is relaxed, is considered. Finally, the F3/no-idle, o = 2/Cmax
problem with the GT restriction of the rst variant is relaxed is considered. The rst known polynomial time
schemes for these three cases are presented.
Various applications of the proposed owshop model are encountered in the testing process in semiconduc-
tor manufacturing, in which an IC is processed through burn-in, testing and packing (including baking). How-
ever, some of these operations may be omitted at the request of the customer for all reasons. For instance, a
DRAM IC needs burn-in, testing and packing, while a FLASH IC needs only testing and packing. Besides, the
packing is not always necessary if the IC is to be used in the aliated company, or the quality of the com-
pound in the surface of an IC is rather high such that baking is unnecessary.
The minimization of makespan for scheduling n jobs on m machines in a ow shop has received consider-
able attention. When m equals 2, the application of Johnsons rule (1954) yields the minimum makespan.
Unfortunately, Johnsons rule cannot be extended to three machines unless that the second machine is not
the bottleneck. Garey, Johnson, and Sethi (1976) proved the NP-hardness of the F3//Cmax problem. Gonz-
alez and Sahni (1978) showed that F3/o 6 2/Cmax was ordinary NP-hard.
Adiri andP Pohoryles (1982) were P the rst to discuss the idea of the no-idle schedule. They studied
F =no-idle= C i and F =no-wait= C i problems, and showed that the F2//Cmax problem was equivalent
i i
to the F2/no-idle/Cmax problem. Riane, Artiba, and Elmaghraby (1998) studied the problem of scheduling
n jobs on a three stages hybrid owshop of a particular structure, one machine in each the rst stage and
the third stage, and two dedicated machines in the second stage to minimize the makespan. Two heuris-
tics procedures have been proposed to solve the problem. Baptiste and Hguny (1997) developed a branch
and bound algorithm to solve the Fm/no-idle/Cmax problem. Saadani, Guinet, and Moalla (2003) then
proposed an O(n log n) heuristic for solving F3/no-idle/Cmax problem. Kamburowski (2004) identied a
simple network representation for the same problem and revealed a certain anomaly that results from
the no-idle condition, and lead to some dominance relations among the machines under which the prob-
lem became eciently solvable. They extended their work to m-machine no-idle owshops were also
included. Saadani, Guinet, and Moalla (2005) recently proposed the well-known nearest insertion rule
to solve the F/no-idle/Cmax problem based on the idea that this problem could be modeled as a traveling
salesman problem.
Liaee and Emmons (1997) considered the processing of families of jobs on single or parallel facilities, a
setup time was incurred whenever a switch was made from the processing of a job in one family to that of
a job in another family. They also considered cases based on the group technology assumption. Yang and
Chern (2000) considered a two-machine owshop group scheduling problem. The jobs were classied into
groups; jobs in the same group must be processed consecutively. Each group involves a setup time and a
removal time for both machines. A transportation time was required to move the jobs from M1 to M2. A
polynomial time algorithm has been proposed to minimize the makespan. Lin and Cheng (2001) consid-
ered a batch scheduling in the no-wait two machine ow shop to minimize the makespan. They showed
that the problem is strongly NP-hard and designed methods for solving the two restricted cases F2/no-
wait, batch, pi = qi = p/Cmax and F2/no-wait, batch, pi = qi, qi = q/Cmax, where pi and qi are the processing
time of job i on M1 and M2, respectively. Andres, Albarracin, Tormo, and Vicens (2005) considered the
problem of a three-stage hybrid ow shop problem with sequence dependent and separable setup time.
They applied the model to the tile industry to identify a set of families that are integrated by products
that have a common feature.
These particular classes of F3/o = 2/Cmax problems were (i) to establish methods for solving simple problems
in order to generalize them to solve complex problems. (ii) to identify special cases which were tractable in
polynomial time. These results may narrow down the gap between easy and hard cases of the general problem.
288 L.-H. Su, C.-T. Lin / Computers & Industrial Engineering 50 (2006) 286295

2. The F3/GT, no-idle, o = 2/Cmax model

This section discusses the F3/GT, no-idle, o = 2/Cmax model is considered. This work propose an optimal
schedule in polynomial time for this model, based on the dominance conditions derived by analyzing the
problem characteristics. According to Adiri and Pohoryles (1982), the resolution to the F2/no-idle/Cmax
problem is identical to that to the F2//Cmax problem, as determined by ordering jobs by Johnsons rule
and then shifting the jobs on the second machine to the right. Fig. 1 shows that the right shifting on the
second machine does not aect Cmax. Denote Pn the idle time immediately before the rst job on the second
machine as a slot, we have Slot C max  i1 P i;2 ; where Cmax is the makespan due to Johnsons algorithm
and Pi,2 denotes the processing time of job i on M2.
For clarity and convenience, the following notation is used throughout the article. The term (ki), k = A, B
and C; i = 1 and 2, is the total processing time of jobs in family k on their ith machine. Specically, as an
example, C1 and C2 denote the total processing time of jobs in family C on M2 and M3, respectively. Let
Si and Cmax,i, i = A, B and C, be the slot and makespan of family i, respectively, when Johnsons rule is applied
to it.
S A C max;A  A2
S B C max;B  B2
S C C max;C  C2
No idle time is incurred at the rst machine on which each job is processed. Therefore, this work focuses on
minimizing the idle time on M2 and M3 to minimize the makespan of the problem accordingly. The optimality
follows from the observation that family C precedes family A on M2 because (i) the idle time on M2 is min-
imized despite the sequence on M1, and (ii) the idle time on M3 is minimized since both family C and B can be
scheduled as early as possible. Moreover, if family A precedes family B on M1, then family C should precede
family B on M3 to minimize the idle time on M3. Therefore, consideration of three schedules, schedule 1, 2 and
3 as in Fig. 2 suces to yield the optimal solution.
The optimal solution to the F3/GT, no-idle, o = 2/Cmax problem is eciently obtained with the aid of the
developed dominance conditions.
Since schedule 2 diers from schedule 3 only on M3, they are compared rst and then the dominance con-
ditions on which the schedule is selected can be derived. Both are further compared with schedule 1 to yield the
general dominance conditions. Dene Ii,j as the total idle time on machine Mi in schedule j, and Cti,j as the
completion time on machine Mi in schedule j, where i = 2 and 3, and j = 1, 2 and 3.

2.1. Comparison of schedule 2 with schedule 3

Ct3;2 I 2;2 S C C2 B2 max fS B  I 2;2 S C C2; 0g I 2;2 S C C2 B2

Ct3;3 S B B2 C2 max fI 2;3 S C  S B B2; 0g
Fig. 2 demonstrates that I2,2 = I2,3. Therefore, schedule 2 had to be compared only with schedule 3 on M3. In
schedule 2, the completion time of C2 exceeds that of C1 and the start time of A2 exceeds, therefore, the start
time of B2 is after the completion of B1 and max{SB-(I2,2 + SC + C2), 0} = 0: no idle time existed between C2
and B2. The following two cases are considered.

a b

M1 M1

M2 M2

Slot

Fig. 1. (a) The scheduling with Johnsons order of F2//Cmax problem. (b) The scheduling of F2/no-idle/Cmax problem.
 L.-H. Su, C.-T. Lin / Computers & Industrial Engineering 50 (2006) 286295 289

Fig. 2. Illustration of schedule 1, 2 and 3 for F3/GT, no-idle, o = 2/Cmax model.

Case 1. If SB > I2,2 + SC = max (B1 + (SA  C1), 0) + SC, then

Ct3;2 I 2;2 S C C2 B2
Ct3;3 S B B2 C2 > I 2;2 S C C2 B2 Ct3;2
Since Ct3,2 < Ct3,3, therefore schedule 2 is selected in this case.
Case 2. If SB < I2,2 + SC = max (B1 + (SA  C1), 0) + SC, then two sub-cases should be classied.

(i) If max{(I2,3 + SC)  (SB + B2), 0} = 0, then

Ct3;3 S B B2 C2 < I 2;2 S C B2 C2 Ct3;2 :
Since Ct3,3 < Ct3,2, therefore schedule 3 is selected.

(ii) If max{(I2,3 + SC)  (SB + B2), 0} = (I2,3 + SC)  (SB + B2), then

Ct3;3 S B B2 C2 I 2;3 S C  S B B2
I 2;3 S C C2 < I 2;2 S C C2 B2 Ct3;2
290 L.-H. Su, C.-T. Lin / Computers & Industrial Engineering 50 (2006) 286295

Since Ct3,3 < Ct3,2, therefore schedule 3 is selected.

By induction, the following conclusion is drawn when making a comparison between schedule 2 and 3:
If SB > max (B1 + (SA  C1), 0) + SC, then schedule 2 is selected; otherwise schedule 3 is selected.

2.2. Comparison of schedule 2 with schedule 1

Schedule 2 is further compared with schedule 1 provided that SB > max (B1 + (SA  C1), 0) + SC according
to the above analysis. By virtue of Fig. 2, the idle time on machine Mi under schedule j, where i = 2 and 3, and
j = 1 and 2 are as follows:

I 3;2 max fS B  I 2;2 S C C2; 0g I 2;2 S C I 2;2 S C

I 3;1 max fA1  I 2;1 S C C2 S B ; 0g I 2;1 S C
max fA1 S B  S C C2; I 2;1 g S C
I 2;2 maxB1  C1 S A ; 0 max fB1 S A  C1; 0g
I 2;1 maxS A  C1; 0

Evidently, I2,1 < I2,2 for SB > max (B1 + (SA  C1), 0) + SC, implying Ct2,2 > Ct2,1. Therefore, Ct2,2 remains to
be compared with Ct3,1 and Ct3,2 with Ct3,1.

Ct3;2 I 3;2 C2 B2 I 2;2 S C C2 B2 max fB1 S A  C1; 0g S C C2 B2

Ct2;2 I 2;2 C1 A2 max fB1 S A  C1; 0g C1 A2
Ct3;1 I 3;1 C2 B2 max fA1 S B  S C C2; I 2;1 g S C C2 B2
max fA1 S B  S C C2; S A  C1; 0g S C C2 B2

By induction, the following conclusion is drawn from the comparison between schedule 2 and schedule 1:
If max{B1 + (SA  C1), 0} + C1 + A2 > max{A1 + (SB  (SC + C2)), SA  C1, 0} + SC + C2 + B2 or
max{B1 + (SA  C1), 0} > max{A1 + (SB  (SC + C2)), SA  C1, 0}, then schedule 1 is selected; otherwise
schedule 2 is selected.

2.3. Comparison of schedule 3 with schedule 1

Schedule 3 is compared with schedule 1, when it is selected by comparison with schedule 2. By virtue of
Fig. 2, the idle time on machine Mi under schedule j, where i = 2 and 3, and j = 1 and 3 are as follows:

I 3;3 max fI 2;3 S C  S B B2; 0g S B

I 3;1 max fA1  I 2;1 S C C2 S B ; 0g I 2;1 S C
max fA1  C2 S B ; I 2;1 S C g
I 2;3 maxB1  C1 S A ; 0 max fB1 S A  C1; 0g
I 2;1 maxS A  C1; 0

Evidently, I2,1 < I2,3 under the condition SB < max (B1 + (SA  C1), 0) + SC, which implies Ct2,3 > Ct2,1.
Therefore, Ct2,3 remains to be compared with Ct3,1 and Ct3,3 with Ct3,1.

Ct3;3 I 3;3 C2 B2 max fI 2;3 S C  S B B2; 0g S B C2 B2

max fI 2;3 S C ; S B B2g C2 max fB1 S A S C  C1; S B B2; S C g C2
Ct2;3 I 2;3 C1 A2 max fB1 S A  C1; 0g C1 A2
Ct3;1 I 3;1 C2 B2 max fA1 S B  S C C2; I 2;1 g S C C2 B2
max fA1 S B  S C C2; S A  C1; 0g S C C2 B2
 L.-H. Su, C.-T. Lin / Computers & Industrial Engineering 50 (2006) 286295 291

By induction, the following conclusion is drawn by comparison between schedule 1 and schedule 3: If
max{B1 + (SA + SC  C1), SB + B2, SC} > max{A1 + (SB  (SC + C2)), SA  C1, 0} + SC + B2 or
max{B1 + (SA  C1), 0} + C1 + A2 > max{A1 + (SB  (SC + C2)), SA  C1, 0} + SC + C2 + B2, then sche-
dule 1 is selected; otherwise schedule 3 is selected.
In summary, the following conclusions are drawn for F3/GT, no-idle, o = 2/Cmax problem according to the
above analysis.

(1) If SB > max (B1 + (SA  C1), 0) + SC and max{B1 + (SA  C1), 0} + C1 + A2 > max {A1 +
(SB  (SC + C2)), SA  C1, 0} + SC + C2 + B2 or if SB > max (B1 + (SA  C1), 0) + SC and
max{B1 + (SA  C1), 0} > max {A1 + (SB  (SC + C2)), SA  C1, 0}, then schedule 1 is selected, other-
wise schedule 2 is selected.
(2) If SB < max (B1 + (SA  C1), 0) + SC and max{B1 + (SA + SC  C1), SB + B2, SC} > max
{A1 + (SB  (SC + C2)), SA  C1, 0} + SC + B2 or if max{B1 + (SA + SC  C1), SB + B2, SC} >
max{A1 + (SB  (SC + C2)), SA  C1, 0} + SC + B2 and max{B1 + (SA  C1), 0} + C1 + A2 > max
{A1 + (SB  (SC + C2)), SA  C1, 0} + SC + C2 + B2, then schedule 1 is selected, otherwise schedule
3 is selected.

Property 1. If A2 + C1 > B2 + max (A1 + SB ;I2,1, SC + C2), where I2,1 = max (SA  C1, 0), then schedule 1,
as shown in Fig. 2, yields an optimal solution to the F3/GT, no-idle, o = 2/Cmax problem.
Proof. If A2 + C1 > B2 + max (A1 + SB  I2,1, SC + C2), then makespan is determined by the maximum
completion time on M2. If no idle time is incurred on M2, then a completion time Ct2,1 of C1 + A2 minimizes
the makespan; otherwise if idle time is incurred on M2, then its value I21 is minimum since Johnsons rule
yields an optimal schedule for jobs in family A. h

3. The F3/GT, o = 2/Cmax problem

This section considers the model F3/GT, o = 2/Cmax in which the machine no-idle constraint in the previous
case is relaxed. Evidently, three candidate schedules, schedule 1, 2 and 3, remain for further consideration.
Since schedule 2 diers from schedule 3 only on M3, these schedules are compared rst, and the dominance
conditions between them are derived. Both are compared to schedule 1 further to induce the general domi-
nance conditions.

3.1. Comparison of schedule 2 with schedule 3

Two cases are considered. The statement for case (i) is proved, the proof for case (ii) follows from
symmetry.

(i) SC < SB. From Fig. 3, the relationships between the completion times of schedule 2 and 3 on M3 are as
follows:

Ct3;2 maxS C C2 B2; S B B2 < Ct3;3 S B B2 C2

Therefore, schedule 2 is selected.

(ii) SC > SB. Schedule 3 is selected in this case.

3.2. Comparison of schedule 2 with schedule 1

Schedule 2 is then compared with schedule 1 if the condition SC < SB holds. According to Fig. 3, the idle
time on machine Mi under schedule j, where i = 2 and 3, and j = 1 and 2 are as follows:
292 L.-H. Su, C.-T. Lin / Computers & Industrial Engineering 50 (2006) 286295

Schedule 1

M1 A1 B1

M2 C1 A2
max (S A C1,0 )
M3 SC C2 B2

max{A1 + (S B (S C + C 2)),0}

Schedule 2

M1 B1 A1

M2 C1 A2
max{B1 + (S A C1),0}

M3 SC C2 B2

max{S B (S C + C 2),0}

Schedule 3

M1 B1 A1

M2 C1 A2

max {B1 + (S A C1),0}

M3 SB B2 C2

max {SC ( S B + B 2), 0}

Fig. 3. Illustration of schedule 1, 2 and 3 for F3/GT, o = 2/Cmax model.

I 3;2 max fS B  S C C2; 0g S C

I 3;1 max fA1  S C C2 S B ; 0g S C
max fA1 S B  S C C2; 0g S C
I 2;2 maxB1  C1 S A ; 0 max fB1 S A  C1; 0g
I 2;1 maxS A  C1; 0
Obviously, I3,2 < I3,1, implying Ct3,2 < Ct3,1, and I2,2 > I2,1, implying Ct2,2 > Ct2,1 under the condition SC < SB.
Accordingly, a further comparison is made between Ct2,2 and Ct3,1 to determine the dominance conditions.
Ct2;2 I 2;2 C1 A2 max fB1 S A  C1; 0g C1 A2 maxB1 S A ; C1 A2
Ct3;1 I 3;1 C2 B2 maxA1 S B ; S C C2 B2
Therefore, if max (B1 + SA, C1) + A2 < max (A1 + SB, SC + C2) + B2, then schedule 2 is selected; otherwise
schedule 1 is selected.
 L.-H. Su, C.-T. Lin / Computers & Industrial Engineering 50 (2006) 286295 293

3.3. Comparison of schedule 3 with schedule 1

Schedule 3 is compared with schedule 1 if SC > SB. Based on Fig. 3, the idle time on machine Mi under
schedule j, where i = 2 and 3, and j = 1 and 3 are as follows:

I 3;3 max fS C  S B B2; 0g S B max fS C  B2; S B g

I 3;1 max fA1  S C C2 S B ; 0g S C max fA1  C2 S B ; S C g
I 2;3 maxB1  C1 S A ; 0 max fB1 S A  C1; 0g
I 2;1 maxS A  C1; 0

Observably, I3,3 < I3,1, implying Ct3,3 < Ct3,1, and I2,3 > I2,1, implying Ct2,3 > Ct2,1, under the condition
SC > SB. Therefore, Ct2,3 is further compared with Ct3,1.

Ct2;3 I 2;3 C1 A2 max fB1 S A  C1; 0g C1 A2 maxB1 S A ; C1 A2

Ct3;1 I 3;1 C2 B2 maxA1 S B ; S C C2 B2;

By induction, if max (B1 + SA, C1) + A2 < max (A1 + SB, SC + C2) + B2, then schedule 3 is selected; other-
wise schedule 1 is selected.
In summary, the following conclusions are drawn for F3/GT, o = 2/Cmax problem.

(1) If SC < SB and max (B1 + SA, C1) + A2 < max (A1 + SB, SC + C2) + B2, then schedule 2 is selected,
otherwise schedule 1 is selected.
(2) If SC > SB and max (B1 + SA, C1) + A2 < max (A1 + SB, SC + C2) + B2, then schedule 3 is selected,
otherwise schedule 1 is selected.

Property 2. If SB + B2 > max (B1 + SA, C1) + A2 and SB > SC + C2, then schedule 2, as shown in Fig. 3, gives
an optimal solution to the F3/GT, o = 2/Cmax problem.
Proof. If SB + B2 > max (B1 + SA, C1) + A2, the makespan is determined by M3. Also, when SB > SC + C2,
the makespan SB + B2 is minimized on M3 since Johnsons rule gives a minimum SB. h
Property 3. If SC + C2 > max (B1 + SA, C1) + A2 and SC > SB + B2, then schedule 3, as shown in Fig. 3, gives
an optimal solution to the F3/GT, o = 2/Cmax problem.
Proof. This property is easily proven by argument similar to that for Property 2. h

4. The F3/no-idle, o = 2/Cmax problem

This section considers the model of F3/no-idle, o = 2/Cmax by relaxing the GT restriction of the rst model.
Gonzalez and Sahni (1978) showed that the problem of F3/o 6 2/Cmax is NP-hard in the ordinary sense. Bap-
tiste and Hguny (1997) demonstrated that the problem F3/no-idle/Cmax is NP-hard in the strong sense. The
problem F3/no-idle, o = 2/Cmax is equivalent to its counterpart without a missing operation in each job
and is NP-hard in the ordinary sense.
Although F3/no-idle, o = 2/Cmax is shown to be NP-hard in the ordinary sense, some restricted versions are
polynomially solvable. A polynomial time algorithm, Algorithm A, is proposed to derive an optimal schedule
for some restricted versions of the problem.
Algorithm A

Step 1. Construct an articial F2//Cmax problem by mixing jobs in families B and C without specifying
machine on which their rst operations are to be processed. Apply Johnsons rule to solve this
articial problem. Arrange the sequence obtained on M3.
294 L.-H. Su, C.-T. Lin / Computers & Industrial Engineering 50 (2006) 286295

Step 2. Order the jobs in family A by applying Johnsons rule. Copy out job sequence in family B obtained in
step 1 to the left, followed by the sequence of jobs in family A just derived to the right on M1.
Step 3. Copy the job sequence of family C obtained in step 1 to the left, and then copy the job sequence of
family A derived in step 2 on M2.

Corollary 1. If C1 > B1 + SA, then Algorithm A gives the optimal result in polynomial time for F3/no-idle,
o = 2/Cmax problem.
Proof. M1 has no idle time and so does on M2under the condition C1 > B1 + SA, as shown in Fig. 4. Since all
jobs in family B and C are sequenced by Johnsons rule and their rst operations are shifted left on M1 and
M2, respectively, without affecting the second processing on M3, the idle time on M3 is minimum. h
Once again, we show that, under the conditions in Corollary 2, Algorithm A is demonstrated to yields an
optimal schedule for F3/no-idle, o = 2/Cmax problem. Denote Pi,j,k as the processing time of job i in family k
on Mj.
Corollary 2. Algorithm A find an optimal schedule if max Pi,2,A < min Pi,1,A, for i = 1, 2,   , n, and jobs
completed during the beginning interval (B1 + Cmax,A  A2  C1) on M3 are of family B for the F3/no-idle,
o = 2/Cmax problem.
Proof. As displayed in Fig. 5, two cases may apply. First, if the makespan is determined by the maximum
completion time on M2, then the value (B1 + A1) plus the processing time of the last job in family A under
the condition max Pi,2,A < min Pi,1,A, for i = 1, 2, . . ., n, gives a minimum completion time on M2. Next, if
the makespan is determined by the maximum completion time on M3, then an idle time interval,
(B1 + Cmax,A  A2  C1), is incurred by a right shift of the jobs in family C on M2 because of the machine
no-idle assumption. The jobs in family C can only begin to be processed on M3 following this interval. If jobs
completed during the interval (B1 + Cmax,A  A2  C1) are in family B, then this interval is used in the most
effective way, minimizing the completion time on M3. h

Fig. 4. Illustration of Algorithm A under the condition C1 > B1 + SA for F3/no-idle, o = 2/Cmax problem.

Fig. 5. Illustration of Algorithm A under the conditions of Corollary 2 for F3/no-idle, o = 2/Cmax problem.
 L.-H. Su, C.-T. Lin / Computers & Industrial Engineering 50 (2006) 286295 295

5. Conclusion

This work considers the minimization of the makespan for three variants of the F3/no-idle, o = 2/Cmax
problem. The general version of this problem has been shown to be ordinary NP-hard. Three variants, based
on GT and the machine no-idle assumption, are proposed.
This study presents approaches for solving simple problems to be extend to more realistic problems, such as
that of the multistage ow shop system in which each job must be processed on dierent machines under the
requirement of group technology or the machine no-idle condition. The scheduling problem in such a realistic
system is necessarily strongly NP-hard. Thus, heuristics must be relied upon to solve such problems. There-
fore, further research should focus on designing ecient heuristics.

References

Adiri, I., & Pohoryles, D. (1982). Flow-shop/no-idle or no-wait scheduling to minimize the sum of completion times. Naval Research
Logistics Quarterly, 29, 495504.
Andres, C., Albarracin, J. M., Tormo, G., & Vicens, E. (2005). Group technology in a hybrid owshop environment: A case study.
European Journal of Operational Research, 167, 272281.
Baptiste, P., & Hguny, L. K. (1997). A branch and bound algorithm for the F/no-idle/Cmax (Vol. 1). In Proceedings of the International
Conference on Industrial Engineering and Production Management IEPM97 (pp. 429438). Lyon.
Garey, M. R. D., Johnson, D. S., & Sethi, R. (1976). The complexity of owshop and jobshop scheduling. Mathematics of Operations
Research, 1, 117129.
Gonzalez, T., & Sahni, S. (1978). Flowshop and jobshop schedules: Complexity and approximation. Operations Research, 26, 3652.
Johnson, S. M. (1954). Optimal two- and three-stage production schedules. Naval Research Logistics Quarterly, 1, 6168.
Kamburowski, J. (2004). More on three-machine no-idle ow shops. Computers & Industrial Engineering, 46, 461466.
Liaee, M. M., & Emmons, H. (1997). Scheduling families of jobs with setup times. International Journal of Production Economics, 41,
165176.
Lin, B. M. T., & Cheng, T. C. E. (2001). Batch scheduling in the no-wait two-machine ow shop to minimize the makespan. Computers &
Operations Research, 28, 613624.
Riane, F., Artiba, A., & Elmaghraby, S. E. (1998). A hybrid three-stage owshop problem: Ecient heuristics to minimize makespan.
European Journal of Operational Research, 109, 321329.
Saadani, H., Guinet, A., & Moalla, M. (2003). Three stage no-idle ow-shops. Computers & Industrial Engineering, 44, 425434.
Saadani, H., Guinet, A., & Moalla, M. (2005). A travelling salesman approach to solve the F/no-idle/Cmax problem. European Journal of
Operational Research, 161, 120.
Yang, D. I., & Chern, M. S. (2000). Two-machine ow shop group scheduling problem. Computers & Operations Research, 27, 975985.