You are on page 1of 13

Available online at www.sciencedirect.

com

Computers & Operations Research 30 (2003) 1335 1347

www.elsevier.com/locate/dsw

Single-machine scheduling with periodic maintenance and


nonresumable jobs
C.J. Liao , W.J. Chen
Department of Industrial Management, National Taiwan University of Science and Technology, 43 Keelung Road,
Section 4, Taipei, Taiwan
Received 1 July 2001; received in revised form 1 January 2002

Abstract
We study a single-machine scheduling problem where periodic maintenance is required in a complete
schedule. Although the scheduling problem with maintenance has attracted some researchers attention, most of
them consider only one maintenance period. In this paper, several maintenance periods are under consideration
where each maintenance is required after a periodic time interval. Speci2cally, the problem is to minimize
the maximum tardiness with periodic maintenance and nonresumable jobs. A branch-and-bound algorithm
that utilizes several inherent theorems is developed to derive the optimal schedule for the problem. To solve
large-sized problems, a heuristic is also developed. Computational results are provided to demonstrate the
e6ectiveness of the heuristic.
Scope and purpose
A survey of scheduling literature reveals that few researches consider the scheduling problem with maintenance. However, machines subject to maintenance are found prevalently in process industries and manufacturing systems. The purpose of this paper is to solve a single-machine scheduling problem with periodic
maintenance. The criterion considered is to minimize the maximum tardiness. We propose a heuristic for
2nding the near-optimal solution for large-sized problems. A branch-and-bound algorithm is also provided to
derive the optimal solution. Computational results demonstrate the e6ectiveness and e8ciency of the heuristic.
? 2002 Published by Elsevier Science Ltd.
Keywords: Scheduling; Periodic maintenance; Maximum tardiness; Nonresumable job

Corresponding author. Tel.: +886-2-2737-6437; fax: +886-2737-6360.


E-mail address: cjl@im.ntust.edu.tw (C.J. Liao).

0305-0548/03/$ - see front matter ? 2002 Published by Elsevier Science Ltd.


PII: S 0 3 0 5 - 0 5 4 8 ( 0 2 ) 0 0 0 7 4 - 6

1336

C.J. Liao, W.J. Chen / Computers & Operations Research 30 (2003) 1335 1347

1. Introduction
Most scheduling models assume that machines are available at all times. However, this assumption
may not be true in real industrial settings. For example, a preventive maintenance activity is required
or a machine breakdown occurs. According to practical experience, we sometimes 2nd that some of
the machines wait for maintenance while there are jobs waiting to be processed by these machines.
This is due to the lack of coordination between maintenance planning and production scheduling.
Therefore, this paper aims to improve the situation by deriving a satisfactory schedule that considers
both jobs and maintenance simultaneously.
An uncertain breakdown will make the shop behavior hard to predict, and thereby reducing the
e8ciency of the production system. Maintenance can reduce the breakdown rate with minor sacri2ces in production time. The importance of maintenance has gradually recognized by the decision
maker. Therefore, it has been scheduled periodically in many manufacturing systems, especially in
a just-in-time environment. The work of periodic maintenance includes periodic inspection, periodic
repair, and preventive maintenance. With proper planning of periodic maintenance, the shop can
concentrate on production e8ciency and safety, resulting in increasing productivity and achieving
high safety awareness [1].
Scheduling with maintenance is usually treated as availability constraint in the scheduling literature.
Although such an availability problem is important, it is relatively unexplored. For the single-machine
problem, Adiri et al. [2] assume that the unavailable time is unknown but with a probabilistic
distribution. They distinguish two cases of a breakdown, i.e., the resumable and nonresumable (or
restarting) cases. By applying the earliest due date (EDD) and modi2ed shortest processing time
(MSPT) rules for the two cases, respectively, the criterion of number of tardy jobs can be minimized.
For the two-machine problem, Mosheiov [3] solves the two-parallel-machine-scheduling problem for
minimizing total completion time by assuming each machine is available in a speci2ed interval. Lee
and Liman [4] consider a similar problem where one machine is not available for a speci2ed period
of time. They provide a pseudo-polynomial dynamic programming algorithm to solve the problem.
Lee [5 8] studies in this area for other machine con2gurations including single and parallel machines,
and di6erent performance measures such as makespan, total weight completion time, tardiness, and
number of tardy jobs.
In all the above papers, there is only one unavailability or availability period for each machine. As
stated earlier, however, maintenance has been scheduled regularly, or periodically, in many manufacturing systems. Therefore, there is a need to develop scheduling method to deal with systems with periodic maintenance, which usually has more than one maintenance periods. In our problem, there are
several maintenance periods where each maintenance is required after a 2xed time interval. The objective is to minimize the maximum tardiness subject to periodic maintenance and nonresumable jobs.

2. Notation and problem setting


The following notation will be used throughout this paper:
n = number of jobs for processing at time zero
Jj = job number j

C.J. Liao, W.J. Chen / Computers & Operations Research 30 (2003) 1335 1347

1337

Fig. 1. A schedule with periodic maintenance for nonresumable case.

Mi = maintenance period i
pj = processing time of job j
dj = due date of job j
Cj = completion time of job j
CMi = completion time of maintenance period i
Lj = lateness of job j, where Lj = Cj dj
Tj = tardiness of job j, where Tj = max{0; Lj }
Tmax = maxj {Tj }
t
= amount of time to perform one maintenance
T = time interval between two maintenance periods
In addition, J[j] denotes the job placed at the jth position, and p[j] ; d[j] ; C[j] ; L[j] , and T[j] are
de2ned accordingly.
We consider a single-machine problem where jobs are nonresumable, i.e., once a job is started
it must be completed without interruption. The setup time of job is sequence independent and
included in the processing time. Also, it is assumed that all jobs are ready for processing at time
zero and that, for simplicity, processing times and due dates can take only integral values. For each
maintenance, it requires an amount of time t for the performance. There is a time interval T between
two consecutive maintenance periods, in which jobs can be scheduled for processing (see Fig. 1).
The problem under consideration is to 2nd a schedule that minimizes the maximum tardiness subject
to periodic maintenance and nonresumable jobs. It is clear that the problem is NP-hard since the
problem that minimizes the maximum lateness subject to one unavailability period and nonresumable
jobs is NP-hard [6].
3. The proposed heuristic
De2ne a batch as a set of jobs in each interval T . Then we have the following theorems that will
be used in the heuristic.
Theorem 1. Suppose Tp is the maximum tardiness of a schedule; where Jp is in batch B. Denote
by Ji a job preceding Jp in B. If di is greater than or equal to dp ; then a smaller Tmax value can
be obtained by interchanging Jp with Ji .
Proof. Let partial schedule S = (PS; Ml ; 1 ; Ji ; 2 ; Jp ); where PS; 1 and 2 are partial schedules. Let
S  be obtained from S by interchanging Ji with Jp . Denote by P1 and P2 the total processing
times of 1 and 2 ; respectively. Then the tardiness of Jp in S is
Tp = CMl + P1 + pi + P2 + pp dp :

1338

C.J. Liao, W.J. Chen / Computers & Operations Research 30 (2003) 1335 1347

The tardiness of Ji in S  is
Ti = CMl + P1 + pp + P2 + pi di :
If di is greater than or equal to dp ; then Ti 6 Tp = Tmax . Thus; S  has a Tmax value smaller than or
equal to that of S. This completes the proof.
Theorem 2. Suppose Tp is the maximum tardiness of a schedule; where Jp is in batch B. Denote
by Jj a job in the batch preceding B. If dj pj is greater than or equal to dp pp ; then a smaller
Tmax value can be obtained by interchanging Jp with Jj .
Proof. Let partial schedule S = (PS; Mk ; 1 ; Jj ; ; Ml ; 2 ; Jp ); where PS and  are partial schedules;
and 1 and 2 are partial schedules of the same batches as Jj and Jp ; respectively. Let S  be obtained
from S by interchanging Jj with Jp . Denote by P1 and P2 the total processing times of 1 and
2 ; respectively. Then the tardiness of Jp in S is
Tp = CMl + P2 + pp dp :
The tardiness of Jj in S  is
Tj = CMl + P2 + pj dj :
If dj pj is greater than or equal to dp pp ; then Tj 6 Tp = Tmax . Thus; S  has a Tmax value smaller
than or equal to that of S. This completes the proof.
For explanatory convenience, we de2ne two terms that are needed in the heuristic.
Denition 1. The slack time of a batch is de2ned as the amount of time unscheduled in a batch.
Denition 2. The potential position is de2ned as the position immediately after the last position of
the 2rst batch where its slack time is greater than or equal to the processing time of the considered
job.
The steps of the proposed heuristic are outlined as follows:
Step 1: Obtain a schedule with the smallest number of maintenance periods by the following
procedure:
(i) Generate a batch by grouping a set of jobs such that the total processing time of jobs in the
batch is equal to T (i.e., the bin-packing technique [9]). Repeat by grouping the remaining jobs
to form other batches.
(ii) Repeat (i) except that the total processing time is changed to T 1; T 2; : : : ; until all jobs are
grouped.
Step 2: Sequence the jobs in each batch in EDD order. If two jobs have the same due date, place
the job with smaller processing time in the front.
Step 3: Sequence the batches in EDD order by letting the smallest due date of jobs in each batch
be the due date of the batch.

C.J. Liao, W.J. Chen / Computers & Operations Research 30 (2003) 1335 1347

1339

Step 4: Denote by  the resulting complete schedule, which is composed of the schedules in all
batches and the necessary maintenance periods. Calculate T[j] for all j in  .
Step 5: If T[j] = 0 for all j, stop; otherwise, 2nd J[p] with Tmax .
Step 6: Let  be the schedule obtained by interchanging J[p] with each of the jobs preceding J[p] ,
say J[q] . If J[p] and J[q] belong to the same batch, we apply Theorem 1 to determine whether any
interchange of J[p] and J[q] is advantageous; otherwise, Theorem 2 is applied to determine whether
any interchange of J[p] and J[q] is advantageous. If it is, go to Step 7. Otherwise, the interchange is
repeated until no J[q] can be found. Go to Step 8.
Step 7: Replace  with  if both the following conditions are satis2ed:
(i) The total processing times in the two batches are not larger than T (if J[p] and J[q] belong to
di6erent batches).
(ii) The Tmax value is reduced.
If the replacement is achieved, calculate T[j] for all j. Return to Step 5.
Step 8: Perform the movement if any of the following cases is satis2ed:
(i) Moving J[p] forward to the potential position can satisfy both the conditions in Step 7.
(ii) Moving any job preceding J[p] in the same batch as J[p] to the potential position can satisfy
both the conditions in Step 7.
(iii) Moving any job preceding J[p] , say J[r] , to the potential position and moving J[p] to J[r] can
satisfy both the conditions in Step 7.
Step 9: Reset the jobs in each batch in EDD order. Stop.
We now elaborate the steps in detail. It is clear that the smallest number of maintenance periods
is obtained in Step 1. In Steps 2 4, we construct an initial schedule, where both jobs and batches
are sequenced in EDD and the maintenance periods are incorporated. In Step 6, Theorems 1 and 2
are used to determine whether any interchange is advantageous. In Step 7, the two conditions must
be satis2ed for any improvement because the batch of jobs cannot be completed in T if condition (i)
cannot be met. In Step 8, we check whether it is advantageous to move the job with Tmax forward
((i) and (iii)) or to reduce the completion time of the job ((ii)).
To calculate the time complexity of the proposed heuristic, note that the bin-packing technique in
Step 1 can be carried out in O(n log n). Thecomplexities of Steps 2, 3, and
9 are O(n log n). Step
5 needs O(n). Steps 6 and 7 require O(n2 nk=1 p[k] ). Step 8needs O(n2 p[k] ). Thus, we have
the overall time complexity of the proposed heuristic as O(n2 ni=1 pi ).

4. A numerical example
Example 1. As an illustration of the heuristic; consider a single-machine scheduling problem with
nine jobs; as given in Table 1; where T = 8 and t = 2.
Applying Steps 15, we obtain an initial schedule as given in Table 2, where there are three
maintenance periods with Tmax =24. Since J9 is associated with Tmax , it is selected to be interchanged.

1340

C.J. Liao, W.J. Chen / Computers & Operations Research 30 (2003) 1335 1347

Table 1
The data for Example 1 (in h)
Ji

J1

J2

J3

J4

J5

J6

J7

J8

J9

pi
di

1
1

05
13

3
2

05
30

02
10

02
13

03
20

04
12

04
14

Table 2
The initial schedule for Example 1 (in h)
Ji

J1

J2

pi
Ci
di
Ti

1
1
1
0

05
06
13
00

M1

J3

J4

03
13
02
11

05
18
30
00

M2

J5

J6

J7

02
22
10
12

02
24
13
11

03
27
20
07

J5

J6

J9

02
22
10
12

02
24
13
11

04
28
14
14

J5

J6

J9

2
22
10
12

2
24
13
11

4
28
14
14

M3

J8

J9

04
34
12
22

04
38
14
24

J8

J7

04
34
12
22

03
37
20
17

J4

J7

5
35
30
5

3
38
20
18

Table 3
The second schedule for Example 1 (in h)
Ji

J1

J2

pi
Ci
di
Ti

1
1
1
0

05
06
13
00

M1

J3

J4

03
13
02
11

05
18
30
00

M2

M3

Table 4
The third schedule for Example 1 (in h)
Ji

J1

J2

pi
Ci
di
Ti

1
1
1
0

5
6
13
0

M1

J3

J8

3
13
2
11

4
17
12
5

M2

M3

In Steps 6 and 7, we interchange J9 with J7 and obtain Tmax = 22 (see Table 3). Now J8 = Tmax
becomes the candidate to be interchanged. There are three jobs preceding J8 (i.e., J2 ; J4 , and J9 )
being considered and d4 p4 is the largest, so J4 is selected (by Theorem 2). After the interchange,
we obtain a schedule with Tmax = 18 (see Table 4). Following the same procedure, we further
interchange J7 with J4 to reduce Tmax to 14 (see Table 5).
According to Theorems 1 and 2, any interchange cannot improve the schedule any further. Thus,
we perform Step 8 to check other possible moves. We 2nd that J5 , which precedes J9 , can be moved
to the potential position. After the jobs in each batch are reset in EDD (Step 9), we obtain the 2nal
schedule with Tmax = 13 (see Table 6).

C.J. Liao, W.J. Chen / Computers & Operations Research 30 (2003) 1335 1347

1341

Table 5
The fourth schedule for Example 1 (in h)
Ji

J1

J2

pi
Ci
di
Ti

1
1
1
0

05
06
13
00

M1

J3

J8

03
13
02
11

04
17
12
05

M2

J5

J6

J9

02
22
10
12

02
24
13
11

04
28
14
14

M2

J6

J9

02
22
13
09

04
26
14
12

M3

J7

J4

03
33
20
13

05
38
30
08

J7

J4

03
33
20
13

05
38
30
08

Table 6
The 2nal schedule for Example 1 (in h)
Ji

J1

J5

J2

pi
Ci
di
Ti

1
1
1
0

02
03
10
00

05
08
13
00

M1

J3

J8

03
13
02
11

04
17
12
05

M3

5. The branch-and-bound algorithm


In this section, a branch-and-bound (B&B) algorithm is presented to provide the optimal schedule,
which can be used to evaluate the proposed heuristic. There are several dominance rules stated in
the theorems that will be used in the B&B algorithm. Let PS; S, and S  be partial schedules.
Theorem 3. For two adjacent jobs in the same batch; there exists an optimal schedule for which
the job with smaller due date is placed before the other job.
Proof. The result follows immediately from the EDD order.
Theorem 4. Let S be created by appending (Jp ; Mi ; Jq ) to PS. Let S  be obtained from S by
interchanging Jq with Jp . If dp pp dq pq ; then S can be dominated by S  .
Proof. Let ST be the minimum of the starting times of Jp and Jq . Let L be the maximum lateness of
PS under S and L be the corresponding quantity under S  . Let Lp ; Lq be the lateness of Jp ; Jq under
S and Lp ; Lq be the corresponding quantity under S  . We; therefore; have the maximum lateness
under S
Lmax = (L; Lp ; Lq );
and the maximum lateness under S 
Lmax = (L ; Lp ; Lq )
= (L; Lp ; Lq ):
Now under S
Lp = ST + pp dp ;

Lq = CMi + pq dq ;

1342

C.J. Liao, W.J. Chen / Computers & Operations Research 30 (2003) 1335 1347

and under S 
Lq = ST + pq dq ;

Lp = CMi + pp dp :

Therefore; Lq Lq (since CMi ST ); and Lq Lp (since dp pp dq pq ). Hence Lq max(Lp ; Lq ).
Since Lq Lp ; we have
max(L; Lp ; Lq ) = max(L; Lq )
max(L; Lp ; Lq )
= max(L ; Lp ; Lq ):
So Lmax Lmax Tmax (S) = max{0; Lmax } max{0; Lmax } = Tmax (S  ). Hence; S can be dominated
by S  .
Theorem 5. Let Jp and Jq be any two jobs in S; where Jp is placed before Jq . Let S  be obtained
from S by interchanging Jq with Jp . Suppose that the total processing time of each batch in S  is
no larger than T. If Tmax = Tq for Jq in both S and S  ; then S can be dominated by S  .
Proof. Suppose that both Jp and Jq are in the same batch in both S and S  . Then the completion
time of Jq in S  is smaller than that of Jq in S. Since Tmax = Tq for Jq in both S and S  ; the Tmax
value of S  is smaller than that of S. Hence; S can be dominated by S  .
Suppose that Jp and Jq are in di6erent batches in both S and S  . The proof is the same as in the
same batch.
Theorem 6. Consider two partial schedules composed of the same set of jobs with the same completion time. Then the partial schedule with larger Tmax value can be dominated by the other.
Proof. Denote by S and S  the two considered partial schedules; where S has larger Tmax value.
Let R be the partial schedule composed of the remaining jobs. It is clear that appending R to S
will result in a complete schedule with Tmax value no smaller than that of the complete schedule by
appending R to S  . Thus; S can be dominated by S  .
For a partial schedule, the lower bound of Tmax , Tmax , is calculated based on a completion of the
partial schedule. The complete schedule is constructed by sequencing the unscheduled jobs in EDD
order. Then, the steps of the B&B algorithm can be stated as follows:
Step 1: Initialization. Use the heuristic to obtain an initial schedule.
Step 2: Branching. Select a partial schedule with the smallest Tmax value among all unfathomed
partial schedules. Let J[i] be the last job of the partial schedule. Create new partial schedules by
placing each of the unscheduled jobs at position i + 1. Note that if the job requires a processing
time greater than the slack time, it is scheduled in the next time interval T . If the number of
maintenance periods of the new partial schedule is greater than that of the incumbent schedule, then
it is eliminated. Use the dominance rules given in Theorems 3 6 to eliminate any possible newly
created partial schedules.
Step 3: Bounding. For each newly created partial schedule, calculate Tmax by using the proposed
scheme stated earlier.

C.J. Liao, W.J. Chen / Computers & Operations Research 30 (2003) 1335 1347

1343

Table 7
The computational results (Data Set I)
n

05

10

B&B algorithm

Heuristic

PEDa

Comp. time (s)

Comp. time (s)

Min.

Mean

Max.

2
4
2
4
2
4

0.08
0.08
0.07
0.05
0.05
0.05

0.10
0.11
0.08
0.08
0.06
0.05

0.00
0.00
0.00
0.00
0.00
0.00

0.00
0.01
0.00
0.00
0.00
0.00

0.00
0.05
0.00
0.00
0.00
0.00

2
4
2
4
2
4

5.21
5.46
4.47
4.83
3.71
4.24

0.61
0.66
0.55
0.59
0.51
0.57

0.00
0.00
0.00
0.00
0.00
0.00

0.62
0.77
0.66
0.58
0.44
0.53

4.51
6.22
3.63
5.81
3.29
3.65

2
4
2
4
2
4

65.57
71.03
59.89
59.37
52.64
59.14

3.66
3.85
2.99
3.07
2.45
2.52

0.00
0.00
0.00
0.00
0.00
0.00

0.95
1.02
0.83
0.89
0.62
0.75

7.71
9.93
3.96
6.52
3.79
4.08

2
411.87
4
413.59
14
2
388.12
4
376.81
18
2
361.86
4
363.36
a
PED: percentage error deviation.

9.13
9.88
9.23
9.37
9.01
9.07

0.01
0.00
0.00
0.00
0.00
0.00

1.05
1.08
0.92
0.98
0.75
0.78

9.19
9.67
5.21
6.35
6.84
4.16

14
18
10

10
14
18

20

10
14
18

30

10

Step 4: Fathoming. Fathom those newly created partial schedules whose Tmax values are no better
than the incumbent schedule. Update the incumbent schedule if a new partial schedule has a smaller
Tmax value.
Step 5: Stop. If there are nodes remaining unfathomed, go to Step 2; otherwise, stop.
To determine the time complexity of the branch-and-bound algorithm, we see that a typical branchn
ing tree
n has 2 nodes for an n-job schedule. Thus, the overall time complexity of the algorithm is
n
O(2
i=1 pi ).
6. Computational results
To compare the heuristic with the B&B algorithm, an experimental work was conducted. Both
the heuristic and the B&B algorithm were coded in Visual BASIC and run on a PC-686. The test

1344

C.J. Liao, W.J. Chen / Computers & Operations Research 30 (2003) 1335 1347

Table 8
The computational results (Data Set II)
n

10

B&B algorithm

Heuristic

PEDa

Comp. time (s)

Comp. time (s)

Min.

Mean

Max.

2
4
2
4
2
4

0.08
0.09
0.06
0.05
0.04
0.05

0.09
0.12
0.08
0.09
0.06
0.06

0.00
0.00
0.00
0.00
0.00
0.00

0.01
0.01
0.00
0.00
0.00
0.00

0.03
0.08
0.00
0.00
0.00
0.00

2
4
2
4
2
4

5.17
5.51
4.58
4.83
3.68
4.11

0.67
0.74
0.59
0.63
0.55
0.62

0.00
0.00
0.00
0.00
0.00
0.00

0.58
0.76
0.59
0.63
0.48
0.51

4.36
7.54
3.71
5.25
4.11
4.33

2
4
2
4
2
4

66.92
68.14
61.35
63.71
52.63
57.35

3.63
3.81
2.92
3.16
2.56
2.71

0.00
0.01
0.00
0.00
0.00
0.00

0.99
0.98
0.82
0.91
0.65
0.77

6.28
9.17
4.68
5.15
3.74
4.19

2
421.17
4
426.36
14
2
391.72
4
378.85
18
2
358.95
4
375.13
a
PED: percentage error deviation.

9.19
9.42
9.04
9.16
8.93
9.11

0.01
0.01
0.00
0.00
0.00
0.00

1.06
1.13
0.92
0.96
0.77
0.81

6.72
8.36
4.52
8.24
5.88
4.31

14
18
10

10
14
18

20

10
14
18

30

10

problems were randomly generated based on the following scheme: processing times were selected
from a discrete uniform 
distribution (DU) over [1,10].
 The due dates were selected from another
DU over [(1 C Q=2) nk=1 p[k] , (1 C + Q=2) nk=1 p[k] , with restriction d[i] 0, where Q and
C denote the due date range and tardiness factor, respectively. The experimental procedure consists
of a full factorial design with two settings of Q (Q = 0:2; 0:6) and two settings of C (C = 0:2; 0:6).
For convenience, the four combinations C = 0:2, Q = 0:2; C = 0:2, Q = 0:6; C = 0:6, Q = 0:2; and
C = 0:6, Q = 0:6 are referred to as Data Sets I, II, III, and IV, respectively. The experiments consist
of three levels of T (10; 14; 18) and two levels of t(2; 4).
The computational results are summarized in Tables 710 for Data Sets I, II, III, and IV, respectively. In each complete trial, we randomly drew n (n = 5; 10; 20; 30) jobs. The tables provide
the information on the average percentage error deviation and computation time for each combination of n, T and t. The number of problems solved in each combination is 24. The percentage
error deviation (PED) used to validate the performance of the heuristic is computed as follows:

C.J. Liao, W.J. Chen / Computers & Operations Research 30 (2003) 1335 1347

1345

Table 9
The computational results (Data Set III)
n

10

B&B algorithm

Heuristic

PEDa

Comp. time (s)

Comp. time (s)

Min.

Mean

Max.

2
4
2
4
2
4

0.08
0.10
0.06
0.07
0.06
0.06

0.11
0.11
0.08
0.09
0.07
0.07

0.00
0.00
0.00
0.00
0.00
0.00

0.01
0.01
0.00
0.00
0.00
0.00

0.02
0.04
0.00
0.00
0.00
0.00

2
4
2
4
2
4

5.93
5.75
5.36
5.40
4.35
4.41

1.21
1.18
0.91
0.87
0.72
0.63

0.00
0.00
0.00
0.00
0.00
0.00

0.88
1.05
0.71
0.76
0.61
0.63

8.17
5.51
5.03
6.94
4.63
4.85

2
4
2
4
2
4

72.11
72.69
68.96
69.71
60.25
62.44

4.15
4.36
3.33
3.52
3.05
2.77

0.00
0.00
0.00
0.00
0.00
0.00

1.27
1.44
1.29
1.31
1.22
1.23

6.45
9.29
6.28
7.05
5.11
5.23

2
426.33
4
438.17
14
2
409.58
4
401.92
18
2
399.06
4
396.68
a
PED: percentage error deviation.

10.75
10.55
10.56
10.35
9.27
9.96

0.02
0.00
0.00
0.00
0.00
0.00

1.57
1.69
1.31
1.45
1.23
1.26

7.14
8.51
6.96
7.13
6.28
8.31

14
18
10

10
14
18

20

10
14
18

30

10

[(Tmax by heuristic Tmax by B&B algorithm)=Tmax by B&B algorithm] 100. It is observed from
the tables that problems with smaller C value produce smaller PED and spend less computation
time. Also, it is found that problems with larger T value and smaller t value produce smaller
PED. This is because in this case a smaller number of maintenance periods is generated, and
hence a smaller number of potential positions is produced. This results in a Tmax value of the
heuristic closer to the optimal value. Moreover, the PED increases slightly as the number of jobs
increases. The average PED is only 0.725 with a maximum of 1.69. It is seen that the heuristic
spends much less computation time than the B&B algorithm. Therefore, the heuristic is applicable
for large-sized problems, while the B&B algorithm can only be used for small-to-medium-sized
problems.

1346

C.J. Liao, W.J. Chen / Computers & Operations Research 30 (2003) 1335 1347

Table 10
The computational results (Data Set IV)
n

10

B&B algorithm

Heuristic

PEDa

Comp. time (s)

Comp. time (s)

Min.

Mean

Max.

2
4
2
4
2
4

0.07
0.09
0.06
0.06
0.05
0.06

0.10
0.11
0.07
0.08
0.06
0.06

0.00
0.00
0.00
0.00
0.00
0.00

0.01
0.01
0.00
0.00
0.00
0.00

0.06
0.09
0.00
0.00
0.00
0.00

2
4
2
4
2
4

5.89
5.74
5.24
5.41
4.23
4.47

1.06
1.14
0.82
0.88
0.65
0.61

0.00
0.00
0.00
0.00
0.00
0.00

0.83
1.01
0.79
0.77
0.56
0.61

7.11
6.80
5.56
7.79
5.64
5.07

2
4
2
4
2
4

70.09
72.13
69.40
70.65
58.47
63.16

3.98
4.03
3.17
3.64
2.82
2.77

0.00
0.00
0.00
0.00
0.00
0.00

1.19
1.30
1.21
1.26
1.07
1.26

5.87
9.76
6.13
7.82
5.81
6.39

2
441.15
4
456.07
14
2
426.39
4
437.90
18
2
402.55
4
394.26
a
PED: percentage error deviation.

11.25
10.79
10.43
10.22
9.58
10.26

0.00
0.02
0.00
0.00
0.00
0.00

1.42
1.56
1.31
1.35
1.24
1.30

7.38
9.11
6.72
9.61
6.58
8.19

14
18
10

10
14
18

20

10
14
18

30

10

7. Conclusions
The importance of maintenance has been gradually recognized by the decision maker. Therefore,
it becomes a common practice to schedule maintenance periodically in many manufacturing systems.
Unfortunately, most papers discussing maintenance assume there is only one maintenance period. In
this paper, periodic maintenance that consists of several maintenance periods has been addressed.
In particular, we have addressed a single-machine scheduling problem with periodic maintenance
and nonresumable jobs. The objective is to minimize the maximum tardiness subject to periodic
maintenance and nonresumable jobs.
An e8cient heuristic has been proposed to provide the near-optimal solution for the problem. The
performance of the heuristic has been evaluated by comparing its solution with the optimal solution
derived by the developed branch-and-bound algorithm. Several properties associated with the problem
have also been investigated and implemented in the algorithm. Computational results have shown

C.J. Liao, W.J. Chen / Computers & Operations Research 30 (2003) 1335 1347

1347

that both the heuristic and branch-and-bound algorithm perform satisfactorily. Direct application of
the results of this study to those companies where maintenance is performed periodically is easy
and worthwhile.
References
[1] Art RHPM, Knapp GM, Lawrence MJ. Some aspects of measuring maintenance performance in the process industry.
Journal of Quality in Maintenance Engineering 1998;4:611.
[2] Adiri I, Frostig E, Rinnooy Kan AHG. Scheduling on a single machine with a single breakdown to minimize
stochastically the number of tardy jobs. Naval Research Logistics 1991;38:26171.
[3] Mosheiov G. Minimizing the sum of job completion times on capacitated parallel machines. Mathematical and
Computer Modeling 1994;20:919.
[4] Lee CY, Liman SD. Capacitated two-parallel machine scheduling to minimize sum of job completion time. Discrete
Applied Mathematics 1993;41:21122.
[5] Lee CY. Parallel machines scheduling with non-simultaneous machine available time. Discrete Applied Mathematics
1991;30:5361.
[6] Lee CY. Machine scheduling with an availability constraint. Journal of Global Optimization 1996;9:395416.
[7] Lee CY. Minimizing the makespan in the two-machine Qowshop scheduling problem with an availability constraint.
Operations Research Letters 1997;20:12939.
[8] Lee CY. Two-machine Qowshop scheduling with availability constraints. European Journal of Operational Research
1999;114:4209.
[9] Knodel W. A bin packing algorithm with complexity O(n log n) in the stochastic limit. Mathematical Foundations of
Computer Science 1981;118:36978.

Ching-Jong Liao is a Professor of Industrial Management at National Taiwan University of Science and Technology.
He holds the Ph.D. in Industrial Engineering from the Pennsylvania State University. His current research interests include
production scheduling and inventory control, on which he has published several papers.
Wen-Jinn Chen is a Ph.D. student at the Department of Industrial Management, National Taiwan University of Science
and Technology. He currently is an instructor of Industrial Management at Oriental Institute of Technology. His current
research interests are scheduling and production management.

You might also like