Int j simul model 11 (2012) 3, 129-140
ISSN 1726-4529 Original scientific paper
DOI:10.2507/ I JSI MM11(3)2.201 129
COMPARISON OF DISPATCHING RULES IN JOB-SHOP
SCHEDULING
PROBLEM USING SIMULATION: A CASE STUDY
Kaban, A. K.
*
; Othman, Z.
*
& Rohmah, D. S.
**
*
School of Information Technology, Faculty of Information Science and Technology,
National University of Malaysia, 43600 Bangi, Selangor Darul Ehsan, Malaysia
**
Department of Industrial Engineering, Langlangbuana University,
40261 Bandung, Karapitan Street 116, Indonesia
E-Mail: aricko.khena@msn.com, zalinda@ftsm.ukm.my, de2_siti_rohmah@yahoo.com
Abstract
Industries nowadays must be able to quickly adapt with the customer and improve product
quality to survive in the competitive edge. Job shop scheduling is crucial in the manufacturing
world and exists within most manufacturing sectors. In the manufacturing world, scheduling
problems are extensively implementing the dispatching rules. The procedures are designed to
provide good solutions to complex problems in real-time. This paper describes the importance
of dispatching rules in improving the performance of the factory. This study evaluates total
of44dispatching rules with the classification of hybrid and single rules. The performance of
each rule compared and summarized to determine the final ranking for all the different
dispatching rules. The result shown that MTWR (Most Total Work Remaining) rule performs
well in almost all measurements as well as hybrid dispatching rules is not generating the best
rules compared to single dispatching rule. A set of data from an automotive industry use to
simulate the job-shop production floor.
(Received in August 2011, accepted in March 2012. This paper was with the authors 2 months for 2 revisions.)
Key Words: Job Shop, Dispatching Rules, Simulation, ARENA
1. INTRODUCTION
Nowadays, industries have to face the intensified global competition and advance in the field
of information and technology. Manufacturing cycle, quality and service are the major
concerns in the manufacturing industries for them to survive in the marketplace. They must be
able to quickly adapt to their customers and improve product quality. Fast responding to rapid
changes in technology, demand fluctuations, and design changes are also important. These
factors require companies to emphasize on automated systems to improve productivity and
quality, while reducing cost simultaneously. Companies that are not being able to revise their
strategies and, accordingly, modify their organizational processes, will face a risk to eliminate
from the competitive edge [1].
Variations of production control techniques applied in order to increase the total
production, reduce the total time completion, and deliver the product on time. One method to
increase the production of an industry is to create proper scheduling for the components on
the available machines so that the order will complete on time, maximizing the use of the
resources and minimizing the average waiting time [2]. Scheduling exists in most
manufacturing and production systems, as well as in most information-processing
environments. In production management, scheduling plays a vital role that is important to
ensure the production system runs orderly and explores its potential capacity [3].
Job shop scheduling is extremely difficult to make both in practice and in theory. It is
based on the fact that so many parameters to be considered. There are several approaches to
Kaban, Othman, Rohmah: Comparison of Dispatching Rules in Job-Shop Scheduling …
130
scheduling problems, such as analytical techniques, meta-heuristic algorithms, rule-based
approach and simulation approach. Traditional analytical techniques and simple mathematical
models are currently inadequate to analyse the complex manufacturing environments. In
addition, analytical models often use mathematical programming techniques and it is not
practical for solving a complex scheduling problem. Simulation is one of the powerful tools
for testing the efficiencies of different scheduling policies [4]. It can simulate a long period in
real life within a reasonably short computer running time of several seconds or minutes. This
saves many long-time observation costs. Moreover, simulation can help to get the result of
future time without any real change to machine layout or the amount of machines [5].
2. DISPATCHING RULES
Scheduling provides a basis for assigning jobs to a work centre. Sequencing (also referred to
as dispatching) specifies the order in which jobs should be complete at each centre. The
sequencing methods referred to as priority rules for sequencing or dispatching jobs to a work
centre. In the manufacturing world, scheduling problems are extensively implementing the
dispatching rules. The procedures designed to provide good solutions to complex problems in
a real-time production environment [6]. Most of the previous researchers, until this current
time, are using dispatching rules to optimize the job-shop scheduling problem [6], [7], [9],
and [11]. Impacts generated by the dispatching procedure in the queuing networks are very
difficult to be explained using analytical techniques [8]. However, the study of the dynamic
job shop scheduling has made rapid progress by using computer simulation. In these
approaches, several dispatching rules or other scheduling policies are compared using
simulation [9].
Reference [10] classified over 100 scheduling rules and attempted to explain the general
idea behind different rules. These rules classified into static and dynamic rules. Static rules
are the ones in which the job priority values do not change as a function of the passage of
time, i.e. it is not time dependent. They are just a function of a job and/or machine data.
Dynamic rules are time dependent. Reference [6] used dispatching rules in semiconductor
manufacturing and it showed that dispatching heuristics provides schedules quickly, that are
easy to understand, easy to apply, and require relatively small computation time. The primary
disadvantage of dispatching rules is that these cannot hope for optimal solutions for all
performance measures in the dynamic job shop [11]. It is because the dispatching decision
made only according to local information.
References [11-12] designed dispatching rules to improve the tardiness performance. They
proposed a new dispatching rule (i.e. RTSLACK), which is based on maximizing the slack
time of the remaining tasks in the manufacturing resources queues in a series of single
machine and hybrid flow shop scheduling problem instances. Reference [12] designed an
effective composite dispatching rule that minimizes total tardiness through a Genetic
Programming approach in a flexible job-shop model. Their research implied that the way to
combine the rules could significantly affect the optimality of the schedules.
Dispatching rules are better than genetic algorithms in three respects [13]. They found that
dispatching rules are able to create various solutions to solve many problems observed,
whereas genetic algorithms only provide one solution to minimize makespan. In addition,
solutions obtained by genetic algorithms yielded scattering results, whereas the solutions
obtained by dispatching rules yielded steady results. Thirdly, genetic algorithms require the
use of a computer because of the large number of parameters to specify, whereas dispatching
rules can obtain simple solutions in an urgent production situation.
Reference [11] combining several single dispatching rules to provide efficient dispatching
rules for dynamic job shop scheduling and they found that no single rule is effective in
Kaban, Othman, Rohmah: Comparison of Dispatching Rules in Job-Shop Scheduling …
131
minimizing all measures of performance. Reference [14] investigated the advantages of using
a combination of dispatching rules for cost performance and they found that the combination
rules performs well in reducing both mean and variability of waiting cost. Reference [15]
proposed a dispatching rule for non-identical parallel machines that considered product
quality, it enabled job shops to keep due dates, while satisfying quality restrictions. The use of
a quality threshold in dispatching can facilitate manufacturing products with a desired quality
level [16-17].
From the literature reviewed, Dispatching heuristic was able to provide not only a good
solution but also the best solutions for the system observed. Dispatching rules have a
significant role within the dynamic context because of their ease of implementation and
compatibility with the dynamic nature of manufacturing systems.
3. PROBLEM STATEMENT
Job shops are an important part in the world of manufacturing. Peoples are not able to
maintain their living standards normally without it [18]. The definition of job shop is “a group
of manufacturing operations where the productive resources are organised according to
function and the work passes through in varying lots and routings” [19].
This paper can state the problem addressed as follows: Given a large job shop and a
number of jobs that consist of local disturbances, we determine how to schedule the jobs, so
that the performance observed is maximum. The job shop problem described as [20-21]: there
are m machines with n operations and j jobs. Therefore, there will be m
n
possible for the
allocation rules, (n!) possible sequences, (m
n .
n!) possible processes, ((j!)
m
) schedules
available for each job, and (m
n .
n!)
j
schedules available for each combination of processes.
These give ((m
n .
n!)
j .
(j!)
m
) schedules evaluated for each job-shop scheduling problem. Let
say if m = n = j = 2, there will be 256 possible numbers of schedules, but if m = n = j
increase to the number of 3, there will be 91833048 possible number of schedules and these
numbers will rapidly increase in proportion to increment of resources.
3.1 Design of experiment
This paper describes a simulation study for automotive supply industry with job shop
environment. This company produces components for vehicle and several industrial
components. In observed company, 10 operations/products produced on 14 machines. Routing
and processing time are dependent on the group of items. Routing length can vary from three
to seven operations and there is no flexibility on the routing. The following assumptions made
while built the simulation model [12], [22]:
• All the items are ready at the start of the simulation.
• No due dates specified.
• Each machine can perform only one operation at a time.
• Each machine has an operator.
• Transportation time between two consecutive work centres is deterministic and assumed to
be 15 time units for all shop types.
• Machines have breakdowns every 3 months, and it take a whole day to repair.
• Set-up times assumed to be negligible.
• All the parts have to leave the system when all their operations are finished.
ARENA simulation software is used to develop the simulation model for the job-shop
scheduling problems. ARENA has many advantages in simulation and modelling of discrete
system. The simulation model executed for one simulation year based on 8 hours/day and 5
Kaban, Othman, Rohmah: Comparison of Dispatching Rules in Job-Shop Scheduling …
132
days/week. Therefore, the total run for a year is 250 days (include public holiday) with n = 5
replications.
3.2 Performance measurement
In this study, simulation experiments collect several measurements of shop performances. The
objective is to evaluate the performance of different dispatching rules based on these
performance measures in real job shop scheduling environment. Several performance
measures used to evaluate which alternative that performs well in observed system. The
performance measures collected are:
• WIP average
It is the average number of jobs in the system. The relationship between the number of jobs
in the system and the WIP inventory will be high. Therefore, the fewer numbers of jobs are
in the system, the lower the inventory.
• The total average time that required to complete an operation
One of the common objectives in job shop scheduling is to minimize the makespan. The
makespan optimization generally ensures high utilization of the production resources, early
satisfaction of the customers’ demands and the reduction of in-process inventory by
minimizing the total production run.
• Total waiting time average for each part
Waiting time is the time that the entity (parts) spends waiting in the system for available
machine and operator.
• Queue waiting average time
Queue time is the time that part i spent waiting in the machine j queue.
• Queue length average for each work centre
Queue length is the number of parts that are waiting in the machine queue.
4. SIMULATION EXPERIMENT
The base simulation model was used FIFO as initial dispatching rule. In this project, beside
LIFO rule, static and dynamic dispatching rules are also use to enhance the plant
performance. Static rule are the ones in which the job priority values does not change as a
function of the passage of time, it is not time dependent. There are 3 parameters that assigned
to be static:
• Processing Time: Time required to operate part i on machine j.
• Process Sequence: Total sequence required for part i to complete the operation.
• Total Processing Time: Total time required for part i to complete the operation.
Dynamic rules are time dependent. It is the ones in which the job priority values change as
a function of the passage of time. There are also 3 parameters that assigned to be dynamic in
this project:
• Creation Time: The time when part i was created.
• Waiting Time: The time that part i spent when waiting in the queue line on machine j.
• Total Work Remaining: Total work remaining for part i to complete the operation.
Based on parameters described, 14 dispatching rules adopted in this project as shown in
Table I.
Up to the current knowledge, there is no single dispatching rule that minimizes most of
the performance measures [11], particularly in the dynamic environment of job shop
scheduling. Therefore, a new dispatching rule is carry out in this study by using a
multiplicative combination of rules that minimize most of the performance measures. Each of
Kaban, Othman, Rohmah: Comparison of Dispatching Rules in Job-Shop Scheduling …
133
parameter in a single rule multiplied to obtain a new hybrid dispatching rules. For example
SPT and SPS:
Z = PT
ij
× PS
i
(1)
where,
i – Set of operation
j – Set of Machine
PT – Time required to operates operation i on machine j
PS – Total sequence require for operation i to complete the operation
Table I: Selected dispatching rules.
No.
Rules
Description
Type
1
FIFO
First In First Out
Static
2
LIFO
Last In First Out
Static
3
SPT
Shortest Processing Time
Static
4
LPT
Longest Processing Time
Static
5
SPS
Shortest Process Sequence
Static
6
LPS
Longest Process Sequence
Static
7
STPT
Shortest Total Processing Time
Static
8
LTPT
Longest Total Processing Time
Static
9
ECT
Earliest Creation Time
Dynamic
10
LCT
Longest Creation Time
Dynamic
11
SWT
Shortest Waiting Time
Dynamic
12
LWT
Longest Waiting Time
Dynamic
13
LTWR
Least Total Work Remaining
Dynamic
14
MTWR
Most Total Work Remaining
Dynamic
The rule that give the lowest Z value has the highest priority to be processed. Table II
shows the experiment for this stage. It is the combination between each parameters used in
this paper. Fifteen (15) combined parameters generated, each of these parameters are set to
high attribute and low attribute value. Therefore, this study will evaluate the 30 hybrid
dispatching rules.
Table II: Combined dispatching rules.
Parameter
PT
PS
TPT
CT
WT
TWR
PT
X
√
√
√
√
√
PS
X
X
√
√
√
√
TPT
X
X
X
√
√
√
CT
X
X
X
X
√
√
WT
X
X
X
X
X
√
TWR
X
X
X
X
X
X
PT = Processing Time. WT = Waiting Time.
PS
= Process Sequence. TWR = Total Work Remaining.
TPT
= Total Processing Time. CT = Creation Time.
5. RESULTS AND DISCUSSION
The results of simulation models are analysed and discussed in this section. In this simulation,
a confidence interval of 0.95 has been set to provide insight to the output variability. For