A MODEL OF JOB SHOP SCHEDULING TO TACKLE DYNAMIC EVENTS THROUGH SIMULATION
Job shop scheduling (JSS) is an important component of manufacturing systems. Traditional JSS models assume static conditions: (1) all the jobs are present at time zero, (2) the attributes of jobs such as release dates (rj ) are known, (3) machines remain always available without breaking down...
Saved in:
Main Author: | |
---|---|
Format: | Dissertations |
Language: | Indonesia |
Online Access: | https://digilib.itb.ac.id/gdl/view/86689 |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Institution: | Institut Teknologi Bandung |
Language: | Indonesia |
Summary: | Job shop scheduling (JSS) is an important component of manufacturing systems.
Traditional JSS models assume static conditions: (1) all the jobs are present at time
zero, (2) the attributes of jobs such as release dates (rj
) are known, (3) machines
remain always available without breaking down or need for maintenance, (4) no
new job arrives during production, and so on. However, these conditions are very
restrictive in practical manufacturing scenario. In practical, the JSS is not static but
rather highly dynamic. The JSS experiences numerous dynamic events which
change the initial state of the system, induce system nervousness, reduce
manufacturing efficiency, and in extreme cases, make the schedule infeasible. Upon
the occurrence of such dynamic events, rescheduling gains theoretical and practical
importance, adding to the further complexity to the JSS.
Numerous studies are dedicated to address these challenges with various solution
approaches. These solution approaches typically focus on exact method, which
although provides optimal solutions for small scale JSS instances, struggles
computationally to generate optimal solutions for medium-to-large scale JSS
instances, or on heuristic approaches, which although provide quicker solutions for
small-to-large scale JSS instances, the solutions obtained are not optimal.
Moreover, there is a growing demand for JSS system that not only manage the
scheduling process under dynamic events but also provide monitoring, control, and
prediction capabilities to address these dynamic aspects of JSS.
To tackle these challenges, our research develops a comprehensive series of
approaches combining the strengths of exact method, greedy algorithm (GrA), and
Greedy Randomized Adaptive Search Procedure (GRASP) algorithm. We start with
the exact method to benchmark the solutions for comparing it with the other
approximate methods. Since the computational limitation of exact method does not
allow us to get the optimal solutions for medium-to-large scale JSS instances, so
only solutions for small-scale JSS instances are used as a benchmark. Next, a
greedy algorithm (GrA) is employed: (1) to scale the problem size for small-to-large
scale JSS instances, and (2) to obtain quick solutions. While GrA is
computationally efficient, it does not always provide optimal solutions,
necessitating the need for a more advanced approach.
The core of our solution approach is a novel GRASP algorithm. This algorithm is
designed to balance the quality of the solution with computational effectiveness.
Our GRASP approach incorporates two strategies: intensification and
diversification. In the intensification strategy: (1) the algorithm systematically
analyzes the tardiness of each job, (2) compute the tardiness contribution of each
operation to the total tardiness, (3) list operations based on their tardiness
contribution value, and (4) apply a directed swapping procedure based on the
tardiness contribution value. For the directed swapping procedure, the operation
with the significant tardiness contribution value swaps the operation with its
predecessor, successor, and other operations on the same machine such that the
precedence relationship and dependency relationship among operations are still
respected. If no further improvements using the intensification strategy can be
made, we employ the diversification strategy. Our diversification strategy
incorporates two techniques: (1) random swapping, and (2) random restarting.
Through novel GRASP, the quality of the solution improves significantly while
maintaining computational efficiency. For example, the percentage gap of 13.82%
for GrA and the exact method is reduced to just 3.43% using the GRASP algorithm.
Additionally, GRASP’s computational time is significantly lower than exact
methods, while being comparable to GrA.
In addressing dynamic events, our study categorized them into three types based on
the impact they make on scheduling process: (1) low impact (LiDE), (2) medium
impact (MiDE), and (3) serious impact (SiDE). Our dissertation focuses on SiDE
as they significantly impact the scheduling processing by inducing system
nervousness and minimizing the manufacturing efficiency. These events include
new jobs arrival (NJA), rush orders (RO), machine failure (MF), and scheduled
machine maintenance (SMM).
A comprehensive rescheduling procedure is developed to tackle SiDE. A proactive-
reactive (Pact ? Ract) rescheduling strategy is used to minimize the impact of SiDE
regularly through Pact and on immediate basis through Ract . To define the
rescheduling trigger, a hybrid (periodic and event-driven) (Hyb.) rescheduling
policy is used, which triggers periodic (pe
) rescheduling for (Pact) to tackle NJA
and SMM, and event-driven (ED) rescheduling for Ract rescheduling to minimize
the impact of MF and RO. To rescheduling jobs upon the rescheduling triggers, a
right-shift (RF) rescheduling method is used for MF and a regeneration (Reg)
rescheduling method is used for NJA, RO, and SMM. This rescheduling procedure
is adaptable, minimizing the total tardiness of jobs. Our results demonstrate that the
Ract rescheduling strategy performs better in minimizing tardiness than Pact
rescheduling strategy.
To further enhance system’s ability to handle SiDE and provide monitoring,
control, and prediction capabilities, we integrate simulation models based on the
concept of Digital Twin (DT) technology. A DT is a virtual replica of the physical
production system, enabling us to simulate different scenarios and predict the
outcomes of various scheduling strategies. Two types of simulation models are
developed in this context: (1) simulation models considering priority rules (FIFO,
LIFO, SPT, SRPT, EDD, and CR), and (2) simulation models considering SiDE.
iii
Through these simulation models, we assess the performance of JSS under different
varying conditions. Most importantly, our simulation models considering SiDE
demonstrate that the combination of Pact and Ract rescheduling strategies
effectively manages SiDE , providing a feasible solution for JSS in dynamic
environments, while providing the monitoring, control, and prediction capabilities.
In conclusion, our research addresses key gaps in JSS by developing a
comprehensive series of solution approaches that integrate exact method, GrA,
GRASP and simulation techniques. Our novel GRASP offers an effective balance
between solution quality and computational efficiency, while our rescheduling
procedures and DT simulations provide practical tools for managing practical
manufacturing disruptions. This approach not only advance the theoretical
understanding of JSS but also offer practical solutions. |
---|