SCHEDULING IN FLEXIBLE MANUFACTURING SYSTEMS

Alberto Gómez
Dpto. de Admon. De Empresas y Contabilidad, University of Oviedo, Spain.
agomez@.estiig.uniovi.es
David de la Fuente, José Parreño and Javier Puente
Dpto. Admon. De Empresas y Contabilidad, University of Oviedo, Spain.
david@.estiig.uniovi.es, parreno@.estiig.uniovi.es, jpuente@.estiig.uniovi.es

Abstract
One of the major problems confronting the search for efficiency in flexible
manufacturing systems is the problem of loading and scheduling. This abstract
presents a new method to solve such problems. The methodology, which is based on
the use of genetic algorithms, allows simple codification of problems and thus
facilitates its use. Two kinds of solution are proposed – that is, loading and
scheduling either jointly or separately – and comparisons are drawn to see which
offers the best solution.

1. Introduction
Flexible Manufacturing Systems (FMS) arose from the requirement to come up with a more
profitable system for manufacturing job lots. Hard automation, which can efficiently churn out
thousands or millions of identical pieces, is not the answer to the demands of the modern world,
where more flexible systems that can handle large numbers and different quantities of articles is
needed.
In the 70s, on-line computers and numerical control techniques made flexible automation
possible; thanks to this step forward, FMS began to become widely used in the electronic,
household electrical appliance, farming and construction components, and above all in the
machine tools industries [5].
This article is structured in the following way. The second section describes FMS’s. The third
section is an introduction to Genetic Algorithms (GA) and the operators that can be used to solve
the problem. The fourth part describes the problem at hand, setting forth the initial hypotheses and
the solution that is applied. The fifth part provides a set of experimental results which allow both
an evaluation of the two suggested versions and an estimation of their system parameters to be
carried out. Finally, section six draws some conclusions.

2. Fms And Scheduling

The simplest definition of an FMS is one that views it as a central-computer-controlled system
linked to several computerised centres or work stations with an automatic materials handling
system [8].
A Flexible Manufacturing Cell (FMC) is made up of at least one element from each of the
following four groups:
• Either specialised or universal machine tools, with automatic tool change.
• Computer-assisted, automated transport of materials along fixed or moving
components of all the pieces and tools in the process.
• A work in process (WIP) area for the products being processed, with automatic
loading and unloading.
 • Computer-controlled sequencing of operations, rules and priorities, flow of
materials, types and quality of components, through a hierarchical system with
different levels of decentralisation.
A FMC basically works as follows. Machine operators transport the raw material for a group
of articles to the loading and unloading areas (work-in-process storage), after which the FMC
starts its work. Following instructions from a central computer, the transport system starts to move
materials towards the different work centres. Articles and pieces are transported according to a
given operating sequence. As can easily be deduced, the key to optimum working of the system is
the computer, and production scheduling is particularly critical.
These problems are even greater for FMS’s, which are made up of two or more FMC’s, where
each of the cells is independent but receives its jobs from a common queue. Thus, the problem of
loading is compounded to the problem of scheduling, as components get to the common queue and
the decision as to which cell they are assigned to has to be taken.

3. Description Of The Problem And Proposition Of Models

This study will deal with an FMS loading and scheduling process. This problem has been
analysed by other authors, as Roh et al. [7] who studied the loading and scheduling problems in
flexible :manufacturing systems (FMSs) in which each part visits only one machine for its entire
processing and Drake et al. [3], they studied this problem using simulation techniques. The aim of
the study is to analyse the possibilities that GA’s provide to solve the problem, not to compare
GA’s with other methodologies, as this approach is considered to have been already sufficiently
dealt with in the literature [4]. The genetic algorithm designed to solve these problems should
carry out the following activities:
• Assigning work to a cell.
• Scheduling work within each cell.
Logically, the whole scope of the problem cannot be dealt with, as the number of possible
combinations is practically infinite. For this reason, a series of restrictions were imposed. This
actually approximates the theoretical approach to industrial reality, as many of the restrictions are
imposed by the working mode of the machines. The following are considered as initial hypotheses:
• Operations of a job cannot be spread amongst several cells.
• At least one cell can carry out all the operations of each job.
• Jobs can be done by more than one cell.
• No cell has more than one machine of the same type.
• Cells need not have all the types of machines.
• All machines of the same type have identical characteristics.
• The scheduling problem is a job shop problems type.
The first solution we designed had two independent GA’s. Loading of each cell was calculated
by a GA, which balanced jobs out between different cells (sum of processing time on the
machines); another GA then proceeded to schedule the jobs independently in each of the cells
(using permutation with repetition [2]). But Concurring with the queue models formulated by
Stecke and Solberg[8], results were considered to be much better providing one managed to get
both loading and scheduling to evolve jointly.
To achieve this joint evolution we propose a structure made up of two GA’s, which will evolve
in synchronisation. With the first, loading is carried out (this algorithm will be called GALoad)
and the second does the scheduling (GASchedu). Each first GA individual (GALoad) is united in a
bijective way to another individual in the second GA (GASchedu). The algorithm proposed
calculates the maximum finishing time, this value acting as the system’s fitness. As a function of
it, the selection and reproduction process is carried out in both GA’s. The following GA design
characteristics should be noted for the GA’s that we have proposed:
 The loading-evaluation algorithm (GALoad) uses decimal codification; each gene of the
individual represents a job, and its value indicates which cell carries out the task (one must check
that the cell the job is assigned to is actually capable of doing the task). As regards the question of
the GA’s’ main characteristics, it should be stressed that the sampling mechanism used was the
stochastic universal sampling, the reproduction operators are uniform mutation (the gene that has
to mutate should select another cell that can carry out the job), and crossover is simple.
So-called permutation with repetition is used for codification in the second GA (GASchedu)
and the reproduction operators used are order-based mutation [2], and Generalized Order
Crossover (GOX) [1][9]. This codification is used because of the excellent solutions it provides
through only working with feasible solutions.
The following process was used to calculate fitness. First, a GALoad individual is analysed
and jobs are assigned to its cells. With this information, the GASchedu individual associated to the
GALoad chromosome is analysed. Thanks to the manufacturing sequence provided by this second
individual, the job finishing time for each job carried out by each cell is calculated. The longest
time is the system’s fitness, and the proposed algorithm should try to minimise it.
An example of how the joint loading and scheduling methods works. The aim of this example
is to show the joint functioning of both algorithms (GALoad and GASchedu) clearly. Let us
suppose that three jobs are wanted, and the first has two operations, the second has three
operations and the third has two operations. One possible codification of a GALoad individual is
(1 2 1), which means that the first job is done in cell one, the second in cell 2, and the third in cell
1.
The GASchedu individual associated to the above chromosome has for example the following
codification (1 2 1 3 2 2 3). This means that the “1” that occupies the first position of the
individual indicates that the first operation of job “1” is carried out in cell “1”; the “2” in second
position indicates that the first operation of job “2” is carried out in cell “2”; the “1” in third
position indicates that the second operation of job “1” is carried out in cell “1”, and so on and so
forth.

4. Experiments Carried Out

Once the new algorithm had been designed, which was a basic aim of this study, it was
decided that experiments should be carried out to calculate optimum values that GA parameters
should have for ideal functioning. An FMS configuration which the experiments could be carried
out on was required to get these values. An FMS structure made up of four FMC was decided
upon. Each of them had a maximum of three machines, and twenty jobs were put into the system.
A series of experiments were carried out using this FMS. Once the optimum GA parameters
were obtained, the initial hypothesis, - that the algorithm works better by considering the loading
and scheduling system as a whole - based on the work of Stecke and Solberg [8], was considered
to need testing. Several experiments, some of which are shown in table 3, were carried out to
contrast this hypothesis. The meaning of the columns is:
The “Method” column shows whether the experiment was done with load and scheduling
evolving separately (part) or jointly (jointly).
The “Cell”, “Job”, and “Machine” columns respectively show the cells, jobs and the maximum
number of machines forming part of the FMS being studied.
Tabla 1. A comparison of different types of problem
Methods Generation Cell Jobs Machine Total
Jointly 200 4 20 3 206
Part 200 4 20 3 250
Jointly 1000 4 20 3 204
Part 1000 4 20 3 250
Jointly 200 4 50 3 572
Part 200 4 50 3 927
Jointly 1000 4 50 3 526
Part 1000 4 50 3 927
Jointly 200 7 50 6 564
Part 200 7 50 6 649
Jointly 1000 7 50 6 488
Part 1000 7 50 6 640

The “Total” column indicates the moment when the FMS finishes all its jobs.
All experiments were carried out with the same population (100), the same crossover (0.8) and
mutation probability (0.05). Table 1 shows that the solutions provided by the system that jointly
loading and scheduling are better than the solutions provided by the separate system. This
approach has also been tested for large size problems. Results were similar (the joint model is
best); results can be considered good and calculation speed continues to be acceptable.

5. Conclusions
This text is a study and description of the application of genetic algorithms to the loading and
scheduling problem in an FMS. It has been empirically demonstrated that better results are
obtained by applying this metaheuristic if the loading and scheduling problems are studied jointly
rather than individually. The text also investigated which combination of GA parameters was ideal
for solving the loading and scheduling problem in an FMS.

References
[1] Bierwirth, C (1995). “A generalized permutation approach to job-shop scheduling with
genetic algorithms”. Or Spektrum 17(2-3), 87-92.
[2] Davis, L. (1991). Handbook of Genetic Algorithms. Van Nostrand Reinhold, New York,
[3] Drake, G.R., Smith, J.S. and Peters, B.A. (1995), "Simulation as a Planning and Scheduling
Tool for Flexible Manufacturing Systems," Proceedings of the 1995 Winter Simulation
Conference, Washington, DC.
[4] Goldberg. D. (1989). Genetic Algorithms in Search, Optimization & Machine Learning.
Addison-Wesley Co., Inc, Reading, MA.
[5] Kouvelis, P. (1992). “Design and planning problems in flexible manufacturing systems: a
critical review”. Journal of Intelligent Manufacturing, 3, 75-99.
[6] Parrish, D. (1993). Flexible Manufacturing. Butterworth-Heinemann Ltd.
[7] Roh HK, Kim YD (1997). Due-date based loading and scheduling methods for a flexible
manufacturing system with an automatic tool transporter International Journal Of Production
Research ,35: (11) 2989-3003.
[8] Stecke, K.E. and Solberg, J. (1982). The optimality of unbalanced workloads and machine
group sizes for flexible manufacturing systems. WP 290, Graduate School of Business
Administration, The University of Michigan.
[9] Varela, R.; Vela, C.; Puente, J.; Gómez, A. and Vidal, A. (2001). Solving Job-Shop
Scheduling Problems by Means of Genetic Algorithms. In: The practical handbook of Genetic
algorithms. Applications. Ed.: Chapman & Hall/CRC. pp. 275-293.