Simulation Modelling Practice and Theory 11 (2003) 2144

www.elsevier.com/locate/simpat

A model for manufacturing systems

simulation with a control dimension
Georges Habchi *, Claire Berchet
Ecole Sup. dIngenieurs dAnnecy, ESIA, LLP/CESALP, 41, avenue de la Plaine BP 806,
74016 Annecy Cedex, France
Received 1 September 2001; received in revised form 1 April 2002

Abstract
The objective of this article is related to the potential improvement of computer simulation
as applied to manufacturing systems. Through our contacts with the operational environment,
we have observed that simulation is not used to its full potential. One remark is that existing
tools are not adapted to modelling the decision process: they fall short of oering eective integration into the control process of production. Control is usually limited to scheduling and
does not lend itself to practical application. In order to enhance the capabilities of computer
simulation and make it more responsive to todays industrial needs, we present a way of introducing such control into simulation by pursuing generic and applicable concepts. The core
concepts that constitute the framework of our research are a global structure supporting
the co-ordination and co-operation relations; a local structure presenting a typology of industrial control adapted to our needs; a control centre, the main concept used to introduce control
into simulation. The modelling language used is UML and the model is implemented using the
object-oriented language JAVA. An industrial application was carried out in the company Alcatel with the help of the Apollo platform.

2002 Elsevier Science B.V. All rights reserved.
Keywords: Manufacturing systems; Industrial control; Decision-making; Modelling and simulation

Corresponding author.
E-mail addresses: georges.habchi@univ-savoie.fr (G. Habchi), claire.berchet@an.cit.alcatel.fr (C.
Berchet).
1569-190X/03/$ - see front matter
2002 Elsevier Science B.V. All rights reserved.
doi:10.1016/S1569-190X(02)00097-7

22

G. Habchi, C. Berchet / Simulation Modelling Practice and Theory 11 (2003) 2144

1. Introduction
Simulation is widely used in the world and therefore it is very familiar [12]. The
most important reasons and advantages of simulation methodology for modelling
manufacturing systems are that:
realistic models are possible, they are a practical approach to representing the important characteristics of a manufacturing system and may incorporate any complex interactions that exist between dierent variables;
options may be considered without direct system experimentation and alternative
designs can be easily evaluated, independently of the real system;
a computer simulation models ability to directly address the performance measures typically used in a real system;
non-existent systems may be modelled;
visual output helps and assists the end-user in model development and validation;
no advanced mathematics is required;
analytical methods are perceived to be unhelpful by management or may require
over-simplication.
Law and Kelton [27] summarise some reasons for the spectacular increase in the
use of simulation in the eld of manufacturing systems as follows:
automated systems are so complex they can typically be analysed only by simulation;
computing costs have been reduced by microcomputers and engineering workstations;
improvements in simulation software have reduced model development time,
thereby allowing for more timely manufacturing analyses;
the availability of animation has resulted in a greater understanding and use of
simulation by engineering managers.
The use of simulation for manufacturing systems design and analysis is rightfully
recognised by scientists and industrial managers and the literature is abundant in this
eld. We can refer to several subjects: productivity analysis [36], Just-In-Time system
design [44], comparison of two kinds of line management [9], exible hybrid assembly system analysis [40], automated overhead warehouse system description [31],
business process modelling [21]. . .
Traditionally, simulation has been used for oine decision-making. One of the
limitations of its use for online decision-making is the considerable amount of time
spent in gathering and analysing data. Consequently, this has resulted in decisionmakers relying on simulation primarily for oine decision support and not for the
critical online decision-making that may arise. In real-time control, the three key issues are data acquisition, quick response and instantaneous feedback. The major
components of online simulation systems generally consist in a data acquisition
module, a simulation model and a cell controller. Over recent years, some articles
have been published in this eld. However, most of them only concern scheduling

G. Habchi, C. Berchet / Simulation Modelling Practice and Theory 11 (2003) 2144

23

problems. At the beginning of the 1990s, online simulation was used in a work order
release mechanism for exible manufacturing systems [33]. Rogers and Flanagan [37]
developed a framework for an online simulation system for real-time scheduling.
Rogers and Gordon [38] discussed the use of discrete event simulation as a component in real-time decision-making tools for manufacturing systems control, focusing
on dynamic scheduling decisions. Jacobs and Lauer [23] proposed an operational enhancement to current job shop scheduling systems. Abdallah [1] used a knowledgebased simulation model for job shop scheduling. Some articles analysed scheduling
heuristics and rules combinations [13,15,20,30]. In more recent literature, some studies concerned loading problems [35,42], work-in-process inventory drive systems [43],
real-time scheduling of batch systems [25], local rules theory [19]. . .
However, in order to overcome some limits of simulation methodology, researchers developed hybrid approaches integrating other techniques such as intelligent simulation [11], neural networks [22,41] genetic algorithms [24], fuzzy logic [8],
experimental designs [7]. . .
The literature on manufacturing systems simulation consulted, reinforces our conviction that simulation is a technique that still has a lot of underexploited potentialities. Thus, the objective of our research is related to the potential improvement of
computer simulation as applied to manufacturing systems. Among the current limits
of simulation, existing tools fall short of oering eective integration into the control
process of production. In order to enhance the capabilities of computer simulation
and make it more responsive to todays industrial needs, the task was to nd a
way of introducing such control into simulation by pursuing generic and applicable
concepts. This is what we are proposing in this article which is organised as follows:
Section 2 introduces the potentialities and limits of manufacturing simulation; Section 3 presents a systemic analysis of manufacturing systems and proposes a conceptual approach for simulation modelling; Section 4 species the modelling approach;
Section 5 describes the UML model and implementing approach and Section 6 summarises the study breakthroughs.

2. Computer simulation potentials and limits

According to the literature overview presented in Section 1, the potentialities of
simulation are very large. Simulation can help users by contributing in design, in
management and in the decision-making of production systems. It is able to model
all kinds of company processes: physical, informational and decisional. Simulation
models can be built at all hierarchical (operational, tactical, strategic) and detailing
levels (machine, cell, shop,. . .) [2,3,10,26]. Moreover, the whole manufacturing system life cycle can be modelled and simulated (design, analysis, implementation, operation) [16,34].
Today however, not all the possibilities of computer simulation are currently used.
Computer simulation is limited in its uses and its applications. The major limits of
simulation can be explained by the fact that the operational decision-making process
and its impact on the production system are not taken into account during simulation

24

G. Habchi, C. Berchet / Simulation Modelling Practice and Theory 11 (2003) 2144

because existing tools are not adapted for decision and control modelling. Nevertheless, successful studies need models describing all types of manufacturing system
processes: operational, informational and decisional. Even if the modication of a
parameter is possible during simulation runs, it is not done according to a control
logic. A good control must imply the introduction of feedback loops around each production resource. As these loops have currently not been introduced in the models,
the action of control requires many experiments in order to optimise the results. In
addition, simulation tools use concepts that are too complex for most potential users.
Therefore the use of simulation in companies remains irregular and limited.
Simulation also focuses on the formulation and solution of problems by trialand-error methods [29]. The simulation process is iterative and often reveals important information and new insights into the problem area as a result. During this
iterative procedure, the relationship between the system under study and the model
are dened and continually redened. Simulation and modelling are therefore inextricably linked with the steps in the simulation process. Therefore, the more experienced practitioners use intuition and experience.
The following sections try to shed light on the problem of control modelling and
simulation.

3. Conceptual approach for manufacturing systems

3.1. Systemic decomposition and analysis of manufacturing systems
Assumptions for further research in industrial management must be that rigid,
static and hierarchical manufacturing systems will be replaced by systems displaying
great adaptability to rapid change and able to produce low-volume and low-cost
products [14]. Then, to design, organise and control current manufacturing systems,
managers must take the following aspects into consideration [2,3]: the diversity and
heterogeneity of production ow, production space optimisation, production process
organisation and management simplication.
A systemic analysis consists in a decomposition of a global system into a network
of subsystems, each subsystem having a specic nature and distinct objectives, with the
latter possibly being conicting. The systemic analysis aims for the following goals [28]:
synthetically describing the functioning of the considered system;
making the study of a decision consequences, an action program, a general policy,
possible by simulation;
enhancing the functioning of the system.
As manufacturing systems are by nature complex and analytic methods cannot be
always applied [18], systemic analysis is then necessary to apply simulation. Basically,
a production system is divided into three subsystems [28]: physical, informational
and decisional. Nevertheless, as simulation models are based on information, we
only considered two subsystems: operation and control (Fig. 1).

G. Habchi, C. Berchet / Simulation Modelling Practice and Theory 11 (2003) 2144

25

Fig. 1. Systemic analysis of a manufacturing system.

We considered the operation subsystem as a combination of two types of physical

objects: entities and resources. Entities represent products to be manufactured and
resources are production means, and the ow of entities on resources describes the
manufacturing process. On the other hand, we consider the control subsystem as a
set of decision-makers able to act on the manufacturing process if it is necessary using
control rules, management procedures. . . Both the manufacturing and control processes together form a rms production process.
Then, according to this analysis and in order to model and simulate the whole
production process (manufacturing and control processes), we propose the following
three conceptual objects: the moving entity (MovEnt), the product processing system
(PPS) and the control centre (CC).
3.2. The moving entity [2,3]
The MovEnt is the rst conceptual object of the operation subsystem. It is a generic concept able to model all kinds of products: an item, a part, a lot, a batch,. . .
The MovEnt has some characteristics:
a routing dening the path on resources and processing times;
a bill of materials dening an assembling process;
an objective describing the customer requirements (products and quantity to be
manufactured, date of need).
The MovEnt also has the following two properties:
it is a passive object vis-
a-vis of itself (it cannot act on itself),
it is an active object vis-
a-vis the other objects (it is the object that triggers the dynamic of the manufacturing system).
In the rest of this article, we will see that the MovEnt is a pre-programmed object
(routings, bill of materials, objectives) remote controlled during production by decision-makers.
3.3. The product processing system [2,3]
The PPS is the second conceptual object of the operation subsystem. It is a generic
object having all structural and functional characteristics of a production resource.

26

G. Habchi, C. Berchet / Simulation Modelling Practice and Theory 11 (2003) 2144

The PPS is able to model all kinds of resources (machine, stock, cell, work-centre. . .)
and presents the following three essential properties:
an atom grouping the natural ow of the three fundamental operations of a resourcereceiving, processing and supplying (Fig. 2);
it synthesises the resource and its control in a non-perturbed context at the same
time (Fig. 3);
it is a recursive structure able to develop models at dierent abstraction and hierarchical levels (Figs. 4 and 5).
The rst property enables us to build simple and concise models very close to production systems.
The second property describes a PPS standard behaviour consisting in the following three-function cycle: receiving, processing and supplying. According to this cycle,
the PPS behaviour is considered as perfect but prone to disturbances that are corrected by the CC.
The third property can be described through the two structures presented in Figs.
4 and 5. Fig. 4 presents the dierent states that a PPS could have in a given simula-

Fig. 2. Schematic representation of a PPS.

Fig. 3. PPS activity cycle according to perfect behaviour.

Fig. 4. Detailed levels and states of a PPS.

G. Habchi, C. Berchet / Simulation Modelling Practice and Theory 11 (2003) 2144

27

Fig. 5. Hierarchical levels of a PPS.

tion. These states are organised according to the level of detail needed in the model,
which is a function of the simulation objectives. This abstract decomposition allows
us to consider the control process in simulation.
Fig. 5 presents a hierarchical structure, which allows the development of models
at ve dierent levels (machine, workstation, cell, work-centre, shop). The PPS is a
recursive object because a PPS models the corresponding resource at each level and
is a function of lower level PPSs.
Generic and modular modelling properties lead designers to consider the use of
recursive structures for both operation and control subsystems. With the two objects
MovEnt and PPS that are proposed, we can build a conceptual model of the operation subsystem.
3.4. The control centre [46,17]
The CC is the third main conceptual object introducing control into simulation
modelling. We dene the CC as: an organised and autonomous structure, depending
on the company global strategy, having a decisional authority, associated with a controlled entity and having the necessary resources to apply actions and achieve the dened goals within the global framework of the company.
The CC disposes of components: decision-makers, referents, objectives, internal
information, external information, performance indicators, measures, actions, control rules, resources.
Fig. 6 presents the four main steps in the CC control process:
PPS performance evaluation consists in analysing the measure obtained from the
PPS, comparing it with the CC objective, and then concluding if a deviation exists.
The main tool used is the performance indicator (i.e. Activity rate 0.80, activity
objective P 0:90, activity rate < activity objective ! deviation exists);
cause search concerns the identication of the cause responsible for the PPS deviation. The identication is done by examining lower level performance indicators
(i.e. Failure rate 0.18, set-up rate 0.02, failure objective 6 0:07, set-up objective 6 0:03, failure rate > failure objective, set-up rate < set-up objective ! the
responsible cause is failure);

28

G. Habchi, C. Berchet / Simulation Modelling Practice and Theory 11 (2003) 2144

Fig. 6. Main steps of the CC control process.

action search consists in the identication of the action able to correct the current
deviation of the PPS and prevent future deviations. The seeking of the right action
may be done with the help of cause and eect relation, and internal simulation (i.e.
actions: preventive maintenance, reliability enhancing, quick repairing,. . .);
action applying concerns the planning and application of actions with the help of
the relevant competent resources (i.e. maintenance service).
The whole control process of a production system is dynamic, built on the basis of
dierent actions decided on by dierent decision-makers, at dierent dates and in different locations. Therefore, in order to situate the evolution environment of the CC,
we dene two structures: the global and local frames.
The global frame (Fig. 7) is a hierarchical co-ordinated structure allowing the decision organisation to progress by enhancing reactivity and co-operation from the
operation level to the highest decision level [32,39]. This frame organises the CCs
in a global environment depending on three dimensions: hierarchy, space and time.
The local frame (Fig. 8) of the control subsystem organises the control methods to
be applied by the CC. The control methods are dened according to control reactivity and are adapted to our modelling approach. The following three control methods
are proposed: the a priori control method, the anticipated reactive control method
and the a posteriori control method.
The a priori control method uses theory and models. It is an oine method applied before the system is run. The control is planned according to structural information, forecasts, orders, management rules, objectives,. . . The a priori control is
carried out using scheduling tools, simulation, analytical methods,. . . However, on
its own it is not adequate because when the system is running, it may be prone to
uncertainty, deviations and risks.
The anticipated reactive control is an online method applied directly to the system. This method is preventive and is applied using tools such as process performance indicators, control charts,. . . The use of such tools prevents deviations by

G. Habchi, C. Berchet / Simulation Modelling Practice and Theory 11 (2003) 2144

29

Fig. 7. Hierarchical global frame of the control subsystem.

Fig. 8. Local frame of the CC.

controlling the process drift. The denition of thresholds and objectives is necessary
to anticipate deviations.
The a posteriori control is also an online control method but it acts after the event
occurs (i.e. failure, absenteeism, production change,. . .). It is also more adapted for
periodic control. Result performance indicators are used.

4. Modelling approach
To illustrate the modelling approach of a PPS/CC network, we considered the example presented in Fig. 9. The production system is composed of the following resources: a stock of raw materials, a welding machine, a grinding machine and a
stock of nished products. Two operators and a storekeeper are assigned to control

30

G. Habchi, C. Berchet / Simulation Modelling Practice and Theory 11 (2003) 2144

Fig. 9. Example of a manufacturing system.

these resources. The rst operator manages the stock of raw materials as well as the
welding machine. The second operator controls the grinding machine. The storekeeper is responsible for the stock of nished products [6]. A production line supervisor manages the three persons. Then, the operation subsystem is composed of four
physical resources and products to be manufactured (P1 ; P2 ; . . . ; Pq ). And the control
subsystem is composed of four decision-makers (three at the rst hierarchical level
and one at the second).
The generic objects of PPS and MovEnt model the operation subsystem. The PPS
models the physical resources and information concerned. The MovEnt models the
products circulating on machines and their information. The CC models the control
subsystem decision-makers represented by the operators, storekeeper and supervisor.
This example may be represented using the meta-model object classes of the PPS/CC
network as indicated in Fig. 10.
The CC is directly linked to objects such as PPSs at the operation level (PPS1,
PPS2, PPS3 and PPS4 controlled by CC1, CC2 and CC3), as well as the CCs at
the rst level of the control submodel (CC1, CC2 and CC3 controlled by CC4).
The following 3-tuple formally denes a PPS/CC network (NPPS/CC):
NPPS=CC : hPPS; MovEnt; CCi

where PPS fPPS1; PPS2; . . . ; PPSng is a nite set of PPS, MovEnt fMovEnt1;
MovEnt2; . . . ; MovEntqg a nite set of MovEnt types, CC fCC1; CC2; . . . ; CCpg
a nite set of CC.
4.1. The PPS model
A PPS is dened by a 5-tuple of object class lists: the attributes list, the functions
list, the methods list, the indicators list and the events list:

Fig. 10. Meta-model classes of the PPS/CC network.

G. Habchi, C. Berchet / Simulation Modelling Practice and Theory 11 (2003) 2144

PPS : hAPPS ; FPPS ; MPPS ; IPPS ; EPPS i

31

The attributes list APPS concerns the structural information of the PPS:
APPS : hI; C; Cal; Lp ; Ts ; Tr ; Re ; Rd i

with I the PPS identier (string), C the PPS capacity dening a number of identical
resources (integer), Cal the PPS opening schedule (vector), Lp the PPS failure and
repairing distributions, Ts the PPS set-up time (positive real), Tr the PPS scrap rate
(positive real), Re the PPS rework rate (positive real), Rd the PPS eciency (positive
real).
The functions list FPPS groups variables used within the state change logic. The following 4-tuple denes the PPS functions list:
FPPS : hL; R; T ; F i

where L f0; 1; . . . ; ng is the PPS load (integer with L 6 C), R f0; 1g is the PPS
receiving function (binary), T fIdle; Inter; Satur; Bloc; PBreak; PMaint; UProd;
BDown; Repair; Supg is the PPS processing function describing the dierent states of
level 4 (string), F f0; 1g the PPS supplying function (binary).
The methods list MPPS groups the methods describing the behavioural logic of the
PPS and used in the events logic. The following 9-tuple denes this list:
MPPS : hsupply; receive; block; fixUp; throwAway; fail; check; break; returni
5
Appendix A presents the algorithms describing the logic of these methods. The
performance indicators list IPPS is a 3-tuple of three types of indicators:
IPPS : hCounter; Time; Ratei

Counter represents the indicators, which count (good entities, scraps,. . .), Time
represents the indicators, which describe delays, and Rate represents the indicators
using a division operation of Counter and Time indicators.
To describe the PPS behaviour, the event-triggering approach is used. Then, to
dene PPS events, we consider a three successive PPSs process (PPSi 1 , PPSi ,
PPSi1 ). The PPS events list EPPS is a n-tuple of n events:
EPPS : hE1 ; E2 ; . . . ; En i

where each event type of the PPS is at least dened by the following 5-tuple:
E : hI; ODate; MovEnt; PPS; Elogici

with I the event identier (string), ODate the event occurrence date (real positive),
MovEnt the entity associated with the event, PPS the PPS associated with the event,
ELogic the algorithm describing the behavioural logic associated with the event.
We considered three event types for the PPS:
OpEndEvent: the event corresponding to the end of an operation on a PPS,
SuEndEvent: the event corresponding to the end of a set-up on a PPS,
FailEndEvent: the event corresponding to the end of a failure on a PPS.

32

G. Habchi, C. Berchet / Simulation Modelling Practice and Theory 11 (2003) 2144

Appendix B presents the algorithms describing the dierent logic associated with
the operation end event OpEndEvent, the set-up end event SuEndEvent and the failure end event FailEndEvent.
4.2. The MovEnt model
The MovEnt is formally dened by the following 5-tuple:
MovEnt : hAMEnt ; FMEnt ; MMEnt ; IMEnt ; EMEnt i

AMEnt designs the attributes list and concerns the structural information of the
product, it is dened by the 2-tuple:
AMEnt : hI; Gi

10

where I is the MovEnt identier (string) and G the MovEnt routing.

The MovEnt routing is formally dened by the following 4-tuple:
G : hI; S1 ; PPS1; TOp1 ; S2 ; PPS2; TOp2 ; repeati

11

with I the routing identier (string), S f1; 2; . . . ; ng the operation number (integer),
PPS the PPS identier planned to process the entity (string), TOp the planned process time (real).
The functions list FMEnt is formally dened as follows:
FMEnt : hMEntStatei

12

where MEntState fgood; scrap; reworkg denes the entity state during simulation.
The methods list MMEnt is formally dened by the following 5-tuple:
MMEnt : hgroup; associate; dissociate; copy; removei

13

where group denitively groups entities in order to form another entity (assembling. . .), associate momentarily groups entities in order to generate a lot (transport. . .), dissociate temporary splits grouped lots, copy creates identical copies of an
entity, remove deletes entities from the model (customers shipment. . .)
As for the PPS, the entity indicators list IMEnt is a 3-tuple:
IMEnt : hCounter; Time; Ratei

14

The MovEnt events list EMEnt is dened by a n-tuple as follows:

EMEnt : hE1 ; E2 ; . . . ; En i

15

Each event type of the MovEnt is dened by the following 8-tuple:

E : hI; ODate; MovEnt; QLot; Lot; Interval; PPS; ELogici

16

where I is the event identier (string), ODate the event occurrence date (real positive), MovEnt the entity type associated with the event, QLot the maximal number
of lots to be created (integer), Lot the quantity of entities forming a lot of the same
type (integer), Interval the time delay between the creation of entities or entity lots
(real), PPS the rst PPS receiving the entities (it is dened by routing and usually

G. Habchi, C. Berchet / Simulation Modelling Practice and Theory 11 (2003) 2144

33

corresponds to a raw material stock), ELogic the algorithm describing the

behavioural logic associated with the event.
We considered one event type for MovEnt, the StartUpEvent which is the event
allowing the creation of entities within the model. Appendix B also presents the algorithm describing the logic associated with this event.
4.3. The CC model
The following 4-tuple formally denes the CC:
CC : hI; PILCC ; CLCC ; ALCC i

17

where I is the CC identier (string), PILCC fPICC1 ; PICC2 ; . . . ; PICCn g the performance indicators list of the CC, CLCC fCCC1 ; CCC2 ; . . . ; CCCn g the causes list of the
CC, ALCC fACC1 ; ACC2 ; . . . ; ACCn g the actions list of the CC.
The following 3-tuple formally denes the performance indicator PICC :
PICC : hO; M; P i

18

O denes the objective associated with a performance indicator. The objective describes an expected state of a whole system, a subsystem, or a resource of the system.
The objective is dened by the following 4-tuple:
O : hI; Vmin ; Vnom ; Vmax i

19

with I the objective identier (string), Vmin a minimal value associated with the objective (real), Vnom a nominal value associated with an objective (real), Vmax a maximal
value associated with objective (real).
M denes the measure associated with the performance indicator. It is gathered
from PPSs and MovEnts each time the logic of an event is executed.
P denes the performance associated with a performance indicator. It is a mathematical function depending on the measure M and objective O. The performance calculation can take dierent forms according to the denition of the objective:
8
< P f M; Vmin;nom;max
P M Vmin;nom;max
:
P Vmin;nom;max M
The performance calculation corresponds to the comparison of the measure versus
one or several objective values. In the case of a posteriori control, the objective is
equivalent to a threshold limit. In the case of anticipated reactive control, the performance corresponds to the comparison of the measure with an objective alert level
value. The alert level is a limit that has more exibility than the threshold limit in
order to anticipate and prevent problems. In both cases, reaching the alert level or
the threshold limit leads to Deviation starting up the CC control process. There is
deviation D when the current behaviour is not the expected one. D is a binary
variable depending on the value of the performance P. D f0; 1g; with D 1 if
deviation exists, and D 0 if deviation does not exist.
A causes list CLCC is associated with the CC. If the deviation D is true for a
given PICC , one searches among the causes list CCC for the cause responsible for this

34

G. Habchi, C. Berchet / Simulation Modelling Practice and Theory 11 (2003) 2144

deviation. The search is carried out with the help of a performance indicator PIC ,
associated with each cause. The formal denition of PIC is identical to the formal
denition of PICC , and the cause CCC is dened as follows:
CCC : hI; PIC i

20

An actions list ALCC is associated with the CC. This list is a set of actions that may
be applied to the system. The application of an action corresponds to the modication of the value associated with a variable or an attribute of an object making up
the PPS/CC network. When the cause of the deviation is located, one searches for the
action ACC among the actions list ALCC to be able to put the system back to its
optimised behaviour. The following 3-tuple denes ACC :
ACC : hI; VA ; DRA i

21

where I is the action identier (string), VA the action variable associated with a PPS
or a lower level CC, DRA the delayed reaction of the action (real). It corresponds to
the time needed for the action to be implemented. Within a simulation experiment, it
corresponds to the time interval separating the choice of the action variable from the
modication of its value.
The logic of the control process of the CC is described in Appendix C.
5. Implementing approach
5.1. The UML packages
The meta-model of the PPS/CC network is composed of ve packages (Fig.
11): MovEnt, PPS, CC, Simulator and PPS/CC. A package is a group of modelling elements (classes, objects). It is a way to organise models as a directory organises les. A package can contain countless diverse diagrams as well as other
packages. A modelling element of one package can be present in another pack-

Fig. 11. UML packages of the PPS/CC network.

G. Habchi, C. Berchet / Simulation Modelling Practice and Theory 11 (2003) 2144

35

age. Only one type of relationship exists between packages: dependency (at least
one customer package element uses the services of the elements of a supplier
package).
The MovEnt package is composed of the following object classes: MovEnt, MovEntType, MovEntLot, MovEntRoute, RouteElement, MovEntEvent and MovEntCreatEvent.
The PPS package groups object classes related to PPS: PPS, PPSSchedule, PPSStateList, PerfIndicatorList,. . .
The aim of the CC package is to model the control subsystem. It is composed of
several classes such as CC, WarningList, InductorList, PerfIndicatorList, ActionList, ObjectiveType,. . .
The PPS/CC network package is composed of a single object class allowing the
modelling of a production system. It represents the aggregation of PPSs, MovEnts
and CCs. The PPS/CC object class is also linked to WarningList class so that the
CCs of a network has the visibility of all warnings coming from higher levels of control.
The Simulator package represents the simulation engine. It is composed of four
classes: Simulator, SimulationObject, EventList and Event. The Simulator object
class allows the creation, starting-up and cancellation of a simulation experiment.
The SimulationObject class associates all the elements necessary for simulation including the PPS/CC network, the MovEntCreatEvent. . . This class may be duplicated in an internal simulation for action evaluation. It also generates events
processing.

5.2. The user interfaces

The main objectives of user interfaces are simplicity, user-friendliness and automatic generation of the model. User-friendly interfaces are a determining factor.
They allow the appropriation of the platform by users. The ergonomic environment
may be a success if graphical representation of concepts (PPSs, MovEnts, CCs) allows a rapid understanding of simulation and results, model animation is available,
data capture is simplied.
Here after some factors enhancing the simulation understanding:

simplicity and user-friendly graphics representing concepts;

moving the models on the screen during experiments;
association of result tables and plots during temporal evolution;
data capture enhanced by the use of help support for data gathering;
availability of data concerning model objects;
automatic generation of models from the denition of the model objects.

5.3. The Apollo platform

The implementation of the Apollo platform was done according to the following three main phases: the simulator implementation, the PPS/CC network objects
implementation and implementation of the user interfaces. The object-oriented lan-

36

G. Habchi, C. Berchet / Simulation Modelling Practice and Theory 11 (2003) 2144

guage JAVA is used for this purpose. The software Rational Rose was used to generate UML diagrams as well as high level object-oriented programs.
The separation of object classes into ve packages allows the progressive implementation giving the possibility of specication evolution. This can be done by the
modication of parameters dening the relationship between object classes.
The simulation engine is event triggered oriented. This technique consists in dening the generic events of the model and their associated logic. The events generated
when a simulation is running are stored in an events list according to their dates of
occurrence. The simulation engine sorts, advances simulation time, executes the logic
of the events and then removes the processed event from the events list. The simulation engine is implemented so that it can be re-used for any application, which can be
simulated.

6. Conclusion
In this article we have presented a new approach for modelling and simulation of
manufacturing systems. This approach, which models and simulates both operation
and control subsystems, is based on the interaction of three main kinds of concepts:
the PPS, the MovEnt and the CC. Following an algebraic specication of the proposed concepts and their modelling using UML, the Apollo platform has been developed to validate this new approach. The main contributions of our approach in
comparison with existing tools are
the exibility of the proposed generic concepts;
an object and non-block description of real objects allowing user-friendly modelling of manufacturing systems;
an automatic generation of the operation and control submodels according to the
denition of product routings and decision-makers;
the modelling of operation decision-making within a control process submodel
based on feedback loop and performance indicators;
an organised structured modication of object parameters during simulation in
order to preserve planned objectives;
a clear separation between operation and control submodels.
Moreover, the introduction of the control process into simulation will lead to
some modications in the classical simulation process, by introducing the following
feedback loopsimulation, performance evaluation, actioninto the simulation
model. The number of simulation runs allowing system optimisation may thus be reduced. The use of performance indicators within the control process introduces the
ability to act on the system into the model. As described previously, a performance
indicator is associated with a single measure and a single objective, whereas CC has
multiple performance indicators and multiple measures and objectives. The current
problem is that the CC decision is single input/single output. Thus, some research

G. Habchi, C. Berchet / Simulation Modelling Practice and Theory 11 (2003) 2144

37

will be helpful in order to enhance the decision capacity of the CC to multiple input/
single output.
Building the model is done by instantiating the objects. As each model object replaces a real object, this simply corresponds to collecting and assigning the adequate
information to the model objects. Each piece of information is a characteristic of the
real object. Therefore, the model is easy to build and object parameters simple to
modify. The Apollo platform will be used at the Alcatel Company for learning
and training purposes. Through the use of the CC models, operators and managers
will enhance understanding of the eect of the modication of parameters on operation subsystems as well as on new control methods.
As the creation of new basic building blocks with many simulation platforms is
not possible, the implementing approach we adopted using JAVA, allows the development of new objects, the extendibility of the existing model objects and the portability of Apollo because it can be used on several computing systems as well as JAVA
itself. The extendibility problem is very important for the CC object and its behaviour. Some of our current research is focused on the development of a cognitive CC
multi-agent system.

Appendix A
Algorithms 19 describe the PPS methods.
Algorithm 1 (supply)
Begin
LL 1
R1
If L 0 Then
T ldle
Else
T Inter
End If
GProduct GProduct 1
Uptime Uptime TOp
ProdRate Uptime=AcTime
End
Algorithm 2 (receive)
Begin
LL1
F 0
If L C Then
R0
T Satur
Else

38

G. Habchi, C. Berchet / Simulation Modelling Practice and Theory 11 (2003) 2144

R1
T Inter
End If
OpEndTime AcTime TOp
Insert an OpEndEvent in the events list
End
Algorithm 3 (block)
Begin
T Bloc
F 1
BlTime AcTime
End
Algorithm 4 (check)
Begin
MEntState f EStateFunction
End
Algorithm 5 (return)
Begin
F 0
RProduct RProduct 1
RTime RTime TOp
RRate RTime=AcTime
OpEndTime AcTime TOp
Insert an OpEndEvent in the events list
End
Algorithm 6 (fix Up)
Begin
R0
F 0
T SUp
NbSUp NbSUp 1
TSuTime TSuTime SUTime
SUpRate TSuTime=AcTime
SUETime AcTime SUTime
Insert a SuEndEvent in the events list
End
Algorithm 7 (throwAway)
Begin
LL 1
R1

G. Habchi, C. Berchet / Simulation Modelling Practice and Theory 11 (2003) 2144

F 0
If L 0 Then
T ldle
Else
T Inter
End If
OpEndTime AcTime
SProduct SProduct 1
ScTime ScTime TOp
ScRate ScTime=AcTime
End
Algorithm 8 (fail)
Begin
R0
F 0
T BDown
NbBDown NbBDown 1
MWaitTime f FailFunction
TMWaitTime TMWaitTime MWaitTime
WaitMRate TMWaitTime=AcTime
RepairTime f RepairFunction
TRepairTime TRepairTime RepairTime
RepairRate TRepairTime=AcTime
BDTime MWaitTime RepairTime
BDETime AcTime BDTime
Insert a FailEndEvent in the events list
End
Algorithm 9 (break)
Begin
PPSFailure f PPSBreak
End

Appendix B
Algorithms 1013 describe the events logic.
Algorithm 10 (OpEndEvent(PPSi ))
Begin
F 1
Call check(PPS i )
If MEntState scrap Then
Call scrap(PPS i )

39

40

G. Habchi, C. Berchet / Simulation Modelling Practice and Theory 11 (2003) 2144

Else If MEntState rework Then

Call return(PPS i )
Exit Sub
Else If MEntState good Then
If RPPSi1 1 Then
If SetUp True Then
Call fixUp(PPS i 1 )
Call block(PPS i )
Exit Sub
End If
Call supply(PPS i )
Call receive(PPS i 1 )
Else
Call block(PPS i )
Exit Sub
End If
End If
Call break(PPS i )
If PPSFailure True Then
Call fail(PPS i )
Exit Sub
Else
if F PPSi 1 1 Then
If SetUp True Then
Call fixUp(PPS i )
Exit Sub
End If
Call supply(PPS i 1 )
Call receive(PPS i )
End If
End If
End
Algorithm 11 (SuEndEvent(PPSi ))
Begin
R1
T ldle
If F PPSi 1 1 Then
Call supply(PPS i 1 )
Call receive(PPS i )
End If
End
Algorithm 12 (FailEndEventPPSi )
Begin

G. Habchi, C. Berchet / Simulation Modelling Practice and Theory 11 (2003) 2144

R1
T ldle
If F PPSi 1 1 Then
Call supply(PPS i 1 )
Call receive(PPS i )
End If
End
Algorithm 13 (StartUpEventMovEnt
Begin
If NbLot < QLot Then
NbLot NbLot 1
LPPS0 LPPS0 Lot 1
ODate AcTime Interval
Insert a StartUpEvent in the events list
End If
Call receive(PPS 0 )
F 1
End

Appendix C
Algorithm 14. This algorithm describes the logic of the CC behaviour.
Begin
calculate the performance and deviation of each measure
If D 0 Then
no existing deviation of the PPS
question the CC horizontal warning list and apply the necessary action
Exit Sub
Else
deviation exists on PPS
add message to the vertical warning list and horizontal warning list
End If
/evaluation of the cause/
For j 1 To n
read the cause CCCj from the causes list CLCC
calculate the performance and deviation of the associate measure
If D 1 Then
deviation exists and the cause CCCj is found
Exit For
End If
Next j
/evaluation of the action/
add a warning message to the horizontal warning list

41

42

G. Habchi, C. Berchet / Simulation Modelling Practice and Theory 11 (2003) 2144

For k 1 To m
read the action ACCk from the actions list ALCC
apply ACCk on a copy of the simulated model
calculate the performance and deviation of this action
If D 0 Then
deviation does not exist and the action is found
Exit For
End If
Next k
apply the action ACCk to the current model
End

Appendix D. Notations for measures

AcTime
BDETime
BDTime
BlTime
GProduct
MWaitTime
NbBDown
NbSUp
OpEndTime
ProdRate
RepairRate
RepairTime
RProduct
RRate
RTime
ScRate
ScTime
SProduct
SUETime
SUpRate
SUTime
TMWaitTime
TOp
TRepairTime
TSuTime
Uptime
WaitMRate

current time of simulation

break down end time
break down time
blocked time of a PPS
total good products on a PPS
waiting time for maintenance
total number of break downs
total number of set-ups on a PPS
operation end time on a PPS
production rate of a PPS at AcTime
repairing rate
time to repair
total products to rework on a PPS
rework rate of a PPS at AcTime
total rework time
scrap rate of a PPS at AcTime
total scrap time on a PPS
total scrap on a PPS
set-up end time on a PPS
set-up rate of a PPS at AcTime
set-up time for an entity on a PPS
total waiting time for maintenance
process time on a PPS
total time to repair
total set-up time on a PPS
total up time of a PPS
maintenance waiting rate

G. Habchi, C. Berchet / Simulation Modelling Practice and Theory 11 (2003) 2144

43

References
[1] M.H. Abdallah, A knowledge-based simulation model for job shop scheduling, International Journal
of Operations and Production Management 15 (10) (1995) 89102.
[2] M. Bakalem, G. Habchi, A. Courtois, PPS: an integrated object oriented approach for modelling and
simulation of manufacturing systems, in: Proceedings of the International Conference on Systems,
Man and Cybernetics IEEE, San Antonio, USA, 1994, pp. 21842189.
[3] M. Bakalem, G. Habchi, A. Courtois, PPS: a contribution for manufacturing systems simulation, in:
Proceedings of the Summer Computer Simulation Conference, Ottawa, Canada, 1995, pp. 390395.
[4] C. Berchet, G. Habchi, A. Courtois, Integration du processus de pilotage a la simulation des systemes
de production, in: Proceeding of the Second French Conference MOSIM, Annecy, France, 1999, pp.
337344.
[5] C. Berchet, G. Habchi, Modelling for simulation of manufacturing system control, in: Proceedings of
EDA2000, Orlando, USA, 2000.
[6] C. Berchet, Modelisation pour la simulation dun systeme daide au pilotage industriel, Ph.D. Thesis,
University of Grenoble, INP, France, 2000.
[7] M. Blosch, J. Antony, Experimental design and computer-based simulation: a case study with the
Royal Navy, Managing Service Quality 9 (5) (1999) 311319.
[8] R.W. Brennan, B. Foroughi, A control framework to support responsive manufacturing, International Journal of Agile Management Systems 1 (3) (1999) 159168.
[9] S.S. Chakravorty, J.B. Atwater, Do JIT lines perform better than traditionally balanced lines,
International Journal of Operations and Production Management 15 (2) (1995) 7788.
[10] D. Cumenal, Un modele de dynamique des systemes pour lanalyser et comprendre les changements
detat de 1organisation, Revue Internationale de Systemique 11 (2) (1997) 177214.
[11] S.M. Disney, M.M. Naim, D.R. Towill, Dynamic simulation modelling for logistics, International
Journal of Physical Distribution and Logistics Management 27 (3/4) (1997) 174196.
[12] P.A. Fishwick, Computer simulation: growth through extension, Transactions of the Society for
Computer Simulation International (14) (1997) 1323.
[13] M. Garetti, A. Pozzetti, A. Bareggi, On-line loading and dispatching in exible manufacturing
systems, International Journal of Production Research 28 (7) (1990) 12711292.
[14] N.N. Gindy, S.M. Saad, Flexibility and responsiveness of machining environments, Integrated
Manufacturing Systems 9 (4) (1998) 218227.
[15] S.K. Goyal, K. Mehta, R. Kodali, S.G. Deshmukh, Simulation for analysis of scheduling rules for a
exible manufacturing system, Integrated Manufacturing Systems 6 (5) (1995) 2126.
[16] G. Habchi, C. Labrune, Study of lot sizes on job shop systems performance using simulation,
Simulation Practice and Theory Journal 2 (1995) 277289.
[17] G. Habchi, C. Berchet, Some concepts for industrial control process modelling, in: Proceedings of the
5th International Conference of Decision Science Institute, Athens, Greece, 1999, pp. 777
779.
[18] G. Habchi, Modelling and simulation of complex production systems, Simulation Practice and
Theory Journal 8 (5) (2000) 281282.
[19] T. Haslett, C. Osborne, Local rules: their application in a kanban system, International Journal of
Operations and Production Management 20 (9) (2000) 10781092.
[20] F. Huq, Z. Huq, The sensitivity of rule combinations for scheduling in a hybrid job shop,
International Journal of Operations and Production Management 15 (3) (1995) 5975.
[21] Z. Irani, V. Hlupic, L.P. Baldwin, P.E.D. Love, Re-engineering manufacturing processes through
simulation modelling, Logistics Information Management 13 (1) (2000) 713.
[22] Y. Ito, Adaptive dynamic inputoutput analysis using neural networks (Japanese industrial
structure), Kybernetes 29 (9/10) (2000) 10871102.
[23] L.W. Jacobs, J. Lauer, DSS for job shop machine scheduling, Industrial Management and Data
Systems 94 (4) (1994) 1523.

44

G. Habchi, C. Berchet / Simulation Modelling Practice and Theory 11 (2003) 2144

[24] M. Jahangirian, G.V. Conroy, Intelligent dynamic scheduling system: the application of genetic
algorithms, Integrated Manufacturing Systems 11 (4) (2000) 247257.
[25] S. Julia, R. Valette, Real time scheduling of batch systems, Simulation Practice and Theory Journal 8
(5) (2000) 307319.
[26] E. Kindler, Simulation of systems with simulating components, in: Proceedings of the Management
and Control of Production and Logistics, Grenoble, France, 2000.
[27] A.M. Law, W.D. Kelton, Simulation Modelling and Analysis, second ed., McGraw-Hill, New York,
1991.
[28] J.L. Le Moigne, Les Systemes de Decision dans les Organisations, Presses Universitaires de France,
1974.
[29] D. Lesley, W. Glyn, Evaluating and selecting simulation software using the analytic hierarchy
process, Integrated Manufacturing Systems 5 (1) (1994) 2332.
[30] B.Y. Leu, J.W. Nazemetz, Comparative analysis of group scheduling heuristics in a ow shop cellular
system, International Journal of Operations and Production Management 15 (9) (1995) 143157.
[31] R.M. Marin, J. Garrido, J.L. Trillo, J. Saez, J. Armesto, Design and simulation of an industrial
automated overhead warehouse, Integrated Manufacturing Systems 9 (5) (1998) 308313.
[32] S.A. Melnyk, P.L. Carter, Production Activity Control, The Business One Irwin/APICS Series in
Production Management, Homewood, Illinois, 1987.
[33] D.J. Muller, J.K. Jackman, C. Fitzwater, A simulation-based work order release mechanism for a
exible manufacturing system, in: Proceedings of the Winter Simulation Conference, 1990, pp. 599
602.
[34] P. Novak, Reective simulation of production and logistic systems with Simula and Java, in:
Proceedings of the Management and Control of Production and Logistics, Grenoble, France, 2000.
[35] P.M. Pelagagge, G. Cardarelli, A. Santalucia, Comparison of conventional and dynamic FMS
periodic loading rules, Integrated Manufacturing Systems 9 (1) (1998) 1522.
[36] A. Polajnar, B. Buchmeister, M. Leber, Analysis of dierent transport solutions in the exible
manufacturing cell by using computer simulation, International Journal of Operations and
Production Management 15 (6) (1995) 5158.
[37] P. Rogers, M.T. Flanagan, Online simulation for real-time scheduling of manufacturing systems,
Industrial Engineering 37 (1991) 3740.
[38] P. Rogers, R.J. Gordon, Simulation for real-time decision making in manufacturing systems, in:
Proceedings of the Winter Simulation Conference, IEEE, New York, USA, 1993, pp. 866874.
[39] D. Roy, Une architecture hierarchisee multi-agents pour le pilotage reactif dateliers de production,
Ph.D. Thesis, University of Metz, France, 1998.
[40] S.M. Saad, M.D. Byrne, Comprehensive simulation analysis of a exible hybrid assembly system,
Integrated Manufacturing Systems 9 (3) (1998) 156167.
[41] T.H. Soon, R. de Souza, Intelligent simulation-based scheduling of work-cells: an approach,
Integrated Manufacturing Systems 8 (1) (1997) 623.
[42] M.T. Tabucanon, D.N. Batanov, S. Basu, Using simulation to evaluate the batching approach to part
type selection in exible manufacturing systems, Integrated Manufacturing Systems 9 (1) (1998) 514.
[43] L.J. Taylor III, A simulation study of WIP inventory drive systems and their eect on nancial
measurements, Integrated Manufacturing Systems 10 (5) (1999) 306315.
[44] P.S. Welgama, R.G.J. Mills, Use of simulation in the design of a JIT system, International Journal of
Operations and Production Management 15 (9) (1995) 245260.