See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/224299469

A holonic approach for manufacturing execution system design: An industrial

application

Conference Paper · October 2007

DOI: 10.1109/EFTA.2007.4416923 · Source: IEEE Xplore

CITATIONS READS

15 796

4 authors, including:

Pierre Castagna Isabel Demongodin

University of Nantes Aix-Marseille Université
93 PUBLICATIONS 859 CITATIONS 70 PUBLICATIONS 740 CITATIONS

SEE PROFILE SEE PROFILE

Jean-Claude Hennet
Aix-Marseille Université
144 PUBLICATIONS 1,810 CITATIONS

SEE PROFILE

All content following this page was uploaded by Jean-Claude Hennet on 14 June 2016.

The user has requested enhancement of the downloaded file.

 A holonic approach for manufacturing execution system design:
an industrial application

Blanc Pascal, Castagna Pierre Demongodin Isabel, Hennet Jean-Claude

IRCCyN – UMR CNRS 6597 LSIS – UMR CNRS 6168
1, rue la Noe BP 92101, Avenue Escadrille Normandie-Niemen,
44321 Nantes Cedex 3, France 13397 Marseille Cedex 20, France
[blanc, castagna]@irccyn.ec-nantes.fr [demongodin, hennet]@lsis.org

Abstract Holonic Manufacturing System (HMS) (Deen,

2003) is such an approach which combines the best
Approaches such as Holonic Manufacturing System features of hierarchical and heterarchical organizational
(HMS) propose a structure organized as a loosely structures. This concept can preserve the stability of
coupled network of communicating agents to answer in hierarchy while providing the dynamic flexibility of
a flexible way to necessities of reactivity and heterarchies and offers a suitable background to
profitability. This paper presents such a Holonic develop a distributed Manufacturing Execution
Manufacturing Execution System and illustrates this System. To that end, the reference holonic architecture
approach on a real industrial application. On one PROSA (Product-Resource-Order-Staff Architecture)
hand, we propose mechanisms to synchronize the is utilized (Valckenaers et al., 2003). PROSA is based
execution of manufacturing orders, to guarantee the on a typology of the manufacturing system elements:
availability of components and to set time-limit products, resources, orders and specialized decision-
constraints on task scheduling. On the other hand, making centers (hierarchical and/or functional). It
based on a typology of Resource holons (supply, allows identifying and defining the roles and behaviors
transformation, assembly or disassembly), we propose of the manufacturing control elements.
specific heuristics to solve the tasks scheduling On the one hand, we propose mechanisms to
problem, taking into account each optimization model. synchronize the execution of manufacturing orders.
These mechanisms guarantee the availability of
components and enable to set time-limit constraints on
1. Introduction task scheduling. In other words, it allows respecting
precedence constraints. On the other hand, we propose
Today’s manufacturing systems are challenged by a a typology of Resource holons, based on characteristics
demanding fast changing environment. Industrial firms of the tasks that they perform: supply, transformation,
have to implement control policies and adapt their assembly or disassembly. For each type of Resource
manufacturing systems to maximize their productivity, holon, specific heuristics are used to solve the tasks
considering also the manufacturing-customer scheduling problem, taking into account the own model
responsiveness as a main issue (McFarlane et al., of each one. In the case of a disassembly holon, the
2003). The need for more reactivity, flexibility, and scheduling problem may be associated to a
productivity leads to drastically increase the disassembling pattern design (stock-cutting problem).
complexity of the (re)design of Manufacturing Thus an adaptation of the basic scheduling heuristic is
Execution Systems (MES). proposed. These mechanisms and algorithms are
Classical centralized and hierarchical approaches adapted to a distributed manufacturing control system
based on time or constraints aggregation show and may be implemented in a multi-agents system.
limitations to solve manufacturing control problems, The contribution is organized as follows. In section
particularly when the system faces numerous random 2, the context of the manufacturing control system is
events. Mass customization, small volume/high variety described; specific problems related to this control are
order management and maximization of the addressed. The use of a PROSA-based holonic
profitability of the firms are long-term challenges approach and the mechanism to perform the
which require new approaches. These approaches manufacturing control is presented in section 3.
support the use of cooperative and autonomous unit Finally, section 4 applies the proposed manufacturing
self-organized in an open structure to offer a very high control system to a real industrial application.
operational and structural level of flexibility.
2. Manufacturing Control represented by a Gantt chart. A feasible solution is a
schedule, which respects the whole set of restrictions.
A manufacturing system can be considered as a set The problem may be over restricted and thus its set of
of resources used to transform raw material into feasible solutions may be empty. The formal model of
finished goods in order to satisfy a demand. In a a scheduling problem is deeply linked to the
systematic way, it is now usual to structure this system characteristics of the system which activity has to be
in three subsystems: physical, informational and scheduled. Let us consider a task i, modeled by a
decisional. quadruplet (ri, ti, di, wi) where ri is a release date or
The physical subsystem is composed of the set of earliest start date, ti is the processing time or operation
physical devices of the manufacturing system: duration, di is the due date or latest finish date and wi
machines, transport devices, workforce. The is a weight or priority. Additional restrictions may be
transformation of raw material into finished goods is related to tasks sequencing or to resource capacity.
the result of successive transformations of single or This model allows formalizing restrictions or to
multiple components into items by way of operations evaluate performance criteria, such as the date
completed by the physical devices of the system. achievement of a task, the lateness of the achievement
Physical transformations are performed by equipments of a task versus a due date, the cost of achievement of a
or work force. They consist in assembling components, task, for instance.
disassembling (cutting process, for instance), or simply Most scheduling problems are classified as NP-
transforming raw material into a product. complete problems (Brucker, 2001). They are highly
The informational subsystem gathers all the complex problems, for which the time and computing
information existing in the manufacturing system, effort are increasing beyond reasonable limits when the
either static such as bills of material, manufacturing size of the problem increases.
routing, standard operation duration, or dynamic such
as current state of resource, location of an item. 2.2. Disassembling pattern design
The decisional subsystem performs the management Efficiently controlling the execution of
of the working of the whole manufacturing system. manufacturing activities also requires solving
This subsystem is in charge of fixing and putting into additional optimization problems. In fact, Gantt chart
practice tactics to produce with quality the right item at permits to define the execution of tasks during time,
the right time, in a rational way. In other words, the which may be inadequate for some processes.
decisional subsystem sets and receives orders to Disassembling processes, such as cutting process,
schedule the execution of the operations, obeying require disassembling patterns design (cutting plan) to
restrictions such as availability of components or define the spatial location of the products, in order to
physical devices, minimizing the tardiness of a finished be performed. Obviously, there is a relation between
good production and maximizing the use of material the location of a product in a scheduled disassembling
or/and resources. pattern and the date of execution of the disassembling
In the next part, two of the main functionalities of a task. Thus, besides scheduling problem, control system
decisional system are addressed: scheduling and stock- should solve optimization problems such as
cutting pattern design, as both problems are faced disassembling pattern design (Parada, et al., 2000).
during the design of the control.
3. Holonic Manufacturing Execution
2.1. Scheduling System
Scheduling consists in allocating tasks to resources
and defining dates for the execution of these tasks. A Holonic Manufacturing System (HMS) is a highly
Scheduling may be performed for a forward-looking decentralized manufacturing system, consisting of
purpose, in order to plan the activities of the resources autonomous and cooperating agents called holons that
and to anticipate the achievement of the tasks. respects some flexible control rules forming a
Scheduling may also be purely reactive. In this case, holarchy. The holonic concept has been proposed for
the allocation of tasks is performed dynamically and no the study of social and natural phenomena. Koestler
planning is proposed. A hybrid approach is also (1969), who is the father of the word “holon”, observed
possible, thus a planning may be performed and that such systems develop adopting stable forms.
dynamically altered in order to respond to a Holonic systems are thus defined as fitted hierarchies.
disturbance, which may be an intentional change in the Low-level holons are defined as reactive and high-level
environment or a random event such as a machine holons have cognitive behavior. For the adaptation of
breakdown or a product loss. the holonic concept to manufacturing control, Deen,
Solving a scheduling problem aims at minimizing or (2003) proposes to define holons as a set of a physical
maximizing one or more criteria, respecting a set of part and a logical part which are connected.
restrictions. A solution to a scheduling problem may be
 The reference architecture for Holonic aggregation exists between Product holons as Product
manufacturing system, developed by Valkenaers et al. holon may be made up of other Product holons.
(2003), is PROSA which stands for Product-Resource- Product holons have identification and are defined as a
Order-Staff Architecture. collection of product and process specifications.
The manufacturing execution system (MES) aims at Product specifications describe the physical
controlling the manufacturing system: what and when characteristics of a product; they are computer-assisted
to produce, how and when to use the resources, which design (CAD), pictures, drawings, measurement.
and when to launch orders. In this section, a holonic Process specifications describe the process that the
manufacturing execution system (HMES) is proposed. component(s) of the product have to undergo to
This control system only considers the equipment manufacture this product. Product holons behave as
responsible for physical transformation of products, information server to other holons, they answer queries
which are assembling, disassembling and transforming concerning product or process specifications.
operations. The static roles of holons, as well as Each physical device considered in the production
dynamic mechanisms are presented. The roles and the system (equipment) with its own logical part is a basic
models of the three basic holons (Product-Resource- Resource holon. Resource holons are responsible for
Order) are detailed. These holons interact into two lot-sizing and scheduling of the activities of their
types of holarchies which are each one responsible for physical part in accordance with their own model. This
an aspect of the manufacturing execution: the model aims at proposing a rational behavior for using
availability of the materials and the availability of the material and resource, taking into account changes of
resources. Finally, the whole dynamic of this HMES is conditions such as failure, rework, or quality rejection
explained. of a product. A typology allows classifying physical
devices into four classes, which are specializations of
3.1. Basic holons definition the Resource holon class. Thus, we consider: Supplier
At the micro level, individual roles are defined for holons, Assembling holons, Disassembling holons and
the basic holons. The class diagram (Fig. 1) shows the Transforming holons.
relations between the different classes of basic holons Resource holons are defined as a collection of
as well as their attributes. process capabilities which may correspond to process
May be
User
specifications of product holons. Resource capabilities
made up of

Product Holon
0..n
0..1
+CreateOrder (Product, Priority, LFD): Order
+SolicitOrder (Product, Priority, LFD): Order
are production recipes or production methods. The
+Id
+SuggestLFD (Order, LFD): LFD
+PropagatePriority (Priority): void
matching between process specifications and process
Is defined as
a collection of
0..1 capabilities means that the resource owns the
0..n 0..n May be
capabilities to manufacture this product. This matching
Process Specification Product Specification
0..n
0..n made up of
Order Holon
0..n
represents the production process which may be used
Corresponds to
+Priority to manufacture the product.
+State 0..n
Process Capability +Release Date - ri
+Due Date - di
Resource holons are responsible for the scheduling
0..n
+Date of Scheduling - si
of the tasks of each Order holons. They attempt to
+CreateOrder (Product, Priority, LFD): Order
Is defined as
a collection of 0..n
+SolicitOrder (Product, Priority, LFD): Order
+SuggestLFD (Order, LFD): LFD 0..n
schedule this task considering release dates, priorities
Ressource Holon
0..1
+PropagatePriority (Priority): void
+SolicitResource (Product, Priority, LFD) and in function of their capacity and the requirements
+RequestScheduling (Product, Priority, ESD, LFD)
+Scheduling ()
+NotifyDoS (DoS) for each task to be performed (such as tool
+CalculateSchedulingMargin (Order)
+NotifyRegister (Order)
+Schedule ()
May be
making up a
0..n
availability). In addition to the schedule produced,
+NotifyScheduling (Order)
resource holons may produce other results, such as a
pattern, or material cutting plan, in the case of a
Supplier Holon Disassembling Holon Transforming Holon Assembling Holon disassembling process.
+DisassemblingPattern ()
Order holons are orders for manufacturing a
+PlaceItem (Order)
+NotifyPlacement (Order) physical product. Their physical part is a physical
product to be manufactured and their logical part
Fig. 1. UML class diagram of basic holons contains information about time-restrictions and date of
Each product holon is a product item considered in scheduling of the manufacturing operation. Thus, each
the production system. Product items define the Order holon is associated to a Product holon which
characteristics and the properties of the products, they contains the definition of its physical part. The
may be bought (raw material) or manufactured, sold manufacturing of a product may require components or
(finished goods) or used to manufacture other goods raw material, and at the same time, a physical product
(raw material, intermediate products). The bill of may be a component used to manufacture other
material indicates the link of composition between products. The reflexive associations between Order
items, as manufactured items are produced using holons model the links of composition between them.
components which are also items. Thus a reflexive These links allow synchronizing the date of
manufacturing of a product and its components. To that
end, Order holons own a time model of the task higher priority and which requires a component. The
required to manufacture its product: each task is manufacturing activities require synchronization
defined by a release date (ri) which corresponds to the between Order holons within an Order holarchy. This
earliest start date for the manufacturing task to be mechanism aims at guaranteeing that components are
performed, a due date (di) or latest finish date, and a available, in order to manufacture a product.
date of scheduling (si), if the task has been scheduled.
Each order holon also has a priority, which expresses 3.3. Dynamic
the importance of each task to be performed. A The whole dynamics of this MES/HMS mainly rely
synchronization of the time restrictions has to be on the activity of the Order holons. From the order
performed so that a due date for an Order holon creation to the product consumption or delivery to a
corresponds to the release date of the Order holon it is final client, the Order holon is in charge of soliciting
a component of. This synchronization guarantees the the services of a Resource holon in order to be
respect of constraints of precedence. manufactured, acquiring the components required for
Order holons subscribe the Resource holon which its manufacture, and propagating changes of priority
owns the specific capabilities to perform the and time limit for its manufacturing. The state diagram
manufacturing of its product. The association between of Fig. 5 illustrates the lifecycle of Order holons, which
Resource holon and Order holon instantiates this is detailed below. The opportunistic behavior of Order
subscription. The Resource holon schedules the Order holons is also considered as they may respond to a
holons using the attributes of each of them. This time- request from an upper Order holon with higher priority.
window approach gives an assessable margin of Thus they may leave an Order holarchy and enter
autonomy for the responsible Resource holon to another one. In the same way an upper Order holon
schedule the tasks. Once the task has been scheduled, with a lower priority may lose its components in favor
the Order holon informs the associated Order holon of an other upper Order holon.
(upstream and downstream), so as to synchronize the
Created
manufacturing of its components and the product of
Initialisation
which it is a component of. The synchronization
mechanism is described in the next section. Associated to a Product Holon
Inactive

3.2. Holarchies Solicited

The interaction between holons allows performing Acquiring Registered Acquiring
the control of the manufacturing system, mainly the resource components
Placed on a Component Components Unplaced
predictive and reactive scheduling of the resources Released pattern or Solicited scheduled from a
(including raw material supply), the predictive and supplied pattern
reactive design of the pattern disassembling, and the Active Placing and
Scheduled Scheduling
synchronization of the manufacturing of products and
their components. The global behavior of this Solicited, Priority changed,
MES/HMS is the result of the concurrent interaction of Manufactured Time-limit changed.
Product
holons inside two kinds of holarchie or groups of available
holons.
First, Resource holarchies are built up by a single
Resource holon and the Order holons which have Fig. 2. State diagram of Order holons
subscribed to its services. This holarchy aims at Order holons are created to respond to a demand
performing the scheduling of the activities of a concerning a single product, when no existing and
resource and the design of the disassembling pattern if available Order holon matches this demand. Such a
required. demand may come from a user (commercial order for a
Then, Order holarchies are groups of Order holons finished product, internal order for stock in process
which are each one a product used for the refill), or from an upper Order holon which requires a
manufacturing of a finished good, or the finished good component in order to be manufactured. Order holon
itself. Each one of these Order holons may be a are initialized with a Product holon which owns the
component of an other Order holon called upper holon. characteristics of the product to be manufactured.
Order holons which are components of an upper Order Newly created Order holons are automatically
holon are called lower Order holons. The Product solicited; inactive holons are solicited when associated
holon associated to each Order holon indicates the to a user or an upper Order holon. Order holons are
components required to manufacture the product. then activated when they are informed of a due date,
Nevertheless, associations between Order holons are for the product to be manufactured.
dynamic, as Order holons are opportunistic and may
answer positively to an offer from an Order holon with
 Order holons first query their associated Product holons follow the same lifecycle. This recursive
holons to retrieve the process specifications required mechanism ends when the components are scheduled
for their manufacturing. Then, they send a request to and send back to their creator (upper Order holon or
the Resource holon, which owns the correspondent user), their date of scheduling. When it happens, the
process capabilities, in order to subscribe its services. creator of an Order holon has a release date (or a
To be scheduled, an Order holon must have a release delivery date for the user) confirmed. This creator may
date confirmed. This confirmation happens when its then request its Resource holon to schedule its task.
components or lower Order holons are scheduled, since When the release date of a manufacturing task has
a release date of an upper Order holon is equal to the been confirmed, the corresponding Order holon
due date of its lower Order holons. The scheduling requests the associated Resource holon to schedule this
mechanism implies that a task cannot be scheduled if task and if necessary to place the product into a
the previous tasks, i.e. the tasks of lower Order holons, disassembling pattern. The placing and scheduling are
are not scheduled. Depending on the type of Resource performed by the Resource holon, taking into account
holon, various cases are described. the release date, and the priority of each Order holon
If the product is a raw material, no component is which has subscribed its services. If the due date is
required, and the corresponding Resource holon is a exceeded, the Order holon creator or the user will be
Supplier holon. Supplier holons are in charge of raw advised as the date of scheduling is propagated inside
material stock control; they define a delivery date for the Order holarchy. During this step, Order holons with
raw material. If the required product is in stock, the lower priority may be unplaced from any
product may be delivered instantly, whereas the disassembling pattern, thus, they should acquire new
supplier holon may put an order and inform a delivery components before requesting once more placing and
date to the Order holon. The scheduling is then scheduling to the Resource holon.
performed instantly. If the Resource holon is a Once scheduled, Order holons propagate their date
disassembling holon, it attempts to place the product of scheduling downstream, to confirm the due date of
into an existing disassembling pattern. This pattern their components, and upstream, for their creator to
must be designed for a lower Order holon which confirm their release or delivery date. When active,
matches the characteristics of the component of the Order holons may be released, due to an Order
subscribing Order holon. If the placement is possible, cancellation, thus, they unsubscribe the Resource holon
the subscribing Order holon is associated to this lower and go to inactive state, waiting for a new solicitation.
Order holon and automatically scheduled. This cancellation is propagated downstream to the
If the placement fails, and more generally if the Order holons components. Order holons components
Resource holon is a Transforming or an Assembling may be solicited by other Order holons with higher
holon, the scheduling can not be performed instantly, priority, their opportunistic behaviour drive them to
as no confirmed release date is defined for the Order leave their current Order holarchy to enter into a new
holon. In this case, the Resource holon registers the one. Active Order holons may have to acquire new
subscription of the Order holon and evaluates a likely components when this occurs. Finally, changes of
release date for this manufacturing task. This priority and time limit (ri or di), may oblige Order
evaluation allows defining a margin of autonomy for holons to request a rescheduling, and eventually a new
the Resource holon to schedule this task, and may be placement.
done using a constant quantity of time units, that may
depend on the workload of the resource. This 3.4. Scheduling and disassembling pattern design
mechanism may be considered as planning for algorithms
manufacturing execution. This likely release date is As previously described, Resource holons are
then informed to the Order holon, as it will be used as a responsible for the scheduling of activities of a
due date for its components. resource. Although the specific behaviour of a
Once planned, Order holons query other Order Resource holon depends on its own model, the
holons to complete their holarchy with lower Order scheduling is performed using a generic rule-based
holons which match the required components. If no algorithm presented below.
available or opportunistic Order holon responds the According to section 2.1, each task i is modeled by
request, Order holons have to create new Order holons a released date or earliest start date ri, a processing
for their components to be manufactured. These Order time or operation duration ti (defined by the Resource
holons are associated to Product holons which own the holon), a due date or latest finish date di and a weight
definition of the required components, and are or a priority wi. In order to perform the scheduling, the
activated with the likely release date the upper Order “earliest by priority order” scheduling rule is used.
holon which created them. This likely release date of Tasks to schedule are first sorted by decreasing
the upper Order holon is the due date for the priority, the Resource holon attempts to schedule them
components to be manufactured. Newly created Order from their release date ; if it is not possible, it attempts
to find a slot sliding the task forward (delaying the acceptable solutions really quickly. In the case of a due
task). The following algorithm (object-like formalism) date which is exceeded, Order holon propagates its
implements the “earliest by priority” rule. Let T be the date of scheduling to its upper holon which will take
set of tasks i, that have to be scheduled, sorted by this date as its new release date to request its own
decreasing priority (wi). Once scheduled, set G scheduling.
contains tasks sorted by increasing date of scheduling
(si). 3.5. Processing of disturbances
T(1).s=T(1).r Disturbances are events which modify the data of
Insert T(1) in G the control. These events may be caused by a user
For i = 2 to n (change of priority or due date) or a random event
j=1 (product loss, or machine breakdown). Some of theses
/* Looking for a first slot to insert T(i) */ events lead to invalidate schedules totally or partially;
While {T(i).r > G(j).s} and {j < card(G)} others do not have any effect on the control, although it
j=j+1 may be interesting to reschedule some resources for
End while optimization purpose.
/* limit case: T(i) should be scheduled at the end */ When the schedule is invalidated, the rescheduling
If T(i).r > G(j).s Then j = j + 1 is compulsory. Examples of invalidation of a schedule
While T(i) is not scheduled are: subscription of a new Order holon which is
If j > card(G) Then /* If the slot to insert T(i) is the last one */ impossible to schedule at its release date, increase of
T(i).s = max(T(i).r, G(card(G)).s+ G(card(G)).t) the priority of an Order holon which is not scheduled at
Insert T(i) in G after G(card(G)) its release date, delay of the release date of an Order
Else holon that exceeds its scheduling date, delay of the
If j = 1 Then /* If the slot to insert T(i) is the first one */ manufacturing of a product which causes an overlap
If T(i).r+T(i).t <= G(j).s Then with a task. In these cases, the Resource holon must be
T(i).s = T(i).r rescheduled. Nevertheless, this rescheduling may affect
Insert T(i) in G before G(1) only Order holons with lower priority than the Order
Else holon responsible for this disturbance. When
j = j + 1 /* Impossible to insert, let’ find the next slot */ rescheduling an Order holon, a new date of scheduling
End if is proposed. If this new date is compatible with the
Else latest finish date, it does not cause any disturbance for
/* If the slot is not the last one nore the first one */ the upper holon. Whereas if the latest finish date is
If max(T(i).r, G(j-1).s+G(j-1).t)+T(i).t <= G(j).s Then exceeded, the Order holon advises its upper holon
T(i).s = max(T(i).r,G(j-1).s+G(j-1).t) which has to request for a rescheduling. In this case,
Insert T(i) in G before G(j) the disturbance is propagating upstream, until being
Else absorbed by an upper holon or until delaying the
j = j + 1 /* Impossible to insert, let’ find the next slot */ delivery to the final client.
End if A Resource holon may also attempt to reschedule in
End If order to locally optimize the schedule. Examples of
End if events which do not invalidate a schedule but may alter
End while the schedule in case of rescheduling are: decrease of
Next i the priority of an Order holon, when a release date is
put forward, when an Order holon unsubscribe a
For disassembling holons, an adaptation of the Resource holon (due to order cancellation), when the
“earliest by priority” rule is necessary, as various Order manufacturing of a product is performed in advance.
holons are gathered together into a single These disturbances do not imply to reschedule a
disassembling pattern. In this case, the Resource holon resource, since the schedule is still feasible,
places Order holons by decreasing priority into the nevertheless, a rescheduling may be performed as it
disassembling pattern. Patterns are then scheduled may improve the scheduling. Alternative schedules
using the “earliest by priority” rule. In this case, the may also be proposed by Staff holons. These schedules
priority of a pattern is the highest priority of the Order are performed using more expensive but more efficient
holon placed in it, and the operation duration is the approaches such as complex heuristics or meta-
sum of all the operation durations of the disassembling heuristics. These approaches allow taking into account
task of Order holons. numerous parameters such as a cost function to
These proposed scheduling rules neither guarantee evaluate a schedule; the result is a better scheduling of
an optimal solution, nor a feasible solution as latest the tasks. In a highly perturbed environment, the
finish date may be exceeded. Nevertheless, they have lifetime of a schedule is shorter, thus the computing
very satisfactory performances as they allow retrieving time to perform such a schedule may not be sufficient.
Then, Resource holons work using their inner rules process aims at preparing the glass layers individually:
such as “earliest by priority”. In a relatively less glass cutting, edge grinding and polishing, and
perturbed environment, Resource holon may accept silkscreen printing are the operations performed during
such a schedule as it may propose a better result this step. Glass layers are then gather together as they
without introducing disturbances. This schedule is build up a single product, and are bent in a furnace.
discarded in case of major disturbance. This glass structure may be stocked for later use
The events registered from the physical system are (assemble-to-order). Plastic layers are cut, in function
not negotiable; they just may or may not lead to modify of a customized recipe and the whole is finally
the schedules and/or the time windows. Various assembled. For additional information, an extensive
scenarios are conceivable when synchronizing with the description of the process is available in (Blanc et al.
physical system. Three main events are considered: 2006).
achievement of an operation, loss of a product, Each resource holon owns a specific decision model
resource breakdown. Achievement of an operation may which corresponds to its own restrictions and
happen earlier or later than scheduled, both cases have objectives. Decision models are coordinated using the
been considered previously. The loss of a product leads mechanism previously described. The complexity of
to create a new Order holon which begins a new these models is only limited by the response time. As
lifecycle. Resource breakdown may have a major the holons presented are low-level Resource holons,
impact on a schedule since it may delay many tasks reactive behavior must prevail. As a result, heuristic
and compress autonomy margin for the Order. rule-based algorithms are proposed to model the
Disturbances generated or which happen in the decision-making process within each holon.
system may be absorbed by the autonomy margin of a Nevertheless, the use of Staff holons allows integrating
task. If not, the Order holon for which an inconsistency any additional model to perform a simultaneous search,
is detected propagates the disturbance to upper Order including the commercial cutting pattern design
holons. software in use today in the firm, which may be
encapsulated in a Staff holon.
4. Application As an example, a model for a glass cutting holon is
presented. In this case, the Resource holon is a
The development and implementation of the Disassembling holon, thus it should solve an allocation
proposed HMS/MES is motivated by a study and placement problem of the glass layer into a cutting
performed in a real manufacturing system, which pattern, and a scheduling problem of the cutting
belongs to the American Glass Product (AGP) pattern. The allocation problem should first consider
company. the material characteristics (type of glass, color, and
American Glass Product, produces laminated thickness) of the glass layer to be cut in order to match
security glass for automotive application. In the with the material characteristic of the stock plates
increasingly competitive environment, AGP aims at available. The placement problem considers the
offering highly customizable products, managing the dimensions of the matching glass plates and the
manufacturing most with design-to-order and make-to- geometric characteristics of each glass layer to be cut,
order policy, and in a smaller part with assemble-to- this information is available as a Drawing eXchange
order policy. For this product, quality criteria are Format file (DXF), which is a two-dimensional
demanding as automotive norms are used for quality graphics file format supported by virtually all PC-
control. Moreover, the industrialization of the process based CAD products. These sets of information are
is recent, and still in progress. The robustness of the collections of product specifications controlled by
process is not guaranteed, and numerous random Product holons. Moreover, process specifications for
events occur. Finally, raw materials which go into final the glass layer holons include rotation restrictions, and
product come from high technology and are expensive. minimum trim loss on each side of the rectangle which
A rational utilization of these materials is then contains the shape, in order to guarantee a safe cutting
necessary. Due to disturbances and mass- process, for both the operator and the glass layer. All
customization, numerous normal and degraded flows this information is made available by way of the Order
intermix. The manufacturing policies imply to consider holon, for the glass cutting holon to control the
time as a critical parameter. In conclusion, AGP execution of the manufacturing process. Finally, the
requires a robust, reactive and flexible system to cutting glass holon may perform the cutting pattern
control costs, delays and quality within its design and the scheduling of the pattern using the
manufacturing system. adapted earliest by priority scheduling rule.
The product is defined by its geometry (dimensions, The whole MES/HMS is based on the interaction of
curve), and its composition (glass and plastic layers). the micro models of each Resource holon. These
The process is split into two main steps: glass pre- models allow controlling the execution of the
process, and glass and plastic assembly. The glass pre- manufacturing tasks within each Resource holon. As
 previously described, Order holons subscribe to development effort, since the manufacturing execution
Resource holons and relay accurate information system evaluated is the same as the one used in the real
necessary in order to perform the manufacturing context. At present, this manufacturing execution
execution control and scheduling. Synchronization system (MES) is integrated to the management system
between the bounds for each manufacturing task to be of the firm, as it interfaces the enterprise resource
executed is realized within the Order holarchies. This planning (ERP) and the physical system.
communication infrastructure builds a control system This proposed HMS/MES is valid for discrete-event
from the loose relation of system components. based manufacturing systems, which consider
transformation, assembly and disassembly operations.
5. Conclusion Possible future development should consider the
control of task allocation, when various resources own
The increasing needs for flexibility, reactivity and matching process capabilities. Other developments
efficiency result in a growing complexity of may reinforce the efficiency of a global scheduling
manufacturing systems and a necessity of integration problem solving, propagating scheduling scenarios,
of their control. These requirements are illustrated by within holarchies in order to evaluate them globally.
an industrial case which is proposed as an example for Finally, the synchronization between the
manufacturing system control design. Within this manufacturing of the components of a single product
manufacturing control system, problems are identified may also be addressed in future works.
as scheduling and/or optimization problems, for which
dynamic data are numerous and highly versatile. References
In this article, the use of a holonic approach is
proposed to perform the control of such an industrial [1] Blanc P., Demongodin I. and Castagna P, 2006. A
manufacturing system. The use of a holonic structure Holonic Approach for manufacturing Control: An
allows building a reactive and configurable control, industrial application. 12th IFAC Symposium on
while considering global aspects of the manufacturing Information Control Problems in Manufacturing,
system. This approach, which is PROSA-based, is INCOM'06, May 17-19, St Etienne, France.
compatible with holonic component-based structure. [2] Brennan R. W., Norrie D. H., 2001. Evaluating the
The use of the PROSA architecture is based on a performance of reactive control architecture for
typology of the manufacturing system elements: manufacturing production control, Computers in
products, resources, orders and specialized decision- Industry, 46, pp 235-245.
making centers (hierarchical and/or functional). It [3] Brucker, 2001. Scheduling Algorithms. Springer
allows us to identify and define the roles and behaviors Verlag, Berlin, 3rd edition, ISBN 3-540-41510-6.
of the manufacturing control elements. [4] Deen S.M. (Ed.) (2003). Agent-Based Manufacturing:
For this architecture, we propose the use of Advances in the Holonic Approach. Springer ISBN 3-
mechanisms to synchronize the execution of 540-44069-0.
manufacturing orders. These mechanisms guarantee the [5] Koestler A., 1969. The Ghost in the Machine, Arkana
availability of components and enable to set time limit Books, London.
constraints on task scheduling. [6] McFarlane Duncan Sarma, Sanjay Chirn, Jin Lung,
Moreover, a typology of Resource holons is Wong C.Y., Ashton Kevin, 2003. Auto ID systems and
proposed. This typology is based on characteristics of intelligent manufacturing control, Engineering
the tasks performed: supplying, disassembling, Applications of Artificial Intelligence, 16, pp 365–376.
transforming or assembly. For each type of Resource [7] Parada V., R.Palma, D.Sales, A.Gómes, 2000. A
holons, specific heuristics are used to solve the comparative numerical analysis for the guillotine two-
manufacturing control problem, taking into account dimensional cutting problem, Annals of Operations
each micro model. For instance, a specific heuristic Research, 96, pp 245-254.
generates a solution to the scheduling/packing problem [8] Simão J. M., Stadzisz P. C., Morel G., 2006.
in the case of a raw material cutting machine. Manufacturing execution systems for customized
These mechanisms and algorithms are implemented production, Journal of Materials Processing
within a multi-agents system, which supports the Technology, 179, pp 268–275.
development of the manufacturing control system. To [9] Valckenaers P., H. Van Brussel, Hadeli, O. Bochmann,
evaluate this manufacturing control, the use of discrete- Bart Saint Germain, C. Zamfirescu, 2003. On the
event simulation is proposed as an emulator of the design of emergent systems: an investigation of
physical system. This proposition makes possible to integration and interoperability issues, Engineering
solve problems of redundancy during the development Applications of Artificial Intelligence, 16, pp 377–393
of manufacturing control, permits to reuse objects of
the simulation model, and allows some savings in the

View publication stats