Implementing the concept of Product-Driven Control using Wireless Sensor
Networks: some experiments and issues
David Gouyon*, Michael David*
* Centre de Recherche en Automatique de Nancy (CRAN), UMR 7039 CNRS Nancy-Université
Faculté des Sciences et Techniques, BP 239, Vandoeuvre-lès-Nancy Cedex.
david.gouyon@cran.uhp-nancy.fr; michael.david@cran.uhp-nancy.fr
Abstract: In the dynamically moving context of mass-customization of products, new manufacturing
control architectures, based on the consideration of highly distributed, autonomous, adaptable and
efficiently cooperating units integrated by a plug-and-operate approach, seem to be efficient alternatives.
Amongst them, the concept of a product-driven distributed control promotes an active role of the product
in its own manufacturing. This paper focuses on the possibilities to implement this concept on a case study
using wireless sensor networks.
1. FROM INTEGRATED TO AGILE
MANUFACTURING CONTROL
Advances in the use of Information Technologies in
manufacturing systems give manufacturers an opportunity to
promote make-to-order business models and mass
customization of products (Da Silveira et al. 2001). Facing
this wide range of customized customer orders impacts the
whole set of enterprise information and control systems (Nof
et al. 2006), which integration capability has to be improved
according to the Enterprise Integration Capability Model
(Hollocks et al. 1997) (EICM Fig. 1), in a dynamically
moving context.
Adaptable Intelligent system
Interoperable Distributed system
Visible Integrated system
Rigid Hierarchic system
Fragmented Fragmented system
Fig. 1. Enterprise Integration Capability Model
Standards, as the IEC/ISO 62264 (ISO 2003) promoted by
the MESA (Manufacturing Enterprise Solutions Association,
http://www.mesa.org), the ISA (Instrumentation, Systems,
and Automation Society, http://www.isa.org) and the ISO
(International Organisation for Standardisation,
http://www.iso.org), enable manufacturing enterprise-control
system integration from the business level to the process level
in order to meet industry-led Business-to-Manufacturing
issues (Morel et al. 2003) (Fig. 2a). In this context,
Manufacturing Execution Systems (MES) ensure information
flow synchronic gateway between enterprise and shop floor
control systems and diachronic integration between execution
activities (service flows). The main issue is then to ensure
consistency of information and product flows.
A possible alternative, in order to reach the ‘interoperable’
level of EICM, is to put into question the
hierarchical/integrated vision of the enterprise-wide control
for a more interoperable or intelligent one by postulating the
customized product as the ‘controller’ of the manufacturing
enterprise resources (McFarlane et al. 2003, Morel et al.
2005) (Fig. 2b). The product, seen as a good by
manufacturing systems, and as information and service
supplier by business systems, ensures consistency between
physical and informational flows.
Plant-control system integration
Enterprise-control
system integration
Service Flow
Good Flow
a)
Product
Flow
Plant-control system
Interoperability
Plant-control system
Interoperability
Good Flow
Service Flow
Enterprise-control
system interoperability
Plant-control system
Interoperability
material data
data information
b)
Products
c)
Enterprise-control
system agility
Enterprise-control
… system agility
Plant-control
system
agility
Plant-control
system
… agility
Fig. 2. From Integrated to agile manufacturing
Another alternative (Fig. 2c), as promoted by the IMS
community, leads to the development of new architectures
based on the consideration of highly distributed, autonomous,
adaptable and efficiently cooperating units integrated by a
plug-and-operate approach, as done in multi-agent (Marik &
Lazansky 2006) and Holonic Manufacturing Systems (Deen
2003). Such an approach is also currently studied by the
European Project “Pabadis-Promise”, which aims at
hal-00321465, version 1 - 15 Sep 2008
Author manuscript, published in "17th IFAC World Congress, Séoul : Korea, Republic of (2008)"
DOI : 10.3182/20080706-5-KR-1001.3154
extending the idea of distributed control to an innovative
architecture which incorporates both resource and product
(http://www.pabadis-promise.org). Emerging infotronic
technologies embedded into product-driven control
(McFarlane et al. 2003) bring more or less research results
closer to actual deployment: Radio Frequency IDentification
(RFID), wireless networking, modern PLC and industrial PC
support of multi-agent systems…
This paper focuses on the possibilities to implement the
product-driven control concept with such infotronics
technologies. After a description in section 2 of the concept
of product-driven control, a comparison is made in section 3
between RFID tags and Wireless Sensor Networks (WSN)
motes to foresee which product intelligence levels can be
implemented. Section 4 presents a case study on which
experiment are being made with WSN motes.
2. PRODUCT-DRIVEN CONTROL
As the work presented in this paper is mainly focused on the
implementation of a product-driven control, this part aims
first at describing the concept.
2.1 Intelligent versus smart product
Considering an active role of the product leads to give it a
form of technical intelligence (Karkkainen et al. 2003),
which corresponds, according to (Wong et al. 2002), to:
1 Possess a unique identity,
2 Be capable of communicating effectively with its
environment,
3 Be able to retain or store data about itself,
4 Deploy a language to display its features, production
requirements etc.,
5 Be capable of participating in or making decisions
relevant to its destiny.
In function of these points, two levels are defined in Wong et
al. 2002:
- Level 1 Product Intelligence allows a product to
communicate its status (form, composition, location, key
features), i.e. it is information oriented. Level 1
essentially covers points 1 to 3 of the intelligent product
definition above.
- Level 2 Product Intelligence allows a product to assess
and influence its function (e.g. self-distributing inventory
and self-manufacturing inventory) in addition to
communicating its status, i.e. it is decision oriented.
Level 2 therefore covers points 1 to 5 of the intelligent
product definition above.
From an operational point of view, things can be very
different because it seems to be difficult to implement
directly into smart products all aspects of product
intelligence. At this time, much embedded devices have
neither enough processing power nor the ability to
communicate all the required information for the
manufacturing. For these reasons, some other cases can be
envisaged if active entities reside in computers and are
remotely linked to physical products and machines. Indeed,
some multi-agent manufacturing systems are already
implemented in real industrial environment (McFarlane et al.
2003), but there are some constraints, related for example to
the reliability of RFID: successful read rate is not yet 100%,
and for this reason, the system may not be fully observable.
In such an approach, the product is considered as central to
the automation rationale, and is logically provided with
information, decision and communication capabilities in
order to make it active in the scheduling and the execution of
its manufacturing operations (point 5 of Wong et al. 2002).
The system is then said « product-driven ». Holonic
Manufacturing Systems (HMS) constitute a repository to
formalize this concept of product-driven control.
2.2 Holonic Manufacturing Systems
Koestler (Koestler 1967) introduced the concept of the
Holon, which is an entity capable of functioning as a whole,
while simultaneously acting as a part of a whole in a
hierarchically ordered system. In other words, a Holonic
system is a combination of an heterarchical system with
centralised elements. Based on this concept, the IMS
community, especially in the area of Holonic Manufacturing
Systems (Valckenaers 2001, Deen 2003, Leitao & Restivo
2006) promotes conceptual architectures, which tend towards
providing manufactured product with an intelligent
behaviour. These HMS (Babiceanu & Chen 2006) are
distributed systems which consider holons, which can be
autonomous production units, cooperating to make products
in a dynamically reconfigurable environment (McFarlane et
al. 2003). In the HMS reference architecture PROSA (Van
Brussel et al. 1998), types of holons are resource holons,
order holons, staff holons and product holons, this last
concept showing explicitly the active role of products.
A very interesting point with HMS is that Chirn and
McFarlane evaluated that this approach can provide higher
reconfigurability and modularity when facing series of design
changes (Chirn & McFarlane 2005).
2.3 Product-Driven Automation
Following conceptual guidelines of HMS, the approach used
in this work focuses on the design of a product-driven
distributed control system (Fig. 3) (Pétin et al. 2007), which
is based on the cooperation between:
- product controllers which control the manufacturing
routes according to a scheduled list of operations the
product has to undergo; these controllers are specific for
each product occurrence in order to take into account their
customization,
- resource controllers which ensure correct execution of
transport and transformation operations and provide the
product controllers with accurate reports; control
flexibility relies on tuning call parameters of the
functional objects which coordinate and control the
elementary operations, or on downloading specific control
policies embedded into products.
hal-00321465, version 1 - 15 Sep 2008
Requests from products / Reports from resources
Product
control
Product
control
Product Material flows
Resource
Control
Resource
Control
Resource
Control
Product/Process
Information flows
Fig. 3. Product-driven control architecture
This cooperation consists in the exchange of requests of
operations (noted RQ) emitted by product controllers to
resource controllers, and reports of operations (noted RP)
emitted by resource controllers to product controllers.
The definition of these controllers are founded, on the one
hand, on the modelling of the manufacturing system
capabilities which describe the system topology and the
manufacturing operations performed by each resource, and,
on the other hand, on the modelling of product requirements
in terms of the operations it has to undergo. Such controllers
can be automatically and formally written by the use of the
product-driven control synthesis, as proposed by Pétin et al.
(2007). This synthesis is out of the scope of this paper which
focuses on the implementation aspects of the product-driven
control.
3. TWO IMPLEMENTATION TECHNOLOGIES
Many technologies can be tried to implement the concept of
product-driven control. Amongst them, this part aims at
comparing RFID tags and Wireless Sensor Networks nodes
possibilities, as given by vendors in technical descriptions.
3.1 RFID tags
RFID corresponds to an automatic identification technology
which relies on the remote reading and writing of information
on electronic tags (also called RFID tags or transponders)
(Finkenzeller 2003). RFID tags are at least composed of a
chip and an antenna. In general, the chip contains a processor,
a memory and a radio transmitter (Fig. 4).
Radio Processor
transmitter Memory
Antenna
Fig. 4. Overview of an RFID tag structure
Some cheaper tags, which are the most used, are said
“passive” because they have no internal power supply, do not
contain an integrated circuit. They can be used for discrete
identification. Many applications in product tracking,
inventory systems and libraries can be found (see for example
http://www.rfidjournal.com).
3.2 WSN motes
Emerging infotronics technology, as advances in
microelectronics and wireless communications, have recently
enabled the design of very tiny sensors. Such autonomous
sensors nodes embed power supply, sensing, data processing,
and wireless communication components (Akyildiz et al.
2002) and are used to build Wireless Sensor Networks
(WSN). They are commonly called ‘motes’ (Fig. 5). With
their capacities, motes can sense their physical environment,
receive messages via the wireless network, and even react by
making a decision or sending messages.
Power supply (power management)
Communication
unit
Processing unit
& memory Sensing unit
(wireless network
protocols) (OS & algorithm) (filtering and signal
adapting)
Fig. 5. Functionnal view of mote components
WSN can be found into numerous military, environmental,
human centric, robotics or logistics applications (Arampatzis
et al. 2005).
3.3 Implementation of product intelligence with tags or motes
Both technologies present interesting capacities which could
enable a more or less direct implementation of the concept of
product-driven control into a physical product.
With the help of the literature and the description given by
vendors about RFID (Finkenzeller 2003) and WSN (Akyildiz
et al. 2002) technologies, Table 1 summarizes the abilities
presented in technical descriptions of passive RFID tags,
active RFID tags, and WSN motes to implement the various
aspects of the product technical intelligence defined in
(McFarlane et al. 2003) and presented in section 2. A passive
RFID tag seems to be able to implement Level 1 product
intelligence, while an active RFID tag containing a processor
or a WSN mote seems to be able to implement Level 2
product intelligence.
Implementing product-driven control implies that, between
products and resources, communications can effectively
occur at each time. While the RFID technology needs a direct
communication between tags and antennas (in this case,
communications are limited by the existing infrastructure),
ad-hoc organisation of WSN motes can be used to propagate
messages. Such an ad-hoc organisation seems to be more
flexible (as architecture, one bridge can be enough). For these
reasons, WSN motes have been chosen in this study to
experiment the implementation of product-driven control on a
particular case study.
hal-00321465, version 1 - 15 Sep 2008
Table 1. Comparison between RFID tags and WSN motes possibilities
Passive RFID tag Active RFID tags Mote
1 Possess a unique identity Yes Yes Yes
2 Be capable of communicating
effectively with its environment
Yes, data can be requested by an
RFID reader
Yes, data can be requested by an
RFID reader
Yes, data can be send via UDP
protocol
3 Be able to retain or store data about
itself Yes, contains a memory Yes, contains a memory Yes, contains a memory
4
Deploy a language to display its
features, production requirements,
etc.
No, the memory only contains data,
not information
Yes, the processor can interpret
memory data into product
information
Yes, the processor can interpret
memory data into product
information
5
Be capable of participating in or
making decisions relevant to its
destiny
Not able to make decision Yes, able to make a decision using
an embedded algorithm
Yes, able to make a decision using
an embedded algorithm
Aspects of technical intelligence
4. CASE STUDY
The implementation of the concept of product-driven control
is tested with WSN motes in this paper on a scenario using
the Flexible Assembly Cell case study of the AIP-Primeca
Lorraine (http://www.aip-primeca.net).
4.1 Presentation of the AIPL Case Study
The cell involves six workstations which are interconnected
by a conveyor: one station for pallet loading, four similar
assembly stations, and one station for pallet unloading (Fig.
6). Six different product families can be assembled (Fig. 7).
Each workstation is able to perform from 1 to 4 assembly
operations and involves a vacuum generator and three air
cylinders to handle parts and products.
Loading workstation n°0 Workstation n°1 Workstation n°2
Unloading workstation n°5 Workstation n°4 Workstation n°3
Fig. 6. AIPL Flexible Assembly Cell
Product 01,09
Product 60,88,09
Product 60,88,11,10
Part 09
Part 01
Part 88
Part 11
Part 60
Part 10
Fig. 7 AIPL Product types
Each pallet is equipped with a P-Particle© WSN mote
(http://particle.teco.edu/) which implements the control part
(‘intelligent part’) of products. A restriction is made so that
each product will only go on one pallet during its assembly.
Workstations are equipped with a Programmable Logic
Controller (PLC), which implement resource controllers. The
communication between product motes and resource
controllers is ensured by an XBridge© which forwards UDP
packets (used for the motes to communicate) from the WSN
to the Industrial Ethernet and vice versa (Fig. 8).
Mote b
Mote a WSN/ Ethernet bridge
WSN
Industrial Ethernet
Workstation 1 PLC
Workstation 4 PLC
SCADA System
MES Server
MES Database
Conveyor PLC
Fig. 8. Principle of the platform technical architecture
As seen in Fig. 9, this platform, currently under specification
and development, plans product controllers to exchange
Requests (RQ) and Reports (RP) with their environment.
Product a program Supervisor
Conveyor PLC program
Work station 4 PLC
program
Work station n PLC
program
Work station 1 PLC
program
RQW_op60
RPW_WS4
RQT_OP60_WS4
RPT_WS4
RQ_op60_WS4
RP_op60_WS4_time_date
OPC Serve r
WS4 variables
WS1 variables
WSn variables
OPC items
Conveyor variables
RQ_config MES
New product process planning
Fig. 9. Principle of the platform applicative architecture
To validate the implementation of level 1 and 2 of product
intelligence, this paper focuses mainly on the product
behaviour and communication. As the intelligent part of the
product is implemented into motes, Teco Particle Analyser
software is used to configure motes and to analyse their
communications with external applications.
Motes are used to implement product intelligence only during
the manufacturing. Once the product is manufactured, the
corresponding mote memory is unloaded in order to store
traceability information into the MES. The mote is then
reconfigured in order to be used with a new product (Fig. 10).
hal-00321465, version 1 - 15 Sep 2008
Initial
A0 - Unconfigured
mote
A2 - Manufacturing of the
product / storage of p roduct
traceability information
[end of configuration]
[end of manufacturing]
[end of unloading]
A3 - Unloading of the
product / downloading
traceability information
A1 - Mote configuration /
emb edding product
p rocess p lan
[new product to be made]
Fig. 10. Activity diagram showing mote and product stages
during manufacturing
4.2 Implementing level 1 product intelligence
According to the definition given by Wong et al. (2002),
level 1 intelligence refers to the ability of a product to cover
points 1 to 3 of the definition: a unique identifier, ability to
communicate and to store data about the itself.
In order to test this level of product intelligence, the
configuration activity (A1 – Mote configuration / embedding
product process planning) presented in Fig. 10 is considered.
During this activity, detailed in Fig. 11, a mote which is not
configured with a product ID and process plan emits
periodically a request of configuration (NCF). Once the
manufacturing of a new product is planned by the supervisor
or the MES, the mote is reconfigurated (a new ID and a new
process plan). In order to acknowledge receipt of the
configuration, the product mote sends an ‘ELO’ message,
with the received configuration.
Product a:Product mote
Supervisor
NCF
CFG (IDProduct, Process Plan)
ELO (ID Product, Process Plan)
Fig. 11. Configuration sequence diagram
This scenario has been implemented on Particle© motes (Fig.
12). The analysis shows that the product emits a ‘NCF’
message, containing the mote ID (2.232.0.0.0.77.220.181)
corresponding to point 1 of (Wong et al. 2002). It receives
the configuration (sequence 14 ‘CFG’ with parameters {ID of
the product class, Class Number, Process Plan, …}), stores it
and is able to communicate it (points 2 & 3) by broadcasting
an ‘ELO’ message containing its ID (100 123) and its type
(97).
Fig. 12. Screenshot showing product and supervisor
exchanges for the first experiment
This first experiment shows that WSN motes can implement
at least level 1 product intelligence. A second scenario is
needed to experiment if WSN motes are able to implement
some more aspects of product intelligence.
4.3 Implementing level 2 product intelligence
As presented above, the level 2 of product intelligence
defined in (Wong et al. 2002) corresponds, in addition to
level 1 intelligence, to the ability of a product to deploy a
language to communicate and to participate in decisions
relevant to its destiny. The second experiment considers the
activity A2 of Fig. 10, in which the manufacturing is driven
by the product itself. The control is then based on the
exchange, between the product and its environment, of
Requests (RQ) and Reports (RP). A language is then defined
as follows:
- RQW_opi: request from the product: “which resource is
able to perform operation i to me?”
- RPW_WSj_opi: report from the supervisor: “the
workstation j is able to perform operation i” (the
workstation is chosen by the supervisor in function of an
optimization criteria, for example the waiting time)
- RQT_WSj_opi: request from the product to the conveyor:
“bring me to workstation j for operation i“
- RPT_WSj_opi: report from the conveyor: “you are now
at workstation j for operation i”
- RQ_opi_WSj: request from the product to workstation j:
“perform me operation i”
- RP_opi_WSj_time_date: workstation j reports to the
product: “I performed you operation i at time and date”
The ‘intelligent part’ of the product, is able to request
operations and to receive reports of operations. The order in
which the reports are emitted and the reports are waited is
defined in the control part of product in function of the
successive physical states of the product (Fig. 13) to ensure
the correct execution of the process plan. A similar sequence
is executed for each operation.
Initia l
1
[RQW_OPi]
[RPW_WSj] 2
3
[RQT_WSj]
4
[RPT_WSj]
5
Final [RP_OPi_WSj_time_date] [RQ_opi]
Fig. 13. Gerenic sequence of product internal behaviour
This internal behaviour can be formally synthesized as
described in Pétin et al. (2007), but it is out of the scope of
this paper. In addition, traceability information is stored in
the product in function of manufacturing parameters which
are given by resources.
An example of message exchange between product and
external applications is shown in Fig. 14. The product
requests resources to perform the operations scheduled in its
process plan, in function of the reports it receives.
Product a:Product mote
Supervisor
RQW_op60
RPW_WS4
Conveyor
RQT_OP60_WS4
RPT_WS4
Workstation 4
RQ_op60_WS4
RP_60_WS4_time_date
If product not at
Workstation 2
Fig. 14. Product and resource motes collaboration to perform
operation ‘op60’ on workstation 4
hal-00321465, version 1 - 15 Sep 2008
This example scenario has been successfully tested. Fig. 15
shows a screenshot with the exchange of product-driven
control messages between a product, a resource and a
supervisor. The product successively requests the operations
which are relevant to its manufacturing (for example
operations 60 and 88). Furthermore, traceability is ensured by
the storage into the product of manufacturing conditions (e.g.
operation 60 on workstation 4, at 15:35 on September 20th).
Fig. 15. Screenshot showing product message exchanges for
the second experiment
5. CONCLUSION AND OPEN ISSUES
This paper focuses on the possibilities to implement the
concept of product-driven control on a case study using
motes of wireless sensor networks. WSN, compared with
RFID, allow guaranteeing the continuity of information
availability during the overall manufacturing process of
products. The use of the Particle Analyser showed that it is
possible with motes, on a simple example, to cover level 1
and 2 of product intelligence, as defined by Wong et al.
(2007). Traceability is also ensured by storing manufacturing
operation information directly into the product.
Issues are now open on the efficiency of the concept
implementation by putting various product instances in a real
manufacturing environment. This will underline
communication and product conflict problems. The last ones
may be solved by the development of a ‘staff holon’ as
defined by Van Brussels et al. (1998). WSN sensing abilities
(temperature, distances between motes, acceleration …) will
also be used to situate and to monitor products in their
environment. Such data may be taken in account by products
in their decision making to ensure quality and to optimize
manufacturing flows.
REFERENCES
Akyildiz I.F., W. Su, Y. Sankarasubramaniam, E. Cayirci
(2002). Wireless sensor networks: a survey. Computer
networks, 38, pp. 393-422.
Arampatzis Th., J. Lygeros, S. Manesis (2005). A survey of
applications of wireless sensors and wireless sensor
networks, Proceedings of the 13th Mediterranean
Conference on Control and Automation, Limassol,
Cyprus, June 27-29, pp. 719-724.
Babiceanu R. F., F. F. Chen (2006). Development and
applications of holonic manufacturing systems: a survey,
Journal of Intelligent Manufacturing, 17, pp. 111-131.
Chirn J.-L., D. McFarlane (2005). Evaluating holonic control
systems: a case study. Proceedings of the 16th IFAC
world congress in Prague, Elsevier, ISBN 008045108X.
Da Silveira G., D. Borenstein, F.S. Fogliatto (2001). Mass
customization: literature review and research directions.
Int. Journal of Production Economics, 72, pp 1-13.
Deen, S.M. (Editor) (2003). Agent-based manufacturing -
Advances in the holonic approach, Springer, ISBN 3-
540-44069-0.
Finkenzeller K. (2003). RFID handbook: fundamentals and
application in contactless smart cards and
indentification, J. Wiley and Son, ISBN 0-470-84402-7.
International Organisation for Standardization (2003). ISO
62264: enterprise-control system integration.
Hollocks B.W., H.T. Goranson, D.N. Shorter, F.B. Vernadat
(1997). Assessing enterprise integration for competitive
advantage, ICEIMT’97, International Conference on
Enterprise Modelling and Modelling Technology, Berlin.
Karkkainen M., J. Holmstrom, K. Framling, K. Artto (2003).
Intelligent products – a step towards a more effective
project delivery chain, Computers in Industry, 50, pp.
141-151.
Koestler A. (1967). The ghost in the machine, ISBN 0-14-
019162-5.
Leitão P., F. Restivo (2006). ADACOR: A Holonic
architecture for agile and adaptive manufacturing
control, Computers in Industry, 57, pp. 121-130.
Marik V., J. Lazansky (2006). Industrial application of agent
technologies, Control Engineering Practice,
doi:10.1016/j.conengprac.2006.10.001.
McFarlane D., S. Sarma, J.L. Chirn, C.Y. Wong, K. Ashton
(2003). Auto id systems and intelligent manufacturing
control, Engineering Application of Artificial
Intelligence, 16 (4), pp. 365-376.
Morel G., H. Panetto, M. Zaremba, F. Mayer (2003).
Manufacturing enterprise control and management
system engineering: rationales and open issues. IFAC
Annual Reviews in Control, 27 (2), pp. 199-209.
Morel G., P. Valckenaers, J.M. Faure, C. Pereira, C. Diedrich
(2005). Survey paper on manufacturing plant control
challenges and issues, Proceedings of the 16th IFAC
world congress in Prague, ISBN 008045108X.
Nof S. Y., G. Morel, L. Monostori, A. Molina, F. Filip
(2006). From plant and logistics control to multienterprise
collaboration, IFAC annual reviews in control,
30 (1), pp. 55-68.
Pétin J.-F., D. Gouyon, G. Morel (2007). Supervisory
synthesis for product-driven automation and its
application to a flexible assembly cell, Control
Engineering Practice, 15, pp. 595-614.
Valckenaers P. (Editor) (2001). Special issue: Holonic
Manufacturing Systems, Computers In Industry, 46 (3),
pp. 233-331.
Van Brussel H., J. Wyns, P. valckenaers, L. Bongaerts, P.
Peeters (1998). Reference architecture for holonic
manufacturing systems: PROSA, Computers in Industry,
37 (3), pp. 255-274.
Wong C.Y., D. McFarlane, A. A. Ahmad Zaharudin, V.
Agarwal (2002). The intelligent product-driven supply
chain, IEEE International Conference on Systems, Man
and Cybernetics.
hal-00321465, version 1 - 15 Sep 2008
ليست هناك تعليقات:
إرسال تعليق