I Congresso da SBI-Agro


Mobile-agent-based Supervision in Supply Chain Management in the Food Industry


R. J. Rabelo ; L.M. Spinosa
Federal University of Santa Catarina / Department of Mechanical Engineering
Grucon / G-SIGMA - Group for Intelligent Manufacturing Systems


The guarantee of reliable and in time information about the suppliers in a supply-chain is one of the most important aspects to be faced in a virtual enterprise scenario. This paper presents a mobile-agent-based approach as a support for that, allowing the supply-chain supervision. In this approach a virtual enterprise is seen as a network of "agentified" enterprises, which means each one has some intelligence, autonomy, privacy and capacity to interact with the other nodes. This work has been developed in the scope of two European-Latinamerican projects (Prodnet-II and SCM+). In the intended approach, once a virtual enterprise in established, intelligent mobile agents are instructed about a "mission" to be executed and sent to the suppliers' site in order to supervise and collect information previously agreed with the client-enterprise. With an efficient supervision the enterprises can improve their logistics and their tactile and operational plans as well.


Mobile agents, multiagent systems, supply-chain management, virtual enterprise, supervision.



The application of the Virtual Enterprise (VE) concept by the industries has been seen as one of the more powerful strategic action in order to face the challenges of a global economy [Browne &al. 95]. In this context, the improvement of the supply-chain management (SCM) appears to be a must. The SCM involves many aspects to be dealt with. This papers aims at addressing the real time supervision of suppliers once the value-chain (the VE) is established, presenting an intelligent mobile-agent-based approach as a support for. More specifically, it intends to allow an enterprise to get reliable and in time information from the suppliers, giving conditions for a more efficient decision-making and agile reaction in the presence of unexpected problems in the chain.

Supervision is a general activity which comprises monitoring, prognosis, diagnosis and exception handling [Barata & al. 93]. Although the proposed approach can be applied on these four actions, this work concentrates the attention on monitoring and diagnosis. Monitoring in order to evaluate if a certain «task» is being executed accordingly it was planned and the diagnosis in order to identify the type of problem occurred.

This work has been developed in the scope of two international projects in the area of virtual enterprises and supply-chain management: the INCO SCM+ (Beyond Supply Chain Management in Food Industry) [SCM+] and Esprit Prodnet-II (Production Planning and Management in an Extended Enterprise) [Prodnet].



The application of the VE concept in the food industry postulates that enterprises do not produce complete products in isolated facilities, rather they operate as cooperative nodes in a network of suppliers and clients, which in turn involves many services, such as storing, packing and distribution. Each node adds a value in a so-called supply-chain. Every node is extended with means to support the information exchange among them. For a certain VE, a certain enterprise becomes the "coordinator" of the production chain. In those projects mentioned before, these means are represented by two software components: an Internal Module which connects the internal services of the existing systems of the enterprise to the Cooperation Layer, which connects the enterprise to the network. Small modifications in the existing systems are normally necessary in order they are able to interact with the cooperation module. The figures 1a and 1b illustrate this scenario.

Figure 1a - The Food chain seen as a Virtual Enterprise.

Figure 1b - The Internal and Cooperation Modules

The introduction of VE concepts in an enterprise can impose changes in the way it normally operates, specially in terms of information technology (IT), working methods and needs of new people skills. In other words, all changes arisen from a isolated, non-cooperative and selfish to a largely open, cooperative and democratic way of working. Suitable IT with truly partnership is one key aspect to establish a successful VE and hence to facilitate the SCM activity. In order to better exploit the potentialities of an intelligent mobile-agent approach within a SCM for the food industry, the following requirements have to be observed [Hardwick &al. 96][Zweben 96]:

  • Information and enterprise integration - a mobile-agent-approach requires an efficient information management flow to support a very intensive information exchanging process. The more an enterprise has its information suitably modeled and integrated - in all the involved SCM activities - the more efficient and reliable tends to be that process.
  • Real-time updating - the food industry normally makes use of sensors connected to cameras, food processing equipment, tractors, etc., in order to make an efficient and reliable SCM. In this sense, adequate communication infrastructures are necessary to support real-time information updating.
  • Standards for communication - an efficient SCM requires the enterprises exchange information with each other, aspect much more relevant when mobile-agent-based approaches are utilized. However, different enterprises which got members of a VE can use different information technologies. Thus, the use of standards is an essential requirement to make possible a more "direct" communication among them, both in terms of code and semantic. In the Food Industry a great effort still have to be done to study, define and divulge standards accordingly its domain needs.



Intelligent mobile-agent is a very recent technology and whose concept comes from the areas of Multiagent Systems (MAS) / Distributed Artificial Intelligence [Huhns 87] [Demazeau 90]. MAS approach provides methodologies, techniques and methods which allow to model complex, distributed and cooperative domains. Several works applying MAS have been developed in many domains, such as [Rabelo & al. 94] [Jennings 94] [Fischer &al. 96] [Spinosa &al. 97]. MAS is based on the notion that the system intelligence is distributed among a set of agents which cooperate between each other in order to coordinate their actions and objectives so that problems can be solved.

In spite there is not a consensus about what an agent is, in this paper an agent has been considered as a piece of software with the following basic properties: autonomy (an agent can act and control its actions by itself), ability to interact (an agent can communicate with information sources and other agents to exchange information) and reasoning ability (an agent has some means for decision-making about how to solve a problem) [Spinosa 96] [Rabelo 97]. These three essential properties make the difference between an agent and a normal computational process.

A mobile-agent is a specialized kind of agent. It is an agent which can move itself from one site to another (keeping those three essential properties) and can take the current memory status with it during its dislocation [Camarinha-Matos & al. 97]. The basic idea of the presented approach for SCM supervision is as follows: once the supply chain is formed, the VE enterprise coordinator creates the mobile-agents, instruct them about a «mission» and send them to the VE suppliers' sites in order to perform their missions. It means that a mobile-agent life cycle is composed by five general phases: creation, mission programming, dislocation, mission execution and mission ending. This mission ending in turn involve two actions which can be performed after the execution of its mission: the mobile-agent (self-)killing or its return to the site which has originally launched it.

Because a VE represents a cooperative action of enterprises, the introduction of a "foreign entity" into the suppliers' sites is previously negotiated / agreed between the affected VE members during the VE forming.

Mobile-agent decides accordingly its mission. A mission is represented by a set of goals each mobile-agent must execute at some supplier's site. Examples of goals/actions a mobile-agent can perform are: monitoring a machine; getting some information from the local databases or WWW pages during a certain period of time; triggering some local reasoning processes; inquiring or negotiation with an user / subsystem; remote dialoguing with the enterprise who has sent it as well as with other mobile-agents and other "normal" agents or processes running at the own supplier's site.

A mission may change along its execution. This can be made either autonomously and/or ordered by the enterprise. Part of the mission information is often related with the definition of the access rights a mobile-agent can have at a supplier's site.

The modeling of a mobile-agent normally resorts on the object-oriented programming (OOP) paradigm. It means the mobile-agent knowledge are represented in terms of attributes and methods. The attributes indicates its characteristics, such as identification, site of origin, restrictions of time to perform its mission, specification of the access rights at the explored site, high level communication protocol ontology, etc.. The methods describe the agent's functionalities, that is, the actions an agent can do (like the examples given above).

The mobile-agent's mission programming can be made regarding the type of the mission. It means that «classes» of configurable missions may exist / be developed along the time. In the approach proposed in this paper, creating mobile-agents means instantiating the agent model accordingly the enterprise coordinator objectives / mobile-agent mission and related characteristics. Once an agent finishes its mission it can either kill itself at the own supplier site or it can return to the enterprise coordinator site (its "head-quarter"). This behavior will also depend on the agent's mission.

Because an enterprise / supplier may belong to several supply-chains simultaneously, many mobile-agents may be located temporary at a certain site. Furthermore, depending on the mobile-agent mission, it may move itself to other enterprises with the information it had in the previous site. It is specially useful when more complex decisions in which a global view upon the supply chain is necessary.

The main advantage of a mobile-agent-based approach from the other approaches is the flexibility and comfort the first one provides to an enterprise. One agent placed at one site can take local decisions, filter unnecessary information, alter its mission strategically, learn about a supplier behavior, dislocate itself to another site, actions which make the supply-chain supervision potentially more efficient and intelligent. Besides that, based on such aspects, the application of this approach appears to provide a more rational utilization of the network, which means lower costs.

3.1 Some application of intelligent mobile-agent

As it was mentioned before, a mobile-agent-based approach can be utilized to provide an enterprise with reliable and in time information about the supply-chain. With accurate information an enterprise can feed its strategic, tactile and operational plans with updated data, one key aspect for a rapid decision-making and agile reaction. However, how intelligently an enterprise make use of the information gathered is up to each one.

In the context of the food industry, many "services" can take advantage of intelligent mobile-agents, specially when they are programmed to play as information providers (push technology). Some examples are:

  • Disease/Pest Control - one mobile-agent can be created and sent to nodes (governmental institutions, research institutes, etc.) with the mission to retrieve and inform the enterprises about the evolution of pests and diseases in different regions;
  • Whether forecasting - several institutions in the world are dedicated to study and prevent natural catastrophes or even whether forecasting. A mobile-agent can constantly collect these information and send them immediately to the farmers and commercial cooperatives;
  • Logistics - a mobile-agent can supervise all steps involved since the first production phase till the last one; in other words, all the activities which provoke material flow between a point of origin and a point of consumption. Hence, it can involves the production scheduling, inventory levels control, warehousing planning, transportation, etc.;
  • Business Broking - new business opportunities can be searched / found in the net through broker agents which visit or make a search in some sites;
  • (Semi-)automatic negotiation - several works have been developed to provide an agent with negotiation capabilities. A (mobile-)agent can be used to make the initial phases of negotiation (usually to define a rough delivery time and prices about a product) in large open markets, leaving the human user only with the more "feeling-based" negotiation part.

3.2 An implementation approach

Since an agent can be easily modeled resorting the OOP paradigm, several object-oriented languages can be used for that. The most popular one being used is C++. Nevertheless, in this last times, the JAVA language starts to gain importance and many followers. Given that it is an interpreter, JAVA is able to interoperate in foreign systems (since they have a JAVA process too !), which is one of the most critic points of the mobile-agents approach. It has to be noted that there are already available in the market and in the Internet some "kits" for C++ and JAVA agents generation and infrastructure communication management.

If the whole supply chain's communication environment is homogeneous, the implementation of such ideas can be also supported by means of the Active-X, from Microsoft, which can be viewed as a concurrent of JAVA agents. Further, the information gathered can be sent through private networks, the Internet or even via satellite.

Nowadays it seems to be reasonable to look for commercial or shareware software already available in the market and in the Internet. The so-called push casting systems have arisen as very interesting solutions to provide information from the selected sites without the necessity to stay in front of the computer surfing on the net. The user can program the software for specific information he/she needs, in the time he/she specifies. Some of these push systems are Pointcasting, Webcasting, Castanet, Air Media and Free Loader. Some of them will be already integrated in the next versions of the WWW browsers. It has also to be considered the existing of specialized information providers sites, which sell information for registered people. This all means that, in a mobile-agent-based context, these kind of solutions can be integrated with the agent's knowledge-base to improve its potentialities and take advantage of already available software.

Any agent-based application requires a specific high-level communication protocol which can support the conversation / information exchange between the agents. Dedicated protocols or some "standards", like KQML (Knowledge and Query Manipulation Language) [Finin & al. 94] can be used. However, if the information to be exchanged is not "homogeneous", an internal messages format has to be established, like KIF (Knowledge Interchange Format) [Genereseth & al. 94]. In the case of those pushing systems, an initiative has already been taken in order to create a "standard" for the information format, called CDF - Channel Definition Format, which will facilitate tremendously the information treatment and management.



This paper presented the initial studies on how a mobile-agent-based approach can allow an enterprise for getting reliable and in time information about the supply-chain as well as how other high level services can use the information gathered. With accurate information an enterprise can improve the quality of its strategic, tactile and operational plans, supporting rapid decision-making and agile reaction.

In order to take advantage of the potentialities of such approach an enterprise has to have its involved systems integrated with suitable technologies. Moreover, since several kinds of information are supposed to flow in the network, the utilization of information standards seems to be extremely important.

This paper has emphasized the use of already available IT to model and support the mobile-agent approach for SCM, which means more stable and lower costs solutions.

Agent-based solutions are very sensitive to be killed or get crashed. Therefore, the use of robust communication infrastructures and agents management systems is fundamental in order to guarantee the overall system reliability.

One of the most controversy aspects in this approach in the fact that a mobile-agent can move itself to another site taking the information it gathered from the previous ones with it. Due to the problems of privacy and confidentiality, strong security mechanisms have to be provided to protect the mobile-agent from the external "spying".

In spite this work have been applied on the food industry, it can be also utilized in many other areas, such as manufacturing, concurrent engineering, etc..



We thank CNPq (The Brazilian Council for Technological Development and Research) and the European Union for their financial support of this work.



  • [Barata &al. 93] M. Barata, T. Rauber, A. Steiger-Garção, Prognostic and Monitoring System for CNC Machines, Proceedings Tooldiag' 93, International Conference on Fault Diagnosis, France, 1993.
  • [Browne &al. 95] Browne, J.; Sackett, P.; Wortmann, J., Future manufacturing systems: Towards the extended enterprise, Computer in Industry, Special Issue on CIM in the Extended Enterprise, Vol. 25 N 3, pp.235-254, 1995.
  • [Camarinha-Matos &al. 97] L. M. Camarinha-Matos, W. Vieira, Mobile Agents and Remote Operation, to appear in Proceedings INES'97, Budapest, Sept 1997.
  • [Demazeau &al. 90] Demazeau and J.-P. Müller (Ed.),  Distributed AI, Vol. 1, Elsevier North-Holland, 1990.
  • [Finin &al. 94] T. Finin, R. Frizson, KQML - A Language and Protocol for Knowledge and Information Exchange, Technical Report CS-94-02, Computer Science Department, University of Maryland, 1994.
  • [Fischer &al. 96] K. Fischer, J. Müller, I. Heimig, A. Scheer, Intelligent Agents in Virtual Enterprises, Proceedings PAAM'96, pp.206-223, 1996.
  • [Genesereth & al. 94] M. Genesereth, N. Singh, A knowledge sharing approach to software interoperability, Technical Report RL-94-6, Logic Group, Department of Computer Science, Stanford University, 1994.
  • [Hardwick &al. 96] M. Hardwick, D. Spooner, T. Rando, C. Morris, Sharing Manufacturing Information in Virtual Enterprises, Communications of the ACM, Vol. 39 N 2, pp.46-54, Feb 1996.
  • [Huhns 87] M.N. Huhns, Distributed Artificial Intelligence, Pitman Pub., Morgan Kaufmann, 1987.
  • [Jennings 94] N. Jennings, Cooperation in Industrial Multi-Agent Systems, World Scientific Publishing Co., 1994.
  • [Prodnet]
  • [Rabelo &al. 94] R. J. Rabelo, L. M. Camarinha-Matos, Negotiation in Multiagent Based Dynamic Scheduling, International Journal on Robotics and CIM, Vol. 11 N 4, pp.303-310, Pergamon, 1994.
  • [Rabelo 97] R. J. Rabelo, Um Enquadramento para o Desenvolvimento de Sistemas de Escalonamento Ágil da Produção - Uma Abordagem Multiagente, Ph.D. Thesis, Universidade Nova de Lisboa, Portugal, 1997.
  • [SCM+]
  • [Spinosa 96] L.M. Spinosa, Contribution à la modélisation d'entreprises manufacturières distribuées fondée sur une approche multi-agents. Thèse de Doctorat, Université d'Aix-Marseille III, 1996.
  • [Spinosa &al. 97] L. M. Spinosa, , For a Decision Support System to Distributed Manufacturing Systems: a multiagent and CIMOSA based approach, to appear in Proceedings IFAC/IFIP Conference MCPL'97 Management and Control of Production and Logistics, Brazil, Sept 1997.
  • [Zweben 96] M. Zweben, Intelligent Agents, Computer Integrated Manufacturing and Engineering, Vol. 1 N 2, pp.14-15, 1996.



Mr. Ricardo J. Rabelo

Science in 1984 at the Federal received his B.Sc. on Computer University of Santa Catarina (UFSC), worked as a consultant for several Brazilian companies as a collaborator of Grucon / UFSC and he finished in 1997 his Ph.D. Thesis at the New University of Lisbon (Portugal) on Robotics and CIM. His main interest are: agile scheduling, virtual enterprise, multiagent systems and information integration.

Dr. Luiz Marcio Spinosa

Received his B.Sc. on Computer Science in 1986, the M.Sc. in Mechanical Engineering in 1991 on CAD/CAM/CIM at the UFSC, the D.E.A in 1992 on Automation & Information Technology and the Ph.D. in 1996 on Production & Information Technology at the University of Aix-Marseille III (France). His current research interest are: virtual enterprise, intelligent manufacturing systems, multiagent systems and industrial modeling approaches.