The scope of application of multi-agent systems in the process industry: three case studies.

Transcription

1 See discussions, stats, and author profiles for this publication at: The scope of application of multi-agent systems in the process industry: three case studies. Article January 2004 Source: DBLP READS 43 6 authors, including: A. Aldea Oxford Brookes University 42 PUBLICATIONS 312 CITATIONS Laureano Jiménez Universitat Rovira i Virgili 182 PUBLICATIONS 1,327 CITATIONS SEE PROFILE SEE PROFILE David Riaño Universitat Rovira i Virgili 83 PUBLICATIONS 443 CITATIONS SEE PROFILE All in-text references underlined in blue are linked to publications on ResearchGate, letting you access and read them immediately. Available from: Laureano Jiménez Retrieved on: 09 May 2016

2 Expert Systems with Applications 26 (2004) The scope of application of multi-agent systems in the process industry: three case studies A. Aldea a, *, R. Bañares-Alcántara b, *, L. Jiménez b, A. Moreno a, J. Martínez a, D. Riaño a a Department of Computer Science and Mathematics, Universitat Rovira i Virgili, Tarragona, Spain b Department of Chemical Engineering, Universitat Rovira i Virgili, Tarragona, Spain Abstract It has been suggested that multi-agent systems (MAS) are specially adequate for the solution of problems with a dynamic, uncertain and distributed nature. Within industrial applications, there is a wide spectrum of problems with these characteristics, in particular those covering the modelling of artifacts, methodologies and organisations. Three case studies on the application of MAS in the process industry are presented. All of them relate to tools that are being developed to support very diverse core tasks in the process industry (and, by extension, the petroleum industry): An intelligent search system composed of Internet information agents which are able to gather, compile and classify data available in web pages related to a specific technological domain. This search engine is the first step towards the construction of a knowledge management platform that will allow chemical process industries to improve their capabilities to monitor, predict and respond to technological trends and challenges. A system to support the concurrent design of processes, to ease communication between engineers who perform design and keep them informed about the progress of the design process. A tool to support the configuration of work teams. This tool will assist in the configuration of the most suitable team for a specific project. It takes into account the ideal size of the team (2 to n members); its specific composition (managers, engineers/scientists, assistants, etc.); and the proposed type of organisation (centralised, tree hierarchy, etc.). These case studies are representative of a large variety of the possible applications of agent based systems in the process industry. q 2003 Elsevier Ltd. All rights reserved. Keywords: Multi-agent systems; Knowledge management; Ontologies; Design and re-design of processes; Social simulation 1. Introduction In the last 25 years we have seen the appearance of several paradigms to design software systems such as Procedural Programming, Structured Programming, Object Orientation and Component-Based software. Agents (Weiss, 1999; Wooldridge, 2002) are now being championed as the next generation paradigm to design and build complex and distributed software systems. An agent based architecture provides additional robustness, scalability, flexibility and is particularly appropriate for problems with a dynamic, uncertain and distributed nature. In particular, they seem to be the ideal computational model for developing software for Internet, an open networked system with no single controlling organisation (Jennings, 2000). Lastly, agent based * Corresponding authors. Address: Department of Chemical Engineering, Universitat Rovira i Virgili, Tarragona, Spain. address: aaldea@etse.urv.es (A. Aldea). architectures allow the incremental development of modular systems not only because of the modular nature of the agents, but also because of the possibility to incorporate legacy code by wrapping it within an agent interface. In a multi-agent system (MAS) agents interact with one another to achieve their individual objectives by exchanging information, cooperating to achieve common objectives or negotiating to resolve conflicts. Alternative flexible patterns of interaction have been used such as the Contract Net Protocol (Reid & Smith 1980), where a task is advertised by a coordinating agent and is assigned to the agent that makes the best bid. However, the details of all possible interactions between agents cannot be foreseen a priori and as a consequence 1. agents need to be able to make decisions about their interactions at run-time, and 2. the organisational relationships between agents need to be represented explicitly (e.g. peer member in a team, /$ - see front matter q 2003 Elsevier Ltd. All rights reserved. doi: /s (03)

3 40 A. Aldea et al. / Expert Systems with Applications 26 (2004) manager, coordinator) by means of constructs such as roles, norms and social laws. Multi-agent architectures have been reported for applications where distributed decision making is advantageous such as flexible manufacturing control, planning and scheduling, diagnosis, process design, modelling and diagnosis, and supply chain management, see for example Julka, Karimi, and Srinivasan (2002) for a refinery application. To date, most of these systems are exploratory in nature. 2. An ontology-based multi-agent search engine In technology intensive industries (e.g. pharmaceutical, biotechnology and most of the process industries) new domains emerge, evolve, mature and decline very fast. Thus, decisions are based not only on well-established information but also on uncertain, dynamic and incomplete information. This decision making would benefit from the capability to monitor, predict and respond to technological, product and market trends and changes, as this would allow to evaluate the current status of a domain and to find new products, services and areas of growth. Much of this information and knowledge is now available from the Web (Internet þ Intranets), however, it is difficult to exploit it to its full due to the large amount of unstructured, redundant and irrelevant information available. The objective of this research is to develop a support tool that extracts, processes, classifies and updates the knowledge contained in the Web with little supervision from the user. As a result, we expect companies that use these knowledge management practices to benefit by improving their competitiveness (see Bañares- Alcántara, Kokossis, Aldea, Jiménez, and Linke (2003)). The combination of two techniques is being used to retrieve information from the web and thus achieve this goal: 1. Ontologies as a representation formalism that eases information retrieval and integration and ensures consistency (Fensel, 2001); and 2. Intelligent Information Agents to search and process knowledge from the Web, generate new knowledge and express it as a new ontology (Klusch, 2001) Ontologies An ontology is a formal explicit hierarchical representation of the knowledge about a specific domain. The essential concepts, properties and characteristics are represented by a vocabulary of entities, classes, properties, functions and their relationships. This static vision of ontologies may not be adequate for some domains such as technology intensive industries which are highly variable. Changes in these domains make the ontologies to become old-fashioned, incomplete, or imperfect. Dynamic ontologies (Heflin & Hendler, 2000) have emerged as an alternative to overcome these problems. In Section 2.2, we describe a multi-agent search engine aiming at managing dynamic ontologies as the information available in the Internet evolves. For this purpose, a domain ontology is defined with OntoEdit (Sure, Staab, & Angele, 2002). This ontology represents the field of interest of the user, and it is used to conduct the search following a master slave procedure that decomposes the domain ontology in several query ontologies. After the search, a response ontology is generated for each query ontology. Finally, all the response ontologies are combined in an information ontology. All the above ontologies are stored in RDF-S (Resource Description Framework) which is becoming the de facto standard for representation of ontologies ( see for example Fensel (2001)) Multi-agent search system architecture We have developed an ontology-based MAS information search prototype that employs an ontology defined by the user to perform web searches. The architecture of the search system is shown in Fig. 1. The agents that perform the search for information and its further classification are: The Internet agents. Given a concept with some associated properties, i.e. a query ontology, this type of agent searches the web with the help of one or more internet search engines and returns the web pages related to this concept in the form of a response ontology. The retrieved web pages are sorted by their weight calculated according to the number of properties of the concepts that can be found in the document text and to their position in the document, i.e. whether they are in the title, headings, plain text, etc. The rationale is to have a two-level relevance evaluation: firstly, by retrieving pages found relevant by a search engine (e.g. Google), secondly, by analysing the contents of those pages in more detail in order to assess more precisely their potential relevance for the user. The Coordinator Agent. This agent receives as input the domain ontology created by the user, and is responsible for partitioning it into several query ontologies which are sent to the Internet Agents. The Coordinator Agent is then in charge of generating the knowledge base that stores all the response ontologies returned by the Internet Agents, and merge and classify them into the information ontology. This last part is currently under implementation (see Kokossis and Paul (2002) for more details). The prototype has been implemented using JADE (Java Agent Development Environment, (Bellifemine, Poggi, & Rimassa, 1999, JADE, 2002)). JADE is a collection of Java libraries that ease the development of multi-agent systems

4 A. Aldea et al. / Expert Systems with Applications 26 (2004) Fig. 1. Ontology-based search engine. (MAS) that follow the specifications suggested by FIPA (Foundation for Intelligent Physical Agents, FIPA is a non-profit organisation that tries to promote the interoperability of agents by defining a set of standard specifications that define aspects such as the architecture of a MAS, the agent communication language (FIPA-ACL) or the communication protocols that agents can follow in their interactions Preliminary results The MAS search prototype was initially tested with a biosensor domain ontology. It is worth mentioning that the ontology-guided search system is independent of the domain and can be applied to any other domain. If the concept to be searched were Biosensor (the top concept of the ontology), the Coordinator Agent would split the domain ontology into 71 query ontologies (one for each class or subclass of the domain ontology). In the case study reported in this section we have analysed a subset of this ontology: The Biosensor class and one of its children (the Company class) together with all its subclasses; as a result, the system works with 10 query ontologies, which are shown in Fig. 2. Several searches were done with different numbers of Internet agents and deadlines. In our prototype every agent Fig. 2. Domain ontology for the case study.

5 42 A. Aldea et al. / Expert Systems with Applications 26 (2004) Fig. 3. Results for four classes of the biosensor ontology. analyses all the links associated with the first 10 results retrieved by Google (this number is configurable, though). Different experiments were made using 2 and 5 Internet agents with different search time deadlines (0.5, 2.5 and 5.0 min). Fig. 3 shows the average number of retrieved web pages in searches associated to four classes of the biosensor ontology: biosensor, company, research and manufacturer. It can be observed that the search results improve with larger values for deadlines and a bigger number of Internet agents; however, the influence of the deadline decreases when there are more Internet agents available to perform the search. To validate the results, we have manually visited the web pages related to these classes and observed that the more relevant web pages were indeed those with larger assigned weights. However, the Internet agents also retrieved irrelevant web pages, these pages were placed at the bottom of the relevance list though. As it stands, the user must decide on a cut value for the weights below which results are most likely irrelevant. This value cannot be generalised to other searches, thus a true solution rather than a patch requires that the web page analyser within the Internet agents be refined to eliminate as many irrelevant pages as possible. From the results of the case study we can conclude that the prototype system finds relevant information (although we cannot prove that it finds all of it!) and successfully prioritises it. 3. A concurrent engineering design support architecture It has been proposed that the integration of the design process can be achieved based on a concurrent engineering approach. One of the primary issues for the application of concurrent process engineering is to count with an environment to communicate engineers who perform different tasks at different sites. This goal may be achieved by sharing design information through an agent-assisted environment. A multi-agent architecture has been under development to assist in the integration of process design for the prototype system CPEDSYS. It is based on cooperating intelligent entities which provide the necessary information and data, and assist in the decision making process using domain-specific knowledge, both distributed among the entities, and centralised but accessible to all of them Proposed architecture As a Concurrent Engineering Process Design System, CPEDSYS is composed of process design tools, process design teams, process design documents, and workflows. In order to ensure the proper realisation of the concurrent engineering, the system requires the integration of teams, workflows, documents and tools to an extent that they can operate within a synchronised, parallel and standardised procedure. The proposed architecture is composed of a number of programs that are distributed in different computers connected to a local network. An interface for chemical process flowsheet construction (structure dimension) using AUTOCAD and other process design tools such as HYSYS for process simulation and AXSYS for PFD and P and ID data generation are also used as commercial tools. The core teams are responsible for the processing of queries about the plant. The management dimension which includes the management team agents and the core team agents encompasses rules for workflow control, procedures and data and information for local design teams and reasoning systems. These local design teams generate actions that can be applied over tools, documents, data and information. These can be distributed throughout the computer network so as to reduce the processing load. Software components also include a process organiser, tools for workflow management, and a tool for economic

6 A. Aldea et al. / Expert Systems with Applications 26 (2004) evaluation in addition to the commercial process design tools mentioned above (Gelete & Bañares-Alcántara, 2002). 4. A tool to support team configuration When a new complex project is started, the Project Manager is put in charge of partitioning this project into tasks and selecting the people who will perform them. The correct selection of people to integrate a team within a complex engineering project is not trivial because it should include not only technical competence and availability aspects, but also personal and social characteristics of each potential team member. The success of a project is greatly due to the personal responsibility of each member, but is also strongly related to adequate communication, collaboration and co-operation between the individual team members (Biegler, Grossmann, & Westerberg, 1997). In addition to social and external factors, emotions play a critical role in rational decisionmaking, perception, human interaction, and human intelligence (Picard, 1995). The emotional state of a person varies with time; furthermore, given the same circumstances, the reactions of different people can be quite different. We think that the MAS technology could be very helpful in the simulation of human social behaviour given its capability to account for characteristics such as autonomy, co-ordination, and communication (Weiss, 1999; Wooldridge, 2002). Thus, we propose to represent a team member with a software agent that includes not only technical competence and availability aspects, but also some personal and social characteristics (Martínez- Miranda et al., 2002) A model of teams and team members We use as a case of study a team in charge of a design problem. In the first step the user (typically a Project Manager) selects, according to his/her own experience, the members of the initial work team. Once the team is formed, several simulations of its behaviour are performed. If the overall results indicate that the team could possibly complete the project with success, the user has the possibility to save the team configuration in a file for future reference. However, if the simulations do not predict an acceptable performance, the user has the possibility of adding, removing or modifying the team members, until a suitable team is identified. Our prototype tool uses agents technology to represent each person from real life by means of a software agent. Each agent has a specific role within the team. This role is given by the type of tasks that the agent can achieve. There are four types of roles in our model: Project Manager, Engineer/Scientist, Technician and Assistant. The characteristics of a software agent are then matched with the ones of a real person using three basic aspects: Cognition, representing the technical knowledge of a person, i.e. creativity and experience. Emotion and personality, to represent some emotional states. In this prototype we consider the following basic emotions: desire, interest, disgust, and anxiety (Johnson- Laird & Oatley, 1992); and the following personality trends: amiable, expressive, analytical and driver (Schubert, 1997). Social characteristics to represent the interaction between team members. An agent s behaviour results from the values of all of these internal properties randomly modified around the initial values of each of the properties using a normal distribution. We have introduced these random variations around each value to account for the non-deterministic nature of human behaviour. As a result, these variations (which are generally small) will generate different results in each simulation in spite of the fact that the system is executing with the same team members performing the same tasks. Our hypothesis is that we can approximate some of the most plausible behaviours of a work team (e.g. best case, worst case, average case) by averaging over many (maybe several hundreds) simulations where each of the parameters is randomly perturbed around a value. Finally, another modification of the agent s internal parameters is effected by its interaction with its tasks and with other agents. For example, if an agent x has (i) an expressive personality, (ii) a low experience value, and (iii) likes to work in a team, but its assigned task requires that it works by itself, then its interest value would be decreased, and its anxiety or stress values would be increased. These changes are reflected on the quality of the results and time required to perform the task that the agent has been assigned. The representation of tasks includes the following parameters: number of participants for each task; duration of the task (measured in working days); sequence of the tasks (sequential or parallel tasks); difficulty of the task (a task can be complex or not); type of task (generic or specialised task); deadline; priority within the project; and finally, the quality of the task Description of the tool The prototype tool is being implemented using the JADE framework for MAS development. Our first prototype has the following assumptions and limitations: (a) (b) The agents will not solve a real design problem but only simulate their interaction with other agents and with their assigned task(s); Given the uncertainty associated with the characterisation of the cognition, emotion, personality and social properties of a person, random probabilities around

7 44 A. Aldea et al. / Expert Systems with Applications 26 (2004) the fixed values of such properties (representing the internal state of the agent) will be used. (c) The set of global behaviours of a team is obtained by averaging its behaviour over a statistically significant number of simulations. (d) The most suitable team configuration can be obtained by comparing the sets of global behaviours for several possible team configurations. The system has three main components: Configuration module. In this component the user must configure the initial team. The team is formed by two or more agents, and the user must also configure the internal state of each agent. Once the team is configured, the user must specify the project that the team will perform, i.e. the characteristics of the project tasks. Simulation module. The user decides the number of simulations to perform, and then the system starts the work team simulations. Results module. This component shows the results of the team s behaviour. In the future this component will include some graphics to ease the interpretation of the data generated by the simulations, e.g. statistics of information such as best, worst and average performance of the work team. In the main window the user creates all the necessary agents to form a team (see Fig. 4). First, the user must select the type of agent that he/she wants to create (Project Manager, Engineer/Scientist, Technician or Assistant). After that, all the internal parameters of the agent must be set, and finally the agent is created. Once the agent is created, the interface allows the user to see and modify the internal parameters of the agent, or if necessary, delete an agent. In addition, in this window the user can select the organisation type of the team. Currently we have implemented a centralised hierarchical organisation. The window shown in Fig. 5, allows the user to set the tasks that conform the project to be developed by the team. When a task is created, all its parameters must be set. Some of these tasks parameters influence the agents behaviour, for example the quality, duration and deadline. When a task is a successor of another, a line between them records this relationship. One of the advantages of the tool is that all the configuration process is made graphically. Similarly to team configurations, a project can be saved for future reference. The last module in the specification of the simulations is the tasks assignment module. Once the team and the project have been defined, the user must assign the tasks to the agents. Each task can be assigned to one or more agents. Currently in our system the user assigns the tasks to team Fig. 4. Graphical user interface of the tool: team configuration window.

8 A. Aldea et al. / Expert Systems with Applications 26 (2004) Fig. 5. Tasks configuration window. members, but in a future the Project Manager agent will perform this assignment. Once the tasks have been assigned, the user can start the simulation step by setting the number of simulations The interaction between agents In our system there are two agents that do not represent human team members: the GUI agent and the Simulator agent. The GUI agent is the agent that interacts with the user, and the Simulator agent controls the number of simulations that the system will perform. The process starts when the GUI agent sends a request message to the Simulator. The content of this message is the number of simulations that the user requires. Then, the Simulator agent sends a request message to the Project Manager agent asking for the tasks assignment. This agent searches the first task(s) to be developed, and in turn sends a request message to the agent(s) that must perform this task. Finally, the behaviour of every agent involved in the process is reproduced according to the model described in Section 4.1. When the task is finished, the agents send an inform message to the Project Manager agent. The Project Manager then searches for the subsequent tasks and their assigned agents and sends the request messages again. This process is continued until all the tasks are finished at which point the Project Manager agent sends an inform message to the Simulator agent. If the Simulator agent observes that all the simulations are done, then sends the inform message to the GUI agent which presents the simulations results. All this process is performed as many times as the number of simulations requested by the user (see Fig. 6). After all the simulations are finished the user can see the results for each simulation. The information about each of the agents is presented to the user, and he/she can analyse which team has generated the most suitable set of behaviours for a particular project. In future versions of the tool several views will be implemented, such as graphics that show the individual behaviour, the real duration of the tasks within the system, the best and worst behaviour of agents in specific moments, etc. All this information has the potential to help the user to make the most suitable decisions about the configuration of a work team Initial results This section presents the initial results of the tool. We took as a case study a team of 10 members: one Project Manager, three Engineers/Scientists, three Technicians and three Assistants, each one with different internal characteristics. The project assigned to these agents has 12 different tasks. After 100 simulations with this team configuration we noticed some interesting behaviours: We observed that agents with a high value in their stress parameter and which were in charge of a specialised task were the agents with least success, independently of their personality. When we decreased the stress parameter of one agent and executed the same number of simulations we observed that the number of failures also decreased. The simulations also showed that agents with an amiable personality and working in a task by themselves presented more failures than successes. When this

9 46 A. Aldea et al. / Expert Systems with Applications 26 (2004) Fig. 6. Agents communication. situation happened, their anxiety and disgust values were increased and so were their number of failures. The agents with a greater number of successes were those with a high value for experience and with an analytical personality. We also observed that the creativity parameter only increases the number of successes when the agent is in charge of specialised tasks, and this parameter has less influence when the agent works with a generic task or with a non complex task. Although some of these results are expected a priori given the way in which the agents behaviour was generated, they are, nevertheless, the basis to add more complexity to the model of a team. In addition we are modifying the agents behaviour to result not only in a success or failure over its assigned tasks, but also in a delay to the task (which in turn may increase the stress value of the agent in charge of the successive task). In any case, each agent must be able to achieve its tasks, otherwise the Project Manager agent reassigns them to other agents. One of the most important steps in the future of this work will be the comparison between the results produced by our system and real teams of people. 5. Other possibilities MAS are, of course, applicable in many other areas related to chemical engineering, those involving distributed decision making exemplify a growing family of applications that is increasingly finding industrial acceptance. Applications such as flexible manufacturing control, where the standard approach is to devise a centralised, pre-planned schedule for the entire process, have benefited from an agent-oriented approach. Within those applications each part, operation and machine is represented by an agent with individual objectives, such as getting to the end of the manufacturing line for a part or maximising its throughput for a machine. Each agent has the capability of negotiation when something goes wrong, e.g. a delay in an operation or a failure in a machine (Parunak, 1999). The resulting schedule is more responsive to circumstances and as a result it exhibits

10 A. Aldea et al. / Expert Systems with Applications 26 (2004) increased throughput and robustness to failure, thanks to the fact that decision-making is more localised. The same approach can be applied more or less directly to batch process operation scheduling supply chain management, to model a network of suppliers, factories, warehouses, distribution centres and retailers (Julka et al., 2002) web-based plant design, to model a network of owneroperators, engineering, procurement and construction companies (Brown, 2002) the autonomous process control system as described in (Stephanopoulos, 1990) integrating planning, scheduling and production management. multi-product and multi-processes scheduling, balancing aspects related to stock and demand (this aspect is very important for the pharmaceutical industry, as there is a seasonal ordering of certain products and this sector maintains a high stock for a large number of specialities) reverse engineering of several aspects of the competitors (prices, offers ) automatisation of offers to customers (e.g. equipment selection according to the technical data sheet). 6. Conclusions Three case studies proposing the use of MAS in the solution of chemical process industry problems have been presented: A support platform to extract, process and classify knowledge from the Web. A system to support the concurrent design of processes. A tool to support the configuration of work teams. These application areas cover the modelling of methodologies, artifacts and organisations, and are thus representative of the large variety of possible applications of agent based systems in industry. Acknowledgements The research described in Section 2 has been funded by the EU project htechsight, IST (coordinator: Prof. Antonis Kokossis, University of Surrey, UK) where the University Rovira i Virgili is one of the 10 nodes. The authors wish to acknowledge the work of Marc Sanchez, David Isern and Jaime Bocio and the feedback of Prof. A. Kokossis and P. Linke. Section 3 is the result of the work of Solomon S. Gelete and Juan Martínez-Miranda. The work in Section 4 is part of the PhD research of Juan Martínez-Miranda, who wishes to thank the Universitat Rovira i Virgili for his scholarship. References Bañares-Alcántara, R., Kokossis, A., Aldea, A., Jiménez, L., & Linke, P. (2003). A knowledge management platform to extract and process information from the web. Process System Engineering, PSE 2003 Kunning, China, in press. Bellifemine, F., Poggi, A., & Rimassa, G. (1999). JADE-A FIPA compliant agent framework. Proceedings of Practical Applications of Intelligent Agents and Multi-Agents, PAAM, Biegler, L. T., Grossmann, I. E., & Westerberg, A. W. (1997). Systematic methods of chemical process design. Englewood Cliffs, NJ: Prentice Hall. Brown, A. (2002). Web-based plant design tools are ready, but are engineers? CEP, 98(6), Fensel, D. (2001). Ontologies: a silver bullet for knowledge management and electronic commerce. Germany: Heidelberg. Gelete, S. S., & Bañares-Alcántara, R. (2002). A concurrent engineering approach for an effective process design support system. Proceedings of ESCAPE-12, Amsterdam: Elsevier, pp Heflin, J., & Hendler, J. (2000). Dynamic ontologies on the web. Proceedings of American Association for Artificial Intelligence Conference AAAI-2000, Menlo Park, CA, pp Jennings, N. R. (2000). On agent-based software engineering. Artificial Intelligence, 117, Johnson-Laird, P., & Oatley, K. (1992). Basic emotions, rationality, and folk theory. In N. L. Stein, & K. Oatley (Eds.), Basic emotions (pp ). Hove: Lawrence Erlbaum. Julka, N., Karimi, I., & Srinivasan, R. (2002). Agent-based supply chain management 1: framework and 2: a refinery application. Computers and Chemical Engineering, 26(12), Klusch, M. (2001). Information agent technology for the internet: a survey. Data and Knowledge Engineering, 36(3), Kokossis, A., & Paul, S. (Eds.), (2002). Dynamic ontology management system (design). H-TechSight Technical Report (D7). Martínez-Miranda, J., Aldea, A., & Bañares-Alcá, R. (2002). A social agent model to simulate human behaviour. In C. Urban (Ed.), (pp ). Proceedings of the Third Workshop on Agent-Based Simulation. Parunak, H. V. D. (1999). Industrial and practical applications of distributed artificial intelligence. In G. Weiss (Ed.), Multi-agent systems (pp ). Cambridge MA: MIT Press. Picard, R. W. (1995). Affective computing. M.I.T. Media Laboratory Perceptual Computing Section Technical Report No Reid, G., & Smith, T. (1980). The contract net protocol: high load communication and control in a distributed problem solver. IEEE Transactions on Computers, 29(12), Schubert, S. (1997). In L. T. Biegler, I. E. Grossmann, & A. W. Westerberg (Eds.), Systematic methods of chemical process design (p. 9) Englewood Cliffs, NJ: Prentice Hall. Sure, Y., Staab, S., & Angele, J. (2002). OntoEdit: guiding ontology development by methodology and inferencing. Proceedings of the International Conference on Ontologies, Databases and Applications of Semantics (ODBASE 02), Irvine, USA. Stephanopoulos, G. (1990). Artificial intelligence in process engineering current state and future trends. Computers and Chemical Engineering, 14(11), JADE (2002). Java Agent Development Environment, JADE Website, Telecom. Italia and Dipartamento di Ingeniria dell Informazione, Università degli Studi di Parma, Torino, Italy. projects/jade/. Weiss, G. (1999). Multiagent systems. A modern approach to distributed artificial intelligence. Cambridge, MA: MIT Press. Wooldridge, M. (2002). An introduction to multiagent systems. New York: Wiley.