Innovative Developments in Design and Manufacturing

Advanced Research in Virtual and Rapid Prototyping -- Proceedings of VRP4, Oct. 2009, Leiria, Portugal

J. N. Reddy, Paulo Jorge da Silva Bartolo, Mateus Artur Jorge, Fernando da Conceicao Batista, Henrique Amorim Almeida, Joao Manuel Matias, Joel Correia Vasco, Jorge Brites Gaspar, Mario Antonio Correia, Nuno Carpinteiro Andre, Nuno Fernandes Alves, Paulo Parente Novo, Pedro Goncalves Martinho, Rui Adriano Carvalho

• Publisher: CRC Press
• Year: 2009
• Print ISBN: 9780415873079
• eBook ISBN: 9781439859216
• Topic: Technology & Engineering
• Subtopic: Civil Engineering
• Index: Technology & Engineering

Essential reading on the latest advances in virtual prototyping and rapid manufacturing. Includes 110 peer reviewed papers covering: 1. Biomanufacturing, 2. CAD and 3D data acquisition technologies, 3. Materials, 4. Rapid tooling and manufacturing, 5. Advanced rapid prototyping technologies and nanofabrication, 6. Virtual environments and

Virtual environments and simulation

Towards ontology-based information extraction in distributed manufacturing systems

B.X. Li, L. Yang, S.K. Ong, Y. Lei & A.Y.C. Nee

Mechanical Engineering Department, National University of Singapore, Singapore

ABSTRACT: In this paper, the role of ontology in constructing a Pay-per-Use (PpU) distributed manufacturing platform is discussed. First, the ontology for each agent in the web-based platform is built, and the relationships between the ontology and the agents in the PpU platform are identified. Second, the ontology for the PpU platform is modeled as a combination of the ontology of all the agents in the PpU platform. Agents are categorized as hybrid agents, which means an integration of agents, or category agents, which means different types of agents. Since the ontology-based architecture allows heterogeneity and polymorphism of the information among different agents, it is suitable for multi-agent platforms and heterogeneous knowledge-based systems. Message translation involves translating multiple interpretations of the same information among diverse agents. Due to the dynamicity of the individual agent ontology and the system ontology construction, ontology matching algorithms are employed to facilitate information communication among the agents.

1 INTRODUCTION

Many manufacturing applications of multi-agent systems (MASs) have been proposed (Lee et al. 2006, Mahesh et al. 2007). In an MAS, communication between the agents is flexible and dynamic due to the ability of the agents to exchange information among themselves. However, the growing complexity and increasing amount of knowledge and information required by the agents have made it increasingly more difficult to extract information within the individual agents and between the diverse agents, and this affects the operability and interoperability of the system. There are two common types of problems, namely, a semantic problem, and a syntactic problem (Lin & Harding 2007).

Ontology (Zhang et al. 2009) allows machine-understandable descriptions of digital content in information systems to be formed, so as to enhance applications interoperability and information retrieval and reasoning (Embley et al. 1998). Ontology-type semantic descriptions of behaviors and services allow software agents in an MAS to better coordinate themselves (Lam & Ho 2001) and improve the efficiency and effectiveness of information communication (Damjanovic et al. 2007). Most MAS applications of ontology involve constructing one ontology for the entire MAS. However, given the diverse domain knowledge of the different agents in an MAS and the need to dynamically update the agent ontology with new domain knowledge, individual agent ontology and system ontology should be separately built to improve the operability and interoperability of an MAS.

This paper proposes an ontology-based approach to improve the information communication and message exchanges within the PpU system. The PpU system ontology is constructed according to the work flow of the system to fulfill distributed manufacturing functions. To allow the agent ontology to fully utilize the domain knowledge and acquire new domain knowledge, an ontology is constructed for each agent in the MAS. The rest of the paper is organized as follows. The work flow in an agent based on the ontology model is described in Section 2. In Section 3, the ontology for the PpU platform is modeled as a combination of all the agents in the PpU platform. Ontology matching methods between agent ontologies are illustrated in Section 4. In Section 5, conclusion and future work are presented.

2 ONTOLOGY FOR INDIVIDUAL AGENT

2.1 Web-based PpU distributed manufacturing system

The PpU platform that has been developed is a web-based platform that provides various professional and technological services for distributed manufacturing. The clients can obtain their needed information and services from the PpU platform and the service providers can provide their services through this platform. Figure 1 shows the framework and data flow of the web-based PpU multi-agent system. The PpU platform consists of both the server/system and the client/designer components, which include the Design Mediator Agent (DMA), Manufacturability Evaluation Agent (MEA), Manufacturing Resource Agent (MRA), Process Planning Agent (PPA), Manufacturing Scheduling Agent (MSA), Manufacturing Managing Agent (MMA), service providers and designers (Mahesh et al. 2007).

2.2 Ontology-based agent 2.2.1 Ontology structure

The research in this paper adopts a two-layer structure (Fig. 2). The domain layer consists of the domain name of an ontology and the various concepts defined by the domain experts. There are several concepts, such as the “I/O Modifier, Work Engine, and Knowledge-based Pool” defined in the concept layer. Each concept contains a concept name, a property set or an operation set for an application domain. Two types of inter-conceptual relations have been used in the domain ontology, namely, aggregation and association. The relationship between an agent and its corresponding concepts is an aggregation relation, which denotes the “is-part-of” relationship. The association relation represents a semantic relationship between the concepts in the concept layer (Lee et al. 2005). The ontology model is developed using Protégé.

Figure 1. PpU multi-agent manufacturing system.

Figure 2. Two-layer structure of an agent ontology.

2.2.2 Structure in ontology-based agent

Agents in the PpU platform consist of a message interpretation mechanism, domain expertise ontology and an autonomous solution generation function, which are developed based on the I/O modifier, knowledge-based pool and work engine respectively. Figure 3 shows the structure of an ontology-based agent.

The I/O modifier is an interpretation mechanism to create the evaluation parameters depending on the input message. To perform a task from a preceding agent, the current agent must obtain the instruction from the input message, e.g., the DMA may send a message containing requirements on cost, time-span or a combination of these two factors to the MSA. After receiving and interpreting the message, the MSA needs to consider these factors as the objectives in the scheduling process.

Besides the information from the input message, the work engine generates a solution based on the domain expertise from the knowledge-based pool. The knowledge-based pool is an expertise ontology that stores the background knowledge and evaluation rules used in performing the agent functions. Figure 4 shows part of the hierarchy structure in the expertise ontology of the MEA.

2.2.3 Ontology model of an agent

Figure 5 shows the ontology model of an agent. The expertise ontology stores the domain knowledge that is implemented using the JAVA programming language as “if/then” rules (Jia et al. 2004). Figure 6 displays the operation process of an agent. The interpretation mechanism performs message translation. The current agent receives the output message of the preceding agent as input message and translates this message. After a solution has been generated, the solution information is packaged as an output message and passed to the succeeding agent. The task performance mechanism is carried out based on the linguistic instruction in the input message of the preceding agent and the evaluation rules, which are retrieved from the expertise ontology. The data flow is shown in Figure 7.

Figure 3. The structure of an ontology-based agent.

Figure 4. The hierarchy structure of MEA expertise.

Figure 5. Ontology model of an agent.

3 THE ONTOLOGY FOR THE PPU PLATFORM

A PpU ontology model has been developed to facilitate the information exchange between different agents in the PpU multi-agent manufacturing system. The structure of the ontology for the PpU manufacturing platform and the relationships between the agents are shown in Figure 8. The PpU platform and the MMA share the same ontology. The MMA is responsible for coordinating the communication between the agents in the system and the conflicts among the agents. Based on the PpU ontology, the operability of the PpU multi-agent system is enhanced as each agent can communicate with other agents to exchange messages more accurately and make corresponding adjustments in the operation process. Different agents can communicate with each other to exchange information without ambiguity through the ontology-based system. If a conflict in the system cannot be solved with the collaboration of the agents, the conflict message will be sent to the MMA for it to coordinate all the agents related to this conflict (Jia et al. 2004).

Figure 6. The operation process of an agent.

Figure 7. Data flow in ontology-based agent.

4 AGENT ONTOLOGY MATCHING

When two agents exchange messages, they may not understand each other if they do not share the same content language and ontology. Thus, it is necessary to provide a means of matching the ontology between them in order to either translate their messages or bridge the axioms in their respective ontology (Shvaiko & Euzenat 2005). The PpU ontology provides the agents in the MAS with a commonly shared ontology. However, dynamic knowledge acquisition carried out by the agents makes it necessary to provide ontology matching methods to guarantee effective knowledge exchanges between these agents. Element-level and structure-level matching approaches, namely, the edit distance (Levenshtein 1996) and the cosine similarity (Karnik et al. 2007) are employed to match the ontologies within the system. Semantic matching based on hybrid agents in the PpU ontology is combined with edit distance and cosine similarity to form an ontology matching method, and the details are presented next.

4.1 Element-level matching

Element-level matching performs matching for individual schema elements (Shvaiko & Euzenat 2005). The edit distance between two strings is the number of operations required to transform the characters in one string into the other string. It is a useful method for element-level matching. In this research, the Levenshtein distance is used to calculate the edit distance between the classes of the source ontology and the target ontology based on the class names in these ontologies. The reason for using the class names to calculate the edit distance is that in the process of building the ontology for the agents, similar class names are usually adopted to represent classes with similar functions. Thus, two agent ontologies are closer when the edit distance between them is smaller. In order to improve the edit distance results, the edit distances between the classes of the source and the target ontologies are further calculated using keywords from the class descriptions that are based on the web ontology language.

Figure 8. Top-level classes from the PpU ontology model.

Figure 9. An example of a hybrid agent.

4.2 Structure-level matching

At the structure level, the hierarchies of class names from two ontologies are used to calculate the cosine similarity. In the matching process, the hierarchies of the classes in the source and the target ontologies form the vectors H1 and H2. The cosine similarity between hierarchy H1 and hierarchy H2 can be viewed as the angle between these vectors (Lee et al. 2007). The similarity value would be between 0, which means these two hierarchies of classes from the source and target ontologies have no relationship, to 1, which means these two hierarchies of classes are identical. The in-between values indicate intermediate similarity between these two hierarchies of classes.

An example on the matching between the ontologies of MRA and MEA based on the material information is presented to illustrate the structure-level ontology matching method. The hierarchy of classes extracted from the MRA ontology is “Manufacturing resource agent → Production resource → Material”. The class names are employed as the keywords that represent the classes in the MRA ontology and other...

Table of contents

• Cover
• Title Page
• Copyright
• Table of contents
• Preface
• Sponsors
• International scientific committee
• Invited lectures
• Biomanufacturing
• CAD and 3D data acquisition technologies
• Materials
• Rapid tooling and manufacturing
• Advanced rapid prototyping technologies and nanofabrication
• Virtual environments and simulation
• Applications
• Author index