Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
COMPUTER SUPPORTED COLLABORATIVE DESIGN: MISMATCH CONTROL. METHODOLOGY AND APPLICATION Victor Taratoukhine, Kamal Bechkoum, Martin Stacey Department of Computer & Information Sciences, De Montfort University, Hammerwood Gate, Kents Hill, Milton Keynes, MK7 6HP, U.K. [email protected] [email protected] [email protected] Abstract The main objective of this paper is the theory and practice of computer supported collaborative design in the field of support of consistency checking process. First part of this paper describes an analysis of existing methods and techniques addressing the problem of mismatch control in collaborative design environment. The second part represents the methodology of Intelligent Distributed Mismatch Control (IDMC). The proposed general methodology consists of two sub-models: process model of IDMC and structural model - conceptual multi-agent framework for IDMC development. The third part demonstrates the application of proposed methodology for aerospace collaborative design. 1 Introduction Computer Supported Collaborative Work (CSCW) as a part of distributed collaborative design process and Concurrent Engineering [Prasad, 1995; Unan, 1992, EDID, 1995] promises to resolve most of the difficulties by replacing the paper and tape and physical meetings based methods by electronic communication and electronic meetings and provide a basis for virtual design environments [Prasad, 1995; Matta, 1997; Sycara, 1991]. This is important, because for many years the design and manufacture of major European complex products, such as sattelites, airplanes and cars has been distributed across the continent. As the result of globalisation and future distribution of design and manufacturing facilities, the cooperation amongst partners is more challenging. The design process tends to be sequential and requires centralised planning teams and/or a great deal of travel to/from distributed designers. In a virtual team designer work together and use a Internet/Intranet for communication. The design is a multidisciplinary task that involves several stages. These stages include input data analysis, conceptual design, basic structural design, detail design, production design, manufacturing processes analysis, and documentation. As a result, the virtual team, normally, is very changeable in terms of designers participation. Moreover, the environment itself changes over time. Consequently, as the result the number of design mismatches are greatly increased. A methodology of Computer Supported Collaborative Design (CSCD) is needed for future progress. In this paper we focus on one aspect of CSCD: mismatch control during the detail design stage. The mismatch detection during detail design stage is important, because the number of mismatches in the early stages of detail design will have huge last implications. This is particularly true for large–complex products such as aeronautics or automotive industry. The decision is to use Artificial Intelligence (AI)/Distributed AI tools, particulary multi-agent systems and techniques to provide automatic and semi-automatic mismatch detection and resolutions [Tomiyama, 1993; Klein, 1992; Bento, Fejo,1997; Beckoum, 1997; Taratoukhine, Bechkoum, 1999]. The next section of paper reviews briefly a current models and methods in Computer Supported Collaborative Work (CSCW) and their applications in design consistency checking. 1 2 CSCW and design consistency checking. Methods and techniques Many of the recent developments in the field of conflict management have been investigated and described by Matta, Lander, Klein and others [Matta, 1997; Lander, 1997, Klein, 1992; Easterbrook, 1991, 2000 ; Bechkoum and Taratoukhine, 1999]. According to [Lander, 1997] there are several ways in which conflict can be managed: such as Avoidance Avoid conflict by sharing information about local constraints and priorities; Conflict classification - Build taxonomy of conflict types. Associated with each conflict type is a specific piece of conflict resolution advice; Negotiation Techniques in this area include bargaining, restructuring, constraint relaxation, mediation, and arbitration. Below a developments of conflict management methods are represented. [Klein, 1992] represents a conflict management as an exception-handling component of a collaborative design tool. [Matta, 1997] defines a library of associations between concurrent engineering sub-tasks and conflict management methods to guide an agent to determine appropriate methods to manage conflict in a particular application [Easterbrook, 1993] explores the support for conflict management in CSCW tools. And identifies a broad areas in which conflict and conflict resolution have been studied. The analysis of current development in conflict management for CSCW sugests that most of of these methods and frameworks are paid more attention to social and psychological aspects of communications between members of team, but not to problems of communications between artificial agents and development a general methodology of conflict management/intelligent control, based on Distributed AI. The main objective of this paper is to propose a framework for the development of an Intelligent Distributed Mismatch Control (IDMC) methodology. This will be based on the integration of two approaches of conflict management - classification and negotiation. In our case, the classification and negotiation processes are associated with multi-agent architecture of IDMC. In this context, the investigation of methods and principles of organisation is as follows: - Development methodology of Intelligent Distributed Mismatch Control (IDMC). The proposed general methodology should include: - Process model of IDMC. Model of taxonomy of mismatches. - Structural model - conceptual multi-agent framework. The next step is an application of proposed methodology for CSCD process. 3 Methodology In order to define a IDMC methodology, the two connected sub-models/levels of formalisation are used. The first model is process model of IDMC and the second is structural model - conceptual multi-agent framework, but first of all, we need to define a terminology of consistency checking process. 2 3.1 Process Model In our understanding, the mismatch control is a process of detection and resolution of design mismatches based on, the on capability of multi-agent framework and communication between agents, communication and resolutions schemes and distributed knowledge-base organisation. Definition of design mismatches are as follows: Definition 1 The Design mismatches are inconsistencies between design goals Gi and current design project M1(t). Obviously, the goals of design are a set of parameters (for design project) and predefined restrictions for these parameters. We propose that concurrency attributes are a basis for definition of restrictions for parameters and structure of design project. Definition 2 The design mismatches at the detail stage of design are inconsistencies between parameters restrictions defined according to concurrency attributes, current parameters, and/or the structure of current design project M 1(t). IDMC uses a concept of distributed artificial intelligence - agents. Agents are represented as “virtual designers” who have internal abilities to receive information, to identify design mismatches and to prepare advice for the designer to find the best modification to resolve the mismatch. Design knowledge model - M2 is used as a personal assistant for the designers D in design team Dt and helps to detect design mismatches and find the best modifications required. The design mismatches are detected using a vocabulary of indicators and taxonomy of design mismatches and resolved using a model M2 - distributed model of designer’s knowledge. We have: M2= {A1, A2, … , Ai , … , An}; where A- agent. As can be seen, each agent is represented as part of design assembly and has a knowledge about assembly part’s geometrical configuration (structure), materials and other parameters. Ai = {W1, ... , Wi, ... , Wn }, and each agent is represented as single knowledge-base which contains a set of words of designer to provide a knowledge about different aspects of design, a project, and concurrency attributes. Each designer world is represented as: Wi = {K(Mstr), K(Mpar), K(Res), K(Indicators)}, where K- means knowledge, Res- design restrictions, I-indicators. Wi - "i" designer world. During the design time Wi is changed according to designer’s knowledge and inter agents communications. The modification of Wi is a general process of adaptation. We have an adaptation of internal design knowledge (agent’s knowledge) and modification of design projects. Adaptation is divided into two types: 3 (1) internal adaptation and (2) external adaptation. Internal adaptation of model of knowledge about design project- M2 is a self-adaptation of knowledge using internal knowledge, as the result of communication with designers and agents. External adaptation of M2 is adaptation knowledge under the supervision of other agents and/or designers. The mismatch control process include a two main actions: consistency chechking (mismatch detection) and resolution of mismatches. If we have changes of structure or parameters of design project M 1, during the mismatch resolution action, we will say about process of modification. The process model of IDMC is represented as follows: Modification of design project Advice for Modification of project Designers Design Project M1 Reception of Information Design Goals Distributed Knowledge Distributed Design Process Intranet/Internet Design Process Design Process Consistency checking Interknowledge communication Taxonomy of mismatches Mismatches of Integration A1,…,An Restrictions External adaptation Indicators Internal adaptation Figure 1 Process Model of IDMC According to Fig. 1 the definition of complex taxonomy of design mismatches is important for the classification and resolution of design mismatches. The development of taxonomy of mismatches is described in next section. 3.2 Design Mismatches. Classification and Taxonomy Important developments in this area include models reported by [Klein, 1997], [Matta, 1997], [Castelfranchi, C (1996)], but these taxonomies, in general, are more oriented towards the conceptual stage of design process. We propose a conceptual framework for the development of a taxonomy for detail stage design. Firstly, we classify design mismatches according the to level/types of information needed for their detection. We have: - syntax level - ordinary geometric mismatches (size, diameter, geometric type, parts orientation,…), 4 - semantic level - complex assembly mismatches - analysis of geometric and materials characteristics for checking assembly possibility, - pragmatic level - the complex mismatches are connected to design/concurrency attributtes, such as a mismatches of manufacturability, manability, costability, serviceability, etc. Accordingly, we can define: mismatches of integration and concurrency mismatches. Mismatches of intergation are assembly mismatches that are necessary to ensure the design at the integration phase. Using the proposed model of taxonomy it is possible to define a real taxonomy, for a known field of implementation (i.e aerospace, automotive, electonics), and a number of concurrency attributtes - A1, An. Of course, the development of taxonomy is a complicated and long process which included a definition of critical parameters, indicators, restrictions and attributes and based on knowledge engineering components, as interviews and questionnaires. An example of taxonomy (Fig. 2) is especially oriented for the implementation of mismatch detection during the integration phase of aerospace/mechanical engineering designWithin the proposed taxonomy critical parameters are identified. It is the variation of these parameters that causes a mismatch. For example, the bolted connection requires consistency between such parameters as thread minor diameter, minor diameter and pitch (for bolt head type and assembly tools type). For aircraft wing box bolt and nut – diameter, length and size, etc. Weld connections require consistency between types of materials and the material thickness, as well as the geometric parameters of material. For the mismatch detection process to be more successful, not only we need to represent a wider variety of mismatch types but we also need to represent geometric information as well as information about the material which the parts are made of. This taxonomy is restricted by Design For Assembly(DFA)/Design For Manufacturability (DFM) mismatches. According the taxonomy development process the criteria of classification should be considered. In our case we have a main criterion - design for assembly/manufacturability, and additional criteria as types of connections and indicators (critical parameters- Mparcr). The full taxonomy DFA/DFM mismatches was described in [Taratoukhine, 1999]. Taxonomy of DFA/DFM Dissasembly mismatches Sequence mismatches Failed assembly/ dissasembly sequence plan Non optimal assembly/ dissasembly sequence plan Impossible dissasembly Bad serviceability Assembly mismatches Interaction mismatches Connection mismatches Forward search: impossible assembly Backward search: impossible dissasembly The time of assembly / dissasembly is too much Unwanted contacts Other mismatches Time of assembly is too much No adequate dissasembly tools The number of parts Interference mismatches Impossible connection Symmetry No tools for automatic/manual assembly Difficulty for hand operations Mating direction Assembly cost is high Unwanted contacts Stability Figure 2 An example (part) of a taxonomy for DFA/DFM. 5 Impossible tools changes operation Directorability The next section describes a formal description of structure of multi-agent framework and dynamics of multi-agent framework. 3.3. The structural model. A multi-agent framework 3.3.1 Formalising the framework This section reviews briefly the agent framework and the components within. Using a multi-agent approach the representation of the required knowledge is distributed amongst several “specialised” independent knowledge bases, or agents. The architecture assumes that the design knowledge is encapsulated within the different members of the agent community. The conceptual framework (CF) may be presented formally as follows: CF = {AP1, ... , APt, ... , APn}, where APt is the tth Assembly Part, t = 1,2, … , n. AP = {DA1, … , DAi, ... , DAm, CA1, ... , CAj, … ,CAk}, DAi is the ith Design Agent (D-agent), i = 1, 2, …, m, CAj is the jth Control Agent (C-agent), j = 1, 2, …, k. Agents exchange information using messages with syntax and semantics defined by the communication protocol. The context of these messages can include declarative and procedural knowledge. The proposed communication protocol (CP) for the above agents is as follows: CP={Lc1, ... , Lcm; Li1, ... ,Lim; Lcc1, ... , Lcck; Lic1, ... , Lick}, where Li - information language, used by DAi, Lc - control language, used by CAi to update the fact base of DAi with the view of resolving a mismatch. Lic - information language, which describes current situation for C-agents, Lcc - control language, which updates meta- and knowledge-base of C-agents. A reactive agent is an entity that may be represented by an independent program that knows everything about itself including its relationships with other agents. The principle of emergence states that intelligence in reactive agents emerges from interaction of agents among themselves and with their environment. The principle of situatedness states that intelligence of a reactive agents is situated in the world and not in any formal model of the world build in the agent [Sycara, 1991] . In our case DAi (D-agent) is an reactive agent, which negotiate with other design agents using design’ schedule (assembly sequence) and scheduling preferences generated and supervised by the Control Agent (C-agent). The next section describes inter-agent relationship within framework and communication between agents. 3.3.2 Dynamics of framework It will be possible to analyse a communication strategy in multi-agent network and to develop mismatch detection and resolution scheme. The general structure of conceptual framework communication and adaptation schemes is shown in Figure 3. 6 Virtual team of D D designers C-agents level CA CA CA D-agents level M1 i Design Project Figure 3 Dynamics of multi-agent framework For multi-agent systems the process of communication is critical, because during this process the system detects and resolves design mismatches. To resolve the conflicts in multi-agent cooperation we will use an general arbitration scheme. Arbitration from C-agent will stop disagrement between D-agents when a conflict situation is presented and a modification of project (fact base of D-agent) is required. We can define the approach as co-ordinated collaboration when we have compatible goals of agents, insufficient resources and insufficient skills. In our case goals are assembly possibilities, resources are sets of parameters, and skills are mismatch detection abilities. In our multi-agent framework we have three layers of vertical communications: (1) reactive layer (D-agentsDesign project), (2) control layer (C-agents-designer), (3) designer layer and two layers of horizontal communication as between D-agents and between C-agents. We are using a general two layers architecture as described earlier, with D-and C-agents, and coordination and combination of groups of agents is realised by Designers using assembly sequences. The different layers are described below: Reactive layer (Vertical): Each D-agent operates as an independent entity and interacts asynchronously with associated assembly parts. The communication between D-agents and the associated parts of the design project is a 7 process of elimination of inconsistencies. These inconsistencies may be the result of a modification of the Design Project or a self-adaptation of the knowledge base of D-agents. Control layer (Vertical): The communication between C-agents and associated D-agents is a process of elimination of inconsistencies between assembly parts, when D-agents are unable to resolve it, using internal knowledge and/or horizontal communication. This is client-server communication (under the supervision of the Cagent). C-agent receives the new information from D-agents using syntax of Li. The result is external adaptation of knowledge-base of D-agents (using syntax of Lc), according to C-agents meta-knowledge base information, if mismatches occur. Designer layer (Vertical): The communication between C-agents and designers (design team) is a process of elimination of inconsistencies between assembly parts, when C-agents are unable to resolve it, using internal knowledge and/or horizontal communication. This is human-computer communication. The result of communication is external adaptation of metaknowledge-base of C-agents. Horizontal Communication: The Horizontal communication takes the form of a negotiation between associated D-agents (D-agent to D- agent relation); C-agent to C-agent, and Designer to Designer. The communication is at peer-to-peer layer. D-agent negotiates with other D-agents, using Li. C-agent negotiate with other C-agents, using Lic and Lcc. Each of these communications aim to eliminate mismatches. The communication between designers is normally organised using e-mail, Internet chat systems, videoconference, telephone and other forms of communications. The optimal organisation of communication of virtual design team is a very important part of the design process and will be investigated in future. 4. Application of Methodology The development of tool for mismatch control is required using the methodology described above. The proposed software - Intelligent Distributed Mismatch Control System (IDMCS) is developed and is outlined below. 4.1 IDMCS The ZEUS toolkit [Hyacinth, Nwana, 1998] was used for the development of IDMCS. The toolkit provides classes that implement generic agent functionality such as communication, co-ordination, planning, scheduling, task execution and monitoring and exception handling. IDMCS analyses designer requirements to the design project given in the form of design information and processes at the layer of the distributed knowledge base. The system (Fig.4) is designed using JAVA 1.2.1 in the Windows NT environment using ZEUS agents building toolkit. In order to retrieve/manipulate CAD data on interface from/to PARASOLID geometric kernel is needed and will be the purpose of future work. ZEUS Building Toolkit agents definitions Agent Based Environment Knowledge Engineering Issues Java based Integration Java- external programs PARASOLID-KID geometric modeller IDMCS Java –PARASOLID Interface Distributed Design Environment Figure 4 IDMCS development 8 4.2 Case study One of the key challenges for Europe is to maintain and develop the European Aerospace sector as a world competitive industry. The European Commission (EC) has fostered several collaborative research initiatives in aeronautics yielding a number of successful projects. In the Fifth Framework Programme of the EC the financial support dedicated to the Aerospace industry alone is set to euro 700 million. Unfortunately the current CAD/CAM/CAE systems do not support the mismatch control process described here. We are using the IDMCS described above for distributed aerospace design support. We are using IDMCS and DFA/DFM taxonomy for development for a distributed knowledge-based design support system which detects geometric and material irregularities at the assembly stage of aerospace design. The IDMCS provides mismatch control during wing-box assembly process, using an initial set of data from aircraft design sources [Nui M, 1999; Raymer, 1999] and AIRBUS design engineers [Airbus, 1999]. When designing using IDMCS, the following steps are being performed: (1) analysis of assembly parts assembly checks of stringers, skins, spars etc., (2) evaluation of assembly possibility - Collision and Tolerance Analysis, (3) manufacturability analysis, (4) choosing the alternatives for mismatch resolution, and (5) semiautomatic mismatch resolution and generation of results. At the present time, a research version of IDMCS is developed and experiments are under progress. 5 Conclusion and future work A novel approach for Intelligent Distributed Mismatch Control as part of Computer Supported Collaborative Design is developed in this paper. The methodology of Intelligent Distributed Mismatch Control (IDMC) is outlined. The general methodology is represented as of two sub-models: process model of IDMC and structural model - conceptual multi-agent framework. The conceptual framework for development of taxonomy is represented as well as the implementation of this framework for a DFA/DFM taxonomy. Development of IDMCS was outlined as well as the possibility of using IDMCS for aerospace design. The future developments of methodology will be in the field of formalisation of the process of dynamics of conceptual multi-agent framework, based on Automata/Petri Nets, that will provide a mathematical foundation of adaptation/learning schemes; and to solve a practical problem - future development of tools for automatic/semiautomatic mismatch control in Computer Supported Collaborative Design. References Akman, V. P.J. Hagen, and T. Tomiyama. A Fundamental and Theoretical Framework for an Intelligent CAD System, Computer Aided Design Journal, Vol. 22, pp. 352-367, (1990). AIRBUS, Knowledge engineering at AIRBUS, 1999 Bento, J, B. Feijo.(1997) An Agent Based Paradigm for Building Intelligent CAD Systems, Artificial Intelligence in Engineering Journal, Vol. 11, pp. 231-244 9 Bechkoum. K. (1997), Intelligent Electronic Mock-up for Concurrent Design, Expert Systems with Applications Journal, Vol. 12, pp. 21-36, (1997). Bechkoum, K (1997),V. V. Taratoukhine. A Framework for Mismatch Control in a Distributed Design Environment, Proc. Advances in Concurrent Engineering, Bath, 1-3 September, Castelfranchi, C (1996) "Conflict Ontology", in Proceedings of ECAI 96, 12 th European Conference on Artificial Intelligence. Easterbrook, S.,(1994) A. C. W. Finkelstein, J. Kramer, and B. A. Nuseibeh (1994) Coordinating Distributed ViewPoints: The anatomy of a consistency check. Journal of Concurrent Engineering: Research and Applications, Vol 2, No 3 (Special Issue on Conflict Management) EDID - An Environment for Distributed Integrated Design, Final report, Cranfield University, 1995 Hyacinth, S. (1998), H. Nwana. ZEUS: An Advanced Tool-Kit for Engineering Distributed Multi-Agent Systems, Proceedings of PAAM'98, London, March, pp. 377-392, (1998). Kock N, (2000), Benefits for Virtual Organizations from Distributed Groups, Communications of the ACM, November 2000/Vol.43, No. 11. Lander S., (1997) Issues in Multi-agent Design systems, in IEEE Expert, 1997 18-26 O'Leary D (1997), Daniel Kuokka, Robert Plan Artificial Intelligence and Virtual Organisations, Communications of the ACM, January 1997/Vol.40, No. 1. Matta, N. (1997), C.Cointe, Concurrent Engineering and Conflict Management Guides, Proceedings of ICED, Tampere, August 1997 Niu, M (1999), Airframe Structural Design , Conmilit Press Ltd. Prasad, B.(1995 ), Concurrent Engineering Fundamentals: Integrated Product and Process Organization, Volume I Hardcover - 502 pages, Prentice Hall, Raymer D. (1999) Aircraft Design: A Conceptual Approach (AIAA Education Series) by Hardcover 3rd edition (August 1999) Sycara, K. P. (1991). Cooperative Negotiation in Concurrent Engineering Design. Computer-Aided Cooperative Product Development. D. Sriram and R. Logcher, ed. New York: Springer-Verlag. 269-297. T. Tomiyama (1993), “Towards knowledge intensive intelligent CAD,” JSME-ASME Workshop on Design, pp. 46-51, 1993. Klein (1992), M., Supporting Conflict Management in Cooperative Design Teams, Proceedings of the 11th International Workshop on DAI, Glen Arbor, MI, 1992. Petrie C.,(1995), Teresa A. Webster, Mark R. Cutkosky, "Using Pareto Optimality to Coordinate Distributed Agents", AIEDAM special issue on conflict management Vol. 9, pp 269-281, 1995. Sycara, K. P. (1991). Cooperative Negotiation in Concurrent Engineering Design. Computer-Aided Cooperative Product Development. D. Sriram and R. Logcher, ed. New York: Springer-Verlag. 269-297. J. Favela, A. Wong, and A Chakravarthy; "Supporting collaborative engineering design" Taratoukhine, V. (1999), K. Bechkoum. Towards a Consistent Distributed Design: A Multi- Agent Approach, Proc Information Visualisation '99, London, IEEE Press, (1999) Unan R., (1992) and E. Dean. Elements of Designing for Cost, presented at The AIAA 1992 Aerospace Design Conference, Irvine CA, 3-6 February, AIAA-92-1057, (1992) 10