To CDR Home    CDR Project

The ProcessLink Constraint Manager Agent


For Secretaries' Nightmare example loading, click here.
Design is a complex process which involves the coorperation of many people from different disciplines to work on multi-disciplinary design. One difficult problem can be coordination of different people on a design team, especially with regard to mutual satisfaction of constraints. For constraint satisfaction, difficulties can arise from several reasons such as implicit representation, ad-hoc based consistency check, inter-dependencies of diverse design considerations, dynamic constraints and other things. The constraints are not static. Given design tasks, the constraints are generated, added, deleted and modified during the design process.

Previous work on constraint based systems(CBS) has led to algorithms for solving collections of constraints. The resulting constraint satisfaction problem(CSP) formalism has been more promising for design than rule-based systems. However, in spite of successful implementations on simple part designs and small problems, there has been little success for design support frameworks. This is largely because implementations of algorithms have treated constraints and preferences as static and because the CBSs have tightly integrated the problem solving methods. For those reasons, CBSs have not proved to be useful for incremental revision support under dynamic constraints and preferences typical of collaborative design.

To develop a design support system that can make use of efficient CSP algorithms under dynamic constraints and preferences, we have defined the key issues as

1) revision reasons by constraint or preference change
2) bookkeeping
These principles provide the basis for problem solving task decomposition by distrubuted domain agents, a generic dependency management agent called "Redux", and a "Constraint Manager" agent that allows various constraints solvers to be used. Currently we are implementing this concept within the Next-Link system by extending the NextLink message protocol to include messages that provide the requisite functionality.

Our concept is proved to be useful for dynamic constraints and preferences management by solving "Secretaries' Nightmare" problem, which has not been completely solved by previous CBS. Certainly, real engineering problem is more complex than "Secretaries' Nightmare", but our system has more functionality than can be shown while solving "Secretaries' Nightmare" and mechanical engineering applications will be attempted

My future work includes

accommodating several solvers to support diverse domains(eg. discrete + continuous)
distributing the Constraint Manager's jobs
implementing Constraint Manager with the Next-Link system and testing on more cases.

Figure List


Heecheol Jeon : jhc@cdr.stanford.edu
Last modified: Fri Feb 21 17:47:06 PST 1997