An Agent Architecture and Algorithm for Solving Distributed Configuration-design Problems

1. INTRODUCTION Configuration design is a type of design problem in which parts are selected from catalogs and connected to meet the following problem requirements:functionality, specifications, and constraints [7-9, 12]. Functionality defines what the design is supposed to do, specifications define optimality goals, and constraints def'me the feasibility relationships that must be satisfied for the design to operate correctly. A design is a collection of parts evaluated with respect to the problem requirements (functionality, specifications and constraints). A design that meets these requirements is a solution to the problem. Configuration design is a ubiquitous and economically important task, playing a prominent role in complex products such as automobiles, airplanes or computer systems that contain millions of parts, including resistors, light bulbs, screws, microprocessors and engines. Parts are described by attributes and implement one or more functions. Configuration design is difficult because parts can implement many functions (the multi-function part problem) [7, 8], functions can be implemented by many parts, parts may depend on other parts for their correct operation (the support function problem) [9], and constraints and specifications defined over part attributes restrict the allowed configurations, introducing horizon effects. There exist several techniques to reduce the number of designs to explore. If the functionality requirement is decomposed into required functions that individual parts implement, designs are constructed by selecting parts that implement each required function. Heuristics based on design experience or properties of the design problem can be used to reduce the search space by ruling out certain part selections. Large-scale configuration design problems that consist of thousands of required functions and millions of parts are too complex for a single agent 1 or small group of agents to manage. Such problems are recursively decomposed into sub-problems, until they are manageable. At this level, parts are selected to implement a small number of required functions, subject to constraints on the selected parts. However, many of these constraints are shared among sub-problems, the agents responsible for solving the sub-problems are often geographically distributed, and the catalogs from which parts are selected often reside outside the organization. Thus, solving large-scale problems requires communication among agents, and algorithms to resolve constraint violations on the shared constraints that link the sub-problems. Computer networks, such as the Internet, facilitate communication among agents making algorithms possible to resolve constraint violations, and provide a way to make the contents of part catalogs outside the design organization available …