A Distributed Generative CSP Framework for Multi-site Product Configuration

Abstract

Abstract. Today's configuration systems are centralized and do not allow manufacturers to collaborate online for offer-generation or sales-configuration activities. However, the integration of configurable products into the supply-chain of a business requires the cooperation of the various manufacturers' configuration systems to jointly offer valuable solutions to customers. As a consequence, there is a need for methods that enable independent specialized agents to compute such configurations. Several approaches to centralized configuration are based on constraint satisfaction problem (CSP) solving. Most of them extend traditional CSP approaches in order to comply to the specific expressivity and dynamism requirements of configuration and similar synthesis tasks. The distributed generative CSP (DisGCSP) framework proposed here builds on a CSP formalism that encompasses the generative aspect of variable creation and extensible domains of problem variables. It also builds on the distributed CSP (DisCSP) framework, supporting configuration tasks where knowledge is distributed over a set of agents. Notably, the notions of constraint and nogood are further generalized, adding an additional level of abstraction and extending inferences to types of variables. An example application of the new framework describes modifications to the ABT algorithms and furthermore our evaluation indicates that the DisGCSP framework is superior to classic DisCSP for typical configuration task problem encoding.

Modeling and Solving Distributed Configuration Problems: A CSP-Based Approach

“…In the following, we summarize the characteristics of asynchronous search algorithms such as Asynchronous Backtracking (ABT) [51], adapted in a way that they allow agents to know only the constraints that they enforce, see also [57].…”

“…In this section, we present technical details of our framework for modeling and solving distributed configuration problems based on the CSP paradigm, see also [57]. We combine the concepts of Generative CSPs (GCSP) and Distributed CSPs in order to support both intuitive and compact configuration knowledge modeling and supply chain integrated, cooperative solution search.…”

“…In the following, we provide a formal definition of a GSCP, which abstracts from the configuration task specific formulation and applies to the wider range of synthesis problems, see also [57].…”

“…In this section, we will formally describe how to extend an existing DisCSP approach described in Section 3.2 for the class of generative CSPs proposed in the previous section; see also [57]. In our work, we focus on the ABT algorithm, which in contrast to heuristic approaches is also capable of finding optimal solutions and for which various performance extensions exist.…”

“…A discussion and proof of the correctness and completeness of the proposed extensions for the ABT scheme are given in [57].…”