Résolution par satisfaction de contraintes appliquée à l'aide à la décision en conception architecturale. (Constraint satisfaction solving applied to decision support for embodiment design)

The embodiment and conceptual design phases of mechanical and energetic systems are concerned with preliminary design. The embodiment design phase aims at investigating the feasibility of concepts emerging from conceptual design. Embodiment achieves to make a decision concerning the main features of the architecture of a product, which will be detailed in the next design phase. The main characteristics of the product are relating to dimensions, choices of components, shapes, topologies or materials. Decisions can better be supported by carrying out models which mainly take into consideration functional specifications and physical behaviors, namely a mathematical definition of the designer knowledge. These models may be processed using a numerical Constraint Satisfaction Problem (CSP) solver. CSP solvers are based on generic algorithms suitable for the numerical treatment of problems regarded as sets of constraints, variables and domains. Through these models, designers aim at investigating design architectures, i.e. they have to match up values with the main variables of the problem (design variables and decision variables). We propose several new heuristics and algorithms taking into account some specific difficulties relating to preliminary design problems. This approach concerns the classification and relative precision of variables and piecewise constraints. The feasibility studies of several mechanical and energetic systems have been investigated. These new algorithms drastically decrease the calculation time of the numerical treatment, and also improves the quality of the computational results.