The present thesis aims at developing a design method dedicated to the optimization of architectured sandwich panels for multifunctional properties following a “materials-by-design” approach. This method is based on a genetic algorithm which enables to deal with materials selection (discrete variables) and geometrical dimensioning (continuous variables) simultaneously. Three core architectures have been investigated: foams, hexagonal honeycombs and tetrahedral truss structures. In this thesis, two main paths for material selection are defined. In the first one, architectured materials are considered as existing materials with properties referenced in a closed materials database. This is called the “real path” optimization. In order to expand the range of possibilities in terms of materials selection, a semi-continuous description of the architectured materials is considered in the second path, which is called “virtual path” optimization. The core material is described by a constitutive material (discrete variable) and a set of continuous geometrical variables representing the architecture. Using these two aforementioned approaches, several working properties of sandwich panels have been evaluated: flexural stiffness and strength, acoustic damping, thermal resistance and insulation, and finally blast mitigation. Bi-objective optimizations were performed in order to optimize each property in a minimal weight design. Some tri-objective cases are also presented, thus assessing the compatibility between different specifications. Indeed, this is achieved by relating trade-off surface shape to the compatibility between specifications. The optimization results also help identify the optimal design regarding the different criteria. Using the “virtual path” approach, a direct comparison between the different core architectures is achievable. Nevertheless, by being global and dealing with mixed variables, the obtained optimization process is complex. Two mixed methods where genetic algorithm is coupled with other approaches are proposed in order to increase the analysis complexity while providing a reasonable optimization complexity.
