A general optimization strategy for composite sandwich structures

An original strategy for the design of large-scale composite sandwich structures is developed. Such structures are typical of launcher and spacecraft structures, but the methodology remains applicable to any thin-walled sandwich structure (such as aircraft fuselage panels or watercraft hulls). The method hinges on multi-step strategies devised for variable-stiffness laminates. Indeed, laminate optimization problems correspond to large combinatorial design spaces and their efficient approximate solutions often combine continuous relaxation and metaheuristics. Typically, the first step consists of a gradient-based optimization that handles the large number of variables with a continuous representation of the composite mechanical behavior. In our work, this continuous representation is adapted to the specific features of laminated sandwich composites. In particular, we modeled the dependency of the material transverse shear stiffness on the design variables (composite skins and core) by employing a machine learning-based approximation. The face sheet laminates are then determined in a second step using an evolutionary algorithm with stacking sequence tables to represent thickness variations. Finally, we introduce a third optimization step to overcome the design feasibility issues related to the use of a stiffness matching method for the determination of the layups of the face sheet laminates. The method is applied to the minimization of the overall weight of a load-carrying structure similar to the Ariane 5 dual-launch system, under several launcher-specific mechanical constraints.