Artificial Intelligence techniques in design optimization of variable stiffness cylindrical shells

The main driver of aerospace structures design is the increase in performances of currently in use components. The behavior of structures is investigated by means of highly accurate finite elements (FE) analysis. The problem related to this kind of simulations is the high computational time required to obtain the structural response associated with nonlinear phenomena. This aspect is particularly significant during the preliminary phase, especially when the analysis involves an optimization procedure. One strategy to overcome this problem is the introduction of artificial intelligence techniques in the design phase. This work proposes an optimization framework based on the approximation of the structural behavior through an artificial neural network (ANN). The net is exploited during the optimization, performed with a particle swarm optimizer, in order to reduce the computational effort. FE analysis are used to train the ANN and to validate the results. The methodology is applied to the optimization of the fibers shape of variable stiffness cylindrical shells, with the goal of maximize the critical load taking into account also manufacturing constraints. The higher accuracy offered by ANN with respect to other global approximation techniques and the time saving, resulting from the developed methodology, are both highlighted.