Minimum Weight Optimization of Composite Stiffened Panels Using Neural Networks

This paper deals with the definition of an optimization procedure for the design of composite stiffened panels subjected to axial compression loads and post-buckling constraints. The structures have to work in post-buckling field with local skin buckling between stiffeners and with high ratios between collapse and buckling loads. The optimization procedure is based on global approximation strategies and genetic algorithms. Systems of neural networks trained by using a minimum set of finite element analyses are proficiently used to predict structural behavior inside the optimization domain. The use of a global approximation strategy reduces considerably computational efforts also offering the complete separation between system modeling and optimization problem. This aspect is here exploited to carry out general guidelines in the design of stiffened composite panels by drawing response surfaces of some meaningful structural indices and performing two different minimum weight optimizations changing typology and values of the constraints.