Composite structural optimization design based on neural network response surfaces

To construct neural-network response surfaces for composite structural optimal design, the Orthotropic Experimental Method (OEM) was used to select the most appropriate structural analysis sample points. The constructed response surfaces were used as the objective function or constraint conditions. Together with other conventional constraints, they form an optimization design model which can be solved by using genetic algorithm (GA). This approach is highly applicable for complex composite structural design. Taking a hat-stiffened composite plate as example, the weight response surface was developed as the objective function, and strength and stability response surfaces as constraints; all these neural networks were trained by PATRAN/NASTRAN computation. The optimization results illustrate that it can significantly reduce the cycles of FEM analysis and achieve highly accurate response approximation results. And eventually, the approach can greatly raise the efficiency of the optimization process.