复合材料格栅结构优化设计中的计算智能技术

针对复合材料格栅结构优化设计多变量、多约束、连续和离散混合变量、高度非线性的难点,提出了用进化神经网络来实现结构设计参数(输入)与结构响应参数(输出)的全局非线性映射关系,以此来代替优化过程中的有限元计算,以提高优化效率.以遗传算法为优化求解器,神经网络屈曲稳定性响应面为主要约束,对复合材料格栅加筋结构进行优化.结果表明,在相同样本数的情况下,进化神经网络可获得比BP网络更高精度的映射模型,具有很强的泛化能力.该方法可以为解决大型复合材料结构优化问题提供一条高效途径. 

Computational intelligence technology for optimal design of grid-stiffened composite structure

To overcome the difficulties of optimal design for grid-stiffened composite structures,such as multi-variables,multi-constraints,mixed discretecontinuous design variables,highly nonlinear,etc,the application of computational intelligence(CI),namely evolutionary neural networks(ENN) was considered for realizing the global nonlinear mapping between structural design parameters and structural responses.They were aimed to replace the finite element computation during an actual optimization process so as to raise the efficiency of optimization.By using genetic(algorithm(GA)) as the optimization procedure and the structural buckling constraint as the neural network response surface,the optimal design of grid-stiffened composite panel under axial compressive loads was studied.The results indicate that with very limited FEM sample space,the accuracy of the evolutionary buckling neural network is much higher than that of traditional BP neural network.The resulted ENN-GA algorithm proves that it can offer an efficient approach to the optimization design of large complex composite structures.