基于代理模型的复合材料带加强筋板铺层优化
Laminate optimization of a composite stiffened panel based on surrogate model

该文基于遗传算法对复合材料带加强筋板中加强筋的铺层顺序进行了优化, 使结构在质量一定的情况下结构屈曲载荷最大。为了减少优化过程中有限元模型的调用次数, 引入径向基神经网络作为代理模型对结构屈曲载荷进行估计, 并且将铺层参数作为其输入以降低目标函数的非线性。由于设计空间形状不规则, 采用D-optimal实验设计方法确定训练径向基神经网络的采样点集。考虑到代理模型存在估计误差, 提出了加强代理模型在暂定最优区域估计精度的方法。算例表明: 以铺层参数为输入的径向基神经网络在建立代理模型方面具有较高的精度和效率; 代理模型的局部精度加强可进一步提高代理模型在暂定最优区域的精度。
Abstract：This study optimized the stacking sequence of stiffeners in a composite stiffened panel to maximize the buckling load of the panel assuming a constant mass panel. The number of finite element models was reduced by using a radial basis function neural network (RBF) as the surragate model with the lamination parameters as inputs to estimate the buckling load. The lamination input parameters reduced the nonlinearities of the objective function. Due to the irregular shape of the design space, the D-optimal method was used to determine the sample points for training the RBF. The model errors were reduced by constructing a zoomed RBF to enhance the RBF accuracy near the provisional optimal laminate. A numerical example shows the accuracy and efficiency of the RBF with the lamination parameters as inputs and how the model accuracy is increased by the zoomed RBF near the optimal region.