中文摘要：

以蜂窝铝芯几何结构参数对其面内等效性能影响为研究对象,将正交试验和均匀化理论与有限元相结合来获得数据样本,建立了蜂窝铝芯几何结构参数与其等效弹性性能参数之间复杂非线性映射关系的网络模型,并利用贝叶斯正则化算法,实现了BP神经网络对蜂窝铝芯力学性能的预测.在较少样本数据的情况,可以较高精度地预测胞元结构参数对蜂窝铝芯等效性能的影响规律.提取该BP模型中各层的权值,运用Garson算法得到各结构参数对蜂窝铝芯等效力学性能影响程度的灵敏度系数,结果表明灵敏度分析可评估结构参数对等效力学性能的影响,可为蜂窝铝芯设计提供参考.

英文摘要：

Considering influence of the geometry structure parameters of aluminum honeycomb core on its equivalent performance as research object,the sample data were got from homogenization theory and FEM combined with the orthogonal experiment.The artificial network model was established to simulate the complex nonlinear mapping relation between the structure parameters and the effective properties for aluminum honeycomb core.By Bayesian regularization algorithm,the prediction of mechanical properties for aluminum honeycomb core was realized.The high precision was acquired to predict the rule of the structure parameters on the effective properties of aluminum honeycomb core with small sample data.Extracting the weight of each layer in the BP model,the sensitivity coefficients which reflect the effect of structural parameters on the equivalent mechanics performance were got with Garson algorithm.The results show that sensitivity analysis can evaluate the influence of the structure parameters on the equivalent mechanical properties and be used as reference in designing honeycomb core.