基于机器学习的梯度点阵材料优化设计

点阵材料具有轻质、抗冲击、高能量吸收等特性，因而在航天飞行器承载部件设计等领域有广阔应用前景. 通过对点阵材料内部杆径进行合理的梯度设计，可以提高点阵材料在高速冲击载荷作用下的动态力学性能. 利用仿真模拟数据，基于随机森林模型实现了梯度点阵材料的动态力学响应预测和结构参数优化. 以面心立方（face center cubic，FCC）结构梯度点阵材料为研究对象，通过对杆径参数的调整实现点阵材料密度的梯度化设计. 通过LS-DYNA软件计算了密度分布不同的梯度点阵材料受到冲击载荷作用时的动态力学响应，包括冲击端面与支撑端面接触应力随时间的变化曲线. 基于随机森林模型，以各层胞元的相对密度为输入，实现对点阵材料端面峰值应力的预测，并基于Gini指数分析出对不同端面处峰值应力影响最大的胞元层. 将网格搜索算法与训练好的随机森林对接，分别以两个端面上的峰值应力最高作为优化目标，获得点阵材料各层胞元相对密度的最优值. 模型对梯度点阵材料端面峰值应力的预测误差在5%以内. 数值模拟验证结果表明，优化后所得梯度点阵材料相应端面上的峰值应力高于仿真数据集内任何结构. 

Optimization Design of Graded Lattice Materials Based on Machine Learning

Lattice materials possess the characteristics of light weight, impact resistance, high energy absorption, so that they can be applied broadly in bearing part design of aero craft. The dynamic mechanical properties of the lattice materials under high speed impact can be improved by reasonably design of the internal bar diameter of the lattice materials. In this paper, employing simulation data, the dynamic mechanical response prediction and structural parameter optimization of graded lattice materials were carried out based on random forest model. Firstly, taking FCC graded lattice structure as study object, a gradient design of lattice material density was realized by adjusting the bar diameter parameters. And then, keeping the relative density of the whole lattice unchanged, the dynamic mechanical response of the graded lattice materials with different density distribution under impact loading was calculated based on LS-DYNA software, including the contact stress curve of the impact face and the support face over time. Finally, based on random forest model, taking the relative density of cells in each layer as input, the peak stress on the end face of lattice materials was predicted, and the cell layer with the greatest influence on the peak stress at different end face positions was analyzed with Gini index. And, connecting the grid search algorithm with a well trained random forest model, and taking the peak stress at the two end faces as the optimization objectives, the optimal value of cell rod diameter of the lattice material was obtained. The prediction error of the model is less than 5%. The numerical simulation results show that the corresponding peak stress of the optimized gradient lattice material is higher than that of any structure in the simulation data set.