非均质点阵结构材料性能快速预测方法研究  

Study on Rapid Prediction Method of Material Properties of Heterogeneous Lattice Structures

作　　者：罗加享 李昱 姚雯 周炜恩 张泽雨 LUO Jiaxiang;LI Yu;YAO Wen;ZHOU Weien;ZHANG Zeyu(College of Aerospace Science and Engineering,National University of Defense Technology,Changsha 410073,China;National Innovation Institute of Defense Technology,Academy of Military Sciences,Beijing 100071,China;Intelligent Game and Decision Laboratory,Beijing,100000,China)

机构地区：[1]国防科技大学空天科学学院,长沙410073 [2]军事科学院国防科技创新研究院,北京100071 [3]智能博弈与决策实验室,北京100000

出　　处：《智能安全》2024年第2期42-54,共13页Artificial Intelligence Security

摘　　要：随着智能设计方法和先进制造技术的快速发展,高端装备呈现出智能化、轻量化、多功能化、仿生化和制造一体化等发展趋势。虽然点阵结构具有轻质高强、减振降噪、抗冲击吸能等优异性能,然而其一直存在微观建模复杂且性能表征分析耗时的难题。为此,本文提出了一种基于切割水平集的非均质点阵微结构建模与数据驱动的材料性能快速预测方法。首先,每个微结构原型由一个水平集函数隐式表示,多个微结构原型组成一个复合点阵微结构,通过改变微结构原型的类型和数量可组合成多种非均质微结构构型。然后,采用均匀化理论计算点阵微结构的等效弹性矩阵并生成数据集,并通过神经网络建立其切割高度变量到等效弹性矩阵与体积分数之间的映射,所得代理模型可快速预测出点阵微结构的材料性能参数,从而替代昂贵的均匀化计算。实验结果表明,本文提出的预测方法可精确表征非均质点阵微结构,构建的代理模型可大幅减少点阵微结构性能计算成本,且具有较高的精度和鲁棒性。With the rapid advancement of intelligent design methods and advanced manufacturing technologies,high-end equipment has witnessed a remarkable shift towards intelligent,lightweight,multi-functional,bio-inspired,and integrated manufacturing.Lattice structures have their exceptional properties,including light weight,high strength,effective vibration and noise reduction,shock resistance,and superior energy absorption capabilities.However,the modeling and performance analysis of lattice structures at the microscopic level can be complex and time-consuming.In this research,a novel approach,based on the cutting level set method,coupled with data-driven rapid prediction of material properties,was proposed for modeling heterogeneous lattice microstructures.Initially,each microstructure prototype was represented implicitly using a level set function.A composite lattice microstructure was then formed by combining multiple microstructure prototypes,enabling the creation of diverse heterogeneous microstructure configurations by altering the types and quantities of prototypes.Subsequently,the homogenization method was employed to calculate the equivalent elasticity tensor of the lattice microstructure,generating a series of datasets.A neural network was utilized to establish the mapping between the cutting height variable and the equivalent elasticity tensor,as well as the volume fraction.This enables the development of a surrogate model that can swiftly predict the material property parameters of lattice microstructures,thereby replacing the computationally expensive homogenization calculations.Experimental results demonstrate the effectiveness of the proposed method in accurately characterizing heterogeneous lattice microstructures.The developed surrogate model significantly reduces the computational cost required for microstructure property analysis,while maintaining high accuracy and robustness.

关 键 词：深度学习 水平集方法 点阵微结构 有限元分析 

分 类 号：V22[航空宇航科学与技术—飞行器设计] V42