面向AI时代的纤维增强树脂基复合材料工艺仿真  

Process simulations of fiber reinforced polymer composites towards AI ages

作　　者：周钰博 李敏[2] 王绍凯[2] 顾轶卓[2] 陶飞[1] 陈祥宝 张佐光[2] ZHOU Yubo;LI Min;WANG Shaokai;GU Yizhuo;TAO Fei;CHEN Xiangbao;ZHANG Zuoguang(Digital Twin International Research Center,International Institute for Interdisciplinary and Frontiers,Beihang University,Beijing 100191,China;School of Materials Science and Engineering,Beihang University,Beijing 100191,China;AECC Beijing Institute of Aeronautical Materials,Beijing 100095,China)

机构地区：[1]北京航空航天大学国际前沿交叉科学研究院数字孪生国际研究中心,北京100191 [2]北京航空航天大学材料科学与工程学院,北京100191 [3]中国航发北京航空材料研究院,北京100095

出　　处：《航空材料学报》2024年第5期17-36,共20页Journal of Aeronautical Materials

摘　　要：纤维增强树脂基复合材料制造工艺是保证其产品结构效率和应用可靠性的关键,通过计算机进行工艺仿真是提高复合材料制造质量与降低制造成本的重要手段。传统工艺仿真依赖于制造过程中的物理化学机理,通过有限元/有限体积等数值计算方法,以及计算机图形学等辅助设计方法来求解相关机理模型的数学方程,目前已在增强体/预浸料的铺覆、树脂的渗透流动、热固性树脂的固化行为、热传导与热交换、非线性力学及残余应力与固化变形预测等方面得到广泛应用。近年来,人工智能(AI)的迅猛发展,其技术基础机器学习(ML)与人工神经网络(ANN)相结合,已用于增强体铺覆、液体成型工艺和热压罐工艺领域,主要目的是数据挖掘和建立降阶模型。前者可以建立工艺条件与制件固化质量、力学性能等之间的关系,后者则可以提高工艺仿真的计算效率。然而受限于纤维增强树脂基复合材料制造过程复杂、不可测、成本高的特点,在AI时代的起点,仅依赖实验获得的数据量难以满足ML的要求,同时数据驱动AI还面临模型代表性、普适性、可解释性不确定的问题。因此,基于物理化学机理的传统工艺仿真可为数据驱动ML仿真提供大量可靠数据,进而通过AI建立更多描述复合材料工艺的定量模型,扩展工艺仿真可计算的过程;同时,通过AI技术提高计算效率后,满足实时性要求的工艺仿真可进化为制造过程的数字孪生(DT),从而可为复合材料降低成本、提高全寿命周期管理的科学性提供新的技术支撑。The manufacturing process of composites is crucial for ensuring structural efficiency and application reliability of their products.Computer-based process simulation plays a significant role in improving the manufacturing quality of composite components and reducing the manufacturing cost.Traditional process simulation relies on physical and chemical mechanisms in manufacturing of composites with the mathematical equations solved by numerical methods such as finite element/finite volume analysis and computer-aided design methods e.g.computer graphics.At present,it has been widely used in simulations of the lay-up of reinforcements/prepregs,the infiltration flows of resin,the curing behaviours of thermosetting resin,the heat transmission and exchange,and the nonlinear mechanics including residual stress and curing deformation predictions.Recently,artificial intelligence(AI)technologies have rapidly developed,its technical basis machine learning(ML),in combination with artificial neural networks(ANN),has been used in the field of lay-up process of fiber reinforcements,liquid molding processes,and autoclave processes,which aimed for data mining and developing reduced-order models.The former can establish relationships between process conditions and the curing quality or mechanical properties of the composite parts,while the latter can improve computational efficiency of the process simulation.However,due to the complexity,immeasurability,and high cost of manufacturing fiber-reinforced resin matrix composites,at the beginning of the AI age,it is difficult to meet the requirements of ML only by relying on the amount of data obtained by experiments.Also,data-driven AI technology faces uncertain issues regarding the representativeness,generalisability,and interpretability of the models.Therefore,traditional process simulation based on physicochemical mechanisms can provide a large amount of reliable data for data-driven ML simulation,and then through AI,more quantitative models describing the composite process can be es

关 键 词：复合材料 工艺仿真 人工智能 机器学习 固化仿真 固化变形 

分 类 号：V258[一般工业技术—材料科学与工程] TB332[航空宇航科学与技术—航空宇航制造工程]