平纹机织碳纤维复合材料的多尺度随机力学性能预测研究

平纹机织碳纤维复合材料在结构上具有多尺度特性和空间随机性. 同时, 组分材料会因存储条件和组成相成分、批次的不同导致力学性能有所差异. 当考虑各尺度结构和组分性能参数不确定性进行随机力学性能预测时, 存在以下两个难点: 一是随机变量众多, 使得对不确定性传递方法的精度和效率提出了要求; 二是由于随机参数之间存在高维相关性, 需要建立高精度的相关性模型. 针对以上问题, 本文提出了基于混沌多项式展开和Vine Copula的平纹机织复合材料多尺度随机力学性能预测方法, 综合考虑了平纹机织碳纤维复合材料微观及介观尺度的材料、结构随机参数, 基于自下而上层级传递的策略逐尺度地研究力学性能不确定性. 该方法采用Vine Copula理论构造相关随机变量的高维联合概率分布, 并运用非嵌入式混沌多项式展开法实现不确定性传递. 结果显示, 本方法构造的相关性模型几乎与原模型一致, 且能够高效准确地实现各尺度力学性能的随机预测. 

RESEARCH ON MULTISCALE STOCHASTIC MECHANICAL PROPERTIES PREDICTION OF PLAIN WOVEN CARBON FIBER COMPOSITES1)

Plain woven carbon fiber composites have multi-scale characteristics and spatial randomness in structure. Meanwhile, the mechanical properties of the component materials vary due to different storage conditions, composition phase components and batches. When the stochastic mechanical properties of plain woven carbon fiber composites are predicted with considering of the parameter uncertainty at different scales, there are two main difficulties: first, the large number of random variables makes the accuracy and efficiency of the uncertainty propagation method required; second, a high-precision correlation model is needed to be established because of multi-dimensional correlations. To solve above problems, this paper proposes a multi-scale prediction method based on polynomial chaos expansion and vine Copula for the stochastic mechanical properties of plain woven composites. The random parameters of materials and structures at the microscopic and mesoscopic scales of the plain woven composites are taken into account, and the uncertainties of mechanical properties are studied scale by scale based on the bottom-up hierarchical propagation strategy. In this method, Vine Copula theory is used to construct the multi-dimensional joint probability distribution of correlated random variables, and the non-embedded polynomial chaos expansion is used to realize uncertainty propagation. Results show that the correlation coefficients of the dependence model constructed by the proposed method are almost the same as that of original data and the stochastic prediction of mechanical properties at different scales are realized efficiently and accurately.