This paper proposes an innovative micromechanics-based artificial neural network (ANN) method to efficiently investigate the transverse modulus of unidirectional fibre/epoxy composites under humid conditions. In this research, a novel approach is developed to establish relations between the geometrical, mechanical, and environmental properties of the microstructure and the material’s performance under transverse tension. A framework is developed to artificially generate periodic representative volume elements (RVEs) while taking into account the interphase region between fibre and matrix. The RVEs are analysed by the finite element method to obtain the transverse moduli of composites. Two-point correlation functions and principal component analysis techniques are applied to extract and compress statistical information from microstructure images. A database establishment framework is developed to create three batches of data with 1000, 2000, and 3000 sizes. An ANN-based prediction framework is developed by integrating 10-fold cross-validation and Bayesian optimisation to optimise the neural network architecture and establish an efficient structure-property linkage under the influence of humid conditions. The prediction results demonstrate the efficiency of ANN in mapping microstructural data to an effective transverse modulus. A parametric study by ANN reveals the role of microstructure geometrical features and humid environmental parameters on the transverse performance of the composite.

中文翻译：

---

人工神经网络预测随机分布单向纤维增强复合材料在潮湿条件下的横向模量：一种微观力学方法

本文提出了一种基于微力学的创新人工神经网络（ANN）方法，可有效研究潮湿条件下单向纤维/环氧树脂复合材料的横向模量。在这项研究中，开发了一种新方法来建立微观结构的几何、机械和环境特性与材料在横向张力下的性能之间的关系。开发了一个框架来人工生成周期性代表体积元素（RVE），同时考虑纤维和基体之间的相间区域。通过有限元方法对RVE进行分析，以获得复合材料的横向模量。应用两点相关函数和主成分分析技术从微观结构图像中提取和压缩统计信息。开发了数据库建立框架，创建了三批1000、2000、3000大小的数据。通过集成 10 倍交叉验证和贝叶斯优化，开发了基于 ANN 的预测框架，以优化神经网络架构，并在潮湿条件的影响下建立有效的结构-性能联系。预测结果证明了 ANN 将微观结构数据映射到有效横向模量的效率。人工神经网络的参数研究揭示了微观结构几何特征和潮湿环境参数对复合材料横向性能的作用。