Virtual Tensile Test Dataset of Stress–Strain Response in Long Discontinuous Fiber Composites with Stochastic Mesostructure

Virtual uniaxial tensile test coupons were generated with stochastic Prepreg Platelet Molded Composite (PPMC) mesostructure. Progressive Failure Analysis was performed on Abaqus Standard using Continuum Damage Mechanics (CDM) and Cohesive Zone Modelling (CZM) methods. The stochastic mesostructure information of each sample was processed such that the explicitly represented platelet geometry and fiber orientations from the Finite Element (FE) model were reduced to compact mesostructure descriptors in the form of in-plane distributions of second-order fiber orientation tensor components a11 and a12. The layerwise a11 and a12 distributions (high-resolution mesostructure descriptors) were further reduced to coarse mesostructure descriptors by locally averaging the a11 and a12 values through the thickness at each voxel location across the length and width of a coupon. The macroscopic (effective) stress-strain data for each coupon were also preprocessed to include a) the strain at peak stress, b) terminal strain (corresponding to the simulation cut-off point at 10% load drop) and c) 40 stress values at prescribed fractions of the previously mentioned reference strains a and b. Both the preprocessed (normalized) and raw (non-normalized) stress-strain data, as well as coarse and high-resolution mesostructure descriptors for 3400 unique virtual PPMC tensile test samples are included in the attachments. An accompanying dataset guide spreadsheet documents the file-naming convention, pre-processing applied on the raw data, material properties of each sample and dataset-wide summary statistics. An example python script is also provided to visualize the inputs (fiber orientation distribution at low and high resolution) and outputs (stress-strain curves) for individual samples. Research Hypothesis: This dataset enables training of data-driven surrogate models to learn mappings between fiber orientation distributions and stress–strain response of PPMCs. In a related study by the authors (to be linked upon publication), a deep learning–based surrogate model trained on this dataset demonstrated that coarse mesostructural descriptors in the form of through-thickness–averaged fiber orientations can serve as effective structural representations for multiscale analysis of stress–strain response in PPMCs. Using only these coarse descriptors, tensile stiffness was predicted with a mean absolute percentage error (MAPE) below 3% while tensile strength and failure strain were predicted with a MAPE below 10%. These results highlight the potential of reduced-order structural descriptors to lower experimental characterization and computational modeling requirements for materials with complex, spatially heterogeneous subscale morphology.研究生成了带有随机预浸料片状模塑复合材料 (Prepreg Platelet Molded Composite, PPMC) 细观结构的虚拟单轴拉伸试样，并采用连续介质损伤力学 (Continuum Damage Mechanics, CDM) 与内聚力区建模 (Cohesive Zone Modelling, CZM) 方法，在Abaqus Standard中开展了渐进失效分析。 对每个试样的随机细观结构信息进行处理，将有限元 (Finite Element, FE) 模型中显式表达的片状几何与纤维取向，简化为以二阶纤维取向张量分量a₁₁和a₁₂的面内分布形式呈现的紧凑细观结构描述符。分层的a₁₁与a₁₂分布（高分辨率细观结构描述符）还可进一步简化为粗粒度细观结构描述符：通过在试样长度与宽度方向的每个体素位置处，沿厚度方向对a₁₁和a₁₂取值进行局部平均实现。 同时对每个试样的宏观（有效）应力-应变数据进行预处理，预处理内容包括：a) 峰值应力对应的应变；b) 终止应变（对应载荷下降至10%时的模拟截断点）；以及c) 基于前述参考应变a与b的指定比例下的40个应力值。 附件中包含3400个独立虚拟PPMC拉伸试样的经预处理（归一化）与原始（未归一化）应力-应变数据，以及粗粒度与高分辨率细观结构描述符。附带的数据集指南电子表格详细说明了文件命名规则、原始数据的预处理流程、每个试样的材料属性，以及全数据集的汇总统计信息。同时还提供了示例Python脚本，用于可视化单个试样的输入数据（低分辨率与高分辨率纤维取向分布）与输出数据（应力-应变曲线）。 研究假设： 本数据集可用于训练数据驱动的替代模型，以学习PPMC的纤维取向分布与应力-应变响应之间的映射关系。在作者团队后续待发表的相关研究中，基于本数据集训练的深度学习替代模型证实：以厚度平均纤维取向形式呈现的粗粒度细观结构描述符，可作为PPMC应力-应变响应多尺度分析的有效结构表征手段。仅使用这些粗粒度描述符，即可实现拉伸刚度的预测，其平均绝对百分比误差 (mean absolute percentage error, MAPE) 低于3%；拉伸强度与失效应变的预测MAPE则低于10%。上述结果凸显了降阶结构描述符的应用潜力，可降低具有复杂空间异质子尺度形貌的材料的实验表征与计算建模需求。