Numerically predicted permeability of over 6500 artificially generated fibrous microstructures

This data set was generated in the project "ML4ProcessSimulation - Machine Learning for Simulation Intelligence in Composite Process Design" (Leibniz Collaborative Excellence funding program: K377/2021), at Leibniz-Institut für Verbundwerkstoffe GmbH. The goals were to create a comprehensive data set for training different neural networks and to gain insight into the influence of fiber structure on permeability. The models represent the fiber structure within fiber bundles in fiber-reinforced plastic composites (FRPC). Over 6500 structure models were generated in the software GeoDict® [1] and the permeability tensor of these models was numerically calculated in the GeoDict® module FlowDict [2]. The zip files contain the structure file (gdt), the model generation result file (FiberGeo_[...].gdr) and the flow simulation result file (LIRStokesResult_[...].gdr). For each zip file is a JSON meta data file available and in addition the gdr files contain all input and output data of the model generation and the flow simulation. The file Table_of_Parameter_studies_and_model_pictures.jpg gives an overview of the parameter studies and exemplarily shows two models each.The data set is divided into three parameter studies: 1_Parameter_study_round_fibers with round fibers by varying the fiber volume content (fvc), fiber diameter (fdia) and fiber orientation (fdir). For each modeling parameter, 5 - 100 models (random seed or RS) were generated, all differing due to the randomized fiber positioning during model generation. 2_Parameter_study_elliptical_fibers with elliptical fibers that was varied based on different aspect ratios (asp1, asp2, asp3). In addition, fdia and fvc were varied and 5 models (RS) were calculated. 3_Parameter_study_undulation with elliptical fibers, whose undulation was varied. In addition, fdia and fvc were varied and 12 models (RS) were calculated. [1] J. Hilden, S. Rief, and B. Planas, GeoDict 2023 User Guide. FiberGeo handbook. DE: Math2Market GmbH, 2023. Accessed: Oct. 26, 2023. [Online]. Available: https://doi.org/10.30423/userguide.geodict [2] J. Hilden, S. Linden, and B. Planas, "GeoDict 2023 User Guide. FlowDict handbook." Math2Market GmbH, 2023. Accessed: Jul. 31, 2023. [Online]. Available: https://doi.org/10.30423/userguide.geodict本数据集生成于莱布尼茨复合材料研究所（Leibniz-Institut für Verbundwerkstoffe GmbH）的「ML4ProcessSimulation——复合材料工艺设计中面向仿真智能的机器学习」项目，该项目受莱布尼茨联合卓越资助计划（编号K377/2021）支持。本研究的核心目标为构建可用于训练各类神经网络的综合数据集，并深入探究纤维结构对渗透率的影响机制。 所构建的模型用于表征纤维增强塑料复合材料（fiber-reinforced plastic composites, FRPC）纤维束内部的纤维结构。研究人员借助GeoDict®软件[1]生成了超过6500组结构模型，并通过其FlowDict®模块[2]对这些模型的渗透率张量进行了数值计算。 各压缩包包含三类文件：结构文件（格式为gdt）、模型生成结果文件（命名格式为FiberGeo_[...].gdr）以及流动仿真结果文件（命名格式为LIRStokesResult_[...].gdr）。每组压缩包均对应一份JSON元数据文件；此外，gdr格式文件还存储了模型生成与流动仿真的全部输入输出数据。 文件Table_of_Parameter_studies_and_model_pictures.jpg可用于概览本次参数研究的整体内容，并示例性展示每组参数研究各两个模型示例。 本数据集分为三项参数研究： 1. 圆形纤维参数研究：通过调整纤维体积分数（fiber volume content, fvc）、纤维直径（fiber diameter, fdia）与纤维取向（fiber orientation, fdir）构建圆形纤维模型。针对每个建模参数，生成5至100组模型（基于随机种子RS生成），由于模型生成过程中纤维排布存在随机性，各组模型均存在差异。 2. 椭圆形纤维参数研究：基于不同长径比（asp1、asp2、asp3）调整椭圆形纤维参数，同时同步改变fdia与fvc，共生成5组模型（基于随机种子RS生成）。 3. 纤维起伏参数研究：针对椭圆形纤维，调整其起伏程度参数，同时同步改变fdia与fvc，共生成12组模型（基于随机种子RS生成）。 参考文献： [1] J. Hilden、S. Rief与B. Planas，《GeoDict 2023用户指南：FiberGeo手册》，德国：Math2Market GmbH，2023年。访问时间：2023年10月26日。[在线资源]，可获取：https://doi.org/10.30423/userguide.geodict [2] J. Hilden、S. Linden与B. Planas，《GeoDict 2023用户指南：FlowDict手册》，Math2Market GmbH，2023年。访问时间：2023年7月31日。[在线资源]，可获取：https://doi.org/10.30423/userguide.geodict