Stochastic representations of fiber-based gas diffusion layers

Gas diffusion layers (GDLs) are relevant for the efficient fluid transport between the channel structure and the membrane electrode assembly (MEA) of fuel cells [1]. Black/white (BW) images of 25 realizations of a stochastic model represent the micro-structure of paper-type GDLs as manufactured by Toray. A binder model (5 representations) is combined with a fiber model (25 representations each). The 3D structures are represented by 130 images of size 512x512 each with a resolution of 1.5 µm/px. Every image represents a layer of 1.5 µm thickness. This leads to a total amount of 5*25*130=16250 images, arranged in a sub-folder structure that represents the binder model. 130 images of size 512x512 layers represent a section of 768 µm x 768 µm m 195 µm of a GDL. The fiber thickness is 7.5 µm. Binder material is located layer-wise along some fibers with a binder width of 6 µm, 18 µm, 30 µm, 40 µm or filled polygons (indicated as FF). The stochastic fundamentals are published in [2]. Transport simulations using the Lattice Boltzmann method were conducted and presented in [1;3-9]. Machine learning (ML) aspects were addressed in [10-11]. For binder with WW in {06, 18 30, 40, FF}, representation N in {1...25}, image number I in {1...130}, image path/names are: binder-WW/SimN/Image_512x512_N_No_I.png. Fig. 1 in [1] shows images with binder width of (A) 6 µm, (B) 18 µm, (C) 30 µm and (D) filled polygons. Fig. 3 in [3] extends the illustration by an 40 µm example, labelled as (D) in [3]. Subsequent simulations in [4-9] favored the binder width of 18 µm. The ML investigations [10, 11] covered the same binder widths as [1].

气体扩散层（Gas Diffusion Layers，GDLs）可实现燃料电池流道结构与膜电极组件（Membrane Electrode Assembly，MEA）之间的高效流体传输[1]。本数据集包含25组随机模型实现的黑白（Black/White，BW）图像，用以表征东丽（Toray）生产的纸质型气体扩散层的微观结构。该数据集将5种粘结剂模型表征结果与每种包含25种表征结果的纤维模型相结合。三维结构由130张尺寸为512×512的图像表征，每张图像的分辨率为1.5 µm/像素，对应厚度为1.5 µm的一层结构。由此总图像数量为5×25×130=16250张，数据集按照粘结剂模型构建子文件夹层级结构进行组织。130张512×512尺寸的图像共同表征了气体扩散层中尺寸为768 µm × 768 µm × 195 µm的区域，其中纤维的直径（厚度）为7.5 µm。粘结剂沿部分纤维分层分布，其宽度分别为6 µm、18 µm、30 µm、40 µm，或采用填充多边形结构（标注为FF）。该随机模型的基础理论已在文献[2]中发表。采用格子玻尔兹曼（Lattice Boltzmann）方法开展的传输模拟相关研究已在文献[1,3-9]中发表。机器学习（Machine Learning，ML）相关研究在文献[10-11]中有所涉及。当粘结剂宽度WW取值为{06、18、30、40、FF}，表征序号N取值为{1…25}，图像编号I取值为{1…130}时，图像的路径/命名格式为：binder-WW/SimN/Image_512x512_N_No_I.png。文献[1]中的图1展示了粘结剂宽度分别为（A）6 µm、（B）18 µm、（C）30 µm以及（D）填充多边形结构的图像示例。文献[3]中的图3补充展示了40 µm宽度粘结剂的示例，在该文献中标记为（D）。文献[4-9]的后续模拟结果显示，粘结剂宽度为18 µm时的性能更为优异。文献[10,11]中的机器学习研究覆盖的粘结剂宽度类别与文献[1]一致。