ibm-research/hemolab-bench
收藏资源简介:
Hemolab Bench数据集包含参数化的3D CFD表面网格(非结构化网格,三角化),每个样本具有每个顶点的压力和壁面剪切应力(WSS)场。每个样本对应于由17个参数族(曲率、扭曲、锥度、雷诺数、泰勒数)生成的不同几何形状。数据集包含49,660个样本,所有样本共享相同的三角拓扑结构(9,600个三角形,25,600个顶点)。数据集预先分为训练集、验证集和测试集,以确保实验的可重复性。测试集还包括两个分类列(alpha_split和reynold_split),用于指示每个测试样本所属的分布区域,以便进行按区域评估。
A dataset of parameterized 3-D CFD surface meshes (UNSTRUCTURED_GRID, triangulated) with per-vertex pressure and wall shear stress (WSS) fields. Each sample corresponds to a distinct geometry generated from a 17-parameter family (curvature, twist, taper, Reynolds number, Taylor number). The dataset contains 49,660 samples, all sharing the same triangulation topology (9,600 triangles, 25,600 vertices). The dataset ships with pre-built splits (train, validation, test) for reproducible experiments. The test set includes two categorical columns (alpha_split, reynold_split) that indicate which distribution region each test sample belongs to, enabling per-regime evaluation.




