2D Binary Images and Effective Thermal Conductivity CFD Results

This dataset was originally created to train a CNN for predicting the effective thermal conductivity of these binary structures based on geometry alone. A total of four different ratios of thermal conductivity between the two phases were simulated. The original dataset contains 40,000 unique 128x128 binary structures, and is further expanded by flipping the color scheme, rotating the image 90 degrees, and doing both simultaneously. That is done to the folders CirclePack, EllipsePack, and QuadrilateralPack, expanding the dataset to 130,000 unique structures. Therefore, in the folders below are the 40,000 original images (10,000 in each folder) and all of the CFD results (520,000 total simulation results). For more detail on structure generation and the CFD algorithm, refer to the manuscript. Pre-proof manuscript: Adam, A., Fang, H., & Li, X. (2023). Effective thermal conductivity estimation using a convolutional neural network and its application in topology optimization. In Energy and AI (p. 100310). Elsevier BV. https://doi.org/10.1016/j.egyai.2023.100310

本数据集最初用于训练卷积神经网络（Convolutional Neural Network, CNN），以仅基于几何结构预测此类二元结构的有效导热系数。研究中共模拟了两相之间四种不同的导热系数比值。原始数据集包含40000组唯一的128×128二元结构样本，随后通过翻转配色方案、将图像旋转90度，以及同时执行这两种操作的方式进行数据扩充。上述扩充操作应用于CirclePack、EllipsePack与QuadrilateralPack三个文件夹中的样本，最终将数据集规模扩充至130000组唯一的结构样本。因此，下方文件夹中包含40000张原始图像（每个文件夹含10000张）以及所有计算流体动力学（Computational Fluid Dynamics, CFD）仿真结果，总计520000组仿真结果。若需了解结构生成方法与CFD算法的更多细节，请参阅相关手稿。预印本手稿：Adam, A., Fang, H. 与 Li, X.（2023）。《基于卷积神经网络的有效导热系数估算及其在拓扑优化中的应用》，载于《Energy and AI》（第100310页），爱思唯尔（Elsevier BV）出版。https://doi.org/10.1016/j.egyai.2023.100310

搜集汇总

背景与挑战

背景概述

该数据集包含130,000个独特的128x128二进制图像，涵盖圆形、椭圆形和四边形等几何结构，并提供了基于四种热导率比（10、50、100、341.7）的520,000个CFD模拟结果。数据集专为训练卷积神经网络（CNN）预测有效热导率而设计，适用于能源工程、材料科学和机器学习等领域的研究。