Data underlying the PhD thesis: Machine learning for complex fluid mechanics and heat transfer

Dataset and code related to the PhD thesis: "Machine learning for complex fluid mechanics and heat transfer", Rafael Diez Sanhueza, 2024.<br>The objective of the research was to study machine learning for fluid mechanics, considering rough surfaces and variable property flows (separately). The baseline data is obtained from post-processed DNS cases, or field inversion (non-linear optimization) in the study for variable-property flows. The full RANS solver, field inversion optimizer, and neural network system of Chapter 3 (variable property flows) is included. The 2-D maps with the local skin friction factors and Nusselt numbers of rough surfaces are generated by the wall force/heat_flux interpolation software attached, starting from 3-D fields with time-averaged DNS data. The DNS solver to simulate turbulent flows past rough surfaces in Chapter 5 is included, along with the full implementation of the immersed-boundary method. All data corresponds to text files, without binaries. Files resembling the JSON format are mainly used. The tecplot files for the rough surfaces can be readily opened in Paraview for 3-D visualization, or read as CSV files (the header has a simple format).

本数据集及配套代码对应拉斐尔·迭斯·桑胡萨（Rafael Diez Sanhueza）2024年的博士论文《面向复杂流体力学与传热的机器学习》。本研究旨在分别针对粗糙表面与变物性流动场景，开展流体力学领域的机器学习研究。基准数据取自后处理的直接数值模拟（Direct Numerical Simulation, DNS）案例，或是变物性流动研究中的场反演（非线性优化）结果。本数据集包含论文第三章（变物性流动场景）所涉及的完整雷诺平均Navier-Stokes（Reynolds-Averaged Navier-Stokes, RANS）求解器、场反演优化器与神经网络系统。附带的壁面力/热通量插值软件可基于搭载时间平均直接数值模拟数据的三维流场，生成粗糙表面的局部皮肤摩擦系数（local skin friction factors）与努塞尔数（Nusselt numbers）二维映射图谱。第五章中用于模拟粗糙表面绕流湍流的直接数值模拟求解器，以及浸入边界法（immersed-boundary method）的完整实现代码均已收录。所有数据均采用文本文件格式，无二进制文件；类JSON格式文件为主要存储载体。粗糙表面的Tecplot文件可直接在Paraview中打开以进行三维可视化，亦可作为逗号分隔值（CSV）文件读取，其表头格式简洁规范。