A Hybrid CFD–ML Approach for Rapid Assessment of Particle Dispersion in a Port-Industrial Environment - Repository

The current repository is deemed to share the data for the training of the ML model and the different scripts used during the preprocess and training phase. Full CFD cases cannot be shared due to the large size, however the cell centers of the mesh will be attached for reference. The files descriptions are:- particle_prediction_dataset_scalar.h5 -> it contains the data needed for training the different ML models. The structure is: X: Input features. Y: Target fields. case_names: identifier for each name. z_slices: height slice values. While the attributes: nx: 1000, grid resolution in x-direction. ny: 1000, grid resolution in y-direction. x_bounds, spatial bounds in x. y_bounds, spatial bounds in y. - best_model.pth -> it contains an example of a final model for the particles field prediction. - C_full -> OpenFOAM file created which contains the cell centers of the full size mesh. - preprocess.py -> example script for preprocessing the raw data from OpenFOAM. - train_mlp_scalar.py -> example script for training the model.本仓库旨在分享机器学习（ML）模型训练所需的数据，以及预处理与训练阶段使用的各类脚本。鉴于完整计算流体动力学（Computational Fluid Dynamics, CFD）算例体积过于庞大，无法对外分享，但我们将附上全尺寸网格的单元中心数据以供参考。文件说明如下： 1. `particle_prediction_dataset_scalar.h5`：包含训练各类机器学习模型所需的数据。其数据结构为： - `X`：输入特征 - `Y`：目标场 - `case_names`：各样本的名称标识符 - `z_slices`：高度切片数值 附带属性说明： - `nx`：1000，x方向网格分辨率 - `ny`：1000，y方向网格分辨率 - `x_bounds`：x方向空间边界 - `y_bounds`：y方向空间边界 2. `best_model.pth`：包含粒子场预测任务的最终模型示例。 3. `C_full`：所创建的OpenFOAM文件，包含全尺寸网格的单元中心数据。 4. `preprocess.py`：用于预处理OpenFOAM原始数据的示例脚本。 5. `train_mlp_scalar.py`：用于训练该模型的示例脚本。