DrivAerStar

This repository contains two complementary sub-projects:

• Production (DrivAerStar_Maker): Tools and scripts for generating, processing, and exporting DrivAer models (case/VTK, etc.).
• Testing/Benchmarking (DrivAerStar_Benchmarking): A benchmarking suite (e.g., for the Transolver model) for deep learning training, validation, and inference based on the DrivAerStar dataset, supporting irregular mesh data and GPU-accelerated training.

Directory Structure (Example)

.
├── DrivAerStar_Maker/           # Production tools (STL/Mesh processing/Conversion)
├── DrivAerStar_Benchmarking/    # Benchmarking training & inference code (Models/Data modules/Configs)
└── README.md

---

1. General Environment Setup

Recommended Python version: >= 3.10. It is advised to run within a virtual or conda environment.

Install general dependencies (each sub-project might also have its own requirements.txt):

pip install -r requirements.txt

If certain dependencies cannot be installed via pip (e.g., specific PyTorch/CUDA versions, mesh libraries), please manually install the corresponding packages as indicated by the error messages or modify requirements.txt.

---

2. DrivAerStar_Maker Production Guide

Purpose: Use Blender and scripts to automatically generate vehicle models (STL), adjust wheel alignment, generate simulation code, and export to formats like VTK, supporting subsequent benchmarking or CFD preprocessing.

Required Software & Versions:

• Blender 4.1
• Star-CCM+ 18.06.007-R8
• pyvista 0.44.0
• Paraview 5.13.3

Recommended Directory Structure:

DrivAerStar_Maker/
├── 1.code_make_stls_by_blender/    # Blender Scripts: Generate STLs
├── 2.code_Wheel_Alignment/         # Wheel Alignment & Geometry Adjustment
├── 3.make_java_code_E/             # Simulation Script Code Generation
├── 4.run/                          # Run/Simulation Script Execution
└── 5.case_to_vtk/                  # Convert Simulation Case to VTK/Visualization Formats

Example Execution:

1. Generate STLs (Run scripts using Blender's Python CLI blender --python or similar):

python doe.py
python make.blender.py

2. Wheel Alignment (Python scripts):

python parallelru_F.py
python parallelru_N.py
python parallelru_E.py

3. Generate Java Code:

python make.py

4. Run Simulation/Processing Scripts:

./allrun_10by10.sh

5. Convert to VTK:

./run.sh

Note:

• You may need to specify your own paths within each script.

---

3. DrivAerStar_Benchmarking (Testing / Benchmarking) Guide

Project Goal: Train and evaluate deep learning models (example: Transolver) on irregular meshes using the DrivAerStar dataset. Uses configurable YAML files to control data loading, model parameters, and training process. Supports GPU acceleration.

Recommended Directory Structure (Example)

DrivAerStar_Benchmarking/
├── configs/
│   └── Transolver/
│       └── Transolver_1V_400.yaml
├── data_module/
│   └── subset/
│       └── DrivAerStar_488.py
├── networks/
│   └── Transolver/
├── main_train.py
├── main_predict.py
└── requirements.txt

Environment Installation

cd DrivAerStar_Benchmarking
pip install -r requirements.txt

Training & Inference Examples

1. Training & Validation:

python main_train.py --config-path configs/Transolver/Transolver_1V_400.yaml

2. Inference/Prediction:

python main_predict.py \
  --config-path configs/Transolver/Transolver_1V_400.yaml \
  --load-ckpt-path logs/checkpoints/epoch=38-step=39.ckpt

Example Configuration File (Transolver_Irregular_Mesh.yaml snippet explanation)

Seed: 42
Data:
  args:
    batch_size: 8
    data_dir: "/dir/to/DrivAerStar/vtk/"
    val_batch_size: 1
    num_train: 50
    num_val: 10
    num_test: 10
    no_cache: False
    data_cache_file: "./dataset/cache_DrivAerStar_488.pt"
  type: DrivAerStar_488

Trainer:
  args:
    accelerator: gpu
    devices: 1
    max_epochs: 500

load_ckpt: null

network:
  args:
    lr: 0.001
    space_dim: 7
    out_dim: 4
    weight_decay: 0.0001
    n_hidden: 64
    n_layers: 4
    n_heads: 4
    max_grad_norm: 0.1
    downsample: 5
    mlp_ratio: 1
    dropout: 0.0
    ntrain: 50
    unified_pos: 0
    ref: 8
    slice_num: 16
    eval: 0
  type: Transolver_Irregular_Mesh