Conceitos Básicos
DrivAerNet provides a comprehensive dataset for aerodynamic car design, enabling efficient drag prediction through machine learning.
Resumo
The study introduces DrivAerNet, a large-scale CFD dataset of 3D car shapes, and RegDGCNN model for aerodynamic design. DrivAerNet addresses the need for extensive datasets in engineering applications, offering detailed 3D car meshes with aerodynamic performance data. RegDGCNN provides high-precision drag estimates directly from 3D meshes, facilitating rapid aerodynamic assessments. The dataset is publicly available and promises to accelerate the car design process.
Directory:
- Introduction
- Importance of aerodynamic design in reducing fuel consumption and CO2 emissions.
- Challenges in current high-fidelity CFD simulations and wind tunnel tests.
- Related Work
- Overview of existing datasets for data-driven aerodynamic design.
- DrivAerNet Dataset
- Detailed numerical simulation methods and characteristics of the dataset.
- Dynamic Graph Convolutional Neural Network (RegDGCNN)
- Introduction to the model architecture for drag prediction.
- Surrogate Modeling of Aerodynamic Drag
- Evaluation of RegDGCNN on DrivAerNet and ShapeNet datasets.
- Limitations and Future Work
Estatísticas
DrivAerNet comprises 4000 detailed 3D car meshes using 0.5 million surface mesh faces.
RegDGCNN achieves an R^2 score of 0.9 on the DrivAerNet dataset.