Abstract:This study introduces DrivAerNet, a large-scale high-fidelity CFD dataset of 3D industry-standard car shapes, and RegDGCNN, a dynamic graph convolutional neural network model, both aimed at aerodynamic car design through machine learning. DrivAerNet, with its 4000 detailed 3D car meshes using 0.5 million surface mesh faces and comprehensive aerodynamic performance data comprising of full 3D pressure, velocity fields, and wall-shear stresses, addresses the critical need for extensive datasets to train deep learning models in engineering applications. It is 60\% larger than the previously available largest public dataset of cars, and is the only open-source dataset that also models wheels and underbody. RegDGCNN leverages this large-scale dataset to provide high-precision drag estimates directly from 3D meshes, bypassing traditional limitations such as the need for 2D image rendering or Signed Distance Fields (SDF). By enabling fast drag estimation in seconds, RegDGCNN facilitates rapid aerodynamic assessments, offering a substantial leap towards integrating data-driven methods in automotive design. Together, DrivAerNet and RegDGCNN promise to accelerate the car design process and contribute to the development of more efficient vehicles. To lay the groundwork for future innovations in the field, the dataset and code used in our study are publicly accessible at \url{

What problem does this paper attempt to address?

The paper aims to address key challenges in automotive aerodynamic design, particularly by accelerating the design process and improving vehicle efficiency through machine learning methods. Specifically, the research team developed two major outcomes: 1. **DrivAerNet**: This is a large-scale, high-fidelity Computational Fluid Dynamics (CFD) dataset containing 4,000 detailed 3D car models and their aerodynamic performance data. Each model includes approximately 0.5 million surface mesh faces and provides complete 3D pressure, velocity fields, and wall shear stress data. Compared to the largest existing public dataset, DrivAerNet is 60% larger and is the only open-source dataset that models both wheels and underbody structures. 2. **RegDGCNN**: This is a surrogate model based on Dynamic Graph Convolutional Neural Network (DGCNN) for directly predicting the drag coefficient from 3D meshes. The model can provide high-precision drag estimates within seconds, significantly speeding up the aerodynamic evaluation process. By introducing DrivAerNet and RegDGCNN, the research team addressed the following issues: - **Insufficient datasets**: Existing datasets are usually small in scale and do not fully consider the complexities in real-world applications, such as the impact of wheels, side mirrors, and underbody structures. These factors significantly affect aerodynamic performance. DrivAerNet compensates for this deficiency by providing a large and detailed dataset. - **Limitations of traditional CFD simulations**: Traditional high-fidelity CFD simulations are time-consuming and costly, while wind tunnel tests, although accurate, are also limited by time and cost. RegDGCNN bypasses these limitations by quickly predicting drag coefficients, enabling faster design iterations and performance evaluations. - **Need for design optimization and iteration**: In the automotive design process, it is necessary to quickly evaluate the aerodynamic performance of different design schemes. The rapid prediction capability of RegDGCNN allows designers to quickly identify areas for improvement, thereby optimizing vehicle design to enhance efficiency. In summary, this research aims to accelerate the automotive design process through data-driven methods and promote the development of more efficient and environmentally friendly vehicles by developing DrivAerNet and RegDGCNN.

DrivAerNet: A Parametric Car Dataset for Data-Driven Aerodynamic Design and Graph-Based Drag Prediction

DrivAerNet++: A Large-Scale Multimodal Car Dataset with Computational Fluid Dynamics Simulations and Deep Learning Benchmarks

DrivAerML: High-Fidelity Computational Fluid Dynamics Dataset for Road-Car External Aerodynamics

A systematic dataset generation technique applied to data-driven automotive aerodynamics

Data-Driven Car Drag Prediction With Depth and Normal Renderings

Surrogate Modeling of Car Drag Coefficient with Depth and Normal Renderings

WindsorML: High-Fidelity Computational Fluid Dynamics Dataset For Automotive Aerodynamics

AhmedML: High-Fidelity Computational Fluid Dynamics Dataset for Incompressible, Low-Speed Bluff Body Aerodynamics

Physics-guided Design and Learning of Neural Networks for Predicting Drag Force on Particle Suspensions in Moving Fluids

Drag prediction of rough-wall turbulent flow using data-driven regression

Data-driven modeling of the aerodynamic deformation and drag for a freely moving drop in the sub-critical Weber number regime

Extended Study of Full-Scale Wind Tunnel Test and Simulation Analysis Based on DrivAer Model

Ship-D: Ship Hull Dataset for Design Optimization using Machine Learning

DCC-DNN: A deep neural network model to predict the drag coefficients of spherical and non-spherical particles aided by empirical correlations

Introduction of a New Car Model Series for Aerodynamics Research

Physics-Guided Deep Learning for Drag Force Prediction in Dense Fluid-Particulate Systems

A supervised neural network for drag prediction of arbitrary 2D shapes in low Reynolds number flows

DeepJEB: 3D Deep Learning-based Synthetic Jet Engine Bracket Dataset

Analysis and Development of Novel Data-Driven Drag Models Based on Direct Numerical Simulations of Fluidized Beds

Knowledge discovery with computational fluid dynamics: Supercritical airfoil database and drag divergence prediction