Abstract:We present DrivAerNet++, the largest and most comprehensive multimodal dataset for aerodynamic car design. DrivAerNet++ comprises 8,000 diverse car designs modeled with high-fidelity computational fluid dynamics (CFD) simulations. The dataset includes diverse car configurations such as fastback, notchback, and estateback, with different underbody and wheel designs to represent both internal combustion engines and electric vehicles. Each entry in the dataset features detailed 3D meshes, parametric models, aerodynamic coefficients, and extensive flow and surface field data, along with segmented parts for car classification and point cloud data. This dataset supports a wide array of machine learning applications including data-driven design optimization, generative modeling, surrogate model training, CFD simulation acceleration, and geometric classification. With more than 39 TB of publicly available engineering data, DrivAerNet++ fills a significant gap in available resources, providing high-quality, diverse data to enhance model training, promote generalization, and accelerate automotive design processes. Along with rigorous dataset validation, we also provide ML benchmarking results on the task of aerodynamic drag prediction, showcasing the breadth of applications supported by our dataset. This dataset is set to significantly impact automotive design and broader engineering disciplines by fostering innovation and improving the fidelity of aerodynamic evaluations.

What problem does this paper attempt to address?

The main objective of this paper is to introduce the DrivAerNet++ dataset, which is the largest and most comprehensive multimodal automotive design dataset to date, aimed at addressing key challenges in automotive aerodynamic design. Specifically, this dataset attempts to solve the following issues: 1. **Lack of diversity and scale**: Existing datasets are often based on the same parametric models, resulting in high similarity in generated car designs, limiting the model's generalization ability and the space for creative design exploration. Additionally, these datasets are usually small in scale, unable to cover the complex geometric variations in car design. 2. **Low simulation accuracy**: Due to the high cost of high-quality Computational Fluid Dynamics (CFD) simulations, existing datasets have compromised between dataset scale and simulation accuracy, reducing their value for practical applications. 3. **Missing key component modeling**: Many datasets overlook the modeling of important components such as wheels, rearview mirrors, and chassis, which have a significant impact on aerodynamic performance. 4. **Lack of experimental validation**: Some large datasets lack validation through physical experiments such as wind tunnel tests, reducing the reliability and accuracy of the datasets. To address these issues, the DrivAerNet++ dataset has been improved in the following aspects: - **Diverse designs**: It includes 8,000 different car designs, covering various car types (such as fastback, hatchback, and station wagon), as well as different chassis and wheel configurations to represent the design characteristics of internal combustion engine cars and electric vehicles. - **High-fidelity CFD simulations**: Each design is based on high-precision CFD simulations, ensuring the quality and reliability of the data. - **Rich data types**: The dataset includes not only detailed 3D meshes and parametric models but also aerodynamic coefficients, flow field data, partial annotations, and point cloud data. - **Experimental validation**: It provides data from physical experiments such as wind tunnel tests as validation benchmarks, enhancing the credibility of the dataset. - **Large-scale dataset**: It offers more than 39 TB of publicly available engineering data, making it one of the largest automotive design datasets currently available. In summary, the DrivAerNet++ dataset aims to fill the gaps in current datasets in terms of diversity and high precision, supporting the wide application of machine learning in the field of automotive design, including data-driven design optimization, generative AI, CFD simulation acceleration, and more, thereby promoting efficiency and innovation in the design process.

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

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

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

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

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

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

Data-Driven Car Drag Prediction With Depth and Normal Renderings

AircraftVerse: A Large-Scale Multimodal Dataset of Aerial Vehicle Designs

Surrogate Modeling of Car Drag Coefficient with Depth and Normal Renderings

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

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

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

DVM-CAR: A large-scale automotive dataset for visual marketing research and applications

Dataset of a parameterized U-bend flow for Deep Learning Applications

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

Deep Learning for Multi-Fidelity Aerodynamic Distribution Modeling from Experimental and Simulation Data

AFBench: A Large-scale Benchmark for Airfoil Design

Introduction of a New Car Model Series for Aerodynamics Research

Data-driven modeling for unsteady aerodynamics and aeroelasticity

Machine learning-accelerated aerodynamic inverse design