Conference paper
Mohamed Elrefaie, Florin Morar, Angela Dai, Faez Ahmed
NeurIPS 2024
NeurIPS 2024
View PDF
Cite
Cite
APA
Click to copy
Elrefaie, M., Morar, F., Dai, A., & Ahmed, F. DrivAerNet++: A Large-Scale Multimodal Car Dataset with Computational Fluid Dynamics Simulations and Deep Learning Benchmarks. In NeurIPS 2024. https://doi.org/10.48550/arXiv.2406.09624
Chicago/Turabian
Click to copy
Elrefaie, Mohamed, Florin Morar, Angela Dai, and Faez Ahmed. “ DrivAerNet++: A Large-Scale Multimodal Car Dataset with Computational Fluid Dynamics Simulations and Deep Learning Benchmarks.” In NeurIPS 2024, n.d.
MLA
Click to copy
Elrefaie, Mohamed, et al. “ DrivAerNet++: A Large-Scale Multimodal Car Dataset with Computational Fluid Dynamics Simulations and Deep Learning Benchmarks.” NeurIPS 2024, doi:10.48550/arXiv.2406.09624.
BibTeX Click to copy
@inproceedings{mohamed-a,
title = { DrivAerNet++: A Large-Scale Multimodal Car Dataset with Computational Fluid Dynamics Simulations and Deep Learning Benchmarks},
journal = {NeurIPS 2024},
doi = {10.48550/arXiv.2406.09624},
author = {Elrefaie, Mohamed and Morar, Florin and Dai, Angela and Ahmed, Faez}
}
In this work, we introduce DrivAerNet++, the largest and most comprehensive multimodal dataset for aerodynamic car design. DrivAerNet++ features 8,000 diverse car designs modeled with high-fidelity computational fluid dynamics (CFD) simulations, covering configurations such as fastback, notchback, and estateback, with varying underbody and wheel designs to represent both internal combustion engines and electric vehicles. Each entry in the dataset includes detailed 3D meshes, parametric models, aerodynamic coefficients, extensive flow and surface field data, as well as segmented parts for car classification and point cloud data. This dataset supports a wide range of machine learning applications, including data-driven design optimization, generative modeling, surrogate model training, CFD simulation acceleration, and geometric classification. With over 39 TB of publicly available engineering data, DrivAerNet++ fills a critical gap in available resources, offering high-quality, diverse data that enhances model training, promotes generalization, and accelerates automotive design processes. We have conducted rigorous dataset validation and provide ML benchmarking results for aerodynamic drag prediction, showcasing the dataset’s broad applicability. This dataset is set to make a significant impact on automotive design and broader engineering disciplines by fostering innovation and improving the fidelity of aerodynamic evaluations.
In order to create the DrivAerNet++ dataset, several geometric parameters with a significant impact on aerodynamics were selected and varied within specific ranges. These parameter ranges were carefully chosen to exclude values that are either difficult to manufacture or not aesthetically pleasing, ensuring both practical applicability and visual appeal in the dataset.
DrivAerNet++ is the first multimodal dataset for aerodynamic car design. It includes:
- Parametric Models: Parametric models with tabular data, allowing extensive exploration of automotive design variations.
- Point Cloud Data: Point cloud data for each car design.
- 3D Car Meshes: Detailed 3D meshes of each car design, suitable for various machine learning applications.
- CFD Simulation Data: High-fidelity CFD simulation data for each car design, including 3D volumetric fields, surface fields, and streamlines.
- Aerodynamic Coefficients: Key aerodynamic metrics such as drag coefficient (Cd), lift coefficient (Cl), and more.
In order to visualize the feature space for different car designs, t-SNE was used to create distinct clusters representing three car design categories: Notchback (green), Estateback (orange), and Fastback (blue). Each data point represents one car design, illustrating the similarity and variation between different designs. This visualization aids designers in linking new designs to existing ones and potentially identifying the most similar designs and their performance using methods like K-Nearest Neighbors (KNN). Using t-SNE to visualize the high-dimensional design space provides an efficient means to explore the vast design space, eliminating the need to compute complex metrics, such as the Chamfer distance, between a new design and all existing designs. This effective clustering demonstrates the feature representation’s capability to capture underlying differences among the designs.
