DrivAerStar: An Industrial-Grade CFD Dataset for Vehicle Aerodynamic Optimization

TL;DR

DrivAerStar delivers a high-fidelity automotive CFD dataset designed to bridge academic ML research and industrial CFD practice. It provides 12,000 STAR-CCM+ simulations with complete engine compartments and cooling systems, generated via Free Form Deformation across 20 CAD parameters and spanning three rear-body configurations and Reynolds-number ranges. The workflow employs refined meshing with strict wall $y^+$ control, yielding about 12 million cells per case and wind-tunnel–level validation with a mean CD error of $1.04\%$, supported by PIV flow-structure validation. Benchmarks with Transolver, GNOT, and PointNet show that models trained on DrivAerStar achieve production-grade predictive accuracy while substantially reducing computational costs, demonstrating the dataset’s potential to accelerate industrial aerodynamic optimization and beyond.

Abstract

Vehicle aerodynamics optimization has become critical for automotive electrification, where drag reduction directly determines electric vehicle range and energy efficiency. Traditional approaches face an intractable trade-off: computationally expensive Computational Fluid Dynamics (CFD) simulations requiring weeks per design iteration, or simplified models that sacrifice production-grade accuracy. While machine learning offers transformative potential, existing datasets exhibit fundamental limitations -- inadequate mesh resolution, missing vehicle components, and validation errors exceeding 5% -- preventing deployment in industrial workflows. We present DrivAerStar, comprising 12,000 industrial-grade automotive CFD simulations generated using STAR-CCM+${}^\unicode{xAE}$ software. The dataset systematically explores three vehicle configurations through 20 Computer Aided Design (CAD) parameters via Free Form Deformation (FFD) algorithms, including complete engine compartments and cooling systems with realistic internal airflow. DrivAerStar achieves wind tunnel validation accuracy below 1.04% -- a five-fold improvement over existing datasets -- through refined mesh strategies with strict wall $y^+$ control. Benchmarks demonstrate that models trained on this data achieve production-ready accuracy while reducing computational costs from weeks to minutes. This represents the first dataset bridging academic machine learning research and industrial CFD practice, establishing a new standard for data-driven aerodynamic optimization in automotive development. Beyond automotive applications, DrivAerStar demonstrates a paradigm for integrating high-fidelity physics simulations with Artificial Intelligence (AI) across engineering disciplines where computational constraints currently limit innovation.

Figures (17)

• Figure 1: DrivAerStar dataset features. (a) Multiple geometry data formats: mesh (top) serves as primary representation while point cloud (bottom) enables alternative computational approaches and geometry analysis. (b) Complete vehicle modeling includes detailed engine bay compartment assembly (top) and underbody components (bottom), distinguishing DrivAerStar from previous automotive datasets. (c) Internal airflow simulation capabilities featuring air inlet (top) and outlet (bottom) simulations, enabled by precise internal component modeling to support comprehensive automotive system research. (d) Velocity field cross-sections showing internal engine compartment flow at vehicle centerline ($y=0$ m) and external flow field at tire axle plane ($z=0.1$ m).
• Figure 2: DrivAerStar data generation. (a) Three canonical DrivAer reference bodies (Estateback, Notchback, and Fastback) serve as geometric foundations. (b) Parametric morphing systematically varies 20 vehicle components, including greenhouse (top), rear diffuser (middle), and trunk lid (bottom). (c) Industrial-grade mesh generation in STAR-CCM+® produces refined hexahedral-dominant meshes with precise wheel alignment and boundary layer resolution for complex flow capture. (d) Comprehensive cfd simulations generate diverse flow visualizations: surface pressure, wall shear stress, velocity magnitude, streamline patterns, flow separation regions ($C_P=0$ iso-surfaces), and pressure coefficient slices revealing key aerodynamic structures.
• Figure 3: Geometric morphing pipeline example. Our parametric deformation framework transforms the baseline model through three sequential operations: (i) Body component morphing (top-left) applies 15 controlled parameter adjustments, including dimensional modifications and component repositioning; (ii) Wheel morphing (bottom-left) enables precise tire parameter control; and (iii) Whole-body scaling and wheel installation (right) applies 3 scaling parameters while aligning wheels to scaled positions. This systematic ffd approach generates geometrically diverse yet aerodynamically realistic vehicle configurations.
• Figure 4: Cross-sectional views of mesh generation strategy. (a) Longitudinal section at $y=0$ m shows four nested refinement zones with progressive cell density: far-field (120-480 mm), intermediate domain ($\leq$ 60 mm), near-body region ($\leq$ 30 mm), and high-resolution zone ($\leq$ 15 mm) surrounding the vehicle. Dedicated 6-layer boundary layer refinement with total thickness under 24 mm captures viscous phenomena. (b) Transverse section at $x=3.12$ m demonstrates consistent refinement strategy with graduated cell density approaching vehicle surfaces, enabling accurate wake structure resolution while maintaining computational efficiency.
• Figure 5: Validation of flow physics predictions against wind tunnel measurements. (a) Velocity magnitude distributions compare wind tunnel piv measurements (top row) with DrivAerStar predictions (bottom row) in wake regions and stagnation zones, demonstrating excellent flow structure agreement. (b) Surface pressure coefficient ($C_P$) distributions across vehicle body surfaces show consistent prediction accuracy between experimental and computational results for all three vehicle configurations, confirming accurate capture of complex aerodynamic phenomena.