HGPTrans: Hierarchical Graph-Pooling Transolver for Automotive Aerodynamic Drag Coefficient Prediction

Bo Liu; Fengli Zhang; Qiuli Luo; Lianrui Nie; Wenjiang Wang

cs.CVSep 24, 2026

HGPTrans: Hierarchical Graph-Pooling Transolver for Automotive Aerodynamic Drag Coefficient Prediction

Authors: Bo Liu, Fengli Zhang, Qiuli Luo, Lianrui Nie, Wenjiang Wang

Organizations: Institute of Auto Engineering, BYD Auto Industry Co., Ltd., Shenzhen, China

Abstract

Accurate and rapid prediction of the aerodynamic drag coefficient ( $C_D$ ) is essential for vehicle design, particularly during early-stage design, where many candidate geometries must be evaluated. Although computational fluid dynamics (CFD) provides reliable aerodynamic estimates, its high computational cost limits large-scale design exploration. This paper proposes the hierarchical graph-pooling Transolver (HGPTrans), which combines hierarchical graph pooling with Transolver-based attention to directly predict $C_D$ from vehicle surface meshes. Motivated by the fact that vehicle aerodynamics depends on both local geometric features and long-range interactions among spatially distant surface regions, HGPTrans integrates three complementary components. Graph isomorphism convolutions encode discriminative local geometry, Transolver-style slice attention captures global interactions with linear computational complexity, and information-redundancy-aware hierarchical pooling progressively removes redundant nodes while preserving informative geometric structures. The model is trained and evaluated on the large-scale DrivAerNet and DrivAerNet++ datasets, where it achieves the lowest mean absolute error and mean squared error among the evaluated baselines. Its generalization capability is further assessed through transfer learning on a real-vehicle dataset containing both sedans and sport utility vehicles (SUVs), achieving relative $L_1$ errors of 1.56% (sedans) and 2.12% (SUVs) with an inference time of approximately $0.293$ s per vehicle. This corresponds to an acceleration of several orders of magnitude relative to high-fidelity CFD while keeping the predicted drag coefficients within a few percent of the CFD reference. Ablation studies confirm each component's contribution and reveal the effects of depth and pooling ratio.

Figures & tables

Figure 1: The overall architecture of HGPTrans. The input vehicle meshes are first normalized, and high-resolution meshes are simplified using QEM to obtain downsampled meshes. The processed mesh is converted into a graph and passed through the stacked HGPTrans blocks, graph-level pooling, and fully connected layers to predict the drag coefficient $C_{D}$ .

Figure 2: Comparison of representative vehicle geometries from the DrivAerNet open-source dataset and the real-vehicle dataset (a production sedan). The real-vehicle geometries contain a substantially higher level of geometric detail.

Figure 3: Computational domain of the full-scale digital wind tunnel.

Dataset	Model	MSE ( $10^{-5}$ ) $\downarrow$	MAE ( $10^{-3}$ ) $\downarrow$	MaxAE ( $10^{-2}$ ) $\downarrow$	$R^{2}$ $\uparrow$
	HGPTrans	2.398	3.718	1.936	0.968
	TripNet	2.602	4.030	1.268	0.972
	FIGConvNet	3.225	4.423	2.134	0.957
	PointNeXt	4.577	5.200	2.410	0.939
	ASSANet	5.433	5.810	2.390	0.927
	MeshGraphNets	6.000	6.080	2.965	0.917

Table 1: Comparison of HGPTrans with mainstream baseline models on the DrivAerNet and DrivAerNet++ datasets. The best value within each dataset block is highlighted in bold. Baseline entries are taken from the cited literature and use the splits reported by the respective sources.

Figure 4: Scatter plots of the predicted versus the ground-truth drag coefficient on the test sets of (a) DrivAerNet and (b) DrivAerNet++. The dashed line indicates the ideal $y=x$ diagonal.

Figure 5: (a) Scatter plot of the predicted versus the ground-truth drag coefficient on the real-vehicle test set; the inset reports the coefficient of determination for the sedan and SUV subsets. (b) Scatter plot of the predicted versus the ground-truth difference $\Delta C_{D}$ for all pairs of variants within the sedan and SUV subsets; blue and red markers denote the pairs for which the model predicts the direction of the change correctly and incorrectly, respectively, and the inset reports the direction accuracy for each subset.

With transfer learning (pre-trained on DrivAerNet++)
Subset	MSE ( $10^{-5}$ ) $\downarrow$	MAE ( $10^{-3}$ ) $\downarrow$	MaxAE ( $10^{-2}$ ) $\downarrow$	Rel. $L_{1}$ ( $\%$ ) $\downarrow$	Rel. $L_{2}$ ( $\%$ ) $\downarrow$	$R^{2}$ $\uparrow$
Total	3.152	4.267	1.826	1.865	2.441	/
Sedan	2.311	3.227	1.826	1.555	2.314	0.770
SUV	3.993	5.306	1.656	2.123	2.524	0.816
Without transfer learning (trained from scratch on the real-vehicle data)
Total	5.549	5.341	3.405	2.335	3.238	/

Table 2: The results of HGPTrans on the real-vehicle dataset, evaluated on the held-out test set and reported for the total test set and the sedan and SUV subsets. The upper block corresponds to the proposed transfer-learning setting (pre-trained on DrivAerNet++); the lower block is a baseline trained from scratch on the real-vehicle data only.

Variant	MSE ( $10^{-5}$ )	MAE ( $10^{-3}$ )	$R^{2}$	Param. ( $10^{6}$ )
Full HGPTrans	2.398	3.718	0.968	2.490
GIN $\rightarrow$ GCN	2.445	3.781	0.967	2.226
GIN $\rightarrow$ MLP	2.635	3.938	0.965	2.490
w/o Transolver	2.481	3.820	0.967	0.628
w/o pooling	2.726	4.021	0.964	2.488
w/o normal features	2.866	4.153	0.962	2.489

Table 3: Architecture ablation study on DrivAerNet. The full model is compared with variants in which the local convolution, the global attention, the pooling, and the input features are modified.

Figure 6: Hyperparameter ablation on DrivAerNet: (a) effect of the number of HGPTrans blocks $L$ and (b) effect of the pooling ratio $r$ on the mean absolute error and the inference time.

Explore similar work

Jun 5, 2026cs.LG

A Geometry-Aware Triplane Field Network for Vehicle Aerodynamic Prediction

High-fidelity computational fluid dynamics (CFD) is crucial to vehicle aerodynamic analysis, but its cost still constrains early-stage design exploration. Machine-learning-based surface-field prediction offers a faster alternative if the model can efficiently capture both global flow context and local geometric detail. This work proposes a machine-learning-based method, named the geometry-aware triplane field network (GTF-Net), for vehicle aerodynamic pressure and wall shear stress prediction. GTF-Net constructs triplane features directly from sampled surface points through a shared multilayer perceptron (MLP) and smooth bilinear rasterization. The planes are then processed by a dual-stream backbone that combines adaptive Fourier neural operator (AFNO) spectral mixing with convolutional neural network (CNN) refinement, so long-range aerodynamic coupling and local geometry-induced variations are modeled in the same representation. At query stage, sampled triplane features are combined with vehicle-aligned directional coordinates, normal-projection features, and a voxel-based curvature proxy. GTF-Net is compared with Transolver, geometry-informed neural operator (GINO), and TripNet, a triplane-based surrogate model. GTF-Net improves the relative L2 error from the strongest baseline value of 0.157 to 0.145 for pressure prediction and from 0.237 to 0.226 for wall shear stress prediction. Ablation results show that AFNO mixing, local CNN refinement, and query-side geometric encoding each contribute to accuracy, supporting the proposed mechanism of combining structured triplane representation with explicit aerodynamic geometry cues.

Kangkang Qi, Huiyu Yang, Keqi Ding +5

Department of Mechanics and Aerospace Engineering, Southern University of Science and Technology, Shenzhen 518055, China · Shenzhen Key Laboratory of Complex Aerospace Flows, Southern University of Science and Technology, Shenzhen 518055, China · Shenzhen Tenfong Technology Co., Ltd., Shenzhen 518000, China +1

Aerodynamics Geometric Deep Learning Surrogate Models

Ask PDF

Apr 20, 2026cs.LG

Faster by Design: Interactive Aerodynamics via Neural Surrogates Trained on Expert-Validated CFD

Computational Fluid Dynamics (CFD) is central to race-car aerodynamic development, yet its cost -- tens of thousands of core-hours per high-fidelity evaluation -- severely limits the design space exploration feasible within realistic budgets. AI-based surrogate models promise to alleviate this bottleneck, but progress has been constrained by the limited complexity of public datasets, which are dominated by smoothed passenger-car shapes that fail to exercise surrogates on the thin, complex, highly loaded components governing motorsport performance. This work presents three primary contributions. First, we introduce a high-fidelity RANS dataset built on a parametric LMP2-class CAD model and spanning six operating conditions (map points) covering straight-line and cornering regimes, generated and validated by aerodynamics experts at Dallara to preserve features relevant to industrial motorsport. Second, we present the Gauge-Invariant Spectral Transformer (GIST), a graph-based neural operator whose spectral embeddings encode mesh connectivity to enhance predictions on tightly packed, complex geometries. GIST guarantees discretization invariance and scales linearly with mesh size, achieving state-of-the-art accuracy on both public benchmarks and the proposed race-car dataset. Third, we demonstrate that GIST achieves a level of predictive accuracy suitable for early-stage aerodynamic design, providing a first validation of the concept of interactive design-space exploration -- where engineers query a surrogate in place of the CFD solver -- within industrial motorsport workflows.

Nicholas Thumiger, Andrea Bartezzaghi, Mattia Rigotti +5

IBM Research · Dallara Automobili

Computational Fluid Dynamic Aerodynamics Surrogate Models Graph Edge Sparsification

Ask PDF

Apr 30, 2026eess.IV

RETO: A Rotary-Enhanced Transformer Operator for High-Fidelity Prediction of Automotive Aerodynamics

Rapid aerodynamic evaluation is crucial for modern vehicle design, yet existing neural operators struggle to capture intricate spatial correlations. We propose the rotary-enhanced transformer operator (RETO), a novel neural solver featuring a dual-stage spatial awareness mechanism: sinusoidal-cosine encodings for global referencing and rotary positional encodings (RoPE) for relative displacements. RoPE encodes spatial relations via unitary rotations, enforcing translation invariance and enhancing local gradient resolution. RETO is validated on ShapeNet and the high-fidelity DrivAerML benchmark. On ShapeNet, RETO achieves a relative

L_2

error of 0.063, outperforming RegDGCNN at 0.125 and representing a 16% improvement over the Transolver baseline, which yields an error of 0.075. These performance gains are further amplified on the DrivAerML dataset, where RETO achieves relative

L_2

errors of 0.089 for surface pressure and 0.097 for velocity. In comparison, Transolver results in errors of 0.116 and 0.121 for the same metrics, indicating that RETO achieves precision enhancements of 23% and 19%, respectively. For comprehensive comparison, the surface pressure and velocity errors for AB-UBT are 0.102 and 0.124, while RegDGCNN yields 0.235 and 0.312, respectively. Information-theoretical analysis shows that the entropy peak of RETO at 0.35 is significantly lower than that of Transolver at 0.75 under

10^4

resolution, indicating a focused attentional mechanism capable of preserving localized gradients against global diffusion.

Bojun Zhang, Huiyu Yang, Yunpeng Wang +4

Department of Mechanics and Aerospace Engineering, Southern University of Science and Technology, Shenzhen, 518055, China · Shenzhen Key Laboratory of Complex Aerospace Flows, Southern University of Science and Technology,Shenzhen 518055, China · Ningbo Key Laboratory of Advanced Manufacturing Simulation, Eastern Institute of Technology, Ningbo, 315200, China +1

Aerodynamics Neural Solvers

Ask PDF