Ordered Physics Mamba: A State–space Model for Vehicle Flow Field and Aerodynamic Force Prediction

The rapid and accurate prediction of vehicle flow field distributions and aerodynamic force coefficients is of great importance in aerodynamic design scenarios. The current deep learning-based surrogate models face three major challenges when handling large-scale irregular meshes: low computational efficiency, the absence of physics priors that are aligned with the dominant flow direction, and the tendency to lose high-frequency geometric details on the surface of the target body. Considering the above issues, the ordered physics Mamba (OP-Mamba) flow field prediction model is proposed in this work based on selective state–space sequence modeling. First, OP-Mamba clusters discrete point clouds into a compact set of physics-aware slices through adaptive soft assignment, thereby reducing the imposed computational burden. A spatial sorting mechanism based on the weighted geometric centroids of these slices is then introduced to reorder them along the principal flow axis. Second, combined with this reordering scheme, bidirectional Mamba scanning not only captures the downstream convective evolution process but also provides a direct backward path for incorporating the nonlocal pressure effects that are associated with the downstream wake structures, thereby efficiently extracting long-range dependencies. Finally, OP-Mamba uses a point cloud-domain residual compensation scheme in which feature mappings and skip connections are implemented in the original discrete space, thus mitigating the high-frequency geometric feature attenuation that is caused by slice-level dimensionality reduction. Extensive experiments conducted on three-dimensional (3D) industrial vehicle datasets (including Baidu CIKM and ShapeNet-Car) and the two-dimensional (2D) airfoil dataset (AirfRANS) demonstrate that OP-Mamba achieves state-of-the-art accuracy in both flow field reconstruction and macroscopic aerodynamic force prediction tasks. Additionally, it maintains competitive performance under physical-field noise augmentation during the training procedure and demonstrates its applicability to both 3D vehicle and 2D airfoil settings.

Authors

Institutions

Publication Details

Journal
Journal of Applied Fluid Mechanics
Published
2026-11-12
DOI
https://doi.org/10.47176/jafm.20.1.4628
Primary Topic
Model Reduction and Neural Networks
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Ordered Physics Mamba: A State–space Model for Vehicle Flow Field and Aerodynamic Force Prediction

Y. Fan, F. L. Zhou, H. Y. Chen, Y. J. Zhu et al.
Journal of Applied Fluid Mechanics
Model Reduction and Neural Networks
article

Ordered Physics Mamba: A State–space Model for Vehicle Flow Field and Aerodynamic Force Prediction

Y. Fan, F. L. Zhou, H. Y. Chen, Y. J. Zhu, H. Z. Li, B. D. Tian
article en

Abstract

The rapid and accurate prediction of vehicle flow field distributions and aerodynamic force coefficients is of great importance in aerodynamic design scenarios. The current deep learning-based surrogate models face three major challenges when handling large-scale irregular meshes: low computational efficiency, the absence of physics priors that are aligned with the dominant flow direction, and the tendency to lose high-frequency geometric details on the surface of the target body. Considering the above issues, the ordered physics Mamba (OP-Mamba) flow field prediction model is proposed in this work based on selective state–space sequence modeling. First, OP-Mamba clusters discrete point clouds into a compact set of physics-aware slices through adaptive soft assignment, thereby reducing the imposed computational burden. A spatial sorting mechanism based on the weighted geometric centroids of these slices is then introduced to reorder them along the principal flow axis. Second, combined with this reordering scheme, bidirectional Mamba scanning not only captures the downstream convective evolution process but also provides a direct backward path for incorporating the nonlocal pressure effects that are associated with the downstream wake structures, thereby efficiently extracting long-range dependencies. Finally, OP-Mamba uses a point cloud-domain residual compensation scheme in which feature mappings and skip connections are implemented in the original discrete space, thus mitigating the high-frequency geometric feature attenuation that is caused by slice-level dimensionality reduction. Extensive experiments conducted on three-dimensional (3D) industrial vehicle datasets (including Baidu CIKM and ShapeNet-Car) and the two-dimensional (2D) airfoil dataset (AirfRANS) demonstrate that OP-Mamba achieves state-of-the-art accuracy in both flow field reconstruction and macroscopic aerodynamic force prediction tasks. Additionally, it maintains competitive performance under physical-field noise augmentation during the training procedure and demonstrates its applicability to both 3D vehicle and 2D airfoil settings.

Journal of Applied Fluid MechanicsVol. 20(1)
Southwest University of Science and Technology (CN), Southwest Jiaotong University (CN)
Baidu, Southwest University of Science and Technology, Natural Science Foundation of Sichuan Province
Model Reduction and Neural Networks
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon