Particle image velocimetry network embedded with Kolmogorov–Arnold networks for near-wall shear flow measurement
Physics of Fluids · 01.07.2026 · 38 (7)
DOI: 10.1063/5.0332071 ↗Kurz gefasst
Fachartikel in der Zeitschrift Physics of Fluids, online erschienen am 17.07.2026. Verfasst von Hengbo Zhang und 1 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 36 Literaturangaben. Gefördert von: National Natural Science Foundation of China, Basic Research Programs of Taicang, Young Talent Fund of Xian Association for Science and Technology.
Abstract (Originalsprache)
Particle image velocimetry (PIV) is a nonintrusive flow visualization technique that extracts dense flow fields directly from raw particle images. However, due to the dominant viscous actions of fluids, tracer particles are inherently sparse within the immediate near-wall region. This localized information void severely hinders the reliable identification of fluid–solid boundaries. Simultaneously, the near-wall shear layer generates severe, nonlinear velocity gradients, which collectively exacerbate the measurement difficulties in these boundary-adjacent zones. Consequently, conventional PIV methods face severe challenges in accurately resolving near-wall shear flows. To overcome these boundary identification deficits, this study introduces a fusion-informed deep learning model named the all-feature adaptive fusion and Kolmogorov–Arnold network for particle image velocimetry (AFKAN-PIV). This neural network framework incorporates an all-feature adaptive fusion (AAF) module and a U-shaped Kolmogorov–Arnold network (U-KAN). The AAF module fuses static particle and dynamic flow features via spatial and channel weighting to optimize fluid–solid boundary segmentation. Building directly upon these geometric boundary constraints, the U-KAN architecture leverages learnable nonlinear spline functions to precisely fit the steep, high-gradient velocity variations, thereby achieving high-fidelity measurements. Evaluated on synthetic datasets, AFKAN-PIV demonstrates superior accuracy over existing methods, achieving an average reduction of approximately 21.73% in mean squared error. Furthermore, extensive evaluations confirm that AFKAN-PIV offers an excellent balance between structural complexity, computational efficiency, measurement performance, and generalization capability, providing a more reliable solution for near-wall shear flows compared to current measurement methods.
Förderung
- National Natural Science Foundation of China
- Basic Research Programs of Taicang
- Young Talent Fund of Xian Association for Science and Technology
Zitieren
APA
Zhang, H., & Wang, L. (2026). Particle image velocimetry network embedded with Kolmogorov–Arnold networks for near-wall shear flow measurement. Physics of Fluids, 38(7). https://doi.org/10.1063/5.0332071
Vancouver
Zhang H, Wang L. Particle image velocimetry network embedded with Kolmogorov–Arnold networks for near-wall shear flow measurement. Physics of Fluids. 2026;38(7). doi:10.1063/5.0332071
BibTeX
@article{Zhang2026,
author = {Zhang, Hengbo and Wang, Lianpo},
title = {Particle image velocimetry network embedded with Kolmogorov–Arnold networks for near-wall shear flow measurement},
journal = {Physics of Fluids},
year = {2026},
volume = {38},
number = {7},
doi = {10.1063/5.0332071}
}
EUVIMED