Capturing Low-Wavenumber Near-Wall Structures via Conditional Generative Modeling

Near-wall turbulence plays a critical role in a wide range of fluid dynamic phenomena, including skin friction drag, heat transfer, and aeroacoustic noise, but remains challenging to resolve due to its multiscale structure and computational demands. In particular, low-wavenumber, elongated turbulent structures in the buffer and logarithmic layers play a dominant role in modulating wall shear stress and pressure fluctuations, with direct implications for drag reduction and the aerodynamic design of bluff bodies. Accurately resolving these large-scale motions using conventional CFD approaches requires prohibitively large computational domains, limiting scalability and practical utility.

We present a generative modeling framework for synthesizing near-wall turbulence with a specific focus on low-wavenumber content. The model employs an autoregressive sampling strategy to generate spatiotemporally coherent velocity fields, enabling flexible and efficient synthesis of large-scale turbulent structures. To address the limitations imposed by sparse wall-based measurements, the generative model is embedded within a data assimilation framework that enforces physical consistency with available observations while preserving statistical fidelity to the underlying flow dynamics. The proposed approach demonstrates strong capabilities in reconstructing physically realistic near-wall flow structures under sparse sensing conditions.