Assimilation of wall-bounded turbulent flows with physics-informed neural networks based on incomplete and imperfect experimental data

• Applicant: Professor Dr.-Ing. Julien Weiss
• Subject Area: Fluid Mechanics
• Term: since 2026
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 585783452

The proposed project is concerned with the development of Physics-Informed Neural Networks (PINNs) to improve measurements of wall-bounded aerodynamic flows by experimental means. In wind tunnels, most velocity field measurements are currently performed with PIV (Particle Image Velocimetry) or PTV (Particle Tracking Velocimetry). These techniques are now well-established in the experimental aerodynamics community and provide useful velocity databases of turbulent flows. Nevertheless, several drawbacks are well known, like poor resolution and relatively high uncertainty near the wall. To address these issues, we plan to devise a PINN framework capable of assimilating near-wall turbulent flow from typical wind-tunnel measurements with PIV and PTV. This will be achieved through a systematic study addressing key challenges in turbulence modeling, data assimilation, and the handling of unsteady, separated flows. The project will be divided into three phases gradually advancing in complexity, each concerned with specific research questions: (1) choice of the appropriate turbulence models, (2) balancing of multiple loss terms, and (3) treatment of unsteady separated-flow dynamics. In each case, model benchmarking will be performed on both synthetic and real-world experimental measurements. By the end of the project, we expect to have provided the experimental aerodynamics community with a robust method for improved near-wall velocity measurements with modern, laser-based measurement techniques.

• DFG Programme: Research Grants