Blade-Tip Noise Characterization Using Physics-Informed Neural Networks and Sparse Velocimetry

Mitigating flow-induced noise from airfoils and blades is critical in industrial applications. However, resolving the near-field pressure fluctuations that underpin flow-induced noise remains challenging: experimental measurements are often intrusive or limited to far-field data, whereas high-fidelity numerical simulations are computationally prohibitive. To address this, we develop a Reynolds-averaged Navier–Stokes (RANS) [Formula: see text] augmented physics-informed neural network (PINN) framework to reconstruct full-field velocity and pressure distributions from sparse, noisy particle image velocimetry snapshots of a blade-tip vortex. By leveraging the governing physical equations as regularizers, the PINN effectively filters measurement noise, yielding physically consistent reconstructed flow and pressure fields. Subsequently, pressure fluctuations and the power spectral density are derived from the reconstructed fields to characterize the near-field flow-induced noise sources associated with the tip leakage flow for a representative blade-tip leakage condition. Comparisons with high-fidelity large-eddy simulation data show that this RANS-augmented PINN framework provides a nonintrusive approach for near-field noise source characterization in complex flows.