Physics-Informed Reconstruction of Three-Dimensional Aircraft Wake from Multiplane Particle Image Velocimetry

Particle image velocimetry (PIV) provides planar velocity data, limiting access to volumetric flowfields and pressure in full-aircraft wake measurements. This study develops physics-informed neural networks (PINNs) to reconstruct a physically consistent three-dimensional mean velocity and pressure field from sparse multiplane PIV-type data (synthetically generated pseudo-PIV data from delayed detached-eddy simulation (DDES) for controlled validation purposes). The network employs sinusoidal representation network activations with skip connections, which enhance the representation of multiscale wake structures. Under a consistent measurement configuration and training protocol, three turbulence-physics formulations embedded in the PINNs loss are compared: laminar Navier–Stokes, Reynolds-averaged Navier–Stokes (RANS) without closure, and RANS with the Spalart–Allmaras model. Tests using pseudo-PIV planes extracted from a time-averaged DDES solution of Common Research Model High-Lift (CRM-HL) configuration show that laminar Navier–Stokes underperforms in turbulence-dominated wakes, while RANS-based formulations improve accuracy. Among them, RANS without closure yields the lowest reconstruction errors for both velocity and pressure under the present conditions, provided that Reynolds-stress components are available as observational constraints.