Physics-informed PIV flow field reconstruction during boiling regime transitions in quenching of thin-walled annular steel components

Boiling-induced flow, particularly vapor-film detachment-driven flow, is central to many thermal engineering processes, yet the liquid-phase flow evolution during transitions among film boiling, nucleate boiling, and natural convection remains insufficiently resolved, particularly in confined geometries. In this study, vertical quenching of a thin-walled annular GCr15 steel specimen was investigated using time-resolved particle image velocimetry (PIV), synchronized cooling-curve analysis, and direct visual observation of the boiling state. The annular configuration permits vapor generation on both the inner and outer surfaces and therefore produces strongly unsteady liquid motion representative of practical sleeve-quenching conditions. The boiling regimes were assigned from combined thermal, visual, and hydrodynamic evidence rather than from a single indicator. Conventional PIV measurements were degraded by bubble scattering, refractive-index gradients, intermittent tracer occlusion, and correlation failure, resulting in noisy and spatially discontinuous velocity fields. A physics-informed reconstruction framework was therefore developed by combining denoised PIV data with the incompressible Navier–Stokes equations through a physics-informed neural network (PINN). The framework reconstructs continuous velocity and pressure fields while enforcing mass and momentum conservation. The combined experimental and reconstructed data were assessed using repeated-test consistency, velocity-profile smoothness, divergence, and momentum residuals. The reconstructed fields reveal regime-dependent liquid-phase structures: film boiling is dominated by coherent buoyancy-driven circulation beneath the vapor layer, nucleate boiling exhibits localized jet-like motions and intense near-wall intermittency, and natural convection develops smoother large-scale circulation after bubble activity weakens. The method should be interpreted as physics-constrained data assimilation for the measured liquid phase, rather than as a complete two-phase simulation. These results provide a more physically consistent basis for interpreting experimentally observed boiling-regime transitions in annular quenching