Spatially Consistent and Phase-Aware Reduced-Order Modeling for Dynamic-Mesh Unsteady Flow Simulation

Dynamic-mesh simulations generate unsteady flow snapshots whose sampling points vary with time, complicating reduced-order modeling because snapshot entries no longer represent the same spatial degrees of freedom. This study develops a spatially consistent and phase-aware POD–RBF reduced-order modeling workflow for fast simulation of periodic dynamic-mesh flow fields. The dynamic-mesh snapshots are first mapped onto a common reference grid by KDTree-based nearest-neighbor search and inverse-distance weighting. The aligned snapshots are then used to examine POD truncation, RBF coefficient regression, external-parameter representation, and phase-localized subspace modeling under single-cycle and cross-cycle prediction settings. Benchmark tests on lid-driven cavity and cylinder wake flows verify the baseline implementation, while the oscillating-airfoil case assesses the complete workflow. The results show that spatial consistency is essential for reliable POD modeling of dynamic-mesh data. For the oscillating-airfoil case, the selected POD truncation is mainly limited by projection error rather than RBF coefficient-regression error. The AoA-only input is insufficient to identify periodic states, whereas combined AoA–phase and normalized-time inputs provide stable representations. Phase-localized subspace modeling further reduces low-AoA mean RMSE by 37.42% and 58.63% in single-cycle and cross-cycle prediction, respectively. Additional threshold, kernel-parameter, and later-cycle tests delineate the sensitivity and same-condition scope of the workflow. The results concern phase-conditioned velocity-field reconstruction; aerodynamic-load accuracy and end-to-end CFD speedup are not established.