Field Inversion and Physics-Driven-Classification Modeling for Correcting Multimechanism Separated Flows

This study addresses the broader challenge of correcting Reynolds-averaged Navier–Stokes (RANS) turbulence models for flows with multiple competing mechanisms, such as those over iced airfoils. In such scenarios, conventional RANS models often overpredict eddy viscosity where turbulence is underdeveloped and underpredict it in regions of strong turbulent fluctuations. Traditional data-driven approaches, which typically target a single flow mechanism, fail to resolve this dual-error behavior. To overcome this limitation, the authors propose a field inversion and physics-driven-classification modeling (FI-PDCM) framework, whose core innovation is a physics-driven classification strategy that analyzes model discrepancies and identifies salient local features governing error modes. Based on these features, highly portable and interpretable explicit symbolic expressions are constructed as zonal correction models. The corrected Spalart–Allmaras model developed using FI-PDCM exhibits markedly improved predictions of lift, drag, and stall behavior for a variety of two-dimensional iced airfoil configurations as well as enhanced separation prediction for three-dimensional iced wing flows. The framework’s generalization capability is further validated on clean airfoil trailing-edge separation. Crucially, the corrected model retains the baseline model’s accuracy in attached flows, addressing a common shortcoming of many data-driven turbulence corrections. In summary, the FI-PDCM framework establishes a rigorous, physics-driven methodology for developing generalizable, data-augmented turbulence models that can effectively capture the complex interplay of multiple flow mechanisms.