An integrated CFD-based reduced-order model for rapid prediction of in-flight ice accretion and aerodynamic degradation on the NACA 0012 airfoil

Ice accretion on aircraft surfaces can degrade aerodynamic performance and may lead to unsafe flight conditions. Although high-fidelity Computational Fluid Dynamics (CFD) simulations are accurate, their computational cost limits rapid prediction and real-time decision-making. This study introduces a modular Reduced-Order Modeling (ROM) framework for the NACA 0012 airfoil that combines aerodynamic prediction, ice type classification, ice shape generation, and performance loss estimation. An ice-shape redistribution procedure is applied to standardize the ice geometry dataset for consistent ROM training. Separate proper orthogonal decomposition models for rime and glaze ice improve shape prediction, with notable gains for complex glaze formations. Targeted sampling expanded the dataset from 100 to 270 cases, reducing glaze model error by 32.8% and increasing the R² for drag prediction from 0.832 to 0.920 and for moment from 0.833 to 0.948. Rime ice and baseline aerodynamic prediction models achieved consistently low errors across lift, drag, and moment. Validation on four representative cases—rime ice, glaze ice with single and double horns, and a glaze-to-no-ice transition—showed that the ice classifier achieved 97.1% accuracy. Aerodynamic and ice accretion characteristics were also well reproduced. Computational cost analysis indicates that, while the offline model generation phase is resource-intensive, the online prediction stage completes in less than one second per case instead of 240 minutes, providing a speed-up of over three orders of magnitude compared to CFD. These results demonstrate the capability of the proposed framework for parametric studies and highlight its potential usage in preliminary design for assessing icing-induced aerodynamic degradation.