This paper presents a reduced-order modeling framework based on proper orthogonal decomposition, multidimensional interpolation, and machine learning algorithms, along with an error-driven iterative sampling method, to adaptively select an optimal set of snapshots in the context of in-flight icing certification. The methodology is applied, to the best of our knowledge for the first time, to a complete aircraft and to the entire icing certification envelope, providing invaluable additional data to those from icing tunnels or natural flight testing. This systematic methodology is applied to the shape/mass of ice and to the aerodynamics penalties in terms of lift, drag, and pitching moments. The level of accuracy achieved strongly supports the drive to incorporate more computational fluid dynamics information into in-flight icing certification and pilot training programs, leading to increased aviation safety.
- reduced-order modeling framework
- machine learning algorithms
- aviation safety
- natural flight testing
- multidimensional interpolation