Comparative analysis of nonlinear data-driven modeling of inlet distortion for nacelle air-intake system

Investigating nacelle intake under crosswind in whole-aircraft models is a frontier topic in modern aerospace propulsion research. However, investigations into the nonlinear flow characteristics of the leeward side nacelle intake influenced by fuselage interference remain scarce, and accurate modeling of this nonlinear flow continues to pose a significant challenge. Although data-driven approaches are cheap and efficient solutions, achieving high-fidelity reconstruction and prediction under small sample conditions remains a major obstacle. This paper investigates the causes of differences in the intake characteristics of windward and leeward side nacelles and evaluates the data-driven modeling performance of representative reduced-order models and deep learning methods, proposing a nonlinear hybrid prediction model framework. The model first extracts the dominant features of the unsteady nacelle intake flow through the proper orthogonal decomposition. The temporal evolution of the modal coefficients is subsequently modeled nonlinearly using neural networks. Finally, global error correction and uncertainty quantification are applied to the predicted flow field. The paper validates the approaches via a numerical example of a typical narrow-body commercial aircraft. The results show that the constructed hybrid prediction model exhibits excellent accuracy and stability across different training set proportions. Given that small sample conditions are a common limitation in aerospace engineering, this approach holds great potential for application in unsteady aerodynamic analysis and rapid evaluation of engine nacelles.