High Angle of Attack Aerodynamic Model Identification for Spin Recovery Simulation Using Non-Parametric Smoothing Functions

摘要

High angle of attack dynamic modeling remains one of the biggest challenges in fixed-wing flight dynamics, due to the complexity of the flow dynamics. This paper presents a modeling approach in which the aerodynamic forces and moments are estimated directly from flight and wind tunnel data using non-parametric smoothing functions. First the methodology is described, which consists of using locally weighted linear regression to predict the aerodynamic force and moment coefficients at a given query point, using a set of input-output training data. To reduce the dimensionality of the input space, a step-wise approach is employed relying on cross-validation over the training set to asses the relative accuracy of different sets of candidate input variables. Having selected the input variables, cross-validation is also used to find the optimal value of the smoothing function hyperparameters. The methodology is applied to modeling the spin arrest phase of a low-wing general aviation aircraft for which spin flight data and wind tunnel data is available. Results show low cross-validation error and high correlation between measured and predicted aerodynamic coefficients for the arrest phase on a number of spin maneuvers. The aerodynamic model obtained is then used to simulate the spin arrest trajectory of two spins not used to train the model, with a good match between simulated and measured trajectories.