Modeling of longitudinal unsteady aerodynamics at high angle-of-attack based on support vector machines

Abstract

Accurately modeling nonlinear and unsteady aerodynamics at high attitude flight plays an important role in design of future high performance fighters. In the meanwhile, it also can improve the prediction of high angle of attack dynamics of normal aircraft configurations. Support vector machines (SVMs), known as a novel type of learning machines based on statistical learning theory and structural risk minimization (SRM) principle, can be used for handle regression problems. By denoting a set of nonlinear transformations from the complex input space to a high-dimensional feature space, SVMs can approximate the regression function by a linear regression in the feature space. Such implementation is so simple that it can be analyzed mathematically. By employing SVMs, the present work models the unsteady pitching oscillation aerodynamic data of a 1/10 scaled aircraft model. Here, the input data are established from the wind tunnel experiments at different frequencies and amplitudes. To make comparison, the artificial neural networks (ANNs) technique is also used. It turned out that SVMs can overcome the ANNs's inherent drawback of slow training convergence speed. Consequently, SVMs demonstrate high potentials for dealing with the chosen modeling of unsteady aerodynamics.