The paper presents a novel nonlinear reduced-order modeling approach for multi-input/multi-output aerodynamic systems. The nonlinear reduced-order model for an aerodynamic system includes a finite sum of Wiener-type cascade models. The nonlinear reduced-order model approach starts with fitting a Wiener-type cascade path between the inputs and outputs of the aerodynamic system first. Then, the approach computes the outputs of the path and subtracts them from the measured outputs. The second path is then fitted between the inputs and the output residuals. This process is repeated until the residuals contain only noise. To obtain an optimal path at each stage, a novel nonlinear model, a linear dynamic state-space element followed by a single-layer neural network model, is selected as the Wiener-type cascade model. The Wiener-type cascade model can be optimized by using the Levenberg–Marquadt algorithm. To demonstrate the performance of the proposed nonlinear reduced-order model in modeling the statically non...
1 October 2012 in Journal of Fluids and Structures
Nonlinear reduced-order modeling approaches have been a promising tool for predicting the transonic aerodynamic and aeroelastic behaviors of aircraft structures. However, the accuracy of the approaches in predicting the transonic limit-cycle oscillations and aeroservoelastic behaviors needs to be further improved. For example, because of the complexity induced by the coupling among deflection of control surfaces, oscillating shock waves, structural vibration, and aerodynamic viscosity, it is difficult to adjust the system order and number of neurons of the Wiener-based reduced-order model. Moreover, the increasing of the system order may lead to overfitting. This paper presents a novel reduced-order aerodynamic modeling methodology based on the theory of nonlinear state-space identification. The accuracy of the method is demonstrated for predicting aerodynamic loads, limit-cycle oscillations, and frequency responses of aeroservoelastic systems at transonic flow conditions. Results are compared with full-order simulations solving the Reynolds-averaged Navier-Stokes equations.