Dynamic nonlinear aerodynamics modeling method based on layered model

It is found that many nonlinear aerodynamic models cannot accurately predict linear characteristics under small disturbances. Based on the above limitation, a nonlinear layered model for identifying transonic nonlinear unsteady aerodynamic forces is presented. Layered modeling process needs training samples of both small and large amplitude oscillations. Firstly, the linear model (autoregressive with exogenous input, ARX) is constructed with small amplitude maneuver and the nonlinear model (radial basis function neural network, RBFNN) is constructed with a deviation of a large amplitude maneuver and linear model samples. Then the superposition is done with the outputs of both linear and nonlinear model. Finally the layered, nonlinear dynamic model is obtained. Results show that the layered aerodynamic model has higher numerical accuracy than the autoregressive RBF (AR-RBF) neural network model. The layered model has the ability of predicting large amplitude maneuvers. For small disturbance, layered model is transformed into linear model automatically and can precisely describe the linear dynamic characteristics of small amplitude oscillation.