Aerodynamic Parameter Estimation Using Two-Stage Radial Basis Function Neural Network

In this paper, we present the application of radial basis function (RBF) neural network for aerodynamic parameter estimation. The Two-Stage RBF neural network (NN) architecture is proposed for complete aerodynamic modelling. The RBF NN is trained by hybrid learning using K-means clustering and recursive least squares (RLS) methods. The proposed architecture yields the significant improvements for model parameter estimation as compared to other neural network based techniques. The results of proposed method are compared with delta method, partial derivative method and equation error method. This approach reduces the post training efforts for finding stability and control derivatives by running a 2nd stage of Two-Stage RBF NN. The significant improvements in mean square error (MSE), relative standard deviation (RSD) and convergence rate are noticed. The estimated derivatives are validated with reconstructed lateral-directional force and moment coefficients.