Nonlinear Identification of an Unmanned Quadcopter Rotor Dynamics using RBF Neural Networks

The unmodelled rotor dynamics in accelerated flight have a negative effect in the robustness and performance of an unmanned quadcopter, which could result in mission failure in adverse conditions or rotor faults. The nonlinear identification of an unmanned quadcopter rotor dynamics is investigated in this paper. The rotor dynamics are considered in terms of a first-order flapping dynamic model with the dynamics estimated using the radial basis function (RBF) neural networks. A RBF structure based on a continuous forward algorithm (CFA) is implemented for the estimation of a longitudinal rotor flapping dynamic coefficient. This was achieved through optimal input design by the maximization of the spectral density function and predicting the resonant frequency response from the RBF output. This was computed at various trim speeds and training data noise levels and compared with a linear model. The prediction accuracy and robustness to noise of the CFA algorithm proved that the proposed approach can result in better understanding of quadcopter flapping dynamic for high fidelity flight controller design.