Nonlinear dynamic inversion with neural networks for the flight control of a low cost tilt rotor UAV

This paper deals with the development of a flight control scheme based on the Nonlinear Dynamic Inversion technique (NDI). Such a scheme has been implemented on the Flight Control System (FCS) of a trirotors prototype model used to simulate tilt rotor air vehicles dynamics during all those phases when aerodynamic surfaces are less effective or not effective at all. An adaptive flight control scheme has been developed based on Radial Basis Function Neural Network (RBFNN) and NDI making both attitude and trajectory tracking control system less sensitive to modelling errors. To demonstrate the effectiveness of the proposed control technique both numerical simulations and experimental tests have been performed. To this end a scaled multi-rotor test bed has been built simulating a tilt rotor unmanned air vehicle (UAV) in hovering and low speed flight conditions. The control system has been implemented on an embedded board based on an ARM Cortex M4 processor and a low cost Inertial Measurement Unit with triaxial MEMS accelerometer, gyroscope and magnetometer sensors. As accuracy of NDI control system can be affected by model uncertainties and sharp transients due to the RBFNN adaptation transients, a model calibration phase is needed. To this end preliminary flight tests have been performed for UAV dynamic model identification and control parameters tuning