Adaptive feedback linearization flight control for a helicopter UAV

A longitudinal, nonlinear helicopter model is linearized using feedback linearization (FBL) allowing the use of only one linear controller (LC) to steer the nonlinear system over the entire envelope. The FBL controller uses the equations of motion to obtain the helicopter inputs, given the helicopter outputs. This paper uses an accurate reference helicopter model for simulation and a simplified model for the linearization process. This is done to decrease the design time of this controller and reduce the complexity. Not all nonlinearities can be cancelled in this way, resulting in a partly linearized system. The performance of the linear controller has decreased compared with the perfect linearization, but these errors can be compensated by neural networks (NN). These NN have the power to learn to control a system or to provide the missing control input, increasing the performance of the total flight controller. The presence of both the FBL and the LC, decreases the complexity of the NN compared to complete neural control without any aid. By applying this technique, the complexity of both the FBL and the NN are reduced. The error back propagation update law resulted in an unstable controller, but a second method, the emodification learning law showed a good learning behaviour. Several manoeuvres were simulated and the performance with the NN is better than without. This validates the use of a simpler FBL together with NN.