CMAC neural network for the rudder dynamic load simulator of unmanned aerial vehicles

In order to solve the nonlinearity and the surplus torque disturbance in the rudder load simulator of the unmanned aerial vehicles, a hybrid controller was proposed with the cerebellar model articulation controller(CMAC) network and the traditional proportional-derivative(PD) controller. The CMAC can simulate the nonlinear continuous function with high approximation and high speed learning rate which is suitable for the real time control of nonlinear systems. In the hybrid controller, the CMAC performed the feed forward control and the PD controller realized the feedback control to ensure the accuracy in process of the control. The instability of hybrid controller was discussed and analyzed, the application of credit-assigned and adaptive learning rate CMAC hybrid controller was researched and a simple one-input and one-output CMAC hybrid control method was proposed. The simulation results show the method has settled the unstable problem of the hybrid controller and effectively improved the system dynamics with high robustness.