Neural network control of a rotating elastic manipulator

Nonminimum phase property of a rotating elastic manipulator causes difficulties for both classical and neural network inverse model control. While most of the neural network methods for control of elastic manipulators do not appear to converge to a solution when the system is lightly damped, in this paper, an appropriate cost function for a neural controller is proposed. In the designed neural control system, there are only three-layer feedforward networks, consisting of an input layer with two nodes, one hidden layer, and output layer with one node. The number of units in the hidden layer and the value of the learning rate are robust to this designed network algorithm. In order to simulate the transient response of the rotating elastic manipulator system, a single-input, single-output state space representation is presented for the system nonlinear model. It can be seen from the simulation results, the designed neural controller can not only achieve very good tracking performance, zero steady-state errors, and strong robustness to system parameter uncertainty, but also reject the effects of the input torque disturbance.