On-line self-learning neural network control for pneumatic robot position system

In this paper, a NNI (neural network identifier) is presented to learn model for an articulated multiple DOF (degrees of freedom) pneumatic robot position system. It can adjust the weights and biases of NNC (neural network controller) on line. This controller can effectively solve the difficult problems of single rod cylinders, which are mainly caused by asymmetric structures and different friction characteristics in two directions. Experimental results prove that the dynamic performance of the system can be much improved. The system using NN (neural network) has strong self-adaptability and robustness. It obtains desired percentage overshoot and repeatability in steady-state responses.