抄録
In this paper, a self-tuning control method for PI parameters based on the back-propagation algorithm of a multi-layered neural network is applied for position control of a pneumatic motor driving system. A set of mathematical models of the controlled object is derived. Then, the computer simulations (off-line learning) are carried out to determine the learning rate and sigmoid parameters of the neural network. In addition, a method using off-line learning is applied to shorten the time period of learning using an experimental system (on-line learning).
Computer simulations and experiments of the position control were carried out for the change in the reference trajectory or load mass. From the results summarized below, the effectiveness of the self-tuning type neural controller could be confirmed :
(1) Appropriate parameters of PI controller were identified by the applied neural controller, and the steady-state positioning accuracies of with in ±0.5mm could be obtained.
(2) The weights of neural network obtained by the off-line learning were applied to the on-line learning as initial values. The learning time was much shorter than the case using random numbers as the initial values.
