2000 年 31 巻 4 号 p. 115-122
The open-loop response of a double-rod linear actuator controlled by an electro-hydraulic servovalve is investigated for unloaded conditions using the identification method, which is conducted by the self-excited oscillation system. According to the identified parameters, the Neural Network velocity controller with a constant gain characteristic is incorporated with the inner feedback line of the position control system. A quasi model reference control approach is proposed for adapting the controller structure to considerable system changes. Thus, the Neural Network is trained under on-line conditions by the simulation. The usefulness of this technique is also experimentally verified with a real-time implementation to the servo-cylinder drive.