The Applications of an On-Line Trained Neural Network in Servo Motor Drives

• Main Authors: Chen ,Hong Pong, 陳鴻鵬
• Other Authors: Lin, Faa-Jeng
• Format: Others
• Language: zh-TW
• Published: 1996
• Online Access: http://ndltd.ncl.edu.tw/handle/98850992388162748477

碩士 === 中原大學 === 電機工程研究所 === 84 === The control performance of the servo motor drive is influenced by the uncertainties of the plant, which usually are composed of unpredictable plant parameter variations, external load disturbance, unmodelled and nonlinear dynamics. In the past decade many modern control theories, such as nonlinear control, optimal control, variable structure system control, adaptive control and robust control, have been developed for the servo motor drives to deal with the uncertainties. Recently much research has been done to apply the neural network to the control field to deal with nonlinearities and uncertainties of the control system. The special learning (on-line) method can overcome the problem in general learning, and if some prior knowledge such as the sensitivity or the Jacobian of the system can be replaced by a simple equation, the connective weights of the neural network are trained during on-line control. That is, the neural network controller can be trained on-line, hence it can obtain good control performance for the plants with time- varying characteristics. In this paper a hybrid controller combining an adaptive neural network controller and an LMFC (linear model following control) or PI position controller is proposed to compensate for the parameter variations and load torque disturbance of the servo motor drive in position control. In the proposed on-line trained adaptive neural network controller, the error between the states of the plant and the reference model is used to train the connective weights of the neural network controller. In the nominal condition the model-following is perfect, and the neural network controller is idle. But when parameter variations or external disturbance occur, an adaptive signal will be generated automatically by the neural network controller to preserve the desired model- following control performance. Finally, the performance of the drive and the effectiveness of the proposed controller are demonstrated by some simulation and