Neural Network-based Direct Model Reference Adaptive Control Of The Servo System Research In A Cannon

This paper proposes a method to improve the speediness, accuracy and stability of certain artillery servo system when it works under high inertia, variable load and mighty impact condition.This paper begins with presenting the constitution of a DC servo system' s hardware and the design of a servo amplifier. Next this paper proposes an off-line learning method to ascertain the network weights and threshold values by incorporating neural network into system identification. The method combines neural network with adaptive model reference to constitute a direct model reference adaptive controller based on neural network. The neural network controller is obtained to simulate the control results of the system model, and modified in real systems. BP algorithm with additional momentum item is applied to neural network identification and control algorithm.A set of simulation and experimental results shows that this controller could meet the required needs with high robustness and stability.