基于DDPG的柔性伺服系统级联陷波器设计

Deep Deterministic Policy Gradient-based Design of Cascade Notch Filters for Suppressing Mechanical Resonance in Flexible Servo System

针对柔性伺服系统的多频谐振抑制问题,提出一种基于DDPG的级联陷波器参数整定方法。以系统速度环开环bode图及陷波器bode图预处理结果作为训练数据,并以相位裕度作为奖励函数训练神经网络,实现所设计的伺服系统级联陷波器深度及宽度参数优化训练。搭建了三质量柔性伺服系统实验平台,并开展了多频谐振抑制实验,实验结果表明所提出的参数选择方法能够找到具有最大相位裕度的陷波器参数,并有效地抑制系统多频谐振。

Aiming at suppressing the multi-frequency resonance problem in the flexible servo system,a cascaded notch filter parameter tuning method based on DDPG was proposed in this paper.The neural network to realize the optimized training for depth and width parameters of the designed servo system cascade notch filter were trained with the training data which can obtain by the preprocessing results of the system speed loop open-loop’s bode diagram and the notch filter’s bode diagram.A three-mass flexible servo platform was built to carry out the multi-frequency resonance suppression experiments.The experimental results demonstrated that the proposed parameter selection method can obtain the notch filter parameters with the maximum phase margin and effectively suppress the system's multi-frequency resonance.