深度强化学习算法在慢走丝机床上的应用研究  

Research on Application of Deep Reinforcement Learning Algorithms in LWEDM

作　　者：谭行 蒋健 魏德骄 TAN Hang;JIANG Jian;WEI De-jiao(School of Chemistry and Chemical Engineering,Beijing Institute of Technology,Beijing 100081,China;Beijing Oriental Technology Co.,Ltd.,Beijing 100088,China)

机构地区：[1]北京理工大学化学与化工学院,北京100081 [2]北京东方嘉科数码科技有限公司,北京100088

出　　处：《自动化与仪表》2019年第4期60-64,共5页Automation & Instrumentation

摘　　要：走丝系统的稳定性是衡量慢走丝线切割机床性能的一项重要指标,电极丝张力变化会引起电极丝不规律振动,从而影响加工工件的精度和表面质量,所以对走丝系统恒张力控制算法进行研究和改进具有很高的实用价值。工业控制普遍使用的是结构简单、稳定性高的经典PID控制算法,但在实际应用中需要对被控对象进行准确建模和精确的参数整定。在被控对象具有滞后、时变等特征时,模型结构和参数会因为扰动和噪声发生不确定性变化,这种情况下传统PID算法的控制效果就会变差。本文利用深度强化学习对环境的自适应能力,设计了一种通过深度强化学习对PID参数进行自适应调整的控制算法,实现了电极丝张力的精确控制,提高了走丝控制系统的鲁棒性和自适应能力。The stability of wire-moving system is an important index to measure the performance of LWEDM,the variation of wire tension would cause irregular vibration of wire,which may affect the accuracy and surface quality of workpiece. Therefore,it is of great practical value to study and improve the constant tension control algorithm of wiredrawing system. The classical PID control algorithm with simple structure and high stability is widely used in industrial control,but in practical application,it is necessary to model the controlled object accurately and tune the parameters accurately. When the controlled object has the characteristics of lag and time varying,the structure and parameters of the model would change uncertainly because of disturbance and noise. In this case,the control effect of the traditional PID algorithm may become worse. In this paper,a control algorithm is designed to adjust the parameters of PID adaptively by deep reinforcement learning(DRL),which makes use of the adaptive ability of DRL to the environment. The precise control of wire tension is realized,and the robustness and adaptive ability of the wire traveling control system are improved.

关 键 词：慢走丝线切割机床 恒张力 自适应PID 深度强化学习 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]