基于RBF神经网络的谐波传动自适应反演控制研究

Research on Adaptive Back-stepping Control of Harmonic Drive Based on the RBF Neural Network

由于自身结构上的特点,谐波传动系统存在柔性变形、摩擦和外界不确定干扰等非线性因素。传统控制器大多对系统进行了一定程度的简化,或未考虑非线性外界扰动,导致所设计的控制器性能达不到预期效果。为了提高系统精度,建立了考虑系统非线性刚度和非线性摩擦的谐波传动系统动力学模型;基于试验数据,采用最小二乘法对模型进行参数辨识;采用径向基函数(Radial Basis Function,RBF)神经网络在线逼近系统非线性摩擦和外界不确定干扰力矩,并提出了一种基于RBF神经网络的自适应反演控制器;利用Lyapunov稳定性理论,证明了其闭环系统的收敛性。仿真结果表明,与普通Back-stepping控制相比,在受到外界未知干扰后,所提出的RBF神经网络自适应反演控制能有效地逼近系统非线性摩擦和外界未知干扰,其跟踪误差峰-峰值能迅速稳定到0.00082 rad;而Back-stepping控制对外界未知干扰比较敏感,其跟踪误差峰-峰值增大至0.0123 rad左右。所提出的RBF神经网络自适应反演控制能抑制参数动态变化和外界干扰对系统传动精度的影响,提高系统的传动精度。

Due to its own structural characteristics,a harmonic drive system has a wide range of nonlinear factors,such as flexible deformation,friction and external uncertain interference.Most of the traditional controllers simplify the system to a certain extent,or do not consider the nonlinear external disturbance,resulting in that the performance of the designed controller cannot achieve the desired results.In order to improve the accuracy of the system,the dynamic model of the harmonic drive system is established considering the nonlinear stiffness and nonlinear friction of the system.Based on the test data,the parameters of the model are identified by the least square method.Radial basis function(RBF)neural network is used to approximate the nonlinear friction and external uncertain disturbance torque of the system on-line,and an adaptive inversion controller based on RBF neural network is proposed.Using Lyapunov stability theory,the convergence of the closed-loop system is proved.The simulation results show that,compared with the ordinary Back-stepping control,the proposed RBF neural network adaptive inversion control can effectively approach the system nonlinear friction and external unknown disturbance after being subjected to external unknown disturbance,and its peak value of tracking error can be quickly stabilized to 0.00082 rad.The Back-stepping control is sensitive to external unknown interference,and the peak value of its tracking error increases to about 0.0123 rad.The proposed RBF neural network adaptive inversion control can suppress the influence of parameter dynamic changes and external disturbances on the transmission accuracy of the system,and improve the transmission accuracy of the system.