基于神经网络柔性关节机械臂反演滑模控制

Backstepping Sliding Mode Control for Flexible-joint Robotic Manipulators Based on Neural Network

针对模型不确定的柔性关节机械臂系统,提出一种基于神经网络的反演滑模控制方法。与传统刚性机械臂相比,考虑柔性关节机械臂数学模型能更好的实现高精度控制性能。将反演法与滑模控制相结合,将复杂的非线性系统分解成不超过系统阶数的子系统,设计虚拟控制量,并在第一个子系统中引入积分滑模面来改善系统的控制性能。其中,通过sigmoid神经网络实现对未知函数和虚拟控制量导数的逼近来简化控制器的设计。仿真结果证明该方法具有良好的跟踪性能,柔性关节机械臂系统的输出能实现稳定跟踪。

In this paper, a neural backstepping sliding mode control （NBSMC） scheme is proposed for flexible-joint robotic manipulators. To achieve a high tracking control performance, joint flexibility is taken into account in both modeling and control of manipulators. The backstepping technique and the sliding-mode control are combined to design the virtual control input in each step of the backstepping method. The complex nonlinear system is decomposed into subsystems whose order numbers are less than the original system, and the integral sliding mode surface is used in the first subsystem to improve the control performance. A simple sigmoid neural network is employed to approximate the nonlinear and uncertain parts in the control system. Comparative simulation examples are given to illustrate the effectiveness of the proposed method and the outputs of robotic manipulators can fast and accurately track the expected trajectories.