在线数据记录和扰动观测器相结合的柔性机械臂复合学习控制

For the dynamics of multiple input multipe output (MIMO) flexible-link manipulator, this paper investigates a composite learning controller based on the neural networks (NN) and disturbance observer. Firstly, the system is decoupled into the fast and slow subsystems by singular perturbation analysis. Then for the slow-varying dynamics, a novel prediction error is constructed based on the online recorded data scheme. The update law for NN weights is designed by combining the tracking error. A sliding mode controller is constructed to suppress the flexible modes. Furthermore, a disturbance observer is built to estimate the compound disturbance, which is also used as the feedforward compensation of the online recorded data scheme. The boundedness of the system signals is proved via the Lyapunov approach. The simulation test illustrates the effectiveness and superiority of the proposed approach.