弹性关节空间机械臂级联智能滑模控制

谐波减速器和力矩传感器等柔性元件因其独特性能而广泛应用在空间机器人关节系统中，以获取高减速比．但同时这些柔性元件的存在为空间机械臂系统引入了关节柔性，使得对其稳定控制变得更为复杂．基于此，文中讨论了基于自适应回归小波神经网络(Self-Recurrent Wavelet Neural Networks, SRWNN)的弹性关节空间机械臂系统动力学建模及级联智能滑模控制．首先，利用级联系统理论及第二类拉格朗日方法推导出了由外环刚性臂子系统和内环关节电机转子子系统组成的系统级联动力学模型；其次，为两个子系统分别设计了内、外环自适应滑模回归小波神经网络控制．外环控制算法以期望轨迹为控制量，而其控制信号作为抑制弹性关节振动的内环控制算法的控制量，整个控制系统由内、外环控制系统叠加而成；而后，基于Lyapunov稳定性理论证明了整个控制系统的稳定性并设计了自适应回归小波神经网络的各权值参数在线学习算法．所提的控制算法有效地消除了模型不确定的影响，避免了复杂的求导计算和角加速度可测的要求，同时，控制系统设计过程中未涉及惯常奇异摄动双时标分解操作，在理论上适合任意大小的关节柔性刚度．最后，系统对比仿真试验证明了所提的级联智能控制算法优于惯常基于奇异摄动法和基于柔性铰补偿奇异摄动法的控制方案．

Intelligent Hierarchical Sliding-Mode Control for Flexible-Joint Space Manipulator

Because of their unique properties and good performance, some elastic components, such as the harmonic reducers and the torque sensors, are widely used in the joints of the space robots and manipulators, in order to obtain high reduction ratio. However, these elastic components bring joint flexibility into the system at the same time, which makes its stability control more complex. Based on this, this paper discussed intelligent hierarchical sliding-mode control based on self-recurrent wavelet neural networks and elastic vibration suppression for flexible-joint space manipulator. Firstly, systematic mechanical model composed of outer loop manipulator dynamics model and inner loop joint dynamics model is deduced by the cascade system theory and the second type of Lagrange method; Secondly, the inner loop and the outer loop subsystems are designed for self-recurrent wavelet neural networks sliding mode controller, respectively. The outer loop control algorithm takes the desired trajectory as the control quantity, using the control quantity of the inner loop control algorithm as the control signal for suppressing elastic vibration of flexible joints. The whole control system is superposed by inner and outer loop control algorithms. Then, using the Lyapunov stability theory, the stability of the whole control system is proved, and the online learning algorithm based on the weight parameters of self-recurrent wavelet neural networks is designed. The proposed control algorithm effectively eliminates the influence of model uncertainty and avoids the complex derivative calculation and angular acceleration measurable requirements. At the same time, this control algorithm is theoretically suitable for the flexible joints of any size, since the conventional singularly perturbed double time scale decomposition operation is not involved in the design of control system. Finally, systematic comparison with simulation demonstrates that the proposed control method is better than the traditional ones based on the singular perturbation method or the flexible hinge compensation singular perturbation method.