柔性脚踝康复机器人自适应迭代学习控制

Adaptive iterative learning control of flexible ankle rehabilitation robot

为解决气动肌肉驱动的脚踝康复机器人实际控制中,无模型自适应迭代学习控制在系统噪声干扰下或初始拟伪偏导选择不当会导致算法收敛速度过慢、控制效果差的问题,提出一种基于高阶拟伪偏导整定的无模型迭代学习控制方法,并设计基于零化神经网络误差递归的迭代学习控制律.通过引入系统观测数据对初始拟伪偏导进行修正,减少拟伪偏导初始值的选取对于算法收敛速度的影响;通过设计抗噪声零化神经网络控制律,减小系统噪声对控制性能的影响,进而实现噪声环境下柔性康复机器人的高性能轨迹跟踪.仿真实验结果表明在噪声环境下能够利用较少的迭代轮次降低最大跟踪误差.机器人实际控制实验结果表明:该方法能够在7次迭代后使气动肌肉平均跟踪误差控制在2%以内,并且在不同初始拟伪偏导条件下均能获得较好的收敛性和轨迹跟踪性能.

To solve the problem that model-free adaptive iterative learning control(MFAILC)in the practical control of ankle rehabilitation robot driven by pneumatic muscles could lead to slow convergence and poor control performance under system noise interference or improper selection of the initial pseudo-partial derivative(PPD),a model-free iterative learning control method based on high-order PPD tuning was proposed,and the control law of MFAILC based on the zeroing neural network(ZNN)error recursion was designed.The system observation data was introduced to modify the initial PPD and to reduce the influence of the selection of initial PPD on the convergence speed.The noise tolerant ZNN control law was designed to reduce the influence of system noise on control performance,enabling the high-performance trajectory tracking of flexible rehabilitation robot in noise environment.Simulation results show that the maximum tracking error can be reduced by fewer iterations in noise environments.The actual control results of robot show that the average tracking error of pneumatic muscle can be reduced within 2%after 7 iterations,and the superior convergence and trajectory tracking performance can be achieved under different initial PPDs.