机械臂系统自适应神经网络滑模控制器设计

Adaptive Neural Network-based Sliding Mode Controller Design for Manipulator Systems

针对机器人系统滑模控制器设计存在的抖振问题,提出了一种新型的具有可变滑模增益的控制器设计方案。在传统滑模控制器设计的基础上,该控制方案的创新之处在于所设计控制器的开关增益可实现动态自适应调整。采用径向基函数神经网络(radialbasis function neural network,RBFNN),使开关增益随关节参数动态改变,以适应系统的未建模动态及未知扰动。通过加入适当的自适应控制算法,有效地抑制逼近误差及外部扰动。并且,通过李雅普诺夫方法证明了系统的轨迹跟踪误差可渐近收敛到0。最后,仿真结果表明,所设计的方案降低了系统抖振,同时可有效地提高跟踪精度。

For the chattering problem in the design of sliding mode control(SMC)for robot system,a novel adaptive controller with variable SMC gain is proposed.Based on the design of traditional SMC,the innovation of the proposed control scheme is that the switching gain of the designed controller can be adjusted dynamically and adaptively.Radial basis function neural network(RBFNN)is used to make the switching gain change dynamically with joint parameters to adapt to the unmodeled dynamics and unknown disturbances of the system.By adding an appropriate adaptive algorithm,the approximation error and external disturbance are effectively suppressed.Furthermore,the Lyapunov method is utilized to deduce that the tracking error of the system asymptotically converges to zero.Finally,simulation results show that the proposed scheme can reduce the system chattering and improve the tracking accuracy effectively.