應用適應性類神經網路於機械手臂之追跡控制器設計

Abstract

本論文研究目的為使用類神經網路(Neural Network)估測機械手臂之未知系統參數，並使用適應性控制(Adaptive Control)作為類神經網路之權重值調變，使機械手臂在未知系統參數的情況下完成追跡。 在運動學方面使用D-H(Denavit-Hartenberg)座標系統定義並以此推導出正向運動學，在此定義基礎上使用Pieper’s Solution推導出機械手臂的逆向運動學，藉由順向與逆向運動學求出機械手臂末端點在空間中的三維座標與各軸馬達移動角度之間的關係。 在控制器設計上使用背推(Backstepping)方法設計，將系統分成一個非線性二階系統，設計一個虛擬控制器用以對抗系統未知項，並藉由穩定性分析在保證子系統穩定的狀況下設計該虛擬控制器的形式。對於未知系統參數與系統未知項使用類神經網路進行估測，並藉由適應控制的更新律對類神經網路之權重值做參數調變，藉由Lyapunov 函數與Barbalat引裡證明整個系統的穩定性，最後經由實驗驗證此控制器的性能。In this study, we design an adaptive neural network controller that is applied on tracking control for robot manipulator. The neural network is used to estimate the unknown system parameters and combine the adaptive control which is used to update the weight value of neural network. The kinematics of robot manipulator is defined by D-H coordinate method and derive forward kinematics. Based on D-H coordinate method, the inverse kinematics is derived by Pieper’s solution. By forward kinematics and inverse kinematics, the relationship between motor rotation angle and the end point position coordinate in three-dimensional space of robot manipulator is described. The controller is designed by backstepping method, we divide the system into a nonlinear second-order system, and design the virtual controller which is defined by Lyapunov method to maintain the system stable. For the unknown system parameters and uncertainty, a neural network is used to estimate, and the weight value of neural network is adjusted by adaptive update law. The stability of system is proofed by Lyapunov function and Barbalat’s Lemma. Consequently, the experiment results show performance of this controller in robot manipulator.