关键词: 神经网络控制, 二次规划, 冗余度解析, 机械臂, 性能约束, 时变欠定线性系统, 容错控制

Abstract: Aiming at the problem of redundancy resolution and fault-tolerant trajectory control of robotic manipulators with joint rotational speed constraints,a Zeroing Neural Network (ZNN) control architecture based on Quadratic Programming (QP) embedded with performance constraints was proposed. Firstly,the resolution model of robotic manipulator redundancy with constraints (time-varying underdetermined linear system) was constructed in the velocity layer. Furthermore,a nonlinear reversible mapping was introduced to transform the constrained system state variables into unconstrained variables,and at the same time,a system error form including the joint velocity term and the end position deviation term was constructed,and the prescribed performance constraints (the upper and lower bounds of the end tracking error of robotic manipulator) were embedded into the system error through nonlinear transformation,and then a QP problem model for redundant resolution of robotic manipulator was constructed,together with a ZNN-based QP problem solving architecture proposed. Then,the global stability and convergence of the proposed control architecture were analyzed by combining convex optimization theory and Lyapunov stability theory. Finally,to solve the problem of fault-tolerant trajectory control of KUKA LBR IIWA 14 R820 robotic manipulator,the performance of the proposed control architecture was verified by simulation analysis and physical experiments. The simulation results show that the proposed QP-ZNN solution architecture can drive the trajectory tracking error of the robotic manipulator to converge to 10^-5 m for different forms of expected trajectories,even if there is initial position deviation of the end-effector,and the embedded performance constraints can greatly improve the performance of the proposed control architecture. Compared with classical and varying parameter ZNN in the existing literature,the control accuracy of the proposed QP-ZNN control architecture with embedded performance constraints can be significantly improved. The results of physical experiments further show that the proposed QP-ZNN control architecture with embedded performance constraints can still drive the trajectory tracking error of the end-effector of the manipulator to converge to 10^-4 m for different types of expected trajectories,even if there are multi-joint faults in the robotic manipulator.

Key words: neural network control, quadratic programming, redundancy resolution, robotic manipulators, performance constraints, time-varying underdetermined linear system, fault-tolerant control