Multi-model Predictive Control Of Robot Manipulator

Robot manipulator can help people to complete dangerous, monotonous and boring work, so it is widely used in industrial, military and other fields. But robot manipulator is a highly nonlinear complex system, which is not to be controlled precisely and easily. A multi-model predictive control method is proposed for nonlinear robot manipulator based on membership function. In this thesis, the following aspects are studied:(1) Multi-model predictive control of single-link robot manipulator driven by a DC motor which is a single variable robot manipulator system is studied. Rotation angle of the DC motor is selected as the scheduling variable according to the characteristics of the system; the operation space of the system is divided into two subspaces; linear models are built in each subspace and local predictive controllers are designed based on linear models. Then the local controllers are combined by trapezoidal functions into a global controller, which avoids the chattering during controller switching. Finally, the multi-model predictive controller is compared with a multi-model PID controller through simulation. The simulation results show that comparing with multi-model PID control, the multi-model predictive controller enables the output of system to track reference signal in a wide range with smaller overshoot and higher control accuracy. So the multi-model predictive control is superior to the multi-model PID control.(2) Multi-model predictive control of a Pendubot system which is a multiple variables robot manipulator system is studied. A multi-model predictive control method based on membership function is proposed for balance control based on membership function. Underactuated arm is selected as the scheduling variable; the operation spaces of the systems are divided into five subspaces; local models are built in each subspace and local predictive controllers are designed based on local models. Finally, the local controllers are combined by trapezoidal functions into a global controller to control the system. In order to verify the effectiveness and highlight the advantages of multi-model predictive control, a multi-model PID controller is designed for the Pendubot system. The simulation results show that the multi-model predictive control enables the output of system to track reference signal in a wide range and realizes the control objective of wide range. But the multi-model PID has lower control performance; it can not even guarantee the stability of the system or meet control requirements.