Abstract
This thesis is concerned to the identification and control of a parallel machine (H4 robot from LIRMM). Parallel robots have a high dynamic that implies control laws with capacity to include this dynamic. These laws are based on a dynamic model of the robot and their dynamic parameters must first be identified. Dynamic parameter identification is made by using two methods: weighted least square estimation and bounded error context estimation that yields a garanteed solution set. Bounded error context is developped with ellipsoidal estimation and interval estimation. A comparative parameter estimation of the 14 parameters of the robot with the three methods is presented. The identified parameters are used for the synthesis of a predictive functional control based on a linear model. An original methodology is proposed: first, the identification of the dynamic model and linearisation of the system, then the identification of the internal model for the predictive control and synthesis of the control law with an inner velocity closed loop for stabilize the linear process. Predictive control is compared to two classical strategies used in robotics: PID and computed torque control (CTC). Experimental tracking performance on simples and complex trajectories are tested. Predictive law shows better results on dynamic, tracking error and robustness.