Contribution a la commande et a l’identification de robots a architectures paralleles

Abstract

In this thesis, we have addressed two aspects in relation with mechatronic systems: Identification and control. Indeed, the MLP-NN (MultiLayer Perceptron Neural Network) is used in several approaches proposed in this thesis. Note that the realized work is purely experimental. The two mechatronic systems considered are the C5 links parallel robot and the wearable robot of exoskeleton type available in the LISSI laboratory. In the first time, we have achieved an identification neural black box of the inverse dynamics of C5 parallel robot. To do this, three identification schemes were tested and compared. On the control part, we have proposed an adaptive control hybrid moment / position of C5 parallel robot directly in the task space. The task space dynamic model of the robot in contact with its environment, seen as a black box, is estimated by a MLP-NN. An adaptation algorithm of the neural parameters resulting from a closed-loop stability analysis is proposed. Another approach of control is proposed to derive the exoskeleton. In this design approach, of the adaptive control, the dynamic model is taken as a gray box. Only its structure is known. The unknown functions of the dynamic model are approximated online. The neural parameters adaptation laws are obtained via stability study in Lyapunov sense of the system in closed loop. The proposed approach is tested on a healthy person in flexion / extension of the knee.