Neural network control of nonlinear objects

• Published in: 2016 17th International Carpathian Control Conference (ICCC) pp. 517 - 522
• Main Authors: Nawrocka, Agata, Nawrocki, Marcin, Kot, Andrzej
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.05.2016
• Subjects: adaptive control; artificial neural networks; Decision support systems; Linearity; Manipulators; mathematical model; nonlinear object; rehabilitation robot manipulator
• DOI: 10.1109/CarpathianCC.2016.7501152

The subject of the conducted research was to develop an advanced algorithm for controlling a rehabilitation manipulator. The concept is a difficult one considering the tasks given to such systems. Seeing as the manipulator is a prototype devised for rehabilitation purposes, the correct functioning of the control system is required for varying load of the robot arm, that is it has been assumed that the manipulator will be used for people with various physical build and a varying degree of physical limitations. Due to the nonlinear character of the object, serious difficulty arises when it comes to devising a satisfactory control system based on traditional solutions. The above-mentioned problems suggest employing adaptive control. In the present case, the concept was extended by using artificial neural networks in the adaptive system of parameters of the manipulator's mathematical model. In the end a system of controllers was arrived at, which thanks to the features of artificial neural networks, and in particular thanks to the possibility of approximating nonlinear functions, carries out the set trajectory in a correct way. The paper contains the numerical solution of the presented problem.