Simulación de un controlador para el sistema ball and beam quanser basado en algoritmos de aprendizaje automático

Abstract

This document presents the development of a controller based on machine learning algorithms for the nonlinear ball and beam system, showing how to implement a control system based on machine learning algorithms to nonlinear and unstable systems and that these adjust to established design parameters. The learning control algorithm that is designed in this work is an iterative learning one, for which a classical controller based on the state space model is first designed and is taken as a basis to implement the selected control algorithm. The performance of the controller based on the iterative learning algorithm is compared with the classical controller, evidencing that the iterative learning algorithm generates a steady-state error that, although minimal, gives the classical controller an advantage, since it does not has steady state error. Finally, this work shows that the selected control strategy, despite being simple, is robust enough to be implemented in a complex nonlinear control problem without presenting much difference compared to a classic controller.