Uma metodologia de projeto de controladores híbridos inteligentes com aplicações no controle ativo de vibrações mecânicas

Abstract

This work proposes a design methodology of fuzzy controllers for systems that do not offer any intuitive information to the control designer, in order to provide guidelines to construct the fuzzy conventional controllers. The design methodology involves the application of artificial intelligence techniques, where the fuzzy controllers are obtained by an optimization process that uses genetic algorithms. For this optimization procedure, the knowledge of the system dynamics is required. From the experimental inputs and outputs of the system, an artificial neural network is trained and, in that way, it is possible to model the dynamic behavior of the plant. The rule base, the weights of the rules and the input membership functions are optimized. The goal of this methodology is the control the vibrations of complex dynamic systems, such as vibration control of plates instrumented with piezoelectric sensors and actuators. In this work, the fundamentals of fuzzy control, artificial neural networks and genetic algorithms are presented. The proposed control methodology is evaluated numerically and experimentally on the control of a vibratory system with one degree of freedom, and on a steel cantilever test beam. The first system is controlled by electromagnetic actuators and the beam is controlled by piezoelectric actuators. Many controllers are evaluated in time and frequency domains. For the studied cases, it is concluded that the proposed methodology is efficient and also some considerations are made about the future works based of the presented search.