Projeto de controle adaptativo e ótimo aplicado a cadeira de rodas motorizada

Abstract

Many users with impaired mobility are unable to safely operate a motorized wheelchair without harming themselves and/or others due to parametric system variations, such as changing the center of mass of the system or changes in the physical road conditions in which it travels. In order to contribute to attenuate the problem of mobility, this dissertation has as main objective to present development of a methodology for the design of dynamic controllers for motorized wheelchairs. Specifically, two proposals are presented, which are based on proportional, integral and derivative (PID) controller tuning models, with optimality and adaptability properties. These properties are entered into the PID controllers through the machine learning, optimal and adaptive control approaches. The first proposal is related to the optimality property, the controller design is based on s-plane models that are customized to state-space form for linear quadratic regulator design (LQR) application, such models are exclusively developed for PID controllers tuning. Thus, the controller must deal not only with conventional performance metrics, but also must meet its operating costs. The second proposal associates the adaptive properties to the PID controller based on Artificial Neural Networks (ANN) theory, to solve problems related to the parametric system variations, since a conventional PID control algorithm does not respond well to the process disturbances. The tuning proposals are evaluated in mathematical wheelchair model in two situations. In the first, the reference signal adopted is a linear trajectory and the second case is a circular trajectory. The tests present satisfactory results.