Controle de sistemas dinâmicos não lineares com compensação por Processo Gaussiano

Abstract

There are many techniques that may be employed in the design of control systems. PID controllers, for example, are widely used in the industry, but have some disadvantages, including instability issues when the system does not operate under design conditions. Regarding the model-based controllers, the great difficulty is to determine the dynamics of the system and the values of its parameters. In this work, a stochastic compensator is proposed to estimate not only parametric uncertainties, but also the unmodeled dynamics. The compensation scheme is based on Gaussian Process Regression (GPR) and embedded within the control law to enhance tracking performance. On this basis, Feedback Linearization and Sliding Modes are used to develop control laws that accommodate the GPR based compensator. The resulting controllers are applied to an electro-hydraulic actuator (EHA) and evaluated by means of both simulation and experiment procedures. Changes in compensator parameters are also tested in order to assess their effects on controllers’ efficacy. Performance metrics are used for comparison purposes. It can be observed, in all evaluated cases, that the proposed compensation scheme, when combined with the adopted nonlinear controllers, are able of improve trajectory tracking without increasing the overall control effort.