Projeto de controladores PID multivariáveis baseado em dados frequenciais.

Abstract

The proportional integral derivative (PID) controller is still the most used in industry. Many of the PID design techniques are based on the process model and formulated for systems with one input and one output. The desired performance may not be achieved due to poor controller tuning, which may occur due to modeling errors. This problem can be avoided by using process data and not the controller's parametric process model. Multivariable processes (MIMO) are often found in industry. In these processes, the controller of a closed loop affects the good performance of other loops, due to the interaction between the different loops. Among the ways to solve the problem of interaction between the meshes are the use of a centralized controller or a decentralized controller with a decoupler. In this Thesis, two design and one redesign techniques for the PI/PID MIMO controller and an evaluation and redesign methodology for the inverted decoupler are developed. The design techniques are formulated as a convex optimization problem, whose objective is to minimize the infinite norm of the difference between the designed and desired mesh gain functions. Constraints expressed by LMI are inserted into the problem in order to guarantee the stability of the designed closed loop. The difference between the two techniques is the added constraint. In the redesign technique, the increments of the initial PI/PID controller parameters are calculated so that the new closed loop approaches a reference model. In these techniques, data in the frequency domain are used and the identification of the parametric model of the process is not necessary. Inverted decoupler evaluation is performed by calculating a decoupler error index at a given frequency. If this error is not close to zero, the inverted decoupler parameters are redesigned in order to reduce this index. Simulation examples show the effectiveness of techniques in processes of different orders.