Análise do Controlador PID Autoajustável Baseado na Rede Neural de Brandt-Lin

Abstract

PID controllers (Proportional, Integral, and Derivative) are widely used in the industry because they have a simple structure. However, these controllers have challenges because the performance goals often conflict with each other, and the plant’s parameters can change during operation.This work analyzes the self-tuning PID (APID) controller. It is based on the Brandt-Lin neural network algorithm com- bined with two-loop compensation. The adaptive PID lets the gains (𝐾𝑝, 𝐾𝑖, 𝐾𝑑) ad- just continuously, looking for optimized performance in both the transient response and the steady-state response.The study aims to use the Simulink environment to simulate the controller’s mathematical model in three types of plants: first-order, second-order, and an unstable system.The analysis was divided into two parts: at the start of the operation and at the end of the controller’s operation, to see how the APID adjusts itself.Using the resulting gains obtained from the simulation, it was possible to find the PID’s transfer function at the two stages of operation. This made it possible to apply the concept of the Root Locus (RL) and the step res- ponse for a complete controller analysis.The analysis showed satisfactory results in all plants, improving the speed of response, eliminating the steady-state error, and stabilizing a plant with a pole in the right half-plane.