Metodologias de sintonia online e ótima para controladores com ações PID baseadas em modelos Neuro-Fuzzy e guiadas por dados de sensores (Data-Driven Ótima)

• Autor(a) principal: MOURA, José Pinheiro de
• Data de Publicação: 2019
• Tipo de documento: Tese
• Idioma: por
• Título da fonte: Biblioteca Digital de Teses e Dissertações da UFMA
• Texto Completo: https://tedebc.ufma.br/jspui/handle/tede/tede/2878
• topic: Controladores lógicos programáveis; Controladores PID; Programação dinâmica adaptativa; Regulador linear quadrático; Sintonia ótima; Programmable logic controllers; PID controllers; Adaptive dynamic programming; Linear quadratic regulator; Optimal tuning; Linguagens de Programação

Optimal online tuning of industrial process controls is an activity that requires the development of dedicated hardware and software for controller parameter setting activity in manufacturing plants that assist in achieving the goals to be achieved in a relatively short time. contributing to the technological evolution of industry 4.0, which lead to an autonomous tuning of industrial processes. In order to support tuning of control systems that follow design approaches such as model-based control (MBC) or datadriven control (DDC), proposed methodologies for online tuning for model-based PID controllers are: Neuro-Fuzzy and guided by sensor data. This Doctoral Thesis presents innovative methodologies for controlling the operational processes of industrial systems with application in mining sector equipment to be embedded in Programmable Logic Controllers (PLC). For better understanding, this Thesis is divided into two parts, where Part I - gives an overview of the industrial process, addressing each of the process phases and a brief introduction to programmable logic controllers. The theoretical framework of proportional, integral and derivative actions (PID), Adaptive Dynamic Programming (ADP) and Computational Intelligence (CI) are presented for the theoretical basis, formulation and problem solving presented in this Thesis. Finally, the proposed methodology is presented. Already in the textbf Part II - we present the Tuning Models Online MBC and DDC of Industrial Processes in two approaches, being: a) Plant model dependent - represented in state spaces and transfer functions, designed with real process data for automatic adjustment of PID controller gains. This adjustment is made as follows: at first, optimal PID controller gains are determined offline through a structured artificial neural network and then a fuzzy rules-based system is used to adjust earnings through a scaling scheme online of this gains vector. In the second moment, the optimal PID controller gains are determined online through a structured feed-forward artificial neural network with low computational complexity training algorithms and high performance RLS algorithms and b) Independent of plant models - taking into account only sensor input/output state signals measured by sensors. Also in this context, an optimal tuning methodology for self-adjusting gain controllers based on Quadratic Linear Regulator (RLQ) (continuous time / discrete time) is presented in Approximate Dynamic Programming (PDA), specifically Action-Dependent Adaptive Heuristic Dynamic Programming (AD-AHDP) has been applied to the experiments presented in this Thesis.