Controle preditivo inferencial adaptativo neural aplicado em um reator químico não isotérmico

Abstract

Industrial chemical processes are essential for the production of various products and have a direct impact on our daily lives. The efficiency and safety of these processes are crucial for the global economy, human health, and the environment. Control strategies play a central role in optimizing resource utilization, minimizing waste, and maximizing product quality. However, complexity increases due to the challenge of accurately measuring some critical variables in real-time. In this scenario, virtual sensors can be valuable as reference models, inferential control, and estimators of hard-to-measure variables, utilizing mathematical models and available data to compute these vital pieces of information. In this study, two control strategies were developed for a non-isothermal CSTR (Continuous Stirred-Tank Reactor) that produces propylene glycol. Model Predictive Control (MPC) was applied to complex and nonlinear systems, replacing conventional control strategies. A Multilayer Perceptron Artificial Neural Network (MLP-ANN) served as a virtual sensor to estimate the concentration of propylene glycol. The neural virtual sensor was trained, validated, and tested, producing remarkable results even under disturbance and setpoint changes. Regression analysis demonstrated a robust model fit with high accuracy. Additionally, a heuristic approach was employed to fine-tune the parameters of the predictive controllers. The Adaptive Quadratic Dynamic Matrix Control (AQDMC) controller exhibited superior performance, with reduced overshoot and stabilization time, underscoring its superiority. This showcases the effectiveness of the virtual sensor in concentration estimation and the advantage of the AQDMC controller, particularly in the presence of process disturbances. The study contributes to the application of Artificial Neural Networks (ANNs) in chemical process optimization, emphasizing the importance of virtual sensors and effective control in achieving efficiency and precision in the industry.