Monitoramento de processos de separação baseado em reconciliação de dados

Abstract

This work presents 4 case studies and a review, with the objective of developing a methodology to robustly solve the problem of dynamic or stationary data reconciliation, in separation processes for real-time monitoring. Three case studies evaluate the problem of dynamic data reconciliation in distillation columns, also evaluating the optimization methods, M-estimators, reduced model of column and strategies to increase speed and robustness, based on metrics and autoregressive predictive model. The fourth case presents a methodology based on data reconciliation, for monitoring and diagnosing gross errors in real time, for a separation process of CO2 by membranes. Finally, a complete review of robust M-estimators was carried out, complementing the methodology developed for realtime monitoring and simultaneous elimination of gross errors. The deterministic optimizers showed better results. The Welsch and Normal Contaminated estimators showed better performances. The reduced model showed improvements in the optimization speed without losing accuracy. The strategy presented an excellent performance, reducing the noise and gross errors, increasing the robustness and speed. The review highlighted 50 M-estimators (2 not robust and 48 robust), tuning them to Relative E ciency levels of 90, 95, 98 and 99%.