Personal tools



Jump to: navigation, search

PhD thesis

Title Estratégias avançadas de modelização e controlo para processos industriais não lineares e descontínuos. Aplicação a cristalizadores industriais de açúcar
Author Luis Alberto Paz Suárez
School FEUP
Month June
Year 2010
Advisor Sebastião Feyo de Azevedo, Petia Georgieva
Group (before 2015) Signal Processing Laboratory

Abstract. The main objective of this PhD thesis was to study and develop efficient and robust non linear predictive control strategies (NMPC) for batch chemical engineering processes, with application in the industrial sugar crystallization processes of three companies: Refinery “RAR.SA”, Refinery “José Martí” and Sugar Company “30 de Noviembre” The work is based on long standing experience and previous studies developed in the Department of Chemical Engineering, Faculty of Engineering of the University of Porto and in the Department of Telecommunications and Electronics of the University of Pinar del Río, related with modeling and control of sugar crystallization units. The control strategies implemented in the three sugar companies involved in the present project do not consider explicitly the final crystal quality and this has been identified as a great limitation for the improvement of the process performance. Therefore, the goal of the present work was to study the possibility to control on-line over the batch duration not only the operational measurable variables, but also the evolution of the final process quality measures. The work can be divided into the following stages: i) analysis of the three industrial processes, with definition of operating strategies and acquisition of real process data; ii) development of a detailed sugar crystallization model, including the main crystal size distribution parameters, namely the mass average crystal size and the coefficient of variation; iii) development of various Nonlinear Model based Predictive Control (NMPC) strategies, iv) validation of the NMPC strategies and v) global analysis of the results and conclusions. Three NMPC strategies were developed. The first control algorithm (DNMPC – Digital NMPC) uses a discrete model to predict the process future behavior. The second control algorithm (NNMPC – Neural Network NMPC) uses a computational neural network based predictive model obtained in two ways: i) from a detailed mechanistic model or ii) from real industrial batch data. In the third algorithm (Batch NNMPC), which is the main result of the thesis, the controller explicitly takes into account the evolution of the mass averaged crystal size parameter over the final crystallization stage in order to reach at the batch end the reference value. The main challenge for real world implementation of the MPC control paradigm is the high computational costs and related numerical problems due to the optimization performed at each iteration step. A modification of the optimization procedure is proposed in this work where a certain tolerance between the value of the controlled variable and its respective reference is admitted (Error Tolerant NMPC). By this alteration, considerable reduction of the optimization time was registered. The NMPC controllers developed in this work, particularly the Batch NNMPC, outperform the classical solutions and represent attractive alternatives for industrial operation.