Vous êtes ici : GIPSA-lab > Formation > Thèses soutenues
DA SILVA MOREIRA Lucas Jos

Multivariable modeling and control for aluminum production

 

Directeur de thèse :     Gildas BESANÇON

Co-encadrant :     Francesco FERRANTE     Mirko FIACCHINI

École doctorale : Electronique, electrotechnique, automatique, traitement du signal (EEATS)

Spécialité : Automatique et productique

Structure de rattachement : CNRS

Établissement d'origine : Université de Campinas (Brésil)

Financement(s) : Contrat doctoral

 

Date d'entrée en thèse : 27/11/2018

Date de soutenance : 23/03/2022

 

Composition du jury :

Monsieur Gildas BESANCON PROFESSEUR DES UNIVERSITES, Grenoble INP, Directeur de thèse
Monsieur Hervé ROUSTAN INGENIEUR DOCTEUR, RIO TINTO, Examinateur
Monsieur Christian JALLUT PROFESSEUR DES UNIVERSITES, UNIVERSITE DE LYON, Examinateur
Madame Isabelle QUEINNEC DIRECTEUR DE RECHERCHE, CNRS DELEGATION ALPES, Rapporteure
Monsieur José Luis GUZMAN PROFESSEUR, Universidad de Almería, Rapporteur
Monsieur Mario Cesar MELLO MASSA DE CAMPOS INGENIEUR DOCTEUR, Smartautomation, Examinateur
Monsieur John Jairo MARTINEZ MOLINA PROFESSEUR DES UNIVERSITES, Grenoble INP, Examinateur

 

Résumé :
The main objectives of this thesis are to propose, design, and test multivariable feedback control laws to optimally regulate the anode-cathode relative distance (ACD) in the Hall-Héroult aluminum production process. Aluminum manufacturing is a challenging industrial area, mostly based on alumina electrolysis, a process that is highly demanding in electrical power. In particular, alumina electrolysis typically relies on a series of large electrolytic cells. Such cells are composed by a bath containing alumina in which a set of carbon anodes is dipped. The base of each cell plays the role of the cathodic electrode in the electrolysis process. In this process, the efficiency and energy consumption depend directly on the anode-cathode relative distance. However, such a distance cannot be measured due the hazardous conditions inside the cells. In addition, the ACD is not constant during the operation being frequently affected by various exogenous phenomena. Another important process state as the dissolved alumina concentration is not measured either. The alumina is required to be in a specific range to avoid the anode effects and sludge formation. Moreover, the concentration can vary along the cell. Therefore, the limited available information hampers an optimal regulation by restricting the efficiency of production. Therefore, an understanding of the chemical process is necessary to relate the measured signals to the ACD and alumina concentration dynamics. This can enhance the process efficiency, reduce the energy consumption, and increase process safety. To achieve these objectives, two approaches are considered in this thesis: average and distributed. The first one aims to control the ACD overall reference value while the second solution intends to optimally reduce the dispersion along the cell. Each of them uses the respective available signals for modeling and control purposes. The average approach aims to model the process using physical and experimental relations. From the model structure, it is possible to use a linear observer strategy to obtain real-time estimations for ACD and alumina concentration. This enables to monitor the process and to develop of an observer-based controller to enhance the production by taking into account the process limitations. This regulation is compared with a control strategy available on the plant. The average approach does not consider local variations nor the influence of daily events. A distributed approach is then performed to study the system conditions along the cell. Individual signals are used to identify the respective models for alumina concentration and anode current distribution. The physical convection-diffusion transport relations are used to obtain a spatio-temporal alumina concentration distribution model. Moreover, the anode current distribution is identified from the data collected in two different test campaigns. Later, two model predictive controllers (MPC) are designed to reduce the current dispersion along the cell using the experimental model. The proposed regulations are based on two configurations available, PIANO and PIEZO, that enable to consider different actuators selection, saturation and deadzone aspects. Both controllers are compared and discussed via numerical simulations. The proposed controllers minimize the anode electric current distribution and are able to indirectly obtain a uniform ACD along the cell. Moreover, the ACD reference value is chosen to optimize the production and energy consumption. All data were collected from APXe pot cell of Rio Tinto Laboratoire des Recherches de Fabrications (LRF) located in Saint Jean de Maurienne, France.


GIPSA-lab, 11 rue des Mathématiques, Grenoble Campus BP46, F-38402 SAINT MARTIN D'HERES CEDEX - 33 (0)4 76 82 71 31