Multivariate latent variable modelling odf the pre-baked anode manufacturing process used in aluminum smelting = : Modélisation multivariée par variables latentes du procédé de fabrication des anodes précuites utilisées pour la production d'aluminium primaire
|Advisor:||Duchesne, Carl; Tessier, Jayson|
|Other Title(s):||Modélisation multivariée par variables latentes du procédé de fabrication des anodes précuites utilisées pour la production d'aluminium primaire|
|Abstract:||Aluminum is manufactured by an electrolytic process. The reaction consumes carbon anodes. Anode quality has a great influence on the optimal operation of the reduction process. However, their properties are poorly characterized by weekly averages of anode sample laboratory analyses. The goal of this thesis is to improve quality control at the baked anode manufacturing plant by predicting anode properties. A multivariate latent variable regression method called Projection to Latent Structure (PLS) is used to relate the raw material and the manufacturing process data to the baked anode properties collected at the Alcoa Deschambault smelter. Several models are investigated for physical properties and gas reactivity. From 27% to 68% of the physical properties variance and 20% to 49% of the reactivity variations are captured. The models explained a significant amount of variability, considering that industrial data is typically very noisy. The interpretation of the models was found in agreement with process knowledge.|
|Document Type:||Mémoire de maîtrise|
|Open Access Date:||18 April 2018|
|Collection:||Thèses et mémoires|
All documents in CorpusUL are protected by Copyright Act of Canada.