Περίληψη σε άλλη γλώσσα
The reduction of alumina for aluminum production is based on the Hall-Héroult process, an electrolytic method with very high energy requirements. It takes place in electrolytic cells operating at high temperatures, often above 940 °C, where the alumina is first dissolved in molten cryolite and then reduced to aluminum. The precise control and determination of the operating temperature is a fundamental problem, particuyolite and then reduced to aluminum. The precise control and determination of the operating temperature is a fundamental problem, particularly criticalin terms of the energy required. The operating temperature must be higher than the solidification start temperature (liquidus temperature) of the cryolite bath, which is ultimately a mixture of aluminum oxide and fluoride compounds of aluminum, sodium,calcium, etc., but also higher than the superheat temperature, which compensates for the energy losses from the bath to the environment. The liquidus temperature, in turn, depends directly on the chemical composition of the bath. For this purpose, mathematical equations have been developed where the bath's solid ...
The reduction of alumina for aluminum production is based on the Hall-Héroult process, an electrolytic method with very high energy requirements. It takes place in electrolytic cells operating at high temperatures, often above 940 °C, where the alumina is first dissolved in molten cryolite and then reduced to aluminum. The precise control and determination of the operating temperature is a fundamental problem, particularly criticalin terms of the energy required. The operating temperature must be higher than the solidification start temperature (liquidus temperature) of the cryolite bath, which is ultimately a mixture of aluminum oxide and fluoride compounds of aluminum, sodium,calcium, etc., but also higher than the superheat temperature, which compensates for the energy losses from the bath to the environment. The liquidus temperature, in turn, depends directly on the chemical composition of the bath. For this purpose, mathematical equations have been developed where the bath's solidification start temperature is calculated based on the weight percentage of the compounds that make up the bath. However, the application of these equations in industrial systems has shown that in many cases they yield values that deviate from reality, ultimately resulting in the failure to guide the pot operation under optimal conditions. The aim of this specific research is the development of a new, accurate, and reliable mathematical model for calculating the liquidus temperature. The methodology applied for the calculation of this model initially involved the generation of liquidus temperature data for various compositions of the cryolite bath, both through the performance of experimental measurements and tests, and through the application of computational methods, which also served to better understand the structure of these melts. The aforementioned data formed the basis for determining, using machine learning, which- 24 -Κωνσταντίνος Μπέτσης: Διδακτορική Διατριβή parameters significantly influence the change in the bath's solidification start temperature, as well as for quantifying the aforementioned effect. More specifically, this project began with an extensive literature review of research conducted broadly and focused on the calculation of the liquidus temperature and otherproperties of the cryolite bath. Subsequently, using the FactSage software, a thermodynamic study of properties such as the bath's solidification start temperature, density, etc., was conducted, and a new, relatively large database of thermodynamic calculations of the liquidus temperature for many different cryolite bath compositions was created. This database was used as input for training various machine learning models, aiming to select the most suitable one that leads to the smallest calculation deviations and to identify its most important parameters. Following this, experimental measurements of the liquidus temperature of complex cryolite mixtures took place using the Differential Scanning Calorimetry (DSC) technique, creating a second database of experimental liquidus temperature measurements. The validation of the final machine learning model was performed using the experimental measurement database, and the final coefficients of the new model were calculated. The use of the developed machine learning model proved, after performing industrial measurements, to predict the bath's solidification start temperature with greater accuracy, for a wider range of concentrations, compared to existing mathematical equations. Furthermore, to understand the structure of the melts at a molecular level, the research turned to Quantum Molecular Dynamics (QMD) simulations with the CP2K software. Using the double zeta pseudopotential, these simulations revealed the fundamental structural components of the bath, with the AlF₅²⁻ ion emerging as the dominant structure in all systems studied, while the OAl₂F₆²⁻ molecule proved to be the species with the greatest presence in melts containing alumina. Using the Polarized Ion Model (PIM) potential and the metalwalls software, an attempt was made to determine the properties of density, viscosity, and electrical conductivity using Classical Molecular Dynamics. In conclusion, this research effort led to the creation of a robust and highly effective modelfor calculating the liquidus temperature. The combination of advanced simulations, experimental data, and innovative algorithmic techniques offers a highly accurate tool. This algorithm has direct potential for industrial application in primary aluminum production, where it can contribute decisively to process optimization, reduction of the energy footprint, and increase in the overall efficiency of the Hall-Héroult process.
περισσότερα
