| Aluminum is a widely used metal in the world,and the Hall-Héroult process is an important method for its industrial large-scale electrolysis.To reduce energy consumption and improve efficiency,the NaF-AlF3 molten salt can be added to the electrolyte to lower the melting point of Al2O3 and increase the conductivity of the system.The microstructure and composition of the NaF-AlF3 and its effects are fundamental issues of the electrolysis process.However,the study of the NaF-AlF3molten salt is limited by the test conditions such as strong corrosiveness,volatility and high-temperature environment,and usually requires methods such as high-temperature Raman spectroscopy and nuclear magnetic resonance combined with theoretical calculations to determine the type of microstructure.The classical molecular dynamics simulations can be employed to obtain the microstructure distribution and its transformation pattern,but the transferability of the current general force field of molten salt is bad.Hence,it is urgent to explore some new methodologies to expand the research.This thesis introduced a machine learning method to accurately describe the interatomic interaction forces of NaF-AlF3 molten salt at the molecular mechanics level with quantum mechanical accuracy,and the effects of the composition and temperature of the NaF-AlF3 molten salt system on the distribution of the microstructure and its change pattern were investigated using this method.The main results of the research are as follows:Firstly,the density functional theory was employed to perform simulations of first-principles molecular dynamics on molten salt of the conventional Na-Al-F system to acquire data set for the machine learning.The potential function of Dee PMD was established and validated by the machine learning method.The predicted values of density and particle distribution by the deep learning potential are in line with those derived from the first-principles high-precision calculations and experiments,which confirmed the reliability and generality of the potential model.Secondly,the molecular dynamics studies on the structure and properties of(NaF)1-x(AlF3)x molten salts with different AlF3 mole fractions were carried out,and the results show that each molten salt system contains three types of ions,i.e.Na+,Al-F ion groups and F-.Among them,Na+keep in a free state,and the ions of Al3+are completely in the form of complex ion groups,mainly[AlF4]-,[AlF5]2-and[AlF6]3-monomers.The F-ions exist in various forms,most of which are free fluorine and coordinated terminal fluorine.When x<0.40,the five-coordinated[AlF5]2-ionic groups dominate in the system.When x?0.40,the tetra-coordinated[AlF4]-ion group dominates,while some new fluorine bridge structures appear,such as the dimeric[Al2F7]-,[Al2F8]2-,[Al2F9]3-and trimeric[Al3F10]-ion groups.As the mole fraction of AlF3 increases,the density of the molten salt decreases and the large-volume of[AlF6]3-ion groups gradually transform into the small-volume of[AlF4]-ionic groups.The self-diffusion coefficients of ions under the same AlF3 mole fraction all follow the law ofNa+>F->7)3+.Besides,the migration of Na+plays a major role in the current.Finally,the molecular dynamics investigations on the structure and properties of the cryolite Na3AlF6 molten salt at different temperatures were conducted,and the results demonstrate that the composition of the Na3AlF6 molten salt at every temperature is five ions of Na+,[AlF4]-,[AlF5]2-,[AlF6]3-and F-.In the Al-F complex ion groups,the highest proportion of five-coordinated[AlF5]2-ions was further confirmed by the comparison of the vibrational density of state diagram with the experiment.As the temperature increases,the density of the molten salt decreases,and the short-range order between all ion pairs decreases slightly.Besides,the coordination number of Al-F ion pairs decreases.Among them,the number of[AlF6]3-in octahedral configuration decreases,while the number of[AlF4]-in tetrahedral configuration increases.The number of[AlF5]2-in trigonal bipyramidal configuration varies slightly,and the mean square displacement of all ions increases significantly.The relationship of the self-diffusion coefficients of ions at the same temperature followsNa+>F->7)3+.In summary,we established a deep learning potential for the Na-Al-F system and performed deep potential molecular dynamics studies on the structure and properties of different NaF-AlF3 molten salt systems,focusing on the analysis of microstructure and composition of the Al-F complex ion groups and their change patterns.It was found that the chemical descriptors of the method can deal with the characteristics of molten salts such as multi-cluster structure and environmental sensitivity.Therefore,this work not only provides a framework for using deep potential molecular dynamics simulation in molten salt systems,but also exhibits an important reference for the establishment of deep learning potentials for different types of molten salt. |
PDF Full Text Request