Thesis - Stefan Winkel Jessen (2008)

Mathematical Modeling of a Hall Hroult Aluminum Reduction Cell
Stefan Winkel Jessen
A Thesis Submitted in Partial Fulfillment of the Requirements for the Degree of Master of Science in Electrical Engineering at the Technical University of Denmark 2008
Mathematical Modeling of a Hall-Hroult Reduction Cell by Stefan W. Jessen Submitted to the Department of Electrical Engineering September, 2008 In Partial Fulfillment of the Requirements for the Degree of Master of Science in Electrical Engineering at the Technical University of Denmark
ABSTRACT
Aluminum reduction cells are used to produce aluminum by electrolysis of aluminum oxide, a process known as the Hall-Hroult process. Due to a limiting number of operational parameters that can be measured in an operating cell, the dynamics of the process are not fully understood and cell operation based largely on experience. In this work, the principles of the Hall-Hroult process are presented and a mathematical model developed to predict the dynamic behavior of an aluminum reduction cell. The model is based on an existing cell design which is currently being used at Norurl reduction plant in Iceland, and measurements data gathered from an operating cell for model validation. The model is constructed in three parts; a material balance model, a cell voltage model and a thermal balance model. The three parts are then combined into one overall cell model which may aid in the future improvements in control strategies and cell operation, as well as developing a predictive tool for the process itself.
Thesis Supervisor: Elbert Hendricks Keywords: Aluminum, Reduction Cell, Hall-Hroult Process, Mathematical Modeling, Lumped Parameter Method, Energy Balance, Thermal Balance, Cell Voltage, Electrolysis
TableofContents
List of Tables................................................................................................................................... VI List of Figures ............................................................................................................................. VIII Acknowledgements ..................................................................................................................... XIII 1 Introduction ............................................................................................................................... 1 1.1 Introduction to the Hall-Hroult process ............................................................................. 1 Potline .......................................................................................................................... 3 Electrolyte Composition ............................................................................................... 3 Cell Operation .............................................................................................................. 3 Measurements............................................................................................................... 5 1.1.1 1.1.2 1.1.3 1.1.4 1.2 1.3 2
Motivation ........................................................................................................................... 5 Overview ............................................................................................................................. 5
Material Balance Model ........................................................................................................... 7 2.1 2.2 2.3 2.4 2.5 Introduction ......................................................................................................................... 7 Aluminum Production Rate .................................................................................................7 Carbon Consumption Rate................................................................................................... 8 Alumina Consumption Rate ................................................................................................ 9 Cell Emissions ................................................................................................................... 10 Hydrogen Fluoride Evolution .................................................................................... 10 Vaporization ............................................................................................................... 12 Entrainment of Particulate Materials.......................................................................... 13 Sulphur Dioxide ......................................................................................................... 14
2.5.1 2.5.2 2.5.3 2.5.4 2.6 2.7 2.8 2.9 2.10
Alumina Feeding ............................................................................................................... 15 Alumina Dissolution .......................................................................................................... 16 Corrective Fluoride Additions ........................................................................................... 18 Total Material Consumption and Production .................................................................... 19 Properties of the Electrolyte .............................................................................................. 20 Density ....................................................................................................................... 20 Viscosity ..................................................................................................................... 20
2.10.1 2.10.2
II
2.10.3 2.10.4 2.11 2.11.1 2.11.2 2.12 2.12.1 2.12.2 2.12.3 2.12.4 2.13 2.13.1 2.13.2 2.14 2.15 3 2.14.1
Alumina Solubility .....................................................................................................21 Aluminum Solubility .................................................................................................. 21 Electrical Conductivity ...............................................................................................23 Heat Capacity ............................................................................................................. 23 Enthalpy of Reaction ..................................................................................................24 Enthalpy of Heating ................................................................................................... 25 Enthalpy of Dissolution.............................................................................................. 26 Net Energy Requirements .......................................................................................... 26 Inputs .......................................................................................................................... 27 Outputs ....................................................................................................................... 29 Current Efficiency ...................................................................................................... 29
Current Efficiency ............................................................................................................. 21
Energy Considerations ....................................................................................................... 24
The Model Construction .................................................................................................... 27
Model Validation ............................................................................................................... 29 Summary............................................................................................................................ 30
Cell Voltage Model.................................................................................................................. 32 3.1 3.2 3.3 Introduction ....................................................................................................................... 32 The Decomposition Voltage .............................................................................................. 32 Overvoltage ....................................................................................................................... 35 Concentration Overvoltage ........................................................................................ 36 Anode Reaction Overvoltage ..................................................................................... 38 Electrolyte Voltage Drop ........................................................................................... 39 Anode voltage drop .................................................................................................... 40 Bubble Voltage Drop ................................................................................................. 42 Cathode Voltage Drop ................................................................................................ 43 External Voltage Drop ............................................................................................... 44
3.3.1 3.3.2 3.4 3.4.1 3.4.2 3.4.3 3.4.4 3.4.5 3.5 3.6
Ohmic Voltage drops ......................................................................................................... 39
Total Cell Voltage ............................................................................................................. 44 The Model Construction ....................................................................................................46 Inputs .......................................................................................................................... 46 III
3.6.1
3.6.2 3.7 3.8 4
Outputs ....................................................................................................................... 47
Model Validation ............................................................................................................... 47 Summary............................................................................................................................ 48
Thermal Balance Model ......................................................................................................... 50 4.1 4.2 Introduction ....................................................................................................................... 50 The Mechanisms of Heat Transfer .................................................................................... 51 Conduction ................................................................................................................. 51 Convection ................................................................................................................. 52 Radiation .................................................................................................................... 53 Heat Capacitance ........................................................................................................ 56 Lumped Thermal Circuit ............................................................................................ 56
4.2.1 4.2.2 4.2.3 4.2.4 4.2.5 4.3 4.4
Modeling Approach ........................................................................................................... 57 Model Building Blocks...................................................................................................... 58 Internal Heat Generation ............................................................................................ 61 The Potshell ................................................................................................................ 62 Metal Pad.................................................................................................................... 64 Electrolyte .................................................................................................................. 65 Air Gap ....................................................................................................................... 66 Frozen Ledge .............................................................................................................. 68 Top Crust .................................................................................................................... 72 Loose Cover ............................................................................................................... 74 Anodes........................................................................................................................ 75 The Yoke and the Connector bar................................................................................ 78 The Cell Hood ............................................................................................................ 78 Heat distribution model .............................................................................................. 79 Ledge Profile Model................................................................................................... 80 Temperature Distribution ........................................................................................... 82 IV
4.4.2 4.4.3 4.5 4.5.1 4.5.2 4.5.3 4.5.4 4.6 4.6.1 4.6.2 4.6.3 4.6.4 4.6.5 4.7 4.7.1 4.7.2 4.8 4.8.1
The Cell Cavity .................................................................................................................. 64
Top Part ............................................................................................................................. 71
The Model Construction ....................................................................................................79
Model Validation ............................................................................................................... 81
4.8.2 4.9 4.10 4.11 5
Ledge Profile .............................................................................................................. 84
Model Reduction ............................................................................................................... 86 Combined Thermal Balance Model ................................................................................... 89 Summary............................................................................................................................ 89
Combined Cell Model ............................................................................................................. 91 5.1 Model validation ................................................................................................................ 91 Operating Conditions of the Physical Cell Chosen for the Validation ...................... 92 Measurement Noise .................................................................................................... 93 The Model Input Signals ............................................................................................ 94 5.1.1 5.1.2 5.1.3 5.2 5.3 5.4 5.5
Comparison Between the Simulated and the Measured Cell Voltage ............................... 98 Predicted Variations in Alumina Concentration ................................................................ 99 Predicted Variations in Bath Temperature ...................................................................... 100 Summary.......................................................................................................................... 100
Conclusions ............................................................................................................................ 102 6.1 Future Work..................................................................................................................... 103
References .............................................................................................................................. 104
Appendix A Tables .................................................................................................................... 107 Appendix B Model Block Diagrams ........................................................................................ 110 B.1 Material Balance Model ..................................................................................................... 110 B.2 Cell Voltage Model ............................................................................................................ 125
ListofTables
2.1 2.2 2.3 2.4 2.5 2.6 2.7 3.1 3.2 3.3 3.4 3.5 3.6 3.7 4.1 4.2 4.3 4.4 4.5 4.6 4.7 5.1 5.2 Calculated material consumption for a 24 hour period of a 182 kA reduction cell .............. 19 with 94.5% current efficiency. Calculated material production for a 24 hour period of a 182 kA reduction cell ................. 19 with 94.5% current efficiency. The enthalpy of the reactions considered in the material balance model .............................. 25 Theoretical net energy requirements of the reduction process at various operating ............. 26 temperatures. Inputs to the material balance model in figure 2.5 ................................................................ 27 Outputs from the material balance model in figure 2.5. ........................................................ 29 Comparison between calculated and measured current efficiency. ...................................... 30 The Gibbs free energy of formation of the compounds in the primary reaction ................... 33 Calculated voltage distribution in a cell in Norurl. ............................................................ 45 Inputs to the material balance model in figure 3.10 .............................................................. 46 Outputs from the cell voltage model in figure 3.10. .............................................................. 47 Cell operating conditions on the day of the measurements (measured). ............................... 47 Comparison between the theoretical and the measured voltage distribution in ................... 48 reduction cell A. Comparison between the theoretical and the measured voltage distribution in .................... 48 reduction cell B. The effective thermal resistances of each air gap region. ...................................................... 68 The effective thermal resistances between the anode surfaces and the air gaps.................... 76 Inputs to the heat distribution model. .................................................................................... 80 Cell operating conditions on the day of the measurements (measured). ............................... 82 Calculated average of measured and simulated temperature values...................................... 83 Differences between measured and simulated bath temperatures at various conditions ....... 84 The norms of the gain errors between each input and output of the reduced order model.... 88 Cell operating conditions and control system set-points on the day of the measurements ... 92 Unknown operational parameters that are required to be defined in order to simulate ......... 93 the behavior of the cell chosen for the model validation.
A.1 The enthalpy of formation of various substances involved in the aluminum reduction ..... 107 process VI
A.2 The entropy of various substances involved in the aluminum reduction process ............... 107 A.3 Shomate coefficients of various substances involved in the aluminum reduction ............. 108 process A.4 Shomate coefficients of various substances involved in the aluminum reduction ............. 108 process A.5 Material specifications of each material used in the reduction cell ..................................... 108 A.6 The heat transfer coefficients considered in the thermal balance model ............................. 109
VII
ListofFigures
1.1 2.1 2.2 2.3 2.4 2.5 2.6 2.7 3.1 3.2 3.3 3.4 3.5 3.6 3.7 3.8 3.9 3.10 4.1 4.2 4.3 4.4 4.5 Cross section of a Hall-Hroult reduction cell with prebaked anodes ..................................... 1 Hydrogen fluoride evolution as a function of alumina concentration and bath ratio ............. 11 Fluoride evolution due to particulate emissions as a function of alumina concentration ..... 13 and bath ratio. Cell voltage as a function of alumina concentration, if the anode effect is assumed ............. 15 to occur at 2%. Time response of alumina dissolution, when 1 kg of alumina is fed to a reduction .............. 18 cell at t = 0s. Block diagram of the material balance model ........................................................................ 27 Signal generator for the material feed inputs of the material balance model ......................... 28 Stepwise accumulation of the total amount of material fed to the cell when 1 kg of ............ 28 material is fed every 1 minute. The relationship between the decomposition voltage and alumina concentration, ............... 35 at different bath temperatures. The relationship between the anode concentration overvoltage and alumina ........................ 37 concentration, at different bath ratios. The relationship between the cathode concentration overvoltage and the bath ratio ............. 37 The relationship between the anode reaction overvoltage and the alumina concentration ... 39 at different bath ratios. An anode Assembly................................................................................................................ 40 The shapes of the 20 anodes, if the first anode is considered as the youngest and the .......... 40 replacements are divided between 28 days. The relationship between the bubble voltage drop and the alumina concentration at............ 42 different bath ratios. A cathode assembly ................................................................................................................ 43 The cell voltage as a function of alumina concentration and interpolar distance (IPD) ........ 45 Block diagram of the cell voltage model................................................................................ 46 Two-dimensional gray, diffuse duct ....................................................................................... 55 A 3-dimensional lumped thermal circuit ................................................................................ 57 2D representation of a 3D rectangular block.......................................................................... 58 A three dimensional sidewall structure is composed of eight rectangular blocks.................. 59 2D representation of a 3D sidewall ........................................................................................ 59 VIII
4.6 4.7 4.8 4.9 4.10 4.11 4.12
A combination of a block element and a sidewall element .................................................... 60 Linearization error introduced by the linearized relationship for heat radiation ................... 61 A simplified schematic of the structure of the potlining at Norurl .................................... 62 The potshell divided into model elements for the lumped parameter method. ...................... 63 The collector bar layer (C1), a) Top view, b) Cross sectional view ....................................... 64 Error introduced to the heat capacitance of the electrolyte as a function of alumina ............ 66 concentration and temperature, if the heat capacitance is assumed constant at 1.767 J/gK. Error introduced to the heat capacitance of the electrolyte as a function of aluminum ......... 66 fluoride concentration and temperature, if the heat capacitance is assumed constant at 1.767 J/gK. The cross section and facing surfaces of each air gap region. a) Air gap A, ........................ 67 b) Air gap B, c) Air gap C. Linearization error introduced to the surface heat flux of side 1 in air gap A ....................... 67 a) The inclined sidewall and frozen ledge, b) A simple representation of the frozen ........... 69 ledge. Time response of the ledge profile at the electrolyte-ledge interface if TEL is increased ...... 70 by 1 at t = 2500s. Time response of the ledge profile at the metal pad-ledge interface if TEL is increased ....... 71 by 1 at t = 2500s. The top crust and anodes, shown from above. The geometry is slightly exaggerated .......... 72 for illustrative purpose. Crust part A ............................................................................................................................ 73 Crust part B ............................................................................................................................ 73 Crust part C ............................................................................................................................ 74 The elements adjacent to the anodes and the division between the 5 horizontal layers......... 76 The Anode crust layer, seen from the top .............................................................................. 77 A block diagram of a typical state-space model..................................................................... 79 A block diagram of the ledge profile model (bath/ledge interface) ....................................... 81 Simulated thermal distribution on the sidewall surface, compared to measured values ........ 83 Calculated superheat 12 days prior to the day of the measurements ...................................... 84 Simulated ledge thickness 12 days prior to the day of the measurements ............................. 85 Measured ledge profile. a) adjacent to end anodes, b) adjacent to center anodes, ................. 85 c) adjacent to end anodes (close to a ventilation duct).
4.13 4.14 4.15 4.16 4.17 4.18 4.19 4.20 4.21 4.22 4.23 4.24 4.25 4.25 4.26 4.27 4.28
IX
4.29 4.30 4.31 4.32
The first 20 Hankel singular values of the full-order heat distribution model ....................... 87 Singular values for model reductions of the thermal balance model from 101 to ................. 87 8 states. a) Output 1, b) Output 2, c) Output 3. Singular values for the error system, shown for each output, a) Output 1, b) Output 2, ...... 88 c) Output 3. Comparison between the step response of the full order and the reduced system for a ........ 88 step increase in the liquidus temperature at t = 0, a) Response of Output 1, b) Response of Output 2, c) response of Output 3. The combined thermal balance model.................................................................................... 89 Block diagram of the combined cell model............................................................................ 91 The cell current on the day of the measurements ................................................................... 94 Rotation of the servo motors on the day of the measurements. The anode bridge is ............. 95 raised by clockwise rotation. A block diagram of a signal converter for converting measured servo rotation to ................ 95 bridge movements. a) Block diagram, b) Subsystem. A block diagram of a signal generator for replicating the behavior of the interpolar ............ 96 distance. The estimated behavior of the interpolar distance on the day of the measurements .............. 96 The intervals between batch additions on the day of the measurements................................ 97 A block diagram of a signal generator for converting feeding intervals to accumulative ..... 97 amount of alumina that is fed to the cell Comparison between the measured and simulated cell voltage ............................................. 98 The errors between the two responses for the first twenty hours ........................................... 98 Predicted variations in alumina concentration ....................................................................... 99 Predicted variations in bath temperature .............................................................................. 100 Block diagram of the material balance model ...................................................................... 110 Block diagram of the mass to concentration conversion subsystem in figure B.1 .............. 110 (Mass to wt%). Block diagram of the current efficiency subsystem in figure B.1 ........................................ 111 Block diagram of the production and consumption subsystem in figure B.1 ...................... 111 Block diagram of the primary reaction subsystem in figure B.4.......................................... 112 Block diagram of the enthalpy of reaction subsystem in figure B.5 .................................... 112 Block diagram of the enthalpy of heating subsystem in figure B.5 ..................................... 112
4.33 5.1 5.2 5.3 5.4 5.5 5.6 5.7 5.8 5.9 5.10 5.11 5.12 B.1 B.2 B.3 B.4 B.5 B.6 B.7
B.8 B.9 B.10 B.11 B.12 B.13 B.14 B.15 B.16 B.17 B.18 B.19 B.20 B.21 B.22 B.23 B.24 B.25 B.26 B.27 B.28 B.29 B.30 B.31 B.32 B.33 B.34 B.35 B.36 B.37 B.38
Block diagram of the fluoride evolution subsystem in figure B.4 ....................................... 113 Block diagram of the enthalpy of reaction subsystem in figure B.8 .................................... 113 Block diagram of the vaporization subsystem in figure B.4 ................................................ 114 Block diagram of the total vapor pressure subsystem in figure B.10................................... 114 Block diagram of the partial pressure of the monomer subsystem in figure B.10 ............... 114 Block diagram of the enthalpy of reaction subsystem in figure B.10 .................................. 114 Block diagram of the entrainment subsystem in figure B.4 ................................................. 115 Block diagram of the material input subsystem in figure B.1 .............................................. 115 Block diagram of the alumina dissolution subsystem in figure B.15 ................................... 116 Block diagram of the alumina dissolution subsystem in figure B.16. .................................. 116 Block diagram of the enthalpy of dissolution subsystem in figure B.16 ............................. 117 Block diagram of the enthalpy of dissolution subsystem in figure B.18 ............................. 117 Block diagram of the sodium oxide reaction subsystem in figure B.16............................... 117 Block diagram of the enthalpy of reaction subsystem in figure B.20 .................................. 117 Block diagram of the sodium carbonate reaction subsystem in figure B.15 ........................ 118 Block diagram of the enthalpy of reaction subsystem in figure B.22 .................................. 118 Block diagram of the aluminum fluoride addition subsystem in figure B.15 ...................... 118 Block diagram of the electrolyte properties subsystem in figure B.1 .................................. 119 Block diagram of the electrolyte density subsystem in figure B.25 ..................................... 120 Block diagram of the bath ratio subsystem in figure B.25 ................................................... 120 Block diagram of the electrolyte viscosity subsystem in figure B.25 .................................. 121 Block diagram of the solubility of Al2O3 subsystem in figure B.25 .................................... 121 Block diagram of the subsystem for determining A in figure B.29 ..................................... 122 Block diagram of the subsystem for determining B in figure B.30 ..................................... 122 Block diagram of the solubility of aluminum subsystem in figure B.25 ............................. 123 Block diagram of the electrical conductivity of the electrolyte subsystem in figure B.25 .. 123 Block diagram of the specific heat of the electrolyte subsystem in figure B.25 .................. 124 Block diagram of the liquidus temperature subsystem in figure B.25 ................................. 124 Block diagram of the cell voltage model.............................................................................. 125 Block diagram of the decomposition voltage subsystem in figure B.36. ............................. 126 Block diagram of the activity of alumina subsystem in figure B.37. ................................... 126
XI
B.39 B.40 B.41 B.42 B.43 B.44 B.45 B.46 B.47 B.48
Block diagram of the anode concentration overvoltage subsystem in figure B.36 .............. 126 Block diagram of the critical current density subsystem in figure B.39 .............................. 127 Block diagram of the cathode concentration overvoltage subsystems in figure B.36 ......... 127 Block diagram of the anode reaction overvoltage subsystem in figure B.36 ....................... 127 Block diagram of the reaction limited current density subsystem in figure B.42 ................ 128 Block diagram of the electrolyte voltage drop subsystem in figure B.36 ............................ 128 Block diagram of the bubble voltage drop subsystem in figure B.36 .................................. 128 Block diagram of the bubble layer thickness subsystem in figure B.45 .............................. 128 Block diagram of the bubble surface coverage subsystem in figure B.45 ........................... 129 Block diagram of the ohmic voltage drops subsystem in figure B.36 .................................. 129
XII
Acknowledgements
I foremost thank my thesis advisor Dr. Elbert Hendricks, whose guidance made this thesis possible, and whose insight, wisdom and integrity has inspired me. Thanks also to rstur Helgason and rur Runlfsson at Mannvit Engineering for their support and help throughout my endeavors. I would also wish to thank the cell operators at Northern Aluminum for their advice and suggestions as well as their extensive knowledge of the operation of aluminum reduction cells. Finally, I extend my personal thanks to my family, for encouraging me every step along the way.
XIII
Mathematical Modeling of a Hall-Hroult Reduction Cell
1 Introduction
Despite the fact that aluminum is the third most abundant element in the Earths crust only exceeded by oxygen and silicon, aluminum does not exist in nature in its elemental form due to its high reactivity. Aluminum exists in its oxidized form, most commonly in aluminates and silicates. The principal ore of aluminum is bauxite, where it exists mainly as aluminum oxide (Al2O3), also known as alumina. The first success of producing elemental aluminum is stated to be had by the Danish physicist Hans Christian rsted in 1825, who managed to produce small amount of metallic aluminum by thermal reaction of aluminum chloride with potassium amalgam. For six decades, a number of different attempts were made to produce the newly discovered metal by more efficient thermal reactions, with different results. The processes were expensive, energy intensive and the quality of the metallic product was low. In 1886, Charles Martin Hall and Paul Hroult discovered almost simultaneously that aluminum could be produced by less energyintensified process, by electrolysis of a cryolite-alumina melt, a process known as the Hall-Hroult process. The new production method then became the foundation of the commercial production of aluminum, when Pittsburgh Reduction Company (Alcoa) was started in 1888 by C.M. Hall. In less than two centuries a rare metal that was more expensive than gold has become the second most produced metal in the world, after iron. According to the Australian Bureau of Agricultural and Resource Economics, the world production of primary aluminum in 2007 reached 37.85 million tons and the demand is expected to grow through the century. The Hall-Hroult process has undergone many refinements and considerable improvements in the last decades and is currently the only method used for large-scale production of metallic aluminum. With ongoing innovative research and development efforts the aluminum industry is constantly searching for ways to maximize productivity and efficiency with reduced energy requirements and costs. With the fast evolution of the personal computer and computer aided simulation, mathematical modeling of the process has become the most important role when determining the best options for improvements. Despite the fact that the Hall-Hroult process is based on a simple theory there is a wide range of complex physiochemical phenomena that must be taken into consideration when modeling the overall process, so models are generally constructed with specific goal in mind rather than including all aspects. Because of the simplifying assumptions that are invariably made, all models developed differ, although they show a common approach to some extent. In the following a short introduction to the Hall-Hroult process will be given in conjunction with the motivation of developing a new mathematical model for the aluminum reduction industry.
1.1 IntroductiontotheHallHroultprocess
The current commercial method of producing elemental aluminum from bauxite requires a two stage process. First, deposits of bauxite ore are mined and refined into alumina, by a process known as the Bayer process, named after its developer. The alumina is then reduced to metallic aluminum by the Hall-Hroult process. The Hall-Hroult process is governed by Faradays law of 1
electrolysis, whereas aluminum is separated from oxygen by passing an electrical current through a melt composed essentially of dissolved alumina in cryolite. The process is carried out in large thermally insulated pots, generally referred to as reduction cells (fig. 1.1).
Figure 1.1 Cross section of a Hall-Hroult reduction cell with prebaked anodes (1).
A modern aluminum reduction cell consists of a steel shell, lined with thermally insulating refractory material and a carbon lined cavity which serves as a containment vessel for the molten bath (electrolyte) and the metallic product. A direct electrical current is passed through the electrolyte, entering the cell through carbon anodes and conducted out of the cell through steel current collector bars that are fixed in the carbon lining. The electric current provides sufficient amount of electrons to liberate the aluminum from the oxygen, which is deposited at the boundary between the electrolyte and the molten aluminum, whereas the aluminum acts as the cathode. The oxygen dissolved in the bath then combines with the bottom surface of the anodes to form carbon monoxide and carbon dioxide, according to the reaction:
2 Al2O + 3C 3CO2 + 4 Al
(1.1)
The anodes are thus consumed by the reduction process. The majority of cells currently in operation use anodes made of baked carbon composite made of petroleum coke and coal-tar pitch which are replaced on a regular schedule. The electrolyte is made up of high concentrations of fluoride composites making it highly corrosive. Reduction cells are designed and operated in such a way that electrolyte freezes on the cavity walls, forming a layer of protection from the severe attack by the electrolyte and the molten aluminum as well as preventing the sidewalls from acting as cathodes. Retaining sufficient thickness of the frozen layer is thus vital for extended operation. Furthermore, a crust of frozen electrolyte and alumina powder covers the top of the bath which serves as thermal insulation. The alumina is fed to the cell in batches at predefined intervals. Generally pneumaticallyoperated crust breakers are used to punch holes through the crust, and the alumina then poured into the molten bath, where it dissolves.
The metal product is periodically tapped of the cell, either by draining the cathode or by siphoning. A large reservoir of metal is though always left in the cell to minimize the effects of the tapping procedure on the cells thermal balance. 1.1.1 Potline The production rate of electrolysis is proportional to the applied electrical current. The chemical bonds between aluminum and oxygen are strong, resulting in a theoretical maximum of approximately 0.336 g of produced aluminum per ampere hour. It is thus evident that for large scale production, the operating current has to be significantly high. Modern reduction cells are typically operated with an operating current in the range of 150 to 250 kA but cells operating at up to 500 kA currently exist. The minimum voltage required by the electrochemical reaction is only around 1.2 volts, but due to electrode polarization and electrical resistance of the various cell components in the current path, the total voltage drop over modern reduction cells is generally in the range of 4 to 5 volts. High voltage rectifiers cost much less and are more efficient than high current rectifiers of the same power rating, so a large number of reduction cells are thus connected in series, forming a potline. 1.1.2 ElectrolyteComposition The predominant constituent of industrial electrolytes is cryolite (Na3AlF6) which is a double fluoride of aluminum and sodium, comprising at least 70% of the total electrolyte mass. Natural cryolite is too rare to satisfy the world demands, but synthetic cryolite can be manufactured by a number of means. Cryolite is the predominant solvent for alumina, with a maximum solvability of 15% by weight. The melting point of pure cryolite is 1011C but with certain additives, such as excess aluminum fluoride (AlF3) and calcium fluoride (CaF2), the melting point can be lowered to 920-970C, increasing the energy efficiency of the process. The three fluorides, Na3AlF6, AlF3 and CaF2 form the basis of the electrolyte currently used by the industry, but an increasing number of reduction plants also introduce lithium fluoride (LiF) and magnesium fluoride (MgF2) to the electrolyte for further benefits such as further lowering the melting point of the electrolyte and increasing the electrical conductivity of the electrolyte to decrease I2R losses. There are advantages and disadvantages of the various additives, so the composition of the electrolyte and the concentrations of the electrolyte constituents vary between plants. Each cell design and operating procedures defines the precise requirements and compromises have to be made between i.e. productivity, energy efficiency and the quality of the product. 1.1.3 CellOperation To guarantee a long life of operation and efficient production, a reduction cell has to be held under constant observation and scheduled maintenance. In the event of catastrophic failures, it is sometimes necessary to rebuild part of the cell or in worst case the cell has to be replaced. Apart from the cost of spare parts and building material, failures result in decreased plant productivity. It is thus a common practice to cut out a cell before actual failure if certain types of signs persist. 3
With constant improvements in automatic control, cell operation has been greatly simplified, but the diligence and experience of potroom workers still plays the most important role. In the following a short description of some the abnormal operating conditions that can arise under operation will be given and how they can be prevented. 1.1.3.1 TheAnodeeffect The anode effect can be described as a blockage effect which greatly inhibits the current flow between an anode and the electrolyte and is normally manifested by a sudden increase in the cell voltage which can vary from only few volts up to 80 volts. The anode effect arises when the concentration of dissolved alumina falls below a critical level, which can vary from 0.5 to 2.5 wt% (2) depending on cell design, operating current density and temperature. Due to the high voltage increase, the energy dissipation at the anode interface can become extremely high, greatly disturbing the cell thermal balance. Apart from higher energy consumption, the anode effect causes an increase in bath temperature and hence melting of the layer of freeze protecting the sidewalls from corrosion. For this reason, it is important to limit the number of anode effects and their duration. 1.1.3.2 Sidewallfailure If the bath temperature becomes too high, the protective layer of freeze can melt away, exposing the sidewalls to the corroding effects of the molten electrolyte and aluminum. The erosion of an exposed sidewall is fast under these conditions and leads to sidewall failure if the condition persists. In case of sidewall failure the cell has to be taken out of service and the sidewalls rebuilt. To prevent this from happening, the cell temperature has to be held within proper limits. 1.1.3.3 Cathodefailure A mixture of undissolved alumina and bath sinks to the bottom of the cavity, forming sludge. If alumina is fed too rapidly and/or the bath temperature is too low, the sludge accumulates and promotes erosion of the cathode lining which can eventually lead to cathode failure. In worst case the molten metal drains directly through the bottom of the cell or along the collector bars. In the case of cathode failure a large portion of the cell has to be rebuilt. With a good feeding strategy and proper cell temperature, cathode erosion can be minimized. 1.1.3.4 PowerOutage The worst case scenario in reduction plant operation is power blackout. In the case of a blackout, the electrolyte can no longer be held at an elevated temperature and begins to freeze. Typically the power has to be restored within some specific period of time to prevent permanent damage to the reduction cells, where the frozen ledge becomes so thick that the anodes become jammed. If it is known by forehand that the power cannot be restored within the time limits, an attempt is made to
drain all the cells in time. In the case of a brownout or if sufficient power can be provided by emergency power generation, the electrolyte can though be held molten for extended time. The damage resulting from a power outage can become so severe that a whole potline has to be replaced or even plant decommissioning. 1.1.4 Measurements The two main operational parameters in an industrial reduction cell are the alumina concentration and bath temperature, but continuous measurements of these parameters have not proven to be technically or economically viable because of the corrosive environment inside the cell. However, they are generally measured intermittently to provide a cross check of average process conditions and cell stability as an indication of operational abnormalities. The parameters that are generally measured in real-time are the cell voltage and the line current. The cell voltage is a function of both the bath temperature and composition and is thus generally used as a basis for process automation.
1.2 Motivation
To summarize, the scope of this work is the development of a comprehensive mathematical model of a Hall-Hroult reduction cell, which may aid in the future improvements in control strategies and cell operation, as well as developing a predictive tool for the process itself. To meet the fast increasing demand of metallic aluminum, an increasing number of reduction plants are being operated beyond their production capabilities by, i.e., operating the reduction cells at higher currents than rated. When operating an aluminum reduction cell, there are numerous operation parameters that have to be held within a tight range to guarantee efficient production of aluminum and long lifetime of the cell, so new operating conditions have to be accompanied by either change in design or operation. With the limit number of continuous measurements available, alterations to the operating conditions are generally made in small steps to avoid dramatic disturbances, but most generally by trial and error. With the aid of a mathematical model of the process and reduction cell as a whole, changes in operating conditions may be predicted and how they can be accounted for to maintain high productivity, energy efficiency and minimize the overall cost of operation. Furthermore the model may aid in future improvement to existing control and operating strategies and specifying limits to changes that can be made in the variables of an operating cell.
1.3 Overview
The objective of this work is to develop a mathematical model of a Hall-Hroult reduction cell as well as develop a better understanding of the reduction process itself and the effects of variations in operational parameters. Despite the coupling that exists between various complex physiochemical phenomena, compromises have to be made between simplicity and accuracy of the model by only focusing on the most relevant aspects of the Hall Hroult process and cell 5
operation. In order to validate if the model provides a good representation of the physical system, the model is based on an existing 180kA cell design which is currently being used at Norurl reduction plant in Iceland. Using measurement data acquired from Norurl, the model can thus be validated for consistency. In chapter 2, the material balance of an aluminum reduction cell will be studied in conjunction with material production and consumption, emissions and energy requirements. Based on the material balance, a model of the dynamic concentrations of the electrolyte constituents and the properties of the electrolyte will then be constructed and validated by comparison to measurement data. In chapter 3, the various contributions to the cell voltage drop will be studied which will then be used as a basis for constructing a cell voltage model. In chapter 4, the dynamic temperature of the electrolyte will be modeled, by modeling the heat distribution in a reduction cell using the lumped parameter method. The model order will then be reduced to allow real-time simulation and the accuracy of the model validated by comparing the simulated heat distribution to measurement data. In chapter 5, the material balance, cell voltage and thermal balance models will be combined into one overall reduction cell model. The dynamics of the combined model will then be compared to the dynamics of an operating cell, using measured cell voltage as a mean for validation. Finally, a conclusive summary and proposal to future work will be given in chapter 6.
2 MaterialBalanceModel
2.1 Introduction
Material balances are based on the law of conservation of mass. The properties of the electrolytic bath depend on the chemical composition of the electrolyte, so the dynamic change in the concentration of each chemical involved is of great importance to the overall cell model. Normally balances are performed for a given production of metal, in this case either per kilogram or ton of aluminum. Mathematically, the material balance for a system with a chemical reaction is:
Input + Generation = Output + Accumulation
(2.1)
In the aluminum reduction process, the materials used in the largest amounts are alumina (Al2O3) and Carbon (C). The theoretical consumption is about 1.89 kg of Al2O3 and 0.33 kg of C per kilogram of aluminum produced. The main paths where all ingredients will leave the cell are either with the liquid metal, as a gas or entrained particulate emissions. On the basis of the amount of metal produced, the consumption of the electrolyte constituents may appear to be small though many times the original weight of the electrolyte is added to the cell during its operating life. Some of the constituents such as aluminum fluoride (AlF3) and sodium fluoride (NaF) are consumed faster than others (as a result of volatilization) and it is easy to perform materials balance on them. However, others such as calcium, lithium and magnesium fluorides have such a low consumption rate that it is difficult to perform a precise materials balance. In this chapter, the material balance of an aluminum reduction cell will be studied, which will then be used as a basis for modeling dynamic changes in the electrolyte composition, properties and energy requirements. Attention is only focused on the dynamic concentrations of the highly consumed materials. The others that have very low consumption rate will be assumed constant. Generally a sample of the electrolyte is regularly taken and the concentration of the electrolyte constituents determined in an onside laboratory. In the material balance model, the concentration of the slowly consumed constituents can thus be corrected periodically.
2.2 AluminumProductionRate
The aluminum electrolysis process is governed by Faradays law of electrolysis, which states that the mass of a substance produced at an electrode during electrolysis is directly proportional to the number of electrons transferred at that electrode:
m= 1 QM 1 QM = eN A n F n
(g)
(2.2)
m : Mass of the substance produced at the electrode (g) M : Molar mass of the substance (g/mol) Q : Total electric charge passed through the solution (C) n : Number of free electrons per atom of the substance e : The elementary charge carried by a single proton (~1.60210-19 C) NA : Avogadros constant (~6.0221023 atoms/mol) F : Faradays constant (~96485 A/s)
The total charge Q is the integral electric current I(t) over time:
Q = I (t ) dt
0 T
(C)
(2.3)
where T is the total amount of time of the electrolysis. According to equation 2.2 and 2.3, the theoretical production rate of the substance produced is thus:
dm 1 M = I (t ) dt F n
(g/s)
(2.4)
The number of free electrons per aluminum atom is three, so the theoretical maximum quantity of aluminum produced is approximately 0.336 grams of aluminum per ampere-hour. In reality the quantity of aluminum produced is lower than this theoretical value, mostly because of the internal recombination of aluminum and carbon dioxide, 2Al + 3CO2 Al2O3 + 3CO. The ratio of the actual quantity of aluminum produced to the theoretical quantity is referred to as the Faradays Efficiency or Current Efficiency (%CE if represented in percentage). In modern industrial cells the average current efficiency for metal production generally varies between 90 and 95%, assuming that only aluminum is deposited at the cathode (2). The current efficiency depends greatly on the cells design as well as a number of other variables including temperature, current density, interpolar distance (distance between the electrodes), etc. An expression for calculating the current efficiency will be given in section 2.11. Taking the current efficiency into account, the actual production rate can then be expressed as follows:
dmAl , p dt
= 9.322 108 I
(kg/s)
(2.5)
where I is the cell current in amperes.
2.3 CarbonConsumptionRate
When producing aluminum by electrolysis, the aluminum oxide is decomposed into aluminum and oxygen. The oxygen formed then reacts immediately with the carbon anode to carbon dioxide and carbon monoxide described by the two principle electrolysis equations:
2 Al2O3 + 3C = 4 Al + 3CO2
(2.6) (2.7)
Al2O3 + 3C = 2 Al + 3CO
Pearson and Waddington (3) assumed that the loss in current efficiency is only due to the internal recombination of aluminum and carbon dioxide. By modifying the two principal equations with this assumption, considering that a (1-) fraction of aluminum is lost if an fraction is produced, results in the modified electrolysis equation:
2 Al2O3 + 3C = 4 Al + 3 ( 2 1) CO2 + 6 (1 ) CO
(2.8)
From the rate of aluminum produced (eq. 2.5) and the electrolysis equation (eq. 2.8), the theoretical carbon consumption rate can then be expressed as follows:
dmC ,c dt = 3.111 108 I t
(kg/s)
(2.9)
Correspondently, the production rate of carbon monoxide and carbon dioxide can be expressed by:
dmCO , p dt dmCO2 , p
dt
= 1.452 107 I t (1 )
= 1.140 107 I t ( 2 1)
(kg/s)
(2.10)
(kg/s)
(2.11)
According to Grjotheim and Welch (2), oxidation of carbon is favored at all temperatures but carbon dioxide formation is favored at temperatures below 1000K, while carbon monoxide will become the dominant product at higher temperatures. However, the proportion between the two gas species is not of great importance in the material balance model since the carbon consumption remains the same. There are various additional loss mechanisms that contribute to the overall consumption, such as air burning and detachment of anode carbon particles from the blocks. According to Grjotheim and Welch (2), the total excess carbon consumption to the theoretical consumption can be expected to be in the range of 7-18%. Furthermore, the anodes are not made up of pure carbon. The anodes used in industrial aluminum cells are made from paste consisting of 70-80% aggregate coke and 30-20% pitch as a binder (4), with about 98-99% carbon concentration. The anode consumption rate is thus considerably higher in practice than expressed by equation 2.9.
2.4 AluminaConsumptionRate
Alumina is the material used in the largest quantity. The alumina consumption is governed by the principal electrolysis equations (eq. 2.6 and 2.7). As for the carbon, the alumina consumption can be expressed from the rate of aluminum produced (eq. 2.5) and the modified electrolysis equation (eq. 2.8) as follows: dmAl2O3 ,c dt = 1.7614 107 I (kg/s) (2.12)
The theoretical consumption is about 1.89 kg of pure alumina per kilogram of aluminum produced. Smelter grade alumina is not pure, so the total amount of alumina that is required for the reduction process is thus higher in practice than the theoretical consumption. Up to 1% of the alumina mass is natural impurities and about 2-3% of the mass is moisture (2). The main impurity is sodium oxide, while calcium, silicon, iron and titanium oxides are also present (2). Furthermore, when using dry scrubbers for removing fluorides from the effluent pot gases and bag filters for the particulates, the alumina is used as a gas absorbent before it is fed to the cell. A part of the feed material is thus recycled fluoride which exists in the alumina bulk material mainly as aluminum
fluoride. When using dry scrubbers, the fluoride content can range from 1.0 to 1.5% of the mass, depending on the absorbability of the alumina used. This means that only about 94-95% of the raw feed material mass is in fact pure alumina so the actual bulk consumption of alumina can be expected to be around 5-6% higher than expressed by equation 2.12.
2.5 CellEmissions
The evolution of gases from the anodes provides a transpiring medium for volatile materials as well as entraining particulate matter. The amounts of gases, dust or fumes emitted vary for different cell types. For a cell with prebaked anodes and normalized to 1 metric ton of aluminum, the values lie between 8 and 15 kg of hydrogen fluoride (HF), 8 and 12 kg of particulate fluorides (i.e., in NaAlF4) and 11 and 12 kg of sulfur dioxide (SO2). The net anode carbon consumption varies between 400 and 500 kg per ton of aluminum, resulting in about 1.7 tons of carbon monoxide/dioxide. There are a number of other chemical species that make up the emissive gases, such as the greenhouse gases tetrafluoromethane and hexafluoroethane as well as the other constituents of the particulate material. Since the mass ratio of these chemicals to the total amount of emissive substances is very small compared to HF, NaAlF4 and SO2, they will be neglected. The fine particulate fluorides and gaseous hydrogen fluoride are of greatest concern in relation to the material balance, since fluorides are lost from the electrolyte. These are present in the anode gases as a result of:

Gaseous fluoride evolution Particulate emission Entrainment of particulate materials
Subsequent discussions will concentrate on these.

2.5.1 HydrogenFluorideEvolution Approximately half of the cell fluoride emissions, predominantly hydrogen fluoride (HF), leaving the cell are gases. The sources of the hydrogen (H) can be from the anode carbon or from moisture present in the air or in the alumina feed. Hydrogen absorbed or entrapped in the anode carbon matrix can be electrochemically oxidized to either water or hydrogen fluoride because the reaction requires less operating potential than the alumina electrolysis. Part of the hydrogen reacts with cryolite to produce hydrogen fluoride according to equation 2.13 (2):
2 Na3 AlF6 (l ) + 3 H 2 ( g ) 2 Al (l ) + 6 NaF (l ) + 6 HF ( g )
(2.13)
Water enters the cell primarily through moisture in the alumina feed material as well as moisture present in the air. The water produced by oxidation of the hydrogen in the anodes also makes a small contribution. The most favored reaction that water may undergo with the electrolyte constituents is that it may react with aluminum fluoride either in the electrolyte or in the emissions
10
to produce hydrogen fluoride. According to Baimakov and Vetyukov (5), the reaction of the dissolved water with the electrolyte may be described in simplified form:
2 AlF3 ( s ) + 3H 2O ( g ) Al2O3 ( s ) + 6 HF ( g )
(2.14)
According to Haupin (6), the total rate of the gaseous fluoride evolution in kilograms per metric ton of aluminum, can be estimated with equation 2.15:
FG = 1.4537 10 e
7
2 (0.78127 Rb 3.1733 Rb 8444/ Tb )
462 191Rb Pb %CE
cH 2 O c c Al2O + H max 3 37.44 21.5 c Al2O3
0.5
0.462
(2.15)
Rb : Bath ratio (cNaF/cAlF3) Pb : Barometric pressure (kPa) %CE : Current efficiency (%) Tb : Bath temperature (K) : Concentration of alumina (wt%) : Maximum solubility of alumina (wt%) : Concentration of water in the alumina feed material (wt%) cH : Concentration of hydrogen in the carbon anodes (wt%)
Haupin (6) suggests that the average concentration of moisture in the alumina feed material is approximately 2.8% and the concentration of hydrogen in the anode carbon 0.093%. From equation 2.13, 2.14 and 2.15, the consumption and production of electrolyte constituents due to cell fluoride emissions can then be approximated as follows, in kilograms per metric ton of aluminum produced.
m Al = 0.01155 FG
(2.16) (2.17) (2.18)
mAl2O3 = 0.8726 FG m AlF3 = 1.473 FG
The calculated hydrogen fluoride evolution as a function of alumina concentration and bath ratio at typical operating conditions is shown in the figure below.
18 kg of HF / ton of Al 16 14 12 10 8 6 2 3 4 5 6 Al2O3 Concentration (wt%) 7 8 Rbath = 1.15 Rbath = 1.20 Rbath = 1.25
Figure 2.1 Hydrogen fluoride evolution as a function of alumina concentration and bath ratio.
11
As can be seen in figure 2.1, the evolution of hydrogen fluoride is closely related to the alumina concentration and bath ratio. To minimize the evolution, the concentration and ratio thus have to be as high as possible.
2.5.2 Vaporization Studies have established that sodium tetrafluoroaluminate (NaAlF4) is the most volatile species existing above cryolite-alumina melts. The vapor exists both as the monomer NaAlF4 and the dimer Na2Al2F8, from the reaction of sodium fluoride with aluminum fluoride, according to equation 2.19:
NaF + AlF3 NaAlF4
(2.19)
For vaporization to be a significant factor, it must have a driving force for removing the electrolyte vapors from the cell. In an operating cell this is provided by the gases from the anodes, which are mostly carbon dioxide and carbon monoxide. The rate of loss of vapors is thus dependent on the saturation vapor pressure as well as the number of moles of the transpiring gases. The total vapor pressure PT (kPa) which is equal to the sum of the partial pressures of the individual gases can be expressed as a function of bath composition and temperature by equation 2.20 (6):
log ( P ) = B A T b T where
A = 7101.6 + 3069.7 Rb 635.77 Rb2 + 764.5 cAl2O3 1 + 1.0817 c Al2O3 1.1385 cAl2O3 1 + 3.2029 cAl2O3 + 13.2cCaF2 + 0.0068cCaF2
(2.20)
(2.21) (2.22)
3 B = 7.0174 + 0.6844 Rb 0.08464 Rb +
Assuming that the transpiring gases are saturated with vapor, the rate of fluoride evolution in kilograms per metric ton of aluminum, due to particulate emissions resulting from volatilization can be approximated by equation 2.23, given by Haupin (6):
FVP = 2040 Pm + 2 Pd Pb
(2.23)
where Pb is the barometric pressure and Pm and Pd are the partial pressure of the monomer and dimer respectively, given by:
Pm =
1 + 4K P 1 T (kPa) 2K Pd = PT Pm (kPa)
(2.24) (2.25)
where K is the dimensionless equilibrium constant for dimerization, given by:
21414 K = exp + 15.6 Tb
(2.26)
12
From equations 2.19 and 2.23, the consumption of electrolyte constituents due to particulate emissions can be described as follows, in kilograms per metric ton of aluminum.
mNaF = 0.5525 FVP
(2.27) (2.28)
m AlF3 = 1.105 FVP
The calculated fluoride evolution due to particulate emissions resulting from volatilization as a function of alumina concentration and bath ratio at typical operational conditions is shown in figure 2.2.
12 kg of F- / ton of Al 11 10 9 8 7 Rbath = 1.15 Rbath = 1.20 Rbath = 1.25
5 6 Al2O3 Concentration (wt%)
Figure 2.2 Fluoride evolution due to particulate emissions as a function of alumina concentration and bath ratio.
As can be seen in figure 2.2, the fluoride evolution due to particulate emissions is closely related to the alumina concentration and bath ratio, increasing with lower concentration and bath ratio, as was the case for the hydrogen fluoride evolution. As will be seen later, this is contradictory to the production efficiency, resulting in a tradeoff between efficiency and the environmental impact of aluminum production.
2.5.3 EntrainmentofParticulateMaterials The mechanism when liquid droplets or solid particulates are entrapped in a flowing gas is known as entrainment. The transpiring anode gases have sufficient gas velocity for electrolyte droplets to be entrained. There will also be some entrainment of condensed phases in the anode gases. When the liquid electrolyte entrained in the anode gas cools it forms particulates. Haupin (6) developed an empirical model for estimating the rate of fluoride evolution due to entrainment, given by equation 2.29. The model was based on the theory that the difference between measured total fluoride losses and the calculated estimate of the gaseous fluoride evolution and the particulate emissions would be the result of electrolyte being entrained by the effluent gases.
FE =
where
2 1 17030 + 29800 Rb 13000 Rb + 67cAl2O3 %CE 173 0.3896 2 + 141.6 Rb
(2.29)
= Tb 1243
(2.30)
13
However, equation 2.29 should be used cautiously to predict the actual contribution of the entrainment to the total fluoride lost by effluent gases since the equation merely accounts for differences between measured values and the estimates produced by the gaseous fluoride and particulate emission models. The entrainment model will thus only be considered as a mean for correction. The total fluoride emission will thus become:
FT = FG + FVP + FE
(2.31)
Since Haupins model can only be considered as a mean for correcting the total amount of fluorides lost by entrainment, it cannot be used to account for the loss of other constituents making up the entrained electrolyte droplets (e.g. alumina). This will thus introduce a small error in the prediction of their dynamic concentrations. From the fluoride loss estimated with equation 2.29 and by assuming that the concentration of the slowly consumed constituents remains constant as before, the consumption of electrolyte constituents due to entrainment can be approximated by proportional calculations as follows, in kilograms per metric ton of aluminum.
mAlF3 = 1.473 cAlF3 FE f F1

m NaF = 2.210 c NaF FE f F1
(2.32) (2.33) (2.34)
mAl2O3 = cAl2O3 M f F1
where fF is the fraction of the electrolyte that is made up of fluoride compounds and M is the total mass change in the fluoride compounds. Assuming that alumina is the only constituent that is not a fluoride compound, fF can be expressed by equation 2.35.
f F = 100 c Al2O3
(2.35)
2.5.4 SulphurDioxide The major source of sulphur in the aluminum electrolysis process appears to be the anode carbon in the form of thiophene (C4H4S). To lesser extent, it is introduced into the electrolyte with the raw materials as an impurity. The anodic reoxidation of the sulphide to form sulphur dioxide may be illustrated by equation 2.36:
S + 2CO2 SO2 + 2CO
(2.36)
According to Burnakin, et al. (7), thiophene can take part directly in the electrochemical reduction of alumina, since the reactions that are theoretically possible require less operating potential than the alumina electrolysis, but so far the behavior of sulphates in the electrolyte is not well understood (4). Since the sulphur originates as an impurity and leaves the cell mainly by reoxidation, its presence will not be considered in the material balance model. It should though be noted that besides deteriorating the quality of the carbon anode, the presence of sulphur, represents a serious pollution hazard in aluminum electrolysis. Fortunately, there exist methods for removing it from the pot gases with very good efficiency. 14
2.6 AluminaFeeding
The alumina is generally fed to a cell in small batches, at pre-determined intervals, by first punching holes through the crust and then pouring the alumina into the electrolyte. A typical automated feeding system comprises a pneumatically-operated crust breaker and an ore bin capable of discharging a fixed volume of alumina. There are usually 3 to 5 feeders fitted to each cell and these are actuated in turn to minimize concentration gradients in the electrolyte. To reduce heat losses, a layer of alumina powder is periodically added on top of the crust and anodes, where it acts as thermal insulation. When breaking the crust, some of this powder falls into the electrolyte as well as alumina bounded in the crust. The total amount of alumina that enters the electrolyte during the break and feed process thus varies. The sequencing of alumina feeding is generally at pre-determined intervals to provide the base load of the alumina requirements. For a cell without a control system for automatic feeding, it is often preferred to slightly underfeed the cell to prevent alumina saturation and then at some intervals (typically at around 48 hour intervals), supplementary additions are made to balance the alumina concentration. When the concentration is to be balanced the alumina feeding is ceased until the anode effect occurs, which occurs at some specified alumina concentration (generally in the range of 1-2%). When the anode effect occurs, the optimum conditions can then be restored by adding sufficient amount of alumina in one or two large batches. Since the anode effect is an abnormal operating condition, decreasing both the energy efficiency and lifetime of the cell, modern cells are generally fitted with a control system for automatic feeding. The control system attempts to keep the alumina concentration within a specific range, by automatically varying the time between batch additions, basing its decisions on abnormalities in the voltage drop over the cell, which is measured in real-time. As will be described in more detail in chapter 3, the cell voltage is a function of the alumina concentration, with an optimal minimum as demonstrated in figure 2.3.
4.6 4.55 Voltage (V) 4.5 4.45 4.4 4.35 4.3 2 3 4 5 6 Alumina Concentration (wt%) 7 8
Figure 2.3 Cell voltage as a function of alumina concentration, if the anode effect is assumed to occur at 2%.
The cell voltage is generally the only parameter used for estimating the alumina concentration, so the control system is unable to determine if the measured voltage is on the right or left side of the minimum, i.e., if the voltage drop in figure 2.3 is measured as 4.4 V, it could either mean that the concentration is 2.4% or 7.3%. It is thus preferred to keep the alumina concentration on only one 15
side of the minimum. The reduction cells at Norurl are fitted with an automatically controlled alumina feeding system. At Norurl the control system attempts to keep the alumina concentration in a narrow band between the voltage minimum and the critical concentration where the anode effect occurs. Using some specific voltage set point, the control system then adjusts the frequency of additions in relation to the deviations from the set point. Hence, if the voltage gets too high, it indicates that the alumina concentration is too low and the frequency has to be increased and correspondently, if the voltage gets too low, the frequency has to be decreased. The frequency of the anode effect can thus be lowered by responding to low alumina concentration in time. However, under some conditions the cell is starved to induce the anode effect on purpose, i.e., to remove sludge buildup with a short burst of the intensive heat generation caused by the anode effect. At Norurl, the ore bins have capacity of 1.50 kg of alumina. With an average impurity content of 5.5% by mass, about 1.43 kg of aluminum oxide is added to the electrolyte in each batch. According to equation 2.12, the theoretical consumption rate of alumina in a cell operating at 180 kA is approximately 30 grams per second, which means that in order to provide the base load of the alumina requirements, the average interval between batches has to be around 48 seconds. With such frequent additions, the holes in the crust remain almost completely open between batches and the amount of crust that falls into the electrolyte during the break and feed process thus minimal. However, alumina can enter the electrolyte by other means than with the feeding process as the result of various operating disturbances. The most severe disturbances occur during an anode change where up to 300 kg of crust and cover material may fall into the electrolyte. It is thus a common practice to underfeed a cell immediately after an anode change to respond to such disturbances. An automatically controlled alumina feeding system can usually detect an anode change from abnormalities in the cell voltage and thus respond to it. The total amount of the excess alumina entering the electrolyte to the alumina fed from the ore bins cannot be predicted by any means but under normal operating conditions the balance between theoretical alumina consumption and alumina addition is relatively stable. For simplicity the alumina will thus be assumed to enter the electrolyte only during the break and feed process and other sources of alumina neglected.
2.7 AluminaDissolution
The density of particulate alumina is greater than that of the cryolite based electrolyte and the aluminum metal pad. However, where it remains as discrete particles it will remain suspended in the turbulent electrolyte where it dissolves quickly before it can sink through the metal pad. In contrast to this, if the particulate alumina is bounded by frozen bath, it will form agglomerates that will sink to the bottom and form sludge when the electrolyte remelts. During the sludge formation, the alumina is transformed into large platelets which have slower dissolution rate in the electrolyte. Similarly, when the crust is broken, the chunks of crust will dissolve slowly in the electrolyte, while new crust is formed. Because of various unpredictable dissolution mechanisms involved, it is very difficult if not impossible to model the actual rate of alumina dissolution. 16
Biedler (8) proposed a simplified method of modeling by assuming two dissolution mechanisms. The first mechanism models the alumina that dissolves quickly which includes all of the powder that came in contact with the bath and did not form agglomerates and the second mechanism models the portion of the alumina addition that dissolves more slowly. Both mechanisms were assumed to be governed by a first-order differential equation, given by equation 2.37, with the difference being the dissolution constant.
dC = kC dt
(2.37)
where C is the concentration of undissolved alumina and k the dissolution constant in seconds-1. According to Biedler, on average, about 75% of an alumina addition dissolves quickly with a dissolution constant of 0.099 s-1 and approximately 25% dissolves slowly with a dissolution constant of 5.010-3 s-1. According to this assumption, the total amount of alumina that has been dissolved in the electrolyte can be expressed by:
Q (t ) = fD1 (t ) + (1 f ) D2 (t )
(2.38)
where D1 and D2 are the alumina concentrations that dissolve quickly and slowly respectively and f is the proportion of the alumina that dissolves quickly (f 0.75). If it is assumed that the alumina is added at the times t = t1, t2, ..., the addition takes T seconds and the rates of the alumina additions m1, m2, ..., the total amount of alumina added to the electrolyte becomes:
M (t ) = m j p ( t t j ) t
j =1
(2.39)
where
1 0 t < T p (t ) = else 0
(2.40)
In this model, mjT is the total amount of alumina that falls into the electrolyte in the period between t = tj and t = tj + T, including both particulate alumina and the alumina bonded in the chunks of crust. mj is thus an unknown variable that has to be estimated from the size of the ore bins and the geometry of the hole in the crust. The two partial alumina concentrations in equation 2.38 can then be expressed by:
dDi = ki ( M (t ) Di (t ) ) , i = 1, 2 dt
(2.41)
where ki are the two dissolution constants. According to equation 2.38 and the dissolution constants proposed it takes about 46 seconds for 99% of the particulate alumina to dissolve in the electrolyte, while it takes about 15 minutes for the sludge and crust to dissolve. The time response of dissolution, when 1 kg of alumina is added to the cell in one batch is shown in figure 2.4.
17

1 Dissolved Al2O3 (kg) 0.8 0.6 0.4 0.2 0
100
200
300
400 500 Time (sec)
600
700
800
900
Figure 2.4 Time response of alumina dissolution, when 1 kg of alumina is fed to a reduction cell at t = 0s.
As can be seen in figure 2.4, the alumina has not been completely dissolved until about 15 minutes after the addition. This means that if alumina is fed too rapidly to the cell, the sludge on the bottom of the cavity accumulates. The presence of the sludge cannot be prevented, but precipitation of alumina can be minimized with a good feeding strategy.
2.8 CorrectiveFluorideAdditions
When using dry scrubbers for removing fluoride from the effluent pot gases, the fluoride losses are almost completely recycled to the cells. The recycled fluoride exists in the alumina bulk material mainly as aluminum fluoride, but also as sodium fluoride and cryolite. At Norurl reduction plant, approximately 98.5-99% of the fluorides lost by emission are recycled. The remaining 1% of fluoride that escapes the cleaning process and the fact that only one scrubber is used for the whole potline, has to be compensated for by direct additions. The amount of fluoride required for correcting the electrolyte composition is mainly determined by the so-called bath ratio Rb, which is the mass ratio of NaF to AlF3. Most of the electrolyte characteristics are closely related to the bath ratio, such as the volatility, melting point, electrical conductivity and density. Maintaining a stable bath ratio is thus of major importance in the alumina reduction process. Another obvious reason for corrective additions is to maintain sufficient electrolyte volume. To increase the ratio, sodium fluoride is added to the electrolyte in the form of sodium carbonate (Na2CO3), which reacts with aluminum fluoride to form sodium fluoride (NaF), according to the reaction:
3 Na2CO3 + 2 AlF3 6 NaF + Al 2O3 + 3CO2
(2.42)
As can be seen from equation 2.42, by adding sodium carbonate instead of adding NaF directly to the electrolyte, the bath ratio can be varied more with less addition, since the AlF3 is reduced at the same time. Sodium fluoride also enters the cell as the natural occurring impurity, sodium oxide (Na2O) in the feed material, which reacts with aluminum fluoride to form sodium fluoride:
3 Na2O + 2 AlF3 6 NaF + Al2O3
(2.43)
According to Grjotheim and Welch (2), the sodium dioxide makes up of around 0.3% of the feed material mass. Correspondently, the bath ratio can be decreased by adding aluminum fluoride, but
18
aluminum fluoride is however added directly. Adding aluminum fluoride also reduces sodium deposition, since excess AlF3 reacts with sodium to form cryolite, according to the reaction:
AlF3 + 3 NaF = Na3 AlF6
(2.44)
The bath ratio is generally a predefined optimal ratio, determined by the overall cell design. As previously mentioned, the composition of the electrolyte is determined in an onside laboratory from regularly taken samples. Since the fluorides are consumed slowly as a result of the recycling process, the bath ratio can be held close to the optimal value with regular corrective additions. At Norurl, the bath ratio is preferred to be close to 1.15.
2.9 TotalMaterialConsumptionandProduction
At Norurl, the only additives used in the electrolyte are aluminum fluoride and calcium fluoride, thus neither MgF2 nor LiF are used. Calcium fluoride enters the bath as an impurity in the feed material and its proportion is sufficient to maintain the CaF2 balance, thus no corrective additions are required. The calculated theoretical material consumption and production for a 24 hour period of one of the reduction cells at Norurl, operating at 970C and 182 kA are listed in table 2.1 and 2.2. The presence of other impurities than CaF2 and the fluoride recycling process are neglected.
Electrolyte constituent Al2O3 C AlF3 CaF2 Na3AlF6 Typical concentration (wt %) 3.0 10.0 6.0 81.0 Material Consumption (kg) 2601 489.7 38.21 < 0.50 20.18 Proportion of the original weight (%) 1378 3.57 6.07 < 0.13 0.40
Table 2.1 Calculated material consumption for a 24 hour period of a 182 kA reduction cell with 94.5% current efficiency.
Products Al CO CO2 HF NaAlF4
Net production (kg) 1385 125.6 1595 19.86 24.22
Table 2.2 Calculated material production for a 24 hour period of a 182 kA reduction cell with 94.5% current efficiency.
If the recycle process and the presence of sodium oxide in the feed material are taken into account, the fluoride consumption in table 2.1 can be reduced significantly. Apart from lowering the overall material cost and fulfilling emission standards, the material balance and hence the bath ratio become more maintainable. Theoretically, the net consumption of the fluorides come close to zero but in practice this does not hold, since the fluorides from the high number of reduction cells in the potroom are recycled with a single dry scrubbing system. The concentration of recycled fluorides in the feed material would thus be close to constant, while the consumption of individual cells can vary. 19
2.10 PropertiesoftheElectrolyte
The physical properties of the electrolyte are strongly influenced by its composition and temperature. In the following, expressions for various properties involved in modeling the aluminum reduction cell will be given.
2.10.1 Density Several empirical equations have been published for calculating the density of the molten electrolyte. The empirical relationship given by Haupin (9) was considered to be the one best suited for the electrolyte composition at Norurl, which is given by:
EL = 100 /
cNa3 AlF6
4
4 3.305 9.37 10 Tb 1.987 3.19 10 Tb + 0.094 cxsAlF3 c Al2O3 cCaF2 + + 4 2 3.177 3.9110 Tb + 0.0005 cCaF2 1.449 + 0.0128 cAl2O3
cxsAlF3
(kg/m3)
(2.45)
cMgF2 3.392 5.24 10 Tb 0.01407 cMgF2

4
3 cLiF 10 4 2.358 4.9 10 Tb
where Tb is the bath temperature in degrees Kelvin (K) and cxsAlF3 is the excess aluminum fluoride concentration (with respect to cryolite stoichiometry). The density of the electrolyte must be below the density of liquid aluminum, for it to float on top of the metal pad. To reduce the mixing between the electrolyte and the metal pad, it is thus desirable to have a maximum difference in density. From equation 2.45 and the typical values listed in table 2.1, the theoretical density of the electrolyte at 970C was found to be approximately 2173 kg/m3. For comparison, the density of molten aluminum is around 2375 kg/m3.
2.10.2 Viscosity The viscosity of the electrolyte influences several hydrodynamic processes in the cell such as the convection in the electrolyte, movement of metal droplets and dissolution and sedimentation of alumina particles. The viscosity of the electrolyte can be expressed by the following simplified equation, given by Grjotheim and Welch (2):
EL = 11.557 9.158 103 (Tb 273)

2 1.587 10 3 c xsAlF3 + 2.094 10 3 + 1.853 10 5 (Tb 1273 ) c xsAlF3 2 2.168 10 3 c Al2O3 + 5.925 10 3 1.938 10 5 (Tb 1273 ) c Al2O3
(mPas)
(2.46)
Despite the fact that the interactive effects of other fluorides composites present in the electrolyte are not considered, Grjotheim and Welch state that the equation gives the viscosity with accuracy between 1.0 and 1.5 percent. From equation 2.46 and the typical values listed in table 2.1, the theoretical viscosity of the electrolyte at 970C was found to be approximately 2.445 mPas. 20
2.10.3 AluminaSolubility The maximum amount of alumina that can be dissolved in the electrolyte depends on both the electrolyte composition and temperature. The maximum alumina solubility can be expressed by the following equation, given by Skybakmoen, et al. (10):
c
max Al2 O3
T 273 = A b 1000
(wt%)
(2.47)
where
2 A = 11.9 0.062 cxsAlF3 0.0031 cxsAlF3 0.20 cCaF2
0.50 cLiF 0.30 cMgF2 + B = 4.8 0.048 cxsAlF3 +
42 cxsAlF3 cLiF 200 + cxsAlF3 cLiF
(2.48)
2.2 c1.5 LiF 3 10 + cLiF + 0.001 cxsAlF3
(2.49)
When the concentration of alumina reaches the maximum alumina solubility, the electrolyte becomes saturated with alumina, and all excess alumina sinks to the bottom of the cavity, where it can form sludge. From equation 2.47 and the typical values listed in table 2.1, the theoretical maximum alumina solubility of the electrolyte at 970C was found to be approximately 8.57 wt%.
2.10.4 AluminumSolubility The tendency for metal to dissolve in the electrolyte is the primary cause for the lowering in cell current efficiency. The lower the metal solubility, the higher the resulting current efficiency will become. Hence the selection of the electrolyte composition is influenced by the need to ensure the metal solubility is minimized, while at the same time ensuring that the melting point is low, since the metal solubility is strongly dependent on temperature. Based on correlation of measurements, the aluminum solubility can be expressed by equation 2.50 (2):
max log ( C Al ) = 1.8251
0.2959 3429 0.0339 cAl2O3 max Rbath Tb cAl2O3
(2.50)
0.0249 cLiF 0.0241 cMgF2 0.031 cCaF2
From equation 2.50 and the typical values listed in table 2.1, the theoretical maximum aluminum solubility of the electrolyte at 970C was found to be approximately 0.037 wt%.
2.11 CurrentEfficiency
As previously explained, the internal recombination of aluminum and carbon dioxide is the main cause of the difference between the actual quantity of aluminum produced and the theoretical quantity stated by Faradays law of electrolysis. The ratio of the actual quantity and the theoretical quantity is referred to as the Current Efficiency, abbreviated by %CE if represented as a 21
percentage. It is obvious that the decrease in current efficiency makes a significant contribution to the excess energy consumption, and it is hence desirable to have the efficiency as high as possible. In industrial cells the current efficiency is affected by a number of variables, but the solubility of the metal in the bath and the mass transfer conditions within the electrolyte play the most important roles. The current efficiency can be expressed by the following relationship (11):
r %CE = 100 1 m rAl
3F rm = 100 1 I
(%)
(2.51)
where rm/rAl is the ratio of the metal transfer through the bulk of the electrolyte and the theoretical rate of aluminum production, F is the Faradays constant and I the cell current. The rate of transfer of the dissolved metal away from the boundary layer between the electrolyte and metal can be calculated using equation 2.52 (2):
max rm = km A cAl (1 f )
(mol/s)
(2.52)
where A is the surface area of the boundary layer, is the maximum solubility of aluminum in the electrolyte and f is a factor relating the concentration of dissolved metal in the bulk electrolyte to the maximum solubility. km is the mass transfer coefficient of the metal, which can be derived from the dimensionless Sherwood number, given by:
Sh = km
l D
(2.53)
where l is the interpolar distance and D is the diffusivity of dissolved metal. For fluids of high viscosity with Reynolds number > 2300, the Sherwood number can be expressed by the Nusselt number for turbulent flow conditions in terms of the Reynolds and Schmitt numbers:
Sh = Nu = 0.023 Re5/6 Sc1/3
(2.54)
Substituting the Sherwood number into equation 2.53, and solving for km, the mass transfer coefficient of the metal becomes:
km = 0.023
D Re5/6 Sc1/3 l
(m/s)
(2.55)
Then by substituting equation 2.52 and 2.55, into equation 2.51, and expressing the dissolved metal concentration as a weight percentage, the theoretical current efficiency can be predicted by the equation:
CE % = 100 2.2 105
A D2/3 max 3/2 5/6 cAl (1 f ) 1/2 1/6 I l
(%)
(2.56)
where , and v are the viscosity, density and interfacial velocity of the electrolyte respectively. The factor f and the interfacial velocity are difficult to predict, but some estimates have been proposed in the literature. At Norurl, the factor f will be assumed to be 0.113 (11) and the
22
interfacial velocity 0.15 m/s (2). Furthermore, the diffusivity of dissolved metal will be assumed to be 310-8 m2/s as proposed by Grjotheim et al. (2). Predicting the current efficiency from equation 2.56 has limitations not only due to the assumptions that have to be made, but also because secondary effects frequently change parameters such as the metal velocity. However, for cells with less turbulent metal pads, such as the ones at Norurl, the predictions are considered reasonable. From equation 2.56 and the typical properties of the electrolyte previously calculated, the theoretical current efficiency at Norurl was found to be in the range of 94.5-96.0 %, depending on the interpolar distance.
2.11.1 ElectricalConductivity In the industrial production of aluminum, it is desirable to have an electrolyte with as high electrical conductivity as possible, so that the ohmic voltage drop through the electrolyte is minimized. Many empirical relationships for the change in the electrical conductivity have been published, which are generally obtained from a multiple regression analysis. Hves (12) proposed an equation for determining the electrical conductivity of a multi-component bath, which can be described by:
= 7.22 e1204.3/T 2.53 xAl O 1.66 xxsAlF 0.76 xCaF 0.206 xKF + 0.97 xLi AlF 6
b 2 3 3 2 3
1.07 xMgF2 1.80 x Al2O3 xCaF2 2.59 x Al2O3 xMgF2 0.942 xxsAlF3 xLi3 AlF 6
(2.57)
where x represents the mole fractions of the corresponding additions and the conductivity is in Siemens per centimeter (S/cm). A simplified relationship was also proposed by Hves, where the contributions of the constituents are represented by mass fractions instead of mole fractions:
ln ( ) = 1.977 0.0200 cAl2O3 0.0131 cxsAlF3 0.0060 cCaF2

0.0106 cMgF2 0.0019 cKF + 0.0121 cLiF 1204.3 T b
(2.58)
Equation 2.58 is less precise than equation 2.57, since the cross products in equation 2.58 are not included. According to Hves, the maximum relative error is less than 2.5% though. Several relationships have been published by various researchers, but the one proposed by Hves has been found to be the most accurate one to date (8). From equation 2.58 and the typical values listed in table 2.1, the theoretical electrical conductivity of the electrolyte at 970C was found to be approximately 2.18 S/cm.
2.11.2 HeatCapacity The heat capacity of the electrolyte depends on both the bath temperature and composition, but an expression with relation to either the temperature or composition was not found in the literature. Based on the heat capacity of each of the electrolyte constituents used at Norurl as well as LiF and MgF2, an overall heat capacity expression can be derived by assuming that each constituents contribution to the overall heat capacity is proportional to its concentration. The heat capacity of
23
each constituent can be expressed by the Shomate equation for heat capacity, given by equation 2.59:
C p = A + Bx + Cx 2 + Dx3 + E / x 2
where
x = Tb / 1000
(2.59) (2.60)
The five coefficients; A, B, C, D and E are known as the Shomate coefficients of a substance, listed in tables A.3 and A.4 in the appendix. By combining the heat capacity of each constituent into an overall heat capacity expression, where each constituents participation is proportional to its concentration, the total heat capacity of the electrolyte composition at Norurl can be expressed by equation 2.61, as a function of both bath temperature and composition:
CEL = (1.202 + 9.5 105 Tb ) cMgF2 + ( 2.89 103 Tb 2.194 ) c Al2O3 + ( 0.7686 + 6.35 10 4 Tb ) cNaF + (1.082 + 1.2 10 4 Tb ) c AlF3
+ ( 0.7129 + 4.4 104 Tb ) cCaF2 + 2.474cLiF / 100
(J/gK)
(2.61)
where Tb is the bath temperature. From equation 2.61 and the typical values listed in table 2.1, the theoretical heat capacity of the electrolyte at 970C was found to be approximately 1.767 J/gK. Taylor, et al. (13) state that the heat capacity of a multi-component electrolyte is 1.657 (J/gK), so equation 2.61 can be considered as a reasonable approximation to the actual value.
2.12 EnergyConsiderations
The aluminum reduction process requires energy to keep the electrolyte at the reaction temperature and to produce aluminum, which is furnished as electric energy. In the following, the theoretical energy requirements for the electrochemical reactions and heating up the raw material will be considered.
2.12.1 EnthalpyofReaction The dominant overall cell reaction that occurs at cell temperature is the primary reaction:
2 Al2O3 + 3C = 4 Al + 3CO2
(2.62)
Using the assumption that the loss in current efficiency is only due to the reoxidation of aluminum and carbon dioxide, the reaction can be represented by the modified electrolysis equation:
2 Al2O3 + 3C = 4 Al + 3 ( 2 1) CO2 + 6 (1 ) CO
(2.63)
By solving out for Al, the energy required for producing one mol of aluminum can then be expressed in terms of the enthalpy of formation of both the reactants and the products:
1 3 1 3 1 H react = H f ( Al2O3 ) 2 H f (CO2 ) 1 H f (CO) 2 4 2
(kJ/mol)
(2.64)
24
The enthalpies of formation of the main substances involved in the overall aluminum reduction process are listed in table A.1 in the appendix for various temperatures. By interpolating linearly between the temperatures 1100K and 1300K, the formation energies for the primary reaction can be determined as follows. It should be noted that the formation energy for a pure substance is zero.
H f ( Al2O3 ) = 1704.8 0.011235 Tb
(kJ/mol)
(2.65) (2.66) (2.67)
H f ( CO2 ) = 392.53 0.0021 Tb H f ( CO ) = 105.55 0.00640 Tb
(kJ/mol) (kJ/mol)
where Tb is the bath temperature. From equation 2.64, the energy consumed by the primary reaction in a cell operating with 94.5% current efficiency and Tb = 970C, was found to be about 560 kJ/mol. The enthalpies of all other reactions considered in the material balance model are listed in table 2.3, determined in the same manner as for the primary reaction.
Chemical Reaction Enthalpy of Reaction [kJ/mol]
Na3 AlF6 (l ) + 3 2 H 2 ( g ) Al (l ) + 3 NaF (l ) + 3HF ( g )
H react = 793.26 + 0.1258 Tb
2 AlF3 ( s ) + 3H 2O ( g ) Al2O3 ( s ) + 6 HF ( g ) NaF + AlF3 NaAlF4 S + 2CO2 SO2 + 2CO

3 Na2CO3 + 2 AlF3 6 NaF + Al2 O3 + 3CO2 AlF3 + 3 NaF Na3 AlF6 3 Na2O + 2 AlF3 6 NaF + Al2O3
H react = 324.03 + 0.07591 Tb H react = 895.21 + 0.93468 Tb H react = 210.9 + 0.00748 Tb

H react = 24.784 + 0.14984 Tb H react = 169.3 + 0.09972 Tb H react = 844.58 0.18601 Tb
Table 2.3 The enthalpy of the reactions considered in the material balance model
As can be seen in table 2.3, most of the reactions are endothermic at temperatures above 900C, except the production of NaAlF4 and the reaction of sodium carbonate with aluminum fluoride, which are exothermic.
2.12.2 EnthalpyofHeating When the reactants and bath additives are added to the cell, they must be heated from their initial temperature to the bath temperature Tb, absorbing heat from the electrolyte. Each substance must thus undergo an enthalpy change, which can be expressed by the Shomate equation for an enthalpy change, given by:
[ H H 298.15 ] = Ax + Bx2 / 2 + Cx3 / 3 + Dx4 / 4 E / x + F H

where
x = Tb / 1000
(kJ/mol)
(2.68)
(2.69)
The seven Shomate coefficients of each of the constituents involved in the reduction process are listed in tables A.3 and A.4 in the appendix. In equation 2.68, the substances are assumed to be 25
initially at room temperature (298.15K), which is a reasonable assumption for all substances that are added directly to the cell, such as the carbon and the corrective additives. The alumina on the other hand is used as a gas absorbent before entering the bath, and is thus preheated, reaching an initial temperature close to the ambient temperature inside the cell hood, which is around 350K. The enthalpy change of the alumina can be determined, by solving the Shomate equation for both the initial temperature and the elevated temperature and subtracting the values of the solutions. The particulate materials entering the bath will be assumed to reach the bath temperature almost instantly, due to the large surface area of the particulate bulk and the high thermal conductivity of the electrolyte. Adding a considerable amount of particulate material will thus result in an instant drop in the bath temperature. For simplicity, the carbon will be assumed to absorb heat proportional to its consumption rate.
2.12.3 EnthalpyofDissolution The dissolution of alumina in the electrolyte is endothermic. The enthalpy change is proportional to the solvability of the electrolyte and increases with higher concentration of alumina. The energy required to convert alumina from its gamma- (undissolved) to alpha (dissolved) phase can be determined from the associated enthalpy change as follows:
H = ( H f ( ) H f ( ) ) c (%)
(kJ/mol)
(2.70)
Hf() : Heat of formation for alumina in alpha phase (-1675.69 kJ/mol) (4) Hf() : Heat of formation for alumina in gamma phase (-1656.86 kJ/mol) (4) c(%) : Concentration of gamma alumina (wt%).
2.12.4 NetEnergyRequirements For a reduction cell at Norurl operating at 94.5% current efficiency and 180 kA nominal current, the theoretical net energy requirements for 24 hour operation were found to be as listed in table 2.4, excluding heat losses and the effects of corrective additions.
930C Reactions Primary Evolution Vaporization Heating Alumina Carbon Dissolution Alumina Total Energy Mean Power 2.867107 kJ 1.534104 kJ -2.213104 kJ 2.486106 kJ 5.799105 kJ 1.044106 kJ 3.277107 kJ 379.3 kW Bath Temperature 950C 970C 2.868107 kJ 1.710104 kJ -3.379104 kJ 2.548106 kJ 5.818105 kJ 1.044106 kJ 3.284107 kJ 380.1 kW 2.868107 kJ 1.899104 kJ -5.132104 kJ 2.611106 kJ 5.837105 kJ 1.044106 kJ 3.289107 kJ 380.7 kW 990C 2.868107 kJ 2.101104 kJ -7.777104 kJ 2.673106 kJ 5.855105 kJ 1.044106 kJ 3.293107 kJ 381.13 kW
Table 2.4 Theoretical net energy requirements of the reduction process at various operating temperatures.
As can be seen in table 2.4, approximately 87% of the energy is dissipated by the primary reaction, while most of the remaining energy goes into heating the raw material. 26
2.13 TheModelConstruction
Based on the previous studies of the mechanisms of the material balance and the electrolyte properties in an aluminum reduction cell, a material balance model was constructed in Simulink. The model is mainly made up of three subsystems: production and consumption, material input and the electrolyte properties as illustrated in figure 2.5. The current efficiency block is a part of the electrolyte properties subsystem, but is shown independently for illustrating its relationship to the internal model variables. Block diagrams of all the subsystems are given in the appendix.
Figure 2.5 Block diagram of the material balance model.
2.13.1 Inputs The Simulink model has six inputs, listed in the table below:
No 1 2 3 4 5 6 7 Name I TBath Al2O3 Feed Na2CO3 Feed AlF3 Feed IPD dbub Description Operating current Bath temperature Aluminum oxide feed Sodium carbonate feed Aluminum fluoride feed Interpolar distance Thickness of the bubble layer Unit A K kg kg kg m m Provided by Control input Thermal balance model Control input Control input Control input Control input Cell voltage model Comments Chapter 4 Accumulative Accumulative Accumulative Chapter 3
Table 2.5 Inputs to the material balance model in figure 2.5
2.13.1.1 OperatingCurrent The operating current of a single cell cannot be varied, since all the cells in a potline are connected in series. The high energy requirements of the reduction plant, puts large strains on the power grid which may introduce some imbalances in the operating current in conjunction with disturbances introduced by the high-voltage rectifiers. Under stable conditions, the operating current can though generally be assumed constant.
27
2.13.1.2 MaterialFeeding The three feed inputs require that the input source is the accumulated amount of material that is fed to the cell. The process of feeding one batch of a material to the cell can be modeled by giving the corresponding model input a discrete pulse, integrated over time, as illustrated by the Simulink model in figure 2.6.
Figure 2.6 Signal generator for the material feed inputs of the material balance model.
The amplitude of the pulse is then the mass rate of the feeding process and the pulse width the time the feeding takes. By integrating the pulse input, the input to the material balance model then becomes stepwise accumulation of the total amount of material fed to the cell as demonstrated in figure 2.7.
6 5 Material Fed (kg) 4 3 2 1 0 0 0.5 1 1.5 2 2.5 Time (min) 3 3.5 4 4.5 5
Figure 2.7 Stepwise accumulation of the total amount of material fed to the cell when 1 kg of material is fed every 1 minute.
2.13.1.3 Interpolardistance The interpolar distance is controlled by an existing black box control system at Norurl. Most control systems currently used by the aluminum industry are based on theories that are undisclosed by the companies, so limited information can be gathered about their functionality. Some cabinets even have a seal, which when broken, power downs the control hardware, erasing the control program from memory. One of the primary objectives of the control system is to hold the interpolar distance constant, thus responding to i.e. metal tapping. The input to the material balance model corresponding to the interpolar distance can thus generally be assumed constant. 2.13.1.4 OtherInputs The bubble layer thickness and the bath temperature are provided by the cell voltage model and the thermal balance model which are the subjects of chapters 3 and 4 respectively.
28
2.13.2 Outputs The model in figure 2.5 has nine outputs. Seven of these outputs are required by the cell voltage model and the thermal balance model, but the other two can be used for tracking the productivity and emissions from the cell. The nine outputs are listed in the table below.
No
1 2 3 4 5 6 7 8 9
Name
mProd QCons cAl2O3 cAl2O3max cLiF %CE RBath KEL TL
Description Material production (Matrix) Energy consumption Concentration of aluminum oxide Solubility of aluminum oxide Concentration of lithium fluoride Current Efficiency Bath ratio Electrical resistivity of the electrolyte Liquidus temperature
Unit kg kJ wt% wt% wt% % S/cm K
Required By Thermal balance model Cell voltage model Cell voltage model Cell voltage model Cell voltage model Cell voltage model Thermal balance model
Comments Accumulative Accumulative
Chapter 4
Table 2.6 Outputs from the material balance model in figure 2.5.
The liquidus temperature is the temperature at which the electrolyte begins to freeze. At temperatures above the liquidus temperature the electrolyte is completely molten but below the liquidus temperature the electrolyte consists of both solid and liquid phases. It is one of the most important parameters in the thermal balance model, but to minimize the connection paths between models, it was included in the material balance model for model simplicity. An expression for determining the liquidus temperature will be given in chapter 4.
2.14 ModelValidation
In order to validate that the model constructed represents the physical system to a sufficient level of accuracy requires continuous measurements of dynamic changes in the electrolyte composition but currently the electrolyte composition cannot be continuously analyzed with any known methods. In recent years, a variety of in situ sensors have been developed for continuous measurement of the alumina concentration but none of them have been proven to be viable for long term exposure to the corrosive nature of the electrolyte. The modeled dynamic variations in the concentration of the electrolyte constituents could thus not be validated. The model can however be validated to some extent by comparing the theoretical current efficiency to measurement data. The current efficiency is strongly influenced by the physical properties of the electrolyte so an accurate estimation of the current efficiency should signify the reliability of the estimates and assumptions involved. In the following the modeled current efficiency will be validated with the measurement data available.
2.14.1 CurrentEfficiency The metal product is tapped from the cells at Norurl on a daily schedule by siphoning the metal directly into an insulated crucible. Weighing this crucible enables the amount of metal removed to be monitored. In conjunction with the tapping process, the height of the metal level is measured to estimate the amount of aluminum that has been produced since the last tapping. The current
29
efficiency can then be determined by calculating the ratio between the theoretical amount of aluminum produced with respect to the supplied current and the measured production, according to:
%CE = 0.3356 I T M 1 100

|I| : Mean cell current between tappings (kA) T : Time between tappings (h) M : Measured metal production (kg)
(2.71)
Four reduction cells were chosen for the comparison which had all been operating for a week without any major disturbances, such as reduced metal tapping, current drop, thermal instabilities or the anode effect. The composition of the electrolyte in each cell was determined from samples taken from the electrolyte at the beginning and at the end of the week and the bath temperature measured daily. The average operating conditions were then used to calculate the theoretical average current efficiency of each cell over the week in question. A comparison between the measured average current efficiencies and the theoretical values calculated from equation 2.56 is listed in table 2.7. The interpolar distance is generally unknown, so the consistency had to be evaluated at different values within the typical range (4-5 cm).
Cell CELL 1 CELL 2 CELL 3 CELL 4 Operating conditions (Measured) I (kA) Tb (C) cAlF3 (%) cCaF2 (%) 181.6 971.8 9.75 5.90 181.9 975.0 9.50 6.00 182.1 973.1 10.50 5.70 209.7 981.0 7.50 6.00 %CE (Theoretical) 4.0 cm 4.5 cm 5.0 cm 95.05 95.15 95.24 95.03 95.13 95.21 94.87 94.97 95.06 95.65 95.74 95.81 %CE (Measured) 95.23 94.85 94.70 95.70 Max Error (%) 0.192 0.380 0.380 0.115
Table 2.7 Comparison between calculated and measured current efficiency. The theoretical values that are closest the measured current efficiencies are shaded with green color.
As can be seen in table 2.7, the theoretical current efficiencies were found to be quite accurate estimates of the measured values. The maximum error in the estimates was found to be 0.38 % which corresponds to that cell 2 and 3 produced each only 5 kg more aluminum a day than predicted if the interpolar distance in these cells was 5 cm. This shows that the modeled current efficiency and consequently the estimates of the electrolyte properties involved can be considered reasonably accurate.
2.15 Summary
The physical properties of the electrolyte are strongly influenced by its chemical composition. Some of the properties such as viscosity, solvability of alumina and liquidus temperature define the operating limits of the reduction cell so maintaining the concentration of the electrolyte constituents within limits is essential for stable and prolonged operation. The fluorides that make up the electrolyte are consumed relatively slowly over time and it is thus possible to predict the requirements for corrective additions with intermittent analysis of samples taken from the electrolyte. The alumina is however consumed very rapidly and maintaining the concentration of alumina in the electrolyte within limits requires frequent additions. The alumina is generally fed to 30
a cell in small batches of fixed size but alumina can also enter the electrolyte by other means as the result of various operating disturbances, which makes it difficult to predict variations in the alumina concentration. Real-time measurements of the electrolyte composition have not yet been proven viable, but valuable process control strategies can be based on a relationship that exists between the alumina concentration and the cell voltage drop. In this chapter the most relevant aspects of the material balance in an aluminum reduction cell were studied, which was used as a basis for modeling dynamic changes in the electrolyte composition, properties and energy requirements. The theoretical current efficiency was compared to measurement data and was found to provide quite accurate estimates of the physical values. The modeled dynamic variations in the concentration of the electrolyte constituents could however not be validated since the electrolyte composition cannot be continuously analyzed with any currently known methods.
31
3 CellVoltageModel
3.1 Introduction
The total voltage drop over a reduction cell is referred to as the cell voltage and is measured between the busbars connected to the electrodes. Although it is usually not a design parameter, the cell voltage is one of the most important operating parameters in aluminum reduction cells, and is often the only parameter measured in real-time. The cell voltage is closely related to both the alumina concentration in the electrolyte and the bath temperature and is thus generally, one of the bases for computer control of the Hall Hroult process. The total cell voltage is composed of three different types of contributions:

The decomposition voltage. (Theoretical minimum potential required for the decomposition of alumina) The overvoltage (The excess voltage due to electrode polarization) The ohmic voltage drops (Voltage drops due to the resistance of various sections in the cell)
The decomposition voltage for alumina is around 1.1-1.2V, but due to electrode polarization and the electrical resistance of various sections of the cell, the cell voltage is typically in the range of 4.1-4.5 volts. The difference between the theoretical energy requirements and the energy generated by the electric current is manifested as heat, so the primary cause for the low energy efficiency of the Hall-Hroult process is the high cell voltage. The reduction cell must operate in thermal balance, so the cell voltage cannot be dramatically decreased without accounting for the change in balance by either a change in design or operating conditions. Under normal operating conditions, the cell voltage does not deviate far from the optimal cell voltage for the given cell design. However, when the anode effect occurs, which is an abnormal operating condition, the cell voltage can reach up to 80 V, which is the primary cause of sidewall erosion as well as lowering the cells production efficiency. In this chapter the various voltage contributions will be studied, which will then be used as a basis for constructing a cell voltage model. The dynamics of the current distribution in the cell are a lot faster than that of the material and thermal balance models, so for simplification; the cell voltage model will be considered steady-state. Since the cell production efficiency is directly related to the cell voltage, the voltage model can be used as an aid to determine the optimal alumina concentration.
3.2 TheDecompositionVoltage
Energy is the driving force for reactions. The tendency for a reaction to reach equilibrium is driven by the so called Gibbs free energy, which comes from the enthalpy and entropy of reaction in the
32
system. The change in Gibbs energy, G is defined by the Gibbs Helmholtz equation, in terms of enthalpy and entropy changes, H and S, at temperature T by:
G = H T S
(kJ/mol)
(3.1)
The Gibbs energy for a chemical reaction can be calculated by summing the free energies of formation of the products and subtracting the free energies of formation of the reactants. For the aluminum reduction process the primary reaction is described by:
2 Al2O3 + 3C = 4 Al + 3CO2
(3.2)
The Gibbs free energies of the relevant compounds are tabulated in the JANAF Thermochemical Tables (14). For the primary reaction, the relevant energies are given in table 3.1, relative to the temperature range of the electrolysis.
Gf (kJ/mol)
Temperature (K) Aluminum oxide (Al2O3) Carbon (C) Aluminum (Al) Carbon dioxide (CO2) 1100 1200 1300
-1313.990 0 0 -396.001
-1281.757 0 0 -396.098
-1249.668 0 0 -396.177
Table 3.1 - The Gibbs free energy of formation of the compounds in the primary reaction equation, taken from (14)
By interpolating linearly between the temperatures 1100K and 1300K, the free energies of the products and reactants can be determined as follows:
G f ( Al2O3 ) = 1691.4 + 0.33011 T kJ / mol
(3.3)
(3.4)
G f (CO2 ) = 395.03 8.8 104 T
kJ / mol
The Gibbs free energy for the electrolysis process then becomes:
G f = 1.5 G f (CO2 ) G f ( Al2O3 ) = 1098.9 0.33143 T
kJ / mol
(3.5)
Since the Gibbs energy is positive, it indicates that the reaction is endothermic, thus the reaction requires energy to progress. During operation of a reduction cell, electrical energy is transformed into chemical energy. The electrical potential required to provide enough Gibbs energy for the electrochemical decomposition at standard conditions is referred to as the standard electrode potential, abbreviated E0. The relationship between the Gibbs free energy and the standard potential can be expressed by equation 3.6:
G = ne E 0 F
(kJ/mol)
(3.6)
where ne is the number of electrons exchanged in the chemical reaction (for the principal reaction, ne = 6) and F is the Faraday constant (F 96485 C/mol). For the primary reaction in equation 3.2, with the Gibbs free energy determined by equation 3.5, the standard electrode potential becomes:
E0 = G 1098.9 0.33143 T = = 1.898 0.0005725 Tb ne F 6 96.485
(V)
(3.7)
33
The standard electrode potential is given for standard conditions, which is when all reactants and products are pure or saturated at atmospheric pressure. For the primary reaction in the HallHroult process, the carbon, aluminum and carbon dioxide are nearly pure phases and can therefore be assumed to be in their standard states, but alumina is in its standard state only when it is at saturation. This means that an aluminum reduction cell generally operates under non-standard conditions with respect to the electrolyte composition. When the reactant concentrations differ from standard conditions, the cell potential will deviate from the standard potential. Given the standard potential of the electrochemical decomposition, the cell potential at non-standard effective concentrations can be calculated using the Nernst equation, given by equation 3.8 where R is the universal gas constant (~8.314 JK-1mol-1) and Q is the reaction quotient. The potential is referred to as the equilibrium or decomposition voltage, abbreviated by E0.
E0 = E R Tb ln ( Q ) ne F
(V)
(3.8)
The reaction quotient is taken at a particular instant in time, not necessarily the moment when equilibrium is reached. For a reaction in chemical equilibrium, the reaction quotient in equation 3.8 can be replaced with the equilibrium constant, K, which may be defined as:
A + BK = S + T
K= k+ {S } {T } L = k { A} {B} L

(3.9) (3.10)
where A, B, S and T are the chemical species involved in the reaction and , , and the corresponding stoichiometric coefficients. The terms in braces represent the activity of the species. For the primary reaction, the decomposition voltage can thus be written as:
4 3 a Al aCO2 R Tb ln 2 E0 = E 3 a Al O aC ne F 2 3 0
(V)
(3.11)
where Tb is the bath temperature and the braces have been replaced by the symbol a, to represent the activity of the corresponding species. Since aluminum, carbon dioxide and carbon are close to their standard states, unit activity is assigned. According to Haupin (15) the relation for the activity of alumina is as follows:
aAl2O3 = 0.03791 RS + 2.364 RS 2 2.194 RS 3 + 2.8686 RS 4

where RS is the relative alumina saturation, given by:
max RS = cAl2O3 cAl2O3
(3.12)
(3.13)
Variations in most of the operating parameters, such as current density and interpolar distance, will not influence the expression in equation 3.11, but the two main dependent parameters are the bath temperature and alumina concentration. The relationship between the decomposition voltage and the alumina concentration at different bath temperatures is shown graphically in figure 3.1. 34
1.25 1.2 Voltage (V) 1.15 1.1 1.05 1 TBath = 1200K TBath = 1250K TBath = 1300K 1 2 3 4 5 Alumina Concentration (wt%) 6 7 8
Figure 3.1 The relationship between the decomposition voltage and alumina concentration, at different bath temperatures.
As can be seen in the figure, the decomposition voltage increases with higher alumina concentration and lower bath temperature, hence more energy is required for the reaction to take place. Even though the primary reaction is the most favorable one, numerous of other possible reactions can take place during cell operation. The following are the most common reactions with the corresponding reaction potentials given by Haupin (6).
Al2 O3 + 3C 3CO + 2 Al
2 Na3 AlF6 + Al2O3 + 3C 3COF2 + 6 NaF + 4 Al
: E0 = 1.02 V : E0 = 1.80 V : E0 = 2.55 V : E0 = 4.41 V
(3.14) (3.15) (3.16) (3.17)
4 Na3 AlF6 + 3C 3CF4 + 12 NaF + 4 Al 2 Na3 AlF6 6 NaF + 2 Al
The first reaction listed above (eq. 3.14) is the secondary reaction in the aluminum reduction process. The secondary reaction requires less Gibbs energy than the primary reaction, but is unfavorable because it has lower production rate than the primary reaction. The remaining reactions require more Gibbs energy than is provided, and can thus be avoided under normal operation. However, when the anode effect occurs, the cell voltage can become sufficiently high to provide more than enough Gibbs energy for the reactions to take place. For simplicity, the three reactions in equations 3.15 to 3.17 will be neglected in the cell model.
3.3 Overvoltage
The decomposition voltage derived in the previous section only describes the theoretical minimum electrical potential required to produce aluminum with electrochemical decomposition of alumina. The excess voltages due to polarization at the electrodes interfaces have to be considered, to determine the total voltage required for the decomposition. This includes overvoltages due to concentration gradients around the electrodes as well as a reaction overvoltage close to the surface
35
of the anodes. The three overvoltages are all derived from the Nernst equation and are proportional to the electrolyte temperature.
3.3.1 ConcentrationOvervoltage During the electrochemical reactions, the reactants around the electrodes are consumed, thus lowering the concentration in that area. The random motion of molecules causes a net transfer of mass from an area of high concentration to an area of low concentration, which balances the concentration. In an aluminum reduction cell, the electrochemical reactions are more rapid than the mass transport of the reacting species, so a concentration gradient is built up around the electrodes. The decomposition voltage is determined at zero current flow, with the concentration of the bulk solution as a reference point, which means that a concentration gradient would result in a potential difference. The potential difference, which is generally known as the concentration overvoltage or polarization, is necessary to make the electroactive species migrate against the gradient concentration. 3.3.1.1 AnodeConcentrationOvervoltage The anode concentration overvoltage occurs at the boundary between the anodes and the electrolyte. Based on the Nernst equation, Haupin (15) introduced an equation for determining the anode concentration overvoltage, given by equation 3.18:
AC =
jcr R Tb ln ne F jcr j A
(V)
(3.18)
where jA is the anode current density and jcr is the so-called critical current density, which is the maximum current density attained before the normal anode reaction is superseded by the anode effect, given by equation 3.19 (15). The number of electrons exchanged in the chemical reactions is two, resulting in ne = 2 in equation 3.18.
jcr =
where
(C c
a
(C
2 cAl2O3 + Cb cAl2O3 AE Al2O3
+ Cb c
AE Al2O3
) j ))
2
(A/m2)
(3.19)
Ca = 1.443 1.985 Rb + 1.131 Rb2

Cb = 0.4122 0.2037 Rb
(3.20) (3.21)
Assuming that the anode effect occurs when the alumina concentration gets close to 2 wt%, the relationship between the alumina concentration and the anode concentration overvoltage will be as shown in figure 3.2, at different bath ratios.
36

0.3 0.25 Voltage (V) 0.2 0.15 0.1 0.05 0 2 3 4 5 6 Alumina Concentration (wt%) 7 8 RBath = 1.15 RBath = 1.40 RBath = 1.65
Figure 3.2 The relationship between the anode concentration overvoltage and alumina concentration, at different bath ratios.
As can be seen in figure 3.2, the anode concentration overvoltage is quite small under normal conditions, but goes towards infinity when the alumina concentration gets close to the critical concentration (here 2 wt%). However, large variations in the bath ratio have little effect on the overvoltage, if compared to the concentration. The anode concentration overvoltage is one of two potentials that contribute to the high cell voltage drop under the anode effect, the other one being the bubble voltage drop, described in a subsequent section.
3.3.1.2 CathodeConcentrationOvervoltage To migrate against concentration gradients at the boundary between the electrolyte and the metal pad cathode, a potential difference occurs at the boundary, which can be expressed by the cathode concentration overvoltage, CC, given by (16):
CC =
R T j 1.375 0.125Rb ln C F ne 1.5 0.283
(V)
(3.22)
where jC is the cathode current density and ne = 2. The relationship between the bath ratio and the cathode concentration overvoltage is shown in figure 3.3.
0.031 0.03 Voltage (V) 0.029 0.028 0.027 1 1.1 1.2 1.3 1.4 1.5 Bath Ratio 1.6 1.7 1.8 1.9 2
Figure 3.3 - The relationship between the cathode concentration overvoltage and the bath ratio.
37
As can be seen in figure 3.3, the cathode concentration overvoltage is normally small and is close to being linearly dependant on the bath ratio. Unlike the anode concentration overvoltage, the cathode concentration overvoltage is independent of the critical alumina concentration.
3.3.2 AnodeReactionOvervoltage For homogeneous reactions, the reaction rate depends on the concentration of the reactants in solution. If the reactants are generated by a different reaction which precedes the electron transfer, the concentration of the reactants will be governed by the rate of this reaction. If an electrode reaction is preceded by such a rate-determining step, this will affect the potential at which the reaction occurs. The potential difference is generally known as the reaction overvoltage. Almost all published reaction mechanisms derived from the electrode polarization indicate that the mass transfer of the oxygen containing anion in the primary reaction becomes the ratedetermining step at the anode surface in aluminum reduction cells. The first step is dissociation of the oxygen-carrying complex, exemplified by the equation (2):
2 F + Al2 OF62 = O 2 + 2 AlF4
(3.23)
This is followed by discharge of the oxyanion to form bonded chemisorbed intermediate oxide (2):
O 2 O ( ad ) + 2e O ( ad ) + xC C x O
(3.24) (3.25)
Because of the stirring of gas bubbles, this happens at higher current density than would be expected theoretically. Haupin (15) introduced an equation for determining the reaction overvoltage appearing at the anode surface, given by equation 3.26:
AR = p
p=
j R T ln A jrl ne F
(V)
(3.26) (3.27)
1 ln (TBk ) 3.773
where Tbk is the final baking temperature of the carbon anodes in degrees Kelvin, and jrl is the reaction limited current density, given by equation 3.28 (15). The anodes used at Norurl are baked at approximately 1746 K.
jrl = exp 0.560 ln c Al2O3 + cLiF / 4 + 0.276 ( Rb 1.5 ) 5.849
(A/m2)
(3.28)
The relationship between the alumina concentration and anode reaction overvoltage is shown in figure 3.4, at different bath ratios.
38
0.58 0.56 Voltage (V) 0.54 0.52 0.5 0.48 0.46 2 2.5 3 3.5 4 4.5 5 Alumina Concentration (wt%) 5.5 6 6.5 7 Rbath = 1.15 Rbath = 1.40 Rbath = 1.65
Figure 3.4 The relationship between the anode reaction overvoltage and the alumina concentration at different bath ratios.
As can be seen in figure 3.4 the reaction overvoltage is normally high (0.48 - 0.55 V) and changes in the reaction overvoltage are mainly due to variations in the alumina concentration, increasing with lower concentration.
3.4 OhmicVoltagedrops
Ohmic voltage drops due to the electrical resistance of various cell components, contribute to the overall cell voltage. The ohmic voltage drops, as well as the electrode polarization, are manifested as heat. Within the cell, the only resistance that can be varied independently and freely at constant current density is the electrical resistance of the electrolyte, which can be achieved by either altering the composition of the electrolyte, or by reducing the interpolar distance. The other resistances are subject to the cell design. The primary conductive regions within the cell are the anode, the electrolyte and the cathode assembly. The insulating materials have very low electrical conductivity, so their contribution will be neglected.
3.4.1 ElectrolyteVoltageDrop The voltage drop over the electrolyte is considered to exist between the anode and the cathode, and is due to the electrical resistivity of the bath. Assuming a uniform current density, the electrical resistance of the electrolyte follows the basic law of electrical resistance, and can thus be written as:
REL =
1 d A
()
(3.29)
where the distance d is the interpolar distance, A is the total surface area of the anodes and the electrical conductivity of the electrolyte, given by equation 2.57 and 2.58. Ohms law then combines the electric current and the electrical resistance of the electrolyte to determine the electrolyte voltage drop:
U EL = I REL
(V)
(3.30)
39
3.4.2 Anodevoltagedrop Each anode is made up of three parts; the anode butt, the yoke and the connecting rod, as illustrated in figure 3.5. The electrical characteristics of each of these parts will be considered in the following. 3.4.2.1 TheAnodeButt The anode butt is rectangular in shape and is made up of a carbon composite. According to the manufacturers data, the electrical conductivity of the carbon composite used in the anodes at Norurl, is approximately = 1.11104 S/m. The resistivity of the anode butt and correspondently the voltage drop can be Figure 3.5 An anode determined by considering the anode butt as an electrical resistor, with crossAssembly sectional area A and height h, where the height is considered as the distance from the bottom of the anode to the contact surface between the anode butt and the yoke (fig 3.5). Since the carbon takes part in the reduction process, the anodes are consumed over time, and the height of the anode thus gradually decreases over time. By assuming uniform current density in the anode butt and that the anode butt keeps its rectangular shape when consumed; the electrical resistance of the anode butt can be expressed in relationship with the anode consumption:
RAB =
dm 1 1 h An A dt A
()
(3.31)
where is the density of the carbon composite and dmAn/dt the anode consumption rate. Under normal operating conditions, an anode butt is consumed in about 28 days at Norurl. It is preferable to replace the anode butt before the upper surface of the butt goes below the electrolyte level to prevent the yoke from corroding, which is generally reused. At Norurl, the immersion depth of the anode is approximately 15 cm on average, which is then the height constraint of the depleted butt. To minimize the effects on the cells thermal balance when new anodes are installed, the anodes are not all installed at once, but only one at the time. At Norurl, there are 20 anodes in a cell, and the anode replacements divided between the 28 days. This results in each anode having a different height, as illustrated in figure 3.6.
Figure 3.6 The shapes of the 20 anodes, if the first anode is considered as the youngest and the replacements are divided between 28 days.
With an operating current of 180 kA, approximately 1.8 cm of the anodes are consumed each day at Norurl, and the average distance from the bottom of the anodes to the contact surface between the anode butt and the yoke is approximately 29.5 cm. This corresponds to an average electrical resistance of 2.1610-5 and a voltage drop of 0.190 V over the whole anode array at 180kA. 40
3.4.2.2 TheYoke The yoke is a fork shaped connector between the connecting rod and the anode butt, made of an aluminum alloy. The yoke is mounted on the anode butt by inserting each of its three forks into corresponding holes in the carbon and held in place with cast iron for low contact resistance. Due to the complex geometry of the yoke and unknown properties of the aluminum alloy, the theoretical electrical resistance of the yoke could not be determined, but according to manufacturers data of the yokes used at Norurl, the electrical resistance of the yoke is approximately 3.010-6 which will be assumed to remain constant during operation. 3.4.2.3 TheConnectingRod The rectangular shaped connecting rod, commonly referred to as the stem, acts as an electrical connector between the yoke and the anode bridge. It is mounted on the bridge with a bolted clamp which can be used to calibrate the interpolar distance of the anode when installed to match the interpolar distance of the older anodes. The length from the clamp to the top of the yoke is in the range between 0.5 0.95 m, depending on the anode height. The connecting rods at Norurl are made of an aluminum alloy, with electrical conductivity of = 3.8107 S/m, resulting in an electrical resistance of approximately 1.32-2.510-6 . The calibration option will be neglected in the model and the effective length of the connecting rods assumed to remain constant over the anodes lifespan. At Norurl, the average anode height is 37.5 cm, resulting in an average electrical resistance of 1.9110-6 . 3.4.2.4 ContactResistances There are three contact resistances associated with the anode assembly; at the boundary between the yoke and the anode butt, between the yoke and the connecting rod and between the connecting rod and the clamp. The contact resistances are strongly dependent on the quality of the welding and casting, so large variations can be expected. Using measurement data from Norurl, the average contact resistance was found to be approximately 2.8510-6 , which will be assumed constant. The total electrical resistance of the anode assembly can then be expressed as follows:
RAn = R AB + RAY + R AR + RCR
()
(3.32)
where RAB, RAY, RAR and RCR are the electrical resistance of the anode butt, yoke, connecting rod and the contact resistance respectively. The total electrical resistance of the assembly can be determined by considering the anodes as parallel connected. According to Ohms law, the voltage drop over the array is then:
U anodes = I N R An
(V)
(3.33)
where I is the cell operating current and N the number of anodes in the array. For a reduction cell at Norurl, operating at 180kA, this results in approximately 0.264 V drop over the whole anode assembly. 41
3.4.3 BubbleVoltageDrop The horizontal orientation of the anodes in a reduction cell causes gas accumulation beneath the anode surface. The gas volume beneath the working anode surface forms an insulating layer of bubbles, reducing the cross-sectional area of the electrolyte in that zone. This effect causes an increase in the effective resistivity of the electrolyte, resulting in a voltage drop generally known as the bubble voltage drop. Hyde and Welch (17) introduced the following expression for determining the bubble voltage drop:
U bub =
dbub j A
(V)
(3.34)
where dbub is the bubble layer thickness and the surface coverage, given by Haupin (15) as:
d bub = 0.5517 + j A 1 + 2.167 j A
(cm)
(3.35)
2 3 = 0.5090 + 0.1823 j A 0.1723 j A + 0.05504 j A + AE 0.4322 0.3781Rb 0.431 0.1437 c Al2O3 c Al2O3 + AE 1 1.637 Rb 1 + 7.353 c Al O c Al O
2 3
2 3
(3.36)
The magnitude is dependent on the current density and alumina concentration. Variations in the alumina concentration cause changes to the viscosity and surface tension, and these factors influence the bubble size and gas volume under the anode. When the alumina concentration gets close to the critical concentration, the anodes become fully wetted by the gaseous products which leads to a significant increase in the electrical resistance of the bubble layer. This is one of the main causes of the high voltage increase when the anode effect occurs. Assuming that the anode effect occurs when the alumina concentration gets close to 2 wt%, the relationship between the alumina concentration and the bubble voltage drop will be as shown in figure 3.7, at different bath ratios.
2 Rbath = 1.15 1.5 Voltage (V) 1 0.5 0 Rbath = 1.40 Rbath = 1.65
2.5
3.5 4 Alumina Concentration (wt%)
4.5
Figure 3.7 The relationship between the bubble voltage drop and the alumina concentration at different bath ratios.
As can be seen in figure 3.7, the magnitude of the bubble voltage drop is strongly influenced by the bath ratio and the voltage drop goes towards infinity when the alumina concentration gets close to the critical concentration (here 2 wt%). 42
3.4.4 CathodeVoltageDrop The cathode is made up of three parts, the aluminum metal pad, carbon cathode blocks and steel collector bars, as illustrated in figure 3.8. The voltage drop over the cathode lining due to the electrical resistance of its components is an important indicator of the cathodes condition. As the cell gets older, the cathode-voltage drop Figure 3.8 A cathode assembly increases in a more or less regular way mainly due to deposition of sodium in the lining. The rate of the increase depends on many factors including the quality of the cathode blocks and insulating materials, but operational factors are also important. Typically, the voltage drop may increase by approximately 0.1 V over the lifetime of cells. In the following, an expression will be given for determining the voltage drop over a new cathode. 3.4.4.1 MetalPad The metal pad is mainly molten aluminum with 99-99.5 % purity. The electrical conductivity of molten aluminum is approximately 1.0108 S/m which is significantly higher than of the other materials in the current path. The metal pad will thus be considered as a perfect electrical conductor and its contribution to the cell voltage drop neglected. 3.4.4.2 CathodeBlocksandCollectorBars The cathode blocks are rectangular in shape and are made up of a carbon composite. According to the manufacturers data, the electrical conductivity of the carbon composite used in the cathode blocks at Norurl, is = 3.07104 S/m. The cells use preformed blocks that run the full width between the sidewalls, each with two steel collector bars casted into the blocks with cast iron for low contact resistance. Each carbon block and collector bar assembly will be considered in two parts. The first part is the part of the carbon block that is located above the collector bars, and the second part the collector bars and the carbon on each side, as illustrated in figure 3.8. For simplicity, the two collector bars are considered as one. The upper part of a cathode block (part A, in fig. 3.8) can be considered as a rectangular electrical resistor with cross-sectional area A and height h. Assuming uniform current density, the electrical resistance of the upper part, RccA, at Norurl was found to be approximately 6.6110-6 . The lower part (part B, in fig. 3.8) can be considered as three rectangular shaped, parallel connected resistors with one end of each collector bar connected to a busbar. The electrical resistance of the lower part, RccB can then be expressed as:
RccB 1 2 = + + Rcb , L Rcb ,V Rcb , H + Rcc , H + Rcc ,V
1
()
(3.37)
where Rcb,V, Rcb,H and Rcb,L are the vertical and horizontal electrical resistances and the electrical resistance along the length of the collector bars respectively. Correspondently, Rcc,V and Rcc,H are the vertical and horizontal electrical resistances of the carbon on each side. The collector bars 43
extent from the lower cathode part to the outside of the cell, where they are connected to the busbars. The electrical resistance between the lower part and the busbars, RccC, is given by equation 3.38:
RccC =
( lcb lcc dcb )

2 wcb hcb cb
()
(3.38)
lcb, wcb, hcb: Length, width and height of the collector bars (m) lcc : Length of the carbon blocks (m) dcb : Separation distance between collector bars (m) cb : Electrical conductivity of the collector bars (S/m)
3.4.4.3 ContactResistances There are two contact resistances associated with the cathode assembly; the metal pad to carbon cathode interface resistance and the carbon blocks to collector bars contact resistance. Since the carbon blocks are bulky and direct measurements of the interface resistances are complicated, the contact resistances were estimated with proportional calculations based on Lewiss work (18) as done by Biedler (8). According to Lewis and Biedler, approximately 46% of the total cathode lining resistance is due to the contact resistance between the metal pad and the cathode, and 15% the contact resistance between the carbon blocks and the collector bars. Based on these proportions, the total electrical resistance of the cathode assembly can be estimated with the expression given by equation 3.39.
RCa =
( RccA + RccB + RccC )

1 0.15 0.46
N Ca
()
(3.39)
where NCa is the number of carbon cathode blocks. The electrical conductivity of the collector bars used in the cells at Norurl is approximately 1.010-7 S/m, resulting in a total electrical resistance of about 1.55010-6 and correspondently a voltage drop of 0.28 Volts over the cathode assembly at Norurl at 180kA, according to Ohms law.
3.4.5 ExternalVoltageDrop The external electrical resistance depends on the overall cell design, e.g. the cables, connectors, clamps and busbars. Since the total voltage drop is measured between the busbars, the voltage drop due to the external components has to be taken into account in the cell voltage model. At the Norurl reduction plant the external electrical resistance is approximately 1.010-6 on average for each cell, which results in a voltage drop of about 0.19 V at 180kA.
3.5 TotalCellVoltage
The total cell voltage, which is composed of three different types of contributions; the decomposition voltage E0, the overvoltage T and the ohmic voltage drop between the busbars UT, can now be expressed by:
Ecell = E0 + T + U T
(V)
(3.40) 44
The calculated cell voltage as a function of alumina concentration and interpolar distance is shown in figure 3.9.
5 4.8 Voltage (V) 4.6 4.4 4.2 4 IPD = 4.0 cm IPD = 4.5 cm IPD = 5.0 cm
5 6 Alumina concentration (wt%)
Figure 3.9 The cell voltage as a function of alumina concentration and interpolar distance (IPD).
The cell voltage versus alumina concentration curves, each exhibits a minimum. It is desirable to maintain the alumina concentration near that indicated by the minimum cell voltage. As can be seen in figure 3.9, if the anode effect is assumed to occur at 2.0 wt%, the optimal alumina concentration with respect to the minimum cell voltage becomes 3.8 wt%. The lower cell voltage will lead to reduced heat generation in the electrolyte and correspondently, higher production efficiency. When the alumina concentration is held close to the optimal concentration, the simplest way to maintain the thermal balance is then to vary the interpolar distance. It is also possible to vary the electrical resistivity of the electrolyte by altering the composition of the electrolyte, but obviously this method is not as flexible. For a cell at Norurl, operating at 180 kA, with an interpolar distance of 4.5 cm and bath temperature of 970C, the voltage distribution becomes as listed in table 3.2 at typical operating conditions.
Voltage Components Deposition voltage (E0) Anode conc. OV (AC) Cathode conc. OV (CC) Anode reaction OV (AR) Electrolyte voltage drop (UEL) Anode voltage drop (UAn) Bubble voltage drop (Ubub) Cathode voltage drop (UCC) External voltage drop (UExt) Total cell voltage (UCell) Voltage (V) 1.105 0.0235 0.0310 0.5391 1.532 0.264 0.255 0.279 0.180 4.210
Table 3.2 Calculated voltage distribution in a cell at Norurl.
Two reduction cells of different sizes, age and/or design cannot be expected to have the same voltage distribution and are thus not completely comparable. However, the distribution listed in table 3.2, is well inside the range of what can be found in the literature.
45
Based on the previous studies of the different contributions to the cell voltage drop, a steady state cell voltage model was constructed in Simulink. The model is made up of seven subsystems: The decomposition voltage, anode and cathode concentration overvoltages, anode reaction overvoltage, electrolyte voltage drop, bubble voltage and ohmic voltage drops as illustrated in figure 3.10. The block diagrams of all the subsystems are given in the appendix.
Figure 3.10 Block diagram of the cell voltage model
3.6.1 Inputs The Simulink model in figure 3.10 has eight inputs, listed in the table below:
No Name Description Unit Provided by
1 2 3 4 5 6 7 8
cAl2O3 cAl2O3max cLiF K Rbath I IPD Tbath
Alumina concentration Alumina solubility Lithium fluoride concentration Electrical conductivity of the electrolyte Bath ratio Cell current Interpolar distance Bath temperature
wt% wt% wt% S/m A m K
Material balance model Material balance model Material balance model Material balance model Material balance model Control input Control input Thermal balance model
Table 3.3 Inputs to the material balance model in figure 3.10
46
The first five inputs listed in table 3.3 are provided by the material balance model, the bath temperature is provided by the thermal balance model and the two control inputs correspond to two control inputs of the material balance model. If the electrolyte does not contain lithium fluoride, input 3 can be grounded.
3.6.2 Outputs The model in figure 3.10 has four outputs. Only two of these outputs are required by other parts of the cell model, while the other two are only used for observation.
No
1 2 3 4
Name
QELGen Voltage d_bub Q_Cons
Description Heat generation between the electrodes Cell voltage drop Thickness of the bubble layer Total cell energy consumption
Unit kW V m kW
Required by Thermal balance model Material balance model
Table 3.4 Outputs from the cell voltage model in figure 4.10.
The first output is required by the thermal balance model and the third output by the material balance model.
3.7 ModelValidation
In order to validate if the cell voltage model constructed represents the real system to a sufficient level of accuracy, the theoretical voltage distribution was compared to measurement data from Norurl. The cell voltage distribution was measured in two reduction cells at Norurl during a measurement campaign. The voltage drops were measured using a standard digital voltmeter (FLUKE Type 76) and an insulated hook used to access the bottom of the anodes. The measured cell operating conditions on the day of the measurements were reported as follows:
Cell A
Age Cell current Current efficiency Set voltage Bath temperature AlF3 concentration CaF2 concentration 0 years 210.0 kA 59.2 % 4.20 V 973C 10.5 wt% 5.8 wt%
Cell B
3 years 210.2 kA 98.2 % 4.32 V 988C 8.5 wt% 6.0 wt%
Table 3.5 Cell operating conditions on the day of the measurements (measured).
Neither of the cells was stable during the measurements. Cell A came from a period with reduced amperage, leading to low current efficiency and unstable thermal balance. Cell B also suffered from thermal instabilities due to a reduced line load few days earlier and a number of anode effects and was thus not completely stable when the measurements had to be carried out. The electrical resistance of the electrolyte can be adjusted by varying the interpolar distance. The interpolar distance is generally unknown, so in order to validate the cell voltage model the distance was varied until the theoretical drop over the electrolyte matched the measured 47
drop, and the remaining voltage drops compared. The comparisons between the theoretical and measured values for the two reduction cells at Norurl are listed in table 3.6 and 3.7. The three overvoltages cannot be distinguished from the decomposition voltage by direct measurement and can only be estimated from the measured backward electromagnetic field, given by:
Eemf = E0 + AC + AR + CC
(V)
(3.41)
Furthermore, the bubble voltage drop cannot be measured directly, and was thus assumed to be equal to the difference between the measured cell voltage and the summation of all other voltage contributions:
U bub = Ucell ( Eemf + U EL + U An + U CC + U Ext )
Voltage Components
Electrolyte (UEL) Backward emf (Eemf) Bubble voltage (Ubub) Anode assembly (UAn) Cathode assembly (UCC) External (UExt) Total cell Voltage (UCell)
(V)
Error (mV)
59 16 2 12 8 97
(3.42)
Error (%)
3.57 5.97 0.654 3.83 3.96 2.32
Theoretical (V)
1.450 1.709 0.284 0.308 0.325 0.210 4.286
Measured (V)
1.450 1.650 (0.268) 0.306 0.313 0.202 4.189
Table 3.6 Comparison between the theoretical and the measured voltage distribution in reduction cell A
Voltage Components
Electrolyte (UEL) Backward emf (Eemf) Bubble voltage (Ubub) Anode assembly (UAn) Cathode assembly (UCC) External (UExt) Total cell Voltage (UCell)
Theoretical (V)
1.470 1.706 0.309 0.309 0.326 0.210 4.330
Measured (V)
1.470 1.650 (0.273) 0.294 0.390 0.223 4.300
Error (mV)
56 36 15 -64 -13 30
Error (%)
3.40 13.2 5.10 -16.41 -5.83 0.698
Table 3.7 Comparison between the theoretical and the measured voltage distribution in reduction cell B
The effects of the age of cell B can clearly be seen in table 3.7, whereas the cathode voltage drop is about 0.08 V higher than in cell A. If the additional voltage drop caused by the aging of cell B is neglected, it can be seen that the theoretical voltage drop is about 10 mV higher than the measured drop in both cases. The main cause of this error is in the estimates of the backward electromagnetic field and the bubble voltage drop. Both are functions of the alumina concentration and the critical alumina concentration that are generally unknown. Since both cells were not completely stable on the day of the measurements, the overall accuracy of the cell voltage model was considered to be reasonable.
3.8 Summary
The cell voltage is generally the only cell parameter that is measured in real-time. It is thus the only reference point that can be used for automatic cell control. The cell voltage depends on various operational parameters, which are influenced by the cell design, mass transfer conditions, 48
operating procedures and the electrolyte composition. The energy consumption is directly proportional to the cell voltage and hence the cell voltage should be held as low as possible without running into operational problems. Lowering the electrical resistance of the electrolyte affords the greatest flexibility for reducing the cell voltage at constant current density. This can be achieved by either altering the electrolyte composition or by reducing the interpolar distance. In this chapter the various mechanisms that make up the cell voltage were studied, and the mathematical expressions presented used as a basis for creating a steady-state cell voltage model. By comparing the simulated voltage distribution to measurement data, the model was found to be reasonably accurate at steady-state.
49
4 ThermalBalanceModel
4.1 Introduction
Thermal balances are based on the first law of thermodynamics, which is a generalization of the principle of conservation of energy to include energy transfer through heat as well as mechanical work. The first law of thermodynamics states that when heat Q is added to a system while it does work W the internal energy U changes by an amount equal to Q-W:
Q = U + W
(4.1)
A thermodynamic system is in thermal balance, only when Q = W. The primary source of heat generation in an aluminum reduction cell is the electrical power dissipated by the cell. With the nominal line current of 180kA and average cell voltage drop of 4.22 V, the total electrical power consumption of a cell at Norurl is 760 kW. Half of the energy, 380 kW, is required for the reduction process itself, leaving the other half, 380 kW, to be manifested as heat which is lost from the cell to the surroundings. For a given cell, the rate of heat loss is determined by its design and the thermal insulation material used, and once constructed the basic thermal characteristics of the cell are fixed. This limits the extent to which heat loss can be minimized while at the same time permitting electrolysis to occur at the minimum possible temperature within the restrictions imposed by the operating procedure. The absence of frozen electrolyte covering the sidewalls around the electrolyte cavity makes operating cells self compensating for design errors associated with the conditions required to maintain thermal balance. For example, if insufficient sidewall insulation is provided the ledge will freeze to a greater thickness, balancing the heat flow through the sidewalls. One of the most important operational parameters in an aluminum reduction cell is the bath temperature. The bath temperature has to be held within strict limits for the bath to remain molten and to maintain a protective layer of frozen electrolyte on the sidewalls. Failure to do so will result in damage to the cell. The bath temperature is determined by the internal energy (U) in the cell and the composition of the electrolyte so an interrelationship exists between the thermal balance, the cell voltage and the material balance in the cell. The thermal balance is frequently disturbed by various operating procedures, such as an anode change, alumina feeding, and variations in the interpolar distance. In order to predict how these disturbances affect the bath temperature and correspondently temperature dependent parameters, such as the cell voltage, the thermal balance in the cell has to be modeled. This requires a comprehensive study of the heat distribution in the cell as well as the mechanisms of heat transfer. In this chapter the thermal balance in an aluminum reduction cell will be modeled to predict the dynamic behavior of the bath temperature and the thickness of the frozen ledge. The model will be based on the same cell design as before, which is currently being used at Norurl reduction plant.
50
4.2 TheMechanismsofHeatTransfer
The three mechanisms of heat transfer are conduction, convection, and radiation. Conduction is the transfer of the energy of molecular motion within bulk materials without bulk motion of the materials, convection involves the transfer of heat by mass motion of a fluid from one region of space to another and radiation is energy transfer via electromagnetic radiation. In this section a very concise summary will be given of topics in heat transfer, which are particularly relevant to the modeling of the heat distribution in an aluminum reduction cell.
4.2.1 Conduction Conduction is the transfer of energy of molecular motion within bulk materials without bulk motion of the materials. The basis of conductive heat transfer was first postulated by the French mathematician Fourier, simply known as the Fouriers law. The law involves the idea that the heat transferred by conduction is proportional to the temperature gradient in the direction of the heat flow, and to the area A, perpendicular to the direction of the heat flow q:
q = kA
T x
(W)
(4.2)
where T is the temperature in a body at any point, distance x from a datum. The constant of proportionality, k, is known as the thermal conductivity of the material, which indicates its ability to conduct heat through the solid by conduction in W/mK. In practice, the thermal conductivity varies with temperature, but for many solids a suitable mean-value can usually be assumed and the conductivity therefore assumed constant. The partial differential form is used since the temperature may vary in the three space directions x, y and z. Considering a one dimensional temperature gradient in a plane wall of thickness x, integrating from 0 to x, the heat conducted at steady state is then:
q = kA
(T1 T2 )
x
(W)
(4.3)
For a plane wall, the area perpendicular to the heat flow is constant, but in many systems the area is a function of the distance in the corresponding direction. An electrical circuit analogue is widely used in thermal conduction analyses. This is realized by considering the temperature difference to be analogous to the potential difference V, the flow of heat to the current I and a thermal resistance to the electrical resistance R. Ohms law can be expressed as, I = V/R and by comparing this equation to equation 4.3, the thermal resistance for a plane wall can be expressed as:
R=
x kA
(K/W)
(4.4)
As for electric circuits, thermal resistances in series can be added, and the reciprocal of the total resistance equals the sum of the reciprocals of the individual resistances when the resistances are in parallel. 51
For any one-dimensional problem, the heat flow is then:

Q= T RT
(W)
(4.5)
where T is the overall temperature difference over the total thermal resistance RT.
4.2.2 Convection Convection involves the transfer of heat by mass motion of a fluid from one region of space to another. If this motion is caused by density variations resulting from temperature differences the process is known as natural convection but if it is caused by an outside force, such as a pump it is known as forced convection. Heat transfer between a surface at a given temperature Ts and a fluid at a bulk temperature Tb is due to convection. The basic relationship for heat transfer by convection has the same form as that for heat transfer by conduction:
q = hc A (Ts Tb )
(W)
(4.6)
The heat transfer coefficient, hc, is expressed non-dimensionally by the Nusselt number Nu:
Nu = hc x / k
(4.7)
where x is a typical dimension of length for the surface and k is the thermal conductivity of the convecting fluid. Heat transfer by convection is more difficult to analyze than heat transfer by conduction because the mechanism cannot be described by a single property of the heat transfer medium, such as thermal conductivity. Heat transfer by convection varies for different conditions, so in practice, analysis of heat transfer by convection is treated empirically. Many empirical laws have been established for different conditions of convective heat transfer. For natural convection, they are most frequently given in relationship with the Rayleigh number Ra, which is the product of the Prandtl number Pr and the Grashof number Gr. The Prandtl number is an approximation of the ratio of kinematic viscosity to thermal diffusivity, given by:
Pr =
cp = k
(4.8)
: Kinematic viscosity (m2/s) : Thermal diffusivity (m2/s) : Viscosity (Pas) cp :Specific heat (J/(kgK)) k : Thermal conductivity (W/(mK))
When the Prandtl number is small, it means that the heat diffuses very quickly compared to the velocity. This means that for liquid metals the thermal boundary layer is much thicker than the velocity boundary layer.
52
Mathematically, the tendency of a particular system towards natural convection relies on the Grashof number, which is an approximation of the ratio of the buoyancy to viscous force acting on a fluid. The Grashof number is given by:
Gr = g Tx 3
g 2 Tx 3
(4.9)
g : Acceleration due to gravity (m/s2) : Coefficient of cubical expansion (K-1) T : Temperature difference between the surface and the fluid (K) : Density of the fluid (kg/m3)
In natural convection the movement of the fluid due to density differences causes a boundary layer to form on the surface. The boundary layer is either laminar or turbulent, which can be determined by the size of the Grashof number. Some practical expressions for vertical and horizontal surfaces in case of natural convection in relationship to the Rayleigh number are given below (19): Vertical surfaces
Nu = 0.59 ( Ra ) Nu = 0.13 ( Ra )
1/ 4 1/3
4 9 for 10 < Ra < 10 9 12 for 10 < Ra < 10
(4.10) (4.11)
Horizontal surfaces: hot side facing up or cold side facing down

Nu = 0.54 ( Ra )
1/ 4 1/3
5 7 for 10 < Ra < 2 10 7 10 for 2 10 < Ra < 3 10
(4.12) (4.13)
Nu = 0.14 ( Ra )
Horizontal surfaces: hot side facing down or cold side facing up

Nu = 0.27 ( Ra )
1/ 4
5 10 for 3 10 < Ra < 3 10
(4.14)
The heat transfer coefficient, hc can then be determined, using equation 4.6. As for the heat transfer by conduction, the heat transfer by convection can be represented as a thermal resistance of a fluid film:
R= 1 hc A
(K/W)
(4.15)
4.2.3 Radiation Radiation is energy transfer through electromagnetic radiation. Radiant heat transfer does not need a medium to take place. Any material that has a temperature above absolute zero gives off some radiant energy. A so called black body is the best emitter of radiation. It emits the maximum amount of heat for its absolute temperature and absorbs the entire radiation incident on it. The
53
Stefan-Boltzmann law states that the emissive power of a black body with surface area A is directly proportional to the forth power of the absolute temperature T of the body:
P = A T 4
(W)
(4.16)
where the constant of proportionality 5.6710-8 W/(m2K4) is known as the Stefan-Boltzmann constant. Real objects do not radiate as much heat as a perfect black body. Such a body is known as a gray body. The emissive power of a grey body can be expressed by:
P = A T 4
(W)
(4.17)
where the emissivity of a body is defined as a dimensionless number between 0 and 1, representing the ratio of the rate of radiation from a particular body to the rate of radiation from an equal area of an ideal radiating black body at the same temperature. The absorptivity and emissivity of real surfaces varies with the temperature of the body and the source from which they are receiving the radiation. In practice it is most often possible to take mean values, assume that = , and a mean value for all wavelengths. While a body at absolute temperature Ts is radiating, it also absorbs some of the radiation emitted by its surroundings. The net rate of radiation from the body with surroundings at temperature Ta is then:
Qnet = A Ts4 A Ta4 = A (Ts4 Ta4 )
(W)
(4.18)
As for the heat transfer by convection, the heat transfer by radiation can be represented as a thermal resistance of the surface of a body:
R= 1 hr A
(K/W)
(4.19)
where hr is the heat transfer coefficient given by the linearized relationship:

hr = (Ts2 + Ta2 ) (Ts Ta )
(W/(m2K))
(4.20)
It should be noted that the linearized relationship is only valid for small variations in Ts and Ta.
4.2.3.1 Radiosity When a body is in very large surroundings, it can be assumed that the surroundings are effectively black and thus a negligible amount of the radiated energy is reflected back to the body. On the other hand, if the surroundings are not large compared to the body, or when other bodies are in close proximity, the heat transferred between the body and any other body depends on the geometrical disposition of the bodies as well as on their temperature and emissivity. The total radiative heat flow leaving the surface of the body, consisting of emitted as well as reflective radiation is referred to as the radiosity of the surface, given by equation 4.21:
54

n
Bi = Ei + Ri B j Fij
j =1
(4.21)
Ei : The blackbody emissive power of side i (E=Ti4) Bi : Radiosity of surface i Ri : Emissivity of surface i Bj : Radiosity of surface j Fij : Form factor of surface j relative to surface i
The form factor is a dimensionless number describing the fraction of energy which leaves one surface and arrives at a second surface. It takes into account the distance between the two surfaces as well as their orientation in space relative to each other. The overall form factor between surfaces i and j can be determined from the double integral (20):
Fij = cos i cos j 1 dAi dA j 2 Ai Ai Ai r
(4.22)
dAi : Differential area of surface i dAj : Differential area of surface j r : vector from dAi to dAj i : angle between normal of surface i and r j : angle between normal of surface j and r
Analytical solutions to equation 4.22 may be found for relatively simple geometries, such as the infinitely long duct shown in figure 4.1.
Figure 4.1 Two-dimensional gray, diffuse duct.
The form factors between each two surfaces in figure 4.1 can be approximated, using the four form factors (21):
1/ 2 H 2 1 H F12 = F14 = F32 = F34 = 1 + 1 + 2 W W
(4.23) (4.24)
F21 = F23 = F41 = F43 =
W F12 H
55

1/ 2
H 2 F13 = F31 = 1 + W
H W H W
(4.25) (4.26)
W 2 F24 = F42 = 1 + H
1/2
The radiosity of each side can then be determined by solving the radiosity equation (eq. 4.20).
4.2.4 HeatCapacitance The heat capacitance of a body is a measure of how much thermal energy it can store for a given volume and temperature. The heat capacitance can be expressed by:
C = C pV
(J/K)
(4.27)
where and V are the density and volume of the body and the constant of proportionality, Cp, is the so called specific heat capacity of the bodys material in J/(gK). For a given heat flow or heat generation Q, the rate of temperature increase in the body can then be expressed by:
dT =Q/C dt
(K/s)
(4.28)
where T is the temperature of the body.

4.2.5 LumpedThermalCircuit Heat propagation in an isotropic and homogeneous medium in the 3 dimensional space, can be expressed by the general differential form of the three dimensional heat conduction equation, given by:
C p
2T 2T 2T dq T = 2 + 2 + 2 + t y z dt x
(4.29)
where and Cp are the density and specific heat of the medium and q a source of internal heat generation. If equation 4.29 is discretized with size x, y and z in x, y, and z directions, respectively, then the temperature T(x,y,z,t) at node (i,j,k) can be replaced by T(ix,jy,kz), and denoted Ti,j,k. By using central-difference discretization on equation 4.29, the difference equation for the temperature at node (i,j,k) can then be expressed by:
C p V
dTi , j ,k dt
Ay Ax Ax (Ti1, j ,k Ti, j ,k ) + x (Ti+i, j ,k Ti, j ,k ) + y (Ti, j 1,k Ti, j ,k ) x
Ay y
(T
i , j +1, k
A A Ti , j ,k ) + z (Ti , j ,k 1 Ti , j , k ) + z (Ti , j , k +1 Ti , j , k ) + Vqi , j , k z z
(4.30)
where Ax, Ay and Az are the effective cross-sectional areas of the transporting medium in the path of the heat flow in directions x, y and z respectively and V is the volume of node (i,j,k).
56
According to equation 4.4 and the law of energy conservation, equation 4.30 can then be translated into the equivalent lumped thermal circuit shown in figure 4.2, where
Q1 = (Ti 1, j , k Ti , j , k ) Ri 1, j , k , Q3 = (Ti , j 1, k Ti , j , k ) Ri , j 1, k , Q5 = (Ti , j , k 1 Ti , j , k ) Ri , j , k 1 , Q2 = (Ti + i , j , k Ti , j , k ) Ri + i , j , k Q4 = (Ti , j +1, k Ti , j , k ) Ri , j +1, k Q6 = (Ti , j , k +1 Ti , j , k ) Ri , j , k +1
(4.31, 4.32) (4.33, 4.34) (4.35, 4.36) (4.37)
and
C = C p V
Q6
Q3
Ti , j 1,k
Ri , j , k +1 Ti , j ,k
Ti 1, j ,k
Ri , j 1, k
Ri +1, j , k
C
Q2
Q1
Ri 1, j ,k
QG
Ti +i , j , k
Ri , j +1,k
Ti , j +1,k
Ri , j ,k 1
z
x
y
Q4
Q5
Ti , j , k 1
Figure 4.2 A 3-dimensional lumped thermal circuit
The temperature at node (i,j,k) in figure 4.2 is then: dTi , j ,k dt C = Q1 + Q2 + Q3 + Q4 + Q5 + Q6 + QG (4.38)
where QG is the internal heat generation at the node.
4.3 ModelingApproach
The approach that was chosen for modeling the heat distribution in the reduction cell is based on the three dimensional lumped thermal circuit in figure 4.2, commonly referred to as lumped parameter modeling. By using algebraic equations to express the numerous states involved rather than using finite elements, the system can more easily be represented by a state-space model, making the model more convenient for use in the design of a state estimator and control since most control system design procedures are based on state variable models. Furthermore, when modeling a dynamic system, a lumped parameter model has a large advantage over a finite element model in terms of computational efficiency. The biggest disadvantages of using the lumped parameter method are the loss of accuracy caused by the discretization, and that all variable parameters, such as thermal characteristics of the materials become fixed after the model has been constructed. A reduction cell is mainly constructed of rectangular bricks and blocks, and has very few rigid components, so the effects of the discretization can be considered negligible. 57
4.4 ModelBuildingBlocks
Because of turbulence and other localized disturbances, significant concentration and temperature gradients can occur in the two fluid regions. Accurately modeling these gradients would require a detailed, three-dimensional fluid model, increasing the model complexity and computation time. Because of the relative unimportance of the gradients relative to changes in the bulk composition and temperature during normal operation, the modeling of the thermal balance can be greatly simplified by assuming that the bath and metal pad are homogenous with uniform temperature distributions. This assumption and the fact that the fluids are surrounded by congruent insulation, makes it reasonable to assume that uniform temperature gradients are established through the vertical and horizontal planes. The model and correspondently the number of internal states are then greatly simplified by limiting the spatial variation of the variables involved to two dimensions rather than including all three dimensions. Nevertheless, some of the more complex parts of the cell have to be considered in all three dimensions to achieve sufficient model accuracy, and some in only one dimension, such as the frozen ledge profile. When defining the nodes for the lumped parameter method, the cell will be divided into multiple parts, based on the geometry and material nature of each part. Each part will be referred to as a model element and contains one node. Each node represents the transient temperature in the center of the element, and correspondently one state in a state-space model. The system will be considered as being composed of two different types of physical elements; rectangular block elements and sidewall elements. The thermal characteristics of the two element types and their two dimensional representations of three dimensional objects will be described in the following. Boundary layers and radiating surfaces will be considered as simple one dimensional elements.
4.4.1.1 RectangularBlockElements As the name suggest, a rectangular block element represents a part of the system that is rectangular in shape, such as a cathode block or an anode. The thermal resistances between the center point and each side of the block can be determined, using directly the lumped parameter approach, where the thermal resistances are expressed as:
Rx =
1 , 2lhk
Ry =
1 , 2whk
Rz =
1 2lwk
(K/W)
(4.39a, 4.39b, 4.39c)
l : Length of the block (m) w : Width of the block (m) h : Height of the block (m) k : Thermal conductivity of the block (W/mK)
Based on electric circuit theories, if the surface temperature of each of the four vertical sides are considered equal, due to, i.e., congruent thermal insulation, the four thermal resistances between the center point and the four sides can be treated as parallel connected, resulting in the two dimensional representation shown in figure 4.3 of the three-dimensional block. The two resistances RH and RV will be
RH
RV
RV TC
Figure 4.3 2D representation of a 3D rectangular block
58
referred to as the horizontal and vertical thermal resistances, representing the thermal resistances between the center point and corresponding sides of the rectangular block, given by:
RV = Rz =
h 2lwk
(K/W)
(4.40) (K/W) (4.41)
RH = ( Rx 2 ) ( Ry 2 ) =
1 4hk ( w + l )
4.4.1.2 SidewallElements A sidewall element represents a layer of sidewall lining, which in most cases surrounds a rectangular block element or another sidewall element. The sidewall layer is composed of eight rectangular blocks, one for each side of the cell, and one for each corner, combining the four sides into the three dimensional structure shown in figure 4.4.
y
Figure 4.4 A three dimensional sidewall structure is composed of eight rectangular blocks.
Each of the eight blocks can be considered to have the same thermal characteristics as a rectangular block. Since the temperature gradients established through the vertical and horizontal planes of the cell are assumed uniform, each side of the sidewall structure; the top, bottom, inside surface and outside surface, can each be assumed to have uniform surface temperature. The three dimensional sidewall structure in figure 4.4, can then be represented by the two RV dimensional element, shown in figure 4.5, where the temperature of the element, RH TC, is considered as the temperature along a center line, running along the length RH RV TC of each side. On the other hand, if the temperature gradients established through the sidewalls are not considered uniform, the sidewall structure would have to be Figure 4.5 2D modeled with at least four block elements (or eight if the corners are considered representation of a 3D sidewall as well). The horizontal and vertical resistances, RH and RV , shown in figure 4.5 represent the thermal resistances between the center line and the corresponding sides of the sidewall structure, given by:
RH =
x 4hk ( w 2 x + l )
(K/W)
(4.42)
59
RV =
h 4kx ( w 2 x + l )
(K/W)
(4.43)
where x is the thickness of the sidewall insulation. A rectangular block, covered by congruent insulation of uniform thickness on each vertical side can then be represented by two connected elements; a rectangular block element and a sidewall element, as shown in figure 4.6. The heat flow, from the center of the block to the center of the insulation layer is then:
RHb
RHa
TCa
Block
TCb
Sidewall
Figure 4.6 - A combination of a block element and a sidewall element
Qa b =
(Tca Tcb )
RHa + RHb
(W)
(4.44)
Both the block and sidewall elements can only be used to represent solid objects, such as sidewall insulation layers or blocks, where heat transfer is by conduction only. The thermal properties of each material used in the reduction cell can be found in the appendix.
4.4.1.3 BoundaryLayers There are two liquid regions in the cell, the molten aluminum metal pad and the electrolyte. Since the two liquids are assumed to be homogenous, no temperature gradients are assumed to exist in them. Instead, for each liquid a boundary layer exists between the liquid and each neighboring element where turbulence cannot operate and transport is by diffusion only. The heat transfer over a boundary layer is thus considered to be due to convection only, proportional to a given convection heat coefficient hc, creating a temperature drop over the layer. The heat transfer by convection can be represented as a thermal resistance of a fluid film, given by:
R= 1 hc A
(K/W)
(4.45)
where A is the cross-sectional area of the boundary layer. Each boundary layer will thus be considered as a one dimensional element. The heat coefficients depend on numerous variables, such as cell design, operating conditions and local flow velocities at different locations in the cell. Measurements and estimates of the heat coefficients have been published by several researchers, but due to different cell designs the estimates can vary significantly. The coefficients used in the thermal balance model were based on those found in the literature, where the cell design and operating conditions were found similar to those at Norurl. The coefficients are listed in table A.6 in the appendix.
60
4.4.1.4 RadiationandConvectiontoAir Heat lost from the surface of the steel shell and the top of the anodes and crust is due to a combination of heat radiation and convection. To simplify the problem, the linearized relationship introduced in section 4.2.3 for heat radiation was applied and the convection heat coefficient between the surface and air linearized with the same linearization points. The linearization points were found by simulating the steady state conditions of the radiating surfaces and the linearization points corrected manually, until the linearization point agreed with the results of the simulation. This method was found to be more computationally efficient than to simulate the heat distribution with a large number of nonlinear states and then linearizing the radiation afterwards. The total linearization error caused by 10% deviation from the linearization point of a radiating surface is shown in figure 4.7.
6 4 Error (%) 2 0 -2 -4 -6 -10 -8 -6 -4 -2 0 2 4 Deviation from the Linerization Point (%) 6 Tlin = 50C Tlin = 150C Tlin = 200C 8 10
Figure 4.7 - Linearization error introduced by the linearized relationship for heat radiation.
For the average surface temperature of the cell, which is approximately 150C, the linearization error can become quite high as shown in the figure. However, if the ambient temperature remains constant and the cell is operated under normal conditions, the variations in the surface temperatures are generally small (< 2%) and the maximum error thus approximately 1.5%. The combination of radiation and convection can be represented by the combined thermal resistance between the radiating surface and the ambient air, given by:
R=
1 ( hc + hr ) A
(K/W)
(4.46)
where hc and hr are the convection and radiation heat coefficients and A is the area of the radiating surface. The radiation and convection of each radiating surface will thus be considered as a one dimensional element.
4.4.2 InternalHeatGeneration Approximately half of the energy required for operating the cell is lost as heat. The heat is generated in different regions in the cell, and in each region it is proportional to the voltage drops described in chapter 3. Since the electrode polarizations are also manifested as heat, almost two thirds of the heat is generated between the electrodes, more specifically in the electrolyte. The
61
remaining heat is generated in the electrodes and external components. The thermal balance in the cell itself is almost unaffected by the heat lost in external components so the effects of the external voltage drop was neglected in the model. The heat was thus considered to be generated in 3 regions; the electrolyte, the anodes and the cathodes. In the following subsections, the nodes for the lumped parameter method will be defined, by dividing the reduction cell into model elements. For convenience, the cell will be considered in three parts, those being: The potshell
(The steel shell lined with thermal insulation, including the carbon cathode blocks and collector bars)
The cell cavity

(The aluminum metal pad, electrolyte, cavity lining, frozen ledge and air gap)
The Top part.

(The crust, loose alumina cover, anodes and the connecting rod)
The three parts will then be combined into an overall thermal balance model.
4.4.3 ThePotshell A typical modern potshell used for aluminum reduction, consists of a rectangular steel shell, supported by cradles and lined with refractory thermal insulation that surrounds the carbon cathode blocks. The lined potshell serves as the containment vessel for the electrolyte and the liquid metal product. The refractory insulation generally consists of multiple layers of different types of insulating refractories, such as castable refractories, porous refractories and fireclay. The insulating material comes in bricks, boards and blocks, resulting in a rectangular structured lining as shown in figure 4.8.
Figure 4.8 A simplified schematic of the structure of the potlining at Norurl.
The cathode consists of large rectangular carbon blocks with rectangular collector bars running through them. To reduce surface resistance, the carbon blocks are separated by a ramming paste mixture and the collector bars cast into the carbon blocks with cast iron. For high electrical conductivity, the collector bars are made of steel, but unfortunately steel also has high thermal 62
conductivity. The collector bars extend through the sidewall insulation to the outside of the cell, so a considerable amount of heat can be lost through the highly conducting steel. To model the dynamic heat distribution in the potshell, the structure in figure 4.8, was divided into an array of model elements. The boundaries between the elements were drawn according to the geometrical configuration of the insulation layers, such that each element would only consist of one type of material as shown in the figure below.
Layers Symb Description
B1 B2 B3 B4 C1 C2 C3 C4 L1 L2 L3 L4 L5 L6 V0 V1 V2 V3 Hrad Hrad2 Vrad Bottom layer 1 Bottom layer 2 Bottom layer 3 Bottom layer 4 Cathode layer 1 Cathode layer 2 Cathode layer 3 Cathode layer 4 Liquid layer 1 Liquid layer 2 Liquid layer 3 Liquid layer 4 Liquid layer 5 Liquid layer 6 Vertical layer 0 Vertical layer 1 Vertical layer 2 Vertical layer 3
Radiation
Horizontal radiation 1 Horizontal radiation 2 Vertical radiation
Figure 4.9 The potshell divided into model elements for the lumped parameter method.
As shown in figure 4.9, the potshell structure was divided into 4 vertical layers and 14 horizontal layers, where the center blocks (V0) were considered as rectangular block elements and the sidewall layers (V1,V2 and V3) as sidewall elements. The red elements shown in the figure represent one dimensional radiation with natural convection. For the most part the physical elements can be considered in only two dimensions due the congruency of the sidewall insulation as previously explained but this does not apply to the horizontal layer containing the collector bars, where the temperature gradient is not uniform due to the high thermal conductivity of the collector bars. The sidewalls of the collector bar layer (C1) in figure 4.9 were thus divided into two parts, one representing the collector bars, and the other the sidewall insulation as shown in figure 4.10. Due to the high temperature of the collector bars, a temperature gradient is established between the collector bars and the sidewall insulation. To model the interactions between the two parts, the heat distribution in the collector bar layer was thus modeled in three dimensions.
63
(a)
Figure 4.10 The collector bar layer (C1), a) Top view, b) Cross sectional view
(b)
With a voltage drop of approximately 0.28 volts, about 50 kW of heat is generated in the carbon cathode blocks. Because of the high thermal conductivity of steel, it is thus reasonable to assume that the temperature of the part of the collector bars that are surrounded by carbon blocks is equal to the temperature of the carbon surrounding them. The center block of the collector bar layer was thus considered as a whole and modeled as a regular two dimensional block element, instead of separating the cathode blocks from the collector bars. The thermal resistances between the center block and the two sidewall parts in figure 4.10a can be expressed as follows:
R01a = R1a + R0, H Aa , Aa + Ab R01b = R1b + R0, H Ab Aa + Ab
(K/W)
(4.47, 4.48)
Aa : Contact area between element V1a and the center block (m2) Ab : Contact area between element V1b and the center block (m2) R1a : Resistance between element V1a and the center block surface (K/W) R1b : Resistance between element V1b and the center block surface (K/W) R0,H : Horizontal resistance of the center block (K/W)
The thermal resistances between each part of the collector bar layer and the sidewall elements above and beneath (fig. 4.10b) can be determined by a similar approach.
4.5 TheCellCavity
The cell cavity contains the molten aluminum metal pad and the electrolyte. The two melts are surrounded by an inclined sidewall made of graphitized ramming paste mix, covered with a layer of frozen electrolyte which protects the lining from penetration by the corrosive electrolyte. Before modeling the heat distribution in the cell cavity, there are numerous factors that have to be taken into consideration, such as the freezing and melting of ledge, moving boundaries and variable thermal characteristics. Subsequent discussions will concentrate on these.
4.5.1 MetalPad The high electric currents used for powering the cells, generate strong electromagnetic fields in the potroom. The interactions between the magnetic fields and the current flowing through the bath
64
and metal disturb the surface profile of the metal pad and cause the metal to circulate with a complex flow pattern. The cells and the potroom are thus designed in such a way that the disturbances caused by electromagnetic fields are minimized, and hence the metal pad is balanced. For every cell design there is thus an optimum depth and volume for the metal pad with respect to the electromagnetic interference. The cells at Norurl are designed in such a way that during normal operation of a cell, the rate of accumulation of the metal is almost balanced by the volumetric rate of consumption of the anodes. Consequently, the change in interpolar distance becomes very small. The cell operates with a large reservoir of metal, so the variation in the total metal volume during a one days operation is small enough to prevent significant alterations in heat distribution. To retain optimum operating conditions, the metal produced is tapped from the cell once a day. The amount of aluminum tapped off, is equal to the amount produced since the last tapping process was performed, maintaining the size of the metal reservoir. The optimum depth of the metal pad in the cells at Norurl is 20 cm and when operated at 180 kA the rate of accumulation of the metal is approximately 2.0 cm a day. When tapping the metal product, the surface area of the boundary layer between the metal pad and the frozen ledge is thus reduced by approximately 10%, lowering the heat transfer over the boundary layer and hence disturbing the heat balance. Due to a low heat transfer coefficient of the boundary layer and the high thermal capacity of the reservoir, the variations in heat flow are small and take very long time to resolve into noticeable variations in the heat distribution. For simplicity, the metal-pad depth was thus considered constant at the optimum value.
4.5.2 Electrolyte Upon commissioning, approximately 6 tons of electrolyte are added to the cell. With the advantage of using dry scrubbers for fluoride recycling and an automated feeding system, variations in the electrolyte volume are in general small and are not as strongly influenced by the production rate, as the metal pad. The constant freezing and melting of the ledge causes small disturbances due to the phase change, but the maximum deviation is less than 2%, so under normal operating conditions, the electrolyte volume is close to constant. The electrolyte depth is proportional to the interpolar distance, since lowering the anodes reduces the electrolyte volume located between the electrodes. The interpolar distance is generally held close to an optimum value, so the electrolyte depth can be considered to be constant. 4.5.2.1 HeatCapacitance The heat capacitance of the electrolyte depends on both the bath composition and temperature, since the heat capacitance is a function of both the density and the heat capacity of the electrolyte. Using the values determined in chapter 2.12, for typical bath composition and temperature, the heat capacitance of the electrolyte was found to be approximately 1.767 J/gK. Assuming that the heat capacitance is constant at this value introduces an error to the temperature response of the electrolyte, as shown in figure 4.11 and 4.12.
65

2 Tb = 900C Error [%] 0 Tb = 930C Tb = 960C -2
-4
5 6 Al2O3 Concentration [wt%]
Figure 4.11 Error introduced to the heat capacitance of the electrolyte as a function of alumina concentration and temperature, if the heat capacitance is assumed constant at 1.767 J/gK.
3 Tb = 900C 2 Error [%] 1 0 -1 Tb = 930C Tb = 960C
9 10 11 AlF3 Concentration [wt%]
12
13
14
15
Figure 4.12 Error introduced to the heat capacitance of the electrolyte as a function of aluminum fluoride concentration and temperature, if the heat capacitance is assumed constant at 1.767 J/gK.
As can be seen in the two figures, the error introduced by large variations in either temperature or bath composition is relatively small when it is considered that these extreme deviations from the optimal conditions are very rare under normal operation. By neglecting the nonlinear relationship in the thermal balance model, the model complexity is reduced so the heat capacitance of the electrolyte was thus assumed constant.
4.5.3 AirGap The bath and the crust are separated by a small air gap. Most diatomic gases, including air are nonparticipating media so the exchange of radiative energy between surfaces around the air gap is virtually unaffected by the air. As air flow is minimal, the primary means of energy transfer is thus infrared radiation. The height of the air gap depends on various factors, such as the bath temperature, the properties of the top crust, the metal pad depth and the interpolar distance. Although an increase in the gap height lowers the crust surface temperature, it does not change the heat flux significantly. For simplicity the air gap height was thus considered constant, at 2.5 cm, which is reasonable to assume to be the average air gap height, with respect to the production rate and variations in the interpolar distance.
66
The air gap was considered as being composed of three different air gap regions. One surrounds the anode array (gap A), one in the center channel running along the cell (gap B) and one in the small zones between anodes (gap C). Each air gap was modeled as an infinetly long rectangular duct. The cross section and facing surfaces of each air gap region are shown in figure 4.13.
(a)
(b)
(c)
Figure 4.13 The cross section and facing surfaces of each air gap region. a) Air gap A, b) Air gap B, c) Air gap C.
Assuming the surfaces to be isothermal with constant heat flux values across them, the surface heat fluxes can be modeled, using the radiosity approach introduced in section 4.2.3.1. For simplicity, the surface temperature of both side 2 and 4 of each air gap in figure 4.13, was assumed to be equal to the mean temperature between the crust surface and the bath. Using bath temperature of 970C and crust surface temperature of 950C as linearization points, the surface heat fluxes of sides 2-4 can be represented by the linearized relationship:
Qi =
(Tb Tc ) ,
Ri
i = 2,3, 4
(W)
(4.49)
where Tb is the bath temperature, Tc is the surface temperature of the crust and Ri is the linearized effective thermal resistance of side i. Similarly, the surface heat flux of side 1 can be represented by:
Q1 =
(Tc Tb )
R1
(W)
(4.50)
The linerization error introduced to the surface heat flux of side 1 in air gap A, due to variations in the two temperatures is shown in figure 4.14.
2 1 Error [%] 0 -1 -2 20
22
24
26 28 30 32 34 Temperature Difference Tb - Tc [K]
36
38
40
Figure 4.14 Linearization error introduced to the surface heat flux of side 1 in air gap A.
As can be seen in figure 4.14 the linearization error is quite small for small variations in the two temperatures. 67
The calculated effective thermal resistances of each air gap region are listed in table 4.1 in K/W.
Air Gap A Air Gap B Air Gap C 1.59310-3 6.07910-3 113.510-3
Side 1
12.4910-3 40.3910-3 351.610-3
Side 2
1.88510-3 8.69710-3 320.410-3
Side 3
57.8510-3 40.3910-3 351.610-3
Side 4
Table 4.1 The effective thermal resistances of each air gap region.
4.5.4 FrozenLedge There is no material available today for lining the cavity walls, which can resist the corroding effects of the molten electrolyte (22). To protect the lining, the cells are designed so that sufficient heat can be extracted through the sidewalls to drop the temperature of the surface in contact with the electrolyte below the liquidus temperature. This causes the electrolyte close to the walls to freeze, forming a protective layer of frozen electrolyte. Of equal importance, the electrolyte can creep under the metal pad edges by capillary action, where it can freeze. The liquidus temperature TL is the absolute temperature at which freezing starts, and correspondently the surface temperature of the freeze. The liquidus temperature of the electrolyte can be expressed by equation 4.51, given by Rye, et al. (23).
2.2 TL = 1285 + 0.50 c AlF3 0.13 c AlF3
3.45 cCaF2 1 + 0.0173 cCaF2
+ 0.124 c AlF3 cCaF2 7.93 cAl2O3

(4.51)
0.00542 cAlF3 cCaF2
1.5
2 1 + 0.0936 cAl2O3 0.0017 cAl2O3 0.0023 cAlF3 cAl2O3
8.90 cLiF 3.95 cMgF2 + cKF 1 + 0.0047 cLiF + 0.0010 2 LiF
The liquidus temperature sets the lowest temperature at which the cell may be operated without a precipitate forming. Low liquidus temperature is desirable because, all else being equal, lowering the operating temperature by 1C improves the current efficiency by about 0.18%. The temperature drop over the electrolyte-ledge boundary layer is referred to as the superheat of the electrolyte, abbreviated T. The convectional heat flow over the boundary layer is then:
QBL = hBL A T
(W)
(4.52)
where hBL and A are the heat transfer coefficient and surface area of the boundary layer. To account for operational disturbances, such as an alumina addition and the anode effect, the superheat has to be kept sufficiently high, providing excess heat. Under steady state conditions, the heat flow through the boundary layer QBL and the heat flow through the frozen ledge QFL are in balance, QBL = QFL. When the heat flows are not in balance, the situation will adjust itself by freezing or melting of the ledge. The formation of side ledge is thus controlled by the superheat: Increase in T (QBL <QFL)
The heat flow across the boundary layer increases. This raises the temperature at the interface and the excess heat melts the ledge until QBL = QFL.
68
Decrease in T (QBL > QFL)

The heat flow across the boundary layer decreases. This lowers the temperature at the interface and the ledge freezes, providing the required heat until QBL = QFL.
The frozen ledge consists mainly of frozen cryolite, so when freezing occurs, the bulk bath becomes enriched in AlF3, increasing the baths acidity and lowering the liquidus temperature. Upon melting, the opposite occurs. This means that melting and freezing of bath on the walls provides a self-regulating mechanism to moderate changes in both bath temperature and composition. As shown in figure 4.15a, the frozen ledge covers an inclined sidewall, made of baked ramming paste and a vertical insulation layer of silicon carbide based insulation blocks. The convection heat transfer coefficient of the boundary layer between the metal pad and the frozen ledge differ from the coefficient of the boundary layer at the electrolyte-ledge interface, so each will be treated individually. The frozen ledge generally has complex profile and rigid surface. For simplicity, the heat flow over the two boundary layers will thus be considered in only one dimension, despite of the inclined structure. Using the average thickness of the ramming paste in each plane, the heat transfer will be determined from the simple representation of the frozen ledge shown in figure 4.15b where x1 and x2 are the average thicknesses of the frozen ledge in contact with the metal pad and electrolyte respectively. The problem of defining the melting rate of the ledge and correspondently the dynamic ledge thickness is a particular kind of a boundary value problem for partial differential equations, generally known as a Stefan Problem. Since the heat flow over the boundary layers is considered in only one dimension, a simpler approach was chosen. The approach applies to both boundary layers.
(a)
(b)
Figure 4.15 a) The inclined sidewall and frozen ledge, b) A simple representation of the frozen ledge.
4.5.4.1 ModelingApproach When the heat flow through the boundary layer, QBL, and the heat flow through the frozen electrolyte, QFL, are not in balance, the excess heat, QM, goes into melting or freezing ledge. The dynamical change in ledge thickness can then be expressed by the differential equation:
& xFL =
QFL QBL H A D
(m/s)
(4.53)
69
where A is the surface area of the boundary layer, D is the density of the ledge and H is the latent heat of fusion of the molten electrolyte, which was assumed to be 510 J/g (24). In practice, the surface area is not constant, since the moving boundary causes small variations in the internal geometry of the cell cavity. The surface Area is thus dependent on the ledge thickness as follows:
A ( xFL ) = 2 h ( l + w 4 xRP 4 xFL )
(m2)
(4.54)
where l and w are the length and width of the cavity, h is the depth of the electrolyte and xRP is the average thickness of the ramming paste mix on the sides. The heat flow through the ledge can be determined by:
QFL = TL TFL RFL
(W)
(4.55)
where TL, TFL and TR are the liquidus, ledge and ramming paste temperature respectively. The heat resistance of the ledge, RFL, with heat conductivity k is proportional to the ledge thickness:
RFL =
xFL kA
(K/W)
(4.56)
and the temperature inside the ledge can be expressed by the differential equation:
& TFL 1 TL TFL TRP TFL = + CFL 2 RFL RFL + RRP
(4.57)
where CFL is the heat capacity of the ledge and RRP the heat resistance of the ramming paste. From equation 4.53 4.57, the time response of the ledge profile at the electrolyte-ledge interface, for a 1 step increase in the electrolyte temperature is as shown in figure 4.16. The initial conditions used are: TEL = 1243K, TL = 1235K, TRP = 720K and xFL=0.086 m, acquired from a steady state analysis of the overall heat transfer model.
Time Response 1.5
X Melted [cm]
0.5
0.2
0.4
0.6
0.8
1 Time [s]
1.2
1.4
1.6
1.8 x 10
2
5
Figure 4.16 Time response of the ledge profile at the electrolyte-ledge interface if TEL is increased by 1 at t = 2500s
As can be seen in figure 4.16, the change in thickness of the frozen ledge has first-order response. With the initial conditions chosen, the time-constant was found to be approximately = 15,600 sec and the final value K = -0.0136 m. Similarly, using the same initial conditions, the time response of the ledge profile at the metal pad-ledge interface, for a 1 step increase in the metal pad temperature is as shown in figure 4.17. 70

Time Response 2.5 2
X Melted [cm]
1.5 1 0.5 0
0.5
1.5
2.5 Time [s]
3.5
4.5 x 10
5
5
Figure 4.17 Time response of the ledge profile at the metal pad-ledge interface if TEL is increased by 1 at t = 2500s.
As can be seen in figure 4.17 the change in thickness of the frozen ledge is significantly slower than at the electrolyte-ledge interface and the ledge is thicker, which demonstrates the importance of modeling the two boundary layers individually. With the initial conditions chosen, the timeconstant was found to be about = 78,000 sec and the final value K = -0.0228. Because of large variations in ledge thickness, linearizing the heat flow through the boundary layers would introduce significant errors in the model. The self-regulating mechanism of the frozen ledge is one of the most important dynamics in the thermal balance model so in order to represent the heat distribution by a state-space model, the heat flow through the frozen ledge was modeled by a separate model and the two heat flows provided as inputs to the state-space model for greater accuracy.
4.5.4.2 EffectsontheBathComposition As previously mentioned, the melting and freezing of the bath, varies the bath composition, since part of the cryolite is bounded in the frozen ledge. Assuming that the frozen ledge consists mainly of frozen cryolite, the amount of cryolite bounded in the frozen ledge can be expressed as a function of the ledge thickness, given by equation 4.58:
mNa3 AlF6 = ( A1 x1 + A2 x2 )
(4.58)
where A1 and A2 are the surface areas of the two boundary layers, metal pad/ledge and electrolyte/ledge respectively and x1 and x2 are the corresponding ledge thicknesses. The density of frozen cryolite is approximately 2850 kg/m3 (13). In order to account for the change in composition, equation 4.58 has to be integrated into the material balance model.
4.6 TopPart
The top part includes the top crust, the anodes and the connecting rods. About half of the total heat lost from the reduction cell, is lost through the top part, mainly by radiation and convection, but also with the hot transpiring gases flowing through gaps and cracks in the crust. A large volume of air has to be sucked over the surface to capture the emissions, which causes considerable cooling, especially when the temperature of the incoming air is low. Hence, forced convective gas flow conditions will exist, especially for the heat conducted up through the anodes and the connecting 71
rods. To reduce the heat loss, as well as reducing air-burning of the anodes, the crust and the anodes are covered with a layer of loose alumina, which has good insulating properties. The top part is the most complicated part of the overall thermal balance model, due to its geometry, heat distribution and various operational disturbances. Since the top part is one of the main sources of heat loss, the heat flow through each element was considered in all three dimensions, which should increase the model accuracy.
4.6.1 TopCrust When cold alumina powder is added to the surface of the molten electrolyte, each alumina grain becomes covered by a thin layer of frozen electrolyte. As the alumina gradually heats up the frozen layer will remelt after some time, but as more is added, a layer of crust will slowly begin to form, consisting of alumina particles embedded in a mixture of frozen and liquid electrolyte. A temperature gradient is then established through the crust which determines both the thickness as well as the eutectic temperature of the crust. The crust is full of voids and cracks and the chemical content varies throughout the crust. The crust is therefore very inhomogeneous by nature, making it difficult to predict its characteristics. The crusts properties, such as thickness, density and conductivity were thus considered constant. One of the earliest comprehensive studies of the crust formed in industrial cells was carried out by Volberg, et al. (25). In his research the average crust thickness was found to be around 10 cm and the total thickness of the crust and the loose alumina cover approximately 15 cm. Rye, et al. (26) experimented with samples of top crust, prepared in a laboratory and got similar results. The average crust thickness will thus be assumed 10 cm and the average thickness of the loose cover, 5 cm with one exception which will be explained later. To model the heat flow through the crust, the crust was divided into four parts, based on their shape, location and relationship to the air gaps that separate the crust from the molten electrolyte, as illustrated in figure 4.18. Each part will be treated individually.
Figure 4.18 The top crust and anodes, shown from above. The geometry is slightly exaggerated for illustrative purpose.
The first part (A) is the crust surrounding the whole anode array and separated from the molten electrolyte by an air gap, the second part (B) is the crust that runs across the center channel, the third part (C) is the crust located in the small zones between the anodes and the fourth part (D) is the crust located above the frozen sidewall ledge, surrounding crust part A. The thermal resistances between the elements involved, will be described in the following. 72
4.6.1.1 CrustPartA The first part of the crust can be considered as a regular two dimensional sidewall element, with vertical and horizontal thermal resistances and uniform temperature along its center line. Instead of surrounding a single block element, crust A surrounds a number of elements with different temperatures. As can be seen in figure 4.18, the two shorter sides of the crusts inner surface (in the y-z plane) are each in contact with two corner anodes and one end of crust B. The two longer sides (in the x-z plane) are on the other hand, each in contact with half of the anodes and the crust between them (crust C). The horizontal thermal resistances between the center zone of crust A and the surface of each of the neighboring elements (fig 4.19) were considered as a fraction of the horizontal thermal resistance of crust A, determined by the area of the boundary between them and hence:
RCrAx1 = RCrA, H l An RCrA,Ci , RCrAy1 = RCrA, H wAn RCrA,Ci , RCrAx 2 = RCrA, H wCrB RCrA,Ci RCrAy 2 = RCrA, H wCrC RCrA,Ci
TCrA
RCrAy1
RCrAy 2
RCrAx1
RCrA, H
RCrAx 2
Figure 4.19 Crust part A
(K/W) (K/W)
(4.59, 4.60) (4.61, 4.62)
wCrB : Width of crust part B (m) wCrC : Width of crust part C (m) RCrA,H : Horizontal resistance of crust A (K/W)
lAn : Anode length (m) wAn : Anode width (m) RCrA,Ci : Inner circumference of crust A (m)
The relationship in equation 4.63 then holds, where N is the number of anodes.
RCrA, H 4 RCrAx1 + 2 RCrAx 2 + N RCrAy1 + ( N 2 ) RCrAy 2
(K/W)
(4.63)
4.6.1.2 CrustPartB The second part of the top crust lies along the center channel between the anodes. The center channel plays an important role T in the alumina feeding process, as holes are frequently broken into the crust by the crust breakers. The effects of the break and R R feed process on the heat transfer and temperature distribution will be neglected. As can be seen in figure 4.18, both ends of R crust B are in contact with crust A and the adjacent element configuration on its longer sides is the same as for crust A. Crust B can be considered as a three dimensional block element with uniform temperature along the center line that lies between two Figure 4.20 Crust part B points where the distance from crust A is equal to the half the width of the anodes, as shown in figure 4.20. The thermal resistances between the center line of crust B and the boundary between the crust and the neighboring elements can then be determined similarly as for Crust A as follows:
CrB
CrBy1
CrBy 2
CrBx
RCrBy1 = RCrBy wAn lCrB
(K/W)
(4.64)
73
RCrBy 2 = RCrBy wCrC lCrB
(K/W)
(4.65)
where RCrBy is the thermal resistance of crust B in the y direction and lCrB is the length of crust B.
4.6.1.3 CrustPartC The third part of the top crust is composed of N-2 small crust parts; each located between two anodes and can be considered as three dimensional block elements. Since the crust thickness is assumed constant the heat transfer characteristics of the N-2 blocks becomes identical. The width of the blocks is relatively small compared to the length (~1:40), providing high thermal conductivity between the two neighboring anodes. The temperature of these blocks can thus be expected to be close to the temperature of the anodes. Since there is only one adjacent element on each side, the thermal resistances between each of the N-2 blocks of crust part C and the boundary between the crust and the neighboring elements can be determined using equations 4.39a-4.39c.
RCrCy
RCrCx
TCrC
RCrCy
Figure 4.21 Crust part C
4.6.1.4 CrustPartD The fourth part of the top crust surrounds crust part A and its outer surface is in contact with the sidewall lining as shown in figure 4.18. Crust part D can be considered as a regular sidewall element as described in section 4.4.1.2. 4.6.2 LooseCover The heat loss from the top is minimized by covering the crust and anodes with a layer of loose alumina. The normal procedure is to form an alumina cover around a newly set anode, but when the anodes are raised and lowered, the protective cover is disturbed and has to be restored. It is important to maintain an adequate and uniform coverage, which is carried out by the cell operators. Uneven cover thickness is a significant disturbance to the thermal balance model, so the accuracy of the model is dependent on the skill and diligence of the operators. Generally, about 10% of the total heat lost from the cell is lost through the top surface of the loose alumina cover, covering the crust due to a combination of radiation and lightly forced convection. The emissivity of the alumina depends upon a number of factors, including particle size and surface roughness. It is a common estimate to use = 0.40 for smelter grade alumina, as done by, i.e., Rye (26) and Biedler (8). As previously described, the average thickness of the loose cover on top of the crust will be assumed to be 5 cm, but it is often preferred to have a thicker layer on top of the anodes. However, in order to cover all of the anode sides with loose alumina, the narrow space between the anodes
74
has to be filled, so for the cover on top of crust C, the thickness was assumed to be equal to the average anode height above the crust level plus the anode cover thickness. For convenience, the loose cover was divided into four parts, corresponding to the four crust parts. The horizontal thermal resistances between elements were determined the same way as for the crust and the additional layer of alumina between the anodes was considered as a separate block element.
4.6.2.1 VerticalHeatTransfer So far, the heat transfer has only be considered in two directions in the crust and loose cover. Since the division of the crust and loose cover into parts is the same, the vertical thermal resistance between a crust part and loose cover part can be calculated as the sum of the vertical thermal resistance of each part. Furthermore, the thermal resistance between each crust part and the electrolyte over the corresponding air gap can be calculated as the sum of the vertical resistance of the crust part and the air gap resistance listed in table 4.1. The vertical thermal resistances between the loose cover parts and the ambient air inside the cell hood follow the law of convection and radiation, whereas each of the vertical thermal resistances can be calculated as the sum of the vertical thermal resistance of the loose cover part and the thermal resistance of the combined radiation and convection of the corresponding surface area. 4.6.3 Anodes Around one third of the total heat loss in a reduction cell is lost through the top of the anodes, including the connector bars, by both radiation and convection. The anodes are made of a carbonrich mixture, which has high thermal conductivity, providing a good heat conductive path between the electrolyte and the air inside the cell hood. The high resistivity of the carbon makes the anodes a large source of heat, generating around 48 kW in the butt and 6 kW in the connecting bars and yoke in a cell at Norurl operating at 180kA. The heat distribution in the anodes is affected by several disturbances, such as the crust and loose cover profile, as well as the age of the anodes. An effort will though be made to make the heat transfer model of the anodes reasonably detailed and accurate. For simplicity, it was assumed that the height of each anode in the model was constant and equal to the average anode height as described in section 3.4.2.1. Based on this assumption and the assumptions made in the modeling of the top crust, each anode will then only have one of two possible heat transfer profiles, based on the anodes location in the anode array. As can be seen in figure 4.18, the four anodes located in the four corners in the anode array, have two sides in contact with crust part A, while the other N-4 (N: number of anodes), have only one side in contact with crust part B. The corner anodes will then fall under one group, which will be referred to as group A and the other anodes under another group, which will be referred to as group B. All anodes under the same group will then be assumed as having the same heat transfer profile.
75
Each anode is in contact with a number of adjacent elements whereas heat transfer can exist between them (fig 4.22). These are the electrolyte, air gap, crust and loose alumina cover.
Figure 4.22 The elements adjacent to the anodes and the division between the 5 horizontal layers.
To model the interactions between the anodes and the adjacent elements, each anode was considered as being made up of a stack of 5 horizontal layers, where the height of each layer corresponds to the height of the adjacent element. The part of the anode that is in contact with the electrolyte was thus considered as the first layer, the part that is in contact with the air gap the second layer and so on. The heat transfer between the 5 layers and the adjacent elements on the vertical sides will be covered in more detail, as follows. The vertical heat transfer between each two layers can be determined using the vertical thermal resistances of the layers as before.
4.6.3.1 BottomLayer(Layer1) The anode bottom layer is submerged in the electrolyte and the heat transfer between the anode and the electrolyte is due to convection. Heat is transferred through the boundary layers both at the bottom of the anode, as well as the four vertical sides. The total thermal resistance between the center of the anode bottom layer and the electrolyte can then be expressed as follows:
2d l An l w 2d wAn RAnEL = 2hc k An + An An + hc wAn + 2k An d hc + 2k An 2kan + l An hc

lAn,wAn : Length and width of the anode (m) d : Immersion depth (m) kAn : Thermal conductivity of the anode carbon (W/(mK)) hc : Heat transfer coefficient (electrolyte/anode) (W/(m2K))
(K/W)
(4.66)
4.6.3.2 AirGapLayer(Layer2) The anode air-gap layer is surrounded by air, whereas the heat transfer between the anode and the other elements facing the air gap is due to radiation. Which air gaps the anode is facing depends on the anodes location and thus the group it belongs to. Using the effective thermal resistances of the air gaps, derived in section 4.5.3, the thermal resistances between each surface of the anodes in each group and the air gaps become as listed in table 4.2 in K/W. Group A
Side 1 Side 2 Side 3 Side 4 0.4194 0.3516 0.3641 0.1821
Group B
0.4194 0.3516 0.3641 0.3516
Table 4.2 The effective thermal resistances between the anode surfaces and the air gaps in K/W.
76
4.6.3.3 CrustLayer(Layer3) The anode crust layer is surrounded by the top crust, whereas the heat transfer between the anode and the crust is due to conduction. One side of the anode is in contact with crust part B, but it depends on which group the anode belongs to, how much area is in contact with the other two crust parts. An anode in group A has two sides in contact with crust part A and one side in contact with crust part C, while an anode in group B has two sides in contact with crust part C and one side in contact with crust part A. Therefore, there exist four different thermal resistances between the anode crust layer and the top crust:
Total thermal resistance between an anode and the short side of crust A:
RAnCy
RAnCx
TAnCr
l An
wAn
Figure 4.23 The Anode crust layer, seen from the top
RAnCr1 = RAnCx + RCrAx1 =
wAn + RCrAx1 2l An hCr k An
(K/W)
(4.65)
Total thermal resistance between an anode and the long side of crust A:
RAnCr 2 = RAnCy + RCrAy1 =
l An + RCrAy1 2 wAn hCr k An l An + RCrBy1 2 wAn hCr k An wAn + RCrCx 2l An hCr k An
(K/W)
(4.68)
Total thermal resistance between an anode and the long side of crust B:
RAnCr 3 = RAnCy + RCrBy1 =
(K/W)
(4.69)
Total thermal resistance between an anode and the long side of crust C:
RAnCr 4 = RAnCx + RCrCx =
(K/W)
(4.70)
where, RAnCx and RAnCy are the horizontal thermal resistances of the anode crust layer in the x and y direction respectively as shown in figure 4.23 and hCr is the thickness of the crust.
4.6.3.4 CoverLayer(Layer4) The anode cover layer is surrounded by the loose alumina that covers the top crust, whereas the heat transfer between the anode and the cover is due to conduction. As for the crust layer, there exist four different thermal resistances between the anode cover layer and the loose cover, which are constructed in the same manner:
Total thermal resistance between an anode and the short side of loose cover A:
RAnLc1 = RAnLx + RLcAx1 =
wAn + RLcAx1 2l An hLc k An

l An + RLcAy1 2 wAn hLc k An
(K/W)
(4.71)
Total thermal resistance between an anode and the short side of loose cover A:
RAnLc 2 = RAnLy + RLcAy1 =
(K/W)
(4.72)
77

Total thermal resistance between an anode and the long side of loose cover B:
RAnLc 3 = RAnLy + RLcBy1 =
l An + RLcBy1 2 wAn hLc k An wAn + RLcCx 2l An hLc k An
(K/W)
(4.73)
Total thermal resistance between an anode and the long side of loose cover C:
RAnLc 4 = RAnLx + RLcCx =
(K/W)
(4.74)
where, RAnLx and RAnLy are the horizontal thermal resistances of the anode cover layer in the x and y direction respectively and hLc is the thickness of the loose alumina cover.
4.6.3.5 TopLayer(Layer5) The anode top layer is thermally insulated from the air inside the hood by a layer of loose alumina powder. Heat is conducted from the top layer to the loose cover, where it is then lost by a combination of radiation and convection. The characteristics of heat flow are different between the two anode groups, since the anodes in group B have two sides in contact with the loose cover filling up the gap between them, while the anodes in group A have only one. The covering of the loose cover can be very irregular, so for simplicity the heat transfer through the top layer was only considered in one dimension and the cross-sectional area between the top layer and the loose alumina cover equal to the total radiating surface area of the anode top layer. It should be noted that the contact area between the yoke and the anode carbon reduces the radiating surface to some extent. 4.6.4 TheYokeandtheConnectorbar Both the yoke and the connector bar were modeled as rectangular block elements. Since one end of the yoke is in contact with the top layer of the anode butt, the heat flow between the butt and the yoke is by conduction. The connector bar extends to the outside of the hood, so the temperature of the protruding end was assumed to be at the ambient temperature inside the potroom and the temperature of the radiating surfaces the average of the two temperatures. The length of the connector bars was considered as the average length derived in chapter 3. 4.6.5 TheCellHood The cell hood and the top part of the reduction cell are separated by turbulent flow of gases. Since the temperature inside the cell hood is assumed constant, the cell hood was not assumed to take any part in the heat distribution in the cell structure itself. This also applies to other external elements, such as the anode beams, busbars, cradles and the break and feed system. To model the interactions of these elements would require a comprehensive study of the potroom as a whole and were thus neglected.
78
After defining the temperature nodes and the thermal resistances between them, a thermal balance model was constructed by combining the defined elements into one lumped thermal circuit. The model was considered as being composed of two subsystems, a heat distribution model and a ledge profile model as described in the following.
4.7.1 Heatdistributionmodel The heat distribution model consists of 101 differential equations, each representing the transient temperature of a defined node. The material and geometrical properties of the model elements were assumed to be fixed, so the differential equations are all linear. The heat distribution can thus be represented by a linear and time-invariant state-space model, expressed in the general form given by equation 4.75 and 4.76 where the differential and algebraic equations are written in matrix form.
& x(t ) = Ax(t ) + Bu (t ) y (t ) = Cx(t ) + Du (t )
(4.75) (4.76)
The vector, (t) is the time derivative of the vector x(t), u(t) a vector composed of the system input functions and y(t) a vector composed of the defined outputs. Furthermore, the four matrices A, B, C and D in equations 4.75 and 4.76 are commonly known as: A: The system matrix, (n n) B: The input matrix, (n r) C: The output matrix, (p n) D: The feed-forward matrix, (p r) where n, r and p are the number of states, inputs and outputs respectively. After converting the heat distribution model from its differential form to LTI state-space representation, the model was then implemented in Simulink, using the block diagram of a statespace model, illustrated in figure 4.24.
Figure 4.24 A block diagram of a typical state-space model
4.7.1.1 Inputs In order to model the heat distribution in the reduction cell, the model in figure 4.24, requires the eight inputs listed in table 4.3.
79

No 1 2 3 4 5 6 7 8 Name I TL TL2 QEL Ta Th QFL1 QFL2 Description Cell current Liquidus temperature Surface temperature of the crust Net heat generation in the electrolyte Ambient temperature around the shell Ambient temperature in the hood Heat flow from ledge to rammed lining (metal) Heat flow from ledge to rammed lining (bath) Unit A K K W K K W W Provided by Control input Material balance model Control input (fixed) Material and cell voltage models Control input (fixed) Control input (fixed) Ledge profile model Ledge profile model
Table 4.3 Inputs to the heat distribution model.
The two ambient temperatures in table 4.3 and the surface temperature of the crust are assumed to remain fixed, but in the case of changes in design or operating conditions, the inputs can be varied to examine their impact on the heat distribution in the cell. However, these temperatures have negligible effects on the bath temperature, since the self-regulating mechanism of the frozen ledge balances the heat flow from the bath. The cell current determines the ohmic heat losses in the electrodes determined by Ohms law, but the net heat generation in the electrolyte is fed directly to the model and is defined as the difference between the total energy generation between the electrodes and the energy consumed by the process itself. The heat generation is thus derived from both the material balance model and cell voltage model and can be expressed by:
QEL = I ( Eemf + U EL + U bub ) Q process
(4.77)
where Qprocess is the energy required by the process itself, derived from the material balance model constructed in chapter 2, as well as the liquidus temperature. The remaining two inputs; QFL1 and QFL2 are provided by the ledge profile model.
4.7.1.2 Outputs The heat distribution model can be used to simulate the transient temperature response of each and every one of the 101 temperature nodes defined in the lumped thermal circuit. However, only three of these temperatures are relevant for the remaining parts of the reduction cell model, those being the bath temperature, which is required by the material balance and cell voltage model, and the temperatures of the rammed sidewall lining, required by the ledge profile model. The minimum number of outputs is thus only three.
4.7.2 LedgeProfileModel As described in section 4.5.4.1, the expressions for determining the profile of the frozen ledge cannot be linearized, due to the possibly large variations in ledge thickness. The thickness of each of the ledges can be determined from a single nonlinear differential equation, but since the frozen ledge was neglected in the heat distribution model, the transient temperatures of the two ledges had to be included in the ledge profile model as well. The ledge profile model consists of two
80
almost identical models, one for each ledge with the difference being the parameters used. A block diagram of one of these two models is shown in the figure below, derived from equations 4.52 to 4.56.
Figure 4.25 A block diagram of the ledge profile model (bath/ledge interface).
Since neither the ledge thickness nor the convective area can become less than zero, the two signals were limited to be above zero as shown in figure 4.25. Briefly, the model first calculates the convective heat flow, Qconv, over the boundary layer between the melt and the frozen ledge. The difference between Qconv and the heat flowing from the ledge to the rammed lining, Qa1, is then used to determine how much energy goes into melting or freezing ledge (QMelt). From QMelt it is then possible to determine the dynamic thickness of the ledge, x(t), and consequently, the amount of heat flowing through it (QFL1).
4.7.2.1 Inputs In order to model the average thicknesses of the two ledge profiles, each model requires three inputs. The bath and liquidus temperatures are required to determine the heat flow over the boundary layer, and the temperature of the rammed sidewall lining to determine the heat flow from the ledge to the sidewall. 4.7.2.2 Outputs Each model has two outputs; one gives the thickness of the ledge, and the other gives the heat flow from the ledge to the rammed sidewall lining, required by the heat distribution model.
4.8 ModelValidation
In order to verify the accuracy and consistency of the thermal balance model, the two model parts; the state space heat distribution model and the ledge profile model were considered separately. The measurements used for comparison were made during a measurement campaign on a three 81
year old cell of the same type as that modeled and lined with the same thermal insulation material. Prior to the measurement day, the cell was reported to be consistent in stability, voltage, temperature, acidity and bath and metal level. On the day of the measurements no anode was changed, which should minimize disturbances to the thermal balance. The surface temperature measurements were carried out with a pyrometer (RAYTEC Raynger MX4) and the bath temperature with a P-type thermocouple in a nickel plated tube, both with an accuracy of 1C. The ledge profile was measured with an L-shaped hook clamped into a portable frame, using a laser level as a reference. The accuracy of the profile measurements was 1 cm.
4.8.1 TemperatureDistribution The temperature distribution simulation was carried out using the full-order 101 state state-space heat distribution model constructed. The dynamic changes in the thermal balance cannot be measured in an industrial size reduction cell, due to the large mass of the insulation material (large time constants) and the frequent disturbances to the thermal balance. In order to verify the accuracy of the model with respect to the heat distribution, the shell surface temperature distribution was simulated to steady-state and then compared to measured values. Since the surface temperatures are proportional to the heat flow through the shell walls, the temperature distribution should be a reasonable estimate of how heat is distributed around the cell. It should be noted that the linearization relationship used for modeling the radiation and convection between the shell surfaces and the ambient air is only valid to approximate the heat flow through the shell sides, but not the actual surface temperatures. Since the model is based on heat distribution instead of temperature distribution, there is inconsistency between the heat flow and the surface temperatures, so to carry out the comparison, the model had to be re-linearized until the surface temperatures became consistent with the modeled heat flow. The re-linearization only lowers the linearization error by a small amount and its effects on the estimated bath temperature are minimal as will be evident later. On the day of measurements, the measured cell operating conditions were reported as follows:
Parameters
Cell current Cell voltage Cathode voltage drop Ambient temperature (Shell/Hood) Excess aluminum fluoride conc. Calcium fluoride conc.
Measured Values
182.2 kA 4.25 V 0.433 V 36/78 C 10.0 wt% 6.0 wt%
Table 4.4 Cell operating conditions on the day of the measurements (measured).
The high voltage drop over the cathode indicates high sodium deposition in the carbon blocks. For newly set carbon blocks, the voltage drop is generally around 300mV, as simulated by the cell voltage model. The cell voltage is in the normal range and hence the heat generation is not affected
82
by the high voltage drop over the cathode. It is thus likely that the interpolar distance was reduced to balance the heat loss. From the operating parameters in table 4.4, the liquidus temperature was calculated 962C and the heat generated between the electrodes, in excess of the heat required by the process, approximately 230 kW. To carry out the simulation without the ledge profile model, the heat flow through the frozen ledge was set equal to the heat flow over the boundary layer between the melts and the ledge, a condition that would occur under steady-state conditions. Using the measured and calculated operating conditions, the simulated temperature distribution in comparison to the measured distribution was found to be as shown in figure 4.25. For comparison, the distribution was simulated with the cathode voltage drop set to both the measured value and that modeled by the cell voltage model to evaluate how much the ageing of the cathode affects the accuracy of the modeled heat distribution.
400 Temperature (C) 300 200 100 0 Measured Simulated (UCC = 433mV) Simulated (UCC = 300mV) 0 50 Height (cm)
Figure 4.25 Simulated thermal distribution on the sidewall surface, compared to measured values.
100
150
As can be seen in figure 4.25, there are only minor discrepancies between the simulated and measured values at the points measured. However, the lowest measurement point was at 15 cm height and the highest at 120 cm, so the distribution could only be compared within this range. The calculated average of both the measured and simulated distribution is listed in table 4.5. The distribution on the shell bottom was only evaluated by the average, due to the few temperature points which could be calculated by the heat distribution model.
Temperatures
Bath temperature Average bottom temperature Average sidewall temperature
Measured
978 C 91.5 C 210.1 C
Simulated
(UCC = 433 mV) 979.26 C 88.940 C 211.05 C
Simulated
(UCC = 300 mV) 979.5 C 88.419 C 208.50 C
Error (%) (Sim 1/Sim 2) 0.129 / 0.153 2.80 / 3.37 0.452 / 0.762
Table 4.5 Calculated average of measured and simulated temperature values.
The difference between the two simulations is quite small, indicating that the model is almost independent on the age and condition of the cathode. As can be seen in table 4.5, the heat distribution model provides reasonable accurate estimates of the steady-state bath temperature and the temperature distribution on the shell surface.
83
Since the bath temperature is the most important temperature with respect to the reduction process, it was simulated and compared to measured values from four reduction cells. To evaluate how accurate the simulated values are, with respect to the assumptions made when constructing the model, the measurements were taken under different operating conditions and the linearization points held constant. The measured and simulated values are listed in table 4.6.
Cell
Cell A Cell B Cell C Cell D
Metal Level
20 cm 20 cm 17 cm 18 cm
Bath Level
17 cm 15 cm 18 cm 22 cm
Ambient Temp.
36C 34C 27C 42C
Cell Current
182.2 kA 182.2 kA 210.0 kA 210.2 kA
Measured
978C 978C 973C 988C
Simulated
979.5C 977.8C 974.1C 978.0C
Error
0.153 % 0.020 % 0.113 % 0.101 %
Table 4.6 Differences between measured and simulated bath temperatures at various conditions
As can be seen in the table, the largest error is in the simulation of cell A, 1.5C (0.153%). Considering that the accuracy of the thermometer used for the measurements is 1C, it is concluded that the thermal balance model is very accurate for simulating the steady-state bath temperature at various operating conditions, and the accuracy is unaffected by small variations in the linearization points used for constructing the model.
4.8.2 LedgeProfile Due to the large time-constants involved in the ledge dynamics, the accuracy and consistency of the ledge profile model cannot be evaluated in steady state. For an example, if the electrolyte superheat decreased significantly, few hours before the measurements were made, the steady-state thickness predicted by simulation would not be representative. Using bath temperature measurements and calculated liquidus temperatures over 12 days ahead of the measurements, the dynamic ledge profile was simulated using the ledge profile part of the thermal balance model. The sidewall thicknesses were assumed to be at steady-state on the first day. The calculated superheat and simulated ledge thicknesses are shown in figure 4.26 and 4.27 respectively.
30 25
T Superheat
20 15 10 5 0 2 4 6 Days
Figure 4.26 Calculated superheat 12 days prior to the day of the measurements.
10
12
84
30 Ledge Thickness (mm) 25 20 15 10 5 0 2 4 6 Days

Figure 4.27 Simulated ledge thickness 12 days prior to the day of the measurements.
Metal/Ledge
Bath/Ledge
10
12
As can be seen in figure 4.27, there are large variations in the ledge thicknesses on the day of the measurements (day 12). Furthermore, the slow dynamics of the melt/freeze mechanism can clearly be seen in conjunction with the large changes in superheat. To minimize the effects of localized degradations in the sidewall lining to the comparison, measured ledge profiles at different locations in the cell were averaged. The averaged profiles are shown in figure 4.28. The two profiles shown for the end anodes are each an average of the profiles adjacent to two end-anodes and the center profile an average of the profiles adjacent to the center anodes and the anodes on each side (10 anodes total).
Interface
Metal/Ledge Bath/Ledge
Av.Thickness
5.1 cm 15 cm
Interface
Av.Thickness
3.8 cm 8.67 cm
Interface
Av.Thickness
9.0 cm 10.67 cm
(a)
(b)
(c)
Figure 4.28 Measured ledge profile. a) adjacent to end anodes, b) adjacent to center anodes, c) adjacent to end anodes (close to a ventilation duct).
As can be seen in the figure, the ledge profile varies between different locations in the cell. The large ledge thicknesses at the cell ends are most likely due to large freeze formation in the corners of the cavity. The average ledge thicknesses, calculated from the 10 profile measurements were found to be 10.34 and 5.10 cm at the bath/ledge and metal/ledge interfaces respectively. The corresponding simulated thicknesses at the time of the measurements (around noon on the 12th day) were found to be 10.25 and 4.03 cm, which were considered reasonably accurate with respect to the simplifications made in the model construction and the accuracy of the measurement technique. 85
4.9 ModelReduction
The large number of states in the heat distribution model makes it computationally inefficient to perform real time predictions that are suitable for either control design or real-time simulation. To allow fast predictions of changes in the dynamical conditions, the model order had to be reduced with the important system dynamics preserved and minimal change in system performance. Given the high-order linear time-invariant stable plant model G, a low-order approximation Gr thus has to be found such that the infinity norm of the difference ||G-Gr|| is small. The outputs of the plant model are used directly in other parts of the system, so it is desirable to have the steady-state gain of the reduced plant model the same as of the full-order model. The method chosen here for the model reduction is known as balanced residualization, which preserves the steady-state gain of the system. Compared to other well known methods such as balanced truncation and optimal Hankel norm approximation, the residualization performs better at low and medium frequencies, i.e., up to the critical frequencies, while the other two are better where high-frequency matching is required. The balanced truncation and optimal Hankel norm approximation need scaling for steady-state gain preservation which can lead to large errors in the critical frequency range around crossover despite the improvements at steady state. The system is relatively slow, so the god low-frequency matching of the balanced residualization was preferred. As with most model reduction techniques, the balanced residualization method is based on the Hankel singular values of the system. Hankel singular values provide a measure of energy for each state in the system. In a balanced model reduction, the high energy states are retained while low energy states are discarded. In the balanced residualization, this is done by setting the derivatives of the discarded states to zero. Keeping the larger energy states of the system then preserves most of its characteristics in terms of stability, frequency and time responses. Given a stable state-space system, the Hankel singular values are calculated as the square roots of the eigenvalues i for the product of the controllability and observability gramians Wc and Wo as follows:
i = i (WcWo )
where the two gramians are given by:
Wc (t ) = e A BB T e A d
T
(4.78)
Wo (t ) = e A C T Ce A d
T
0 t 0
(4.79) (4.80)
In order to reduce the number of states in the heat distribution model, only a limited number of states can be modeled with the reduced model. Hence, only three states were chosen to be modeled, those being the bath temperature and the temperatures of the rammed sidewall lining. For the heat distribution model with 8 inputs, 3 outputs and 101 states, the first twenty Hankel singular values are shown in figure 4.29, in descending order.
86
Hankel Singular Values 0.6
0.4 abs 0.2 0 0
10 Order
12
14
16
18
20
Figure 4.29 The first 20 Hankel singular values of the full-order heat distribution model.
The figure shows that the system G has most of its energy stored in the first 8 states or so, indicating that the model can be reduced significantly. The balanced residualization gives a stable approximation and a guaranteed bound on the error in the approximation, ||G-Gr||. The error bound after the reduction is twice the sum of the discarded states, given by equation 4.81:
G Gr
2 i
k +1
(4.81)
where n is the full-order, k is the reduced order and i is the ith Hankel singular value of the original system G. From figure 4.29 it was found reasonable to reduce the model to 8 states, which gives a theoretical upper bound of the error of approximately 0.0414.
(a)
(b)
(c)
Figure 4.30 Singular values for model reductions of the thermal balance model from 101 to 8 states. a) Output 1, b) Output 2, c) Output 3
Figure 4.30 shows the singular values of the reduced and full-order models plotted against frequency for outputs 1, 2 and 3 respectively. As can be seen, the residualized system gives a good match in the steady state. The singular values of the gain error (G-Gr) as a function of frequency for each of the three outputs respectively are shown in figure 4.31.
87
(a)
(b)
(c)
Figure 4.31 Singular values for the error system, shown for each output, a) Output 1, b) Output 2, c) Output 3
The H norm of the error system was computed to be 0.0185 occurring at 5.8510-3 rad/s, which is well below the theoretical upper bound for the error norm. As can be seen in figure 4.31, the largest error in the frequency range is in output 3, which is the temperature of the ramming paste sidewall at the electrolyte level. Furthermore, the H norms of the errors from each input to all outputs are listed in table 4.6.
Outputs/Inputs 1 : TBath 2 : TL1V1 3 : TL2V1 1:I 1.4910-11 7.1710-12 4.5210-12 2 : TL 1.2310-2 2.8410-3 2.6310-3 3 : TL2 2.8110-3 2.7210-3 5.2210-3 4 : QEL 5.0310-6 5.0010-7 6.7910-7 5 : Ta 2.9710-3 3.9710-3 1.5810-2 6 : Th 1.1210-3 2.6310-3 5.5410-3 7 : QFL1 1.0110-4 4.1810-4 4.5410-4 8 : QFL2 2.4110-5 1.1810-4 6.7110-4
Table 4.7 The norms of the gain errors between each input and output.
As can be seen in table 4.6, the primary source of error in the bath temperature in the reduced model is the liquidus temperature input and the ambient temperature input in the ramming paste temperatures. The step responses from the liquidus temperature to the bath temperature and the ambient temperature to the two ramming paste temperatures are shown in figure 4.32 respectively for the reduced system in comparison to the full order system.
(a)
(b)
(c)
Figure 4.32 Comparison between the step response of the full order and the reduced system for a step increase in the liquidus temperature at t = 0, a) Response of Output 1, b) Response of Output 2, c) response of Output 3.
88
The simulations confirm that the residualized models response is very close to the full-order models response.
4.10 CombinedThermalBalanceModel
By combining the reduced order heat distribution model and the ledge profile model, a complete thermal balance model was constructed in Simulink. A block diagram of the combined model is shown in the figure below.
Figure 4.33 - The combined thermal balance model.
The combined model will be validated in conjunction with the other two models in the following chapter, where the three models will be combined into one overall model of the reduction cell.
4.11 Summary
In this chapter a thermal balance model of an aluminum reduction cell was constructed, which can be used to predict the dynamic behavior of the bath temperature and the thickness of the frozen ledge. For simplicity, the model was considered in two parts; a heat distribution model and a ledge profile model. The approach that was chosen for modeling the heat distribution in the cell is based on the three dimensional lumped parameter method. As in all modeling endeavors, compromises had to be made between the accuracy of the solution and the computational effort required to obtain a solution. The model and correspondently the number of internal states was greatly simplified by limiting the spatial variation of the variables involved to two dimensions rather than including all three dimensions. Nevertheless, some of the more complex parts of the cell had to be considered in all three dimensions to achieve sufficient model accuracy. From the structural point of view, the model was mostly based on structural drawings of a cell currently being used at Norurl, as well as manufacturers data for the insulation material and electrodes. For other parts it was necessary to rely on variables proposed in the literature, such as heat coefficients, and properties of the frozen ledge and crust. The material and geometrical properties of each part of the cell were assume to be fixed, so each state in the heat distribution model could be represented by a single linear differential equation and the heat distribution in the cell represented by a LTI 89
state-space model. However, in order to preserve the self-regulating mechanism of the frozen ledge, the dynamic variations in the ledge thickness could not be linearized so the heat flow through the frozen ledge was modeled by a separate model and the modeled heat flow provided as an input to the state-space model. The two models were validated separately by comparing both the predicted heat distribution and ledge profile to measurement data from Norurl. The heat distribution model was found to provide reasonably accurate estimates of the steady-state bath temperature and the temperature distribution on the shell surface and the ledge profile model was found to provide reasonably accurate estimates of the thickness of the frozen ledge, both at the metal/ledge interface and the bath ledge/bath interface. The modeled dynamic variations in the bath temperature and ledge thickness could however not be validated since neither could be measured continuously in the physical cell. The heat distribution model consisted of 101 differential equations which made it computationally inefficient to perform real time predictions that are suitable for either control design or real-time simulation. To allow fast predictions of changes in the dynamic conditions, the model order had to be reduced with the important system dynamics preserved and minimal change in system performance. Using a model reduction method known as balanced residualization the model order could be reduced to only 8 states without greatly compromising the accuracy of the model. The residualized model was found to perform well at steady-state and the models response was found to be very close to the full-order models response. The reduced order heat distribution model and the ledge profile model were then combined into a thermal balance model.
90
5 CombinedCellModel
After constructing and validating the material balance model, the cell voltage model and the thermal balance model, the three models were combined into one overall reduction cell model as illustrated by the block diagram in figure 5.1.
Figure 5.1 Block diagram of the combined cell model
The inputs to the cell model in figure 5.1 are the same as required by the physical system, namely the cell current, material feeding and the interpolar distance. Sodium carbonate (Na2CO3) additions are very rarely required at Norurl and under optimal conditions the bath ratio remains constant. The minimum number of inputs required to simulate the behavior of a reduction cell at Norurl is thus only three. However, the number of outputs only depends on which parameters are to be simulated.
5.1 Modelvalidation
With the few parameters that can be measured continuously in an industrial size aluminum reduction cell, the possibilities for validating the dynamics of the cell model are limited. At Norurl the only operational parameters that are measured continuously are the cell voltage and the line current. Other measurements such as measurements of the bath temperature and composition are only carried out intermittently. The bath temperature is generally measured once a day and the bath composition is determined from samples taken from the electrolyte at least once a week. Because of the long intervals between the intermittent measurements, they can only be used for steady-state analysis but not for validating the dynamic behavior of the model. 91
As previously mentioned, the existing cell control uses measured voltage abnormalities to initiate two types of action. One is varying the interpolar distance and the other one is feeding the cell with alumina. Every decision made by the control system as well as the two continuous measurements, are logged by the SCADA system with a time stamp. These logs provide sufficient information for generating the three input signals required by the cell model to simulate the behavior of the reduction cell at the time the data was logged. The model can then be validated to some extent by comparing the simulated cell voltage to the measured cell voltage and evaluating the consistency between the two systems. The primary causes of variations in the cell voltage are fluctuations in the alumina concentration and bath temperature. Other parameters such as the concentrations of the electrolyte additives and correspondently the electrolyte properties are relatively constant over time and their contribution to voltage variations are thus relatively small. The effects of these slowly varying parameters cannot be distinguished from the fluctuations in the alumina concentration and the effects of various unpredictable operational disturbances. The cell model thus has to be validated as a whole, which limits the extent of this method of validation. Another limiting factor is that even under apparently normal operating conditions, the simulated alumina concentration will eventually deviate from the alumina concentration in the physical system, because of the many uncertainties involved in the estimates of the alumina balance. The model can thus only be expected to be able to track the behavior of the physical system over short periods of time. In the following, the dynamics of the combined cell model will be validated to the degree possible with the measurement data available. First the operating conditions of the cell chosen will be described and the inputs signals constructed from the measurement data.
5.1.1 OperatingConditionsofthePhysicalCellChosenfortheValidation In order to validate the cell model, a cell had to be chosen which had not been compromised by any major disturbances to the alumina balance for as long period of time as possible. The metal product is tapped off once a day at Norurl so the maximum time between disturbances is 24 hours (1 day). A cell and date was chosen from the SCADA historical database where the operating conditions of the cell were favorable for model validation. The cell chosen was relatively stable on the day of the measurements and neither an anode change nor the anode effect was reported. Furthermore, the amount of alumina that was fed to the cell was almost equal to the theoretical alumina requirements on the day in question, which indicates that any disturbances to the alumina balance were minimal between metal tappings. The measured cell operating conditions and control system set-points on the day of the measurements were reported as follows:
Parameters Aluminum fluoride concentration Calcium fluoride concentration Bath temperature Voltage set point Current set point Base feeding interval Values 10.0 wt% 6.0 wt% 981C 4.32 V 187 kA 48 sec
Table 5.1 Cell operating conditions and control system set-points on the day of the measurements.
92
In addition, there are several unknown operational parameters that are required to be defined in order to simulate the cell behavior. These parameters were presented in the previous chapters and are summarized in table 5.2. Most of the parameters listed are as suggested in the literature while others are based on typical conditions at Norurl.
Parameters Ambient temperature around the potshell Ambient temperature inside the cell hood Eutectic temperature of the top crust Final baking temperature of the anodes Barometric pressure Concentration of moisture in the raw alumina Concentration of hydrogen in the anode carbon Concentration of aluminum fluoride in the raw alumina Concentration of sodium oxide in the raw alumina Concentration of impurities in the raw alumina Critical alumina concentration f-factor Interfacial velocity at the metal/bath boundary Diffusivity of dissolved metal Excess carbon consumption Values 36C 77C 947C 1473C 101.3 kPa 2.75 wt% 0.09 wt% 1.50 wt% 0.295 wt% 1.0 wt% 2.0 wt% 0.113 0.15 m/s 310-8 m2/s 10 wt%
Table 5.2 Unknown operational parameters that are required to be defined in order to simulate the behavior of the cell chosen for the model validation
Under most conditions, the parameters in table 5.2 can be assumed constant. The critical alumina concentration however does not occur at fixed alumina content in practice as suggested in table 5.2, but within a fairly wide concentration range. The anode effect arises most frequently when the alumina concentration falls below 1-2 wt%, but the reason for these variations is not well understood and cannot be predicted. Any deviations from the suggested value, results in a voltage bias in the cell voltage, which can be compensated for by altering the interpolar distance. In the model validation, these variations are thus not of great concern and the critical alumina concentration assumed constant.
5.1.2 MeasurementNoise The interactions between the magnetic fields and the current flowing through the metal pad and electrolyte causes the boundary between the two melts to oscillate. These oscillations introduce noise to the voltage measurements, which is generally in the range of 10-40mV, but in some cases, such as if the interpolar distance becomes too small, they can range up to a couple of volts, making the voltage measurements very unreliable. In modern reduction cells these magnetohydrodynamic oscillations have however been greatly reduced with automatically controlled interpolar distance. On the day of the measurements the standard deviation of the measurement error was reported to be 6.0 mV, which is in the lower range.
93
5.1.3 TheModelInputSignals As most modern cell control systems, the control system at Norurl is decision based; hence it only acts on changes in conditions. The control signals from the cell control are thus not logged on a continuous basis but instead the adjustments that are made to the existing conditions are logged with a time stamp. The cell model in figure 5.1 requires that the control signals are continuous, so in order to simulate the behavior of the cell chosen to be compared to the physical system the intermittent decisions have to be converted to continuous signals. This can be achieved with decision based signal generators and/or signal converters. In the following, the three control signals required for the model validation will be constructed from the measurement data available. 5.1.3.1 LineCurrent The line current is measured continuously at Norurl. The current measurements are stored in a convenient format in the historical database, which can be used directly as an input signal to the corresponding control input of the cell model. Hence, no scaling or conversion was required. On the day of the measurements, the line current was measured as shown in figure 5.2.
190 Line Current (kA) 188 186 184 182
Measured Target 0 5 10 Time (hours)

Figure 5.2 The cell current on the day of the measurements.
15
20
As can be seen in figure 5.2, the line current was slightly higher than the current set point but relatively stable. The large current transients are mainly due to disturbances occurring in other cells in the potline, such as anode changes, and can thus not be prevented.
5.1.3.2 InterpolarDistance As mentioned in chapter 4.5.1 the cells at Norurl are designed in such a way that during normal operation of a cell, the rate of accumulation of the metal is almost balanced by the volumetric rate of consumption of the anodes. The anodes currently used are however larger than specified by the design, to allow the cells to be operated with higher current than rated and consequently the interpolar distance varies over time. When operated at 187 kA the interpolar distance can be expected to decrease by approximately 2.9 mm a day. At Norurl, the interpolar distance is however controlled by the cell control, which responds automatically to any variations in the interpolar distance.
94
The interpolar distance is altered by lowering or raising the anode bridges with servo actuators. Each actuator is composed of a servo motor and a gearbox with 720 controllable steps per revolution, whereas one degree of rotation corresponds to 37.8 m linear bridge movement. Since the interpolar distance cannot be measured continuously the variations made to the distance are not logged, but instead the feedback signals from the servo motors are used to determine and log the rotation of the motors. On the day of the measurements the rotation of the servo motors was measured as shown in figure 5.3.
Figure 5.3 Rotation of the servo motors on the day of the measurements. The anode bridge is raised by clockwise rotation.
As can be seen in figure 5.3, the anodes were raised intermittently by a small amount to compensate for the decrease in the interpolar distance. At t = 21:07 the metal product was tapped from the cell which lowered the metal level by approximately 2 cm, which was compensated for by lowering the anodes by the equal amount. The measurement data in figure 5.3 cannot be applied directly as an input signal to the cell model, since the data does not represent the actual variations made to the interpolar distance. The measured servo rotation thus has to be scaled and converted to dynamic bridge movements. For simplicity, this was done by constructing a signal converter in Simulink as illustrated by the block diagram in figure 5.4.
(a)
(b)
Figure 5.4 A block diagram of a signal converter for converting measured servo rotation to bridge movements. a) Block diagram, b) Subsystem.
First the measured servo movements are converted into discrete pulses with the same time period as the time steps used for the simulation and the stream of pulses is then integrated for accumulation and the accumulative rotation scaled to bridge movements. In order to replicate the behavior of the interpolar distance on the day of the measurements, the distance variations due to the increased anode size, the measured bridge movements and the effects of the metal tapping had to be taken into account. This was accomplished by constructing a signal generator in Simulink as illustrated by the block diagram in figure 5.5.
95
d_IPD : Bridge Movement 1 Deg

Deg d_IPD
rotation 2 movement d_IPD : Increased Anode Size
-3.36 e-5 m/s d_IPD : Metal Tapping
1 d_IPD
0.002 m/s @ t ~ 21 :07
u < 0.02
Intial interpolar distance Init
Figure 5.5 A block diagram of a signal generator for replicating the behavior of the interpolar distance.
The distance variations due to the increased anode size were assumed to follow a linear ramp with a slope of -2.9 mm/86,400 s and the effects of the metal tapping was modeled as a ramp decrease in the interpolar distance, occurring few seconds before the cell control lowered the anode bridges. The initial interpolar distance on the day of the measurements was found by calculating the voltage distribution in the cell with respect to the cell operating conditions listed in table 5.1. In order to match the cell voltage set point, the interpolar distance required was found to be approximately 4.55 cm. Then by combining the initial interpolar distance, the distance variations due to the increased anode size, the measured bridge movements and the effects of the metal tapping, the behavior of the interpolar distance on the day in question was found to be as shown in figure 5.6. The estimated behavior in the figure could then be used as a control signal for the model validation by connecting the output of the signal generator to the corresponding control input of the cell model.
4.65 Interpolar Distance (cm) 4.6 4.55 4.5 4.45
Estimated Initial 0 5 10 Time (hours)

Figure 5.6 The estimated behavior of the interpolar distance on the day of the measurements.
15
20
According to the estimated behavior in figure 5.6, the interpolar distance did not deviate far from the initial condition on the day of the measurements. It should though be noted that a 1 mm increase in the interpolar distance, results in about 30 mV increase in the cell voltage so dynamic variations in the interpolar distance cannot be neglected.
96
5.1.3.3 AluminaFeeding When a batch of alumina is to be added from an ore bin to the cell, the automated break and feed process is initiated with a control signal pulse from the cell control system. The intervals between the batch additions are logged by the SCADA system at Norurl and on the day of the measurements, the intervals were logged as illustrated by figure 5.7.
250 200 Interval (sec) 150 100 50 0 Control Signal Base Feed
10 Time (hours)
15
20
Figure 5.7 The intervals between batch additions on the day of the measurements.
Under normal operating conditions, the feeding intervals are regularly varied such that the cell is always either underfed or overfed with alumina. As can be seen in figure 5.7, the feeding intervals were oscillating regularly between 35 and 68 seconds during the first 21 hours or so, which signifies that the cell was operating without any major disturbances to the alumina balance prior to the metal tapping. After the metal tapping, the cell was greatly underfed for more than 45 minutes to compensate for the cover material and chunks of crust that fell into the bath when the crust was broken. The measurement data in figure 5.7 cannot be applied directly as an input signal to the cell model since the cell model requires that the signal represents the accumulative amount of alumina that is fed to the cell. The feeding intervals thus had to be converted into discrete pulses and the pulses accumulated. This was accomplished by constructing a variable pulse generator in Simulink with an integrated output as illustrated by the block diagram in figure 5.8.
A : Mass rate (kg/s) T : Duration of each addition (s)
Figure 5.8 A block diagram of a signal generator for converting feeding intervals to accumulative amount of alumina that is fed to the cell
The signal generator in figure 5.8 was then implemented as an interface between the alumina feed input of the cell model and the measurement signal in figure 5.7. Each batch addition was assumed to take 3 seconds and the mass rate 0.5 kg/s with respect to the capacity of the ore bins.
97
5.2 ComparisonbetweentheSimulatedandtheMeasuredCellVoltage
Using the operating conditions in table 5.1 as the initial conditions and the three control signals generated as the model inputs, the behavior of the cell voltage was simulated using the combined cell model in figure 5.1. The simulated time response of the cell voltage in comparison to the measured cell voltage was found to be as shown in figure 5.9.
4.55 4.5 Cell Voltage (V) 4.45 4.4 4.35 4.3 4.25 0 5 10 Time (hours)
Figure 5.9 - Comparison between the measured and simulated cell voltage.
Simulated Measured
15
20
As can be seen in figure 5.9 the two time responses are very similar in the first 20 hours or so and the majority of the peaks are consistent. The simulated voltage then begins to deviate far from the measured voltage because of the disturbances introduced to the alumina balance by the metal tapping process. The numerous transients occurring in the simulated response are caused by the fluctuations in the cell current. Since the cell voltage model is steady state, the model is more sensitive to current disturbances than the physical cell, where these disturbances appear to take longer to resolve. The alumina consumption is proportional to the operating current so these transients in the line current have more effect on the alumina balance in the model than in the physical system. These effects can be seen more clearly in figure 5.10, which shows the errors between the two responses for the first twenty hours.
40 20 Error (mV) 0 -20 -40
10 12 Time (hours)
14
16
18
20
Figure 5.10 The errors between the two responses for the first twenty hours.
The variations in the error voltage in figure 5.10 appear to become more systematic after the large transient in the line current occurring at around t = 10 hours, whereas it begins to follow a similar pattern as the voltage response in figure 5.9. This non-random structure thus suggests that the 98
model does not perform well when the operating current is compromised with transients. Prior to the large current transient, the operating current however varies more slowly and consequently the residuals are distributed more randomly. Because of the few parameters that can be measured continuously in a cell in operation and the short time between any major disturbances, it is difficult to validate the accuracy of the predicted behavior of the cell voltage. Apart from the effects of the transients in the line current, the errors in the estimates are however quite small with respect to the assumptions made and the many uncertainties involved. The standard deviation of the error voltage prior to the transient occurring at t~10 hours was found to be 5.5 mV, which is close to the standard deviation of the measurement error on the day of the measurements.
5.3 PredictedVariationsinAluminaConcentration
In conjunction with the simulated voltage response in figure 5.9, the predicted variations in alumina concentration were found to be as shown in figure 5.11.
3.5 Al2O3 Concentration (wt%)
2.5
10 Time (hours)
15
20
Figure 5.11 Predicted variations in alumina concentration.
As can be seen in figure 5.11, the predicted alumina concentration oscillates around 2.9 wt% and the maximum deviation from the mean value is approximately 0.33 wt%, prior to the metal tapping. The control strategy is to maintain the alumina concentration in a narrow concentration band, but the range set by the cell control at Norurl was unknown. However, according to Grjotheim and Welch (2), the range can be expected to be within 0.5 wt% in a less precise feeding strategy, so the predicted variations in figure 5.11 are within typical limits. The effects of the long period of alumina underfeeding, following the metal tapping process, can clearly be seen in figure 5.11, where the alumina concentration falls down to approximately 2.2 wt%. This is close the critical aluminum concentration, which results in the large voltage peak in figure 5.9. If the mass of the electrolyte was 6.3 tons, the long underfeeding period suggests that approximately 30 kg of alumina fell into the electrolyte during the metal tapping. Despite the fact that the predicted variations shown in figure 5.11 could not be validated, the behavior of the alumina concentration is consistent with the cell voltage drop and the variations are realistic with respect to the feeding strategy at Norurl. 99
5.4 PredictedVariationsinBathTemperature
In conjunction with the simulated voltage response in figure 5.9, the predicted variations in bath temperature were found to be as shown in figure 5.12.
984 Bath Temperature (C) 983 982 981 980 979 978 0 5 10 Time (hours)
Figure 5.12 Predicted variations in bath temperature.
15
20
25
According to the predicted variations in bath temperature in figure 5.12, the bath temperature was relatively stable on the day of the measurements. The average temperature prior to the metal tapping was found to be 981.3C, which is very close to the measured value. It was however unknown at what instance the temperature was measured, but the maximum deviation from the mean value is lower than the uncertainties in the measurements. As can be seen in figure 5.12, the lowest bath temperature occurs at t = 5:10, where the temperature is approximately one degree lower than the mean temperature. The line current was slightly lower than the average and the alumina concentration was relatively high at t = 5:10, which explains the temperature minimum. After the metal tapping, the sudden increase in the cell voltage results in increased heat generation in the electrolyte and correspondently, the bath temperature increases suddenly after the metal tapping. The same behavior occurs during the anode effect. Despite the fact that the predicted variations shown in figure 5.12 could not be validated, the behavior of the bath temperature is consistent with variations in heat generation in the electrolyte. Furthermore, the accuracy of the predicted bath temperature indicates that the modeled balance between heat generation and heat loss in the reduction cell is quite accurate.
5.5 Summary
In this chapter, the material balance, cell voltage and thermal balance models were combined into one overall reduction cell model. The dynamics of the combined model were then compared to the dynamics of an operating cell, using measured cell voltage as a mean for validation. The predicted behavior of the cell voltage was found to be very similar to the measured cell voltage under normal operating conditions. However, when the alumina balance and/or the line current became compromised with disturbances, the predicted voltage began to deviate from the measured voltage. The model was found to be more sensitive to current transients in the line current than the physical system, but in most cases these transients can be neglected. The predicted behavior of the alumina 100
concentration and the bath temperature were found to be realistic and consistent with the behavior of the predicted cell voltage.
101
6 Conclusions
In this work, the most relevant aspects of the Hall Hroult process have been presented and a mathematical model of the process developed. The model was constructed in three parts; a material balance model, a cell voltage model and a thermal balance model. The three models can be used separately or combined, but each enhances the other. With the few parameters that can be measured continuously in an industrial size aluminum reduction cell, the possibilities for validating the dynamics of the cell model were limited, but each model was validated to the extent possible with the measurement data available. The model was constructed with a specific cell design in mind, but with little effort the model can be suited to other cell specifications. The conclusions that were drawn from the development of the cell model and its predictions are:
The material balance model can be used to:
Predict the production and consumption of materials from the most relevant chemical reactions involved in the reduction process. Estimate the production efficiency of a reduction cell and how the efficiency is affected by changes in operating conditions. Predict how the most essential properties of the electrolyte are affected by variations in the electrolyte composition and changes in operating conditions.
The cell voltage model can be used to:
Predict the steady-state voltage distribution in a cell and the relationship between alumina concentration and cell voltage. Determine the optimal alumina concentration in the electrolyte, with respect to energy consumption. Estimate the average interpolar distance by using measured cell voltage as a reference point.
The thermal balance model can be used to:
Predict the heat distribution in a reduction cell and how the bath temperature is affected by operational disturbances and changes in operating conditions. Predict the thickness of the frozen ledge, both at the bath/ledge interface and the metal/ledge interface.
102
The combined cell model can be used to:
Predict how variations in operating parameters affect the stability and operating conditions of an aluminum reduction cell. Specify limits to changes that can be made in the variables of an operating cell. Develop a better understanding of the Hall-Hroult process itself.
In conclusion, the cell model provides a good representation of the physical system, at normal operating conditions, with respect to the main operational parameters. The model provides a useful tool for studying various process interactions, which may aid in future improvements to existing control and operating strategies.
6.1 FutureWork
Even though the cell model developed provides a good representation of the physical system, there are many phenomena that were not taken into consideration. Suggestions for future studies will be given in the following, which are a direct continuation of the work described in this thesis.
Suggestions for the future studies:
Empirical modeling of disturbances to the alumina balance, which are imposed by i.e., an anode change or metal tapping. Empirical modeling of variations in the critical alumina concentration. Empirical modeling of sludge buildup. Modeling of the current distribution and magnetic fields in the cell and how they are influenced by sludge buildup and other operational disturbances.
103
7 References
1. Principles of Aluminum Electrolysis. Haupin, W. E. [ed.] J. Evans. Warrendale, PA : The Minerals, Metals & Materials Society, 1995. Light Metals. pp. 195-203. Grjotheim, Kai and Welch, Barry J. Aluminium Smelter Technology, 2nd Edition. Dsseldorf : Aluminium-Verlag, 1988. Electrode Reaction in the Aluminium Reduction Cell. Pearson, T. G. and Waddington, J. 1947, Discussions of the Faraday Society, Vol. 1, pp. 307-320. Grjotheim, Kai, et al. Aluminium Electrolysis. 2nd Edition. Dsseldorf : Aluminium-Verlag, 1982. Baimakov, Yu. V. and Vetyukov, M. M. Elektroliz rasplavlennykh solei,. Moskow : Metallurgiya, 1966. Haupin, W. E. Production of Aluminum and Alumina. Chichester : Hohn Wiley & Sons, 1987. Study of the Dependence of Anode Overvoltage on the Sulfur Content in the Anode Mass. Burnakin, V. V., et al. 57, Moscow : Tsvetnye Met., 1979, Vol. 52. Biedler, Philip. Modeling of an Aluminum Reduction Cell for the Development of a State Estimator. West Virginia : West Virginia University, 2003. PhD Thesis. Chemical and Physical Properties of the Electrolyte. Haupin, W. E. [ed.] A. R. Burkin. Chichester : John Wiley & Sons, 1987. Production of Aluminum and Alumina, Critical Reports on Applied Chemistry. Vol. 20, pp. 85-119.
2.
3.
4.
5.
6.
7.
8.
9.
10. Alumina Solubility in Molten Salt Systems of Interest for Alumina Electrolysis and Related Phase Diagram. Solheim, A., Skybakmoen, E. and Sterten, A. Warrendale : The Minerals, Metals & Material Society, 1997. Metallurgical and Materials Transactions B. Vol. 28B, pp. 81-86. 11. Current Efficiency. Haupin, W. E. [ed.] A. R. Burkin. Chichester : John Wiley & Sons, 1987. Production of Aluminum and Alumina, Critical Reports on Applied Chemistry. pp. 134-149. 12. Electrical Conductivity of Molten Cryolite-Based Mixtures Obtained With a Tube-Type Cell Made of Pyrolytic Boron Nitride. Hves, J., et al. [ed.] U. Mannweiler. Warrendale, PA : The Minerals, Metals & Materials Society, 1994. Ligth Metals. pp. 187-193. 13. A Dynamic Model for the Energy Balance of an Electrolysis Cell. Taylor, M. P., et al. 1996. Trasactions of the Institution of Chemical Engineers. Vol. 74, pp. 913-933.
104
14. Chase, Malcolm W. Jr. NIST-JANAF Thermochemical Tables. 4th Edition. Washington DC : American Institute of Physics, 1998. 15. Interpreting the Components of Cell Voltage. Haupin, W. E. Warrendale, PA : The Minerals, Metals & Materials Society, 1998. Light Metals. pp. 531-537. 16. Grjotheim, K. and Kvande, H. Introduction to Aluminum Electrolysis, Understanding the Hall-Hroult Process. 2nd Edition. Dsseldorf : Aluminium-Verlag, 1993. 17. The Gas under Anodes in Aluminium Smelting Cells Part I: Measuring and Modelling Bubble Resistance under Horizontally Oriented Electrodes. Hyde, T. M. and Welch, B. J. Warrendale, PA : The Minerals, Metals & Materials Society, 1997. Light Metals. pp. 333-340. 18. Lewis, R. A. Reduction Division Technical Manual I: Technical Fundamentals of the Aluminum Reduction Cell Process. s.l. : Kaiser Aluminum and Chemical Corporation, Inc, 1973. 19. McAdams, W. H. Heat Transmission 3rd edition. New York : McGraw-Hill, 1954. 20. Kreith, Frank, [ed.]. The CRC Handbook of Thermal Engineering. Boca Raton : CRC Press LLC, 2000. 21. Domalski, E.S. and Hearing, E. D., [ed.]. NIST Standard Reference Database. July 2001. Vol. 69. 22. Understanding Boundary Layers. Haupin, Warren. [ed.] Reidar Huglen. Warrendale, PA : The Minerals, Metals & Materials Society, 1997. Light Metals. pp. 319-323. 23. Liquidus Temperature and Alumina Solubility in the System Na3AlF6-AlF3-LiF-CaF2-MgF2. Solheim, A., et al. Warrendale, PA : The Minerals, Metals & Materials Society, 1995. Light Metals. pp. 451-460. 24. Wei, C. C. Modelling of Bath/Ledge Heat Transfer in Hall-Hroult Cells. The University of Auckland. Auckland : s.n., 1996. PhD Thesis. 25. Structure and Thermophysical Properties of Surface Crust Formed on Industrial Aluminium Cells. Volberg, A. A., Sukhanov, E. L. and Belyaev, A. I. 1964. Izv. Nauk. SSSR. Met. Gorn. Delo. Vol. 5, pp. 44-56. 26. Heat Transfer, Thermal Conductivity, and Emissivity of Hall-Heroult Top Crust. Rye, Ketil, Thonstad, Jomar and Liu, Xiaoling. [ed.] J. Evans. Warrendale, PA : The Minerals, Metals & Materials Society, 1995. Light Metals. pp. 441-449. 27. Chase, Malcolm W. Jr., et al. JANAF Thermochemical Tables. 3rd Edition. Washington, DC : American Chemical Society, 1985.
105
28. Alumina Crusting in Cryolitic Melts, Part II : Bulk Properties of Crust. Rye, Ketil . [ed.] Euel R. Cutshall. Warrendale, PA : The Minerals, Metals & Materials Society, 1992. Light Metals. pp. 503-509. 29. A Dynamic Model for the Energy Balance of an Electrolysis Cell. Taylor, M. P., Zhang, W. D. and Schmid, S. 1996. Transactions of the Institution of Chemical Engineers. pp. 913-933. 30. A Study of Cell Ledge Heat Transfer Using an Analogue Ice-Water Model. Wei, C. C., et al. [ed.] U. Mannweiler. Warrendale, PA : The Minerals, Metals & Materials Society, 1994. Light Metals. pp. 285-293.
106
AppendixATables
Chemical Substance Aluminum (l) Aluminum Fluoride (s) Aluminum Oxide () Aluminum Oxide () Carbon (s) Carbon Monoxide (g) Carbon Dioxide (g) Hydrogen (g) Hydrogen Fluoride (g) Water (g) Sodium Tetrafluoroaluminate (g) Cryolite (,l) Sodium Carbonate (s) Sodium Oxide (,) Sodium Fluoride (s) Sulfur (s) Sulfur Dioxide (g) Al AlF3 Al2O3 Al2O3 C CO CO2 H2 HF H2O NaAlF4 Na3AlF6 Na2CO3 Na2O NaF S SO2 Hf (kJmol-1) T=1100K T=1200K T=1300K 0 0 0 -1509.431 -1507.984 -1506.471 -1692.437 -1691.366 -1690.190 -1669.354 -1667.684 -1665.902 0 0 0 -112.586 -113.217 -113.870 -394.838 -395.050 -395.257 0 0 0 -274.896 -275.231 -275.558 -248.460 -248.997 -249.473 -1859.349 -1956.723 -1956.466 -3196.537 -3470.530 -3451.697 -1110.305 -1297.547 -1289.016 -412.418 -603.913 -588.116 -572.322 -667.607 -448.581 0 0 0 -361.835 -361.720 -361.601
Table A.1 The enthalpy of formation of various substances involved in the aluminum reduction process. Taken from JANAF (27)
Chemical Substance Aluminum (l) Aluminum Fluoride (s) Aluminum Oxide () Aluminum Oxide () Carbon (s) Carbon Monoxide (g) Carbon Dioxide (g) Hydrogen (g) Hydrogen Fluoride (g) Water (g) Sodium Tetrafluoroaluminate (g) Cryolite (,l) Sodium Carbonate (s) Sodium Oxide(,) Sodium Fluoride (s) Sulfur (s) Sulfur Dioxide (g) Al AlF3 Al2O3 Al2O3 C CO CO2 H2 HF H2O NaAlF4 Na3AlF6 Na2CO3 Na2O NaF S SO2
S (JK-1mol-1) T=1100K T=1200K T=1300K 76.426 79.189 81.730 189.147 198.057 206.334 192.189 203.277 213.602 200.178 211.787 222.597 26.548 28.506 30.346 237.726 240.679 243.431 274.528 279.390 283.932 169.112 171.790 174.288 212.168 214.845 217.340 236.731 240.485 244.035 502.077 513.392 523.836 659.829 694.243 725.901 334.537 377.525 392.696 186.205 194.645 212.098 120.289 125.768 157.380 137.187 138.829 140.352 310.995 315.824 320.310
Table A.2 The Entropy of various substances involved in the aluminum reduction process. Taken from JANAF (27)
107
TK A B C D E F G H
Al2O3 () 298-2327 102.4290 38.74980 -15.91090 2.628181 -3.007551 -1717.930 146.9970 -1675.690
AlF3 298-728 728-2523 -27.17362 92.96722 594.2661 8.667826 -1042.272 0.148503 662.4444 -0.006692 0.019262 -0.942781 -1520.775 -1541.369 -103.0051 170.2754 -1510.424 -1510.424
NaF 298-1269 53.00220 -5.952150 12.56350 0.378227 -0.489113 -592.6740 113.8140 -575.3840
CaF2 298-1424 89.88404 -46.12023 65.27793 -17.86271 -1.139438 -1255.008 182.0057 -1225.912
MgF2 298-1536 68.55149 22.72586 -12.16218 2.573717 -1.140865 -1149.445 127.4522 -1124.241
LiF 1121.3-3000 64.18298 4.19542310-11 -2.30227110-11 3.92424210-12 1.15023710-12 -617.7885 120.6335 -598.6509
Table A.3 - Shomate coefficients of various substances involved in the aluminum reduction process. Taken from NIST-JANAF (14)
TK A B C D E F G H
Na3AlF6 1285- 3000 395.5144 1.72426910-8 -7.86855710-9 1.19487110-9 2.44908910-9 -3437.779 622.1315 -3250.658
298-1023 25.57540 177.7100 -166.3350 57.61160 0.338149 -431.0160 61.79400 -417.9820
Na2O 1023-1243 -125.7730 302.0740 -140.6420 21.32400 38.28310 -301.6530 -42.66700 -417.9820
1243-1405 2240.950 -3209.970 1803.690 -359.0730 -386.2480 -1877.630 2421.660 -417.9820
298-723 175.2010 -348.0580 743.0720 -305.5510 -1.634221 -1178.980 415.0610 -1130.770
Na2CO3 723-1123 -1067.000 2469.340 -1829.060 505.7480 100.1820 -607.1240 -1356.440 -1130.770
1123.15-2500 189.5350 -0.000007 0.000002 -5.20510010-9 -0.000003 -1183.060 342.9690 -1108.510
Table A.4 Shomate coefficients of various substances involved in the aluminum reduction process. Taken from NIST-JANAF (14)
Material Silicon carbide blocks Insulating moler bricks Fire clay bricks Insulating fibre mats Insulating blocks Carbon cathode blocks Fire clay granulate Ramming paste mixture Alumina powder filler Steel collector bars Steel shell Molten aluminum Carbon anode blocks Frozen electrolyte Loose alumina cover Cryolitic crust Electrolytic bath
Thermal Conductivity W/(mK) 28.0 0.19 1.50 0.12 0.21 12.20 1.30 10.00 0.20 50.0 31.0 500 (8) 6.30 1.07 (29) 0.30 (26) 1.30 (26) 1000 (8)
Specific Heat kJ/(kgK) 1.261 0.800 1.400 1.000 0.750 0.889 1.400 0.800 1850 (8) 0.120 0.122 1.088 (18) 0.889 1.850 (29) 1.850 (8) 1.850 (8) 1.767
Bulk Density kg/m3 2680 750 2150 225 425 1560 1850 1600 2400 (28) 7840 7840 2375 1450 2850 (29) 2400 (28) 2400 (28) 2173
Table A.5 Material specifications of each material used in the reduction cell. The specifications are from manufacturers data unless otherwise specified.
108

Interface Electrolyte/Anodes Electrolyte/Ledge Electrolyte/Metal Pad Metal Pad/Cathode Metal Pad/Ledge Heat Transfer Coefficient W/(m2K) 1200 (8) 1200 (8) 1000 (29) 400 (29) 400 (30)
Table A.6 The heat transfer coefficients considered in the thermal balance model
109
AppendixBModelBlockDiagrams
B.1MaterialBalanceModel
Figure B.1 Block diagram of the material balance model.
Figure B.2 Block diagram of the mass to concentration conversion subsystem in figure B.1 (Mass to wt%).
110
Figure B.3 Block diagram of the current efficiency subsystem in figure B.1.
Figure B.4 Block diagram of the production and consumption subsystem in figure B.1.
111
Figure B.5 Block diagram of the primary reaction subsystem in figure B.4.
Figure B.6 Block diagram of the enthalpy of reaction subsystem in figure B.5.
Figure B.7 Block diagram of the enthalpy of heating subsystem in figure B.5.
112
Figure B.8 Block diagram of the fluoride evolution subsystem in figure B.4.
113
Figure B.10 Block diagram of the vaporization subsystem in figure B.4.
Figure B.11 Block diagram of the total vapor pressure subsystem in figure B.10.
Figure B.12 Block diagram of the partial pressure of the monomer subsystem in figure B.10.
114
Figure B.14 Block diagram of the entrainment subsystem in figure B.4.
Figure B.15 Block Diagram of the material input subsystem in figure B.1.
115
Figure B.16 Block diagram of the alumina dissolution subsystem in figure B.15.
Figure B.17 Block diagram of the alumina dissolution subsystem in figure B.16.
116
Figure B.18 Block diagram of the enthalpy of dissolution subsystem in figure B.16.
Figure B.19 Block diagram of the enthalpy of dissolution subsystem in figure B.18.
Figure B.20 Block diagram of the sodium oxide reaction subsystem in figure B.16.
117
Figure B.22 Block diagram of the sodium carbonate reaction subsystem in figure B.15.
Figure B.24 Block diagram of the aluminum fluoride addition subsystem in figure B.15.
118
Figure B.25 Block diagram of the electrolyte properties subsystem in figure B.1.
119
Figure B.26 Block diagram of the electrolyte density subsystem in figure B.25.
Figure B.27 Block diagram of the bath ratio subsystem in figure B.25.
120
Figure B.28 Block diagram of the electrolyte viscosity subsystem in figure B.25.
Figure B.29 Block diagram of the solubility of Al2O3 subsystem in figure B.25.
121
Figure B.30 Block diagram of the subsystem for determining A in figure B.29.
Figure B.31 Block diagram of the subsystem for determining B in figure B.30.
122
Figure B.32 Block diagram of the solubility of aluminum subsystem in figure B.25.
Figure B.33 Block diagram of the electrical conductivity of the electrolyte subsystem in figure B.25.
123
Figure B.34 Block diagram of the specific heat of the electrolyte subsystem in figure B.25.
Figure B.35 Block diagram of the liquidus temperature subsystem in figure B.25.
124
B.2CellVoltageModel
Figure B.36 Block diagram of the cell voltage model.
125
Figure B.37 Block diagram of the decomposition voltage subsystem in figure B.36.
Figure B.38 Block diagram of the activity of alumina subsystem in figure B.37.
Figure B.39 Block diagram of the anode concentration overvoltage subsystem in figure B.36.
126
Figure B.40 Block diagram of the critical current density subsystem in figure B.39.
Figure B.41 Block diagram of the cathode concentration overvoltage subsystems in figure B.36.
Figure B.42 Block diagram of the anode reaction overvoltage subsystem in figure B.36.
127
Figure B.43 Block diagram of the reaction limited current density subsystem in figure B.42.
Figure B.44 Block diagram of the electrolyte voltage drop subsystem in figure B.36.
Figure B.45 Block diagram of the bubble voltage drop subsystem in figure B.36.
Figure B.46 Block diagram of the bubble layer thickness subsystem in figure B.45.
128
Figure B.47 Block diagram of the bubble surface coverage subsystem in figure B.45.
Figure B.48 Block diagram of the ohmic voltage drops subsystem in figure B.36.
129

Thesis - Stefan Winkel Jessen (2008)

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Thesis - Stefan Winkel Jessen (2008)

Uploaded by

Copyright:

Available Formats

Mathematical Modeling of a Hall Hroult Aluminum Reduction Cell

Stefan Winkel Jessen

Motivation ........................................................................................................................... 5 Overview ............................................................................................................................. 5

2.5.1 2.5.2 2.5.3 2.5.4 2.6 2.7 2.8 2.9 2.10

Current Efficiency ............................................................................................................. 21

Energy Considerations ....................................................................................................... 24

The Model Construction .................................................................................................... 27

Model Validation ............................................................................................................... 29 Summary............................................................................................................................ 30

Ohmic Voltage drops ......................................................................................................... 39

3.6.2 3.7 3.8 4

Model Validation ............................................................................................................... 47 Summary............................................................................................................................ 48

4.2.1 4.2.2 4.2.3 4.2.4 4.2.5 4.3 4.4

The Cell Cavity .................................................................................................................. 64

Top Part ............................................................................................................................. 71

The Model Construction ....................................................................................................79

Model Validation ............................................................................................................... 81

4.8.2 4.9 4.10 4.11 5

Ledge Profile .............................................................................................................. 84

Conclusions ............................................................................................................................ 102 6.1 Future Work..................................................................................................................... 103

References .............................................................................................................................. 104

4.6 4.7 4.8 4.9 4.10 4.11 4.12

4.29 4.30 4.31 4.32

Mathematical Modeling of a Hall-Hroult Reduction Cell

Mathematical Modeling of a Hall-Hroult Reduction Cell

Mathematical Modeling of a Hall-Hroult Reduction Cell

Mathematical Modeling of a Hall-Hroult Reduction Cell

Mathematical Modeling of a Hall-Hroult Reduction Cell

Mathematical Modeling of a Hall-Hroult Reduction Cell

Mathematical Modeling of a Hall-Hroult Reduction Cell

Input + Generation = Output + Accumulation

Mathematical Modeling of a Hall-Hroult Reduction Cell

where I is the cell current in amperes.

Mathematical Modeling of a Hall-Hroult Reduction Cell

Mathematical Modeling of a Hall-Hroult Reduction Cell

Gaseous fluoride evolution Particulate emission Entrainment of particulate materials

Subsequent discussions will concentrate on these.

Mathematical Modeling of a Hall-Hroult Reduction Cell

462 191Rb Pb %CE

cH 2 O c c Al2O + H max 3 37.44 21.5 c Al2O3

(2.16) (2.17) (2.18)

mAl2O3 = 0.8726 FG m AlF3 = 1.473 FG

Mathematical Modeling of a Hall-Hroult Reduction Cell

3 B = 7.0174 + 0.6844 Rb 0.08464 Rb +

where K is the dimensionless equilibrium constant for dimerization, given by:

21414 K = exp + 15.6 Tb

Mathematical Modeling of a Hall-Hroult Reduction Cell

m AlF3 = 1.105 FVP

5 6 Al2O3 Concentration (wt%)

2 1 17030 + 29800 Rb 13000 Rb + 67cAl2O3 %CE 173 0.3896 2 + 141.6 Rb

Mathematical Modeling of a Hall-Hroult Reduction Cell

mAlF3 = 1.473 cAlF3 FE f F1

(2.32) (2.33) (2.34)

Mathematical Modeling of a Hall-Hroult Reduction Cell

Mathematical Modeling of a Hall-Hroult Reduction Cell

Mathematical Modeling of a Hall-Hroult Reduction Cell

Mathematical Modeling of a Hall-Hroult Reduction Cell

400 500 Time (sec)

Mathematical Modeling of a Hall-Hroult Reduction Cell

Products Al CO CO2 HF NaAlF4

Net production (kg) 1385 125.6 1595 19.86 24.22

Mathematical Modeling of a Hall-Hroult Reduction Cell

cMgF2 3.392 5.24 10 Tb 0.01407 cMgF2

3 cLiF 10 4 2.358 4.9 10 Tb